openkylin
/
linux
mirror of https://gitee.com/openkylin/linux.git
9
0
Fork 0
Code Issues Projects Releases Wiki Activity
linux/drivers/infiniband/core/cache.c

1299 lines
32 KiB
C
Raw Normal View History

Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
/*
* Copyright (c) 2004 Topspin Communications. All rights reserved.
[PATCH] IB: Add copyright notices Make some lawyers happy and add copyright notices for people who forgot to include them when they actually touched the code. Signed-off-by: Roland Dreier <rolandd@cisco.com>
2005-08-11 14:03:10 +08:00
* Copyright (c) 2005 Intel Corporation. All rights reserved.
* Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2005 Voltaire, Inc. All rights reserved.
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <linux/module.h>
#include <linux/errno.h>
#include <linux/slab.h>
Detach sched.h from mm.h First thing mm.h does is including sched.h solely for can_do_mlock() inline function which has "current" dereference inside. By dealing with can_do_mlock() mm.h can be detached from sched.h which is good. See below, why. This patch a) removes unconditional inclusion of sched.h from mm.h b) makes can_do_mlock() normal function in mm/mlock.c c) exports can_do_mlock() to not break compilation d) adds sched.h inclusions back to files that were getting it indirectly. e) adds less bloated headers to some files (asm/signal.h, jiffies.h) that were getting them indirectly Net result is: a) mm.h users would get less code to open, read, preprocess, parse, ... if they don't need sched.h b) sched.h stops being dependency for significant number of files: on x86_64 allmodconfig touching sched.h results in recompile of 4083 files, after patch it's only 3744 (-8.3%). Cross-compile tested on all arm defconfigs, all mips defconfigs, all powerpc defconfigs, alpha alpha-up arm i386 i386-up i386-defconfig i386-allnoconfig ia64 ia64-up m68k mips parisc parisc-up powerpc powerpc-up s390 s390-up sparc sparc-up sparc64 sparc64-up um-x86_64 x86_64 x86_64-up x86_64-defconfig x86_64-allnoconfig as well as my two usual configs. Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-21 05:22:52 +08:00
#include <linux/workqueue.h>
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
#include <linux/netdevice.h>
#include <net/addrconf.h>
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
[PATCH] IB: move include files to include/rdma Move the InfiniBand headers from drivers/infiniband/include to include/rdma. This allows InfiniBand-using code to live elsewhere, and lets us remove the ugly EXTRA_CFLAGS include path from the InfiniBand Makefiles. Signed-off-by: Roland Dreier <rolandd@cisco.com>
2005-08-26 04:40:04 +08:00
#include <rdma/ib_cache.h>
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
#include "core_priv.h"
struct ib_pkey_cache {
int table_len;
u16 table[0];
};
struct ib_update_work {
struct work_struct work;
struct ib_device *device;
u8 port_num;
IB/core: Enforce PKey security on QPs Add new LSM hooks to allocate and free security contexts and check for permission to access a PKey. Allocate and free a security context when creating and destroying a QP. This context is used for controlling access to PKeys. When a request is made to modify a QP that changes the port, PKey index, or alternate path, check that the QP has permission for the PKey in the PKey table index on the subnet prefix of the port. If the QP is shared make sure all handles to the QP also have access. Store which port and PKey index a QP is using. After the reset to init transition the user can modify the port, PKey index and alternate path independently. So port and PKey settings changes can be a merge of the previous settings and the new ones. In order to maintain access control if there are PKey table or subnet prefix change keep a list of all QPs are using each PKey index on each port. If a change occurs all QPs using that device and port must have access enforced for the new cache settings. These changes add a transaction to the QP modify process. Association with the old port and PKey index must be maintained if the modify fails, and must be removed if it succeeds. Association with the new port and PKey index must be established prior to the modify and removed if the modify fails. 1. When a QP is modified to a particular Port, PKey index or alternate path insert that QP into the appropriate lists. 2. Check permission to access the new settings. 3. If step 2 grants access attempt to modify the QP. 4a. If steps 2 and 3 succeed remove any prior associations. 4b. If ether fails remove the new setting associations. If a PKey table or subnet prefix changes walk the list of QPs and check that they have permission. If not send the QP to the error state and raise a fatal error event. If it's a shared QP make sure all the QPs that share the real_qp have permission as well. If the QP that owns a security structure is denied access the security structure is marked as such and the QP is added to an error_list. Once the moving the QP to error is complete the security structure mark is cleared. Maintaining the lists correctly turns QP destroy into a transaction. The hardware driver for the device frees the ib_qp structure, so while the destroy is in progress the ib_qp pointer in the ib_qp_security struct is undefined. When the destroy process begins the ib_qp_security structure is marked as destroying. This prevents any action from being taken on the QP pointer. After the QP is destroyed successfully it could still listed on an error_list wait for it to be processed by that flow before cleaning up the structure. If the destroy fails the QPs port and PKey settings are reinserted into the appropriate lists, the destroying flag is cleared, and access control is enforced, in case there were any cache changes during the destroy flow. To keep the security changes isolated a new file is used to hold security related functionality. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Acked-by: Doug Ledford <dledford@redhat.com> [PM: merge fixup in ib_verbs.h and uverbs_cmd.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:52 +08:00
bool enforce_security;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
};
IB/mlx4: Implement ib_device callbacks get_netdev: get the net_device on the physical port of the IB transport port. In port aggregation mode it is required to return the netdev of the active port. modify_gid: note for a change in the RoCE gid cache. Handle this by writing to the harsware GID table. It is possible that indexes in cahce and hardware tables won't match so a translation is required when modifying a QP or creating an address handle. Signed-off-by: Moni Shoua <monis@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:29 +08:00
union ib_gid zgid;
EXPORT_SYMBOL(zgid);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
enum gid_attr_find_mask {
GID_ATTR_FIND_MASK_GID = 1UL << 0,
GID_ATTR_FIND_MASK_NETDEV = 1UL << 1,
GID_ATTR_FIND_MASK_DEFAULT = 1UL << 2,
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
GID_ATTR_FIND_MASK_GID_TYPE = 1UL << 3,
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
};
enum gid_table_entry_props {
GID_TABLE_ENTRY_INVALID = 1UL << 0,
GID_TABLE_ENTRY_DEFAULT = 1UL << 1,
};
struct ib_gid_table_entry {
unsigned long props;
union ib_gid gid;
struct ib_gid_attr attr;
void *context;
};
struct ib_gid_table {
int sz;
/* In RoCE, adding a GID to the table requires:
* (a) Find if this GID is already exists.
* (b) Find a free space.
* (c) Write the new GID
*
* Delete requires different set of operations:
* (a) Find the GID
* (b) Delete it.
*
**/
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/* Any writer to data_vec must hold this lock and the write side of
* rwlock. readers must hold only rwlock. All writers must be in a
* sleepable context.
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
*/
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
struct mutex lock;
/* rwlock protects data_vec[ix]->props. */
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
rwlock_t rwlock;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
struct ib_gid_table_entry *data_vec;
};
IB/core: Refactor GID cache's ib_dispatch_event Refactor ib_dispatch_event into a new function in order to avoid duplicating code in the next patch. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:39 +08:00
static void dispatch_gid_change_event(struct ib_device *ib_dev, u8 port)
{
IB/core: Generate GID change event regardless of RoCE GID table property Due to following reasons, GID table event is generated regardless of GID table property. 1. GID table cache is maintained at ib core layer regardless of link layer. 2. GID change event has no relation with IB link layer. 3. GID change event also doesn't depend on whether HCA supports GID table or not. Fixes: f3906bd36087 ("IB/core: Refactor GID cache's ib_dispatch_event") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-25 18:40:20 +08:00
struct ib_event event;
IB/core: Refactor GID cache's ib_dispatch_event Refactor ib_dispatch_event into a new function in order to avoid duplicating code in the next patch. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:39 +08:00
IB/core: Generate GID change event regardless of RoCE GID table property Due to following reasons, GID table event is generated regardless of GID table property. 1. GID table cache is maintained at ib core layer regardless of link layer. 2. GID change event has no relation with IB link layer. 3. GID change event also doesn't depend on whether HCA supports GID table or not. Fixes: f3906bd36087 ("IB/core: Refactor GID cache's ib_dispatch_event") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-25 18:40:20 +08:00
event.device = ib_dev;
event.element.port_num = port;
event.event = IB_EVENT_GID_CHANGE;
IB/core: Refactor GID cache's ib_dispatch_event Refactor ib_dispatch_event into a new function in order to avoid duplicating code in the next patch. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:39 +08:00
IB/core: Generate GID change event regardless of RoCE GID table property Due to following reasons, GID table event is generated regardless of GID table property. 1. GID table cache is maintained at ib core layer regardless of link layer. 2. GID change event has no relation with IB link layer. 3. GID change event also doesn't depend on whether HCA supports GID table or not. Fixes: f3906bd36087 ("IB/core: Refactor GID cache's ib_dispatch_event") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-25 18:40:20 +08:00
ib_dispatch_event(&event);
IB/core: Refactor GID cache's ib_dispatch_event Refactor ib_dispatch_event into a new function in order to avoid duplicating code in the next patch. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:39 +08:00
}
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
static const char * const gid_type_str[] = {
[IB_GID_TYPE_IB] = "IB/RoCE v1",
IB/core: Add ROCE_UDP_ENCAP (RoCE V2) type Adding RoCE v2 GID type and port type. Vendors which support this type will get their GID table populated with RoCE v2 GIDs automatically. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:50 +08:00
[IB_GID_TYPE_ROCE_UDP_ENCAP] = "RoCE v2",
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
};
const char *ib_cache_gid_type_str(enum ib_gid_type gid_type)
{
if (gid_type < ARRAY_SIZE(gid_type_str) && gid_type_str[gid_type])
return gid_type_str[gid_type];
return "Invalid GID type";
}
EXPORT_SYMBOL(ib_cache_gid_type_str);
IB/core: Reduce the places that use zgid Instead of open coding memcmp() to check whether a given GID is zero or not, use a helper function to do so, and replace instances of memcpy(z,&zgid) with memset. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:45 +08:00
/** rdma_is_zero_gid - Check if given GID is zero or not.
* @gid: GID to check
* Returns true if given GID is zero, returns false otherwise.
*/
bool rdma_is_zero_gid(const union ib_gid *gid)
{
return !memcmp(gid, &zgid, sizeof(*gid));
}
EXPORT_SYMBOL(rdma_is_zero_gid);
IB/cma: Add configfs for rdma_cm Users would like to control the behaviour of rdma_cm. For example, old applications which don't set the required RoCE gid type could be executed on RoCE V2 network types. In order to support this configuration, we implement a configfs for rdma_cm. In order to use the configfs, one needs to mount it and mkdir <IB device name> inside rdma_cm directory. The patch adds support for a single configuration file, default_roce_mode. The mode can either be "IB/RoCE v1" or "RoCE v2". Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:55 +08:00
int ib_cache_gid_parse_type_str(const char *buf)
{
unsigned int i;
size_t len;
int err = -EINVAL;
len = strlen(buf);
if (len == 0)
return -EINVAL;
if (buf[len - 1] == '\n')
len--;
for (i = 0; i < ARRAY_SIZE(gid_type_str); ++i)
if (gid_type_str[i] && !strncmp(buf, gid_type_str[i], len) &&
len == strlen(gid_type_str[i])) {
err = i;
break;
}
return err;
}
EXPORT_SYMBOL(ib_cache_gid_parse_type_str);
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
static struct ib_gid_table *rdma_gid_table(struct ib_device *device, u8 port)
{
return device->cache.ports[port - rdma_start_port(device)].gid;
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
static void del_roce_gid(struct ib_device *device, u8 port_num,
struct ib_gid_table *table, int ix)
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
{
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
pr_debug("%s device=%s port=%d index=%d gid %pI6\n", __func__,
device->name, port_num, ix,
table->data_vec[ix].gid.raw);
if (rdma_cap_roce_gid_table(device, port_num))
RDMA: Use ib_gid_attr during GID modification Now that ib_gid_attr contains device, port and index, simplify the provider APIs add_gid() and del_gid() to use device, port and index fields from the ib_gid_attr attributes structure. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:24 +08:00
device->del_gid(&table->data_vec[ix].attr,
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
&table->data_vec[ix].context);
dev_put(table->data_vec[ix].attr.ndev);
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
static int add_roce_gid(struct ib_gid_table *table,
const union ib_gid *gid,
const struct ib_gid_attr *attr)
{
struct ib_gid_table_entry *entry;
int ix = attr->index;
int ret = 0;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
if (!attr->ndev) {
pr_err("%s NULL netdev device=%s port=%d index=%d\n",
__func__, attr->device->name, attr->port_num,
attr->index);
return -EINVAL;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
entry = &table->data_vec[ix];
if ((entry->props & GID_TABLE_ENTRY_INVALID) == 0) {
WARN(1, "GID table corruption device=%s port=%d index=%d\n",
attr->device->name, attr->port_num,
attr->index);
return -EINVAL;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Fix removal of default GID cache entry When deleting a default GID from the cache, its gid_type field is set to 0. This could set the gid_type to RoCE v1 for a RoCE v2 default GID, essentially making it inaccessible to future modifications, since it is no longer found by find_gid(). This fix preserves the gid_type value for default gids during cache operations. Fixes: b39ffa1df505 ('IB/core: Add gid_type to gid attribute') Signed-off-by: Aviv Heller <avivh@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-06-04 20:15:21 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
if (rdma_cap_roce_gid_table(attr->device, attr->port_num)) {
RDMA: Use ib_gid_attr during GID modification Now that ib_gid_attr contains device, port and index, simplify the provider APIs add_gid() and del_gid() to use device, port and index fields from the ib_gid_attr attributes structure. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:24 +08:00
ret = attr->device->add_gid(gid, attr, &entry->context);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
if (ret) {
pr_err("%s GID add failed device=%s port=%d index=%d\n",
__func__, attr->device->name, attr->port_num,
attr->index);
goto add_err;
}
IB/core: Fix removal of default GID cache entry When deleting a default GID from the cache, its gid_type field is set to 0. This could set the gid_type to RoCE v1 for a RoCE v2 default GID, essentially making it inaccessible to future modifications, since it is no longer found by find_gid(). This fix preserves the gid_type value for default gids during cache operations. Fixes: b39ffa1df505 ('IB/core: Add gid_type to gid attribute') Signed-off-by: Aviv Heller <avivh@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-06-04 20:15:21 +08:00
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
dev_hold(attr->ndev);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
add_err:
if (!ret)
pr_debug("%s device=%s port=%d index=%d gid %pI6\n", __func__,
attr->device->name, attr->port_num, ix, gid->raw);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return ret;
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/**
* add_modify_gid - Add or modify GID table entry
*
* @table: GID table in which GID to be added or modified
* @gid: GID content
* @attr: Attributes of the GID
*
* Returns 0 on success or appropriate error code. It accepts zero
* GID addition for non RoCE ports for HCA's who report them as valid
* GID. However such zero GIDs are not added to the cache.
*/
static int add_modify_gid(struct ib_gid_table *table,
const union ib_gid *gid,
const struct ib_gid_attr *attr)
{
int ret;
if (rdma_protocol_roce(attr->device, attr->port_num)) {
ret = add_roce_gid(table, gid, attr);
if (ret)
return ret;
} else {
/*
* Some HCA's report multiple GID entries with only one
* valid GID, but remaining as zero GID.
* So ignore such behavior for IB link layer and don't
* fail the call, but don't add such entry to GID cache.
*/
IB/core: Reduce the places that use zgid Instead of open coding memcmp() to check whether a given GID is zero or not, use a helper function to do so, and replace instances of memcpy(z,&zgid) with memset. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:45 +08:00
if (rdma_is_zero_gid(gid))
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
return 0;
}
lockdep_assert_held(&table->lock);
memcpy(&table->data_vec[attr->index].gid, gid, sizeof(*gid));
memcpy(&table->data_vec[attr->index].attr, attr, sizeof(*attr));
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
write_lock_irq(&table->rwlock);
table->data_vec[attr->index].props &= ~GID_TABLE_ENTRY_INVALID;
write_unlock_irq(&table->rwlock);
return 0;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/**
* del_gid - Delete GID table entry
*
* @ib_dev: IB device whose GID entry to be deleted
* @port: Port number of the IB device
* @table: GID table of the IB device for a port
* @ix: GID entry index to delete
*
*/
static void del_gid(struct ib_device *ib_dev, u8 port,
struct ib_gid_table *table, int ix)
{
lockdep_assert_held(&table->lock);
write_lock_irq(&table->rwlock);
table->data_vec[ix].props |= GID_TABLE_ENTRY_INVALID;
write_unlock_irq(&table->rwlock);
if (rdma_protocol_roce(ib_dev, port))
del_roce_gid(ib_dev, port, table, ix);
IB/core: Reduce the places that use zgid Instead of open coding memcmp() to check whether a given GID is zero or not, use a helper function to do so, and replace instances of memcpy(z,&zgid) with memset. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:45 +08:00
memset(&table->data_vec[ix].gid, 0, sizeof(table->data_vec[ix].gid));
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
memset(&table->data_vec[ix].attr, 0, sizeof(table->data_vec[ix].attr));
table->data_vec[ix].context = NULL;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/* rwlock should be read locked, or lock should be held */
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
static int find_gid(struct ib_gid_table *table, const union ib_gid *gid,
const struct ib_gid_attr *val, bool default_gid,
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
unsigned long mask, int *pempty)
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
{
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
int i = 0;
int found = -1;
int empty = pempty ? -1 : 0;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
while (i < table->sz && (found < 0 || empty < 0)) {
struct ib_gid_table_entry *data = &table->data_vec[i];
struct ib_gid_attr *attr = &data->attr;
int curr_index = i;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
i++;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/* find_gid() is used during GID addition where it is expected
* to return a free entry slot which is not duplicate.
* Free entry slot is requested and returned if pempty is set,
* so lookup free slot only if requested.
*/
if (pempty && empty < 0) {
IB/core: Don't allow default GID addition at non reseved slots Default GIDs are marked reserved at the start of the GID table at index 0 and 1 by gid_table_reserve_default(). Currently when default GID is requested, it can still allocates an empty slot which was not marked as RESERVED for default GID, which is incorrect. At least in current code flow of roce_gid_mgmt.c, in theory we can still request to allocate more than one/two default GIDs depending on how upper devices are setup. Therefore, it is better for cache layer to only allow our reserved slots to be used by default GID allocation requests. Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:17 +08:00
if (data->props & GID_TABLE_ENTRY_INVALID &&
(default_gid ==
!!(data->props & GID_TABLE_ENTRY_DEFAULT))) {
/*
* Found an invalid (free) entry; allocate it.
* If default GID is requested, then our
* found slot must be one of the DEFAULT
* reserved slots or we fail.
* This ensures that only DEFAULT reserved
* slots are used for default property GIDs.
*/
empty = curr_index;
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
}
}
/*
* Additionally find_gid() is used to find valid entry during
* lookup operation, where validity needs to be checked. So
* find the empty entry first to continue to search for a free
* slot and ignore its INVALID flag.
*/
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
if (data->props & GID_TABLE_ENTRY_INVALID)
continue;
if (found >= 0)
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
continue;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
if (mask & GID_ATTR_FIND_MASK_GID_TYPE &&
attr->gid_type != val->gid_type)
continue;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (mask & GID_ATTR_FIND_MASK_GID &&
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
memcmp(gid, &data->gid, sizeof(*gid)))
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
continue;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (mask & GID_ATTR_FIND_MASK_NETDEV &&
attr->ndev != val->ndev)
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
continue;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (mask & GID_ATTR_FIND_MASK_DEFAULT &&
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
!!(data->props & GID_TABLE_ENTRY_DEFAULT) !=
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
default_gid)
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
continue;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
found = curr_index;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
if (pempty)
*pempty = empty;
return found;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
static void make_default_gid(struct net_device *dev, union ib_gid *gid)
{
gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
addrconf_ifid_eui48(&gid->raw[8], dev);
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
static int __ib_cache_gid_add(struct ib_device *ib_dev, u8 port,
union ib_gid *gid, struct ib_gid_attr *attr,
unsigned long mask, bool default_gid)
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
{
struct ib_gid_table *table;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
int ret = 0;
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
int empty;
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
int ix;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/* Do not allow adding zero GID in support of
* IB spec version 1.3 section 4.1.1 point (6) and
* section 12.7.10 and section 12.7.20
*/
IB/core: Reduce the places that use zgid Instead of open coding memcmp() to check whether a given GID is zero or not, use a helper function to do so, and replace instances of memcpy(z,&zgid) with memset. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:45 +08:00
if (rdma_is_zero_gid(gid))
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
return -EINVAL;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
mutex_lock(&table->lock);
ix = find_gid(table, gid, attr, default_gid, mask, &empty);
if (ix >= 0)
goto out_unlock;
if (empty < 0) {
ret = -ENOSPC;
goto out_unlock;
}
attr->device = ib_dev;
attr->index = empty;
attr->port_num = port;
ret = add_modify_gid(table, gid, attr);
if (!ret)
dispatch_gid_change_event(ib_dev, port);
out_unlock:
mutex_unlock(&table->lock);
if (ret)
pr_warn("%s: unable to add gid %pI6 error=%d\n",
__func__, gid->raw, ret);
return ret;
}
int ib_cache_gid_add(struct ib_device *ib_dev, u8 port,
union ib_gid *gid, struct ib_gid_attr *attr)
{
struct net_device *idev;
unsigned long mask;
int ret;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (ib_dev->get_netdev) {
idev = ib_dev->get_netdev(ib_dev, port);
if (idev && attr->ndev != idev) {
union ib_gid default_gid;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
/* Adding default GIDs in not permitted */
make_default_gid(idev, &default_gid);
if (!memcmp(gid, &default_gid, sizeof(*gid))) {
dev_put(idev);
return -EPERM;
}
}
if (idev)
dev_put(idev);
}
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
mask = GID_ATTR_FIND_MASK_GID |
GID_ATTR_FIND_MASK_GID_TYPE |
GID_ATTR_FIND_MASK_NETDEV;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
ret = __ib_cache_gid_add(ib_dev, port, gid, attr, mask, false);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
return ret;
}
IB/core: Fix to avoid deleting IPv6 look alike default GIDs When IPv6 link local address is removed, if it matches with the default GID, default GID(s)s gets removed which may not be a desired behavior. This behavior is introduced by refactor work in Fixes tag. When IPv6 link address is removed, removing its equivalent RoCEv2 GID which exactly matches with default RoCEv2 GID, is right thing to do. However achieving it correctly requires lot more changes, likely in roce_gid_mgmt.c and core/cache.c. This should be done as independent patch. Therefore, this patch preserves behavior of not deleteing default GIDs. This is done by providing explicit hint to consider default GID property using mask and default_gid; similar to add_gid(). Fixes: 598ff6bae68 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:18 +08:00
static int
_ib_cache_gid_del(struct ib_device *ib_dev, u8 port,
union ib_gid *gid, struct ib_gid_attr *attr,
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
unsigned long mask, bool default_gid)
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
{
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
struct ib_gid_table *table;
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
int ret = 0;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
int ix;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
mutex_lock(&table->lock);
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
ix = find_gid(table, gid, attr, default_gid, mask, NULL);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
if (ix < 0) {
ret = -EINVAL;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
goto out_unlock;
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
del_gid(ib_dev, port, table, ix);
dispatch_gid_change_event(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
out_unlock:
mutex_unlock(&table->lock);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
if (ret)
pr_debug("%s: can't delete gid %pI6 error=%d\n",
__func__, gid->raw, ret);
return ret;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Fix to avoid deleting IPv6 look alike default GIDs When IPv6 link local address is removed, if it matches with the default GID, default GID(s)s gets removed which may not be a desired behavior. This behavior is introduced by refactor work in Fixes tag. When IPv6 link address is removed, removing its equivalent RoCEv2 GID which exactly matches with default RoCEv2 GID, is right thing to do. However achieving it correctly requires lot more changes, likely in roce_gid_mgmt.c and core/cache.c. This should be done as independent patch. Therefore, this patch preserves behavior of not deleteing default GIDs. This is done by providing explicit hint to consider default GID property using mask and default_gid; similar to add_gid(). Fixes: 598ff6bae68 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:18 +08:00
int ib_cache_gid_del(struct ib_device *ib_dev, u8 port,
union ib_gid *gid, struct ib_gid_attr *attr)
{
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
unsigned long mask = GID_ATTR_FIND_MASK_GID |
GID_ATTR_FIND_MASK_GID_TYPE |
GID_ATTR_FIND_MASK_DEFAULT |
GID_ATTR_FIND_MASK_NETDEV;
return _ib_cache_gid_del(ib_dev, port, gid, attr, mask, false);
IB/core: Fix to avoid deleting IPv6 look alike default GIDs When IPv6 link local address is removed, if it matches with the default GID, default GID(s)s gets removed which may not be a desired behavior. This behavior is introduced by refactor work in Fixes tag. When IPv6 link address is removed, removing its equivalent RoCEv2 GID which exactly matches with default RoCEv2 GID, is right thing to do. However achieving it correctly requires lot more changes, likely in roce_gid_mgmt.c and core/cache.c. This should be done as independent patch. Therefore, this patch preserves behavior of not deleteing default GIDs. This is done by providing explicit hint to consider default GID property using mask and default_gid; similar to add_gid(). Fixes: 598ff6bae68 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:18 +08:00
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
int ib_cache_gid_del_all_netdev_gids(struct ib_device *ib_dev, u8 port,
struct net_device *ndev)
{
struct ib_gid_table *table;
int ix;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
bool deleted = false;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
mutex_lock(&table->lock);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
for (ix = 0; ix < table->sz; ix++) {
if (table->data_vec[ix].attr.ndev == ndev) {
del_gid(ib_dev, port, table, ix);
deleted = true;
}
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
mutex_unlock(&table->lock);
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
if (deleted)
dispatch_gid_change_event(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return 0;
}
static int __ib_cache_gid_get(struct ib_device *ib_dev, u8 port, int index,
union ib_gid *gid, struct ib_gid_attr *attr)
{
struct ib_gid_table *table;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (index < 0 || index >= table->sz)
return -EINVAL;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
if (table->data_vec[index].props & GID_TABLE_ENTRY_INVALID)
IB/core: Fix error code for invalid GID entry When a GID entry is invalid EAGAIN is returned. This is an incorrect error code, there is nothing that will make this GID entry valid again in bounded time. Some user space tools fail incorrectly if EAGAIN is returned here, and this represents a small ABI change from earlier kernels. The first patch in the Fixes list makes entries that were valid before to become invalid, allowing this code to trigger, while the second patch in the Fixes list introduced the wrong EAGAIN. Therefore revert the return result to EINVAL which matches the historical expectations of the ibv_query_gid_type() API of the libibverbs user space library. Cc: <stable@vger.kernel.org> Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Fixes: 03db3a2d81e6 ("IB/core: Add RoCE GID table management") Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-27 19:49:16 +08:00
return -EINVAL;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
memcpy(gid, &table->data_vec[index].gid, sizeof(*gid));
if (attr) {
memcpy(attr, &table->data_vec[index].attr, sizeof(*attr));
if (attr->ndev)
dev_hold(attr->ndev);
}
return 0;
}
static int _ib_cache_gid_table_find(struct ib_device *ib_dev,
const union ib_gid *gid,
const struct ib_gid_attr *val,
unsigned long mask,
u8 *port, u16 *index)
{
struct ib_gid_table *table;
u8 p;
int local_index;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
unsigned long flags;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
for (p = 0; p < ib_dev->phys_port_cnt; p++) {
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
table = ib_dev->cache.ports[p].gid;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_lock_irqsave(&table->rwlock, flags);
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
local_index = find_gid(table, gid, val, false, mask, NULL);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (local_index >= 0) {
if (index)
*index = local_index;
if (port)
*port = p + rdma_start_port(ib_dev);
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_unlock_irqrestore(&table->rwlock, flags);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return 0;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_unlock_irqrestore(&table->rwlock, flags);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return -ENOENT;
}
static int ib_cache_gid_find(struct ib_device *ib_dev,
const union ib_gid *gid,
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
enum ib_gid_type gid_type,
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
struct net_device *ndev, u8 *port,
u16 *index)
{
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
unsigned long mask = GID_ATTR_FIND_MASK_GID |
GID_ATTR_FIND_MASK_GID_TYPE;
struct ib_gid_attr gid_attr_val = {.ndev = ndev, .gid_type = gid_type};
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (ndev)
mask |= GID_ATTR_FIND_MASK_NETDEV;
return _ib_cache_gid_table_find(ib_dev, gid, &gid_attr_val,
mask, port, index);
}
IB/core: Fix comments of GID query functions Exported symbol's comments should be with function definition and not in the header file. Therefore comments of ib_find_cached_gid() and ib_find_cached_gid_by_port() functions are moved closer to their definitions. The function name in then comment is different than the actual function name, fix it to be same as ib_cache_gid_find_by_filter(). Also current comment section of ib_find_cached_gid_by_port() contains the desciption of ib_find_cached_gid(), fix that as well. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-13 22:06:11 +08:00
/**
* ib_find_cached_gid_by_port - Returns the GID table index where a specified
* GID value occurs. It searches for the specified GID value in the local
* software cache.
* @device: The device to query.
* @gid: The GID value to search for.
* @gid_type: The GID type to search for.
* @port_num: The port number of the device where the GID value should be
* searched.
* @ndev: In RoCE, the net device of the device. Null means ignore.
* @index: The index into the cached GID table where the GID was found. This
* parameter may be NULL.
*/
IB/core: Expose and rename ib_find_cached_gid_by_port cache API Sometime consumers might want to search for a GID in a specific port. For example, when a WC arrives and we want to search the GID that matches that port - it's better to search only the relevant port. Exposing and renaming ib_cache_gid_find_by_port in order to match the naming convention of the module. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:46 +08:00
int ib_find_cached_gid_by_port(struct ib_device *ib_dev,
const union ib_gid *gid,
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
enum ib_gid_type gid_type,
IB/core: Expose and rename ib_find_cached_gid_by_port cache API Sometime consumers might want to search for a GID in a specific port. For example, when a WC arrives and we want to search the GID that matches that port - it's better to search only the relevant port. Exposing and renaming ib_cache_gid_find_by_port in order to match the naming convention of the module. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:46 +08:00
u8 port, struct net_device *ndev,
u16 *index)
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
{
int local_index;
struct ib_gid_table *table;
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
unsigned long mask = GID_ATTR_FIND_MASK_GID |
GID_ATTR_FIND_MASK_GID_TYPE;
struct ib_gid_attr val = {.ndev = ndev, .gid_type = gid_type};
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
unsigned long flags;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(ib_dev, port))
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return -ENOENT;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (ndev)
mask |= GID_ATTR_FIND_MASK_NETDEV;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_lock_irqsave(&table->rwlock, flags);
IB/core: don't search the GID table twice Previously, we've searched the GID table twice: first when we searched the table for a GID matching the proposed new one, and second when we didn't find a match, we searched again for an empty GID slot in the table. Instead, search the table once noting the first empty slot as we search for our target GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:41 +08:00
local_index = find_gid(table, gid, &val, false, mask, NULL);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (local_index >= 0) {
if (index)
*index = local_index;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_unlock_irqrestore(&table->rwlock, flags);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return 0;
}
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_unlock_irqrestore(&table->rwlock, flags);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return -ENOENT;
}
IB/core: Expose and rename ib_find_cached_gid_by_port cache API Sometime consumers might want to search for a GID in a specific port. For example, when a WC arrives and we want to search the GID that matches that port - it's better to search only the relevant port. Exposing and renaming ib_cache_gid_find_by_port in order to match the naming convention of the module. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:46 +08:00
EXPORT_SYMBOL(ib_find_cached_gid_by_port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
/**
IB/core: Fix comments of GID query functions Exported symbol's comments should be with function definition and not in the header file. Therefore comments of ib_find_cached_gid() and ib_find_cached_gid_by_port() functions are moved closer to their definitions. The function name in then comment is different than the actual function name, fix it to be same as ib_cache_gid_find_by_filter(). Also current comment section of ib_find_cached_gid_by_port() contains the desciption of ib_find_cached_gid(), fix that as well. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-13 22:06:11 +08:00
* ib_cache_gid_find_by_filter - Returns the GID table index where a specified
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
* GID value occurs
* @device: The device to query.
* @gid: The GID value to search for.
* @port_num: The port number of the device where the GID value could be
* searched.
* @filter: The filter function is executed on any matching GID in the table.
* If the filter function returns true, the corresponding index is returned,
* otherwise, we continue searching the GID table. It's guaranteed that
* while filter is executed, ndev field is valid and the structure won't
* change. filter is executed in an atomic context. filter must not be NULL.
IB/core: Fix comments of GID query functions Exported symbol's comments should be with function definition and not in the header file. Therefore comments of ib_find_cached_gid() and ib_find_cached_gid_by_port() functions are moved closer to their definitions. The function name in then comment is different than the actual function name, fix it to be same as ib_cache_gid_find_by_filter(). Also current comment section of ib_find_cached_gid_by_port() contains the desciption of ib_find_cached_gid(), fix that as well. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-13 22:06:11 +08:00
* @index: The index into the cached GID table where the GID was found. This
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
* parameter may be NULL.
*
* ib_cache_gid_find_by_filter() searches for the specified GID value
* of which the filter function returns true in the port's GID table.
* This function is only supported on RoCE ports.
*
*/
static int ib_cache_gid_find_by_filter(struct ib_device *ib_dev,
const union ib_gid *gid,
u8 port,
bool (*filter)(const union ib_gid *,
const struct ib_gid_attr *,
void *),
void *context,
u16 *index)
{
struct ib_gid_table *table;
unsigned int i;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
unsigned long flags;
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
bool found = false;
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(ib_dev, port) ||
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
!rdma_protocol_roce(ib_dev, port))
return -EPROTONOSUPPORT;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_lock_irqsave(&table->rwlock, flags);
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
for (i = 0; i < table->sz; i++) {
struct ib_gid_attr attr;
if (table->data_vec[i].props & GID_TABLE_ENTRY_INVALID)
IB/core: Refactor to avoid unnecessary check on GID lookup miss Currently on every gid entry comparison miss found variable is checked; which is not needed as those two comparison fail already indicate that GID is not found yet. So refactor to avoid such check and copy the GID index when found. Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2017-11-14 20:52:05 +08:00
continue;
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
if (memcmp(gid, &table->data_vec[i].gid, sizeof(*gid)))
IB/core: Refactor to avoid unnecessary check on GID lookup miss Currently on every gid entry comparison miss found variable is checked; which is not needed as those two comparison fail already indicate that GID is not found yet. So refactor to avoid such check and copy the GID index when found. Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2017-11-14 20:52:05 +08:00
continue;
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
memcpy(&attr, &table->data_vec[i].attr, sizeof(attr));
IB/core: Refactor to avoid unnecessary check on GID lookup miss Currently on every gid entry comparison miss found variable is checked; which is not needed as those two comparison fail already indicate that GID is not found yet. So refactor to avoid such check and copy the GID index when found. Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2017-11-14 20:52:05 +08:00
if (filter(gid, &attr, context)) {
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
found = true;
IB/core: Refactor to avoid unnecessary check on GID lookup miss Currently on every gid entry comparison miss found variable is checked; which is not needed as those two comparison fail already indicate that GID is not found yet. So refactor to avoid such check and copy the GID index when found. Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2017-11-14 20:52:05 +08:00
if (index)
*index = i;
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
break;
IB/core: Refactor to avoid unnecessary check on GID lookup miss Currently on every gid entry comparison miss found variable is checked; which is not needed as those two comparison fail already indicate that GID is not found yet. So refactor to avoid such check and copy the GID index when found. Signed-off-by: Parav Pandit <parav@mellanox.com> Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2017-11-14 20:52:05 +08:00
}
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
}
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_unlock_irqrestore(&table->rwlock, flags);
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
if (!found)
return -ENOENT;
return 0;
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
static struct ib_gid_table *alloc_gid_table(int sz)
{
struct ib_gid_table *table =
kzalloc(sizeof(struct ib_gid_table), GFP_KERNEL);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
int i;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (!table)
return NULL;
table->data_vec = kcalloc(sz, sizeof(*table->data_vec), GFP_KERNEL);
if (!table->data_vec)
goto err_free_table;
mutex_init(&table->lock);
table->sz = sz;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
rwlock_init(&table->rwlock);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
/* Mark all entries as invalid so that allocator can allocate
* one of the invalid (free) entry.
*/
for (i = 0; i < sz; i++)
table->data_vec[i].props |= GID_TABLE_ENTRY_INVALID;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return table;
err_free_table:
kfree(table);
return NULL;
}
static void release_gid_table(struct ib_gid_table *table)
{
if (table) {
kfree(table->data_vec);
kfree(table);
}
}
static void cleanup_gid_table_port(struct ib_device *ib_dev, u8 port,
struct ib_gid_table *table)
{
int i;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
bool deleted = false;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
if (!table)
return;
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
mutex_lock(&table->lock);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
for (i = 0; i < table->sz; ++i) {
IB/core: Reduce the places that use zgid Instead of open coding memcmp() to check whether a given GID is zero or not, use a helper function to do so, and replace instances of memcpy(z,&zgid) with memset. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:45 +08:00
if (!rdma_is_zero_gid(&table->data_vec[i].gid)) {
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
del_gid(ib_dev, port, table, i);
deleted = true;
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
mutex_unlock(&table->lock);
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
if (deleted)
dispatch_gid_change_event(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
void ib_cache_gid_set_default_gid(struct ib_device *ib_dev, u8 port,
struct net_device *ndev,
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
unsigned long gid_type_mask,
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
enum ib_cache_gid_default_mode mode)
{
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
union ib_gid gid = { };
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
struct ib_gid_attr gid_attr;
struct ib_gid_table *table;
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
unsigned int gid_type;
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
unsigned long mask;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(ib_dev, port);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
mask = GID_ATTR_FIND_MASK_GID_TYPE |
GID_ATTR_FIND_MASK_DEFAULT |
GID_ATTR_FIND_MASK_NETDEV;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
memset(&gid_attr, 0, sizeof(gid_attr));
gid_attr.ndev = ndev;
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
for (gid_type = 0; gid_type < IB_GID_TYPE_SIZE; ++gid_type) {
if (1UL << gid_type & ~gid_type_mask)
continue;
gid_attr.gid_type = gid_type;
if (mode == IB_CACHE_GID_DEFAULT_MODE_SET) {
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
make_default_gid(ndev, &gid);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
__ib_cache_gid_add(ib_dev, port, &gid,
&gid_attr, mask, true);
} else if (mode == IB_CACHE_GID_DEFAULT_MODE_DELETE) {
IB/core: Fix deleting default GIDs when changing mac adddress Before [1], When MAC address of the netdevice is changed, default GID is supposed to get deleted and added back which affects the node and/or port GUID in below sequence. netdevice_event() -> NETDEV_CHANGEADDR default_del_cmd() del_netdev_default_ips() bond_delete_netdev_default_gids() ib_cache_gid_set_default_gid() ib_cache_gid_del() add_cmd() [..] However, ib_cache_gid_del() was not getting invoked in non bonding scenarios because event_ndev and rdma_ndev are same. Therefore, fix such condition to ignore checking upper device when event ndev and rdma_dev are same; similar to bond_set_netdev_default_gids(). Which this fix ib_cache_gid_del() is invoked correctly; however ib_cache_gid_del() doesn't find the default GID for deletion because find_gid() was given default_gid = false with GID_ATTR_FIND_MASK_DEFAULT set. But it was getting overwritten by ib_cache_gid_set_default_gid() later on as part of add_cmd(). Therefore, mac address change used to work for default GID. With refactor series [1], this incorrect behavior is detected. Therefore, when deleting default GID, set default_gid and set MASK flag. when deleting IP based GID, clear default_gid and set MASK flag. [1] https://patchwork.kernel.org/patch/10319151/ Fixes: 238fdf48f2b5 ("IB/core: Add RoCE table bonding support") Fixes: 598ff6bae689 ("IB/core: Refactor GID modify code for RoCE") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-04-23 21:58:19 +08:00
_ib_cache_gid_del(ib_dev, port, &gid,
&gid_attr, mask, true);
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
}
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
IB/core: Make gid_table_reserve_default() return void gid_table_reserve_default() always returns zero. Make it return void and simplify error checking. rdma_port is already calculated, use that while calling gid_table_reserve_default() instead of recalculating it. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:55 +08:00
static void gid_table_reserve_default(struct ib_device *ib_dev, u8 port,
struct ib_gid_table *table)
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
{
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
unsigned int i;
unsigned long roce_gid_type_mask;
unsigned int num_default_gids;
unsigned int current_gid = 0;
roce_gid_type_mask = roce_gid_type_mask_support(ib_dev, port);
num_default_gids = hweight_long(roce_gid_type_mask);
for (i = 0; i < num_default_gids && i < table->sz; i++) {
IB/core: Make gid_table_reserve_default() return void gid_table_reserve_default() always returns zero. Make it return void and simplify error checking. rdma_port is already calculated, use that while calling gid_table_reserve_default() instead of recalculating it. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:55 +08:00
struct ib_gid_table_entry *entry = &table->data_vec[i];
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
entry->props |= GID_TABLE_ENTRY_DEFAULT;
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
current_gid = find_next_bit(&roce_gid_type_mask,
BITS_PER_LONG,
current_gid);
entry->attr.gid_type = current_gid++;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
}
IB/core: Reuse gid_table_release_one() in table allocation failure _gid_table_setup_one() only performs GID table cache memory allocation, marks entries as invalid (free) and marks the reserved entries. At this point GID table is empty and no entries are added. On dual port device if _gid_table_setup_one() fails to allocate the gid table for 2nd port, there is no need to perform cleanup_gid_table_port() to delete GID entries, as GID table is empty. Therefore make use of existing gid_table_release_one() routine which frees the GID table memory and avoid code duplication. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:56 +08:00
static void gid_table_release_one(struct ib_device *ib_dev)
{
struct ib_gid_table *table;
u8 port;
for (port = 0; port < ib_dev->phys_port_cnt; port++) {
table = ib_dev->cache.ports[port].gid;
release_gid_table(table);
ib_dev->cache.ports[port].gid = NULL;
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
static int _gid_table_setup_one(struct ib_device *ib_dev)
{
u8 port;
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
struct ib_gid_table *table;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
for (port = 0; port < ib_dev->phys_port_cnt; port++) {
u8 rdma_port = port + rdma_start_port(ib_dev);
IB/core: Make gid_table_reserve_default() return void gid_table_reserve_default() always returns zero. Make it return void and simplify error checking. rdma_port is already calculated, use that while calling gid_table_reserve_default() instead of recalculating it. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:55 +08:00
table = alloc_gid_table(
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
ib_dev->port_immutable[rdma_port].gid_tbl_len);
IB/core: Reuse gid_table_release_one() in table allocation failure _gid_table_setup_one() only performs GID table cache memory allocation, marks entries as invalid (free) and marks the reserved entries. At this point GID table is empty and no entries are added. On dual port device if _gid_table_setup_one() fails to allocate the gid table for 2nd port, there is no need to perform cleanup_gid_table_port() to delete GID entries, as GID table is empty. Therefore make use of existing gid_table_release_one() routine which frees the GID table memory and avoid code duplication. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:56 +08:00
if (!table)
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
goto rollback_table_setup;
IB/core: Make gid_table_reserve_default() return void gid_table_reserve_default() always returns zero. Make it return void and simplify error checking. rdma_port is already calculated, use that while calling gid_table_reserve_default() instead of recalculating it. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:55 +08:00
gid_table_reserve_default(ib_dev, rdma_port, table);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
ib_dev->cache.ports[port].gid = table;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
return 0;
rollback_table_setup:
IB/core: Reuse gid_table_release_one() in table allocation failure _gid_table_setup_one() only performs GID table cache memory allocation, marks entries as invalid (free) and marks the reserved entries. At this point GID table is empty and no entries are added. On dual port device if _gid_table_setup_one() fails to allocate the gid table for 2nd port, there is no need to perform cleanup_gid_table_port() to delete GID entries, as GID table is empty. Therefore make use of existing gid_table_release_one() routine which frees the GID table memory and avoid code duplication. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2018-05-02 18:12:56 +08:00
gid_table_release_one(ib_dev);
return -ENOMEM;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
static void gid_table_cleanup_one(struct ib_device *ib_dev)
{
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
struct ib_gid_table *table;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
u8 port;
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
for (port = 0; port < ib_dev->phys_port_cnt; port++) {
table = ib_dev->cache.ports[port].gid;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
cleanup_gid_table_port(ib_dev, port + rdma_start_port(ib_dev),
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
table);
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
static int gid_table_setup_one(struct ib_device *ib_dev)
{
int err;
err = _gid_table_setup_one(ib_dev);
if (err)
return err;
{net, IB}/mlx5: Manage port association for multiport RoCE When mlx5_ib_add is called determine if the mlx5 core device being added is capable of dual port RoCE operation. If it is, determine whether it is a master device or a slave device using the num_vhca_ports and affiliate_nic_vport_criteria capabilities. If the device is a slave, attempt to find a master device to affiliate it with. Devices that can be affiliated will share a system image guid. If none are found place it on a list of unaffiliated ports. If a master is found bind the port to it by configuring the port affiliation in the NIC vport context. Similarly when mlx5_ib_remove is called determine the port type. If it's a slave port, unaffiliate it from the master device, otherwise just remove it from the unaffiliated port list. The IB device is registered as a multiport device, even if a 2nd port is not available for affiliation. When the 2nd port is affiliated later the GID cache must be refreshed in order to get the default GIDs for the 2nd port in the cache. Export roce_rescan_device to provide a mechanism to refresh the cache after a new port is bound. In a multiport configuration all IB object (QP, MR, PD, etc) related commands should flow through the master mlx5_core_dev, other commands must be sent to the slave port mlx5_core_mdev, an interface is provide to get the correct mdev for non IB object commands. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-01-04 23:25:36 +08:00
rdma_roce_rescan_device(ib_dev);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return err;
}
int ib_get_cached_gid(struct ib_device *device,
u8 port_num,
int index,
IB/core: Add netdev and gid attributes paramteres to cache Adding an ability to query the IB cache by a netdev and get the attributes of a GID. These parameters are necessary in order to successfully resolve the required GID (when the netdevice is known) and get the Ethernet L2 attributes from a GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:45 +08:00
union ib_gid *gid,
struct ib_gid_attr *gid_attr)
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
{
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
int res;
unsigned long flags;
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
struct ib_gid_table *table;
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(device, port_num))
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return -EINVAL;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(device, port_num);
IB/core: Change per-entry lock in RoCE GID table to one lock Previously, IB GID cached used a lock per entry. This could result in spending a lot of CPU cycles for locking and unlocking just in order to find a GID. Changing this in favor of one lock per a GID table. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-28 22:52:40 +08:00
read_lock_irqsave(&table->rwlock, flags);
res = __ib_cache_gid_get(device, port_num, index, gid, gid_attr);
read_unlock_irqrestore(&table->rwlock, flags);
return res;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
}
EXPORT_SYMBOL(ib_get_cached_gid);
IB/core: Fix comments of GID query functions Exported symbol's comments should be with function definition and not in the header file. Therefore comments of ib_find_cached_gid() and ib_find_cached_gid_by_port() functions are moved closer to their definitions. The function name in then comment is different than the actual function name, fix it to be same as ib_cache_gid_find_by_filter(). Also current comment section of ib_find_cached_gid_by_port() contains the desciption of ib_find_cached_gid(), fix that as well. Reviewed-by: Daniel Jurgens <danielj@mellanox.com> Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-13 22:06:11 +08:00
/**
* ib_find_cached_gid - Returns the port number and GID table index where
* a specified GID value occurs.
* @device: The device to query.
* @gid: The GID value to search for.
* @gid_type: The GID type to search for.
* @ndev: In RoCE, the net device of the device. NULL means ignore.
* @port_num: The port number of the device where the GID value was found.
* @index: The index into the cached GID table where the GID was found. This
* parameter may be NULL.
*
* ib_find_cached_gid() searches for the specified GID value in
* the local software cache.
*/
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
int ib_find_cached_gid(struct ib_device *device,
const union ib_gid *gid,
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
enum ib_gid_type gid_type,
IB/core: Add netdev and gid attributes paramteres to cache Adding an ability to query the IB cache by a netdev and get the attributes of a GID. These parameters are necessary in order to successfully resolve the required GID (when the netdevice is known) and get the Ethernet L2 attributes from a GID. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:45 +08:00
struct net_device *ndev,
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
u8 *port_num,
u16 *index)
{
IB/core: Add gid_type to gid attribute In order to support multiple GID types, we need to store the gid_type with each GID. This is also aligned with the RoCE v2 annex "RoCEv2 PORT GID table entries shall have a "GID type" attribute that denotes the L3 Address type". The currently supported GID is IB_GID_TYPE_IB which is also RoCE v1 GID type. This implies that gid_type should be added to roce_gid_table meta-data. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-12-23 20:56:47 +08:00
return ib_cache_gid_find(device, gid, gid_type, ndev, port_num, index);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
EXPORT_SYMBOL(ib_find_cached_gid);
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
int ib_find_gid_by_filter(struct ib_device *device,
const union ib_gid *gid,
u8 port_num,
bool (*filter)(const union ib_gid *gid,
const struct ib_gid_attr *,
void *),
void *context, u16 *index)
{
/* Only RoCE GID table supports filter function */
IB/core: Refer to RoCE port property instead of GID table property ib_find_gid_by_filter() searches GID with filter only for RoCE link layer regardless of HCA's support for GID table. Therefore, right way to lookup is compare RoCE port property and not the GID table property. Fixes: 99b27e3b5da0 ("IB/cache: Add ib_find_gid_by_filter cache API") Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-25 18:40:21 +08:00
if (!rdma_protocol_roce(device, port_num) && filter)
IB/cache: Add ib_find_gid_by_filter cache API GID cache API users might want to search for GIDs with specific attributes rather than just specifying GID, net device and port. This is used in a later patch, where we find the sgid index by L2 Ethernet attributes. Signed-off-by: Matan Barak <matanb@mellanox.com> Reviewed-By: Devesh Sharma <devesh.sharma@avagotech.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-10-15 23:38:50 +08:00
return -EPROTONOSUPPORT;
return ib_cache_gid_find_by_filter(device, gid,
port_num, filter,
context, index);
}
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
int ib_get_cached_pkey(struct ib_device *device,
u8 port_num,
int index,
u16 *pkey)
{
struct ib_pkey_cache *cache;
unsigned long flags;
int ret = 0;
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(device, port_num))
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
return -EINVAL;
read_lock_irqsave(&device->cache.lock, flags);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
cache = device->cache.ports[port_num - rdma_start_port(device)].pkey;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
if (index < 0 || index >= cache->table_len)
ret = -EINVAL;
else
*pkey = cache->table[index];
read_unlock_irqrestore(&device->cache.lock, flags);
return ret;
}
EXPORT_SYMBOL(ib_get_cached_pkey);
IB/core: IB cache enhancements to support Infiniband security Cache the subnet prefix and add a function to access it. Enforcing security requires frequent queries of the subnet prefix and the pkeys in the pkey table. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Eli Cohen <eli@mellanox.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: James Morris <james.l.morris@oracle.com> Acked-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:51 +08:00
int ib_get_cached_subnet_prefix(struct ib_device *device,
u8 port_num,
u64 *sn_pfx)
{
unsigned long flags;
int p;
IB/core: Use rdma_is_port_valid() Use rdma_is_port_valid() which performs port validity check instead of open coding the same check. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-19 13:59:59 +08:00
if (!rdma_is_port_valid(device, port_num))
IB/core: IB cache enhancements to support Infiniband security Cache the subnet prefix and add a function to access it. Enforcing security requires frequent queries of the subnet prefix and the pkeys in the pkey table. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Eli Cohen <eli@mellanox.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: James Morris <james.l.morris@oracle.com> Acked-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:51 +08:00
return -EINVAL;
p = port_num - rdma_start_port(device);
read_lock_irqsave(&device->cache.lock, flags);
*sn_pfx = device->cache.ports[p].subnet_prefix;
read_unlock_irqrestore(&device->cache.lock, flags);
return 0;
}
EXPORT_SYMBOL(ib_get_cached_subnet_prefix);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
int ib_find_cached_pkey(struct ib_device *device,
u8 port_num,
u16 pkey,
u16 *index)
{
struct ib_pkey_cache *cache;
unsigned long flags;
int i;
int ret = -ENOENT;
IB/core: Handle table with full and partial membership for the same P_Key Extend the cached and non-cached P_Key table lookups to handle limited and full membership of the same P_Key to co-exist in the P_Key table. This is necessary for SR-IOV, to allow for some guests would to have the full membership P_Key in their virtual P_Key table, while other guests on the same physical HCA would have the limited one. To support this, we need both the limited and full membership P_Keys to be present in the master's (hypervisor physical port) P_Key table. The algorithm for handling P_Key tables which contain both the limited and the full membership versions of the same P_Key works as follows: When scanning the P_Key table for a 15-bit P_Key: A. If there is a full member version of that P_Key anywhere in the table, return its index (even if a limited-member version of the P_Key exists earlier in the table). B. If the full member version is not in the table, but the limited-member version is in the table, return the index of the limited P_Key. Signed-off-by: Liran Liss <liranl@mellanox.com> Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-08-03 16:40:38 +08:00
int partial_ix = -1;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(device, port_num))
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
return -EINVAL;
read_lock_irqsave(&device->cache.lock, flags);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
cache = device->cache.ports[port_num - rdma_start_port(device)].pkey;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
*index = -1;
for (i = 0; i < cache->table_len; ++i)
if ((cache->table[i] & 0x7fff) == (pkey & 0x7fff)) {
IB/core: Handle table with full and partial membership for the same P_Key Extend the cached and non-cached P_Key table lookups to handle limited and full membership of the same P_Key to co-exist in the P_Key table. This is necessary for SR-IOV, to allow for some guests would to have the full membership P_Key in their virtual P_Key table, while other guests on the same physical HCA would have the limited one. To support this, we need both the limited and full membership P_Keys to be present in the master's (hypervisor physical port) P_Key table. The algorithm for handling P_Key tables which contain both the limited and the full membership versions of the same P_Key works as follows: When scanning the P_Key table for a 15-bit P_Key: A. If there is a full member version of that P_Key anywhere in the table, return its index (even if a limited-member version of the P_Key exists earlier in the table). B. If the full member version is not in the table, but the limited-member version is in the table, return the index of the limited P_Key. Signed-off-by: Liran Liss <liranl@mellanox.com> Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-08-03 16:40:38 +08:00
if (cache->table[i] & 0x8000) {
*index = i;
ret = 0;
break;
} else
partial_ix = i;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
IB/core: Handle table with full and partial membership for the same P_Key Extend the cached and non-cached P_Key table lookups to handle limited and full membership of the same P_Key to co-exist in the P_Key table. This is necessary for SR-IOV, to allow for some guests would to have the full membership P_Key in their virtual P_Key table, while other guests on the same physical HCA would have the limited one. To support this, we need both the limited and full membership P_Keys to be present in the master's (hypervisor physical port) P_Key table. The algorithm for handling P_Key tables which contain both the limited and the full membership versions of the same P_Key works as follows: When scanning the P_Key table for a 15-bit P_Key: A. If there is a full member version of that P_Key anywhere in the table, return its index (even if a limited-member version of the P_Key exists earlier in the table). B. If the full member version is not in the table, but the limited-member version is in the table, return the index of the limited P_Key. Signed-off-by: Liran Liss <liranl@mellanox.com> Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-08-03 16:40:38 +08:00
if (ret && partial_ix >= 0) {
*index = partial_ix;
ret = 0;
}
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
read_unlock_irqrestore(&device->cache.lock, flags);
return ret;
}
EXPORT_SYMBOL(ib_find_cached_pkey);
IB/core: Add ib_find_exact_cached_pkey() When P_Key tables potentially contain both full and partial membership copies for the same P_Key, we need a function to find the index for an exact (16-bit) P_Key. This is necessary when the master forwards QP1 MADs sent by guests. If the guest has sent the MAD with a limited membership P_Key, we need to to forward the MAD using the same limited membership P_Key. Since the master may have both the limited and the full member P_Keys in its table, we must make sure to retrieve the limited membership P_Key in this case. Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-08-03 16:40:39 +08:00
int ib_find_exact_cached_pkey(struct ib_device *device,
u8 port_num,
u16 pkey,
u16 *index)
{
struct ib_pkey_cache *cache;
unsigned long flags;
int i;
int ret = -ENOENT;
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(device, port_num))
IB/core: Add ib_find_exact_cached_pkey() When P_Key tables potentially contain both full and partial membership copies for the same P_Key, we need a function to find the index for an exact (16-bit) P_Key. This is necessary when the master forwards QP1 MADs sent by guests. If the guest has sent the MAD with a limited membership P_Key, we need to to forward the MAD using the same limited membership P_Key. Since the master may have both the limited and the full member P_Keys in its table, we must make sure to retrieve the limited membership P_Key in this case. Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-08-03 16:40:39 +08:00
return -EINVAL;
read_lock_irqsave(&device->cache.lock, flags);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
cache = device->cache.ports[port_num - rdma_start_port(device)].pkey;
IB/core: Add ib_find_exact_cached_pkey() When P_Key tables potentially contain both full and partial membership copies for the same P_Key, we need a function to find the index for an exact (16-bit) P_Key. This is necessary when the master forwards QP1 MADs sent by guests. If the guest has sent the MAD with a limited membership P_Key, we need to to forward the MAD using the same limited membership P_Key. Since the master may have both the limited and the full member P_Keys in its table, we must make sure to retrieve the limited membership P_Key in this case. Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-08-03 16:40:39 +08:00
*index = -1;
for (i = 0; i < cache->table_len; ++i)
if (cache->table[i] == pkey) {
*index = i;
ret = 0;
break;
}
read_unlock_irqrestore(&device->cache.lock, flags);
return ret;
}
EXPORT_SYMBOL(ib_find_exact_cached_pkey);
IB: Add caching of ports' LMC Add an LMC cache to struct ib_device, and add a function ib_get_cached_lmc() to query the cache. Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
2006-06-18 11:37:34 +08:00
int ib_get_cached_lmc(struct ib_device *device,
u8 port_num,
u8 *lmc)
{
unsigned long flags;
int ret = 0;
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(device, port_num))
IB: Add caching of ports' LMC Add an LMC cache to struct ib_device, and add a function ib_get_cached_lmc() to query the cache. Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
2006-06-18 11:37:34 +08:00
return -EINVAL;
read_lock_irqsave(&device->cache.lock, flags);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
*lmc = device->cache.ports[port_num - rdma_start_port(device)].lmc;
IB: Add caching of ports' LMC Add an LMC cache to struct ib_device, and add a function ib_get_cached_lmc() to query the cache. Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
2006-06-18 11:37:34 +08:00
read_unlock_irqrestore(&device->cache.lock, flags);
return ret;
}
EXPORT_SYMBOL(ib_get_cached_lmc);
RDMA/core: export ib_get_cached_port_state Export function for rdma_cm, patch for rdma_cm to follow. Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Acked-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-02 20:19:24 +08:00
int ib_get_cached_port_state(struct ib_device *device,
u8 port_num,
enum ib_port_state *port_state)
{
unsigned long flags;
int ret = 0;
IB/core: Use rdma_is_port_valid() Use rdma_is_port_valid() which performs port validity check instead of open coding the same check. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-03-19 13:59:59 +08:00
if (!rdma_is_port_valid(device, port_num))
RDMA/core: export ib_get_cached_port_state Export function for rdma_cm, patch for rdma_cm to follow. Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Acked-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-02 20:19:24 +08:00
return -EINVAL;
read_lock_irqsave(&device->cache.lock, flags);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
*port_state = device->cache.ports[port_num
- rdma_start_port(device)].port_state;
RDMA/core: export ib_get_cached_port_state Export function for rdma_cm, patch for rdma_cm to follow. Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Acked-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-02 20:19:24 +08:00
read_unlock_irqrestore(&device->cache.lock, flags);
return ret;
}
EXPORT_SYMBOL(ib_get_cached_port_state);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
static int config_non_roce_gid_cache(struct ib_device *device,
u8 port, int gid_tbl_len)
{
struct ib_gid_attr gid_attr = {};
struct ib_gid_table *table;
union ib_gid gid;
int ret = 0;
int i;
gid_attr.device = device;
gid_attr.port_num = port;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(device, port);
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
mutex_lock(&table->lock);
for (i = 0; i < gid_tbl_len; ++i) {
if (!device->query_gid)
continue;
ret = device->query_gid(device, port, i, &gid);
if (ret) {
pr_warn("query_gid failed (%d) for %s (index %d)\n",
ret, device->name, i);
goto err;
}
gid_attr.index = i;
add_modify_gid(table, &gid, &gid_attr);
}
err:
mutex_unlock(&table->lock);
return ret;
}
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
static void ib_cache_update(struct ib_device *device,
IB/core: Enforce PKey security on QPs Add new LSM hooks to allocate and free security contexts and check for permission to access a PKey. Allocate and free a security context when creating and destroying a QP. This context is used for controlling access to PKeys. When a request is made to modify a QP that changes the port, PKey index, or alternate path, check that the QP has permission for the PKey in the PKey table index on the subnet prefix of the port. If the QP is shared make sure all handles to the QP also have access. Store which port and PKey index a QP is using. After the reset to init transition the user can modify the port, PKey index and alternate path independently. So port and PKey settings changes can be a merge of the previous settings and the new ones. In order to maintain access control if there are PKey table or subnet prefix change keep a list of all QPs are using each PKey index on each port. If a change occurs all QPs using that device and port must have access enforced for the new cache settings. These changes add a transaction to the QP modify process. Association with the old port and PKey index must be maintained if the modify fails, and must be removed if it succeeds. Association with the new port and PKey index must be established prior to the modify and removed if the modify fails. 1. When a QP is modified to a particular Port, PKey index or alternate path insert that QP into the appropriate lists. 2. Check permission to access the new settings. 3. If step 2 grants access attempt to modify the QP. 4a. If steps 2 and 3 succeed remove any prior associations. 4b. If ether fails remove the new setting associations. If a PKey table or subnet prefix changes walk the list of QPs and check that they have permission. If not send the QP to the error state and raise a fatal error event. If it's a shared QP make sure all the QPs that share the real_qp have permission as well. If the QP that owns a security structure is denied access the security structure is marked as such and the QP is added to an error_list. Once the moving the QP to error is complete the security structure mark is cleared. Maintaining the lists correctly turns QP destroy into a transaction. The hardware driver for the device frees the ib_qp structure, so while the destroy is in progress the ib_qp pointer in the ib_qp_security struct is undefined. When the destroy process begins the ib_qp_security structure is marked as destroying. This prevents any action from being taken on the QP pointer. After the QP is destroyed successfully it could still listed on an error_list wait for it to be processed by that flow before cleaning up the structure. If the destroy fails the QPs port and PKey settings are reinserted into the appropriate lists, the destroying flag is cleared, and access control is enforced, in case there were any cache changes during the destroy flow. To keep the security changes isolated a new file is used to hold security related functionality. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Acked-by: Doug Ledford <dledford@redhat.com> [PM: merge fixup in ib_verbs.h and uverbs_cmd.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:52 +08:00
u8 port,
bool enforce_security)
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
{
struct ib_port_attr *tprops = NULL;
struct ib_pkey_cache *pkey_cache = NULL, *old_pkey_cache;
int i;
int ret;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
struct ib_gid_table *table;
IB/core: Add inline function to validate port Signed-off-by: Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-26 00:41:37 +08:00
if (!rdma_is_port_valid(device, port))
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return;
IB/core: Introduce and use rdma_gid_table() There are several places a gid table is accessed. Have a helper tiny function rdma_gid_table() to avoid code duplication at such places. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-05-23 01:33:46 +08:00
table = rdma_gid_table(device, port);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
tprops = kmalloc(sizeof *tprops, GFP_KERNEL);
if (!tprops)
return;
ret = ib_query_port(device, port, tprops);
if (ret) {
IB/core: trivial prink cleanup. 1. Replaced printk with appropriate pr_warn, pr_err, pr_info. 2. Removed unnecessary prints around memory allocation failure which are not required, as reported by the checkpatch script. Signed-off-by: Parav Pandit <pandit.parav@gmail.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Reviewed-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-03-02 03:20:29 +08:00
pr_warn("ib_query_port failed (%d) for %s\n",
ret, device->name);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
goto err;
}
IB/core: Refactor GID modify code for RoCE Code is refactored to prepare separate functions for RoCE which can do more complex operations related to reference counting, while still maintainining code readability. This includes (a) Simplification to not perform netdevice checks and modifications for IB link layer. (b) Do not add RoCE GID entry which has NULL netdevice; instead return an error. (c) If GID addition fails at provider level add_gid(), do not add the entry in the cache and keep the entry marked as INVALID. (d) Simplify and reuse the ib_cache_gid_add()/del() routines so that they can be used even for modifying default GIDs. This avoid some code duplication in modifying default GIDs. (e) find_gid() routine refers to the data entry flags to qualify a GID as valid or invalid GID rather than depending on attributes and zeroness of the GID content. (f) gid_table_reserve_default() sets the GID default attribute at beginning while setting up the GID table. There is no need to use default_gid flag in low level functions such as write_gid(), add_gid(), del_gid(), as they never need to update the DEFAULT property of the GID entry while during GID table update. As as result of this refactor, reserved GID 0:0:0:0:0:0:0:0 is no longer searchable as described below. A unicast GID entry of 0:0:0:0:0:0:0:0 is Reserved GID as per the IB spec version 1.3 section 4.1.1, point (6) whose snippet is below. "The unicast GID address 0:0:0:0:0:0:0:0 is reserved - referred to as the Reserved GID. It shall never be assigned to any endport. It shall not be used as a destination address or in a global routing header (GRH)." GID table cache now only stores valid GID entries. Before this patch, Reserved GID 0:0:0:0:0:0:0:0 was searchable in the GID table using ib_find_cached_gid_by_port() and other similar find routines. Zero GID is no longer searchable as it shall not to be present in GRH or path recored entry as described in IB spec version 1.3 section 4.1.1, point (6), section 12.7.10 and section 12.7.20. ib_cache_update() is simplified to check link layer once, use unified locking scheme for all link layers, removed temporary gid table allocation/free logic. Additionally, (a) Expand ib_gid_attr to store port and index so that GID query routines can get port and index information from the attribute structure. (b) Expand ib_gid_attr to store device as well so that in future code when GID reference counting is done, device is used to reach back to the GID table entry. Signed-off-by: Parav Pandit <parav@mellanox.com> Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2018-04-01 20:08:21 +08:00
if (!rdma_protocol_roce(device, port)) {
ret = config_non_roce_gid_cache(device, port,
tprops->gid_tbl_len);
if (ret)
goto err;
}
treewide: Use struct_size() for kmalloc()-family One of the more common cases of allocation size calculations is finding the size of a structure that has a zero-sized array at the end, along with memory for some number of elements for that array. For example: struct foo { int stuff; void *entry[]; }; instance = kmalloc(sizeof(struct foo) + sizeof(void *) * count, GFP_KERNEL); Instead of leaving these open-coded and prone to type mistakes, we can now use the new struct_size() helper: instance = kmalloc(struct_size(instance, entry, count), GFP_KERNEL); This patch makes the changes for kmalloc()-family (and kvmalloc()-family) uses. It was done via automatic conversion with manual review for the "CHECKME" non-standard cases noted below, using the following Coccinelle script: // pkey_cache = kmalloc(sizeof *pkey_cache + tprops->pkey_tbl_len * // sizeof *pkey_cache->table, GFP_KERNEL); @@ identifier alloc =~ "kmalloc|kzalloc|kvmalloc|kvzalloc"; expression GFP; identifier VAR, ELEMENT; expression COUNT; @@ - alloc(sizeof(*VAR) + COUNT * sizeof(*VAR->ELEMENT), GFP) + alloc(struct_size(VAR, ELEMENT, COUNT), GFP) // mr = kzalloc(sizeof(*mr) + m * sizeof(mr->map[0]), GFP_KERNEL); @@ identifier alloc =~ "kmalloc|kzalloc|kvmalloc|kvzalloc"; expression GFP; identifier VAR, ELEMENT; expression COUNT; @@ - alloc(sizeof(*VAR) + COUNT * sizeof(VAR->ELEMENT[0]), GFP) + alloc(struct_size(VAR, ELEMENT, COUNT), GFP) // Same pattern, but can't trivially locate the trailing element name, // or variable name. @@ identifier alloc =~ "kmalloc|kzalloc|kvmalloc|kvzalloc"; expression GFP; expression SOMETHING, COUNT, ELEMENT; @@ - alloc(sizeof(SOMETHING) + COUNT * sizeof(ELEMENT), GFP) + alloc(CHECKME_struct_size(&SOMETHING, ELEMENT, COUNT), GFP) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-05-09 04:45:50 +08:00
pkey_cache = kmalloc(struct_size(pkey_cache, table,
tprops->pkey_tbl_len),
GFP_KERNEL);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
if (!pkey_cache)
goto err;
pkey_cache->table_len = tprops->pkey_tbl_len;
for (i = 0; i < pkey_cache->table_len; ++i) {
ret = ib_query_pkey(device, port, i, pkey_cache->table + i);
if (ret) {
IB/core: trivial prink cleanup. 1. Replaced printk with appropriate pr_warn, pr_err, pr_info. 2. Removed unnecessary prints around memory allocation failure which are not required, as reported by the checkpatch script. Signed-off-by: Parav Pandit <pandit.parav@gmail.com> Reviewed-by: Haggai Eran <haggaie@mellanox.com> Reviewed-by: Sagi Grimberg <sagig@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-03-02 03:20:29 +08:00
pr_warn("ib_query_pkey failed (%d) for %s (index %d)\n",
ret, device->name, i);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
goto err;
}
}
write_lock_irq(&device->cache.lock);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
old_pkey_cache = device->cache.ports[port -
rdma_start_port(device)].pkey;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
device->cache.ports[port - rdma_start_port(device)].pkey = pkey_cache;
device->cache.ports[port - rdma_start_port(device)].lmc = tprops->lmc;
device->cache.ports[port - rdma_start_port(device)].port_state =
RDMA/core: add port state cache We need a port state cache in ib_core, later we will use in rdma_cm. Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Acked-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-02 20:17:36 +08:00
tprops->state;
IB: Add caching of ports' LMC Add an LMC cache to struct ib_device, and add a function ib_get_cached_lmc() to query the cache. Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Sean Hefty <sean.hefty@intel.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
2006-06-18 11:37:34 +08:00
IB/core: IB cache enhancements to support Infiniband security Cache the subnet prefix and add a function to access it. Enforcing security requires frequent queries of the subnet prefix and the pkeys in the pkey table. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Eli Cohen <eli@mellanox.com> Reviewed-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: James Morris <james.l.morris@oracle.com> Acked-by: Doug Ledford <dledford@redhat.com> Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:51 +08:00
device->cache.ports[port - rdma_start_port(device)].subnet_prefix =
tprops->subnet_prefix;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
write_unlock_irq(&device->cache.lock);
IB/core: Enforce PKey security on QPs Add new LSM hooks to allocate and free security contexts and check for permission to access a PKey. Allocate and free a security context when creating and destroying a QP. This context is used for controlling access to PKeys. When a request is made to modify a QP that changes the port, PKey index, or alternate path, check that the QP has permission for the PKey in the PKey table index on the subnet prefix of the port. If the QP is shared make sure all handles to the QP also have access. Store which port and PKey index a QP is using. After the reset to init transition the user can modify the port, PKey index and alternate path independently. So port and PKey settings changes can be a merge of the previous settings and the new ones. In order to maintain access control if there are PKey table or subnet prefix change keep a list of all QPs are using each PKey index on each port. If a change occurs all QPs using that device and port must have access enforced for the new cache settings. These changes add a transaction to the QP modify process. Association with the old port and PKey index must be maintained if the modify fails, and must be removed if it succeeds. Association with the new port and PKey index must be established prior to the modify and removed if the modify fails. 1. When a QP is modified to a particular Port, PKey index or alternate path insert that QP into the appropriate lists. 2. Check permission to access the new settings. 3. If step 2 grants access attempt to modify the QP. 4a. If steps 2 and 3 succeed remove any prior associations. 4b. If ether fails remove the new setting associations. If a PKey table or subnet prefix changes walk the list of QPs and check that they have permission. If not send the QP to the error state and raise a fatal error event. If it's a shared QP make sure all the QPs that share the real_qp have permission as well. If the QP that owns a security structure is denied access the security structure is marked as such and the QP is added to an error_list. Once the moving the QP to error is complete the security structure mark is cleared. Maintaining the lists correctly turns QP destroy into a transaction. The hardware driver for the device frees the ib_qp structure, so while the destroy is in progress the ib_qp pointer in the ib_qp_security struct is undefined. When the destroy process begins the ib_qp_security structure is marked as destroying. This prevents any action from being taken on the QP pointer. After the QP is destroyed successfully it could still listed on an error_list wait for it to be processed by that flow before cleaning up the structure. If the destroy fails the QPs port and PKey settings are reinserted into the appropriate lists, the destroying flag is cleared, and access control is enforced, in case there were any cache changes during the destroy flow. To keep the security changes isolated a new file is used to hold security related functionality. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Acked-by: Doug Ledford <dledford@redhat.com> [PM: merge fixup in ib_verbs.h and uverbs_cmd.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:52 +08:00
if (enforce_security)
ib_security_cache_change(device,
port,
tprops->subnet_prefix);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
kfree(old_pkey_cache);
kfree(tprops);
return;
err:
kfree(pkey_cache);
kfree(tprops);
}
WorkStruct: make allyesconfig Fix up for make allyesconfig. Signed-Off-By: David Howells <dhowells@redhat.com>
2006-11-22 22:57:56 +08:00
static void ib_cache_task(struct work_struct *_work)
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
{
WorkStruct: make allyesconfig Fix up for make allyesconfig. Signed-Off-By: David Howells <dhowells@redhat.com>
2006-11-22 22:57:56 +08:00
struct ib_update_work *work =
container_of(_work, struct ib_update_work, work);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Enforce PKey security on QPs Add new LSM hooks to allocate and free security contexts and check for permission to access a PKey. Allocate and free a security context when creating and destroying a QP. This context is used for controlling access to PKeys. When a request is made to modify a QP that changes the port, PKey index, or alternate path, check that the QP has permission for the PKey in the PKey table index on the subnet prefix of the port. If the QP is shared make sure all handles to the QP also have access. Store which port and PKey index a QP is using. After the reset to init transition the user can modify the port, PKey index and alternate path independently. So port and PKey settings changes can be a merge of the previous settings and the new ones. In order to maintain access control if there are PKey table or subnet prefix change keep a list of all QPs are using each PKey index on each port. If a change occurs all QPs using that device and port must have access enforced for the new cache settings. These changes add a transaction to the QP modify process. Association with the old port and PKey index must be maintained if the modify fails, and must be removed if it succeeds. Association with the new port and PKey index must be established prior to the modify and removed if the modify fails. 1. When a QP is modified to a particular Port, PKey index or alternate path insert that QP into the appropriate lists. 2. Check permission to access the new settings. 3. If step 2 grants access attempt to modify the QP. 4a. If steps 2 and 3 succeed remove any prior associations. 4b. If ether fails remove the new setting associations. If a PKey table or subnet prefix changes walk the list of QPs and check that they have permission. If not send the QP to the error state and raise a fatal error event. If it's a shared QP make sure all the QPs that share the real_qp have permission as well. If the QP that owns a security structure is denied access the security structure is marked as such and the QP is added to an error_list. Once the moving the QP to error is complete the security structure mark is cleared. Maintaining the lists correctly turns QP destroy into a transaction. The hardware driver for the device frees the ib_qp structure, so while the destroy is in progress the ib_qp pointer in the ib_qp_security struct is undefined. When the destroy process begins the ib_qp_security structure is marked as destroying. This prevents any action from being taken on the QP pointer. After the QP is destroyed successfully it could still listed on an error_list wait for it to be processed by that flow before cleaning up the structure. If the destroy fails the QPs port and PKey settings are reinserted into the appropriate lists, the destroying flag is cleared, and access control is enforced, in case there were any cache changes during the destroy flow. To keep the security changes isolated a new file is used to hold security related functionality. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Acked-by: Doug Ledford <dledford@redhat.com> [PM: merge fixup in ib_verbs.h and uverbs_cmd.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:52 +08:00
ib_cache_update(work->device,
work->port_num,
work->enforce_security);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
kfree(work);
}
static void ib_cache_event(struct ib_event_handler *handler,
struct ib_event *event)
{
struct ib_update_work *work;
if (event->event == IB_EVENT_PORT_ERR ||
event->event == IB_EVENT_PORT_ACTIVE ||
event->event == IB_EVENT_LID_CHANGE ||
event->event == IB_EVENT_PKEY_CHANGE ||
IB/core: Fix SM LID/LID change with client reregister set After commit 12bbb2b7be7f5564952ebe0196623e97464b8ac5, when SM LID change or LID change MAD also has a client reregistration bit set, only CLIENT_REREGISTER event is generated. As a result, the sa_query module and the cache module don't update the port information, and ULPs (e.g. IPoIB) stop working. This is the regression we observe as compared to 2.6.17. Rather than generate multiple events (which would have negative performance impact), let us simply let cache and SA query respond to reregister event in the same way as to LID and SM change events. Signed-off-by: Jack Morgenstein <jackm@mellanox.co.il> Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il> Signed-off-by: Roland Dreier <rolandd@cisco.com>
2006-08-15 22:20:50 +08:00
event->event == IB_EVENT_SM_CHANGE ||
IB/core: Add GID change event Add IB GID change event type. This is needed for IBoE when the HW driver updates the GID (e.g when new VLANs are added/deleted) table and the change should be reflected to the IB core cache. Signed-off-by: Eli Cohen <eli@mellanox.co.il> Signed-off-by: Or Gerlitz <ogerlitz@mellanox.co.il> Signed-off-by: Roland Dreier <roland@purestorage.com>
2011-06-15 22:39:29 +08:00
event->event == IB_EVENT_CLIENT_REREGISTER ||
event->event == IB_EVENT_GID_CHANGE) {
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
work = kmalloc(sizeof *work, GFP_ATOMIC);
if (work) {
WorkStruct: make allyesconfig Fix up for make allyesconfig. Signed-Off-By: David Howells <dhowells@redhat.com>
2006-11-22 22:57:56 +08:00
INIT_WORK(&work->work, ib_cache_task);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
work->device = event->device;
work->port_num = event->element.port_num;
IB/core: Enforce PKey security on QPs Add new LSM hooks to allocate and free security contexts and check for permission to access a PKey. Allocate and free a security context when creating and destroying a QP. This context is used for controlling access to PKeys. When a request is made to modify a QP that changes the port, PKey index, or alternate path, check that the QP has permission for the PKey in the PKey table index on the subnet prefix of the port. If the QP is shared make sure all handles to the QP also have access. Store which port and PKey index a QP is using. After the reset to init transition the user can modify the port, PKey index and alternate path independently. So port and PKey settings changes can be a merge of the previous settings and the new ones. In order to maintain access control if there are PKey table or subnet prefix change keep a list of all QPs are using each PKey index on each port. If a change occurs all QPs using that device and port must have access enforced for the new cache settings. These changes add a transaction to the QP modify process. Association with the old port and PKey index must be maintained if the modify fails, and must be removed if it succeeds. Association with the new port and PKey index must be established prior to the modify and removed if the modify fails. 1. When a QP is modified to a particular Port, PKey index or alternate path insert that QP into the appropriate lists. 2. Check permission to access the new settings. 3. If step 2 grants access attempt to modify the QP. 4a. If steps 2 and 3 succeed remove any prior associations. 4b. If ether fails remove the new setting associations. If a PKey table or subnet prefix changes walk the list of QPs and check that they have permission. If not send the QP to the error state and raise a fatal error event. If it's a shared QP make sure all the QPs that share the real_qp have permission as well. If the QP that owns a security structure is denied access the security structure is marked as such and the QP is added to an error_list. Once the moving the QP to error is complete the security structure mark is cleared. Maintaining the lists correctly turns QP destroy into a transaction. The hardware driver for the device frees the ib_qp structure, so while the destroy is in progress the ib_qp pointer in the ib_qp_security struct is undefined. When the destroy process begins the ib_qp_security structure is marked as destroying. This prevents any action from being taken on the QP pointer. After the QP is destroyed successfully it could still listed on an error_list wait for it to be processed by that flow before cleaning up the structure. If the destroy fails the QPs port and PKey settings are reinserted into the appropriate lists, the destroying flag is cleared, and access control is enforced, in case there were any cache changes during the destroy flow. To keep the security changes isolated a new file is used to hold security related functionality. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Acked-by: Doug Ledford <dledford@redhat.com> [PM: merge fixup in ib_verbs.h and uverbs_cmd.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:52 +08:00
if (event->event == IB_EVENT_PKEY_CHANGE ||
event->event == IB_EVENT_GID_CHANGE)
work->enforce_security = true;
else
work->enforce_security = false;
RDMA: Update workqueue usage * ib_wq is added, which is used as the common workqueue for infiniband instead of the system workqueue. All system workqueue usages including flush_scheduled_work() callers are converted to use and flush ib_wq. * cancel_delayed_work() + flush_scheduled_work() converted to cancel_delayed_work_sync(). * qib_wq is removed and ib_wq is used instead. This is to prepare for deprecation of flush_scheduled_work(). Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: Roland Dreier <rolandd@cisco.com>
2010-10-19 23:24:36 +08:00
queue_work(ib_wq, &work->work);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
}
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
int ib_cache_setup_one(struct ib_device *device)
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
{
int p;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
int err;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
rwlock_init(&device->cache.lock);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
device->cache.ports =
kzalloc(sizeof(*device->cache.ports) *
IB/core: Create common start/end port functions Previously start_port and end_port were defined in 2 places, cache.c and device.c and this prevented their use in other modules. Make these common functions, change the name to reflect the rdma name space, and update existing users. Signed-off-by: Ira Weiny <ira.weiny@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-05-14 08:02:55 +08:00
(rdma_end_port(device) - rdma_start_port(device) + 1), GFP_KERNEL);
RDMA/(core, ulp): Convert register/unregister event handler to be void The functions ib_register_event_handler() and ib_unregister_event_handler() always returned success and they can't fail. Let's convert those functions to be void, remove redundant checks and cleanup tons of goto statements. Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-08-17 20:50:36 +08:00
if (!device->cache.ports)
return -ENOMEM;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
err = gid_table_setup_one(device);
RDMA/(core, ulp): Convert register/unregister event handler to be void The functions ib_register_event_handler() and ib_unregister_event_handler() always returned success and they can't fail. Let's convert those functions to be void, remove redundant checks and cleanup tons of goto statements. Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-08-17 20:50:36 +08:00
if (err) {
kfree(device->cache.ports);
device->cache.ports = NULL;
return err;
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
IB/core: Make ib_alloc_device init the kobject This gets rid of the weird in-between state where struct ib_device was allocated but the kobject didn't work. Consequently ib_device_release is now guaranteed to be called in all situations and we needn't duplicate its kfrees on error paths. Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-08-05 05:23:34 +08:00
for (p = 0; p <= rdma_end_port(device) - rdma_start_port(device); ++p)
IB/core: Enforce PKey security on QPs Add new LSM hooks to allocate and free security contexts and check for permission to access a PKey. Allocate and free a security context when creating and destroying a QP. This context is used for controlling access to PKeys. When a request is made to modify a QP that changes the port, PKey index, or alternate path, check that the QP has permission for the PKey in the PKey table index on the subnet prefix of the port. If the QP is shared make sure all handles to the QP also have access. Store which port and PKey index a QP is using. After the reset to init transition the user can modify the port, PKey index and alternate path independently. So port and PKey settings changes can be a merge of the previous settings and the new ones. In order to maintain access control if there are PKey table or subnet prefix change keep a list of all QPs are using each PKey index on each port. If a change occurs all QPs using that device and port must have access enforced for the new cache settings. These changes add a transaction to the QP modify process. Association with the old port and PKey index must be maintained if the modify fails, and must be removed if it succeeds. Association with the new port and PKey index must be established prior to the modify and removed if the modify fails. 1. When a QP is modified to a particular Port, PKey index or alternate path insert that QP into the appropriate lists. 2. Check permission to access the new settings. 3. If step 2 grants access attempt to modify the QP. 4a. If steps 2 and 3 succeed remove any prior associations. 4b. If ether fails remove the new setting associations. If a PKey table or subnet prefix changes walk the list of QPs and check that they have permission. If not send the QP to the error state and raise a fatal error event. If it's a shared QP make sure all the QPs that share the real_qp have permission as well. If the QP that owns a security structure is denied access the security structure is marked as such and the QP is added to an error_list. Once the moving the QP to error is complete the security structure mark is cleared. Maintaining the lists correctly turns QP destroy into a transaction. The hardware driver for the device frees the ib_qp structure, so while the destroy is in progress the ib_qp pointer in the ib_qp_security struct is undefined. When the destroy process begins the ib_qp_security structure is marked as destroying. This prevents any action from being taken on the QP pointer. After the QP is destroyed successfully it could still listed on an error_list wait for it to be processed by that flow before cleaning up the structure. If the destroy fails the QPs port and PKey settings are reinserted into the appropriate lists, the destroying flag is cleared, and access control is enforced, in case there were any cache changes during the destroy flow. To keep the security changes isolated a new file is used to hold security related functionality. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Acked-by: Doug Ledford <dledford@redhat.com> [PM: merge fixup in ib_verbs.h and uverbs_cmd.c] Signed-off-by: Paul Moore <paul@paul-moore.com>
2017-05-19 20:48:52 +08:00
ib_cache_update(device, p + rdma_start_port(device), true);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
INIT_IB_EVENT_HANDLER(&device->cache.event_handler,
device, ib_cache_event);
RDMA/(core, ulp): Convert register/unregister event handler to be void The functions ib_register_event_handler() and ib_unregister_event_handler() always returned success and they can't fail. Let's convert those functions to be void, remove redundant checks and cleanup tons of goto statements. Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-08-17 20:50:36 +08:00
ib_register_event_handler(&device->cache.event_handler);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
return 0;
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
void ib_cache_release_one(struct ib_device *device)
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
{
int p;
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
/*
* The release function frees all the cache elements.
* This function should be called as part of freeing
* all the device's resources when the cache could no
* longer be accessed.
*/
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
for (p = 0; p <= rdma_end_port(device) - rdma_start_port(device); ++p)
kfree(device->cache.ports[p].pkey);
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
gid_table_release_one(device);
RDMA/core: create struct ib_port_cache As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by: Michael Wang <yun.wang@profitbricks.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2017-01-17 17:11:12 +08:00
kfree(device->cache.ports);
Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
2005-04-17 06:20:36 +08:00
}
IB/core: Add RoCE GID table management RoCE GIDs are based on IP addresses configured on Ethernet net-devices which relate to the RDMA (RoCE) device port. Currently, each of the low-level drivers that support RoCE (ocrdma, mlx4) manages its own RoCE port GID table. As there's nothing which is essentially vendor specific, we generalize that, and enhance the RDMA core GID cache to do this job. In order to populate the GID table, we listen for events: (a) netdev up/down/change_addr events - if a netdev is built onto our RoCE device, we need to add/delete its IPs. This involves adding all GIDs related to this ndev, add default GIDs, etc. (b) inet events - add new GIDs (according to the IP addresses) to the table. For programming the port RoCE GID table, providers must implement the add_gid and del_gid callbacks. RoCE GID management requires us to state the associated net_device alongside the GID. This information is necessary in order to manage the GID table. For example, when a net_device is removed, its associated GIDs need to be removed as well. RoCE mandates generating a default GID for each port, based on the related net-device's IPv6 link local. In contrast to the GID based on the regular IPv6 link-local (as we generate GID per IP address), the default GID is also available when the net device is down (in order to support loopback). Locking is done as follows: The patch modify the GID table code both for new RoCE drivers implementing the add_gid/del_gid callbacks and for current RoCE and IB drivers that do not. The flows for updating the table are different, so the locking requirements are too. While updating RoCE GID table, protection against multiple writers is achieved via mutex_lock(&table->lock). Since writing to a table requires us to find an entry (possible a free entry) in the table and then modify it, this mutex protects both the find_gid and write_gid ensuring the atomicity of the action. Each entry in the GID cache is protected by rwlock. In RoCE, writing (usually results from netdev notifier) involves invoking the vendor's add_gid and del_gid callbacks, which could sleep. Therefore, an invalid flag is added for each entry. Updates for RoCE are done via a workqueue, thus sleeping is permitted. In IB, updates are done in write_lock_irq(&device->cache.lock), thus write_gid isn't allowed to sleep and add_gid/del_gid are not called. When passing net-device into/out-of the GID cache, the device is always passed held (dev_hold). The code uses a single work item for updating all RDMA devices, following a netdev or inet notifier. The patch moves the cache from being a client (which was incorrect, as the cache is part of the IB infrastructure) to being explicitly initialized/freed when a device is registered/removed. Signed-off-by: Matan Barak <matanb@mellanox.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2015-07-30 23:33:26 +08:00
void ib_cache_cleanup_one(struct ib_device *device)
{
/* The cleanup function unregisters the event handler,
* waits for all in-progress workqueue elements and cleans
* up the GID cache. This function should be called after
* the device was removed from the devices list and all
* clients were removed, so the cache exists but is
* non-functional and shouldn't be updated anymore.
*/
ib_unregister_event_handler(&device->cache.event_handler);
flush_workqueue(ib_wq);
gid_table_cleanup_one(device);
}