Or Gerlitz says:
==============================
Mellanox driver update for net-next
Some small fixes and small enhancements from the team.
Series applies over net-next commit acb4a6b "tcp: ensure prior synack rtx behavior
with small backlogs".
==============================
Signed-off-by: David S. Miller <davem@davemloft.net>
The 'for' loop when undoing rb-tree insertions and list-adds in the
error flow in add_res_range had errors, fix them.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The procedure mlx4_update_qp leaks mailboxes in its error-flow, fix that.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Changed the receive control flow in a way that steering
rules are added only when the RSS object is already in RTR/RTS mode.
Some optimization features, which are enabled by the device firmware,
require this condition in order to be effective.
Signed-off-by: Ido Shamay <idos@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use a single threaded work queue for each device in the system instead of
using one thread for any device. This is required so we can concurrently
process system error handling for all the devices that need that.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use ioread function to read health buffer data. In addition, print the
firmware version as a string for readability and also use dev_err to have
the device string to be printed.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation to handling system errors at the mlx5_core level, change the
interface of cmd_work_handler to accept a 64 bit argument for the vector.
This allows to encode a flag that signifies when the handler is called
as a result of a driver logic that wishes to terminate commands that
the hardware may not be able to terminate. Such command completions
are detected at the handler and proper return status is encoded.
To be able to terminate page handler commands, we make sure to set
the corresponding bit in the bitmask.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Improve the messages printed by the mlx5 macros to include the device
string. In addition, prefix names used by the macros with two underscores
to avoid possible name collisions.
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
Intel Wired LAN Driver Updates 2015-10-08
This series contains updates to i40e and i40evf only (again).
Jesse fixes an issue where the driver was issuing a WARN_ON during ring
size changes because the code was cloning the rx_ring struct but not
zeroing out the pointers before allocating new memory, so simply zero
out the pointers. Also reduced the function call overhead by moving
the interrupt enable function by moving it to the header file, which it
in turn allows us to inline it. Also does a thorough job of code
cleanup to fix spaces after declarations, remove unnecessary braces
and breaks, remove another __func__ use and general code tidiness.
Mitch adds mover verbose error messages when the number of supported VFs
is reported in driver init and it different from the number reported in
config space. Updated the VF driver to now detect a reset with the
VF_ARQLEN register since the enable bit is cleared when the VF is reset
and it stays cleared until the VF driver processes the reset and
re-enables the admin queue which is more reliable than using the
VFGEN_RSTAT as previously.
Neerav adds parsing for CEE DCBx TLVs from the LLDP MIB since there is
a need to get the CEE DesiredCfg Tx by firmware and DCB configuration
Rx from peer for debug and other application purposes.
Carolyn fixes a problem where the PF's Flow Director filter table would
have an entry that the hardware was unable to add, when this occurs an
invalid entry gets replayed and a valid one is lost.
Matt fixes an issue where multiple link up messages can be logged
resulting from admin queue link status timing when link properties are
changed.
Shannon adds the ability to control the period link polling through
ethtool to be able to switch it off and on for debugging link issues.
Serey explicitly assigns the enum index for each VSI type so that the PF
and VF always reference to the same VSI type event if the enum lists
are different.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
To help users and developers know what compile options
and hardware features are enabled at compile time, print
VxLAN is available.
Change-ID: I3162f3b7678dc725a597f964217920eb218b480b
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The VF really doesn't care about the QOS handle but it will in the
future. Since the VF only uses TC0, send it that handle. On the VF
side, save the handle and use it to populate the QOS params when we call
into the client interface.
Change-ID: I76f41b070baeaa09b19383e9168bc677837e0761
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Use the capabilities passed to us by the PF driver to control VF driver
behavior. In the process, clean up the VLAN add/remove code so it's not
a horrible morass of ifdefs.
Change-ID: I1050eaf12b658a26fea6813047c9964163c70a73
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
I found a code indent that was avoidable because a whole function is inside
an if block, reverse the if and move the code back a tab.
Change-ID: I9989c8750ee61678fbf96a3b0fd7bf7cc7ef300a
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add missings spaces after declarations, remove another __func__ use,
remove uncessary braces, remove unneeded breaks, and useless returns,
and generally fix up some code.
Change-ID: Ie715d6b64976c50e1c21531685fe0a2bd38c4244
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Using VFGEN_RSTAT to detect a VF reset is an endeavor that is fraught
with peril. It's entirely too easy to miss a reset because none of the
bits are sticky. By the time the VF driver reads the register, the reset
may have been processed and cleaned up by the PF driver, leaving the
register in the same state that it was before the reset.
Instead, detect a reset with the VF_ARQLEN register. When the VF is
reset, the enable bit in this register is cleared, and it stays cleared
until the VF driver processes the reset and re-enables the admin queue.
Because we now deal with multiple registers in the reset and watchdog
tasks, rename the rstat_val variable to reg_val.
Change-ID: Id1df17045c0992e607da0162d31807f7fc20d199
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Changes parsing of AQ command Get CEE DCBX OPER CFG (0x0A07). Change is
required because FW creates the oper_prio_tc nibbles reversed from those
in the CEE Priority Group sub-TLV.
Change-ID: I7d9d8641bb430d30e286fc3fac909866ef8a0de8
Signed-off-by: Greg Bowers <gregory.j.bowers@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Ran into an issue where PF's VSI type list was different from VF's,
which was resulted in different enum index. The VSI type list can
be different depending on what build flag is used for PF and VF.
The change is to explicitly assign enum index for each VSI type
so that PF and VF always reference to the same VSI type event if the
enum lists are different.
Change-ID: I8c0e5fdb515f324f7964df863a458073cf467e57
Signed-off-by: Serey Kong <serey.kong@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
There's been some need for controlling the periodic link polling for
debugging link issues. This patch enables switching it off and on
through an ethtool private flag. The link poll remains on by default,
but can be turned off with
ethtool --set-priv-flags p261p1 LinkPolling off
and later turned back on with
ethtool --set-priv-flags p261p1 LinkPolling on
To check the current status, use
ethtool --show-priv-flags p261p1
Change-ID: I32e4ab654ff3eec90a06cf144899971b82d71c40
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch addresses an issue where multiple link up messages can be logged
resulting from aq link status timing when link properties are changed (fc,
speed, etc.); solved by using a single function to handle status printing
and adding a mechanism to track whether link state (up or down) has
actually changed.
Change-ID: Ied6ed6e49dc397c77d992adc0bc9ed3767152b9d
Signed-off-by: Matt Jared <matthew.a.jared@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch fixes a problem where the PF's fdir filter table would have an
entry that the hw was unable to add. This notification happens in the hot
path, so instead of trying to fix it then, we note the location in the
failure case and delete it during regular fdir subtask callback. Without
this patch, a case can occur where an invalid entry gets replayed and a
valid one is not.
Change-ID: I67831c183b5d0309876de807cc434809b74c9cb7
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds capability to query and store the CEE DCBX DesiredCfg
and RemoteCfg data from the LLDP MIB.
Added new member "desired_dcbx_config" in the i40e_hw data structure
to hold CEE only DesiredCfg data.
Change-ID: I19c550369594384eaff4cc63e690ca740231195d
Signed-off-by: Neerav Parikh <neerav.parikh@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds parsing for CEE DCBX TLVs from the LLDP MIB.
While the driver gets the DCB CEE operational configuration from Firmware
using the "Get CEE DCBX Oper Config" AQ command there is a need to get
the CEE DesiredCfg Tx by firmware and DCB configuration Rx from peer; for
debug and other application purposes.
Change-ID: I9140edf1a25a2852c7eff805d81e5eff6266178d
Signed-off-by: Neerav Parikh <neerav.parikh@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Under certain circumstances, the device may not have enough resources to
enable all of the VFs that it advertises in config space. Although the
number of supported VFs is reported upon driver init, it is not obvious
when this is different from the number reported in config space. To
eliminate this confusion, add an error message explaining the problem.
Additionally, move the 'Allocating VFs' message down below the error
checks so as to prevent further confusion.
Change-ID: I45b7efca53a7aebf7777be33a8bc9d615ae48ea1
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The interrupt enable function can be inlined by moving it to the header
file, which decreases the function call overhead for a frequently called
function.
Change-ID: I3214cc99593725768642680e7b8ce7e9bba7e44d
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The driver was issuing a WARN_ON during ring size changes
because the code was cloning the rx_ring struct but
not zeroing out the pointers before allocating new memory.
Zero out the pointers in the cloned copy before allocating
new memory for them. In this case the code was correctly
avoiding memory leaks but still triggering the warning.
Change-ID: I186dd493948e9b7254ab0593d4aad8b68808918d
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Daniel Borkmann says:
====================
BPF/random32 updates
BPF update to split the prandom state apart, and to move the
*once helpers to the core. For details, please see individual
patches. Given the changes and since it's in the tree for
quite some time, net-next is a better choice in our opinion.
v1 -> v2:
- Make DO_ONCE() type-safe, remove the kvec helper. Credits
go to Alexei Starovoitov for the __VA_ARGS__ hint, thanks!
- Add a comment to the DO_ONCE() helper as suggested by Alexei.
- Rework prandom_init_once() helper to the new API.
- Keep Alexei's Acked-by on the last patch.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
While recently arguing on a seccomp discussion that raw prandom_u32()
access shouldn't be exposed to unpriviledged user space, I forgot the
fact that SKF_AD_RANDOM extension actually already does it for some time
in cBPF via commit 4cd3675ebf ("filter: added BPF random opcode").
Since prandom_u32() is being used in a lot of critical networking code,
lets be more conservative and split their states. Furthermore, consolidate
eBPF and cBPF prandom handlers to use the new internal PRNG. For eBPF,
bpf_get_prandom_u32() was only accessible for priviledged users, but
should that change one day, we also don't want to leak raw sequences
through things like eBPF maps.
One thought was also to have own per bpf_prog states, but due to ABI
reasons this is not easily possible, i.e. the program code currently
cannot access bpf_prog itself, and copying the rnd_state to/from the
stack scratch space whenever a program uses the prng seems not really
worth the trouble and seems too hacky. If needed, taus113 could in such
cases be implemented within eBPF using a map entry to keep the state
space, or get_random_bytes() could become a second helper in cases where
performance would not be critical.
Both sides can trigger a one-time late init via prandom_init_once() on
the shared state. Performance-wise, there should even be a tiny gain
as bpf_user_rnd_u32() saves one function call. The PRNG needs to live
inside the BPF core since kernels could have a NET-less config as well.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Chema Gonzalez <chema@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a prandom_init_once() facility that works on the rnd_state, so that
users that are keeping their own state independent from prandom_u32() can
initialize their taus113 per cpu states.
The motivation here is similar to net_get_random_once(): initialize the
state as late as possible in the hope that enough entropy has been
collected for the seeding. prandom_init_once() makes use of the recently
introduced prandom_seed_full_state() helper and is generic enough so that
it could also be used on fast-paths due to the DO_ONCE().
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Factor out the full reseed handling code that populates the state
through get_random_bytes() and runs prandom_warmup(). The resulting
prandom_seed_full_state() will be used later on in more than the
current __prandom_reseed() user. Fix also two minor whitespace
issues along the way.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make the get_random_once() helper generic enough, so that functions
in general would only be called once, where one user of this is then
net_get_random_once().
The only implementation specific call is to get_random_bytes(), all
the rest of this *_once() facility would be duplicated among different
subsystems otherwise. The new DO_ONCE() helper will be used by prandom()
later on, but might also be useful for other scenarios/subsystems as
well where a one-time initialization in often-called, possibly fast
path code could occur.
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
There's no good reason why users outside of networking should not
be using this facility, f.e. for initializing their seeds.
Therefore, make it accessible from there as get_random_once().
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit deaa0a6a93 ("net: Lookup actual route when oif is VRF device")
exposed a bug in __ip_route_output_key_hash for VRF devices: on FIB lookup
failure if the oif is specified the current logic drops to make_route on
the assumption that the route tables are wrong. For VRF/L3 master devices
this leads to wrong dst entries and route lookups. For example:
$ ip route ls table vrf-red
unreachable default
broadcast 10.2.1.0 dev eth1 proto kernel scope link src 10.2.1.2
10.2.1.0/24 dev eth1 proto kernel scope link src 10.2.1.2
local 10.2.1.2 dev eth1 proto kernel scope host src 10.2.1.2
broadcast 10.2.1.255 dev eth1 proto kernel scope link src 10.2.1.2
$ ip route get oif vrf-red 1.1.1.1
1.1.1.1 dev vrf-red src 10.0.0.2
cache
With this patch:
$ ip route get oif vrf-red 1.1.1.1
RTNETLINK answers: No route to host
which is the correct response based on the default route
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Similar to commit c29390c6df ("xps: must clear sender_cpu before
forwarding"), we also need to clear the skb->sender_cpu when moving
from RX to TX via skb_do_redirect() due to the shared location of
napi_id (used on RX) and sender_cpu (used on TX).
Fixes: 27b29f6305 ("bpf: add bpf_redirect() helper")
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The recently added hns driver causes a build warning in ARM
allmodconfig builds:
drivers/net/ethernet/hisilicon/hns/hnae.c: In function 'handles_show':
drivers/net/ethernet/hisilicon/hns/hnae.c:452:13: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
j, (u64)h->qs[i]->io_base);
^
This removes the pointless cast and prints the pointer address using
the "%p" format string in all three locations.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This ethernet driver supports the Micorchip enc424j600/626j600 Ethernet
controller over a SPI bus interface. This driver makes use of the regmap API to
optimize access to registers by caching registers where possible.
Datasheet:
http://ww1.microchip.com/downloads/en/DeviceDoc/39935b.pdf
Signed-off-by: Jon Ringle <jringle@gridpoint.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Arun Parameswaran says:
====================
Add support for Broadcom's iProc MDIO and Cygnus Ethernet PHY
This patchset adds support for the iProc MDIO interface and the
Broadcom Cygnus SoC's internal Ethernet PHY.
The internal Ethernet PHY(s) in the Cygnus SoC's are accessed
via the MDIO interface found in most of the iProc based chips.
The patch also consolidates the common API's used by the
Broadcom phys to a common library. Existing Broadcom phy
drivers have been modified to use the common library API's.
This patch series is based on Linux v4.3-rc1 and is avaliable in:
https://github.com/Broadcom/cygnus-linux/tree/cygnus-net-phy-mdio-v3
The Ethernet driver for the iProc family will be submitted soon,
as will the device tree configurations for the different iProc
family SoCs.
Changes from v2:
- Modified drivers/net/phy/Kconfig to modify the BCM_CYGNUS_PHY
driver to 'depends on MDIO_BCM_IPROC' instead of 'select'.
- Added github branch to the cover letter
Changes from v1:
- Updated device tree documentation for the iProc MDIO driver
based on Florian's feedback.
- Moved the core register defines from the Cygnus PHY driver to
'include/linux/brcmphy.h' based on Florian's feedback.
- Created a new patch/commit to modify the bcm7xxx phy driver
to use the new core register defines.
- Modified the Kconfig entry for the Broadcom PHY library to
'tristate' instead of 'bool'
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Modified the bcm7xxx phy driver to remove local core register
defines and use the common ones from "include/linux/brcmphy.h"
Signed-off-by: Arun Parameswaran <arunp@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for the Broadcom Cygnus SoCs internal PHY's.
The PHYs are 1000M/100M/10M capable with support for 'EEE'
and 'APD' (Auto Power Down).
This driver supports the following Broadcom Cygnus SoCs:
- BCM583XX (BCM58300, BCM58302, BCM58303, BCM58305)
- BCM113XX (BCM11300, BCM11320, BCM11350, BCM11360)
The PHY's on these SoC's require some workarounds for
stable operation, both during configuration time and
during suspend/resume. This driver handles the
application of the workarounds.
Signed-off-by: Arun Parameswaran <arunp@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the Broadcom phy library to consolidate common
interfaces shared by Broadcom phy's.
Moved the common interfaces to the 'bcm-phy-lib.c' and updated
the Broadcom PHY drivers to use the new APIs.
Signed-off-by: Arun Parameswaran <arunp@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support for the Broadcom iProc MDIO bus interface.
The MDIO interface can be found in the Broadcom iProc family Soc's.
The MDIO bus is accessed using a combination of command and data
registers. This MDIO driver provides access to the Etherent GPHY's
connected to the MDIO bus.
Signed-off-by: Arun Parameswaran <arunp@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add device tree binding documentation for the Broadcom iProc MDIO
bus driver.
Signed-off-by: Arun Parameswaran <arunp@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Santosh Shilimkar says:
====================
RDS: connection scalability and performance improvements
[v4]
Re-sending the same patches from v3 again since my repost of
patch 05/14 from v3 was whitespace damaged.
[v3]
Updated patch "[PATCH v2 05/14] RDS: defer the over_batch work to
send worker" as per David Miller's comment [4] to avoid the magic
value usage. Patch now makes use of already available but unused
send_batch_count module parameter. Rest of the patches are same as
earlier version v2 [3]
[v2]:
Dropped "[PATCH 05/15] RDS: increase size of hash-table to 8K" from
earlier version [1]. I plan to address the hash table scalability using
re-sizable hash tables as suggested by David Laight and David Miller [2]
This series addresses RDS connection bottlenecks on massive workloads and
improve the RDMA performance almost by 3X. RDS TCP also gets a small gain
of about 12%.
RDS is being used in massive systems with high scalability where several
hundred thousand end points and tens of thousands of local processes
are operating in tens of thousand sockets. Being RC(reliable connection),
socket bind and release happens very often and any inefficiencies in
bind hash look ups hurts the overall system performance. RDS bin hash-table
uses global spin-lock which is the biggest bottleneck. To make matter worst,
it uses rcu inside global lock for hash buckets.
This is being addressed by simply using per bucket rw lock which makes the
locking simple and very efficient. The hash table size is still an issue and
I plan to address it by using re-sizable hash tables as suggested on the list.
For RDS RDMA improvement, the completion handling is revamped so that we
can do batch completions. Both send and receive completion handlers are
split logically to achieve the same. RDS 8K messages being one of the
key usecase, mr pool is adapted to have the 8K mrs along with default 1M
mrs. And while doing this, few fixes and couple of bottlenecks seen with
rds_sendmsg() are addressed.
Series applies against 4.3-rc1 as well net-next. Its tested on Oracle
hardware with IB fabric for both bcopy as well as RDMA mode. RDS TCP is
tested with iXGB NIC. Like last time, iWARP transport is untested with
these changes. The patchset is also available at below git repo:
git://git.kernel.org/pub/scm/linux/kernel/git/ssantosh/linux.git net/rds/4.3-v3
As a side note, the IB HCA driver I used for testing misses at least 3
important patches in upstream to see the full blown IB performance and
am hoping to get that in mainline with help of them.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric W. Biederman says:
====================
net: Pass net through the output path v2
This is the next installment of my work to pass struct net through the
output path so the code does not need to guess how to figure out which
network namespace it is in, and ultimately routes can have output
devices in another network namespace.
The first patch in this series is a fix for a bug that came in when sk
was passed through the functions in the output path, and as such is
probably a candidate for net. At the same time my later patches depend
on it so sending the fix separately would be confusing.
The second patch in this series is another fix that for an issue that
came in when sk was passed through the output path. I don't think it
needs a backport as I don't think anyone uses the path where the code
was incorrect.
The rest of the patchset focuses on the path from xxx_local_out to
dst_output and in the end succeeds in passing sock_net(sk) from the
socket a packet locally originates on to the dst->output function.
Given the size reduction in the code I think this counts as a cleanup as
much as feature work.
There remain a number of helper functions (like ip option processing) to
take care of before the network stack can support destination devices in
other network namespaces but with this set of changes the backbone of
the work is done.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The network namespace is already passed into dst_output pass it into
dst->output lwt->output and friends.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Compute net once in ipvlan_process_v4_outbound and
ipvlan_process_v6_outbound and store it in a variable so that net does
not need to be recomputed next time it is used.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Compute net and store it in a variable in pptp_xmit, so that the value
can be reused the next time it is needed.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Compute net and store it in a variable in the functions
ip_build_and_send_pkt and ip_queue_xmit so that it does not need to be
recomputed next time it is needed.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Store net in a variable in ip_tunnel_xmit so it does not need
to be recomputed when it is used again.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller