2015-07-31 03:17:43 +08:00
|
|
|
/*
|
2016-02-15 12:22:17 +08:00
|
|
|
* Copyright(c) 2015, 2016 Intel Corporation.
|
2015-07-31 03:17:43 +08:00
|
|
|
*
|
|
|
|
* This file is provided under a dual BSD/GPLv2 license. When using or
|
|
|
|
* redistributing this file, you may do so under either license.
|
|
|
|
*
|
|
|
|
* GPL LICENSE SUMMARY
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of version 2 of the GNU General Public License as
|
|
|
|
* published by the Free Software Foundation.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful, but
|
|
|
|
* WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* General Public License for more details.
|
|
|
|
*
|
|
|
|
* BSD LICENSE
|
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
* - Neither the name of Intel Corporation nor the names of its
|
|
|
|
* contributors may be used to endorse or promote products derived
|
|
|
|
* from this software without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
|
|
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
|
|
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
|
|
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
|
|
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
|
|
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
|
|
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
|
|
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
|
|
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include <linux/pci.h>
|
|
|
|
#include <linux/io.h>
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/netdevice.h>
|
|
|
|
#include <linux/vmalloc.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/prefetch.h>
|
2016-01-20 06:42:11 +08:00
|
|
|
#include <rdma/ib_verbs.h>
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
#include "hfi.h"
|
|
|
|
#include "trace.h"
|
|
|
|
#include "qp.h"
|
|
|
|
#include "sdma.h"
|
|
|
|
|
|
|
|
#undef pr_fmt
|
|
|
|
#define pr_fmt(fmt) DRIVER_NAME ": " fmt
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The size has to be longer than this string, so we can append
|
|
|
|
* board/chip information to it in the initialization code.
|
|
|
|
*/
|
|
|
|
const char ib_hfi1_version[] = HFI1_DRIVER_VERSION "\n";
|
|
|
|
|
|
|
|
DEFINE_SPINLOCK(hfi1_devs_lock);
|
|
|
|
LIST_HEAD(hfi1_dev_list);
|
|
|
|
DEFINE_MUTEX(hfi1_mutex); /* general driver use */
|
|
|
|
|
|
|
|
unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU;
|
|
|
|
module_param_named(max_mtu, hfi1_max_mtu, uint, S_IRUGO);
|
2016-04-13 02:17:09 +08:00
|
|
|
MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is " __stringify(
|
|
|
|
HFI1_DEFAULT_MAX_MTU));
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
unsigned int hfi1_cu = 1;
|
|
|
|
module_param_named(cu, hfi1_cu, uint, S_IRUGO);
|
|
|
|
MODULE_PARM_DESC(cu, "Credit return units");
|
|
|
|
|
|
|
|
unsigned long hfi1_cap_mask = HFI1_CAP_MASK_DEFAULT;
|
|
|
|
static int hfi1_caps_set(const char *, const struct kernel_param *);
|
|
|
|
static int hfi1_caps_get(char *, const struct kernel_param *);
|
|
|
|
static const struct kernel_param_ops cap_ops = {
|
|
|
|
.set = hfi1_caps_set,
|
|
|
|
.get = hfi1_caps_get
|
|
|
|
};
|
|
|
|
module_param_cb(cap_mask, &cap_ops, &hfi1_cap_mask, S_IWUSR | S_IRUGO);
|
|
|
|
MODULE_PARM_DESC(cap_mask, "Bit mask of enabled/disabled HW features");
|
|
|
|
|
|
|
|
MODULE_LICENSE("Dual BSD/GPL");
|
|
|
|
MODULE_DESCRIPTION("Intel Omni-Path Architecture driver");
|
|
|
|
MODULE_VERSION(HFI1_DRIVER_VERSION);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* MAX_PKT_RCV is the max # if packets processed per receive interrupt.
|
|
|
|
*/
|
|
|
|
#define MAX_PKT_RECV 64
|
|
|
|
#define EGR_HEAD_UPDATE_THRESHOLD 16
|
|
|
|
|
|
|
|
struct hfi1_ib_stats hfi1_stats;
|
|
|
|
|
|
|
|
static int hfi1_caps_set(const char *val, const struct kernel_param *kp)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
unsigned long *cap_mask_ptr = (unsigned long *)kp->arg,
|
|
|
|
cap_mask = *cap_mask_ptr, value, diff,
|
|
|
|
write_mask = ((HFI1_CAP_WRITABLE_MASK << HFI1_CAP_USER_SHIFT) |
|
|
|
|
HFI1_CAP_WRITABLE_MASK);
|
|
|
|
|
|
|
|
ret = kstrtoul(val, 0, &value);
|
|
|
|
if (ret) {
|
|
|
|
pr_warn("Invalid module parameter value for 'cap_mask'\n");
|
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
/* Get the changed bits (except the locked bit) */
|
|
|
|
diff = value ^ (cap_mask & ~HFI1_CAP_LOCKED_SMASK);
|
|
|
|
|
|
|
|
/* Remove any bits that are not allowed to change after driver load */
|
|
|
|
if (HFI1_CAP_LOCKED() && (diff & ~write_mask)) {
|
|
|
|
pr_warn("Ignoring non-writable capability bits %#lx\n",
|
|
|
|
diff & ~write_mask);
|
|
|
|
diff &= write_mask;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Mask off any reserved bits */
|
|
|
|
diff &= ~HFI1_CAP_RESERVED_MASK;
|
|
|
|
/* Clear any previously set and changing bits */
|
|
|
|
cap_mask &= ~diff;
|
|
|
|
/* Update the bits with the new capability */
|
|
|
|
cap_mask |= (value & diff);
|
|
|
|
/* Check for any kernel/user restrictions */
|
|
|
|
diff = (cap_mask & (HFI1_CAP_MUST_HAVE_KERN << HFI1_CAP_USER_SHIFT)) ^
|
|
|
|
((cap_mask & HFI1_CAP_MUST_HAVE_KERN) << HFI1_CAP_USER_SHIFT);
|
|
|
|
cap_mask &= ~diff;
|
|
|
|
/* Set the bitmask to the final set */
|
|
|
|
*cap_mask_ptr = cap_mask;
|
|
|
|
done:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int hfi1_caps_get(char *buffer, const struct kernel_param *kp)
|
|
|
|
{
|
|
|
|
unsigned long cap_mask = *(unsigned long *)kp->arg;
|
|
|
|
|
|
|
|
cap_mask &= ~HFI1_CAP_LOCKED_SMASK;
|
|
|
|
cap_mask |= ((cap_mask & HFI1_CAP_K2U) << HFI1_CAP_USER_SHIFT);
|
|
|
|
|
|
|
|
return scnprintf(buffer, PAGE_SIZE, "0x%lx", cap_mask);
|
|
|
|
}
|
|
|
|
|
|
|
|
const char *get_unit_name(int unit)
|
|
|
|
{
|
|
|
|
static char iname[16];
|
|
|
|
|
2015-11-17 10:59:27 +08:00
|
|
|
snprintf(iname, sizeof(iname), DRIVER_NAME "_%u", unit);
|
2015-07-31 03:17:43 +08:00
|
|
|
return iname;
|
|
|
|
}
|
|
|
|
|
2016-01-20 06:42:06 +08:00
|
|
|
const char *get_card_name(struct rvt_dev_info *rdi)
|
|
|
|
{
|
|
|
|
struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
|
|
|
|
struct hfi1_devdata *dd = container_of(ibdev,
|
|
|
|
struct hfi1_devdata, verbs_dev);
|
|
|
|
return get_unit_name(dd->unit);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi)
|
|
|
|
{
|
|
|
|
struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
|
|
|
|
struct hfi1_devdata *dd = container_of(ibdev,
|
|
|
|
struct hfi1_devdata, verbs_dev);
|
|
|
|
return dd->pcidev;
|
|
|
|
}
|
|
|
|
|
2015-07-31 03:17:43 +08:00
|
|
|
/*
|
|
|
|
* Return count of units with at least one port ACTIVE.
|
|
|
|
*/
|
|
|
|
int hfi1_count_active_units(void)
|
|
|
|
{
|
|
|
|
struct hfi1_devdata *dd;
|
|
|
|
struct hfi1_pportdata *ppd;
|
|
|
|
unsigned long flags;
|
|
|
|
int pidx, nunits_active = 0;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&hfi1_devs_lock, flags);
|
|
|
|
list_for_each_entry(dd, &hfi1_dev_list, list) {
|
|
|
|
if (!(dd->flags & HFI1_PRESENT) || !dd->kregbase)
|
|
|
|
continue;
|
|
|
|
for (pidx = 0; pidx < dd->num_pports; ++pidx) {
|
|
|
|
ppd = dd->pport + pidx;
|
|
|
|
if (ppd->lid && ppd->linkup) {
|
|
|
|
nunits_active++;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&hfi1_devs_lock, flags);
|
|
|
|
return nunits_active;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Return count of all units, optionally return in arguments
|
|
|
|
* the number of usable (present) units, and the number of
|
|
|
|
* ports that are up.
|
|
|
|
*/
|
|
|
|
int hfi1_count_units(int *npresentp, int *nupp)
|
|
|
|
{
|
|
|
|
int nunits = 0, npresent = 0, nup = 0;
|
|
|
|
struct hfi1_devdata *dd;
|
|
|
|
unsigned long flags;
|
|
|
|
int pidx;
|
|
|
|
struct hfi1_pportdata *ppd;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&hfi1_devs_lock, flags);
|
|
|
|
|
|
|
|
list_for_each_entry(dd, &hfi1_dev_list, list) {
|
|
|
|
nunits++;
|
|
|
|
if ((dd->flags & HFI1_PRESENT) && dd->kregbase)
|
|
|
|
npresent++;
|
|
|
|
for (pidx = 0; pidx < dd->num_pports; ++pidx) {
|
|
|
|
ppd = dd->pport + pidx;
|
|
|
|
if (ppd->lid && ppd->linkup)
|
|
|
|
nup++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&hfi1_devs_lock, flags);
|
|
|
|
|
|
|
|
if (npresentp)
|
|
|
|
*npresentp = npresent;
|
|
|
|
if (nupp)
|
|
|
|
*nupp = nup;
|
|
|
|
|
|
|
|
return nunits;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Get address of eager buffer from it's index (allocated in chunks, not
|
|
|
|
* contiguous).
|
|
|
|
*/
|
|
|
|
static inline void *get_egrbuf(const struct hfi1_ctxtdata *rcd, u64 rhf,
|
|
|
|
u8 *update)
|
|
|
|
{
|
|
|
|
u32 idx = rhf_egr_index(rhf), offset = rhf_egr_buf_offset(rhf);
|
|
|
|
|
|
|
|
*update |= !(idx & (rcd->egrbufs.threshold - 1)) && !offset;
|
|
|
|
return (void *)(((u64)(rcd->egrbufs.rcvtids[idx].addr)) +
|
|
|
|
(offset * RCV_BUF_BLOCK_SIZE));
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Validate and encode the a given RcvArray Buffer size.
|
|
|
|
* The function will check whether the given size falls within
|
|
|
|
* allowed size ranges for the respective type and, optionally,
|
|
|
|
* return the proper encoding.
|
|
|
|
*/
|
|
|
|
inline int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encoded)
|
|
|
|
{
|
2016-03-05 01:10:17 +08:00
|
|
|
if (unlikely(!PAGE_ALIGNED(size)))
|
2015-07-31 03:17:43 +08:00
|
|
|
return 0;
|
|
|
|
if (unlikely(size < MIN_EAGER_BUFFER))
|
|
|
|
return 0;
|
|
|
|
if (size >
|
|
|
|
(type == PT_EAGER ? MAX_EAGER_BUFFER : MAX_EXPECTED_BUFFER))
|
|
|
|
return 0;
|
|
|
|
if (encoded)
|
|
|
|
*encoded = ilog2(size / PAGE_SIZE) + 1;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd,
|
|
|
|
struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
struct hfi1_message_header *rhdr = packet->hdr;
|
|
|
|
u32 rte = rhf_rcv_type_err(packet->rhf);
|
|
|
|
int lnh = be16_to_cpu(rhdr->lrh[0]) & 3;
|
|
|
|
struct hfi1_ibport *ibp = &ppd->ibport_data;
|
2016-01-20 06:43:44 +08:00
|
|
|
struct hfi1_devdata *dd = ppd->dd;
|
|
|
|
struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (packet->rhf & RHF_TID_ERR) {
|
|
|
|
/* For TIDERR and RC QPs preemptively schedule a NAK */
|
|
|
|
struct hfi1_ib_header *hdr = (struct hfi1_ib_header *)rhdr;
|
|
|
|
struct hfi1_other_headers *ohdr = NULL;
|
|
|
|
u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */
|
|
|
|
u16 lid = be16_to_cpu(hdr->lrh[1]);
|
|
|
|
u32 qp_num;
|
|
|
|
u32 rcv_flags = 0;
|
|
|
|
|
|
|
|
/* Sanity check packet */
|
|
|
|
if (tlen < 24)
|
|
|
|
goto drop;
|
|
|
|
|
|
|
|
/* Check for GRH */
|
2016-02-15 12:22:00 +08:00
|
|
|
if (lnh == HFI1_LRH_BTH) {
|
2015-07-31 03:17:43 +08:00
|
|
|
ohdr = &hdr->u.oth;
|
2016-02-15 12:22:00 +08:00
|
|
|
} else if (lnh == HFI1_LRH_GRH) {
|
2015-07-31 03:17:43 +08:00
|
|
|
u32 vtf;
|
|
|
|
|
|
|
|
ohdr = &hdr->u.l.oth;
|
|
|
|
if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
|
|
|
|
goto drop;
|
|
|
|
vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow);
|
|
|
|
if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
|
|
|
|
goto drop;
|
|
|
|
rcv_flags |= HFI1_HAS_GRH;
|
2016-02-15 12:22:00 +08:00
|
|
|
} else {
|
2015-07-31 03:17:43 +08:00
|
|
|
goto drop;
|
2016-02-15 12:22:00 +08:00
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
/* Get the destination QP number. */
|
2016-01-20 06:43:44 +08:00
|
|
|
qp_num = be32_to_cpu(ohdr->bth[1]) & RVT_QPN_MASK;
|
2016-01-20 06:42:11 +08:00
|
|
|
if (lid < be16_to_cpu(IB_MULTICAST_LID_BASE)) {
|
2016-01-20 06:42:28 +08:00
|
|
|
struct rvt_qp *qp;
|
2015-10-26 22:28:43 +08:00
|
|
|
unsigned long flags;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
rcu_read_lock();
|
2016-01-20 06:43:44 +08:00
|
|
|
qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num);
|
2015-07-31 03:17:43 +08:00
|
|
|
if (!qp) {
|
|
|
|
rcu_read_unlock();
|
|
|
|
goto drop;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Handle only RC QPs - for other QP types drop error
|
|
|
|
* packet.
|
|
|
|
*/
|
2015-10-26 22:28:43 +08:00
|
|
|
spin_lock_irqsave(&qp->r_lock, flags);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
/* Check for valid receive state. */
|
2016-01-20 06:43:33 +08:00
|
|
|
if (!(ib_rvt_state_ops[qp->state] &
|
|
|
|
RVT_PROCESS_RECV_OK)) {
|
2016-01-20 06:42:39 +08:00
|
|
|
ibp->rvp.n_pkt_drops++;
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
switch (qp->ibqp.qp_type) {
|
|
|
|
case IB_QPT_RC:
|
|
|
|
hfi1_rc_hdrerr(
|
|
|
|
rcd,
|
|
|
|
hdr,
|
|
|
|
rcv_flags,
|
|
|
|
qp);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
/* For now don't handle any other QP types */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2015-10-26 22:28:43 +08:00
|
|
|
spin_unlock_irqrestore(&qp->r_lock, flags);
|
2015-07-31 03:17:43 +08:00
|
|
|
rcu_read_unlock();
|
|
|
|
} /* Unicast QP */
|
|
|
|
} /* Valid packet with TIDErr */
|
|
|
|
|
|
|
|
/* handle "RcvTypeErr" flags */
|
|
|
|
switch (rte) {
|
|
|
|
case RHF_RTE_ERROR_OP_CODE_ERR:
|
|
|
|
{
|
|
|
|
u32 opcode;
|
|
|
|
void *ebuf = NULL;
|
|
|
|
__be32 *bth = NULL;
|
|
|
|
|
|
|
|
if (rhf_use_egr_bfr(packet->rhf))
|
|
|
|
ebuf = packet->ebuf;
|
|
|
|
|
2016-02-15 12:19:49 +08:00
|
|
|
if (!ebuf)
|
2015-07-31 03:17:43 +08:00
|
|
|
goto drop; /* this should never happen */
|
|
|
|
|
|
|
|
if (lnh == HFI1_LRH_BTH)
|
|
|
|
bth = (__be32 *)ebuf;
|
|
|
|
else if (lnh == HFI1_LRH_GRH)
|
|
|
|
bth = (__be32 *)((char *)ebuf + sizeof(struct ib_grh));
|
|
|
|
else
|
|
|
|
goto drop;
|
|
|
|
|
|
|
|
opcode = be32_to_cpu(bth[0]) >> 24;
|
|
|
|
opcode &= 0xff;
|
|
|
|
|
|
|
|
if (opcode == IB_OPCODE_CNP) {
|
|
|
|
/*
|
|
|
|
* Only in pre-B0 h/w is the CNP_OPCODE handled
|
2015-12-22 10:57:44 +08:00
|
|
|
* via this code path.
|
2015-07-31 03:17:43 +08:00
|
|
|
*/
|
2016-01-20 06:42:28 +08:00
|
|
|
struct rvt_qp *qp = NULL;
|
2015-07-31 03:17:43 +08:00
|
|
|
u32 lqpn, rqpn;
|
|
|
|
u16 rlid;
|
|
|
|
u8 svc_type, sl, sc5;
|
|
|
|
|
2016-07-26 04:40:34 +08:00
|
|
|
sc5 = hdr2sc(rhdr, packet->rhf);
|
2015-07-31 03:17:43 +08:00
|
|
|
sl = ibp->sc_to_sl[sc5];
|
|
|
|
|
2016-01-20 06:43:44 +08:00
|
|
|
lqpn = be32_to_cpu(bth[1]) & RVT_QPN_MASK;
|
2015-07-31 03:17:43 +08:00
|
|
|
rcu_read_lock();
|
2016-01-20 06:43:44 +08:00
|
|
|
qp = rvt_lookup_qpn(rdi, &ibp->rvp, lqpn);
|
2016-02-15 12:19:49 +08:00
|
|
|
if (!qp) {
|
2015-07-31 03:17:43 +08:00
|
|
|
rcu_read_unlock();
|
|
|
|
goto drop;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (qp->ibqp.qp_type) {
|
|
|
|
case IB_QPT_UD:
|
|
|
|
rlid = 0;
|
|
|
|
rqpn = 0;
|
|
|
|
svc_type = IB_CC_SVCTYPE_UD;
|
|
|
|
break;
|
|
|
|
case IB_QPT_UC:
|
|
|
|
rlid = be16_to_cpu(rhdr->lrh[3]);
|
|
|
|
rqpn = qp->remote_qpn;
|
|
|
|
svc_type = IB_CC_SVCTYPE_UC;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
goto drop;
|
|
|
|
}
|
|
|
|
|
|
|
|
process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
|
|
|
|
packet->rhf &= ~RHF_RCV_TYPE_ERR_SMASK;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
drop:
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void init_packet(struct hfi1_ctxtdata *rcd,
|
2016-02-15 12:21:52 +08:00
|
|
|
struct hfi1_packet *packet)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
packet->rsize = rcd->rcvhdrqentsize; /* words */
|
|
|
|
packet->maxcnt = rcd->rcvhdrq_cnt * packet->rsize; /* words */
|
|
|
|
packet->rcd = rcd;
|
|
|
|
packet->updegr = 0;
|
|
|
|
packet->etail = -1;
|
2015-10-26 22:28:44 +08:00
|
|
|
packet->rhf_addr = get_rhf_addr(rcd);
|
2015-07-31 03:17:43 +08:00
|
|
|
packet->rhf = rhf_to_cpu(packet->rhf_addr);
|
|
|
|
packet->rhqoff = rcd->head;
|
|
|
|
packet->numpkt = 0;
|
|
|
|
packet->rcv_flags = 0;
|
|
|
|
}
|
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
void hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
|
|
|
|
bool do_cnp)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
|
2016-07-26 04:38:07 +08:00
|
|
|
struct hfi1_ib_header *hdr = pkt->hdr;
|
|
|
|
struct hfi1_other_headers *ohdr = pkt->ohdr;
|
|
|
|
struct ib_grh *grh = NULL;
|
|
|
|
u32 rqpn = 0, bth1;
|
|
|
|
u16 rlid, dlid = be16_to_cpu(hdr->lrh[1]);
|
|
|
|
u8 sc, svc_type;
|
|
|
|
bool is_mcast = false;
|
|
|
|
|
|
|
|
if (pkt->rcv_flags & HFI1_HAS_GRH)
|
|
|
|
grh = &hdr->u.l.grh;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
switch (qp->ibqp.qp_type) {
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
case IB_QPT_SMI:
|
|
|
|
case IB_QPT_GSI:
|
2015-07-31 03:17:43 +08:00
|
|
|
case IB_QPT_UD:
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
rlid = be16_to_cpu(hdr->lrh[3]);
|
2016-01-20 06:43:44 +08:00
|
|
|
rqpn = be32_to_cpu(ohdr->u.ud.deth[1]) & RVT_QPN_MASK;
|
2015-07-31 03:17:43 +08:00
|
|
|
svc_type = IB_CC_SVCTYPE_UD;
|
2016-07-26 04:38:07 +08:00
|
|
|
is_mcast = (dlid > be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
|
|
|
|
(dlid != be16_to_cpu(IB_LID_PERMISSIVE));
|
2015-07-31 03:17:43 +08:00
|
|
|
break;
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
case IB_QPT_UC:
|
|
|
|
rlid = qp->remote_ah_attr.dlid;
|
|
|
|
rqpn = qp->remote_qpn;
|
|
|
|
svc_type = IB_CC_SVCTYPE_UC;
|
|
|
|
break;
|
|
|
|
case IB_QPT_RC:
|
|
|
|
rlid = qp->remote_ah_attr.dlid;
|
|
|
|
rqpn = qp->remote_qpn;
|
|
|
|
svc_type = IB_CC_SVCTYPE_RC;
|
|
|
|
break;
|
2015-07-31 03:17:43 +08:00
|
|
|
default:
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
sc = hdr2sc((struct hfi1_message_header *)hdr, pkt->rhf);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
bth1 = be32_to_cpu(ohdr->bth[1]);
|
|
|
|
if (do_cnp && (bth1 & HFI1_FECN_SMASK)) {
|
2015-07-31 03:17:43 +08:00
|
|
|
u16 pkey = (u16)be32_to_cpu(ohdr->bth[0]);
|
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
return_cnp(ibp, qp, rqpn, pkey, dlid, rlid, sc, grh);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
if (!is_mcast && (bth1 & HFI1_BECN_SMASK)) {
|
2015-07-31 03:17:43 +08:00
|
|
|
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
|
2016-01-20 06:43:44 +08:00
|
|
|
u32 lqpn = bth1 & RVT_QPN_MASK;
|
2016-07-26 04:38:07 +08:00
|
|
|
u8 sl = ibp->sc_to_sl[sc];
|
2015-07-31 03:17:43 +08:00
|
|
|
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
2016-07-26 04:38:07 +08:00
|
|
|
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
struct ps_mdata {
|
|
|
|
struct hfi1_ctxtdata *rcd;
|
|
|
|
u32 rsize;
|
|
|
|
u32 maxcnt;
|
|
|
|
u32 ps_head;
|
|
|
|
u32 ps_tail;
|
|
|
|
u32 ps_seq;
|
|
|
|
};
|
|
|
|
|
|
|
|
static inline void init_ps_mdata(struct ps_mdata *mdata,
|
|
|
|
struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
struct hfi1_ctxtdata *rcd = packet->rcd;
|
|
|
|
|
|
|
|
mdata->rcd = rcd;
|
|
|
|
mdata->rsize = packet->rsize;
|
|
|
|
mdata->maxcnt = packet->maxcnt;
|
2015-11-05 10:10:09 +08:00
|
|
|
mdata->ps_head = packet->rhqoff;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
|
2015-11-05 10:10:09 +08:00
|
|
|
mdata->ps_tail = get_rcvhdrtail(rcd);
|
2015-11-11 13:35:19 +08:00
|
|
|
if (rcd->ctxt == HFI1_CTRL_CTXT)
|
|
|
|
mdata->ps_seq = rcd->seq_cnt;
|
|
|
|
else
|
|
|
|
mdata->ps_seq = 0; /* not used with DMA_RTAIL */
|
2015-07-31 03:17:43 +08:00
|
|
|
} else {
|
|
|
|
mdata->ps_tail = 0; /* used only with DMA_RTAIL*/
|
|
|
|
mdata->ps_seq = rcd->seq_cnt;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
static inline int ps_done(struct ps_mdata *mdata, u64 rhf,
|
|
|
|
struct hfi1_ctxtdata *rcd)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
2015-11-11 13:35:19 +08:00
|
|
|
if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL))
|
2015-07-31 03:17:43 +08:00
|
|
|
return mdata->ps_head == mdata->ps_tail;
|
|
|
|
return mdata->ps_seq != rhf_rcv_seq(rhf);
|
|
|
|
}
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
static inline int ps_skip(struct ps_mdata *mdata, u64 rhf,
|
|
|
|
struct hfi1_ctxtdata *rcd)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Control context can potentially receive an invalid rhf.
|
|
|
|
* Drop such packets.
|
|
|
|
*/
|
|
|
|
if ((rcd->ctxt == HFI1_CTRL_CTXT) && (mdata->ps_head != mdata->ps_tail))
|
|
|
|
return mdata->ps_seq != rhf_rcv_seq(rhf);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void update_ps_mdata(struct ps_mdata *mdata,
|
|
|
|
struct hfi1_ctxtdata *rcd)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
mdata->ps_head += mdata->rsize;
|
2015-11-05 10:10:09 +08:00
|
|
|
if (mdata->ps_head >= mdata->maxcnt)
|
2015-07-31 03:17:43 +08:00
|
|
|
mdata->ps_head = 0;
|
2015-11-11 13:35:19 +08:00
|
|
|
|
|
|
|
/* Control context must do seq counting */
|
|
|
|
if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL) ||
|
|
|
|
(rcd->ctxt == HFI1_CTRL_CTXT)) {
|
2015-07-31 03:17:43 +08:00
|
|
|
if (++mdata->ps_seq > 13)
|
|
|
|
mdata->ps_seq = 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* prescan_rxq - search through the receive queue looking for packets
|
|
|
|
* containing Excplicit Congestion Notifications (FECNs, or BECNs).
|
|
|
|
* When an ECN is found, process the Congestion Notification, and toggle
|
|
|
|
* it off.
|
2016-02-04 06:32:57 +08:00
|
|
|
* This is declared as a macro to allow quick checking of the port to avoid
|
|
|
|
* the overhead of a function call if not enabled.
|
2015-07-31 03:17:43 +08:00
|
|
|
*/
|
2016-02-04 06:32:57 +08:00
|
|
|
#define prescan_rxq(rcd, packet) \
|
|
|
|
do { \
|
|
|
|
if (rcd->ppd->cc_prescan) \
|
|
|
|
__prescan_rxq(packet); \
|
|
|
|
} while (0)
|
|
|
|
static void __prescan_rxq(struct hfi1_packet *packet)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
struct hfi1_ctxtdata *rcd = packet->rcd;
|
|
|
|
struct ps_mdata mdata;
|
|
|
|
|
|
|
|
init_ps_mdata(&mdata, packet);
|
|
|
|
|
|
|
|
while (1) {
|
|
|
|
struct hfi1_devdata *dd = rcd->dd;
|
|
|
|
struct hfi1_ibport *ibp = &rcd->ppd->ibport_data;
|
2016-02-15 12:19:41 +08:00
|
|
|
__le32 *rhf_addr = (__le32 *)rcd->rcvhdrq + mdata.ps_head +
|
2015-07-31 03:17:43 +08:00
|
|
|
dd->rhf_offset;
|
2016-01-20 06:42:28 +08:00
|
|
|
struct rvt_qp *qp;
|
2015-07-31 03:17:43 +08:00
|
|
|
struct hfi1_ib_header *hdr;
|
|
|
|
struct hfi1_other_headers *ohdr;
|
2016-01-20 06:43:44 +08:00
|
|
|
struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
|
2015-07-31 03:17:43 +08:00
|
|
|
u64 rhf = rhf_to_cpu(rhf_addr);
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
u32 etype = rhf_rcv_type(rhf), qpn, bth1;
|
2015-07-31 03:17:43 +08:00
|
|
|
int is_ecn = 0;
|
|
|
|
u8 lnh;
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
if (ps_done(&mdata, rhf, rcd))
|
2015-07-31 03:17:43 +08:00
|
|
|
break;
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
if (ps_skip(&mdata, rhf, rcd))
|
|
|
|
goto next;
|
|
|
|
|
2015-07-31 03:17:43 +08:00
|
|
|
if (etype != RHF_RCV_TYPE_IB)
|
|
|
|
goto next;
|
|
|
|
|
|
|
|
hdr = (struct hfi1_ib_header *)
|
|
|
|
hfi1_get_msgheader(dd, rhf_addr);
|
|
|
|
lnh = be16_to_cpu(hdr->lrh[0]) & 3;
|
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
if (lnh == HFI1_LRH_BTH)
|
2015-07-31 03:17:43 +08:00
|
|
|
ohdr = &hdr->u.oth;
|
2016-07-26 04:38:07 +08:00
|
|
|
else if (lnh == HFI1_LRH_GRH)
|
2015-07-31 03:17:43 +08:00
|
|
|
ohdr = &hdr->u.l.oth;
|
2016-07-26 04:38:07 +08:00
|
|
|
else
|
2015-07-31 03:17:43 +08:00
|
|
|
goto next; /* just in case */
|
2016-07-26 04:38:07 +08:00
|
|
|
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
bth1 = be32_to_cpu(ohdr->bth[1]);
|
|
|
|
is_ecn = !!(bth1 & (HFI1_FECN_SMASK | HFI1_BECN_SMASK));
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
if (!is_ecn)
|
|
|
|
goto next;
|
|
|
|
|
2016-01-20 06:43:44 +08:00
|
|
|
qpn = bth1 & RVT_QPN_MASK;
|
2015-07-31 03:17:43 +08:00
|
|
|
rcu_read_lock();
|
2016-01-20 06:43:44 +08:00
|
|
|
qp = rvt_lookup_qpn(rdi, &ibp->rvp, qpn);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2016-02-15 12:19:49 +08:00
|
|
|
if (!qp) {
|
2015-07-31 03:17:43 +08:00
|
|
|
rcu_read_unlock();
|
|
|
|
goto next;
|
|
|
|
}
|
|
|
|
|
2016-07-26 04:38:07 +08:00
|
|
|
process_ecn(qp, packet, true);
|
2015-07-31 03:17:43 +08:00
|
|
|
rcu_read_unlock();
|
staging/rdma/hfi1: optionally prescan rx queue for {B, F}ECNs - UC, RC
To more rapidly respond to Explicit Congestion Notifications, prescan the
receive queue, and process FECNs, and BECNs first. When a UC, or RC packet
containing a FECN, or BECN is found, immediately react to the ECN (either by
returning a CNP, or adjusting the injection rate). Afterward, the packet will
be processed normally.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Arthur Kepner <arthur.kepner@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2015-11-05 10:10:10 +08:00
|
|
|
|
|
|
|
/* turn off BECN, FECN */
|
|
|
|
bth1 &= ~(HFI1_FECN_SMASK | HFI1_BECN_SMASK);
|
|
|
|
ohdr->bth[1] = cpu_to_be32(bth1);
|
2015-07-31 03:17:43 +08:00
|
|
|
next:
|
2015-11-11 13:35:19 +08:00
|
|
|
update_ps_mdata(&mdata, rcd);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
}
|
2015-11-11 13:35:19 +08:00
|
|
|
|
|
|
|
static inline int skip_rcv_packet(struct hfi1_packet *packet, int thread)
|
|
|
|
{
|
|
|
|
int ret = RCV_PKT_OK;
|
|
|
|
|
|
|
|
/* Set up for the next packet */
|
|
|
|
packet->rhqoff += packet->rsize;
|
|
|
|
if (packet->rhqoff >= packet->maxcnt)
|
|
|
|
packet->rhqoff = 0;
|
|
|
|
|
|
|
|
packet->numpkt++;
|
|
|
|
if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0)) {
|
|
|
|
if (thread) {
|
|
|
|
cond_resched();
|
|
|
|
} else {
|
|
|
|
ret = RCV_PKT_LIMIT;
|
|
|
|
this_cpu_inc(*packet->rcd->dd->rcv_limit);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
|
|
|
|
packet->rcd->dd->rhf_offset;
|
|
|
|
packet->rhf = rhf_to_cpu(packet->rhf_addr);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
static inline int process_rcv_packet(struct hfi1_packet *packet, int thread)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
int ret = RCV_PKT_OK;
|
|
|
|
|
|
|
|
packet->hdr = hfi1_get_msgheader(packet->rcd->dd,
|
|
|
|
packet->rhf_addr);
|
|
|
|
packet->hlen = (u8 *)packet->rhf_addr - (u8 *)packet->hdr;
|
|
|
|
packet->etype = rhf_rcv_type(packet->rhf);
|
|
|
|
/* total length */
|
|
|
|
packet->tlen = rhf_pkt_len(packet->rhf); /* in bytes */
|
|
|
|
/* retrieve eager buffer details */
|
|
|
|
packet->ebuf = NULL;
|
|
|
|
if (rhf_use_egr_bfr(packet->rhf)) {
|
|
|
|
packet->etail = rhf_egr_index(packet->rhf);
|
|
|
|
packet->ebuf = get_egrbuf(packet->rcd, packet->rhf,
|
|
|
|
&packet->updegr);
|
|
|
|
/*
|
|
|
|
* Prefetch the contents of the eager buffer. It is
|
|
|
|
* OK to send a negative length to prefetch_range().
|
|
|
|
* The +2 is the size of the RHF.
|
|
|
|
*/
|
|
|
|
prefetch_range(packet->ebuf,
|
2016-02-15 12:21:52 +08:00
|
|
|
packet->tlen - ((packet->rcd->rcvhdrqentsize -
|
|
|
|
(rhf_hdrq_offset(packet->rhf)
|
|
|
|
+ 2)) * 4));
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Call a type specific handler for the packet. We
|
|
|
|
* should be able to trust that etype won't be beyond
|
|
|
|
* the range of valid indexes. If so something is really
|
|
|
|
* wrong and we can probably just let things come
|
|
|
|
* crashing down. There is no need to eat another
|
|
|
|
* comparison in this performance critical code.
|
|
|
|
*/
|
|
|
|
packet->rcd->dd->rhf_rcv_function_map[packet->etype](packet);
|
|
|
|
packet->numpkt++;
|
|
|
|
|
|
|
|
/* Set up for the next packet */
|
|
|
|
packet->rhqoff += packet->rsize;
|
|
|
|
if (packet->rhqoff >= packet->maxcnt)
|
|
|
|
packet->rhqoff = 0;
|
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0)) {
|
|
|
|
if (thread) {
|
|
|
|
cond_resched();
|
|
|
|
} else {
|
|
|
|
ret = RCV_PKT_LIMIT;
|
|
|
|
this_cpu_inc(*packet->rcd->dd->rcv_limit);
|
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2016-02-15 12:19:41 +08:00
|
|
|
packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
|
2015-07-31 03:17:43 +08:00
|
|
|
packet->rcd->dd->rhf_offset;
|
|
|
|
packet->rhf = rhf_to_cpu(packet->rhf_addr);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void process_rcv_update(int last, struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Update head regs etc., every 16 packets, if not last pkt,
|
|
|
|
* to help prevent rcvhdrq overflows, when many packets
|
|
|
|
* are processed and queue is nearly full.
|
|
|
|
* Don't request an interrupt for intermediate updates.
|
|
|
|
*/
|
|
|
|
if (!last && !(packet->numpkt & 0xf)) {
|
|
|
|
update_usrhead(packet->rcd, packet->rhqoff, packet->updegr,
|
|
|
|
packet->etail, 0, 0);
|
|
|
|
packet->updegr = 0;
|
|
|
|
}
|
|
|
|
packet->rcv_flags = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void finish_packet(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* Nothing we need to free for the packet.
|
|
|
|
*
|
|
|
|
* The only thing we need to do is a final update and call for an
|
|
|
|
* interrupt
|
|
|
|
*/
|
|
|
|
update_usrhead(packet->rcd, packet->rcd->head, packet->updegr,
|
|
|
|
packet->etail, rcv_intr_dynamic, packet->numpkt);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void process_rcv_qp_work(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
struct hfi1_ctxtdata *rcd;
|
2016-01-20 06:42:28 +08:00
|
|
|
struct rvt_qp *qp, *nqp;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
rcd = packet->rcd;
|
|
|
|
rcd->head = packet->rhqoff;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Iterate over all QPs waiting to respond.
|
|
|
|
* The list won't change since the IRQ is only run on one CPU.
|
|
|
|
*/
|
|
|
|
list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
|
|
|
|
list_del_init(&qp->rspwait);
|
2016-01-20 06:43:01 +08:00
|
|
|
if (qp->r_flags & RVT_R_RSP_NAK) {
|
|
|
|
qp->r_flags &= ~RVT_R_RSP_NAK;
|
2015-07-31 03:17:43 +08:00
|
|
|
hfi1_send_rc_ack(rcd, qp, 0);
|
|
|
|
}
|
2016-01-20 06:43:01 +08:00
|
|
|
if (qp->r_flags & RVT_R_RSP_SEND) {
|
2015-07-31 03:17:43 +08:00
|
|
|
unsigned long flags;
|
|
|
|
|
2016-01-20 06:43:01 +08:00
|
|
|
qp->r_flags &= ~RVT_R_RSP_SEND;
|
2015-07-31 03:17:43 +08:00
|
|
|
spin_lock_irqsave(&qp->s_lock, flags);
|
2016-01-20 06:43:33 +08:00
|
|
|
if (ib_rvt_state_ops[qp->state] &
|
|
|
|
RVT_PROCESS_OR_FLUSH_SEND)
|
2015-07-31 03:17:43 +08:00
|
|
|
hfi1_schedule_send(qp);
|
|
|
|
spin_unlock_irqrestore(&qp->s_lock, flags);
|
|
|
|
}
|
|
|
|
if (atomic_dec_and_test(&qp->refcount))
|
|
|
|
wake_up(&qp->wait);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Handle receive interrupts when using the no dma rtail option.
|
|
|
|
*/
|
2015-10-26 22:28:44 +08:00
|
|
|
int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
u32 seq;
|
2015-10-26 22:28:44 +08:00
|
|
|
int last = RCV_PKT_OK;
|
2015-07-31 03:17:43 +08:00
|
|
|
struct hfi1_packet packet;
|
|
|
|
|
|
|
|
init_packet(rcd, &packet);
|
|
|
|
seq = rhf_rcv_seq(packet.rhf);
|
2015-10-26 22:28:44 +08:00
|
|
|
if (seq != rcd->seq_cnt) {
|
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
goto bail;
|
2015-10-26 22:28:44 +08:00
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2016-02-04 06:32:57 +08:00
|
|
|
prescan_rxq(rcd, &packet);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
while (last == RCV_PKT_OK) {
|
|
|
|
last = process_rcv_packet(&packet, thread);
|
2015-07-31 03:17:43 +08:00
|
|
|
seq = rhf_rcv_seq(packet.rhf);
|
|
|
|
if (++rcd->seq_cnt > 13)
|
|
|
|
rcd->seq_cnt = 1;
|
|
|
|
if (seq != rcd->seq_cnt)
|
2015-10-26 22:28:44 +08:00
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
process_rcv_update(last, &packet);
|
|
|
|
}
|
|
|
|
process_rcv_qp_work(&packet);
|
|
|
|
bail:
|
|
|
|
finish_packet(&packet);
|
2015-10-26 22:28:44 +08:00
|
|
|
return last;
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
u32 hdrqtail;
|
2015-10-26 22:28:44 +08:00
|
|
|
int last = RCV_PKT_OK;
|
2015-07-31 03:17:43 +08:00
|
|
|
struct hfi1_packet packet;
|
|
|
|
|
|
|
|
init_packet(rcd, &packet);
|
|
|
|
hdrqtail = get_rcvhdrtail(rcd);
|
2015-10-26 22:28:44 +08:00
|
|
|
if (packet.rhqoff == hdrqtail) {
|
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
goto bail;
|
2015-10-26 22:28:44 +08:00
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
smp_rmb(); /* prevent speculative reads of dma'ed hdrq */
|
|
|
|
|
2016-02-04 06:32:57 +08:00
|
|
|
prescan_rxq(rcd, &packet);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
while (last == RCV_PKT_OK) {
|
|
|
|
last = process_rcv_packet(&packet, thread);
|
2015-07-31 03:17:43 +08:00
|
|
|
if (packet.rhqoff == hdrqtail)
|
2015-10-26 22:28:44 +08:00
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
process_rcv_update(last, &packet);
|
|
|
|
}
|
|
|
|
process_rcv_qp_work(&packet);
|
|
|
|
bail:
|
|
|
|
finish_packet(&packet);
|
2015-10-26 22:28:44 +08:00
|
|
|
return last;
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline void set_all_nodma_rtail(struct hfi1_devdata *dd)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
|
2015-07-31 03:17:43 +08:00
|
|
|
dd->rcd[i]->do_interrupt =
|
|
|
|
&handle_receive_interrupt_nodma_rtail;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void set_all_dma_rtail(struct hfi1_devdata *dd)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
|
2015-07-31 03:17:43 +08:00
|
|
|
dd->rcd[i]->do_interrupt =
|
|
|
|
&handle_receive_interrupt_dma_rtail;
|
|
|
|
}
|
|
|
|
|
2016-01-12 07:32:21 +08:00
|
|
|
void set_all_slowpath(struct hfi1_devdata *dd)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
/* HFI1_CTRL_CTXT must always use the slow path interrupt handler */
|
|
|
|
for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
|
|
|
|
dd->rcd[i]->do_interrupt = &handle_receive_interrupt;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int set_armed_to_active(struct hfi1_ctxtdata *rcd,
|
2016-08-09 23:19:55 +08:00
|
|
|
struct hfi1_packet *packet,
|
2016-01-12 07:32:21 +08:00
|
|
|
struct hfi1_devdata *dd)
|
|
|
|
{
|
|
|
|
struct work_struct *lsaw = &rcd->ppd->linkstate_active_work;
|
2016-08-09 23:19:55 +08:00
|
|
|
struct hfi1_message_header *hdr = hfi1_get_msgheader(packet->rcd->dd,
|
|
|
|
packet->rhf_addr);
|
2016-08-09 23:19:56 +08:00
|
|
|
u8 etype = rhf_rcv_type(packet->rhf);
|
2016-01-12 07:32:21 +08:00
|
|
|
|
2016-08-09 23:19:56 +08:00
|
|
|
if (etype == RHF_RCV_TYPE_IB && hdr2sc(hdr, packet->rhf) != 0xf) {
|
2016-01-12 07:32:21 +08:00
|
|
|
int hwstate = read_logical_state(dd);
|
|
|
|
|
|
|
|
if (hwstate != LSTATE_ACTIVE) {
|
|
|
|
dd_dev_info(dd, "Unexpected link state %d\n", hwstate);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
queue_work(rcd->ppd->hfi1_wq, lsaw);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-07-31 03:17:43 +08:00
|
|
|
/*
|
|
|
|
* handle_receive_interrupt - receive a packet
|
|
|
|
* @rcd: the context
|
|
|
|
*
|
|
|
|
* Called from interrupt handler for errors or receive interrupt.
|
|
|
|
* This is the slow path interrupt handler.
|
|
|
|
*/
|
2015-10-26 22:28:44 +08:00
|
|
|
int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
|
|
|
struct hfi1_devdata *dd = rcd->dd;
|
|
|
|
u32 hdrqtail;
|
2015-11-11 13:35:19 +08:00
|
|
|
int needset, last = RCV_PKT_OK;
|
2015-07-31 03:17:43 +08:00
|
|
|
struct hfi1_packet packet;
|
2015-11-11 13:35:19 +08:00
|
|
|
int skip_pkt = 0;
|
|
|
|
|
|
|
|
/* Control context will always use the slow path interrupt handler */
|
|
|
|
needset = (rcd->ctxt == HFI1_CTRL_CTXT) ? 0 : 1;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
init_packet(rcd, &packet);
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
|
2015-07-31 03:17:43 +08:00
|
|
|
u32 seq = rhf_rcv_seq(packet.rhf);
|
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
if (seq != rcd->seq_cnt) {
|
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
goto bail;
|
2015-10-26 22:28:44 +08:00
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
hdrqtail = 0;
|
|
|
|
} else {
|
|
|
|
hdrqtail = get_rcvhdrtail(rcd);
|
2015-10-26 22:28:44 +08:00
|
|
|
if (packet.rhqoff == hdrqtail) {
|
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
goto bail;
|
2015-10-26 22:28:44 +08:00
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
smp_rmb(); /* prevent speculative reads of dma'ed hdrq */
|
2015-11-11 13:35:19 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Control context can potentially receive an invalid
|
|
|
|
* rhf. Drop such packets.
|
|
|
|
*/
|
|
|
|
if (rcd->ctxt == HFI1_CTRL_CTXT) {
|
|
|
|
u32 seq = rhf_rcv_seq(packet.rhf);
|
|
|
|
|
|
|
|
if (seq != rcd->seq_cnt)
|
|
|
|
skip_pkt = 1;
|
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2016-02-04 06:32:57 +08:00
|
|
|
prescan_rxq(rcd, &packet);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2015-10-26 22:28:44 +08:00
|
|
|
while (last == RCV_PKT_OK) {
|
2016-02-15 12:21:52 +08:00
|
|
|
if (unlikely(dd->do_drop &&
|
|
|
|
atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
|
|
|
|
DROP_PACKET_ON)) {
|
2015-07-31 03:17:43 +08:00
|
|
|
dd->do_drop = 0;
|
|
|
|
|
|
|
|
/* On to the next packet */
|
|
|
|
packet.rhqoff += packet.rsize;
|
2016-02-15 12:19:41 +08:00
|
|
|
packet.rhf_addr = (__le32 *)rcd->rcvhdrq +
|
2015-07-31 03:17:43 +08:00
|
|
|
packet.rhqoff +
|
|
|
|
dd->rhf_offset;
|
|
|
|
packet.rhf = rhf_to_cpu(packet.rhf_addr);
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
} else if (skip_pkt) {
|
|
|
|
last = skip_rcv_packet(&packet, thread);
|
|
|
|
skip_pkt = 0;
|
2015-07-31 03:17:43 +08:00
|
|
|
} else {
|
2016-01-12 07:32:21 +08:00
|
|
|
/* Auto activate link on non-SC15 packet receive */
|
|
|
|
if (unlikely(rcd->ppd->host_link_state ==
|
|
|
|
HLS_UP_ARMED) &&
|
2016-08-09 23:19:55 +08:00
|
|
|
set_armed_to_active(rcd, &packet, dd))
|
2016-01-12 07:32:21 +08:00
|
|
|
goto bail;
|
2015-10-26 22:28:44 +08:00
|
|
|
last = process_rcv_packet(&packet, thread);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2015-11-11 13:35:19 +08:00
|
|
|
if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
|
2015-07-31 03:17:43 +08:00
|
|
|
u32 seq = rhf_rcv_seq(packet.rhf);
|
|
|
|
|
|
|
|
if (++rcd->seq_cnt > 13)
|
|
|
|
rcd->seq_cnt = 1;
|
|
|
|
if (seq != rcd->seq_cnt)
|
2015-10-26 22:28:44 +08:00
|
|
|
last = RCV_PKT_DONE;
|
2015-07-31 03:17:43 +08:00
|
|
|
if (needset) {
|
2016-02-15 12:21:52 +08:00
|
|
|
dd_dev_info(dd, "Switching to NO_DMA_RTAIL\n");
|
2015-07-31 03:17:43 +08:00
|
|
|
set_all_nodma_rtail(dd);
|
|
|
|
needset = 0;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (packet.rhqoff == hdrqtail)
|
2015-10-26 22:28:44 +08:00
|
|
|
last = RCV_PKT_DONE;
|
2015-11-11 13:35:19 +08:00
|
|
|
/*
|
|
|
|
* Control context can potentially receive an invalid
|
|
|
|
* rhf. Drop such packets.
|
|
|
|
*/
|
|
|
|
if (rcd->ctxt == HFI1_CTRL_CTXT) {
|
|
|
|
u32 seq = rhf_rcv_seq(packet.rhf);
|
|
|
|
|
|
|
|
if (++rcd->seq_cnt > 13)
|
|
|
|
rcd->seq_cnt = 1;
|
|
|
|
if (!last && (seq != rcd->seq_cnt))
|
|
|
|
skip_pkt = 1;
|
|
|
|
}
|
|
|
|
|
2015-07-31 03:17:43 +08:00
|
|
|
if (needset) {
|
|
|
|
dd_dev_info(dd,
|
|
|
|
"Switching to DMA_RTAIL\n");
|
|
|
|
set_all_dma_rtail(dd);
|
|
|
|
needset = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
process_rcv_update(last, &packet);
|
|
|
|
}
|
|
|
|
|
|
|
|
process_rcv_qp_work(&packet);
|
|
|
|
|
|
|
|
bail:
|
|
|
|
/*
|
|
|
|
* Always write head at end, and setup rcv interrupt, even
|
|
|
|
* if no packets were processed.
|
|
|
|
*/
|
|
|
|
finish_packet(&packet);
|
2015-10-26 22:28:44 +08:00
|
|
|
return last;
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2016-01-12 07:32:21 +08:00
|
|
|
/*
|
|
|
|
* We may discover in the interrupt that the hardware link state has
|
|
|
|
* changed from ARMED to ACTIVE (due to the arrival of a non-SC15 packet),
|
|
|
|
* and we need to update the driver's notion of the link state. We cannot
|
|
|
|
* run set_link_state from interrupt context, so we queue this function on
|
|
|
|
* a workqueue.
|
|
|
|
*
|
|
|
|
* We delay the regular interrupt processing until after the state changes
|
|
|
|
* so that the link will be in the correct state by the time any application
|
|
|
|
* we wake up attempts to send a reply to any message it received.
|
|
|
|
* (Subsequent receive interrupts may possibly force the wakeup before we
|
|
|
|
* update the link state.)
|
|
|
|
*
|
|
|
|
* The rcd is freed in hfi1_free_ctxtdata after hfi1_postinit_cleanup invokes
|
|
|
|
* dd->f_cleanup(dd) to disable the interrupt handler and flush workqueues,
|
|
|
|
* so we're safe from use-after-free of the rcd.
|
|
|
|
*/
|
|
|
|
void receive_interrupt_work(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
|
|
|
|
linkstate_active_work);
|
|
|
|
struct hfi1_devdata *dd = ppd->dd;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
/* Received non-SC15 packet implies neighbor_normal */
|
|
|
|
ppd->neighbor_normal = 1;
|
|
|
|
set_link_state(ppd, HLS_UP_ACTIVE);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Interrupt all kernel contexts that could have had an
|
|
|
|
* interrupt during auto activation.
|
|
|
|
*/
|
|
|
|
for (i = HFI1_CTRL_CTXT; i < dd->first_user_ctxt; i++)
|
|
|
|
force_recv_intr(dd->rcd[i]);
|
|
|
|
}
|
|
|
|
|
2015-07-31 03:17:43 +08:00
|
|
|
/*
|
|
|
|
* Convert a given MTU size to the on-wire MAD packet enumeration.
|
|
|
|
* Return -1 if the size is invalid.
|
|
|
|
*/
|
|
|
|
int mtu_to_enum(u32 mtu, int default_if_bad)
|
|
|
|
{
|
|
|
|
switch (mtu) {
|
|
|
|
case 0: return OPA_MTU_0;
|
|
|
|
case 256: return OPA_MTU_256;
|
|
|
|
case 512: return OPA_MTU_512;
|
|
|
|
case 1024: return OPA_MTU_1024;
|
|
|
|
case 2048: return OPA_MTU_2048;
|
|
|
|
case 4096: return OPA_MTU_4096;
|
|
|
|
case 8192: return OPA_MTU_8192;
|
|
|
|
case 10240: return OPA_MTU_10240;
|
|
|
|
}
|
|
|
|
return default_if_bad;
|
|
|
|
}
|
|
|
|
|
|
|
|
u16 enum_to_mtu(int mtu)
|
|
|
|
{
|
|
|
|
switch (mtu) {
|
|
|
|
case OPA_MTU_0: return 0;
|
|
|
|
case OPA_MTU_256: return 256;
|
|
|
|
case OPA_MTU_512: return 512;
|
|
|
|
case OPA_MTU_1024: return 1024;
|
|
|
|
case OPA_MTU_2048: return 2048;
|
|
|
|
case OPA_MTU_4096: return 4096;
|
|
|
|
case OPA_MTU_8192: return 8192;
|
|
|
|
case OPA_MTU_10240: return 10240;
|
|
|
|
default: return 0xffff;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* set_mtu - set the MTU
|
|
|
|
* @ppd: the per port data
|
|
|
|
*
|
|
|
|
* We can handle "any" incoming size, the issue here is whether we
|
|
|
|
* need to restrict our outgoing size. We do not deal with what happens
|
|
|
|
* to programs that are already running when the size changes.
|
|
|
|
*/
|
|
|
|
int set_mtu(struct hfi1_pportdata *ppd)
|
|
|
|
{
|
|
|
|
struct hfi1_devdata *dd = ppd->dd;
|
|
|
|
int i, drain, ret = 0, is_up = 0;
|
|
|
|
|
|
|
|
ppd->ibmtu = 0;
|
|
|
|
for (i = 0; i < ppd->vls_supported; i++)
|
|
|
|
if (ppd->ibmtu < dd->vld[i].mtu)
|
|
|
|
ppd->ibmtu = dd->vld[i].mtu;
|
|
|
|
ppd->ibmaxlen = ppd->ibmtu + lrh_max_header_bytes(ppd->dd);
|
|
|
|
|
|
|
|
mutex_lock(&ppd->hls_lock);
|
2016-02-15 12:20:15 +08:00
|
|
|
if (ppd->host_link_state == HLS_UP_INIT ||
|
|
|
|
ppd->host_link_state == HLS_UP_ARMED ||
|
|
|
|
ppd->host_link_state == HLS_UP_ACTIVE)
|
2015-07-31 03:17:43 +08:00
|
|
|
is_up = 1;
|
|
|
|
|
|
|
|
drain = !is_ax(dd) && is_up;
|
|
|
|
|
|
|
|
if (drain)
|
|
|
|
/*
|
|
|
|
* MTU is specified per-VL. To ensure that no packet gets
|
|
|
|
* stuck (due, e.g., to the MTU for the packet's VL being
|
|
|
|
* reduced), empty the per-VL FIFOs before adjusting MTU.
|
|
|
|
*/
|
|
|
|
ret = stop_drain_data_vls(dd);
|
|
|
|
|
|
|
|
if (ret) {
|
|
|
|
dd_dev_err(dd, "%s: cannot stop/drain VLs - refusing to change per-VL MTUs\n",
|
|
|
|
__func__);
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
|
|
|
|
hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_MTU, 0);
|
|
|
|
|
|
|
|
if (drain)
|
|
|
|
open_fill_data_vls(dd); /* reopen all VLs */
|
|
|
|
|
|
|
|
err:
|
|
|
|
mutex_unlock(&ppd->hls_lock);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc)
|
|
|
|
{
|
|
|
|
struct hfi1_devdata *dd = ppd->dd;
|
|
|
|
|
|
|
|
ppd->lid = lid;
|
|
|
|
ppd->lmc = lmc;
|
|
|
|
hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LIDLMC, 0);
|
|
|
|
|
2016-05-13 01:23:35 +08:00
|
|
|
dd_dev_info(dd, "port %u: got a lid: 0x%x\n", ppd->port, lid);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
void shutdown_led_override(struct hfi1_pportdata *ppd)
|
|
|
|
{
|
|
|
|
struct hfi1_devdata *dd = ppd->dd;
|
|
|
|
|
2016-02-27 05:33:28 +08:00
|
|
|
/*
|
2016-03-08 03:35:03 +08:00
|
|
|
* This pairs with the memory barrier in hfi1_start_led_override to
|
|
|
|
* ensure that we read the correct state of LED beaconing represented
|
|
|
|
* by led_override_timer_active
|
2016-02-27 05:33:28 +08:00
|
|
|
*/
|
2016-03-08 03:35:03 +08:00
|
|
|
smp_rmb();
|
2016-02-04 06:35:57 +08:00
|
|
|
if (atomic_read(&ppd->led_override_timer_active)) {
|
|
|
|
del_timer_sync(&ppd->led_override_timer);
|
|
|
|
atomic_set(&ppd->led_override_timer_active, 0);
|
2016-03-08 03:35:03 +08:00
|
|
|
/* Ensure the atomic_set is visible to all CPUs */
|
|
|
|
smp_wmb();
|
2016-02-04 06:35:57 +08:00
|
|
|
}
|
|
|
|
|
2016-03-08 03:35:03 +08:00
|
|
|
/* Hand control of the LED to the DC for normal operation */
|
|
|
|
write_csr(dd, DCC_CFG_LED_CNTRL, 0);
|
2016-02-04 06:35:57 +08:00
|
|
|
}
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
static void run_led_override(unsigned long opaque)
|
|
|
|
{
|
|
|
|
struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)opaque;
|
|
|
|
struct hfi1_devdata *dd = ppd->dd;
|
2016-02-04 06:35:57 +08:00
|
|
|
unsigned long timeout;
|
|
|
|
int phase_idx;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
if (!(dd->flags & HFI1_INITTED))
|
|
|
|
return;
|
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
phase_idx = ppd->led_override_phase & 1;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
setextled(dd, phase_idx);
|
|
|
|
|
|
|
|
timeout = ppd->led_override_vals[phase_idx];
|
2016-03-08 03:35:03 +08:00
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
/* Set up for next phase */
|
|
|
|
ppd->led_override_phase = !ppd->led_override_phase;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2016-03-08 03:35:03 +08:00
|
|
|
mod_timer(&ppd->led_override_timer, jiffies + timeout);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
/*
|
|
|
|
* To have the LED blink in a particular pattern, provide timeon and timeoff
|
2016-03-08 03:35:03 +08:00
|
|
|
* in milliseconds.
|
|
|
|
* To turn off custom blinking and return to normal operation, use
|
|
|
|
* shutdown_led_override()
|
2016-02-04 06:35:57 +08:00
|
|
|
*/
|
2016-03-08 03:35:03 +08:00
|
|
|
void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
|
|
|
|
unsigned int timeoff)
|
2015-07-31 03:17:43 +08:00
|
|
|
{
|
2016-03-08 03:35:03 +08:00
|
|
|
if (!(ppd->dd->flags & HFI1_INITTED))
|
2015-07-31 03:17:43 +08:00
|
|
|
return;
|
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
/* Convert to jiffies for direct use in timer */
|
|
|
|
ppd->led_override_vals[0] = msecs_to_jiffies(timeoff);
|
|
|
|
ppd->led_override_vals[1] = msecs_to_jiffies(timeon);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
2016-03-08 03:35:03 +08:00
|
|
|
/* Arbitrarily start from LED on phase */
|
|
|
|
ppd->led_override_phase = 1;
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If the timer has not already been started, do so. Use a "quick"
|
2016-03-08 03:35:03 +08:00
|
|
|
* timeout so the handler will be called soon to look at our request.
|
2015-07-31 03:17:43 +08:00
|
|
|
*/
|
2016-03-08 03:35:03 +08:00
|
|
|
if (!timer_pending(&ppd->led_override_timer)) {
|
2015-10-25 18:43:24 +08:00
|
|
|
setup_timer(&ppd->led_override_timer, run_led_override,
|
2016-02-15 12:21:52 +08:00
|
|
|
(unsigned long)ppd);
|
2015-07-31 03:17:43 +08:00
|
|
|
ppd->led_override_timer.expires = jiffies + 1;
|
|
|
|
add_timer(&ppd->led_override_timer);
|
2016-03-08 03:35:03 +08:00
|
|
|
atomic_set(&ppd->led_override_timer_active, 1);
|
|
|
|
/* Ensure the atomic_set is visible to all CPUs */
|
|
|
|
smp_wmb();
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* hfi1_reset_device - reset the chip if possible
|
|
|
|
* @unit: the device to reset
|
|
|
|
*
|
|
|
|
* Whether or not reset is successful, we attempt to re-initialize the chip
|
|
|
|
* (that is, much like a driver unload/reload). We clear the INITTED flag
|
|
|
|
* so that the various entry points will fail until we reinitialize. For
|
|
|
|
* now, we only allow this if no user contexts are open that use chip resources
|
|
|
|
*/
|
|
|
|
int hfi1_reset_device(int unit)
|
|
|
|
{
|
|
|
|
int ret, i;
|
|
|
|
struct hfi1_devdata *dd = hfi1_lookup(unit);
|
|
|
|
struct hfi1_pportdata *ppd;
|
|
|
|
unsigned long flags;
|
|
|
|
int pidx;
|
|
|
|
|
|
|
|
if (!dd) {
|
|
|
|
ret = -ENODEV;
|
|
|
|
goto bail;
|
|
|
|
}
|
|
|
|
|
|
|
|
dd_dev_info(dd, "Reset on unit %u requested\n", unit);
|
|
|
|
|
|
|
|
if (!dd->kregbase || !(dd->flags & HFI1_PRESENT)) {
|
|
|
|
dd_dev_info(dd,
|
2016-02-15 12:21:52 +08:00
|
|
|
"Invalid unit number %u or not initialized or not present\n",
|
|
|
|
unit);
|
2015-07-31 03:17:43 +08:00
|
|
|
ret = -ENXIO;
|
|
|
|
goto bail;
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_lock_irqsave(&dd->uctxt_lock, flags);
|
|
|
|
if (dd->rcd)
|
|
|
|
for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
|
|
|
|
if (!dd->rcd[i] || !dd->rcd[i]->cnt)
|
|
|
|
continue;
|
|
|
|
spin_unlock_irqrestore(&dd->uctxt_lock, flags);
|
|
|
|
ret = -EBUSY;
|
|
|
|
goto bail;
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&dd->uctxt_lock, flags);
|
|
|
|
|
|
|
|
for (pidx = 0; pidx < dd->num_pports; ++pidx) {
|
|
|
|
ppd = dd->pport + pidx;
|
|
|
|
|
2016-02-04 06:35:57 +08:00
|
|
|
shutdown_led_override(ppd);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
if (dd->flags & HFI1_HAS_SEND_DMA)
|
|
|
|
sdma_exit(dd);
|
|
|
|
|
|
|
|
hfi1_reset_cpu_counters(dd);
|
|
|
|
|
|
|
|
ret = hfi1_init(dd, 1);
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
dd_dev_err(dd,
|
2016-02-15 12:21:52 +08:00
|
|
|
"Reinitialize unit %u after reset failed with %d\n",
|
|
|
|
unit, ret);
|
2015-07-31 03:17:43 +08:00
|
|
|
else
|
|
|
|
dd_dev_info(dd, "Reinitialized unit %u after resetting\n",
|
2016-02-15 12:21:52 +08:00
|
|
|
unit);
|
2015-07-31 03:17:43 +08:00
|
|
|
|
|
|
|
bail:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
void handle_eflags(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
struct hfi1_ctxtdata *rcd = packet->rcd;
|
|
|
|
u32 rte = rhf_rcv_type_err(packet->rhf);
|
|
|
|
|
|
|
|
rcv_hdrerr(rcd, rcd->ppd, packet);
|
2015-11-07 09:07:03 +08:00
|
|
|
if (rhf_err_flags(packet->rhf))
|
|
|
|
dd_dev_err(rcd->dd,
|
|
|
|
"receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n",
|
|
|
|
rcd->ctxt, packet->rhf,
|
|
|
|
packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "",
|
|
|
|
packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "",
|
|
|
|
packet->rhf & RHF_DC_ERR ? "dc " : "",
|
|
|
|
packet->rhf & RHF_TID_ERR ? "tid " : "",
|
|
|
|
packet->rhf & RHF_LEN_ERR ? "len " : "",
|
|
|
|
packet->rhf & RHF_ECC_ERR ? "ecc " : "",
|
|
|
|
packet->rhf & RHF_VCRC_ERR ? "vcrc " : "",
|
|
|
|
packet->rhf & RHF_ICRC_ERR ? "icrc " : "",
|
|
|
|
rte);
|
2015-07-31 03:17:43 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The following functions are called by the interrupt handler. They are type
|
|
|
|
* specific handlers for each packet type.
|
|
|
|
*/
|
|
|
|
int process_receive_ib(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
trace_hfi1_rcvhdr(packet->rcd->ppd->dd,
|
|
|
|
packet->rcd->ctxt,
|
|
|
|
rhf_err_flags(packet->rhf),
|
|
|
|
RHF_RCV_TYPE_IB,
|
|
|
|
packet->hlen,
|
|
|
|
packet->tlen,
|
|
|
|
packet->updegr,
|
|
|
|
rhf_egr_index(packet->rhf));
|
|
|
|
|
|
|
|
if (unlikely(rhf_err_flags(packet->rhf))) {
|
|
|
|
handle_eflags(packet);
|
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
hfi1_ib_rcv(packet);
|
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
int process_receive_bypass(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
if (unlikely(rhf_err_flags(packet->rhf)))
|
|
|
|
handle_eflags(packet);
|
|
|
|
|
|
|
|
dd_dev_err(packet->rcd->dd,
|
2016-02-15 12:21:52 +08:00
|
|
|
"Bypass packets are not supported in normal operation. Dropping\n");
|
2016-07-02 07:01:22 +08:00
|
|
|
incr_cntr64(&packet->rcd->dd->sw_rcv_bypass_packet_errors);
|
2015-07-31 03:17:43 +08:00
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
int process_receive_error(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
handle_eflags(packet);
|
|
|
|
|
|
|
|
if (unlikely(rhf_err_flags(packet->rhf)))
|
|
|
|
dd_dev_err(packet->rcd->dd,
|
|
|
|
"Unhandled error packet received. Dropping.\n");
|
|
|
|
|
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
int kdeth_process_expected(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
if (unlikely(rhf_err_flags(packet->rhf)))
|
|
|
|
handle_eflags(packet);
|
|
|
|
|
|
|
|
dd_dev_err(packet->rcd->dd,
|
|
|
|
"Unhandled expected packet received. Dropping.\n");
|
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
int kdeth_process_eager(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
if (unlikely(rhf_err_flags(packet->rhf)))
|
|
|
|
handle_eflags(packet);
|
|
|
|
|
|
|
|
dd_dev_err(packet->rcd->dd,
|
|
|
|
"Unhandled eager packet received. Dropping.\n");
|
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
int process_receive_invalid(struct hfi1_packet *packet)
|
|
|
|
{
|
|
|
|
dd_dev_err(packet->rcd->dd, "Invalid packet type %d. Dropping\n",
|
2016-02-15 12:21:52 +08:00
|
|
|
rhf_rcv_type(packet->rhf));
|
2015-07-31 03:17:43 +08:00
|
|
|
return RHF_RCV_CONTINUE;
|
|
|
|
}
|