linux/drivers/infiniband/hw/hfi1/qp.c

978 lines
26 KiB
C
Raw Normal View History

/*
* Copyright(c) 2015 - 2019 Intel Corporation.
*
* 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/err.h>
#include <linux/vmalloc.h>
#include <linux/hash.h>
#include <linux/module.h>
#include <linux/seq_file.h>
#include <rdma/rdma_vt.h>
#include <rdma/rdmavt_qp.h>
#include <rdma/ib_verbs.h>
#include "hfi.h"
#include "qp.h"
#include "trace.h"
#include "verbs_txreq.h"
unsigned int hfi1_qp_table_size = 256;
module_param_named(qp_table_size, hfi1_qp_table_size, uint, S_IRUGO);
MODULE_PARM_DESC(qp_table_size, "QP table size");
static void flush_tx_list(struct rvt_qp *qp);
static int iowait_sleep(
struct sdma_engine *sde,
struct iowait_work *wait,
struct sdma_txreq *stx,
unsigned int seq,
bool pkts_sent);
static void iowait_wakeup(struct iowait *wait, int reason);
static void iowait_sdma_drained(struct iowait *wait);
static void qp_pio_drain(struct rvt_qp *qp);
const struct rvt_operation_params hfi1_post_parms[RVT_OPERATION_MAX] = {
[IB_WR_RDMA_WRITE] = {
.length = sizeof(struct ib_rdma_wr),
.qpt_support = BIT(IB_QPT_UC) | BIT(IB_QPT_RC),
},
[IB_WR_RDMA_READ] = {
.length = sizeof(struct ib_rdma_wr),
.qpt_support = BIT(IB_QPT_RC),
.flags = RVT_OPERATION_ATOMIC,
},
[IB_WR_ATOMIC_CMP_AND_SWP] = {
.length = sizeof(struct ib_atomic_wr),
.qpt_support = BIT(IB_QPT_RC),
.flags = RVT_OPERATION_ATOMIC | RVT_OPERATION_ATOMIC_SGE,
},
[IB_WR_ATOMIC_FETCH_AND_ADD] = {
.length = sizeof(struct ib_atomic_wr),
.qpt_support = BIT(IB_QPT_RC),
.flags = RVT_OPERATION_ATOMIC | RVT_OPERATION_ATOMIC_SGE,
},
[IB_WR_RDMA_WRITE_WITH_IMM] = {
.length = sizeof(struct ib_rdma_wr),
.qpt_support = BIT(IB_QPT_UC) | BIT(IB_QPT_RC),
},
[IB_WR_SEND] = {
.length = sizeof(struct ib_send_wr),
.qpt_support = BIT(IB_QPT_UD) | BIT(IB_QPT_SMI) | BIT(IB_QPT_GSI) |
BIT(IB_QPT_UC) | BIT(IB_QPT_RC),
},
[IB_WR_SEND_WITH_IMM] = {
.length = sizeof(struct ib_send_wr),
.qpt_support = BIT(IB_QPT_UD) | BIT(IB_QPT_SMI) | BIT(IB_QPT_GSI) |
BIT(IB_QPT_UC) | BIT(IB_QPT_RC),
},
[IB_WR_REG_MR] = {
.length = sizeof(struct ib_reg_wr),
.qpt_support = BIT(IB_QPT_UC) | BIT(IB_QPT_RC),
.flags = RVT_OPERATION_LOCAL,
},
[IB_WR_LOCAL_INV] = {
.length = sizeof(struct ib_send_wr),
.qpt_support = BIT(IB_QPT_UC) | BIT(IB_QPT_RC),
.flags = RVT_OPERATION_LOCAL,
},
[IB_WR_SEND_WITH_INV] = {
.length = sizeof(struct ib_send_wr),
.qpt_support = BIT(IB_QPT_RC),
},
[IB_WR_OPFN] = {
.length = sizeof(struct ib_atomic_wr),
.qpt_support = BIT(IB_QPT_RC),
.flags = RVT_OPERATION_USE_RESERVE,
},
[IB_WR_TID_RDMA_WRITE] = {
.length = sizeof(struct ib_rdma_wr),
.qpt_support = BIT(IB_QPT_RC),
.flags = RVT_OPERATION_IGN_RNR_CNT,
},
};
static void flush_list_head(struct list_head *l)
{
while (!list_empty(l)) {
struct sdma_txreq *tx;
tx = list_first_entry(
l,
struct sdma_txreq,
list);
list_del_init(&tx->list);
hfi1_put_txreq(
container_of(tx, struct verbs_txreq, txreq));
}
}
static void flush_tx_list(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
flush_list_head(&iowait_get_ib_work(&priv->s_iowait)->tx_head);
flush_list_head(&iowait_get_tid_work(&priv->s_iowait)->tx_head);
}
static void flush_iowait(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
unsigned long flags;
seqlock_t *lock = priv->s_iowait.lock;
if (!lock)
return;
write_seqlock_irqsave(lock, flags);
if (!list_empty(&priv->s_iowait.list)) {
list_del_init(&priv->s_iowait.list);
priv->s_iowait.lock = NULL;
rvt_put_qp(qp);
}
write_sequnlock_irqrestore(lock, flags);
}
static inline int opa_mtu_enum_to_int(int mtu)
{
switch (mtu) {
case OPA_MTU_8192: return 8192;
case OPA_MTU_10240: return 10240;
default: return -1;
}
}
/**
* This function is what we would push to the core layer if we wanted to be a
* "first class citizen". Instead we hide this here and rely on Verbs ULPs
* to blindly pass the MTU enum value from the PathRecord to us.
*/
static inline int verbs_mtu_enum_to_int(struct ib_device *dev, enum ib_mtu mtu)
{
int val;
/* Constraining 10KB packets to 8KB packets */
if (mtu == (enum ib_mtu)OPA_MTU_10240)
mtu = OPA_MTU_8192;
val = opa_mtu_enum_to_int((int)mtu);
if (val > 0)
return val;
return ib_mtu_enum_to_int(mtu);
}
int hfi1_check_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
int attr_mask, struct ib_udata *udata)
{
struct ib_qp *ibqp = &qp->ibqp;
struct hfi1_ibdev *dev = to_idev(ibqp->device);
struct hfi1_devdata *dd = dd_from_dev(dev);
u8 sc;
if (attr_mask & IB_QP_AV) {
sc = ah_to_sc(ibqp->device, &attr->ah_attr);
if (sc == 0xf)
return -EINVAL;
if (!qp_to_sdma_engine(qp, sc) &&
dd->flags & HFI1_HAS_SEND_DMA)
return -EINVAL;
if (!qp_to_send_context(qp, sc))
return -EINVAL;
}
if (attr_mask & IB_QP_ALT_PATH) {
sc = ah_to_sc(ibqp->device, &attr->alt_ah_attr);
if (sc == 0xf)
return -EINVAL;
if (!qp_to_sdma_engine(qp, sc) &&
dd->flags & HFI1_HAS_SEND_DMA)
return -EINVAL;
if (!qp_to_send_context(qp, sc))
return -EINVAL;
}
return 0;
}
/*
* qp_set_16b - Set the hdr_type based on whether the slid or the
* dlid in the connection is extended. Only applicable for RC and UC
* QPs. UD QPs determine this on the fly from the ah in the wqe
*/
static inline void qp_set_16b(struct rvt_qp *qp)
{
struct hfi1_pportdata *ppd;
struct hfi1_ibport *ibp;
struct hfi1_qp_priv *priv = qp->priv;
/* Update ah_attr to account for extended LIDs */
hfi1_update_ah_attr(qp->ibqp.device, &qp->remote_ah_attr);
/* Create 32 bit LIDs */
hfi1_make_opa_lid(&qp->remote_ah_attr);
if (!(rdma_ah_get_ah_flags(&qp->remote_ah_attr) & IB_AH_GRH))
return;
ibp = to_iport(qp->ibqp.device, qp->port_num);
ppd = ppd_from_ibp(ibp);
priv->hdr_type = hfi1_get_hdr_type(ppd->lid, &qp->remote_ah_attr);
}
void hfi1_modify_qp(struct rvt_qp *qp, struct ib_qp_attr *attr,
int attr_mask, struct ib_udata *udata)
{
struct ib_qp *ibqp = &qp->ibqp;
struct hfi1_qp_priv *priv = qp->priv;
if (attr_mask & IB_QP_AV) {
priv->s_sc = ah_to_sc(ibqp->device, &qp->remote_ah_attr);
priv->s_sde = qp_to_sdma_engine(qp, priv->s_sc);
IB/hfi1: Fix panic in adaptive pio The following panic occurs while running ib_send_bw -a with adaptive pio turned on: [ 8551.143596] BUG: unable to handle kernel NULL pointer dereference at (null) [ 8551.152986] IP: [<ffffffffa0902a94>] pio_wait.isra.21+0x34/0x190 [hfi1] [ 8551.160926] PGD 80db21067 PUD 80bb45067 PMD 0 [ 8551.166431] Oops: 0000 [#1] SMP [ 8551.276725] task: ffff880816bf15c0 ti: ffff880812ac0000 task.ti: ffff880812ac0000 [ 8551.285705] RIP: 0010:[<ffffffffa0902a94>] pio_wait.isra.21+0x34/0x190 [hfi1] [ 8551.296462] RSP: 0018:ffff880812ac3b58 EFLAGS: 00010282 [ 8551.303029] RAX: 000000000000002d RBX: 0000000000000000 RCX: 0000000000000800 [ 8551.311633] RDX: ffff880812ac3c08 RSI: 0000000000000000 RDI: ffff8800b6665e40 [ 8551.320228] RBP: ffff880812ac3ba0 R08: 0000000000001000 R09: ffffffffa09039a0 [ 8551.328820] R10: ffff880817a0c000 R11: 0000000000000000 R12: ffff8800b6665e40 [ 8551.337406] R13: ffff880817a0c000 R14: ffff8800b6665800 R15: ffff8800b6665e40 [ 8551.355640] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 8551.362674] CR2: 0000000000000000 CR3: 000000080abe8000 CR4: 00000000001406e0 [ 8551.371262] Stack: [ 8551.374119] ffff880812ac3bf0 ffff88080cf54010 ffff880800000800 ffff880812ac3c08 [ 8551.383036] ffff8800b6665800 ffff8800b6665e40 0000000000000202 ffffffffa08e7b80 [ 8551.391941] 00000001007de431 ffff880812ac3bc8 ffffffffa0904645 ffff8800b6665800 [ 8551.400859] Call Trace: [ 8551.404214] [<ffffffffa08e7b80>] ? hfi1_del_timers_sync+0x30/0x30 [hfi1] [ 8551.412417] [<ffffffffa0904645>] hfi1_verbs_send+0x215/0x330 [hfi1] [ 8551.420154] [<ffffffffa08ec126>] hfi1_do_send+0x166/0x350 [hfi1] [ 8551.427618] [<ffffffffa055a533>] rvt_post_send+0x533/0x6a0 [rdmavt] [ 8551.435367] [<ffffffffa050760f>] ib_uverbs_post_send+0x30f/0x530 [ib_uverbs] [ 8551.443999] [<ffffffffa0501367>] ib_uverbs_write+0x117/0x380 [ib_uverbs] [ 8551.452269] [<ffffffff815810ab>] ? sock_recvmsg+0x3b/0x50 [ 8551.459071] [<ffffffff81581152>] ? sock_read_iter+0x92/0xe0 [ 8551.466068] [<ffffffff81212857>] __vfs_write+0x37/0x100 [ 8551.472692] [<ffffffff81213532>] ? rw_verify_area+0x52/0xd0 [ 8551.479682] [<ffffffff81213782>] vfs_write+0xa2/0x1a0 [ 8551.486089] [<ffffffff81003176>] ? do_audit_syscall_entry+0x66/0x70 [ 8551.493891] [<ffffffff812146c5>] SyS_write+0x55/0xc0 [ 8551.500220] [<ffffffff816ae0ee>] entry_SYSCALL_64_fastpath+0x12/0x71 [ 8551.531284] RIP [<ffffffffa0902a94>] pio_wait.isra.21+0x34/0x190 [hfi1] [ 8551.539508] RSP <ffff880812ac3b58> [ 8551.544110] CR2: 0000000000000000 The priv s_sendcontext pointer was not setup properly. Fix with this patch by using the s_sendcontext and eliminating its send engine use. Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-03-08 03:35:35 +08:00
priv->s_sendcontext = qp_to_send_context(qp, priv->s_sc);
qp_set_16b(qp);
}
if (attr_mask & IB_QP_PATH_MIG_STATE &&
attr->path_mig_state == IB_MIG_MIGRATED &&
qp->s_mig_state == IB_MIG_ARMED) {
qp->s_flags |= HFI1_S_AHG_CLEAR;
priv->s_sc = ah_to_sc(ibqp->device, &qp->remote_ah_attr);
priv->s_sde = qp_to_sdma_engine(qp, priv->s_sc);
IB/hfi1: Fix panic in adaptive pio The following panic occurs while running ib_send_bw -a with adaptive pio turned on: [ 8551.143596] BUG: unable to handle kernel NULL pointer dereference at (null) [ 8551.152986] IP: [<ffffffffa0902a94>] pio_wait.isra.21+0x34/0x190 [hfi1] [ 8551.160926] PGD 80db21067 PUD 80bb45067 PMD 0 [ 8551.166431] Oops: 0000 [#1] SMP [ 8551.276725] task: ffff880816bf15c0 ti: ffff880812ac0000 task.ti: ffff880812ac0000 [ 8551.285705] RIP: 0010:[<ffffffffa0902a94>] pio_wait.isra.21+0x34/0x190 [hfi1] [ 8551.296462] RSP: 0018:ffff880812ac3b58 EFLAGS: 00010282 [ 8551.303029] RAX: 000000000000002d RBX: 0000000000000000 RCX: 0000000000000800 [ 8551.311633] RDX: ffff880812ac3c08 RSI: 0000000000000000 RDI: ffff8800b6665e40 [ 8551.320228] RBP: ffff880812ac3ba0 R08: 0000000000001000 R09: ffffffffa09039a0 [ 8551.328820] R10: ffff880817a0c000 R11: 0000000000000000 R12: ffff8800b6665e40 [ 8551.337406] R13: ffff880817a0c000 R14: ffff8800b6665800 R15: ffff8800b6665e40 [ 8551.355640] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 8551.362674] CR2: 0000000000000000 CR3: 000000080abe8000 CR4: 00000000001406e0 [ 8551.371262] Stack: [ 8551.374119] ffff880812ac3bf0 ffff88080cf54010 ffff880800000800 ffff880812ac3c08 [ 8551.383036] ffff8800b6665800 ffff8800b6665e40 0000000000000202 ffffffffa08e7b80 [ 8551.391941] 00000001007de431 ffff880812ac3bc8 ffffffffa0904645 ffff8800b6665800 [ 8551.400859] Call Trace: [ 8551.404214] [<ffffffffa08e7b80>] ? hfi1_del_timers_sync+0x30/0x30 [hfi1] [ 8551.412417] [<ffffffffa0904645>] hfi1_verbs_send+0x215/0x330 [hfi1] [ 8551.420154] [<ffffffffa08ec126>] hfi1_do_send+0x166/0x350 [hfi1] [ 8551.427618] [<ffffffffa055a533>] rvt_post_send+0x533/0x6a0 [rdmavt] [ 8551.435367] [<ffffffffa050760f>] ib_uverbs_post_send+0x30f/0x530 [ib_uverbs] [ 8551.443999] [<ffffffffa0501367>] ib_uverbs_write+0x117/0x380 [ib_uverbs] [ 8551.452269] [<ffffffff815810ab>] ? sock_recvmsg+0x3b/0x50 [ 8551.459071] [<ffffffff81581152>] ? sock_read_iter+0x92/0xe0 [ 8551.466068] [<ffffffff81212857>] __vfs_write+0x37/0x100 [ 8551.472692] [<ffffffff81213532>] ? rw_verify_area+0x52/0xd0 [ 8551.479682] [<ffffffff81213782>] vfs_write+0xa2/0x1a0 [ 8551.486089] [<ffffffff81003176>] ? do_audit_syscall_entry+0x66/0x70 [ 8551.493891] [<ffffffff812146c5>] SyS_write+0x55/0xc0 [ 8551.500220] [<ffffffff816ae0ee>] entry_SYSCALL_64_fastpath+0x12/0x71 [ 8551.531284] RIP [<ffffffffa0902a94>] pio_wait.isra.21+0x34/0x190 [hfi1] [ 8551.539508] RSP <ffff880812ac3b58> [ 8551.544110] CR2: 0000000000000000 The priv s_sendcontext pointer was not setup properly. Fix with this patch by using the s_sendcontext and eliminating its send engine use. Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-03-08 03:35:35 +08:00
priv->s_sendcontext = qp_to_send_context(qp, priv->s_sc);
qp_set_16b(qp);
}
opfn_qp_init(qp, attr, attr_mask);
}
/**
* hfi1_setup_wqe - set up the wqe
* @qp - The qp
* @wqe - The built wqe
* @call_send - Determine if the send should be posted or scheduled.
*
* Perform setup of the wqe. This is called
* prior to inserting the wqe into the ring but after
* the wqe has been setup by RDMAVT. This function
* allows the driver the opportunity to perform
* validation and additional setup of the wqe.
*
* Returns 0 on success, -EINVAL on failure
*
*/
int hfi1_setup_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe, bool *call_send)
{
struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
struct rvt_ah *ah;
struct hfi1_pportdata *ppd;
struct hfi1_devdata *dd;
switch (qp->ibqp.qp_type) {
case IB_QPT_RC:
hfi1_setup_tid_rdma_wqe(qp, wqe);
/* fall through */
case IB_QPT_UC:
if (wqe->length > 0x80000000U)
return -EINVAL;
if (wqe->length > qp->pmtu)
*call_send = false;
break;
case IB_QPT_SMI:
/*
* SM packets should exclusively use VL15 and their SL is
* ignored (IBTA v1.3, Section 3.5.8.2). Therefore, when ah
* is created, SL is 0 in most cases and as a result some
* fields (vl and pmtu) in ah may not be set correctly,
* depending on the SL2SC and SC2VL tables at the time.
*/
ppd = ppd_from_ibp(ibp);
dd = dd_from_ppd(ppd);
if (wqe->length > dd->vld[15].mtu)
return -EINVAL;
break;
case IB_QPT_GSI:
case IB_QPT_UD:
ah = rvt_get_swqe_ah(wqe);
if (wqe->length > (1 << ah->log_pmtu))
return -EINVAL;
if (ibp->sl_to_sc[rdma_ah_get_sl(&ah->attr)] == 0xf)
return -EINVAL;
default:
break;
}
/*
* System latency between send and schedule is large enough that
* forcing call_send to true for piothreshold packets is necessary.
*/
if (wqe->length <= piothreshold)
*call_send = true;
return 0;
}
/**
* _hfi1_schedule_send - schedule progress
* @qp: the QP
*
* This schedules qp progress w/o regard to the s_flags.
*
* It is only used in the post send, which doesn't hold
* the s_lock.
*/
bool _hfi1_schedule_send(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
struct hfi1_ibport *ibp =
to_iport(qp->ibqp.device, qp->port_num);
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);
return iowait_schedule(&priv->s_iowait, ppd->hfi1_wq,
priv->s_sde ?
priv->s_sde->cpu :
cpumask_first(cpumask_of_node(dd->node)));
}
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
static void qp_pio_drain(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
if (!priv->s_sendcontext)
return;
while (iowait_pio_pending(&priv->s_iowait)) {
write_seqlock_irq(&priv->s_sendcontext->waitlock);
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
hfi1_sc_wantpiobuf_intr(priv->s_sendcontext, 1);
write_sequnlock_irq(&priv->s_sendcontext->waitlock);
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
iowait_pio_drain(&priv->s_iowait);
write_seqlock_irq(&priv->s_sendcontext->waitlock);
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
hfi1_sc_wantpiobuf_intr(priv->s_sendcontext, 0);
write_sequnlock_irq(&priv->s_sendcontext->waitlock);
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
}
}
/**
* hfi1_schedule_send - schedule progress
* @qp: the QP
*
* This schedules qp progress and caller should hold
* the s_lock.
* @return true if the first leg is scheduled;
* false if the first leg is not scheduled.
*/
bool hfi1_schedule_send(struct rvt_qp *qp)
{
lockdep_assert_held(&qp->s_lock);
if (hfi1_send_ok(qp)) {
_hfi1_schedule_send(qp);
return true;
}
if (qp->s_flags & HFI1_S_ANY_WAIT_IO)
iowait_set_flag(&((struct hfi1_qp_priv *)qp->priv)->s_iowait,
IOWAIT_PENDING_IB);
return false;
}
static void hfi1_qp_schedule(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
bool ret;
if (iowait_flag_set(&priv->s_iowait, IOWAIT_PENDING_IB)) {
ret = hfi1_schedule_send(qp);
if (ret)
iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB);
}
if (iowait_flag_set(&priv->s_iowait, IOWAIT_PENDING_TID)) {
ret = hfi1_schedule_tid_send(qp);
if (ret)
iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
}
}
void hfi1_qp_wakeup(struct rvt_qp *qp, u32 flag)
{
unsigned long flags;
spin_lock_irqsave(&qp->s_lock, flags);
if (qp->s_flags & flag) {
qp->s_flags &= ~flag;
trace_hfi1_qpwakeup(qp, flag);
hfi1_qp_schedule(qp);
}
spin_unlock_irqrestore(&qp->s_lock, flags);
/* Notify hfi1_destroy_qp() if it is waiting. */
rvt_put_qp(qp);
}
void hfi1_qp_unbusy(struct rvt_qp *qp, struct iowait_work *wait)
{
struct hfi1_qp_priv *priv = qp->priv;
if (iowait_set_work_flag(wait) == IOWAIT_IB_SE) {
qp->s_flags &= ~RVT_S_BUSY;
/*
* If we are sending a first-leg packet from the second leg,
* we need to clear the busy flag from priv->s_flags to
* avoid a race condition when the qp wakes up before
* the call to hfi1_verbs_send() returns to the second
* leg. In that case, the second leg will terminate without
* being re-scheduled, resulting in failure to send TID RDMA
* WRITE DATA and TID RDMA ACK packets.
*/
if (priv->s_flags & HFI1_S_TID_BUSY_SET) {
priv->s_flags &= ~(HFI1_S_TID_BUSY_SET |
RVT_S_BUSY);
iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
}
} else {
priv->s_flags &= ~RVT_S_BUSY;
}
}
static int iowait_sleep(
struct sdma_engine *sde,
struct iowait_work *wait,
struct sdma_txreq *stx,
uint seq,
bool pkts_sent)
{
struct verbs_txreq *tx = container_of(stx, struct verbs_txreq, txreq);
struct rvt_qp *qp;
struct hfi1_qp_priv *priv;
unsigned long flags;
int ret = 0;
qp = tx->qp;
priv = qp->priv;
spin_lock_irqsave(&qp->s_lock, flags);
if (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) {
/*
* If we couldn't queue the DMA request, save the info
* and try again later rather than destroying the
* buffer and undoing the side effects of the copy.
*/
/* Make a common routine? */
list_add_tail(&stx->list, &wait->tx_head);
write_seqlock(&sde->waitlock);
if (sdma_progress(sde, seq, stx))
goto eagain;
if (list_empty(&priv->s_iowait.list)) {
struct hfi1_ibport *ibp =
to_iport(qp->ibqp.device, qp->port_num);
ibp->rvp.n_dmawait++;
qp->s_flags |= RVT_S_WAIT_DMA_DESC;
iowait_get_priority(&priv->s_iowait);
iowait_queue(pkts_sent, &priv->s_iowait,
&sde->dmawait);
priv->s_iowait.lock = &sde->waitlock;
trace_hfi1_qpsleep(qp, RVT_S_WAIT_DMA_DESC);
rvt_get_qp(qp);
}
write_sequnlock(&sde->waitlock);
hfi1_qp_unbusy(qp, wait);
spin_unlock_irqrestore(&qp->s_lock, flags);
ret = -EBUSY;
} else {
spin_unlock_irqrestore(&qp->s_lock, flags);
hfi1_put_txreq(tx);
}
return ret;
eagain:
write_sequnlock(&sde->waitlock);
spin_unlock_irqrestore(&qp->s_lock, flags);
list_del_init(&stx->list);
return -EAGAIN;
}
static void iowait_wakeup(struct iowait *wait, int reason)
{
struct rvt_qp *qp = iowait_to_qp(wait);
WARN_ON(reason != SDMA_AVAIL_REASON);
hfi1_qp_wakeup(qp, RVT_S_WAIT_DMA_DESC);
}
static void iowait_sdma_drained(struct iowait *wait)
{
struct rvt_qp *qp = iowait_to_qp(wait);
unsigned long flags;
/*
* This happens when the send engine notes
* a QP in the error state and cannot
* do the flush work until that QP's
* sdma work has finished.
*/
spin_lock_irqsave(&qp->s_lock, flags);
if (qp->s_flags & RVT_S_WAIT_DMA) {
qp->s_flags &= ~RVT_S_WAIT_DMA;
hfi1_schedule_send(qp);
}
spin_unlock_irqrestore(&qp->s_lock, flags);
}
static void hfi1_init_priority(struct iowait *w)
{
struct rvt_qp *qp = iowait_to_qp(w);
struct hfi1_qp_priv *priv = qp->priv;
if (qp->s_flags & RVT_S_ACK_PENDING)
w->priority++;
if (priv->s_flags & RVT_S_ACK_PENDING)
w->priority++;
}
/**
* qp_to_sdma_engine - map a qp to a send engine
* @qp: the QP
* @sc5: the 5 bit sc
*
* Return:
* A send engine for the qp or NULL for SMI type qp.
*/
struct sdma_engine *qp_to_sdma_engine(struct rvt_qp *qp, u8 sc5)
{
struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);
struct sdma_engine *sde;
if (!(dd->flags & HFI1_HAS_SEND_DMA))
return NULL;
switch (qp->ibqp.qp_type) {
case IB_QPT_SMI:
return NULL;
default:
break;
}
sde = sdma_select_engine_sc(dd, qp->ibqp.qp_num >> dd->qos_shift, sc5);
return sde;
}
/*
* qp_to_send_context - map a qp to a send context
* @qp: the QP
* @sc5: the 5 bit sc
*
* Return:
* A send context for the qp
*/
struct send_context *qp_to_send_context(struct rvt_qp *qp, u8 sc5)
{
struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);
switch (qp->ibqp.qp_type) {
case IB_QPT_SMI:
/* SMA packets to VL15 */
return dd->vld[15].sc;
default:
break;
}
return pio_select_send_context_sc(dd, qp->ibqp.qp_num >> dd->qos_shift,
sc5);
}
static const char * const qp_type_str[] = {
"SMI", "GSI", "RC", "UC", "UD",
};
static int qp_idle(struct rvt_qp *qp)
{
return
qp->s_last == qp->s_acked &&
qp->s_acked == qp->s_cur &&
qp->s_cur == qp->s_tail &&
qp->s_tail == qp->s_head;
}
/**
* qp_iter_print - print the qp information to seq_file
* @s: the seq_file to emit the qp information on
* @iter: the iterator for the qp hash list
*/
void qp_iter_print(struct seq_file *s, struct rvt_qp_iter *iter)
{
struct rvt_swqe *wqe;
struct rvt_qp *qp = iter->qp;
struct hfi1_qp_priv *priv = qp->priv;
struct sdma_engine *sde;
struct send_context *send_context;
struct rvt_ack_entry *e = NULL;
struct rvt_srq *srq = qp->ibqp.srq ?
ibsrq_to_rvtsrq(qp->ibqp.srq) : NULL;
sde = qp_to_sdma_engine(qp, priv->s_sc);
wqe = rvt_get_swqe_ptr(qp, qp->s_last);
send_context = qp_to_send_context(qp, priv->s_sc);
if (qp->s_ack_queue)
e = &qp->s_ack_queue[qp->s_tail_ack_queue];
seq_printf(s,
"N %d %s QP %x R %u %s %u %u f=%x %u %u %u %u %u %u SPSN %x %x %x %x %x RPSN %x S(%u %u %u %u %u %u %u) R(%u %u %u) RQP %x LID %x SL %u MTU %u %u %u %u %u SDE %p,%u SC %p,%u SCQ %u %u PID %d OS %x %x E %x %x %x RNR %d %s %d\n",
iter->n,
qp_idle(qp) ? "I" : "B",
qp->ibqp.qp_num,
atomic_read(&qp->refcount),
qp_type_str[qp->ibqp.qp_type],
qp->state,
wqe ? wqe->wr.opcode : 0,
qp->s_flags,
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
iowait_sdma_pending(&priv->s_iowait),
iowait_pio_pending(&priv->s_iowait),
!list_empty(&priv->s_iowait.list),
qp->timeout,
wqe ? wqe->ssn : 0,
qp->s_lsn,
qp->s_last_psn,
qp->s_psn, qp->s_next_psn,
qp->s_sending_psn, qp->s_sending_hpsn,
qp->r_psn,
qp->s_last, qp->s_acked, qp->s_cur,
qp->s_tail, qp->s_head, qp->s_size,
qp->s_avail,
/* ack_queue ring pointers, size */
qp->s_tail_ack_queue, qp->r_head_ack_queue,
rvt_max_atomic(&to_idev(qp->ibqp.device)->rdi),
/* remote QP info */
qp->remote_qpn,
rdma_ah_get_dlid(&qp->remote_ah_attr),
rdma_ah_get_sl(&qp->remote_ah_attr),
qp->pmtu,
qp->s_retry,
qp->s_retry_cnt,
qp->s_rnr_retry_cnt,
qp->s_rnr_retry,
sde,
sde ? sde->this_idx : 0,
send_context,
send_context ? send_context->sw_index : 0,
ib_cq_head(qp->ibqp.send_cq),
ib_cq_tail(qp->ibqp.send_cq),
qp->pid,
qp->s_state,
qp->s_ack_state,
/* ack queue information */
e ? e->opcode : 0,
e ? e->psn : 0,
e ? e->lpsn : 0,
qp->r_min_rnr_timer,
srq ? "SRQ" : "RQ",
srq ? srq->rq.size : qp->r_rq.size
);
}
void *qp_priv_alloc(struct rvt_dev_info *rdi, struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv;
priv = kzalloc_node(sizeof(*priv), GFP_KERNEL, rdi->dparms.node);
if (!priv)
return ERR_PTR(-ENOMEM);
priv->owner = qp;
priv->s_ahg = kzalloc_node(sizeof(*priv->s_ahg), GFP_KERNEL,
rdi->dparms.node);
if (!priv->s_ahg) {
kfree(priv);
return ERR_PTR(-ENOMEM);
}
iowait_init(
&priv->s_iowait,
1,
_hfi1_do_send,
_hfi1_do_tid_send,
iowait_sleep,
iowait_wakeup,
iowait_sdma_drained,
hfi1_init_priority);
/* Init to a value to start the running average correctly */
priv->s_running_pkt_size = piothreshold / 2;
return priv;
}
void qp_priv_free(struct rvt_dev_info *rdi, struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
hfi1_qp_priv_tid_free(rdi, qp);
kfree(priv->s_ahg);
kfree(priv);
}
unsigned free_all_qps(struct rvt_dev_info *rdi)
{
struct hfi1_ibdev *verbs_dev = container_of(rdi,
struct hfi1_ibdev,
rdi);
struct hfi1_devdata *dd = container_of(verbs_dev,
struct hfi1_devdata,
verbs_dev);
int n;
unsigned qp_inuse = 0;
for (n = 0; n < dd->num_pports; n++) {
struct hfi1_ibport *ibp = &dd->pport[n].ibport_data;
rcu_read_lock();
if (rcu_dereference(ibp->rvp.qp[0]))
qp_inuse++;
if (rcu_dereference(ibp->rvp.qp[1]))
qp_inuse++;
rcu_read_unlock();
}
return qp_inuse;
}
void flush_qp_waiters(struct rvt_qp *qp)
{
lockdep_assert_held(&qp->s_lock);
flush_iowait(qp);
hfi1_tid_rdma_flush_wait(qp);
}
void stop_send_queue(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
iowait_cancel_work(&priv->s_iowait);
if (cancel_work_sync(&priv->tid_rdma.trigger_work))
rvt_put_qp(qp);
}
void quiesce_qp(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
hfi1_del_tid_reap_timer(qp);
hfi1_del_tid_retry_timer(qp);
iowait_sdma_drain(&priv->s_iowait);
staging/rdma/hfi1: Adaptive PIO for short messages The change requires a new pio_busy field in the iowait structure to track the number of outstanding pios. The new counter together with the sdma counter serve as the basis for a packet by packet decision as to which egress mechanism to use. Since packets given to different egress mechanisms are not ordered, this scheme will preserve the order. The iowait drain/wait mechanisms are extended for a pio case. An additional qp wait flag is added for the PIO drain wait case. Currently the only pio wait is for buffers, so the no_bufs_available() routine name is changed to pio_wait() and a third argument is passed with one of the two pio wait flags to generalize the routine. A module parameter is added to hold a configurable threshold. For now, the module parameter is zero. A heuristic routine is added to return the func pointer of the proper egress routine to use. The heuristic is as follows: - SMI always uses pio - GSI,UD qps <= threshold use pio - UD qps > threadhold use sdma o No coordination with sdma is required because order is not required and this qp pio count is not maintained for UD - RC/UC ONLY packets <= threshold chose as follows: o If sdmas pending, use SDMA o Otherwise use pio and enable the pio tracking count at the time the pio buffer is allocated - RC/UC ONLY packets > threshold use SDMA o If pio's are pending the pio_wait with the new wait flag is called to delay for pios to drain The threshold is potentially reduced by the QP's mtu. The sc_buffer_alloc() has two additional args (a callback, a void *) which are exploited by the RC/UC cases to pass a new complete routine and a qp *. When the shadow ring completes the credit associated with a packet, the new complete routine is called. The verbs_pio_complete() will then decrement the busy count and trigger any drain waiters in qp destroy or reset. Reviewed-by: Jubin John <jubin.john@intel.com> Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-02-15 04:45:36 +08:00
qp_pio_drain(qp);
flush_tx_list(qp);
}
void notify_qp_reset(struct rvt_qp *qp)
{
hfi1_qp_kern_exp_rcv_clear_all(qp);
qp->r_adefered = 0;
clear_ahg(qp);
/* Clear any OPFN state */
if (qp->ibqp.qp_type == IB_QPT_RC)
opfn_conn_error(qp);
}
/*
* Switch to alternate path.
* The QP s_lock should be held and interrupts disabled.
*/
void hfi1_migrate_qp(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
struct ib_event ev;
qp->s_mig_state = IB_MIG_MIGRATED;
qp->remote_ah_attr = qp->alt_ah_attr;
qp->port_num = rdma_ah_get_port_num(&qp->alt_ah_attr);
qp->s_pkey_index = qp->s_alt_pkey_index;
qp->s_flags |= HFI1_S_AHG_CLEAR;
priv->s_sc = ah_to_sc(qp->ibqp.device, &qp->remote_ah_attr);
priv->s_sde = qp_to_sdma_engine(qp, priv->s_sc);
qp_set_16b(qp);
ev.device = qp->ibqp.device;
ev.element.qp = &qp->ibqp;
ev.event = IB_EVENT_PATH_MIG;
qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
}
int mtu_to_path_mtu(u32 mtu)
{
return mtu_to_enum(mtu, OPA_MTU_8192);
}
u32 mtu_from_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp, u32 pmtu)
{
u32 mtu;
struct hfi1_ibdev *verbs_dev = container_of(rdi,
struct hfi1_ibdev,
rdi);
struct hfi1_devdata *dd = container_of(verbs_dev,
struct hfi1_devdata,
verbs_dev);
struct hfi1_ibport *ibp;
u8 sc, vl;
ibp = &dd->pport[qp->port_num - 1].ibport_data;
sc = ibp->sl_to_sc[rdma_ah_get_sl(&qp->remote_ah_attr)];
vl = sc_to_vlt(dd, sc);
mtu = verbs_mtu_enum_to_int(qp->ibqp.device, pmtu);
if (vl < PER_VL_SEND_CONTEXTS)
mtu = min_t(u32, mtu, dd->vld[vl].mtu);
return mtu;
}
int get_pmtu_from_attr(struct rvt_dev_info *rdi, struct rvt_qp *qp,
struct ib_qp_attr *attr)
{
int mtu, pidx = qp->port_num - 1;
struct hfi1_ibdev *verbs_dev = container_of(rdi,
struct hfi1_ibdev,
rdi);
struct hfi1_devdata *dd = container_of(verbs_dev,
struct hfi1_devdata,
verbs_dev);
mtu = verbs_mtu_enum_to_int(qp->ibqp.device, attr->path_mtu);
if (mtu == -1)
return -1; /* values less than 0 are error */
if (mtu > dd->pport[pidx].ibmtu)
return mtu_to_enum(dd->pport[pidx].ibmtu, IB_MTU_2048);
else
return attr->path_mtu;
}
void notify_error_qp(struct rvt_qp *qp)
{
struct hfi1_qp_priv *priv = qp->priv;
seqlock_t *lock = priv->s_iowait.lock;
if (lock) {
write_seqlock(lock);
if (!list_empty(&priv->s_iowait.list) &&
!(qp->s_flags & RVT_S_BUSY) &&
!(priv->s_flags & RVT_S_BUSY)) {
IB/hfi1: Failed to drain send queue when QP is put into error state When a QP is put into error state, all pending requests in the send work queue should be drained. The following sequence of events could lead to a failure, causing a request to hang: (1) The QP builds a packet and tries to send through SDMA engine. However, PIO engine is still busy. Consequently, this packet is put on the QP's tx list and the QP is put on the PIO waiting list. The field qp->s_flags is set with HFI1_S_WAIT_PIO_DRAIN; (2) The QP is put into error state by the user application and notify_error_qp() is called, which removes the QP from the PIO waiting list and the packet from the QP's tx list. In addition, qp->s_flags is cleared of RVT_S_ANY_WAIT_IO bits, which does not include HFI1_S_WAIT_PIO_DRAIN bit; (3) The hfi1_schdule_send() function is called to drain the QP's send queue. Subsequently, hfi1_do_send() is called. Since the flag bit HFI1_S_WAIT_PIO_DRAIN is set in qp->s_flags, hfi1_send_ok() fails. As a result, hfi1_do_send() bails out without draining any request from the send queue; (4) The PIO engine completes the sending and tries to wake up any QP on its waiting list. But the QP has been removed from the PIO waiting list and therefore is kept in sleep forever. The fix is to clear qp->s_flags of HFI1_S_ANY_WAIT_IO bits in step (2). HFI1_S_ANY_WAIT_IO includes RVT_S_ANY_WAIT_IO and HFI1_S_WAIT_PIO_DRAIN. Fixes: 2e2ba09e48b7 ("IB/rdmavt, IB/hfi1: Create device dependent s_flags") Cc: <stable@vger.kernel.org> # 4.19.x+ Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Reviewed-by: Alex Estrin <alex.estrin@intel.com> Signed-off-by: Kaike Wan <kaike.wan@intel.com> Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
2019-03-19 00:55:19 +08:00
qp->s_flags &= ~HFI1_S_ANY_WAIT_IO;
iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_IB);
iowait_clear_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
list_del_init(&priv->s_iowait.list);
priv->s_iowait.lock = NULL;
rvt_put_qp(qp);
}
write_sequnlock(lock);
}
if (!(qp->s_flags & RVT_S_BUSY) && !(priv->s_flags & RVT_S_BUSY)) {
qp->s_hdrwords = 0;
if (qp->s_rdma_mr) {
rvt_put_mr(qp->s_rdma_mr);
qp->s_rdma_mr = NULL;
}
flush_tx_list(qp);
}
}
/**
* hfi1_qp_iter_cb - callback for iterator
* @qp - the qp
* @v - the sl in low bits of v
*
* This is called from the iterator callback to work
* on an individual qp.
*/
static void hfi1_qp_iter_cb(struct rvt_qp *qp, u64 v)
{
int lastwqe;
struct ib_event ev;
struct hfi1_ibport *ibp =
to_iport(qp->ibqp.device, qp->port_num);
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
u8 sl = (u8)v;
if (qp->port_num != ppd->port ||
(qp->ibqp.qp_type != IB_QPT_UC &&
qp->ibqp.qp_type != IB_QPT_RC) ||
rdma_ah_get_sl(&qp->remote_ah_attr) != sl ||
!(ib_rvt_state_ops[qp->state] & RVT_POST_SEND_OK))
return;
spin_lock_irq(&qp->r_lock);
spin_lock(&qp->s_hlock);
spin_lock(&qp->s_lock);
lastwqe = rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
spin_unlock(&qp->s_lock);
spin_unlock(&qp->s_hlock);
spin_unlock_irq(&qp->r_lock);
if (lastwqe) {
ev.device = qp->ibqp.device;
ev.element.qp = &qp->ibqp;
ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
}
}
/**
* hfi1_error_port_qps - put a port's RC/UC qps into error state
* @ibp: the ibport.
* @sl: the service level.
*
* This function places all RC/UC qps with a given service level into error
* state. It is generally called to force upper lay apps to abandon stale qps
* after an sl->sc mapping change.
*/
void hfi1_error_port_qps(struct hfi1_ibport *ibp, u8 sl)
{
struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
struct hfi1_ibdev *dev = &ppd->dd->verbs_dev;
rvt_qp_iter(&dev->rdi, sl, hfi1_qp_iter_cb);
}