linux/drivers/infiniband/hw/hns/hns_roce_qp.c

1481 lines
38 KiB
C
Raw Normal View History

/*
* Copyright (c) 2016 Hisilicon Limited.
* Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <linux/pci.h>
#include <rdma/ib_addr.h>
#include <rdma/ib_umem.h>
#include <rdma/uverbs_ioctl.h>
#include "hns_roce_common.h"
#include "hns_roce_device.h"
#include "hns_roce_hem.h"
static void flush_work_handle(struct work_struct *work)
{
struct hns_roce_work *flush_work = container_of(work,
struct hns_roce_work, work);
struct hns_roce_qp *hr_qp = container_of(flush_work,
struct hns_roce_qp, flush_work);
struct device *dev = flush_work->hr_dev->dev;
struct ib_qp_attr attr;
int attr_mask;
int ret;
attr_mask = IB_QP_STATE;
attr.qp_state = IB_QPS_ERR;
if (test_and_clear_bit(HNS_ROCE_FLUSH_FLAG, &hr_qp->flush_flag)) {
ret = hns_roce_modify_qp(&hr_qp->ibqp, &attr, attr_mask, NULL);
if (ret)
dev_err(dev, "Modify QP to error state failed(%d) during CQE flush\n",
ret);
}
/*
* make sure we signal QP destroy leg that flush QP was completed
* so that it can safely proceed ahead now and destroy QP
*/
if (refcount_dec_and_test(&hr_qp->refcount))
complete(&hr_qp->free);
}
void init_flush_work(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
struct hns_roce_work *flush_work = &hr_qp->flush_work;
flush_work->hr_dev = hr_dev;
INIT_WORK(&flush_work->work, flush_work_handle);
refcount_inc(&hr_qp->refcount);
queue_work(hr_dev->irq_workq, &flush_work->work);
}
void flush_cqe(struct hns_roce_dev *dev, struct hns_roce_qp *qp)
{
/*
* Hip08 hardware cannot flush the WQEs in SQ/RQ if the QP state
* gets into errored mode. Hence, as a workaround to this
* hardware limitation, driver needs to assist in flushing. But
* the flushing operation uses mailbox to convey the QP state to
* the hardware and which can sleep due to the mutex protection
* around the mailbox calls. Hence, use the deferred flush for
* now.
*/
if (!test_and_set_bit(HNS_ROCE_FLUSH_FLAG, &qp->flush_flag))
init_flush_work(dev, qp);
}
void hns_roce_qp_event(struct hns_roce_dev *hr_dev, u32 qpn, int event_type)
{
struct device *dev = hr_dev->dev;
struct hns_roce_qp *qp;
xa_lock(&hr_dev->qp_table_xa);
qp = __hns_roce_qp_lookup(hr_dev, qpn);
if (qp)
refcount_inc(&qp->refcount);
xa_unlock(&hr_dev->qp_table_xa);
if (!qp) {
dev_warn(dev, "Async event for bogus QP %08x\n", qpn);
return;
}
if (event_type == HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR ||
event_type == HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR ||
event_type == HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR ||
event_type == HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION ||
event_type == HNS_ROCE_EVENT_TYPE_INVALID_XRCETH) {
qp->state = IB_QPS_ERR;
flush_cqe(hr_dev, qp);
}
qp->event(qp, (enum hns_roce_event)event_type);
if (refcount_dec_and_test(&qp->refcount))
complete(&qp->free);
}
static void hns_roce_ib_qp_event(struct hns_roce_qp *hr_qp,
enum hns_roce_event type)
{
struct ib_qp *ibqp = &hr_qp->ibqp;
struct ib_event event;
if (ibqp->event_handler) {
event.device = ibqp->device;
event.element.qp = ibqp;
switch (type) {
case HNS_ROCE_EVENT_TYPE_PATH_MIG:
event.event = IB_EVENT_PATH_MIG;
break;
case HNS_ROCE_EVENT_TYPE_COMM_EST:
event.event = IB_EVENT_COMM_EST;
break;
case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
event.event = IB_EVENT_SQ_DRAINED;
break;
case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
event.event = IB_EVENT_QP_LAST_WQE_REACHED;
break;
case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
event.event = IB_EVENT_QP_FATAL;
break;
case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
event.event = IB_EVENT_PATH_MIG_ERR;
break;
case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
event.event = IB_EVENT_QP_REQ_ERR;
break;
case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
case HNS_ROCE_EVENT_TYPE_XRCD_VIOLATION:
case HNS_ROCE_EVENT_TYPE_INVALID_XRCETH:
event.event = IB_EVENT_QP_ACCESS_ERR;
break;
default:
dev_dbg(ibqp->device->dev.parent, "roce_ib: Unexpected event type %d on QP %06lx\n",
type, hr_qp->qpn);
return;
}
ibqp->event_handler(&event, ibqp->qp_context);
}
}
static u8 get_least_load_bankid_for_qp(struct hns_roce_bank *bank)
{
u32 least_load = bank[0].inuse;
u8 bankid = 0;
u32 bankcnt;
u8 i;
for (i = 1; i < HNS_ROCE_QP_BANK_NUM; i++) {
bankcnt = bank[i].inuse;
if (bankcnt < least_load) {
least_load = bankcnt;
bankid = i;
}
}
return bankid;
}
static int alloc_qpn_with_bankid(struct hns_roce_bank *bank, u8 bankid,
unsigned long *qpn)
{
int id;
id = ida_alloc_range(&bank->ida, bank->next, bank->max, GFP_KERNEL);
if (id < 0) {
id = ida_alloc_range(&bank->ida, bank->min, bank->max,
GFP_KERNEL);
if (id < 0)
return id;
}
/* the QPN should keep increasing until the max value is reached. */
bank->next = (id + 1) > bank->max ? bank->min : id + 1;
/* the lower 3 bits is bankid */
*qpn = (id << 3) | bankid;
return 0;
}
static int alloc_qpn(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
unsigned long num = 0;
u8 bankid;
int ret;
if (hr_qp->ibqp.qp_type == IB_QPT_GSI) {
num = 1;
hr_qp->doorbell_qpn = 1;
} else {
mutex_lock(&qp_table->bank_mutex);
bankid = get_least_load_bankid_for_qp(qp_table->bank);
ret = alloc_qpn_with_bankid(&qp_table->bank[bankid], bankid,
&num);
if (ret) {
ibdev_err(&hr_dev->ib_dev,
"failed to alloc QPN, ret = %d\n", ret);
mutex_unlock(&qp_table->bank_mutex);
return ret;
}
qp_table->bank[bankid].inuse++;
mutex_unlock(&qp_table->bank_mutex);
hr_qp->doorbell_qpn = (u32)num;
}
hr_qp->qpn = num;
return 0;
}
static void add_qp_to_list(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
struct ib_cq *send_cq, struct ib_cq *recv_cq)
{
struct hns_roce_cq *hr_send_cq, *hr_recv_cq;
unsigned long flags;
hr_send_cq = send_cq ? to_hr_cq(send_cq) : NULL;
hr_recv_cq = recv_cq ? to_hr_cq(recv_cq) : NULL;
spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
hns_roce_lock_cqs(hr_send_cq, hr_recv_cq);
list_add_tail(&hr_qp->node, &hr_dev->qp_list);
if (hr_send_cq)
list_add_tail(&hr_qp->sq_node, &hr_send_cq->sq_list);
if (hr_recv_cq)
list_add_tail(&hr_qp->rq_node, &hr_recv_cq->rq_list);
hns_roce_unlock_cqs(hr_send_cq, hr_recv_cq);
spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
}
static int hns_roce_qp_store(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr)
{
struct xarray *xa = &hr_dev->qp_table_xa;
int ret;
if (!hr_qp->qpn)
return -EINVAL;
ret = xa_err(xa_store_irq(xa, hr_qp->qpn, hr_qp, GFP_KERNEL));
if (ret)
dev_err(hr_dev->dev, "Failed to xa store for QPC\n");
else
/* add QP to device's QP list for softwc */
add_qp_to_list(hr_dev, hr_qp, init_attr->send_cq,
init_attr->recv_cq);
return ret;
}
static int alloc_qpc(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
struct device *dev = hr_dev->dev;
int ret;
if (!hr_qp->qpn)
return -EINVAL;
/* Alloc memory for QPC */
ret = hns_roce_table_get(hr_dev, &qp_table->qp_table, hr_qp->qpn);
if (ret) {
dev_err(dev, "Failed to get QPC table\n");
goto err_out;
}
/* Alloc memory for IRRL */
ret = hns_roce_table_get(hr_dev, &qp_table->irrl_table, hr_qp->qpn);
if (ret) {
dev_err(dev, "Failed to get IRRL table\n");
goto err_put_qp;
}
if (hr_dev->caps.trrl_entry_sz) {
/* Alloc memory for TRRL */
ret = hns_roce_table_get(hr_dev, &qp_table->trrl_table,
hr_qp->qpn);
if (ret) {
dev_err(dev, "Failed to get TRRL table\n");
goto err_put_irrl;
}
}
if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
/* Alloc memory for SCC CTX */
ret = hns_roce_table_get(hr_dev, &qp_table->sccc_table,
hr_qp->qpn);
if (ret) {
dev_err(dev, "Failed to get SCC CTX table\n");
goto err_put_trrl;
}
}
return 0;
err_put_trrl:
if (hr_dev->caps.trrl_entry_sz)
hns_roce_table_put(hr_dev, &qp_table->trrl_table, hr_qp->qpn);
err_put_irrl:
hns_roce_table_put(hr_dev, &qp_table->irrl_table, hr_qp->qpn);
err_put_qp:
hns_roce_table_put(hr_dev, &qp_table->qp_table, hr_qp->qpn);
err_out:
return ret;
}
static void qp_user_mmap_entry_remove(struct hns_roce_qp *hr_qp)
{
rdma_user_mmap_entry_remove(&hr_qp->dwqe_mmap_entry->rdma_entry);
}
void hns_roce_qp_remove(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
struct xarray *xa = &hr_dev->qp_table_xa;
unsigned long flags;
list_del(&hr_qp->node);
if (hr_qp->ibqp.qp_type != IB_QPT_XRC_TGT)
list_del(&hr_qp->sq_node);
if (hr_qp->ibqp.qp_type != IB_QPT_XRC_INI &&
hr_qp->ibqp.qp_type != IB_QPT_XRC_TGT)
list_del(&hr_qp->rq_node);
xa_lock_irqsave(xa, flags);
__xa_erase(xa, hr_qp->qpn);
xa_unlock_irqrestore(xa, flags);
}
static void free_qpc(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
if (hr_dev->caps.trrl_entry_sz)
hns_roce_table_put(hr_dev, &qp_table->trrl_table, hr_qp->qpn);
hns_roce_table_put(hr_dev, &qp_table->irrl_table, hr_qp->qpn);
}
static inline u8 get_qp_bankid(unsigned long qpn)
{
/* The lower 3 bits of QPN are used to hash to different banks */
return (u8)(qpn & GENMASK(2, 0));
}
static void free_qpn(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
u8 bankid;
if (hr_qp->ibqp.qp_type == IB_QPT_GSI)
return;
if (hr_qp->qpn < hr_dev->caps.reserved_qps)
return;
bankid = get_qp_bankid(hr_qp->qpn);
ida_free(&hr_dev->qp_table.bank[bankid].ida, hr_qp->qpn >> 3);
mutex_lock(&hr_dev->qp_table.bank_mutex);
hr_dev->qp_table.bank[bankid].inuse--;
mutex_unlock(&hr_dev->qp_table.bank_mutex);
}
static u32 proc_rq_sge(struct hns_roce_dev *dev, struct hns_roce_qp *hr_qp,
bool user)
{
u32 max_sge = dev->caps.max_rq_sg;
if (dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
return max_sge;
/* Reserve SGEs only for HIP08 in kernel; The userspace driver will
* calculate number of max_sge with reserved SGEs when allocating wqe
* buf, so there is no need to do this again in kernel. But the number
* may exceed the capacity of SGEs recorded in the firmware, so the
* kernel driver should just adapt the value accordingly.
*/
if (user)
max_sge = roundup_pow_of_two(max_sge + 1);
else
hr_qp->rq.rsv_sge = 1;
return max_sge;
}
static int set_rq_size(struct hns_roce_dev *hr_dev, struct ib_qp_cap *cap,
struct hns_roce_qp *hr_qp, int has_rq, bool user)
{
u32 max_sge = proc_rq_sge(hr_dev, hr_qp, user);
u32 cnt;
/* If srq exist, set zero for relative number of rq */
if (!has_rq) {
hr_qp->rq.wqe_cnt = 0;
hr_qp->rq.max_gs = 0;
hr_qp->rq_inl_buf.wqe_cnt = 0;
cap->max_recv_wr = 0;
cap->max_recv_sge = 0;
return 0;
}
/* Check the validity of QP support capacity */
if (!cap->max_recv_wr || cap->max_recv_wr > hr_dev->caps.max_wqes ||
cap->max_recv_sge > max_sge) {
ibdev_err(&hr_dev->ib_dev,
"RQ config error, depth = %u, sge = %u\n",
cap->max_recv_wr, cap->max_recv_sge);
return -EINVAL;
}
cnt = roundup_pow_of_two(max(cap->max_recv_wr, hr_dev->caps.min_wqes));
if (cnt > hr_dev->caps.max_wqes) {
ibdev_err(&hr_dev->ib_dev, "rq depth %u too large\n",
cap->max_recv_wr);
return -EINVAL;
}
hr_qp->rq.max_gs = roundup_pow_of_two(max(1U, cap->max_recv_sge) +
hr_qp->rq.rsv_sge);
if (hr_dev->caps.max_rq_sg <= HNS_ROCE_SGE_IN_WQE)
hr_qp->rq.wqe_shift = ilog2(hr_dev->caps.max_rq_desc_sz);
else
hr_qp->rq.wqe_shift = ilog2(hr_dev->caps.max_rq_desc_sz *
hr_qp->rq.max_gs);
hr_qp->rq.wqe_cnt = cnt;
if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE &&
hr_qp->ibqp.qp_type != IB_QPT_UD &&
hr_qp->ibqp.qp_type != IB_QPT_GSI)
hr_qp->rq_inl_buf.wqe_cnt = cnt;
else
hr_qp->rq_inl_buf.wqe_cnt = 0;
cap->max_recv_wr = cnt;
cap->max_recv_sge = hr_qp->rq.max_gs - hr_qp->rq.rsv_sge;
return 0;
}
static u32 get_wqe_ext_sge_cnt(struct hns_roce_qp *qp)
{
/* GSI/UD QP only has extended sge */
if (qp->ibqp.qp_type == IB_QPT_GSI || qp->ibqp.qp_type == IB_QPT_UD)
return qp->sq.max_gs;
if (qp->sq.max_gs > HNS_ROCE_SGE_IN_WQE)
return qp->sq.max_gs - HNS_ROCE_SGE_IN_WQE;
return 0;
}
static void set_ext_sge_param(struct hns_roce_dev *hr_dev, u32 sq_wqe_cnt,
struct hns_roce_qp *hr_qp, struct ib_qp_cap *cap)
{
u32 total_sge_cnt;
u32 wqe_sge_cnt;
hr_qp->sge.sge_shift = HNS_ROCE_SGE_SHIFT;
hr_qp->sq.max_gs = max(1U, cap->max_send_sge);
wqe_sge_cnt = get_wqe_ext_sge_cnt(hr_qp);
/* If the number of extended sge is not zero, they MUST use the
* space of HNS_HW_PAGE_SIZE at least.
*/
if (wqe_sge_cnt) {
total_sge_cnt = roundup_pow_of_two(sq_wqe_cnt * wqe_sge_cnt);
hr_qp->sge.sge_cnt = max(total_sge_cnt,
(u32)HNS_HW_PAGE_SIZE / HNS_ROCE_SGE_SIZE);
}
}
static int check_sq_size_with_integrity(struct hns_roce_dev *hr_dev,
struct ib_qp_cap *cap,
struct hns_roce_ib_create_qp *ucmd)
{
u32 roundup_sq_stride = roundup_pow_of_two(hr_dev->caps.max_sq_desc_sz);
u8 max_sq_stride = ilog2(roundup_sq_stride);
/* Sanity check SQ size before proceeding */
if (ucmd->log_sq_stride > max_sq_stride ||
ucmd->log_sq_stride < HNS_ROCE_IB_MIN_SQ_STRIDE) {
ibdev_err(&hr_dev->ib_dev, "failed to check SQ stride size.\n");
return -EINVAL;
}
if (cap->max_send_sge > hr_dev->caps.max_sq_sg) {
ibdev_err(&hr_dev->ib_dev, "failed to check SQ SGE size %u.\n",
cap->max_send_sge);
return -EINVAL;
}
return 0;
}
static int set_user_sq_size(struct hns_roce_dev *hr_dev,
struct ib_qp_cap *cap, struct hns_roce_qp *hr_qp,
struct hns_roce_ib_create_qp *ucmd)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
u32 cnt = 0;
int ret;
if (check_shl_overflow(1, ucmd->log_sq_bb_count, &cnt) ||
cnt > hr_dev->caps.max_wqes)
return -EINVAL;
ret = check_sq_size_with_integrity(hr_dev, cap, ucmd);
if (ret) {
ibdev_err(ibdev, "failed to check user SQ size, ret = %d.\n",
ret);
return ret;
}
set_ext_sge_param(hr_dev, cnt, hr_qp, cap);
hr_qp->sq.wqe_shift = ucmd->log_sq_stride;
hr_qp->sq.wqe_cnt = cnt;
return 0;
}
static int set_wqe_buf_attr(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
struct hns_roce_buf_attr *buf_attr)
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
{
int buf_size;
int idx = 0;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
hr_qp->buff_size = 0;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
/* SQ WQE */
hr_qp->sq.offset = 0;
buf_size = to_hr_hem_entries_size(hr_qp->sq.wqe_cnt,
hr_qp->sq.wqe_shift);
if (buf_size > 0 && idx < ARRAY_SIZE(buf_attr->region)) {
buf_attr->region[idx].size = buf_size;
buf_attr->region[idx].hopnum = hr_dev->caps.wqe_sq_hop_num;
idx++;
hr_qp->buff_size += buf_size;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
}
/* extend SGE WQE in SQ */
hr_qp->sge.offset = hr_qp->buff_size;
buf_size = to_hr_hem_entries_size(hr_qp->sge.sge_cnt,
hr_qp->sge.sge_shift);
if (buf_size > 0 && idx < ARRAY_SIZE(buf_attr->region)) {
buf_attr->region[idx].size = buf_size;
buf_attr->region[idx].hopnum = hr_dev->caps.wqe_sge_hop_num;
idx++;
hr_qp->buff_size += buf_size;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
}
/* RQ WQE */
hr_qp->rq.offset = hr_qp->buff_size;
buf_size = to_hr_hem_entries_size(hr_qp->rq.wqe_cnt,
hr_qp->rq.wqe_shift);
if (buf_size > 0 && idx < ARRAY_SIZE(buf_attr->region)) {
buf_attr->region[idx].size = buf_size;
buf_attr->region[idx].hopnum = hr_dev->caps.wqe_rq_hop_num;
idx++;
hr_qp->buff_size += buf_size;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
}
if (hr_qp->buff_size < 1)
return -EINVAL;
buf_attr->page_shift = HNS_HW_PAGE_SHIFT + hr_dev->caps.mtt_buf_pg_sz;
buf_attr->region_count = idx;
return 0;
}
static int set_kernel_sq_size(struct hns_roce_dev *hr_dev,
struct ib_qp_cap *cap, struct hns_roce_qp *hr_qp)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
u32 cnt;
if (!cap->max_send_wr || cap->max_send_wr > hr_dev->caps.max_wqes ||
cap->max_send_sge > hr_dev->caps.max_sq_sg) {
ibdev_err(ibdev, "failed to check SQ WR or SGE num.\n");
return -EINVAL;
}
cnt = roundup_pow_of_two(max(cap->max_send_wr, hr_dev->caps.min_wqes));
if (cnt > hr_dev->caps.max_wqes) {
ibdev_err(ibdev, "failed to check WQE num, WQE num = %u.\n",
cnt);
return -EINVAL;
}
hr_qp->sq.wqe_shift = ilog2(hr_dev->caps.max_sq_desc_sz);
hr_qp->sq.wqe_cnt = cnt;
set_ext_sge_param(hr_dev, cnt, hr_qp, cap);
/* sync the parameters of kernel QP to user's configuration */
cap->max_send_wr = cnt;
cap->max_send_sge = hr_qp->sq.max_gs;
return 0;
}
static int hns_roce_qp_has_sq(struct ib_qp_init_attr *attr)
{
if (attr->qp_type == IB_QPT_XRC_TGT || !attr->cap.max_send_wr)
return 0;
return 1;
}
static int hns_roce_qp_has_rq(struct ib_qp_init_attr *attr)
{
if (attr->qp_type == IB_QPT_XRC_INI ||
attr->qp_type == IB_QPT_XRC_TGT || attr->srq ||
!attr->cap.max_recv_wr)
return 0;
return 1;
}
static int alloc_rq_inline_buf(struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr)
{
u32 max_recv_sge = init_attr->cap.max_recv_sge;
u32 wqe_cnt = hr_qp->rq_inl_buf.wqe_cnt;
struct hns_roce_rinl_wqe *wqe_list;
int i;
/* allocate recv inline buf */
wqe_list = kcalloc(wqe_cnt, sizeof(struct hns_roce_rinl_wqe),
GFP_KERNEL);
if (!wqe_list)
goto err;
/* Allocate a continuous buffer for all inline sge we need */
wqe_list[0].sg_list = kcalloc(wqe_cnt, (max_recv_sge *
sizeof(struct hns_roce_rinl_sge)),
GFP_KERNEL);
if (!wqe_list[0].sg_list)
goto err_wqe_list;
/* Assign buffers of sg_list to each inline wqe */
for (i = 1; i < wqe_cnt; i++)
wqe_list[i].sg_list = &wqe_list[0].sg_list[i * max_recv_sge];
hr_qp->rq_inl_buf.wqe_list = wqe_list;
return 0;
err_wqe_list:
kfree(wqe_list);
err:
return -ENOMEM;
}
static void free_rq_inline_buf(struct hns_roce_qp *hr_qp)
{
if (hr_qp->rq_inl_buf.wqe_list)
kfree(hr_qp->rq_inl_buf.wqe_list[0].sg_list);
kfree(hr_qp->rq_inl_buf.wqe_list);
}
static int alloc_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata, unsigned long addr)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
struct hns_roce_buf_attr buf_attr = {};
int ret;
if (!udata && hr_qp->rq_inl_buf.wqe_cnt) {
ret = alloc_rq_inline_buf(hr_qp, init_attr);
if (ret) {
ibdev_err(ibdev,
"failed to alloc inline buf, ret = %d.\n",
ret);
return ret;
}
} else {
hr_qp->rq_inl_buf.wqe_list = NULL;
}
ret = set_wqe_buf_attr(hr_dev, hr_qp, &buf_attr);
if (ret) {
ibdev_err(ibdev, "failed to split WQE buf, ret = %d.\n", ret);
goto err_inline;
}
ret = hns_roce_mtr_create(hr_dev, &hr_qp->mtr, &buf_attr,
PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz,
udata, addr);
if (ret) {
ibdev_err(ibdev, "failed to create WQE mtr, ret = %d.\n", ret);
goto err_inline;
}
if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_DIRECT_WQE)
hr_qp->en_flags |= HNS_ROCE_QP_CAP_DIRECT_WQE;
return 0;
err_inline:
free_rq_inline_buf(hr_qp);
return ret;
}
static void free_qp_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
hns_roce_mtr_destroy(hr_dev, &hr_qp->mtr);
free_rq_inline_buf(hr_qp);
}
static inline bool user_qp_has_sdb(struct hns_roce_dev *hr_dev,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata,
struct hns_roce_ib_create_qp_resp *resp,
struct hns_roce_ib_create_qp *ucmd)
{
return ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) &&
udata->outlen >= offsetofend(typeof(*resp), cap_flags) &&
hns_roce_qp_has_sq(init_attr) &&
udata->inlen >= offsetofend(typeof(*ucmd), sdb_addr));
}
static inline bool user_qp_has_rdb(struct hns_roce_dev *hr_dev,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata,
struct hns_roce_ib_create_qp_resp *resp)
{
return ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) &&
udata->outlen >= offsetofend(typeof(*resp), cap_flags) &&
hns_roce_qp_has_rq(init_attr));
}
static inline bool kernel_qp_has_rdb(struct hns_roce_dev *hr_dev,
struct ib_qp_init_attr *init_attr)
{
return ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) &&
hns_roce_qp_has_rq(init_attr));
}
static int qp_mmap_entry(struct hns_roce_qp *hr_qp,
struct hns_roce_dev *hr_dev,
struct ib_udata *udata,
struct hns_roce_ib_create_qp_resp *resp)
{
struct hns_roce_ucontext *uctx =
rdma_udata_to_drv_context(udata,
struct hns_roce_ucontext, ibucontext);
struct rdma_user_mmap_entry *rdma_entry;
u64 address;
address = hr_dev->dwqe_page + hr_qp->qpn * HNS_ROCE_DWQE_SIZE;
hr_qp->dwqe_mmap_entry =
hns_roce_user_mmap_entry_insert(&uctx->ibucontext, address,
HNS_ROCE_DWQE_SIZE,
HNS_ROCE_MMAP_TYPE_DWQE);
if (!hr_qp->dwqe_mmap_entry) {
ibdev_err(&hr_dev->ib_dev, "failed to get dwqe mmap entry.\n");
return -ENOMEM;
}
rdma_entry = &hr_qp->dwqe_mmap_entry->rdma_entry;
resp->dwqe_mmap_key = rdma_user_mmap_get_offset(rdma_entry);
return 0;
}
static int alloc_user_qp_db(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata,
struct hns_roce_ib_create_qp *ucmd,
struct hns_roce_ib_create_qp_resp *resp)
{
struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context(udata,
struct hns_roce_ucontext, ibucontext);
struct ib_device *ibdev = &hr_dev->ib_dev;
int ret;
if (user_qp_has_sdb(hr_dev, init_attr, udata, resp, ucmd)) {
ret = hns_roce_db_map_user(uctx, ucmd->sdb_addr, &hr_qp->sdb);
if (ret) {
ibdev_err(ibdev,
"failed to map user SQ doorbell, ret = %d.\n",
ret);
goto err_out;
}
hr_qp->en_flags |= HNS_ROCE_QP_CAP_SQ_RECORD_DB;
}
if (user_qp_has_rdb(hr_dev, init_attr, udata, resp)) {
ret = hns_roce_db_map_user(uctx, ucmd->db_addr, &hr_qp->rdb);
if (ret) {
ibdev_err(ibdev,
"failed to map user RQ doorbell, ret = %d.\n",
ret);
goto err_sdb;
}
hr_qp->en_flags |= HNS_ROCE_QP_CAP_RQ_RECORD_DB;
}
return 0;
err_sdb:
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_SQ_RECORD_DB)
hns_roce_db_unmap_user(uctx, &hr_qp->sdb);
err_out:
return ret;
}
static int alloc_kernel_qp_db(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
int ret;
if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
hr_qp->sq.db_reg = hr_dev->mem_base +
HNS_ROCE_DWQE_SIZE * hr_qp->qpn;
else
hr_qp->sq.db_reg = hr_dev->reg_base + hr_dev->sdb_offset +
DB_REG_OFFSET * hr_dev->priv_uar.index;
hr_qp->rq.db_reg = hr_dev->reg_base + hr_dev->odb_offset +
DB_REG_OFFSET * hr_dev->priv_uar.index;
if (kernel_qp_has_rdb(hr_dev, init_attr)) {
ret = hns_roce_alloc_db(hr_dev, &hr_qp->rdb, 0);
if (ret) {
ibdev_err(ibdev,
"failed to alloc kernel RQ doorbell, ret = %d.\n",
ret);
return ret;
}
*hr_qp->rdb.db_record = 0;
hr_qp->en_flags |= HNS_ROCE_QP_CAP_RQ_RECORD_DB;
}
return 0;
}
static int alloc_qp_db(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata,
struct hns_roce_ib_create_qp *ucmd,
struct hns_roce_ib_create_qp_resp *resp)
{
int ret;
if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SDI_MODE)
hr_qp->en_flags |= HNS_ROCE_QP_CAP_OWNER_DB;
if (udata) {
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_DIRECT_WQE) {
ret = qp_mmap_entry(hr_qp, hr_dev, udata, resp);
if (ret)
return ret;
}
ret = alloc_user_qp_db(hr_dev, hr_qp, init_attr, udata, ucmd,
resp);
if (ret)
goto err_remove_qp;
} else {
ret = alloc_kernel_qp_db(hr_dev, hr_qp, init_attr);
if (ret)
return ret;
}
return 0;
err_remove_qp:
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_DIRECT_WQE)
qp_user_mmap_entry_remove(hr_qp);
return ret;
}
static void free_qp_db(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
struct ib_udata *udata)
{
struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context(
udata, struct hns_roce_ucontext, ibucontext);
if (udata) {
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
hns_roce_db_unmap_user(uctx, &hr_qp->rdb);
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_SQ_RECORD_DB)
hns_roce_db_unmap_user(uctx, &hr_qp->sdb);
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_DIRECT_WQE)
qp_user_mmap_entry_remove(hr_qp);
} else {
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
hns_roce_free_db(hr_dev, &hr_qp->rdb);
}
}
static int alloc_kernel_wrid(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
u64 *sq_wrid = NULL;
u64 *rq_wrid = NULL;
int ret;
sq_wrid = kcalloc(hr_qp->sq.wqe_cnt, sizeof(u64), GFP_KERNEL);
if (ZERO_OR_NULL_PTR(sq_wrid)) {
ibdev_err(ibdev, "failed to alloc SQ wrid.\n");
return -ENOMEM;
}
if (hr_qp->rq.wqe_cnt) {
rq_wrid = kcalloc(hr_qp->rq.wqe_cnt, sizeof(u64), GFP_KERNEL);
if (ZERO_OR_NULL_PTR(rq_wrid)) {
ibdev_err(ibdev, "failed to alloc RQ wrid.\n");
ret = -ENOMEM;
goto err_sq;
}
}
hr_qp->sq.wrid = sq_wrid;
hr_qp->rq.wrid = rq_wrid;
return 0;
err_sq:
kfree(sq_wrid);
return ret;
}
static void free_kernel_wrid(struct hns_roce_qp *hr_qp)
{
kfree(hr_qp->rq.wrid);
kfree(hr_qp->sq.wrid);
}
static int set_qp_param(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata,
struct hns_roce_ib_create_qp *ucmd)
{
struct ib_device *ibdev = &hr_dev->ib_dev;
int ret;
if (init_attr->cap.max_inline_data > hr_dev->caps.max_sq_inline)
init_attr->cap.max_inline_data = hr_dev->caps.max_sq_inline;
hr_qp->max_inline_data = init_attr->cap.max_inline_data;
if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
hr_qp->sq_signal_bits = IB_SIGNAL_ALL_WR;
else
hr_qp->sq_signal_bits = IB_SIGNAL_REQ_WR;
ret = set_rq_size(hr_dev, &init_attr->cap, hr_qp,
hns_roce_qp_has_rq(init_attr), !!udata);
if (ret) {
ibdev_err(ibdev, "failed to set user RQ size, ret = %d.\n",
ret);
return ret;
}
if (udata) {
ret = ib_copy_from_udata(ucmd, udata,
min(udata->inlen, sizeof(*ucmd)));
if (ret) {
ibdev_err(ibdev,
"failed to copy QP ucmd, ret = %d\n", ret);
return ret;
}
ret = set_user_sq_size(hr_dev, &init_attr->cap, hr_qp, ucmd);
if (ret)
ibdev_err(ibdev,
"failed to set user SQ size, ret = %d.\n",
ret);
} else {
ret = set_kernel_sq_size(hr_dev, &init_attr->cap, hr_qp);
if (ret)
ibdev_err(ibdev,
"failed to set kernel SQ size, ret = %d.\n",
ret);
}
return ret;
}
static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
struct ib_pd *ib_pd,
struct ib_qp_init_attr *init_attr,
struct ib_udata *udata,
struct hns_roce_qp *hr_qp)
{
struct hns_roce_ib_create_qp_resp resp = {};
struct ib_device *ibdev = &hr_dev->ib_dev;
struct hns_roce_ib_create_qp ucmd;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
int ret;
mutex_init(&hr_qp->mutex);
spin_lock_init(&hr_qp->sq.lock);
spin_lock_init(&hr_qp->rq.lock);
hr_qp->state = IB_QPS_RESET;
hr_qp->flush_flag = 0;
if (init_attr->create_flags)
return -EOPNOTSUPP;
ret = set_qp_param(hr_dev, hr_qp, init_attr, udata, &ucmd);
if (ret) {
ibdev_err(ibdev, "failed to set QP param, ret = %d.\n", ret);
return ret;
}
if (!udata) {
ret = alloc_kernel_wrid(hr_dev, hr_qp);
if (ret) {
ibdev_err(ibdev, "failed to alloc wrid, ret = %d.\n",
ret);
return ret;
}
}
ret = alloc_qp_buf(hr_dev, hr_qp, init_attr, udata, ucmd.buf_addr);
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
if (ret) {
ibdev_err(ibdev, "failed to alloc QP buffer, ret = %d.\n", ret);
goto err_buf;
}
ret = alloc_qpn(hr_dev, hr_qp);
if (ret) {
ibdev_err(ibdev, "failed to alloc QPN, ret = %d.\n", ret);
goto err_qpn;
}
ret = alloc_qp_db(hr_dev, hr_qp, init_attr, udata, &ucmd, &resp);
if (ret) {
ibdev_err(ibdev, "failed to alloc QP doorbell, ret = %d.\n",
ret);
goto err_db;
RDMA/hns: Fix bug when wqe num is larger than 16K hip08 can support up to 32768 wqes in one qp. currently if the wqe num is larger than 16384, the driver will lead a calltrace as follows. [21361.393725] Call trace: [21361.398605] hns_roce_v2_modify_qp+0xbcc/0x1360 [hns_roce_hw_v2] [21361.410627] hns_roce_modify_qp+0x1d8/0x2f8 [hns_roce] [21361.420906] _ib_modify_qp+0x70/0x118 [21361.428222] ib_modify_qp+0x14/0x1c [21361.435193] rt_ktest_modify_qp+0xb8/0x650 [rdma_test] [21361.445472] exec_modify_qp_cmd+0x110/0x4d8 [rdma_test] [21361.455924] rt_ktest_dispatch_cmd_3+0xa94/0x2edc [rdma_test] [21361.467422] rt_ktest_dispatch_cmd_2+0x9c/0x108 [rdma_test] [21361.478570] rt_ktest_dispatch_cmd+0x138/0x904 [rdma_test] [21361.489545] rt_ktest_dev_write+0x328/0x4b0 [rdma_test] [21361.499998] __vfs_write+0x38/0x15c [21361.506966] vfs_write+0xa8/0x1a0 [21361.513586] ksys_write+0x50/0xb0 [21361.520206] sys_write+0xc/0x14 [21361.526479] el0_svc_naked+0x30/0x34 [21361.533622] Code: 1ac10841 d37d7c22 0b000021 d37df021 (f86268c0) [21361.545815] ---[ end trace e2a1feb2c3d7f13c ]--- When the wqe num is larger than 16384, hns_roce_table_find will return an invalid mtt, this will lead an kernel paging requet error if the driver try to access it. It's the mtt design defect which can't support up to the max wqe num of hip08. This patch fixs it by replacing mtt with mtr for wqe. Fixes: 926a01dc000d ("RDMA/hns: Add QP operations support for hip08 SoC") Signed-off-by: Xi Wang <wangxi11@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Doug Ledford <dledford@redhat.com>
2019-06-08 14:46:10 +08:00
}
ret = alloc_qpc(hr_dev, hr_qp);
if (ret) {
ibdev_err(ibdev, "failed to alloc QP context, ret = %d.\n",
ret);
goto err_qpc;
}
ret = hns_roce_qp_store(hr_dev, hr_qp, init_attr);
if (ret) {
ibdev_err(ibdev, "failed to store QP, ret = %d.\n", ret);
goto err_store;
}
if (udata) {
resp.cap_flags = hr_qp->en_flags;
ret = ib_copy_to_udata(udata, &resp,
min(udata->outlen, sizeof(resp)));
if (ret) {
ibdev_err(ibdev, "copy qp resp failed!\n");
goto err_store;
}
}
if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
ret = hr_dev->hw->qp_flow_control_init(hr_dev, hr_qp);
if (ret)
goto err_flow_ctrl;
}
hr_qp->ibqp.qp_num = hr_qp->qpn;
hr_qp->event = hns_roce_ib_qp_event;
refcount_set(&hr_qp->refcount, 1);
init_completion(&hr_qp->free);
return 0;
err_flow_ctrl:
hns_roce_qp_remove(hr_dev, hr_qp);
err_store:
free_qpc(hr_dev, hr_qp);
err_qpc:
free_qp_db(hr_dev, hr_qp, udata);
err_db:
free_qpn(hr_dev, hr_qp);
err_qpn:
free_qp_buf(hr_dev, hr_qp);
err_buf:
free_kernel_wrid(hr_qp);
return ret;
}
void hns_roce_qp_destroy(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
struct ib_udata *udata)
{
if (refcount_dec_and_test(&hr_qp->refcount))
complete(&hr_qp->free);
wait_for_completion(&hr_qp->free);
free_qpc(hr_dev, hr_qp);
free_qpn(hr_dev, hr_qp);
free_qp_buf(hr_dev, hr_qp);
free_kernel_wrid(hr_qp);
free_qp_db(hr_dev, hr_qp, udata);
}
static int check_qp_type(struct hns_roce_dev *hr_dev, enum ib_qp_type type,
bool is_user)
{
switch (type) {
case IB_QPT_XRC_INI:
case IB_QPT_XRC_TGT:
if (!(hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC))
goto out;
break;
case IB_QPT_UD:
if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08 &&
is_user)
goto out;
break;
case IB_QPT_RC:
case IB_QPT_GSI:
break;
default:
goto out;
}
return 0;
out:
ibdev_err(&hr_dev->ib_dev, "not support QP type %d\n", type);
return -EOPNOTSUPP;
}
int hns_roce_create_qp(struct ib_qp *qp, struct ib_qp_init_attr *init_attr,
struct ib_udata *udata)
{
struct ib_device *ibdev = qp->device;
struct hns_roce_dev *hr_dev = to_hr_dev(ibdev);
struct hns_roce_qp *hr_qp = to_hr_qp(qp);
struct ib_pd *pd = qp->pd;
int ret;
ret = check_qp_type(hr_dev, init_attr->qp_type, !!udata);
if (ret)
return ret;
if (init_attr->qp_type == IB_QPT_XRC_TGT)
hr_qp->xrcdn = to_hr_xrcd(init_attr->xrcd)->xrcdn;
if (init_attr->qp_type == IB_QPT_GSI) {
hr_qp->port = init_attr->port_num - 1;
hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
}
ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata, hr_qp);
if (ret)
ibdev_err(ibdev, "Create QP type 0x%x failed(%d)\n",
init_attr->qp_type, ret);
return ret;
}
int to_hr_qp_type(int qp_type)
{
switch (qp_type) {
case IB_QPT_RC:
return SERV_TYPE_RC;
case IB_QPT_UD:
case IB_QPT_GSI:
return SERV_TYPE_UD;
case IB_QPT_XRC_INI:
case IB_QPT_XRC_TGT:
return SERV_TYPE_XRC;
default:
return -1;
}
}
static int check_mtu_validate(struct hns_roce_dev *hr_dev,
struct hns_roce_qp *hr_qp,
struct ib_qp_attr *attr, int attr_mask)
{
enum ib_mtu active_mtu;
int p;
p = attr_mask & IB_QP_PORT ? (attr->port_num - 1) : hr_qp->port;
active_mtu = iboe_get_mtu(hr_dev->iboe.netdevs[p]->mtu);
if ((hr_dev->caps.max_mtu >= IB_MTU_2048 &&
attr->path_mtu > hr_dev->caps.max_mtu) ||
attr->path_mtu < IB_MTU_256 || attr->path_mtu > active_mtu) {
ibdev_err(&hr_dev->ib_dev,
"attr path_mtu(%d)invalid while modify qp",
attr->path_mtu);
return -EINVAL;
}
return 0;
}
static int hns_roce_check_qp_attr(struct ib_qp *ibqp, struct ib_qp_attr *attr,
int attr_mask)
{
struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
int p;
if ((attr_mask & IB_QP_PORT) &&
(attr->port_num == 0 || attr->port_num > hr_dev->caps.num_ports)) {
ibdev_err(&hr_dev->ib_dev, "invalid attr, port_num = %u.\n",
attr->port_num);
return -EINVAL;
}
if (attr_mask & IB_QP_PKEY_INDEX) {
p = attr_mask & IB_QP_PORT ? (attr->port_num - 1) : hr_qp->port;
if (attr->pkey_index >= hr_dev->caps.pkey_table_len[p]) {
ibdev_err(&hr_dev->ib_dev,
"invalid attr, pkey_index = %u.\n",
attr->pkey_index);
return -EINVAL;
}
}
if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
attr->max_rd_atomic > hr_dev->caps.max_qp_init_rdma) {
ibdev_err(&hr_dev->ib_dev,
"invalid attr, max_rd_atomic = %u.\n",
attr->max_rd_atomic);
return -EINVAL;
}
if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
attr->max_dest_rd_atomic > hr_dev->caps.max_qp_dest_rdma) {
ibdev_err(&hr_dev->ib_dev,
"invalid attr, max_dest_rd_atomic = %u.\n",
attr->max_dest_rd_atomic);
return -EINVAL;
}
if (attr_mask & IB_QP_PATH_MTU)
return check_mtu_validate(hr_dev, hr_qp, attr, attr_mask);
return 0;
}
int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
int attr_mask, struct ib_udata *udata)
{
struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
enum ib_qp_state cur_state, new_state;
int ret = -EINVAL;
mutex_lock(&hr_qp->mutex);
if (attr_mask & IB_QP_CUR_STATE && attr->cur_qp_state != hr_qp->state)
goto out;
cur_state = hr_qp->state;
new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
if (ibqp->uobject &&
(attr_mask & IB_QP_STATE) && new_state == IB_QPS_ERR) {
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_SQ_RECORD_DB) {
hr_qp->sq.head = *(int *)(hr_qp->sdb.virt_addr);
if (hr_qp->en_flags & HNS_ROCE_QP_CAP_RQ_RECORD_DB)
hr_qp->rq.head = *(int *)(hr_qp->rdb.virt_addr);
} else {
ibdev_warn(&hr_dev->ib_dev,
"flush cqe is not supported in userspace!\n");
goto out;
}
}
if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
attr_mask)) {
ibdev_err(&hr_dev->ib_dev, "ib_modify_qp_is_ok failed\n");
goto out;
}
ret = hns_roce_check_qp_attr(ibqp, attr, attr_mask);
if (ret)
goto out;
if (cur_state == new_state && cur_state == IB_QPS_RESET)
goto out;
ret = hr_dev->hw->modify_qp(ibqp, attr, attr_mask, cur_state,
new_state);
out:
mutex_unlock(&hr_qp->mutex);
return ret;
}
void hns_roce_lock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq)
__acquires(&send_cq->lock) __acquires(&recv_cq->lock)
{
if (unlikely(send_cq == NULL && recv_cq == NULL)) {
__acquire(&send_cq->lock);
__acquire(&recv_cq->lock);
} else if (unlikely(send_cq != NULL && recv_cq == NULL)) {
spin_lock_irq(&send_cq->lock);
__acquire(&recv_cq->lock);
} else if (unlikely(send_cq == NULL && recv_cq != NULL)) {
spin_lock_irq(&recv_cq->lock);
__acquire(&send_cq->lock);
} else if (send_cq == recv_cq) {
spin_lock_irq(&send_cq->lock);
__acquire(&recv_cq->lock);
} else if (send_cq->cqn < recv_cq->cqn) {
spin_lock_irq(&send_cq->lock);
spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
} else {
spin_lock_irq(&recv_cq->lock);
spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING);
}
}
void hns_roce_unlock_cqs(struct hns_roce_cq *send_cq,
struct hns_roce_cq *recv_cq) __releases(&send_cq->lock)
__releases(&recv_cq->lock)
{
if (unlikely(send_cq == NULL && recv_cq == NULL)) {
__release(&recv_cq->lock);
__release(&send_cq->lock);
} else if (unlikely(send_cq != NULL && recv_cq == NULL)) {
__release(&recv_cq->lock);
spin_unlock(&send_cq->lock);
} else if (unlikely(send_cq == NULL && recv_cq != NULL)) {
__release(&send_cq->lock);
spin_unlock(&recv_cq->lock);
} else if (send_cq == recv_cq) {
__release(&recv_cq->lock);
spin_unlock_irq(&send_cq->lock);
} else if (send_cq->cqn < recv_cq->cqn) {
spin_unlock(&recv_cq->lock);
spin_unlock_irq(&send_cq->lock);
} else {
spin_unlock(&send_cq->lock);
spin_unlock_irq(&recv_cq->lock);
}
}
static inline void *get_wqe(struct hns_roce_qp *hr_qp, u32 offset)
{
return hns_roce_buf_offset(hr_qp->mtr.kmem, offset);
}
void *hns_roce_get_recv_wqe(struct hns_roce_qp *hr_qp, unsigned int n)
{
return get_wqe(hr_qp, hr_qp->rq.offset + (n << hr_qp->rq.wqe_shift));
}
void *hns_roce_get_send_wqe(struct hns_roce_qp *hr_qp, unsigned int n)
{
return get_wqe(hr_qp, hr_qp->sq.offset + (n << hr_qp->sq.wqe_shift));
}
void *hns_roce_get_extend_sge(struct hns_roce_qp *hr_qp, unsigned int n)
{
return get_wqe(hr_qp, hr_qp->sge.offset + (n << hr_qp->sge.sge_shift));
}
bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, u32 nreq,
struct ib_cq *ib_cq)
{
struct hns_roce_cq *hr_cq;
u32 cur;
cur = hr_wq->head - hr_wq->tail;
if (likely(cur + nreq < hr_wq->wqe_cnt))
return false;
hr_cq = to_hr_cq(ib_cq);
spin_lock(&hr_cq->lock);
cur = hr_wq->head - hr_wq->tail;
spin_unlock(&hr_cq->lock);
return cur + nreq >= hr_wq->wqe_cnt;
}
int hns_roce_init_qp_table(struct hns_roce_dev *hr_dev)
{
struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
unsigned int reserved_from_bot;
unsigned int i;
qp_table->idx_table.spare_idx = kcalloc(hr_dev->caps.num_qps,
sizeof(u32), GFP_KERNEL);
if (!qp_table->idx_table.spare_idx)
return -ENOMEM;
mutex_init(&qp_table->scc_mutex);
mutex_init(&qp_table->bank_mutex);
xa_init(&hr_dev->qp_table_xa);
reserved_from_bot = hr_dev->caps.reserved_qps;
for (i = 0; i < reserved_from_bot; i++) {
hr_dev->qp_table.bank[get_qp_bankid(i)].inuse++;
hr_dev->qp_table.bank[get_qp_bankid(i)].min++;
}
for (i = 0; i < HNS_ROCE_QP_BANK_NUM; i++) {
ida_init(&hr_dev->qp_table.bank[i].ida);
hr_dev->qp_table.bank[i].max = hr_dev->caps.num_qps /
HNS_ROCE_QP_BANK_NUM - 1;
hr_dev->qp_table.bank[i].next = hr_dev->qp_table.bank[i].min;
}
return 0;
}
void hns_roce_cleanup_qp_table(struct hns_roce_dev *hr_dev)
{
int i;
for (i = 0; i < HNS_ROCE_QP_BANK_NUM; i++)
ida_destroy(&hr_dev->qp_table.bank[i].ida);
kfree(hr_dev->qp_table.idx_table.spare_idx);
}