tmakin/tegra-linux-noble
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

RDMA/hns: Fix flush cqe error when racing with destroy qp

Browse Source 
BugLink: https://bugs.launchpad.net/bugs/2101915

[ Upstream commit 377a2097705b915325a67e4d44f9f2844e567809 ]

QP needs to be modified to IB_QPS_ERROR to trigger HW flush cqe. But
when this process races with destroy qp, the destroy-qp process may
modify the QP to IB_QPS_RESET first. In this case flush cqe will fail
since it is invalid to modify qp from IB_QPS_RESET to IB_QPS_ERROR.

Add lock and bit flag to make sure pending flush cqe work is completed
first and no more new works will be added.

Fixes: ffd541d457 ("RDMA/hns: Add the workqueue framework for flush cqe handler")
Signed-off-by: wenglianfa <wenglianfa@huawei.com>
Signed-off-by: Junxian Huang <huangjunxian6@hisilicon.com>
Link: https://patch.msgid.link/20241024124000.2931869-3-huangjunxian6@hisilicon.com
Reviewed-by: Zhu Yanjun <yanjun.zhu@linux.dev>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[koichiroden: adjusted context due to missing commit:
ee20cc17e9d8 ("RDMA/hns: Support DSCP")]
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
This commit is contained in:
wenglianfa
2025-03-11 08:51:19 +09:00
committed by Stefan Bader
parent c0471c0ffd
commit 4f0c922e9b
3 changed files with 22 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
drivers/infiniband/hw/hns/hns_roce_device.h
+2
View File
@@ -581,6 +581,7 @@ struct hns_roce_dev;
enum {
HNS_ROCE_FLUSH_FLAG = 0,
HNS_ROCE_STOP_FLUSH_FLAG = 1,
};
struct hns_roce_work {
@@ -642,6 +643,7 @@ struct hns_roce_qp {
struct hns_user_mmap_entry *dwqe_mmap_entry;
u32 config;
enum hns_roce_cong_type cong_type;
spinlock_t flush_lock;
};
struct hns_roce_ib_iboe {
drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+7
View File
@@ -5531,8 +5531,15 @@ int hns_roce_v2_destroy_qp(struct ib_qp *ibqp, 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);
unsigned long flags;
int ret;
/* Make sure flush_cqe() is completed */
spin_lock_irqsave(&hr_qp->flush_lock, flags);
set_bit(HNS_ROCE_STOP_FLUSH_FLAG, &hr_qp->flush_flag);
spin_unlock_irqrestore(&hr_qp->flush_lock, flags);
flush_work(&hr_qp->flush_work.work);
ret = hns_roce_v2_destroy_qp_common(hr_dev, hr_qp, udata);
if (ret)
ibdev_err_ratelimited(&hr_dev->ib_dev,
drivers/infiniband/hw/hns/hns_roce_qp.c
+13 -2
View File
@@ -90,11 +90,18 @@ static void flush_work_handle(struct work_struct *work)
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;
unsigned long flags;
spin_lock_irqsave(&hr_qp->flush_lock, flags);
/* Exit directly after destroy_qp() */
if (test_bit(HNS_ROCE_STOP_FLUSH_FLAG, &hr_qp->flush_flag)) {
spin_unlock_irqrestore(&hr_qp->flush_lock, flags);
return;
}
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);
spin_unlock_irqrestore(&hr_qp->flush_lock, flags);
}
void flush_cqe(struct hns_roce_dev *dev, struct hns_roce_qp *qp)
@@ -1080,6 +1087,7 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
struct ib_udata *udata,
struct hns_roce_qp *hr_qp)
{
struct hns_roce_work *flush_work = &hr_qp->flush_work;
struct hns_roce_ib_create_qp_resp resp = {};
struct ib_device *ibdev = &hr_dev->ib_dev;
struct hns_roce_ib_create_qp ucmd = {};
@@ -1088,9 +1096,12 @@ static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
mutex_init(&hr_qp->mutex);
spin_lock_init(&hr_qp->sq.lock);
spin_lock_init(&hr_qp->rq.lock);
spin_lock_init(&hr_qp->flush_lock);
hr_qp->state = IB_QPS_RESET;
hr_qp->flush_flag = 0;
flush_work->hr_dev = hr_dev;
INIT_WORK(&flush_work->work, flush_work_handle);
if (init_attr->create_flags)
return -EOPNOTSUPP;