RDMA/hns: Avoid unnecessary memset on WQEs in post_send

All fields of WQE will be rewrote, so the memset is unnecessary. And when
SQ is working in OWNER mode, the pipeline may prefetch the WQEs beyond PI,
the memset operation may flip the owner bit too early, then the pipeline
may get a wrong WQ.

Link: https://lore.kernel.org/r/1612517974-31867-11-git-send-email-liweihang@huawei.com
Signed-off-by: Lang Cheng <chenglang@huawei.com>
Signed-off-by: Weihang Li <liweihang@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
This commit is contained in:
Lang Cheng 2021-02-05 17:39:32 +08:00 committed by Jason Gunthorpe
parent 993703370a
commit 62490fd5a8

View File

@ -469,7 +469,6 @@ static inline int set_ud_wqe(struct hns_roce_qp *qp,
int ret; int ret;
valid_num_sge = calc_wr_sge_num(wr, &msg_len); valid_num_sge = calc_wr_sge_num(wr, &msg_len);
memset(ud_sq_wqe, 0, sizeof(*ud_sq_wqe));
ret = set_ud_opcode(ud_sq_wqe, wr); ret = set_ud_opcode(ud_sq_wqe, wr);
if (WARN_ON(ret)) if (WARN_ON(ret))
@ -575,7 +574,6 @@ static inline int set_rc_wqe(struct hns_roce_qp *qp,
int ret; int ret;
valid_num_sge = calc_wr_sge_num(wr, &msg_len); valid_num_sge = calc_wr_sge_num(wr, &msg_len);
memset(rc_sq_wqe, 0, sizeof(*rc_sq_wqe));
rc_sq_wqe->msg_len = cpu_to_le32(msg_len); rc_sq_wqe->msg_len = cpu_to_le32(msg_len);