@@ -705,8 +705,6 @@ static int hns_roce_u_v2_post_send(struct ibv_qp *ibvqp, struct ibv_send_wr *wr,
rc_sq_wqe->rkey = htole32(wr->wr.atomic.rkey);
rc_sq_wqe->va =
htole64(wr->wr.atomic.remote_addr);
- wqe += sizeof(struct hns_roce_v2_wqe_data_seg);
- set_atomic_seg(wqe, wr);
break;
case IBV_WR_ATOMIC_FETCH_AND_ADD:
@@ -717,8 +715,6 @@ static int hns_roce_u_v2_post_send(struct ibv_qp *ibvqp, struct ibv_send_wr *wr,
rc_sq_wqe->rkey = htole32(wr->wr.atomic.rkey);
rc_sq_wqe->va =
htole64(wr->wr.atomic.remote_addr);
- wqe += sizeof(struct hns_roce_v2_wqe_data_seg);
- set_atomic_seg(wqe, wr);
break;
default:
roce_set_field(rc_sq_wqe->byte_4,
@@ -737,14 +733,13 @@ static int hns_roce_u_v2_post_send(struct ibv_qp *ibvqp, struct ibv_send_wr *wr,
break;
}
+ dseg = wqe;
if (wr->opcode == IBV_WR_ATOMIC_FETCH_AND_ADD ||
- wr->opcode == IBV_WR_ATOMIC_CMP_AND_SWP)
- dseg = wqe - sizeof(struct hns_roce_v2_wqe_data_seg);
- else
- dseg = wqe;
-
- /* Inline */
- if (wr->send_flags & IBV_SEND_INLINE && wr->num_sge) {
+ wr->opcode == IBV_WR_ATOMIC_CMP_AND_SWP) {
+ set_data_seg_v2(dseg, wr->sg_list);
+ wqe += sizeof(struct hns_roce_v2_wqe_data_seg);
+ set_atomic_seg(wqe, wr);
+ } else if (wr->send_flags & IBV_SEND_INLINE && wr->num_sge) {
if (le32toh(rc_sq_wqe->msg_len) > qp->max_inline_data) {
ret = EINVAL;
*bad_wr = wr;
The atomic operation not support to inline. Besides, the standard atomic operation only support a sge and the sge place in wqe. This patch mainly ajdust the code. Fix: d92b0f5("libhns: Add atomic support for hip08 user mode") Signed-off-by: Lijun Ou <oulijun@huawei.com> --- providers/hns/hns_roce_u_hw_v2.c | 17 ++++++----------- 1 file changed, 6 insertions(+), 11 deletions(-)