On 13/10/2022 09:47, Daisuke Matsuda wrote:
Requesting nodes do not handle a reported error correctly if it is generated in the middle of multi-packet Read responses, and the node tries to resend the request endlessly. Let completer terminate the connection in that case. Signed-off-by: Daisuke Matsuda <matsuda-daisuke@xxxxxxxxxxx> --- FOR REVIEWERS: I referred to IB Specification Vol 1-Revision-1.5 to make this patch. Please see Ch.9.9.2.2, Ch.9.9.2.4.2 and Table 59. drivers/infiniband/sw/rxe/rxe_comp.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/drivers/infiniband/sw/rxe/rxe_comp.c b/drivers/infiniband/sw/rxe/rxe_comp.c index fb0c008af78c..c9170dd99f3a 100644 --- a/drivers/infiniband/sw/rxe/rxe_comp.c +++ b/drivers/infiniband/sw/rxe/rxe_comp.c @@ -200,6 +200,10 @@ static inline enum comp_state check_psn(struct rxe_qp *qp, */ if (pkt->psn == wqe->last_psn) return COMPST_COMP_ACK; + else if (pkt->opcode == IB_OPCODE_RC_ACKNOWLEDGE && + (qp->comp.opcode == IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST || + qp->comp.opcode == IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE))
When IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST or IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE will be assigned to qp->comp.opcode ? I wonder if "(pkt->opcode == IB_OPCODE_RC_ACKNOWLEDGE) " is enough ? Thanks Zhijian
+ return COMPST_CHECK_ACK; else return COMPST_DONE; } else if ((diff > 0) && (wqe->mask & WR_ATOMIC_OR_READ_MASK)) { @@ -228,6 +232,10 @@ static inline enum comp_state check_ack(struct rxe_qp *qp,case IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST:case IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE: + /* Check NAK code to handle a remote error */ + if (pkt->opcode == IB_OPCODE_RC_ACKNOWLEDGE) + break; + if (pkt->opcode != IB_OPCODE_RC_RDMA_READ_RESPONSE_MIDDLE && pkt->opcode != IB_OPCODE_RC_RDMA_READ_RESPONSE_LAST) { /* read retries of partial data may restart from
