Commit db09de98 authored by Alexey Marchuk's avatar Alexey Marchuk Committed by Tomasz Zawadzki
Browse files

nvmf/rdma: Update Work Completion error logging 



1. Retrieve actual IBV state when we receive WC with bad status
2. Don't log an error if WC status is IBV_WC_WR_FLUSH_ERR. This
means that we are performing qpair cleanup and this WC is expected.

Change-Id: Id23634092f537861e66ca0f83ab79db9e052507b
Signed-off-by: default avatarAlexey Marchuk <alexeymar@mellanox.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3736


Community-CI: Mellanox Build Bot
Reviewed-by: default avatarShuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: default avatar <dongx.yi@intel.com>
Reviewed-by: default avatarMichael Haeuptle <michaelhaeuptle@gmail.com>
Reviewed-by: default avatarBen Walker <benjamin.walker@intel.com>
Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
parent 47ce1fe3

lib/nvmf/rdma.c

+37 −10
Original line number Diff line number Diff line
@@ -3750,6 +3750,41 @@ _poller_submit_sends(struct spdk_nvmf_rdma_transport *rtransport, 
	}

}



static const char *

nvmf_rdma_wr_type_str(enum spdk_nvmf_rdma_wr_type wr_type)

{

	switch (wr_type) {

	case RDMA_WR_TYPE_RECV:

		return "RECV";

	case RDMA_WR_TYPE_SEND:

		return "SEND";

	case RDMA_WR_TYPE_DATA:

		return "DATA";

	default:

		SPDK_ERRLOG("Unknown WR type %d\n", wr_type);

		SPDK_UNREACHABLE();

	}

}



static inline void

nvmf_rdma_log_wc_status(struct spdk_nvmf_rdma_qpair *rqpair, struct ibv_wc *wc)

{

	enum spdk_nvmf_rdma_wr_type wr_type = ((struct spdk_nvmf_rdma_wr *)wc->wr_id)->type;



	if (wc->status == IBV_WC_WR_FLUSH_ERR) {

		/* If qpair is in ERR state, we will receive completions for all posted and not completed

		 * Work Requests with IBV_WC_WR_FLUSH_ERR status. Don't log an error in that case */

		SPDK_DEBUGLOG(SPDK_LOG_RDMA,

			      "Error on CQ %p, (qp state %d ibv_state %d) request 0x%lu, type %s, status: (%d): %s\n",

			      rqpair->poller->cq, rqpair->qpair.state, rqpair->ibv_state, wc->wr_id,

			      nvmf_rdma_wr_type_str(wr_type), wc->status, ibv_wc_status_str(wc->status));

	} else {

		SPDK_ERRLOG("Error on CQ %p, (qp state %d ibv_state %d) request 0x%lu, type %s, status: (%d): %s\n",

			    rqpair->poller->cq, rqpair->qpair.state, rqpair->ibv_state, wc->wr_id,

			    nvmf_rdma_wr_type_str(wr_type), wc->status, ibv_wc_status_str(wc->status));

	}

}



static int

nvmf_rdma_poller_poll(struct spdk_nvmf_rdma_transport *rtransport,

		      struct spdk_nvmf_rdma_poller *rpoller)
@@ -3873,16 +3908,8 @@ nvmf_rdma_poller_poll(struct spdk_nvmf_rdma_transport *rtransport, 


		/* Handle error conditions */

		if (wc[i].status) {

			if ((rdma_wr->type == RDMA_WR_TYPE_RECV && !rpoller->srq)) {

				/* When we don't use SRQ and close a qpair, we will receive completions with error

				 * status for all posted ibv_recv_wrs. This is expected and we don't want to log

				 * an error in that case. */

				SPDK_DEBUGLOG(SPDK_LOG_RDMA, "Error on CQ %p, request 0x%lu, type %d, status: (%d): %s\n",

					      rpoller->cq, wc[i].wr_id, rdma_wr->type, wc[i].status, ibv_wc_status_str(wc[i].status));

			} else {

				SPDK_ERRLOG("Error on CQ %p, request 0x%lu, type %d, status: (%d): %s\n",

					    rpoller->cq, wc[i].wr_id, rdma_wr->type, wc[i].status, ibv_wc_status_str(wc[i].status));

			}

			nvmf_rdma_update_ibv_state(rqpair);

			nvmf_rdma_log_wc_status(rqpair, &wc[i]);



			error = true;