From 411de511c6943554cdc4173c3f522029db2f75c7 Mon Sep 17 00:00:00 2001 From: James Smart Date: Tue, 30 Jan 2018 15:58:52 -0800 Subject: [PATCH] scsi: lpfc: Fix RQ empty firmware trap When nvme target deferred receive logic waits for exchange resources, the corresponding receive buffer is not replenished with the hardware. This can result in a lack of asynchronous receive buffer resources in the hardware, resulting in a "2885 Port Status Event: ... error 1=0x52004a01 ..." message. Correct by replenishing the buffer whenenver the deferred logic kicks in. Update corresponding debug messages and statistics as well. [mkp: applied by hand] Signed-off-by: Dick Kennedy Signed-off-by: James Smart Reviewed-by: Hannes Reinecke Signed-off-by: Martin K. Petersen --- drivers/scsi/lpfc/lpfc_attr.c | 6 ++++++ drivers/scsi/lpfc/lpfc_mem.c | 10 +++++++--- drivers/scsi/lpfc/lpfc_nvmet.c | 31 +++++++++++++++++++++---------- drivers/scsi/lpfc/lpfc_nvmet.h | 7 +++++-- drivers/scsi/lpfc/lpfc_sli.c | 12 ++++++++++++ 5 files changed, 51 insertions(+), 15 deletions(-) diff --git a/drivers/scsi/lpfc/lpfc_attr.c b/drivers/scsi/lpfc/lpfc_attr.c index ac77081e6e9e..c63a028385ab 100644 --- a/drivers/scsi/lpfc/lpfc_attr.c +++ b/drivers/scsi/lpfc/lpfc_attr.c @@ -259,6 +259,12 @@ lpfc_nvme_info_show(struct device *dev, struct device_attribute *attr, atomic_read(&tgtp->xmt_abort_rsp), atomic_read(&tgtp->xmt_abort_rsp_error)); + len += snprintf(buf + len, PAGE_SIZE - len, + "DELAY: ctx %08x fod %08x wqfull %08x\n", + atomic_read(&tgtp->defer_ctx), + atomic_read(&tgtp->defer_fod), + atomic_read(&tgtp->defer_wqfull)); + /* Calculate outstanding IOs */ tot = atomic_read(&tgtp->rcv_fcp_cmd_drop); tot += atomic_read(&tgtp->xmt_fcp_release); diff --git a/drivers/scsi/lpfc/lpfc_mem.c b/drivers/scsi/lpfc/lpfc_mem.c index 87c08ff37ddd..60078e61da5e 100644 --- a/drivers/scsi/lpfc/lpfc_mem.c +++ b/drivers/scsi/lpfc/lpfc_mem.c @@ -753,12 +753,16 @@ lpfc_rq_buf_free(struct lpfc_hba *phba, struct lpfc_dmabuf *mp) drqe.address_hi = putPaddrHigh(rqb_entry->dbuf.phys); rc = lpfc_sli4_rq_put(rqb_entry->hrq, rqb_entry->drq, &hrqe, &drqe); if (rc < 0) { + (rqbp->rqb_free_buffer)(phba, rqb_entry); lpfc_printf_log(phba, KERN_ERR, LOG_INIT, - "6409 Cannot post to RQ %d: %x %x\n", + "6409 Cannot post to HRQ %d: %x %x %x " + "DRQ %x %x\n", rqb_entry->hrq->queue_id, rqb_entry->hrq->host_index, - rqb_entry->hrq->hba_index); - (rqbp->rqb_free_buffer)(phba, rqb_entry); + rqb_entry->hrq->hba_index, + rqb_entry->hrq->entry_count, + rqb_entry->drq->host_index, + rqb_entry->drq->hba_index); } else { list_add_tail(&rqb_entry->hbuf.list, &rqbp->rqb_buffer_list); rqbp->buffer_count++; diff --git a/drivers/scsi/lpfc/lpfc_nvmet.c b/drivers/scsi/lpfc/lpfc_nvmet.c index 9c2acf90212c..0539585d32d4 100644 --- a/drivers/scsi/lpfc/lpfc_nvmet.c +++ b/drivers/scsi/lpfc/lpfc_nvmet.c @@ -270,8 +270,6 @@ lpfc_nvmet_ctxbuf_post(struct lpfc_hba *phba, struct lpfc_nvmet_ctxbuf *ctx_buf) "NVMET RCV BUSY: xri x%x sz %d " "from %06x\n", oxid, size, sid); - /* defer repost rcv buffer till .defer_rcv callback */ - ctxp->flag &= ~LPFC_NVMET_DEFER_RCV_REPOST; atomic_inc(&tgtp->rcv_fcp_cmd_out); return; } @@ -837,6 +835,7 @@ lpfc_nvmet_xmt_fcp_op(struct nvmet_fc_target_port *tgtport, list_add_tail(&nvmewqeq->list, &wq->wqfull_list); wq->q_flag |= HBA_NVMET_WQFULL; spin_unlock_irqrestore(&pring->ring_lock, iflags); + atomic_inc(&lpfc_nvmep->defer_wqfull); return 0; } @@ -975,11 +974,9 @@ lpfc_nvmet_defer_rcv(struct nvmet_fc_target_port *tgtport, tgtp = phba->targetport->private; atomic_inc(&tgtp->rcv_fcp_cmd_defer); - if (ctxp->flag & LPFC_NVMET_DEFER_RCV_REPOST) - lpfc_rq_buf_free(phba, &nvmebuf->hbuf); /* repost */ - else - nvmebuf->hrq->rqbp->rqb_free_buffer(phba, nvmebuf); - ctxp->flag &= ~LPFC_NVMET_DEFER_RCV_REPOST; + + /* Free the nvmebuf since a new buffer already replaced it */ + nvmebuf->hrq->rqbp->rqb_free_buffer(phba, nvmebuf); } static struct nvmet_fc_target_template lpfc_tgttemplate = { @@ -1309,6 +1306,9 @@ lpfc_nvmet_create_targetport(struct lpfc_hba *phba) atomic_set(&tgtp->xmt_abort_sol, 0); atomic_set(&tgtp->xmt_abort_rsp, 0); atomic_set(&tgtp->xmt_abort_rsp_error, 0); + atomic_set(&tgtp->defer_ctx, 0); + atomic_set(&tgtp->defer_fod, 0); + atomic_set(&tgtp->defer_wqfull, 0); } return error; } @@ -1810,6 +1810,8 @@ lpfc_nvmet_unsol_fcp_buffer(struct lpfc_hba *phba, lpfc_nvmeio_data(phba, "NVMET FCP RCV: xri x%x sz %d CPU %02x\n", oxid, size, smp_processor_id()); + tgtp = (struct lpfc_nvmet_tgtport *)phba->targetport->private; + if (!ctx_buf) { /* Queue this NVME IO to process later */ spin_lock_irqsave(&phba->sli4_hba.nvmet_io_wait_lock, iflag); @@ -1825,10 +1827,11 @@ lpfc_nvmet_unsol_fcp_buffer(struct lpfc_hba *phba, lpfc_post_rq_buffer( phba, phba->sli4_hba.nvmet_mrq_hdr[qno], phba->sli4_hba.nvmet_mrq_data[qno], 1, qno); + + atomic_inc(&tgtp->defer_ctx); return; } - tgtp = (struct lpfc_nvmet_tgtport *)phba->targetport->private; payload = (uint32_t *)(nvmebuf->dbuf.virt); sid = sli4_sid_from_fc_hdr(fc_hdr); @@ -1892,12 +1895,20 @@ lpfc_nvmet_unsol_fcp_buffer(struct lpfc_hba *phba, /* Processing of FCP command is deferred */ if (rc == -EOVERFLOW) { + /* + * Post a brand new DMA buffer to RQ and defer + * freeing rcv buffer till .defer_rcv callback + */ + qno = nvmebuf->idx; + lpfc_post_rq_buffer( + phba, phba->sli4_hba.nvmet_mrq_hdr[qno], + phba->sli4_hba.nvmet_mrq_data[qno], 1, qno); + lpfc_nvmeio_data(phba, "NVMET RCV BUSY: xri x%x sz %d from %06x\n", oxid, size, sid); - /* defer reposting rcv buffer till .defer_rcv callback */ - ctxp->flag |= LPFC_NVMET_DEFER_RCV_REPOST; atomic_inc(&tgtp->rcv_fcp_cmd_out); + atomic_inc(&tgtp->defer_fod); return; } ctxp->rqb_buffer = nvmebuf; diff --git a/drivers/scsi/lpfc/lpfc_nvmet.h b/drivers/scsi/lpfc/lpfc_nvmet.h index 354cce443c9f..5da35de5ea45 100644 --- a/drivers/scsi/lpfc/lpfc_nvmet.h +++ b/drivers/scsi/lpfc/lpfc_nvmet.h @@ -72,7 +72,6 @@ struct lpfc_nvmet_tgtport { atomic_t xmt_fcp_rsp_aborted; atomic_t xmt_fcp_rsp_drop; - /* Stats counters - lpfc_nvmet_xmt_fcp_abort */ atomic_t xmt_fcp_xri_abort_cqe; atomic_t xmt_fcp_abort; @@ -81,6 +80,11 @@ struct lpfc_nvmet_tgtport { atomic_t xmt_abort_unsol; atomic_t xmt_abort_rsp; atomic_t xmt_abort_rsp_error; + + /* Stats counters - defer IO */ + atomic_t defer_ctx; + atomic_t defer_fod; + atomic_t defer_wqfull; }; struct lpfc_nvmet_ctx_info { @@ -131,7 +135,6 @@ struct lpfc_nvmet_rcv_ctx { #define LPFC_NVMET_XBUSY 0x4 /* XB bit set on IO cmpl */ #define LPFC_NVMET_CTX_RLS 0x8 /* ctx free requested */ #define LPFC_NVMET_ABTS_RCV 0x10 /* ABTS received on exchange */ -#define LPFC_NVMET_DEFER_RCV_REPOST 0x20 /* repost to RQ on defer rcv */ #define LPFC_NVMET_DEFER_WQFULL 0x40 /* Waiting on a free WQE */ struct rqb_dmabuf *rqb_buffer; struct lpfc_nvmet_ctxbuf *ctxbuf; diff --git a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c index fbda2fbcbfec..8b2919a553d6 100644 --- a/drivers/scsi/lpfc/lpfc_sli.c +++ b/drivers/scsi/lpfc/lpfc_sli.c @@ -6535,9 +6535,11 @@ lpfc_post_rq_buffer(struct lpfc_hba *phba, struct lpfc_queue *hrq, struct lpfc_rqe hrqe; struct lpfc_rqe drqe; struct lpfc_rqb *rqbp; + unsigned long flags; struct rqb_dmabuf *rqb_buffer; LIST_HEAD(rqb_buf_list); + spin_lock_irqsave(&phba->hbalock, flags); rqbp = hrq->rqbp; for (i = 0; i < count; i++) { /* IF RQ is already full, don't bother */ @@ -6561,6 +6563,15 @@ lpfc_post_rq_buffer(struct lpfc_hba *phba, struct lpfc_queue *hrq, drqe.address_hi = putPaddrHigh(rqb_buffer->dbuf.phys); rc = lpfc_sli4_rq_put(hrq, drq, &hrqe, &drqe); if (rc < 0) { + lpfc_printf_log(phba, KERN_ERR, LOG_INIT, + "6421 Cannot post to HRQ %d: %x %x %x " + "DRQ %x %x\n", + hrq->queue_id, + hrq->host_index, + hrq->hba_index, + hrq->entry_count, + drq->host_index, + drq->hba_index); rqbp->rqb_free_buffer(phba, rqb_buffer); } else { list_add_tail(&rqb_buffer->hbuf.list, @@ -6568,6 +6579,7 @@ lpfc_post_rq_buffer(struct lpfc_hba *phba, struct lpfc_queue *hrq, rqbp->buffer_count++; } } + spin_unlock_irqrestore(&phba->hbalock, flags); return 1; }