scsi: lpfc: Fix target reset failing
authorJames Smart <jsmart2021@gmail.com>
Mon, 4 Jan 2021 18:02:34 +0000 (10:02 -0800)
committerMartin K. Petersen <martin.petersen@oracle.com>
Fri, 8 Jan 2021 04:02:36 +0000 (23:02 -0500)
Target reset is failed by the target as an invalid command.

The Target Reset TMF has been obsoleted in T10 for a while, but continues
to be used. On (newer) devices, the TMF is rejected causing the reset
handler to escalate to adapter resets.

Fix by having Target Reset TMF rejections be translated into a LOGO and
re-PLOGI with the target device. This provides the same semantic action
(although, if the device also supports nvme traffic, it will terminate nvme
traffic as well - but it's still recoverable).

Link: https://lore.kernel.org/r/20210104180240.46824-10-jsmart2021@gmail.com
Co-developed-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/lpfc/lpfc_disc.h
drivers/scsi/lpfc/lpfc_els.c
drivers/scsi/lpfc/lpfc_scsi.c

index ea07afc..4cea61b 100644 (file)
@@ -135,14 +135,17 @@ struct lpfc_nodelist {
        struct lpfc_scsicmd_bkt *lat_data;      /* Latency data */
        uint32_t fc4_prli_sent;
        uint32_t fc4_xpt_flags;
+       uint32_t upcall_flags;
 #define NLP_WAIT_FOR_UNREG    0x1
 #define SCSI_XPT_REGD         0x2
 #define NVME_XPT_REGD         0x4
+#define NLP_WAIT_FOR_LOGO     0x2
 
 
        uint32_t nvme_fb_size; /* NVME target's supported byte cnt */
 #define NVME_FB_BIT_SHIFT 9    /* PRLI Rsp first burst in 512B units. */
        uint32_t nlp_defer_did;
+       wait_queue_head_t *logo_waitq;
 };
 
 struct lpfc_node_rrq {
index e099caa..c944f22 100644 (file)
@@ -2817,6 +2817,7 @@ lpfc_cmpl_els_logo(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
        IOCB_t *irsp;
        unsigned long flags;
        uint32_t skip_recovery = 0;
+       int wake_up_waiter = 0;
 
        /* we pass cmdiocb to state machine which needs rspiocb as well */
        cmdiocb->context_un.rsp_iocb = rspiocb;
@@ -2824,6 +2825,10 @@ lpfc_cmpl_els_logo(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
        irsp = &(rspiocb->iocb);
        spin_lock_irq(&ndlp->lock);
        ndlp->nlp_flag &= ~NLP_LOGO_SND;
+       if (ndlp->upcall_flags & NLP_WAIT_FOR_LOGO) {
+               wake_up_waiter = 1;
+               ndlp->upcall_flags &= ~NLP_WAIT_FOR_LOGO;
+       }
        spin_unlock_irq(&ndlp->lock);
 
        lpfc_debugfs_disc_trc(vport, LPFC_DISC_TRC_ELS_CMD,
@@ -2889,6 +2894,8 @@ out:
         * Initiator, we are assuming the NPortID is not going to change.
         */
 
+       if (wake_up_waiter && ndlp->logo_waitq)
+               wake_up(ndlp->logo_waitq);
        /*
         * If the node is a target, the handling attempts to recover the port.
         * For any other port type, the rpi is unregistered as an implicit
index 78f34b1..4ee890d 100644 (file)
@@ -5924,6 +5924,8 @@ lpfc_target_reset_handler(struct scsi_cmnd *cmnd)
        struct lpfc_scsi_event_header scsi_event;
        int status;
        u32 logit = LOG_FCP;
+       unsigned long flags;
+       DECLARE_WAIT_QUEUE_HEAD_ONSTACK(waitq);
 
        rdata = lpfc_rport_data_from_scsi_device(cmnd->device);
        if (!rdata || !rdata->pnode) {
@@ -5942,10 +5944,10 @@ lpfc_target_reset_handler(struct scsi_cmnd *cmnd)
                lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
                        "0722 Target Reset rport failure: rdata x%px\n", rdata);
                if (pnode) {
-                       spin_lock_irq(&pnode->lock);
+                       spin_lock_irqsave(&pnode->lock, flags);
                        pnode->nlp_flag &= ~NLP_NPR_ADISC;
                        pnode->nlp_fcp_info &= ~NLP_FCP_2_DEVICE;
-                       spin_unlock_irq(&pnode->lock);
+                       spin_unlock_irqrestore(&pnode->lock, flags);
                }
                lpfc_reset_flush_io_context(vport, tgt_id, lun_id,
                                          LPFC_CTX_TGT);
@@ -5965,6 +5967,38 @@ lpfc_target_reset_handler(struct scsi_cmnd *cmnd)
                                        FCP_TARGET_RESET);
        if (status != SUCCESS)
                logit =  LOG_TRACE_EVENT;
+       spin_lock_irqsave(&pnode->lock, flags);
+       if (status != SUCCESS &&
+           (!(pnode->upcall_flags & NLP_WAIT_FOR_LOGO)) &&
+            !pnode->logo_waitq) {
+               pnode->logo_waitq = &waitq;
+               pnode->nlp_fcp_info &= ~NLP_FCP_2_DEVICE;
+               pnode->nlp_flag |= NLP_ISSUE_LOGO;
+               pnode->upcall_flags |= NLP_WAIT_FOR_LOGO;
+               spin_unlock_irqrestore(&pnode->lock, flags);
+               lpfc_unreg_rpi(vport, pnode);
+               wait_event_timeout(waitq,
+                                  (!(pnode->upcall_flags & NLP_WAIT_FOR_LOGO)),
+                                   msecs_to_jiffies(vport->cfg_devloss_tmo *
+                                   1000));
+
+               if (pnode->upcall_flags & NLP_WAIT_FOR_LOGO) {
+                       lpfc_printf_vlog(vport, KERN_ERR, LOG_TRACE_EVENT,
+                               "0725 SCSI layer TGTRST failed & LOGO TMO "
+                               " (%d, %llu) return x%x\n", tgt_id,
+                                lun_id, status);
+                       spin_lock_irqsave(&pnode->lock, flags);
+                       pnode->upcall_flags &= ~NLP_WAIT_FOR_LOGO;
+               } else {
+                       spin_lock_irqsave(&pnode->lock, flags);
+               }
+               pnode->logo_waitq = NULL;
+               spin_unlock_irqrestore(&pnode->lock, flags);
+               status = SUCCESS;
+       } else {
+               status = FAILED;
+               spin_unlock_irqrestore(&pnode->lock, flags);
+       }
 
        lpfc_printf_vlog(vport, KERN_ERR, logit,
                         "0723 SCSI layer issued Target Reset (%d, %llu) "