scsi: lpfc: Clear outstanding active mailbox during PCI function reset

Mailbox commands sent via ioctl/bsg from user applications may be
interrupted from processing by a concurrently triggered PCI function
reset. The command will not generate a completion due to the reset.  This
results in a user application hang waiting for the mailbox command to
complete.

Resolve by changing the function reset handler to detect that there was an
outstanding mailbox command and simulate a mailbox completion.  Add some
additional debug when a mailbox command times out.

Link: https://lore.kernel.org/r/20210707184351.67872-13-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

2 files changed, 40 insertions, 3 deletions

diff --git a/drivers/scsi/lpfc/lpfc_init.c b/drivers/scsi/lpfc/lpfc_init.c
index 55f720ac6c8c..4fd9a8098e86 100644
--- a/drivers/scsi/lpfc/lpfc_init.c
+++ b/drivers/scsi/lpfc/lpfc_init.c
@@ -1852,6 +1852,7 @@ lpfc_sli4_port_sta_fn_reset(struct lpfc_hba *phba, int mbx_action,
 {
 	int rc;
 	uint32_t intr_mode;
+	LPFC_MBOXQ_t *mboxq;
 
 	if (bf_get(lpfc_sli_intf_if_type, &phba->sli4_hba.sli_intf) >=
 	    LPFC_SLI_INTF_IF_TYPE_2) {
@@ -1871,11 +1872,19 @@ lpfc_sli4_port_sta_fn_reset(struct lpfc_hba *phba, int mbx_action,
 				"Recovery...\n");
 
 	/* If we are no wait, the HBA has been reset and is not
-	 * functional, thus we should clear LPFC_SLI_ACTIVE flag.
+	 * functional, thus we should clear
+	 * (LPFC_SLI_ACTIVE | LPFC_SLI_MBOX_ACTIVE) flags.
 	 */
 	if (mbx_action == LPFC_MBX_NO_WAIT) {
 		spin_lock_irq(&phba->hbalock);
 		phba->sli.sli_flag &= ~LPFC_SLI_ACTIVE;
+		if (phba->sli.mbox_active) {
+			mboxq = phba->sli.mbox_active;
+			mboxq->u.mb.mbxStatus = MBX_NOT_FINISHED;
+			__lpfc_mbox_cmpl_put(phba, mboxq);
+			phba->sli.sli_flag &= ~LPFC_SLI_MBOX_ACTIVE;
+			phba->sli.mbox_active = NULL;
+		}
 		spin_unlock_irq(&phba->hbalock);
 	}
 
diff --git a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c
index 1fdb2232729f..c34240819d92 100644
--- a/drivers/scsi/lpfc/lpfc_sli.c
+++ b/drivers/scsi/lpfc/lpfc_sli.c
@@ -8794,8 +8794,11 @@ static int
 lpfc_sli4_async_mbox_block(struct lpfc_hba *phba)
 {
 	struct lpfc_sli *psli = &phba->sli;
+	LPFC_MBOXQ_t *mboxq;
 	int rc = 0;
 	unsigned long timeout = 0;
+	u32 sli_flag;
+	u8 cmd, subsys, opcode;
 
 	/* Mark the asynchronous mailbox command posting as blocked */
 	spin_lock_irq(&phba->hbalock);
@@ -8813,12 +8816,37 @@ lpfc_sli4_async_mbox_block(struct lpfc_hba *phba)
 	if (timeout)
 		lpfc_sli4_process_missed_mbox_completions(phba);
 
-	/* Wait for the outstnading mailbox command to complete */
+	/* Wait for the outstanding mailbox command to complete */
 	while (phba->sli.mbox_active) {
 		/* Check active mailbox complete status every 2ms */
 		msleep(2);
 		if (time_after(jiffies, timeout)) {
-			/* Timeout, marked the outstanding cmd not complete */
+			/* Timeout, mark the outstanding cmd not complete */
+
+			/* Sanity check sli.mbox_active has not completed or
+			 * cancelled from another context during last 2ms sleep,
+			 * so take hbalock to be sure before logging.
+			 */
+			spin_lock_irq(&phba->hbalock);
+			if (phba->sli.mbox_active) {
+				mboxq = phba->sli.mbox_active;
+				cmd = mboxq->u.mb.mbxCommand;
+				subsys = lpfc_sli_config_mbox_subsys_get(phba,
+									 mboxq);
+				opcode = lpfc_sli_config_mbox_opcode_get(phba,
+									 mboxq);
+				sli_flag = psli->sli_flag;
+				spin_unlock_irq(&phba->hbalock);
+				lpfc_printf_log(phba, KERN_ERR, LOG_TRACE_EVENT,
+						"2352 Mailbox command x%x "
+						"(x%x/x%x) sli_flag x%x could "
+						"not complete\n",
+						cmd, subsys, opcode,
+						sli_flag);
+			} else {
+				spin_unlock_irq(&phba->hbalock);
+			}
+
 			rc = 1;
 			break;
 		}