[PATCH 07/14] lpfc: Rework I/O flush ordering when unloading driver

[Justin Tee <justintee8345@gmail.com>]

The lpfc_els_abort routine has a code path that cancels outstanding
I/Os on the ELS ring when attempted aborts fail.  The failed aborts are
queued to a drv_cmpl_list and then cancelled after the ELS pring->txcmplq
is fully traversed.  However if the abort failure returns IOCB_ABORTING,
then the driver should not have cancelled it.  Doing so starts two threads
working on the same iocb and ndlp, leading to unintended race conditions.

Fix by capturing the IOCB_ABORTING return value in lpfc_els_abort and not
adding it to the list of iocbs for cancelling.  We should allow the iocb
scheduled for abort to complete naturally.  This avoids simultaneous
threads acting on the same iocb and ndlp objects.

The lpfc_free_iocb_list is moved to execute after lpfc_sli4_hba_unset
allowing the routine to flush I/O before freeing it.  And, in
lpfc_pci_remove_one_s4 a call to flush the phba->wq is added.  This makes
the unload logic consistent with offline handling logic.

Signed-off-by: Justin Tee <justintee8345@gmail.com>
---
 drivers/scsi/lpfc/lpfc_init.c      | 16 ++++++++++++++--
 drivers/scsi/lpfc/lpfc_nportdisc.c | 11 +++++++++--
 2 files changed, 23 insertions(+), 4 deletions(-)

diff --git a/drivers/scsi/lpfc/lpfc_init.c b/drivers/scsi/lpfc/lpfc_init.c
index 968a25235a2d..44f213f42347 100644
--- a/drivers/scsi/lpfc/lpfc_init.c
+++ b/drivers/scsi/lpfc/lpfc_init.c
@@ -13515,6 +13515,9 @@ lpfc_sli4_hba_unset(struct lpfc_hba *phba)
 	/* Stop the SLI4 device port */
 	if (phba->pport)
 		phba->pport->work_port_events = 0;
+
+	/* All IO completed and queues released. Free the IOCBs. */
+	lpfc_free_iocb_list(phba);
 }
 
 /*
@@ -14949,11 +14952,20 @@ lpfc_pci_remove_one_s4(struct pci_dev *pdev)
 
 	/* Perform scsi free before driver resource_unset since scsi
 	 * buffers are released to their corresponding pools here.
+	 * lpfc_sli4_hba_unset() issues aborts via lpfc_sli_hba_iocb_abort(),
+	 * which allocates abort IOCBs from phba->lpfc_iocb_list; the pool
+	 * must still exist, so lpfc_free_iocb_list() runs only after unset.
 	 */
 	lpfc_io_free(phba);
-	lpfc_free_iocb_list(phba);
-	lpfc_sli4_hba_unset(phba);
 
+	/* Flush the PHBA WQ - there could be a race with ELS IOs while lpfc
+	 * is unloading.  This stops a race between completions, aborts and
+	 * resource recovery.
+	 */
+	if (phba->wq)
+		flush_workqueue(phba->wq);
+
+	lpfc_sli4_hba_unset(phba);
 	lpfc_unset_driver_resource_phase2(phba);
 	lpfc_sli4_driver_resource_unset(phba);
 
diff --git a/drivers/scsi/lpfc/lpfc_nportdisc.c b/drivers/scsi/lpfc/lpfc_nportdisc.c
index 2c8d995a45bf..f917a5bcfd02 100644
--- a/drivers/scsi/lpfc/lpfc_nportdisc.c
+++ b/drivers/scsi/lpfc/lpfc_nportdisc.c
@@ -255,8 +255,9 @@ lpfc_els_abort(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
 	spin_lock_irq(&phba->hbalock);
 	if (phba->sli_rev == LPFC_SLI_REV4)
 		spin_lock(&pring->ring_lock);
+
 	list_for_each_entry_safe(iocb, next_iocb, &pring->txcmplq, list) {
-	/* Add to abort_list on on NDLP match. */
+		/* Add to abort_list on NDLP match. */
 		if (lpfc_check_sli_ndlp(phba, pring, iocb, ndlp))
 			list_add_tail(&iocb->dlist, &abort_list);
 	}
@@ -271,7 +272,13 @@ lpfc_els_abort(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
 		retval = lpfc_sli_issue_abort_iotag(phba, pring, iocb, NULL);
 		spin_unlock_irq(&phba->hbalock);
 
-		if (retval && test_bit(FC_UNLOADING, &phba->pport->load_flag)) {
+		/* An abort that fails here is just cancelled when the driver is
+		 * going offline.  However, if the abort failure is because the
+		 * IOCB is already getting aborted, don't cancel.  Just let it
+		 * complete.
+		 */
+		if (test_bit(FC_UNLOADING, &phba->pport->load_flag) &&
+		    retval && retval != IOCB_ABORTING) {
 			list_del_init(&iocb->list);
 			list_add_tail(&iocb->list, &drv_cmpl_list);
 		}
-- 
2.38.0