OSDN Git Service

scsi: lpfc: Fix soft lockup in lpfc worker thread during LIP testing
authorJames Smart <jsmart2021@gmail.com>
Tue, 30 Jan 2018 23:58:54 +0000 (15:58 -0800)
committerMartin K. Petersen <martin.petersen@oracle.com>
Mon, 12 Feb 2018 16:43:23 +0000 (11:43 -0500)
During link bounce testing in a point-to-point topology, the host may
enter a soft lockup on the lpfc_worker thread:

    Call Trace:
     lpfc_work_done+0x1f3/0x1390 [lpfc]
     lpfc_do_work+0x16f/0x180 [lpfc]
     kthread+0xc7/0xe0
     ret_from_fork+0x3f/0x70

The driver was simultaneously setting a combination of flags that caused
lpfc_do_work()to effectively spin between slow path work and new event
data, causing the lockup.

Ensure in the typical wq completions, that new event data flags are set
if the slow path flag is running. The slow path will eventually
reschedule the wq handling.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/lpfc/lpfc_hbadisc.c

index b159a5c..9265906 100644 (file)
@@ -696,8 +696,9 @@ lpfc_work_done(struct lpfc_hba *phba)
                      phba->hba_flag & HBA_SP_QUEUE_EVT)) {
                if (pring->flag & LPFC_STOP_IOCB_EVENT) {
                        pring->flag |= LPFC_DEFERRED_RING_EVENT;
-                       /* Set the lpfc data pending flag */
-                       set_bit(LPFC_DATA_READY, &phba->data_flags);
+                       /* Preserve legacy behavior. */
+                       if (!(phba->hba_flag & HBA_SP_QUEUE_EVT))
+                               set_bit(LPFC_DATA_READY, &phba->data_flags);
                } else {
                        if (phba->link_state >= LPFC_LINK_UP ||
                            phba->link_flag & LS_MDS_LOOPBACK) {