scsi: lpfc: Fix hard lock up NMI in els timeout handling.
authorDick Kennedy <dick.kennedy@broadcom.com>
Tue, 7 Nov 2017 20:59:02 +0000 (12:59 -0800)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 8 Nov 2017 23:25:12 +0000 (18:25 -0500)
System crashed due to a hard lockup at lpfc_els_timeout_handler+0x128.

The els ring's txcmplq list is corrupted: the last element in the list
does not point back the the head causing a loop. Issue is the els
processing path for sli4 hbas are using the hbalock instead of the
ring_lock for removing elements from the txcmplq list.

Use the adapter SLI_REV to determine which lock should be used for
removing iocbqs from the els rings txcmplq.

note: the future refactoring will address this so that we don't have
this ugly type-based lock code.

Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/lpfc/lpfc_sli.c

index 8c37885..98c488a 100644 (file)
@@ -2732,7 +2732,8 @@ lpfc_sli_process_unsol_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
  *
  * This function looks up the iocb_lookup table to get the command iocb
  * corresponding to the given response iocb using the iotag of the
- * response iocb. This function is called with the hbalock held.
+ * response iocb. This function is called with the hbalock held
+ * for sli3 devices or the ring_lock for sli4 devices.
  * This function returns the command iocb object if it finds the command
  * iocb else returns NULL.
  **/
@@ -2828,9 +2829,15 @@ lpfc_sli_process_sol_iocb(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
        unsigned long iflag;
 
        /* Based on the iotag field, get the cmd IOCB from the txcmplq */
-       spin_lock_irqsave(&phba->hbalock, iflag);
+       if (phba->sli_rev == LPFC_SLI_REV4)
+               spin_lock_irqsave(&pring->ring_lock, iflag);
+       else
+               spin_lock_irqsave(&phba->hbalock, iflag);
        cmdiocbp = lpfc_sli_iocbq_lookup(phba, pring, saveq);
-       spin_unlock_irqrestore(&phba->hbalock, iflag);
+       if (phba->sli_rev == LPFC_SLI_REV4)
+               spin_unlock_irqrestore(&pring->ring_lock, iflag);
+       else
+               spin_unlock_irqrestore(&phba->hbalock, iflag);
 
        if (cmdiocbp) {
                if (cmdiocbp->iocb_cmpl) {