]> err.no Git - linux-2.6/commitdiff
[SCSI] lpfc 8.1.1 : Fixes to error handlers
authorJames.Smart@Emulex.Com <James.Smart@Emulex.Com>
Mon, 28 Nov 2005 16:42:05 +0000 (11:42 -0500)
committerJames Bottomley <jejb@mulgrave.(none)>
Wed, 14 Dec 2005 01:26:47 +0000 (18:26 -0700)
- Release task management command before counting outstanding commands.
  TMF was being erroneously counted as an active outstanding command.
- Serialize EH calls and block requests when EH function is running.

Signed-off-by: James Smart <James.Smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
drivers/scsi/lpfc/lpfc.h
drivers/scsi/lpfc/lpfc_init.c
drivers/scsi/lpfc/lpfc_scsi.c

index 3062b39fbdb97ca7e12e0a66a7af6f43c346cb5b..dc73a2f8f12cb3a2f13b993c08649fdabc39acf3 100644 (file)
@@ -167,6 +167,7 @@ struct lpfc_hba {
        dma_addr_t slim2p_mapping;
        uint16_t pci_cfg_value;
 
+       struct semaphore hba_can_block;
        uint32_t hba_state;
 
 #define LPFC_INIT_START           1    /* Initial state after board reset */
index 4d4e217edd84514a890f5a3cb725d62543eaed0d..dfd59d21ec49fb787d5101061d9651fd30c264f8 100644 (file)
@@ -1345,7 +1345,7 @@ lpfc_pci_probe_one(struct pci_dev *pdev, const struct pci_device_id *pid)
                goto out_put_host;
 
        host->unique_id = phba->brd_no;
-
+       init_MUTEX(&phba->hba_can_block);
        INIT_LIST_HEAD(&phba->ctrspbuflist);
        INIT_LIST_HEAD(&phba->rnidrspbuflist);
        INIT_LIST_HEAD(&phba->freebufList);
index a4d8455de4464962235a538c8453b4821aaba9b9..7dc7810b748239e6c865514f39ea064f807ff556 100644 (file)
 #define LPFC_ABORT_WAIT  2
 
 
+static inline void
+lpfc_block_requests(struct lpfc_hba * phba)
+{
+       down(&phba->hba_can_block);
+       scsi_block_requests(phba->host);
+}
+
+static inline void
+lpfc_unblock_requests(struct lpfc_hba * phba)
+{
+       scsi_unblock_requests(phba->host);
+       up(&phba->hba_can_block);
+}
+
 /*
  * This routine allocates a scsi buffer, which contains all the necessary
  * information needed to initiate a SCSI I/O.  The non-DMAable buffer region
@@ -774,6 +788,7 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
        unsigned int loop_count = 0;
        int ret = SUCCESS;
 
+       lpfc_block_requests(phba);
        spin_lock_irq(shost->host_lock);
 
        lpfc_cmd = (struct lpfc_scsi_buf *)cmnd->host_scribble;
@@ -853,6 +868,7 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
                        cmnd->device->lun, cmnd->serial_number);
 
        spin_unlock_irq(shost->host_lock);
+       lpfc_unblock_requests(phba);
 
        return ret;
 }
@@ -866,9 +882,11 @@ lpfc_reset_lun_handler(struct scsi_cmnd *cmnd)
        struct lpfc_iocbq *iocbq, *iocbqrsp;
        struct lpfc_rport_data *rdata = cmnd->device->hostdata;
        struct lpfc_nodelist *pnode = rdata->pnode;
+       uint32_t cmd_result = 0, cmd_status = 0;
        int ret = FAILED;
        int cnt, loopcnt;
 
+       lpfc_block_requests(phba);
        spin_lock_irq(shost->host_lock);
        /*
         * If target is not in a MAPPED state, delay the reset until
@@ -912,26 +930,28 @@ lpfc_reset_lun_handler(struct scsi_cmnd *cmnd)
        if (ret == IOCB_SUCCESS)
                ret = SUCCESS;
 
-       lpfc_cmd->result = iocbqrsp->iocb.un.ulpWord[4];
-       lpfc_cmd->status = iocbqrsp->iocb.ulpStatus;
-       if (lpfc_cmd->status == IOSTAT_LOCAL_REJECT)
-               if (lpfc_cmd->result & IOERR_DRVR_MASK)
-                       lpfc_cmd->status = IOSTAT_DRIVER_REJECT;
+
+       cmd_result = iocbqrsp->iocb.un.ulpWord[4];
+       cmd_status = iocbqrsp->iocb.ulpStatus;
+
+       lpfc_sli_release_iocbq(phba, iocbqrsp);
+       lpfc_release_scsi_buf(phba, lpfc_cmd);
 
        /*
-        * All outstanding txcmplq I/Os should have been aborted by the target.
+        * All outstanding txcmplq I/Os should have been aborted by the device.
         * Unfortunately, some targets do not abide by this forcing the driver
         * to double check.
         */
-       lpfc_sli_abort_iocb(phba, &phba->sli.ring[phba->sli.fcp_ring],
-                           cmnd->device->id, cmnd->device->lun, 0,
-                           LPFC_CTX_LUN);
-
+       cnt = lpfc_sli_sum_iocb(phba, &phba->sli.ring[phba->sli.fcp_ring],
+                               cmnd->device->id, cmnd->device->lun,
+                               LPFC_CTX_LUN);
+       if (cnt)
+               lpfc_sli_abort_iocb(phba,
+                                   &phba->sli.ring[phba->sli.fcp_ring],
+                                   cmnd->device->id, cmnd->device->lun,
+                                   0, LPFC_CTX_LUN);
        loopcnt = 0;
-       while((cnt = lpfc_sli_sum_iocb(phba,
-                                      &phba->sli.ring[phba->sli.fcp_ring],
-                                      cmnd->device->id, cmnd->device->lun,
-                                      LPFC_CTX_LUN))) {
+       while(cnt) {
                spin_unlock_irq(phba->host->host_lock);
                schedule_timeout_uninterruptible(LPFC_RESET_WAIT*HZ);
                spin_lock_irq(phba->host->host_lock);
@@ -939,6 +959,11 @@ lpfc_reset_lun_handler(struct scsi_cmnd *cmnd)
                if (++loopcnt
                    > (2 * phba->cfg_nodev_tmo)/LPFC_RESET_WAIT)
                        break;
+
+               cnt = lpfc_sli_sum_iocb(phba,
+                                       &phba->sli.ring[phba->sli.fcp_ring],
+                                       cmnd->device->id, cmnd->device->lun,
+                                       LPFC_CTX_LUN);
        }
 
        if (cnt) {
@@ -948,18 +973,16 @@ lpfc_reset_lun_handler(struct scsi_cmnd *cmnd)
                ret = FAILED;
        }
 
-       lpfc_sli_release_iocbq(phba, iocbqrsp);
-
 out_free_scsi_buf:
        lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
                        "%d:0713 SCSI layer issued LUN reset (%d, %d) "
                        "Data: x%x x%x x%x\n",
-                       phba->brd_no, lpfc_cmd->pCmd->device->id,
-                       lpfc_cmd->pCmd->device->lun, ret, lpfc_cmd->status,
-                       lpfc_cmd->result);
-       lpfc_release_scsi_buf(phba, lpfc_cmd);
+                       phba->brd_no, cmnd->device->id,cmnd->device->lun,
+                       ret, cmd_status, cmd_result);
+
 out:
        spin_unlock_irq(shost->host_lock);
+       lpfc_unblock_requests(phba);
        return ret;
 }
 
@@ -975,6 +998,7 @@ lpfc_reset_bus_handler(struct scsi_cmnd *cmnd)
        unsigned int midlayer_id = 0;
        struct lpfc_scsi_buf * lpfc_cmd;
 
+       lpfc_block_requests(phba);
        spin_lock_irq(shost->host_lock);
 
        lpfc_cmd = lpfc_sli_get_scsi_buf (phba);
@@ -1008,18 +1032,31 @@ lpfc_reset_bus_handler(struct scsi_cmnd *cmnd)
                lpfc_cmd->pCmd->device->hostdata = ndlp->rport->dd_data;
                ret = lpfc_scsi_tgt_reset(lpfc_cmd, phba);
                if (ret != SUCCESS) {
-                       lpfc_printf_log(phba, KERN_INFO, LOG_FCP,
+                       lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
                                "%d:0713 Bus Reset on target %d failed\n",
                                phba->brd_no, i);
                        err_count++;
                }
        }
 
+       if (err_count == 0)
+               ret = SUCCESS;
+
+       lpfc_release_scsi_buf(phba, lpfc_cmd);
+
+       /*
+        * All outstanding txcmplq I/Os should have been aborted by
+        * the targets.  Unfortunately, some targets do not abide by
+        * this forcing the driver to double check.
+        */
        cmnd->device->id = midlayer_id;
+       cnt = lpfc_sli_sum_iocb(phba, &phba->sli.ring[phba->sli.fcp_ring],
+                               0, 0, LPFC_CTX_HOST);
+       if (cnt)
+               lpfc_sli_abort_iocb(phba, &phba->sli.ring[phba->sli.fcp_ring],
+                                   0, 0, 0, LPFC_CTX_HOST);
        loopcnt = 0;
-       while((cnt = lpfc_sli_sum_iocb(phba,
-                               &phba->sli.ring[phba->sli.fcp_ring],
-                               0, 0, LPFC_CTX_HOST))) {
+       while(cnt) {
                spin_unlock_irq(phba->host->host_lock);
                schedule_timeout_uninterruptible(LPFC_RESET_WAIT*HZ);
                spin_lock_irq(phba->host->host_lock);
@@ -1027,25 +1064,19 @@ lpfc_reset_bus_handler(struct scsi_cmnd *cmnd)
                if (++loopcnt
                    > (2 * phba->cfg_nodev_tmo)/LPFC_RESET_WAIT)
                        break;
+
+               cnt = lpfc_sli_sum_iocb(phba,
+                                       &phba->sli.ring[phba->sli.fcp_ring],
+                                       0, 0, LPFC_CTX_HOST);
        }
 
        if (cnt) {
-               /* flush all outstanding commands on the host */
-               i = lpfc_sli_abort_iocb(phba,
-                               &phba->sli.ring[phba->sli.fcp_ring], 0, 0, 0,
-                               LPFC_CTX_HOST);
-
-               lpfc_printf_log(phba, KERN_INFO, LOG_FCP,
+               lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
                   "%d:0715 Bus Reset I/O flush failure: cnt x%x left x%x\n",
                   phba->brd_no, cnt, i);
-       }
-
-       if (cnt == 0)
-               ret = SUCCESS;
-       else
                ret = FAILED;
+       }
 
-       lpfc_release_scsi_buf(phba, lpfc_cmd);
        lpfc_printf_log(phba,
                        KERN_ERR,
                        LOG_FCP,
@@ -1053,6 +1084,7 @@ lpfc_reset_bus_handler(struct scsi_cmnd *cmnd)
                        phba->brd_no, ret);
 out:
        spin_unlock_irq(shost->host_lock);
+       lpfc_unblock_requests(phba);
        return ret;
 }