scsi: storvsc: Update error logging
authorMichael Kelley <mikelley@microsoft.com>
Fri, 4 Jun 2021 17:21:02 +0000 (10:21 -0700)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 16 Jun 2021 02:21:44 +0000 (22:21 -0400)
When an I/O error is reported by the underlying Hyper-V host, current code
provides details only when the logging level is set to WARN, making it more
difficult to diagnose problems in live customer situations. Fix this by
reporting details at ERROR level, which is the default.  Also add more
information, including the Hyper-V error code, and the tag # so that the
message can be matched with messages at the SCSI and blk-mq levels.

Also, sense information logging is inconsistent and duplicative. The
existence of sense info is first logged at WARN level, and then full sense
info is logged at ERROR level. Fix this by removing the logging of the
existence of sense info, and change the logging of full sense info to WARN
level in favor of letting the generic SCSI layer handle such logging. With
the change to WARN level, it's no longer necessary to filter out as noise
any NOT READY sense info generated by the virtual DVD device.

Link: https://lore.kernel.org/r/1622827263-12516-2-git-send-email-mikelley@microsoft.com
Signed-off-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/storvsc_drv.c

index d73a0f4bcf8363b78a1c4caba7e4e8fc08c740ee..b4ea67169e817604c50a6b3706f351991e617b67 100644 (file)
@@ -1090,6 +1090,7 @@ static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
        struct Scsi_Host *host;
        u32 payload_sz = cmd_request->payload_sz;
        void *payload = cmd_request->payload;
+       bool sense_ok;
 
        host = stor_dev->host;
 
@@ -1099,11 +1100,10 @@ static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request,
        scmnd->result = vm_srb->scsi_status;
 
        if (scmnd->result) {
-               if (scsi_normalize_sense(scmnd->sense_buffer,
-                               SCSI_SENSE_BUFFERSIZE, &sense_hdr) &&
-                   !(sense_hdr.sense_key == NOT_READY &&
-                                sense_hdr.asc == 0x03A) &&
-                   do_logging(STORVSC_LOGGING_ERROR))
+               sense_ok = scsi_normalize_sense(scmnd->sense_buffer,
+                               SCSI_SENSE_BUFFERSIZE, &sense_hdr);
+
+               if (sense_ok && do_logging(STORVSC_LOGGING_WARN))
                        scsi_print_sense_hdr(scmnd->device, "storvsc",
                                             &sense_hdr);
        }
@@ -1173,18 +1173,19 @@ static void storvsc_on_io_completion(struct storvsc_device *stor_device,
 
        if (vstor_packet->vm_srb.scsi_status != 0 ||
            vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS)
-               storvsc_log(device, STORVSC_LOGGING_WARN,
-                       "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
+               storvsc_log(device, STORVSC_LOGGING_ERROR,
+                       "tag#%d cmd 0x%x status: scsi 0x%x srb 0x%x hv 0x%x\n",
+                       request->cmd->request->tag,
                        stor_pkt->vm_srb.cdb[0],
                        vstor_packet->vm_srb.scsi_status,
-                       vstor_packet->vm_srb.srb_status);
+                       vstor_packet->vm_srb.srb_status,
+                       vstor_packet->status);
 
        if (vstor_packet->vm_srb.scsi_status == SAM_STAT_CHECK_CONDITION &&
            (vstor_packet->vm_srb.srb_status & SRB_STATUS_AUTOSENSE_VALID))
                memcpy(request->cmd->sense_buffer,
                       vstor_packet->vm_srb.sense_data,
                       stor_pkt->vm_srb.sense_info_length);
-       }
 
        stor_pkt->vm_srb.data_transfer_length =
                vstor_packet->vm_srb.data_transfer_length;