From: Roland Dreier <rol...@purestorage.com>
If a SCSI device's old state is already SDEV_RUNNING and we're moving
to the same SDEV_RUNNING state, still wake the blockdev queue in
scsi_internal_device_unblock(). This fixes a case where we silently
hang SCSI commands forever during device discovery. One way this can
happen is when mpt2sas is discovering a reasonably big SAS topology,
and the sd driver has queued up a bunch of sd_probe_async() instances
that are queueing SCSI commands to various devices.
If at the same time a SAS fabric event goes to the HBA, what can
happen is the following:
- mpt2sas calls _scsih_block_io_all_device() ->
scsi_internal_device_block(sdev)
(In response to some HBA firmware event like
MPI2_EVENT_SAS_BROADCAST_PRIMITIVE)
Now sdev state is SDEV_BLOCK and blockdev queue has QUEUE_FLAG_STOPPED
set.
- Someone like scsi_add_lun() calls scsi_device_set_state(sdev,
SDEV_RUNNING)
(SCSI bus scanning runs asynchronously to firmware event handling)
Now sdev state is SDEV_RUNNING but blockdev queue still has
QUEUE_FLAG_STOPPED set
- mpt2sas calls _scsih_ublock_io_all_device() ->
scsi_internal_device_unblock(sdev, SDEV_RUNNING)
(Finishes handling the firmware event)
With the old scsi_lib code, scsi_internal_device_unblock() will return
an error at this point because the sdev state is already SDEV_RUNNING.
This means we skip the call to blk_start_queue() and never actually
start executing commands again.
Fix this by still going ahead and finishing scsi_internal_device_unblock()
even if the sdev state is already SDEV_RUNNING.
Signed-off-by: Roland Dreier <rol...@purestorage.com>
---
drivers/scsi/scsi_lib.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 765398c..75108ea 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -2495,7 +2495,9 @@ scsi_internal_device_unblock(struct scsi_device *sdev,
else
sdev->sdev_state = SDEV_CREATED;
} else if (sdev->sdev_state != SDEV_CANCEL &&
- sdev->sdev_state != SDEV_OFFLINE)
+ sdev->sdev_state != SDEV_OFFLINE &&
+ (sdev->sdev_state != SDEV_RUNNING ||
+ new_state != SDEV_RUNNING))
return -EINVAL;
spin_lock_irqsave(q->queue_lock, flags);
--
1.8.1.2
--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
