Proposed fix for scsi related boot crash

Hi

We have fedora bug report where system crash at boot when SCSI card is plugged into the system:
https://bugzilla.redhat.com/show_bug.cgi?id=781625

Since it is 3.1.2 -> 3.1.4 regression, it was easy to find out that the problem is caused by:

commit 4e6c82b3614a18740ef63109d58743a359266daf

Author: James Bottomley <[email protected]>
Date:   Mon Nov 7 08:51:24 2011 -0600

[SCSI] fix WARNING: at drivers/scsi/scsi_lib.c:1704

User confirmed that reverting the commit fixed the problem (and generates lot's of scsi/scsi_lib.c:1704 warnings).

The remain question is how to properly fix. My proposition is to revert 4e6c82b361 and nullify sdev->request_queue->queuedata what should fix the warning.

Thanks
Stanislaw

index 89da43f..fcc8b53 100644
@@ -319,7 +319,15 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
 	return sdev;
 
 out_device_destroy:
-	__scsi_remove_device(sdev);
+	scsi_device_set_state(sdev, SDEV_DEL);
+	transport_destroy_device(&sdev->sdev_gendev);
+	put_device(&sdev->sdev_dev);
+
+	/* stop the block layer for our device */
+	sdev->request_queue->queuedata = NULL;
+	scsi_free_queue(sdev->request_queue);
+
+	put_device(&sdev->sdev_gendev);
 out:
 	if (display_failure_msg)
 		printk(ALLOC_FAILURE_MSG, __func__);

It will be fixed in stable by

commit cced5041ed5a2d1352186510944b0ddfbdbe4c0b

Author: Stratos Psomadakis <[email protected]>
Date:   Sun Dec 4 02:23:54 2011 +0200

[SCSI] sym53c8xx: Fix NULL pointer dereference in slave_destroy

James