From: Chip Coldwell <coldwell@redhat.com>
Subject: [RHEL-5.1 PATCH] bz323231 (regression) cat /proc/scsi/megaraid_sas/0 	panics the machine
Date: Mon, 8 Oct 2007 15:16:33 -0400 (EDT)
Bugzilla: 323231
Message-Id: <alpine.LFD.0.9999.0710081512270.3296@localhost.localdomain>
Changelog: [scsi] megaraid_sas: kabi fix for /proc entries



The KABI workaround for bug 245184 (megaraid_sas: intercept cmd
timeout and throttle io) stuffed a sentinel value into the
scsi_host_template.proc_info pointer:

The SCSI midlayer tests the value of this pointer, and if it is
non-NULL, it will create a /proc/scsi/[driver-name]/[host-number] file
using the proc_info value as a pointer to a read method.  The proposed
solution is to modify the midlayer so that it ignores both null
pointers and pointers containing the sentinel value.

--- linux-2.6.18.x86_64/drivers/scsi/scsi_proc.c~	2007-10-08 13:16:27.872633000 -0400
+++ linux-2.6.18.x86_64/drivers/scsi/scsi_proc.c	2007-10-08 13:13:48.447218000 -0400
@@ -82,7 +82,7 @@ out:
 
 void scsi_proc_hostdir_add(struct scsi_host_template *sht)
 {
-	if (!sht->proc_info)
+	if (!sht->proc_info || sht->proc_info == (void *) RH_EXTENDED_MAGIC)
 		return;
 
 	mutex_lock(&global_host_template_mutex);
@@ -99,7 +99,7 @@ void scsi_proc_hostdir_add(struct scsi_h
 
 void scsi_proc_hostdir_rm(struct scsi_host_template *sht)
 {
-	if (!sht->proc_info)
+	if (!sht->proc_info || sht->proc_info == (void *) RH_EXTENDED_MAGIC)
 		return;
 
 	mutex_lock(&global_host_template_mutex);


-- 
Charles M. "Chip" Coldwell
Senior Software Engineer
Red Hat, Inc
978-392-2426