Fix for aic7xxx problems in BK latest

Fix for aic7xxx problems in BK latest

[James Bottomley]

It turns out the scsi_device refcounting and list lockdown altered the
conditions under which scsi_report_bus_reset() works, since it calls
shost_for_each_device, which now takes the host lock, that lock cannot
be held while calling the report functions.  However, the assumption by
aic is that it can.

I fixed the problem by using the no lock version of the iterator.

James

===== drivers/scsi/scsi_error.c 1.65 vs edited =====
--- 1.65/drivers/scsi/scsi_error.c	Sun Sep 21 12:49:36 2003
+++ edited/drivers/scsi/scsi_error.c	Sun Nov 23 23:39:09 2003
@@ -1622,7 +1622,7 @@
 {
 	struct scsi_device *sdev;
 
-	shost_for_each_device(sdev, shost) {
+	__shost_for_each_device(sdev, shost) {
 		if (channel == sdev->channel) {
 			sdev->was_reset = 1;
 			sdev->expecting_cc_ua = 1;
@@ -1656,7 +1656,7 @@
 {
 	struct scsi_device *sdev;
 
-	shost_for_each_device(sdev, shost) {
+	__shost_for_each_device(sdev, shost) {
 		if (channel == sdev->channel &&
 		    target == sdev->id) {
 			sdev->was_reset = 1;

Re: Fix for aic7xxx problems in BK latest

[Christoph Hellwig]

On Sun, Nov 23, 2003 at 11:53:20PM -0600, James Bottomley wrote:
> It turns out the scsi_device refcounting and list lockdown altered the
> conditions under which scsi_report_bus_reset() works, since it calls
> shost_for_each_device, which now takes the host lock, that lock cannot
> be held while calling the report functions.  However, the assumption by
> aic is that it can.
> 
> I fixed the problem by using the no lock version of the iterator.

Well, that makes the problem go away but adds races again, the right
fix is to rename you versions with an __ and make the normal versions
take the host_lock around that - we don't actually need the refcounting
here because we just set the two flags so holding the lock over it is
fine.