[8576] in linux-scsi channel archive
Re: Bad segment list supplied to aha1542.c
daemon@ATHENA.MIT.EDU (Andries.Brouwer@cwi.nl)
Sun Apr 9 10:46:19 2000
Date: Sun, 9 Apr 2000 16:34:22 +0200 (MET DST)
From: Andries.Brouwer@cwi.nl
Message-Id: <UTC200004091434.QAA10954.aeb@arend.cwi.nl>
To: eric@andante.org, nigel.chan@alcatel.com.au
Cc: linux-scsi@vger.rutgers.edu
From: "Eric Youngdale" <eric@andante.org>
Sigh. It sounds like some of the queueing code doesn't have sanity
checking. I guess I will have to dust off the 1542 in my test machine and
see if I can reproduce it. I am hoping that this won't be *too* hard to
reproduce and find.
Also on one of my machines I get a panic with aha1542 after
upgrading (downgrading?) to linux-2.3.99pre3.
At first sight it looks like all checks in scsi_merge
are combined in the wrong way - like
if (sth > max1 && sth_else > max2) error;
with && where I would expect ||.
Without investigating any further I made the patch below,
and after applying it the panic was gone.
[Don't apply it blindly - I wrote it without any attempt
to understand the code.]
Andries
----------------------------------------------------------
--- /g1/linux/linux-2.3.99p3/linux/drivers/scsi/scsi_merge.c Fri Mar 17 18:17:26 2000
+++ scsi_merge.c Sun Apr 9 16:01:47 2000
@@ -282,7 +282,7 @@
* This can come up if you get a MEDIUM_ERROR, for example,
* as we will have "completed" all of the sectors up to and
* including the bad sector, and the leftover bit is what
- * we have to do now. This tends to be a rare occurence, so
+ * we have to do now. This tends to be a rare occurrence, so
* we aren't busting our butts to instantiate separate versions
* of this function for the 4 different flag values. We
* probably should, however.
@@ -320,7 +320,7 @@
* scsi.c allocates for this purpose
* min(64,sg_tablesize) entries.
*/
- if (req->nr_segments >= max_segments &&
+ if (req->nr_segments >= max_segments ||
req->nr_segments >= SHpnt->sg_tablesize)
return 0;
req->nr_segments++;
@@ -339,7 +339,7 @@
* check if things fit into sg_tablesize.
*/
if (req->nr_hw_segments >= SHpnt->sg_tablesize ||
- (req->nr_segments >= max_segments &&
+ (req->nr_segments >= max_segments ||
req->nr_segments >= SHpnt->sg_tablesize))
return 0;
if (req->nr_segments >= max_segments)
@@ -609,7 +609,7 @@
/* If it would not fit into prepared memory space for sg chain,
* then don't allow the merge.
*/
- if (req->nr_segments + next->nr_segments - 1 > max_segments &&
+ if (req->nr_segments + next->nr_segments - 1 > max_segments ||
req->nr_segments + next->nr_segments - 1 > SHpnt->sg_tablesize) {
return 0;
}
@@ -674,7 +674,7 @@
}
dont_combine:
#ifdef DMA_CHUNK_SIZE
- if (req->nr_segments + next->nr_segments > max_segments &&
+ if (req->nr_segments + next->nr_segments > max_segments ||
req->nr_segments + next->nr_segments > SHpnt->sg_tablesize) {
return 0;
}
@@ -697,7 +697,7 @@
* Make sure we can fix something that is the sum of the two.
* A slightly stricter test than we had above.
*/
- if (req->nr_segments + next->nr_segments > max_segments &&
+ if (req->nr_segments + next->nr_segments > max_segments ||
req->nr_segments + next->nr_segments > SHpnt->sg_tablesize) {
return 0;
} else {
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
