* [PATCH] 2.4.27-pre2: another loss of SCSI I/O initiative and system freeze
@ 2004-05-04 18:04 Martin Peschke3
2004-05-05 0:31 ` Doug Ledford
0 siblings, 1 reply; 3+ messages in thread
From: Martin Peschke3 @ 2004-05-04 18:04 UTC (permalink / raw)
To: marcelo.tosatti, dledford; +Cc: linux-scsi, andmike, axboe, garloff
Marcelo, Doug,
Please consider the following patch for inclusion. It fixes a severe
loss-of-initiative SCSI problem that can cause a system freeze.
The problem is that blk_finish_sectors() needs a valid block
queue pointer in order to kick more I/Os off, while _scsi_insert_special()
appears to have a license to kill the block queue pointers of SCSI
requests.
Hence, the problem is likely to occur if SCSI commands are retried,
e.g. after a QUEUE_FULL or BUSY response. Since this hitch slowly
builds up by losing more and more valid pointers and generating
misfires more often, I also suspect a SCSI performance decrease
prior to a total freeze.
The fix is to have blk_finish_sectors() take a block queue pointer
instead of a request pointer.
We found the problem during tests and subsequent debugging
with the zfcp driver on the SUSE Linux Enterprise Server.
This is Jens' version of the fix, which made it into newer SUSE kernels.
Please apply.
Martin
diff -ur linux-2.4.27-pre2-ref/drivers/block/ll_rw_blk.c linux-2.4.27-pre2/drivers/block/ll_rw_blk.c
--- linux-2.4.27-pre2-ref/drivers/block/ll_rw_blk.c 2004-04-14 15:05:29.000000000 +0200
+++ linux-2.4.27-pre2/drivers/block/ll_rw_blk.c 2004-05-04 19:13:42.000000000 +0200
@@ -1465,7 +1465,7 @@
if ((bh = req->bh) != NULL) {
nsect = bh->b_size >> 9;
blk_finished_io(nsect);
- blk_finished_sectors(req, nsect);
+ blk_finished_sectors(req->q, nsect);
req->bh = bh->b_reqnext;
bh->b_reqnext = NULL;
bh->b_end_io(bh, uptodate);
diff -ur linux-2.4.27-pre2-ref/drivers/scsi/scsi_lib.c linux-2.4.27-pre2/drivers/scsi/scsi_lib.c
--- linux-2.4.27-pre2-ref/drivers/scsi/scsi_lib.c 2004-04-14 15:05:31.000000000 +0200
+++ linux-2.4.27-pre2/drivers/scsi/scsi_lib.c 2004-05-04 19:13:42.000000000 +0200
@@ -378,7 +378,7 @@
if ((bh = req->bh) != NULL) {
nsect = bh->b_size >> 9;
blk_finished_io(nsect);
- blk_finished_sectors(req, nsect);
+ blk_finished_sectors(q, nsect);
req->bh = bh->b_reqnext;
bh->b_reqnext = NULL;
sectors -= nsect;
diff -ur linux-2.4.27-pre2-ref/include/linux/blkdev.h linux-2.4.27-pre2/include/linux/blkdev.h
--- linux-2.4.27-pre2-ref/include/linux/blkdev.h 2004-02-18 14:36:32.000000000 +0100
+++ linux-2.4.27-pre2/include/linux/blkdev.h 2004-05-04 19:13:42.000000000 +0200
@@ -323,9 +323,8 @@
}
}
-static inline void blk_finished_sectors(struct request *rq, int count)
+static inline void blk_finished_sectors(request_queue_t *q, int count)
{
- request_queue_t *q = rq->q;
if (q && q->can_throttle) {
atomic_sub(count, &q->nr_sectors);
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] 2.4.27-pre2: another loss of SCSI I/O initiative and system freeze
2004-05-04 18:04 [PATCH] 2.4.27-pre2: another loss of SCSI I/O initiative and system freeze Martin Peschke3
@ 2004-05-05 0:31 ` Doug Ledford
2004-05-05 5:43 ` Jens Axboe
0 siblings, 1 reply; 3+ messages in thread
From: Doug Ledford @ 2004-05-05 0:31 UTC (permalink / raw)
To: Martin Peschke3
Cc: Marcelo Tosatti, linux-scsi mailing list, andmike, axboe, garloff
On Tue, 2004-05-04 at 14:04, Martin Peschke3 wrote:
>
>
> Marcelo, Doug,
>
> Please consider the following patch for inclusion. It fixes a severe
> loss-of-initiative SCSI problem that can cause a system freeze.
Marcelo, I have no objection to this, but I haven't personally verified
it either (and my bk trees are out of date so integrating it yourself is
probably quickest).
> The problem is that blk_finish_sectors() needs a valid block
> queue pointer in order to kick more I/Os off, while _scsi_insert_special()
> appears to have a license to kill the block queue pointers of SCSI
> requests.
> Hence, the problem is likely to occur if SCSI commands are retried,
> e.g. after a QUEUE_FULL or BUSY response. Since this hitch slowly
> builds up by losing more and more valid pointers and generating
> misfires more often, I also suspect a SCSI performance decrease
> prior to a total freeze.
>
> The fix is to have blk_finish_sectors() take a block queue pointer
> instead of a request pointer.
>
> We found the problem during tests and subsequent debugging
> with the zfcp driver on the SUSE Linux Enterprise Server.
> This is Jens' version of the fix, which made it into newer SUSE kernels.
>
> Please apply.
>
> Martin
>
>
> diff -ur linux-2.4.27-pre2-ref/drivers/block/ll_rw_blk.c linux-2.4.27-pre2/drivers/block/ll_rw_blk.c
> --- linux-2.4.27-pre2-ref/drivers/block/ll_rw_blk.c 2004-04-14 15:05:29.000000000 +0200
> +++ linux-2.4.27-pre2/drivers/block/ll_rw_blk.c 2004-05-04 19:13:42.000000000 +0200
> @@ -1465,7 +1465,7 @@
> if ((bh = req->bh) != NULL) {
> nsect = bh->b_size >> 9;
> blk_finished_io(nsect);
> - blk_finished_sectors(req, nsect);
> + blk_finished_sectors(req->q, nsect);
> req->bh = bh->b_reqnext;
> bh->b_reqnext = NULL;
> bh->b_end_io(bh, uptodate);
> diff -ur linux-2.4.27-pre2-ref/drivers/scsi/scsi_lib.c linux-2.4.27-pre2/drivers/scsi/scsi_lib.c
> --- linux-2.4.27-pre2-ref/drivers/scsi/scsi_lib.c 2004-04-14 15:05:31.000000000 +0200
> +++ linux-2.4.27-pre2/drivers/scsi/scsi_lib.c 2004-05-04 19:13:42.000000000 +0200
> @@ -378,7 +378,7 @@
> if ((bh = req->bh) != NULL) {
> nsect = bh->b_size >> 9;
> blk_finished_io(nsect);
> - blk_finished_sectors(req, nsect);
> + blk_finished_sectors(q, nsect);
> req->bh = bh->b_reqnext;
> bh->b_reqnext = NULL;
> sectors -= nsect;
> diff -ur linux-2.4.27-pre2-ref/include/linux/blkdev.h linux-2.4.27-pre2/include/linux/blkdev.h
> --- linux-2.4.27-pre2-ref/include/linux/blkdev.h 2004-02-18 14:36:32.000000000 +0100
> +++ linux-2.4.27-pre2/include/linux/blkdev.h 2004-05-04 19:13:42.000000000 +0200
> @@ -323,9 +323,8 @@
> }
> }
>
> -static inline void blk_finished_sectors(struct request *rq, int count)
> +static inline void blk_finished_sectors(request_queue_t *q, int count)
> {
> - request_queue_t *q = rq->q;
> if (q && q->can_throttle) {
> atomic_sub(count, &q->nr_sectors);
--
Doug Ledford <dledford@redhat.com> 919-754-3700 x44233
Red Hat, Inc.
1801 Varsity Dr.
Raleigh, NC 27606
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] 2.4.27-pre2: another loss of SCSI I/O initiative and system freeze
2004-05-05 0:31 ` Doug Ledford
@ 2004-05-05 5:43 ` Jens Axboe
0 siblings, 0 replies; 3+ messages in thread
From: Jens Axboe @ 2004-05-05 5:43 UTC (permalink / raw)
To: Doug Ledford
Cc: Martin Peschke3, Marcelo Tosatti, linux-scsi mailing list,
andmike, garloff
On Tue, May 04 2004, Doug Ledford wrote:
> On Tue, 2004-05-04 at 14:04, Martin Peschke3 wrote:
> >
> >
> > Marcelo, Doug,
> >
> > Please consider the following patch for inclusion. It fixes a severe
> > loss-of-initiative SCSI problem that can cause a system freeze.
>
> Marcelo, I have no objection to this, but I haven't personally verified
> it either (and my bk trees are out of date so integrating it yourself is
> probably quickest).
It's really a slight API change, not a SCSI change per se. It's fine to
apply.
--
Jens Axboe
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2004-05-05 5:44 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-05-04 18:04 [PATCH] 2.4.27-pre2: another loss of SCSI I/O initiative and system freeze Martin Peschke3
2004-05-05 0:31 ` Doug Ledford
2004-05-05 5:43 ` Jens Axboe
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox