2007-01-15 21:51:21

by linas

[permalink] [raw]
Subject: [PATCH] adjust use of unplug in elevator code


Hi Chris, Jens,
Can you look at this, and push upstream if this looks reasonable
to you? It fixes a bug I've been tripping over.

--linas


A flag was recently added to the elevator code to avoid
performing an unplug when reuests are being re-queued.
The goal of this flag was to avoid a deep recursion that
can occur when re-queueing requests after a SCSI device/host
reset. See http://lkml.org/lkml/2006/5/17/254

However, that fix added the flag near the bottom of a case
statement, where an earlier break (in an if statement) could
transport one out of the case, without setting the flag.
This patch sets the flag earlier in the case statement.

I re-discovered the deep recursion recently during testing;
I was told that it was a known problem, and the fix to it was
in the kernel I was testing. Indeed it was ... but it didn't
fix the bug. With the patch below, I no longer see the bug.

Signed-off by: Linas Vepstas <[email protected]>
Cc: Jens Axboe <[email protected]>
Cc: Chris Wright <[email protected]>

----
block/elevator.c | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)

Index: linux-2.6.20-rc4/block/elevator.c
===================================================================
--- linux-2.6.20-rc4.orig/block/elevator.c 2007-01-15 14:16:03.000000000 -0600
+++ linux-2.6.20-rc4/block/elevator.c 2007-01-15 14:20:04.000000000 -0600
@@ -590,6 +590,12 @@ void elv_insert(request_queue_t *q, stru
*/
rq->cmd_flags |= REQ_SOFTBARRIER;

+ /*
+ * Most requeues happen because of a busy condition,
+ * don't force unplug of the queue for that case.
+ */
+ unplug_it = 0;
+
if (q->ordseq == 0) {
list_add(&rq->queuelist, &q->queue_head);
break;
@@ -604,11 +610,6 @@ void elv_insert(request_queue_t *q, stru
}

list_add_tail(&rq->queuelist, pos);
- /*
- * most requeues happen because of a busy condition, don't
- * force unplug of the queue for that case.
- */
- unplug_it = 0;
break;

default:


2007-01-16 00:40:46

by Jens Axboe

[permalink] [raw]
Subject: Re: [PATCH] adjust use of unplug in elevator code

On Mon, Jan 15 2007, Linas Vepstas wrote:
>
> Hi Chris, Jens,
> Can you look at this, and push upstream if this looks reasonable
> to you? It fixes a bug I've been tripping over.
>
> --linas
>
>
> A flag was recently added to the elevator code to avoid
> performing an unplug when reuests are being re-queued.
> The goal of this flag was to avoid a deep recursion that
> can occur when re-queueing requests after a SCSI device/host
> reset. See http://lkml.org/lkml/2006/5/17/254
>
> However, that fix added the flag near the bottom of a case
> statement, where an earlier break (in an if statement) could
> transport one out of the case, without setting the flag.
> This patch sets the flag earlier in the case statement.
>
> I re-discovered the deep recursion recently during testing;
> I was told that it was a known problem, and the fix to it was
> in the kernel I was testing. Indeed it was ... but it didn't
> fix the bug. With the patch below, I no longer see the bug.
>
> Signed-off by: Linas Vepstas <[email protected]>
> Cc: Jens Axboe <[email protected]>
> Cc: Chris Wright <[email protected]>
>
> ----
> block/elevator.c | 11 ++++++-----
> 1 file changed, 6 insertions(+), 5 deletions(-)
>
> Index: linux-2.6.20-rc4/block/elevator.c
> ===================================================================
> --- linux-2.6.20-rc4.orig/block/elevator.c 2007-01-15 14:16:03.000000000 -0600
> +++ linux-2.6.20-rc4/block/elevator.c 2007-01-15 14:20:04.000000000 -0600
> @@ -590,6 +590,12 @@ void elv_insert(request_queue_t *q, stru
> */
> rq->cmd_flags |= REQ_SOFTBARRIER;
>
> + /*
> + * Most requeues happen because of a busy condition,
> + * don't force unplug of the queue for that case.
> + */
> + unplug_it = 0;
> +
> if (q->ordseq == 0) {
> list_add(&rq->queuelist, &q->queue_head);
> break;
> @@ -604,11 +610,6 @@ void elv_insert(request_queue_t *q, stru
> }
>
> list_add_tail(&rq->queuelist, pos);
> - /*
> - * most requeues happen because of a busy condition, don't
> - * force unplug of the queue for that case.
> - */
> - unplug_it = 0;
> break;

Ah, yes it definitely should be moved up, thanks for that!

Acked-by: Jens Axboe <[email protected]>

I'll get this merged for 2.6.21.

--
Jens Axboe