Re: Linux 2.6.13-rc1

[Hugh Dickins <hugh@veritas.com>]

On Wed, 29 Jun 2005, Tomasz Torcz wrote:
> On Wed, Jun 29, 2005 at 04:23:11PM +0200, Ronny V. Vindenes wrote:
> > 
> > On x86_64 with reiserfs and ext3 on dm (using cfq scheduler) the log
> > is full of this:
> > 
> > Badness in blk_remove_plug at drivers/block/ll_rw_blk.c:1424
> 
>  Also on x86, Reiserfs on LVM2, cfq, on sata_sil; Preemption set to
> Voluntary.

You should find the patch I posted just a little earlier fixes all that...

get_request is now expected to be holding on to queue_lock, with interrupts
disabled, when it returns NULL; but one path forgot that, causing all kinds
of nastiness under swap load - badness backtraces, strange failures, BUGs.

Signed-off-by: Hugh Dickins <hugh@veritas.com>

--- 2.6.13-rc1/drivers/block/ll_rw_blk.c	2005-06-29 11:54:08.000000000 +0100
+++ linux/drivers/block/ll_rw_blk.c	2005-06-29 14:41:04.000000000 +0100
@@ -1917,10 +1917,9 @@ get_rq:
 	 * limit of requests, otherwise we could have thousands of requests
 	 * allocated with any setting of ->nr_requests
 	 */
-	if (rl->count[rw] >= (3 * q->nr_requests / 2)) {
-		spin_unlock_irq(q->queue_lock);
+	if (rl->count[rw] >= (3 * q->nr_requests / 2))
 		goto out;
-	}
+
 	rl->count[rw]++;
 	rl->starved[rw] = 0;
 	if (rl->count[rw] >= queue_congestion_on_threshold(q))