Re: Commit 7eaceaccab5f40 causing boot hang.

[Jens Axboe <jaxboe@fusionio.com>]

On 2011-03-29 10:52, Rob Landley wrote:
> On 03/29/2011 01:35 AM, Jens Axboe wrote:
>> On 2011-03-29 08:19, Rob Landley wrote:
>>> Using the attached config (run through oldconfig and holding down
>>> enter), 2.6.38 boots but current -git hangs.  I've bisected it to this
>>> commit:
>>>
>>> 7eaceaccab5f40bbfda044629a6298616aeaed50 is the first bad commit
>>> commit 7eaceaccab5f40bbfda044629a6298616aeaed50
>>> Author: Jens Axboe <jaxboe@fusionio.com>
>>> Date:   Thu Mar 10 08:52:07 2011 +0100
>>>
>>>     block: remove per-queue plugging
>>>
>>>     Code has been converted over to the new explicit on-stack plugging,
>>>     and delay users have been converted to use the new API for that.
>>>     So lets kill off the old plugging along with aops->sync_page().
>>>
>>>     Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
>>>
>>> I suspect it broke CONFIG_IDE_GD_ATA.  (The symptoms change a bit in
>>> later versions, when that was first introduced it hung right after
>>> discovering hda, now it makes it to freeing kernel memory and then hangs
>>> trying to launch init.  But "the disk hangs a lot" still seems to
>>> describe the problem.)
>>
>> That is the case even with 9ced0b95b3993660e7990cf6bf68c1a80f31544e,
>> which was otherwise supposed to fix this queue stall?
>>
> 
> The first time I tried it, that version did manage to launch init, and
> then hung two lines later with ".udev/ already exists on the static
> /dev! ..." and the cursor still on the end of that line.  Could be an
> unrelated bug, hard to tell.
> 
> I tried it three more times and it hung right after "freeing unused
> kernel memory", with no output from init.  (I also tried adding
> "init=/bin/bash" to the kernel command line and it still hung after the
> "freeing unused kernel memory" line, with no output from bash.  Pressed
> a lot of keys, nothing echoed back.)
> 
> This is an improvement over hanging earlier in the boot, but userspace
> isn't really happy at the moment.
> 
> I'm booting all this under kvm or qemu, by the way:
> 
> qemu-system-x86_64 -m 1024 -kernel arch/x86/boot/bzImage \
>   -hda ~/sid.ext3 -append "root=/dev/hda rw"
> 
> Sometimes with init=/bin/bash in that last quoted bit.  The root
> filesystem's debian sid but that's probably not relevant because it
> worked fine with .38.

Does this help?

diff --git a/drivers/ide/ide-io.c b/drivers/ide/ide-io.c
index 0e406d73..ca27d30 100644
--- a/drivers/ide/ide-io.c
+++ b/drivers/ide/ide-io.c
@@ -570,8 +570,7 @@ void ide_requeue_and_plug(ide_drive_t *drive, struct request *rq)
 	spin_unlock_irqrestore(q->queue_lock, flags);
 
 	/* Use 3ms as that was the old plug delay */
-	if (rq)
-		blk_delay_queue(q, 3);
+	blk_delay_queue(q, 3);
 }
 
 static int drive_is_ready(ide_drive_t *drive)

-- 
Jens Axboe