Re: Commit 7eaceaccab5f40 causing boot hang.

[Jens Axboe <jaxboe@fusionio.com>]

On 2011-03-31 12:47, Rob Landley wrote:
> On 03/31/2011 05:37 AM, Jens Axboe wrote:
>> On 2011-03-31 12:07, Jens Axboe wrote:
>>> On 2011-03-31 11:11, Rob Landley wrote:
>>>> On 03/31/2011 04:02 AM, Jens Axboe wrote:
>>>>> On 2011-03-30 15:52, Rob Landley wrote:
>>>>>> On 03/30/2011 06:38 AM, Jens Axboe wrote:
>>>>>>> On 2011-03-30 08:06, Rob Landley wrote:
>>>>>>>> On 03/29/2011 10:51 AM, Pete Clements wrote:
>>>>>>>>> Quoting Jens Axboe
>>>>>>>>>   > 
>>>>>>>>>   > On 2011-03-29 16:13, Rob Landley wrote:
>>>>>>>>>   > > On 03/29/2011 08:59 AM, Jens Axboe wrote:
>>>>>>>>>   > >> On 2011-03-29 10:52, Rob Landley wrote:
>>>>>>>>>   > >>> 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)
>>>>>>>>>   > >>
>>>>>>>>>   > > 
>>>>>>>>>   > > Nope, still hung the same way.
>>>>>>>>>   > 
>>>>>>>>>   > Funky. I'll try and reproduce this tonight.
>>>>>>>>>   > 
>>>>>>>>>   > -- 
>>>>>>>>>   > Jens Axboe
>>>>>>>>>   > 
>>>>>>>>>
>>>>>>>>> I have had a similiar problem (reported earlier) unable to boot.
>>>>>>>>> With git15-18 hung with IDE drives (hda), git19-21 moved the hang down to
>>>>>>>>> the IDE CDROM (hdc). Applied the above patch and now booted into git21 without
>>>>>>>>> any hang and all appears ok.
>>>>>>>>
>>>>>>>> It may have made it better for me, it's hard to tell.
>>>>>>>>
>>>>>>>> I did a fresh pull, re-applied the patch, and tried again with
>>>>>>>> init=/bin/sh and it booted to the shell prompt... which then hung when I
>>>>>>>> did "ls -l /".
>>>>>>>>
>>>>>>>> If I let it boot normally, init announces itself, gives a spurious
>>>>>>>> warning about a fstab field (which it's been doing for a while, my fault
>>>>>>>> but harmless), then hangs.
>>>>>>>>
>>>>>>>>> This is i386, UP.
>>>>>>>>
>>>>>>>> I'm doing x86-64 SMP.
>>>>>>>
>>>>>>> I think we have the same issue the other location. How about this, then:
>>>>>>>
>>>>>>> diff --git a/drivers/ide/ide-io.c b/drivers/ide/ide-io.c
>>>>>>> index 0e406d73..4978ec3 100644
>>>>>>> --- a/drivers/ide/ide-io.c
>>>>>>> +++ b/drivers/ide/ide-io.c
>>>>>>> @@ -549,12 +549,11 @@ plug_device:
>>>>>>>  	spin_unlock_irq(&hwif->lock);
>>>>>>>  	ide_unlock_host(host);
>>>>>>>  plug_device_2:
>>>>>>> +	blk_delay_queue(q, queue_run_ms);
>>>>>>>  	spin_lock_irq(q->queue_lock);
>>>>>>>  
>>>>>>> -	if (rq) {
>>>>>>> +	if (rq)
>>>>>>>  		blk_requeue_request(q, rq);
>>>>>>> -		blk_delay_queue(q, queue_run_ms);
>>>>>>> -	}
>>>>>>>  }
>>>>>>>  
>>>>>>>  void ide_requeue_and_plug(ide_drive_t *drive, struct request *rq)
>>>>>>> @@ -570,8 +569,7 @@ 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)
>>>>>>>
>>>>>>
>>>>>> Did a fresh pull and applied that patch.  (It conflicts with your
>>>>>> previous one, but looks like it includes it.)
>>>>>>
>>>>>> Now it hangs after the "EXT3-fs: barriers not enabled" line, doesn't
>>>>>> make it to init.
>>>>>
>>>>> I have tried hard to reproduce this, but even stock 2.6.39-rc1 works
>>>>> fine for me here. Setup a KVM image with a debian 6 install, then
>>>>> converted it to IDE and booting it with a custom kernel like you are.
>>>>> Works fine, boots and I can do disk activity tests and it all works.
>>>>>
>>>>> Can you send me your .config?
>>>>
>>>> It was attached to the first message in this series, here it is again.
>>>>
>>>> I update it via "make oldconfig" and hold down return.
>>>>
>>>> I boot it via:
>>>>
>>>> qemu-system-x86_64 -m 1024 -kernel arch/x86/boot/bzImage \
>>>>   -hda ~/sid.ext3 -append "root=/dev/hda rw"
>>>
>>> Much better, I see the hang now! Now to try and diagnose...
>>
>> It seems to hard hang, looks very odd:
> 
> I should have mentioned that: when it hangs, you can't page back up in
> the console display anymore.

Data point - only seems to happen on UP. If you add -smp X to kvm, then
it boots and works just fine. It smells oddly like some sort of irq
event.

-- 
Jens Axboe