lockdep and oops_in_progress

lockdep and oops_in_progress

[Dave Airlie]

Hi guys,

I was debugging a wierd issue with suspend/resume and fbcon/X with
some recent work Jesse and myself did to try and make sure a pinned
fbcon always gets the text on it with a panic or oops occurs.

It does this by testing the oops_in_progress flag, however once any
lockdep issue occurs it looks to me that we leave this flag set
forever, in most places I can see in oops code etc they call
bust_spinlocks(1) then bust_spinlocks(0) to balance the
oops_in_progress value, but lockdep never seems to reset it.

I'm wondering if there is an inherent reason for this or if I whip up
a patch to reset once the lockdep is printed if this would cause any
issues?

Dave.

Re: lockdep and oops_in_progress

[Peter Zijlstra]

On Thu, 2010-08-05 at 16:21 +1000, Dave Airlie wrote:
> Hi guys,
> 
> I was debugging a wierd issue with suspend/resume and fbcon/X with
> some recent work Jesse and myself did to try and make sure a pinned
> fbcon always gets the text on it with a panic or oops occurs.
> 
> It does this by testing the oops_in_progress flag, however once any
> lockdep issue occurs it looks to me that we leave this flag set
> forever, in most places I can see in oops code etc they call
> bust_spinlocks(1) then bust_spinlocks(0) to balance the
> oops_in_progress value, but lockdep never seems to reset it.
> 
> I'm wondering if there is an inherent reason for this or if I whip up
> a patch to reset once the lockdep is printed if this would cause any
> issues?

That's debug_locks_off(), right? I don't think there's a particular
reason we keep it set, cleaning that up might take a bit of work but
shouldn't be too hard.

Ingo, do you remember anything about that?, I think that bit comes from
before my time.

Re: lockdep and oops_in_progress

[Yong Zhang]

On Thu, Aug 05, 2010 at 10:10:49AM +0200, Peter Zijlstra wrote:
> On Thu, 2010-08-05 at 16:21 +1000, Dave Airlie wrote:
> > Hi guys,
> > 
> > I was debugging a wierd issue with suspend/resume and fbcon/X with
> > some recent work Jesse and myself did to try and make sure a pinned
> > fbcon always gets the text on it with a panic or oops occurs.
> > 
> > It does this by testing the oops_in_progress flag, however once any
> > lockdep issue occurs it looks to me that we leave this flag set
> > forever, in most places I can see in oops code etc they call
> > bust_spinlocks(1) then bust_spinlocks(0) to balance the
> > oops_in_progress value, but lockdep never seems to reset it.
> > 
> > I'm wondering if there is an inherent reason for this or if I whip up
> > a patch to reset once the lockdep is printed if this would cause any
> > issues?
> 
> That's debug_locks_off(), right? I don't think there's a particular
> reason we keep it set, cleaning that up might take a bit of work but
> shouldn't be too hard.

commit e0fdace10e75dac67d906213b780ff1b1a4cc360
Author: David Miller <davem@davemloft.net>
Date:   Fri Aug 1 01:11:22 2008 -0700

    debug_locks: set oops_in_progress if we will log messages.
    
    Otherwise lock debugging messages on runqueue locks can deadlock the
    system due to the wakeups performed by printk().
    
    Signed-off-by: David S. Miller <davem@davemloft.net>
    Signed-off-by: Ingo Molnar <mingo@elte.hu>

Seems we can revert that commit now because of 'robustify printk'.

Dave, what do you think about it?

Thanks,
Yong

> 
> Ingo, do you remember anything about that?, I think that bit comes from
> before my time.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: lockdep and oops_in_progress

[Dave Airlie]

On Fri, Aug 6, 2010 at 12:59 AM, Yong Zhang <yong.zhang0@gmail.com> wrote:
> On Thu, Aug 05, 2010 at 10:10:49AM +0200, Peter Zijlstra wrote:
>> On Thu, 2010-08-05 at 16:21 +1000, Dave Airlie wrote:
>> > Hi guys,
>> >
>> > I was debugging a wierd issue with suspend/resume and fbcon/X with
>> > some recent work Jesse and myself did to try and make sure a pinned
>> > fbcon always gets the text on it with a panic or oops occurs.
>> >
>> > It does this by testing the oops_in_progress flag, however once any
>> > lockdep issue occurs it looks to me that we leave this flag set
>> > forever, in most places I can see in oops code etc they call
>> > bust_spinlocks(1) then bust_spinlocks(0) to balance the
>> > oops_in_progress value, but lockdep never seems to reset it.
>> >
>> > I'm wondering if there is an inherent reason for this or if I whip up
>> > a patch to reset once the lockdep is printed if this would cause any
>> > issues?
>>
>> That's debug_locks_off(), right? I don't think there's a particular
>> reason we keep it set, cleaning that up might take a bit of work but
>> shouldn't be too hard.
>
> commit e0fdace10e75dac67d906213b780ff1b1a4cc360
> Author: David Miller <davem@davemloft.net>
> Date:   Fri Aug 1 01:11:22 2008 -0700
>
>    debug_locks: set oops_in_progress if we will log messages.
>
>    Otherwise lock debugging messages on runqueue locks can deadlock the
>    system due to the wakeups performed by printk().
>
>    Signed-off-by: David S. Miller <davem@davemloft.net>
>    Signed-off-by: Ingo Molnar <mingo@elte.hu>
>
> Seems we can revert that commit now because of 'robustify printk'.
>
> Dave, what do you think about it?

DaveM ping?

I should probably just push for the revert if printk is robust enough now.

Dave.

Re: lockdep and oops_in_progress

[Dave Airlie]

On Fri, Aug 6, 2010 at 12:59 AM, Yong Zhang <yong.zhang0@gmail.com> wrote:
> On Thu, Aug 05, 2010 at 10:10:49AM +0200, Peter Zijlstra wrote:
>> On Thu, 2010-08-05 at 16:21 +1000, Dave Airlie wrote:
>> > Hi guys,
>> >
>> > I was debugging a wierd issue with suspend/resume and fbcon/X with
>> > some recent work Jesse and myself did to try and make sure a pinned
>> > fbcon always gets the text on it with a panic or oops occurs.
>> >
>> > It does this by testing the oops_in_progress flag, however once any
>> > lockdep issue occurs it looks to me that we leave this flag set
>> > forever, in most places I can see in oops code etc they call
>> > bust_spinlocks(1) then bust_spinlocks(0) to balance the
>> > oops_in_progress value, but lockdep never seems to reset it.
>> >
>> > I'm wondering if there is an inherent reason for this or if I whip up
>> > a patch to reset once the lockdep is printed if this would cause any
>> > issues?
>>
>> That's debug_locks_off(), right? I don't think there's a particular
>> reason we keep it set, cleaning that up might take a bit of work but
>> shouldn't be too hard.
>
> commit e0fdace10e75dac67d906213b780ff1b1a4cc360
> Author: David Miller <davem@davemloft.net>
> Date:   Fri Aug 1 01:11:22 2008 -0700
>
>    debug_locks: set oops_in_progress if we will log messages.
>
>    Otherwise lock debugging messages on runqueue locks can deadlock the
>    system due to the wakeups performed by printk().
>
>    Signed-off-by: David S. Miller <davem@davemloft.net>
>    Signed-off-by: Ingo Molnar <mingo@elte.hu>
>
> Seems we can revert that commit now because of 'robustify printk'.
>
> Dave, what do you think about it?

Hey DaveM, as mentioned at KS,

Dave.

Re: lockdep and oops_in_progress

[David Miller]

From: Dave Airlie <airlied@gmail.com>
Date: Wed, 3 Nov 2010 06:51:07 +1000

> Hey DaveM, as mentioned at KS,

Thanks, will take a look.

Re: lockdep and oops_in_progress

[David Miller]

From: Dave Airlie <airlied@gmail.com>
Date: Wed, 3 Nov 2010 06:51:07 +1000

>> commit e0fdace10e75dac67d906213b780ff1b1a4cc360
>> Author: David Miller <davem@davemloft.net>
>> Date:   Fri Aug 1 01:11:22 2008 -0700
>>
>>    debug_locks: set oops_in_progress if we will log messages.
>>
>>    Otherwise lock debugging messages on runqueue locks can deadlock the
>>    system due to the wakeups performed by printk().
>>
>>    Signed-off-by: David S. Miller <davem@davemloft.net>
>>    Signed-off-by: Ingo Molnar <mingo@elte.hu>
>>
>> Seems we can revert that commit now because of 'robustify printk'.

I completely agree, this can definitely be reverted.

Thanks!