Re: WARNING: at kernel/lockdep.c:2323 trace_hardirqs_on_caller+0xb9/0x16c()

[Tejun Heo <tj@kernel.org>]

Hello,

On Thu, Feb 17, 2011 at 10:43:57AM +0800, Yong Zhang wrote:
> On Thu, Feb 17, 2011 at 10:03 AM, David Miller <davem@davemloft.net> wrote:
> > From: Yong Zhang <yong.zhang0@gmail.com>
> > Date: Thu, 17 Feb 2011 09:37:30 +0800
> >
> >> On Tue, Feb 15, 2011 at 10:42 PM, Daniel Lezcano <daniel.lezcano@free.fr> wrote:
> >>> Hi All,
> >>>
> >>> I am running a 2.6.38-rc4-next-20110215+ kernel on qemu x86_64 and the
> >>> following traces appear in the console:
> >>>
> >>> Feb 15 15:00:24 lucid kernel: ------------[ cut here ]------------
> >>> Feb 15 15:00:24 lucid kernel: WARNING: at kernel/lockdep.c:2323
> >>> trace_hardirqs_on_caller+0xb9/0x16c()
> >>> Feb 15 15:00:24 lucid kernel: Hardware name: Bochs
> >>> Feb 15 15:00:24 lucid kernel: Pid: 1477, comm: mountall Not tainted
> >>> 2.6.38-rc4-next-20110215+ #74
> >>> Feb 15 15:00:24 lucid kernel: Call Trace:
> >>> Feb 15 15:00:24 lucid kernel: <IRQ>  [<ffffffff8102b8a5>] ?
> >>> warn_slowpath_common+0x7b/0x93
> >>> Feb 15 15:00:24 lucid kernel: [<ffffffff8146c097>] ?
> >>> _raw_spin_unlock_irq+0x2b/0x30
> >>> Feb 15 15:00:24 lucid kernel: [<ffffffff8102b8d2>] ?
> >>> warn_slowpath_null+0x15/0x17
> >>> Feb 15 15:00:24 lucid kernel: [<ffffffff8104f796>] ?
> >>> trace_hardirqs_on_caller+0xb9/0x16c
> >>> Feb 15 15:00:24 lucid kernel: [<ffffffff8104f856>] ?
> >>> trace_hardirqs_on+0xd/0xf
> >>> Feb 15 15:00:24 lucid kernel: [<ffffffff8146c097>] ?
> >>> _raw_spin_unlock_irq+0x2b/0x30
> >>> Feb 15 15:00:24 lucid kernel: [<ffffffff812e167e>] ?
> >>> do_ide_request+0x32/0x590
> >>
> >> Seems related to IDE SUBSYSTEM
> >
> > Which hasn't had any changes in the past release.
> 
> OK.
> 
> Cc'ing Tejun Heo
> 
> For the back trace, I think __blk_run_queue() is the ligament.
> As from the comment of __blk_run_queue(), it must be called
> with the queue lock and interrupts disabled. And the lock
> is hold through spin_lock_irqsave(q->queue_lock, flags); at
> blk_end_bidi_request().
> 
> But in do_ide_request(), it realse the lock through
> spin_unlock_irq(q->queue_lock); which make the state
> inconsistent.
> 
> BTW, do_ide_request() also say it might_sleep(), this warning
> also trigger in Daniel's log.

This seems to be the same problem Jan reported and fixed by the
following patches.

  http://article.gmane.org/gmane.linux.kernel/1101766/raw
  http://article.gmane.org/gmane.linux.ide/48819/raw

Can you please test whether these two patches fix the problem?

Thanks.

-- 
tejun