Re: 2.6.32 regression, hard lock

[Takashi Iwai <tiwai@suse.de>]

At Mon, 7 Dec 2009 16:48:26 -0800,
Andrew Morton wrote:
> 
> On Tue, 8 Dec 2009 01:23:53 +0100
> Richard Zidlicky <rz@linux-m68k.org> wrote:
> 
> > 
> > > > first visible symptom is that ppp over UMTS connection "stops working", connection
> > > > did not die. Trying to restart the connection fails and the related processes hang.
> > > > 
> > > > After that one after another everything stops working, "telinit 6" does not do its
> > > > job and SysRQ sync-umount-reboot is logged in messages but has no visible effect.
> > > > 
> > > > Attaching messages and .config.
> > > > 
> > > 
> > > I uploaded your dmesg output to
> > > http://userweb.kernel.org/~akpm/stuff/messages.txt
> > 
> > > 
> > > I'm looking at those traces and am not able to develop a theory to
> > > explain it :(
> > 
> > thanks for looking at it. 
> > 
> > I did revert to 2.31.5 and to my big surprise found a similar thing happened
> > so it is not a regression, at least not from 2.6.31.
> 
> This trace is different.  It's a lock ranking warning and is surely
> unrelated to the IRQs-off lockup which you earlier experienced.
> 
(snip)
> And fortunately I know who to cc on this one ;)

Oh, it's a pity to me that you know that ;)

It's a known issue for several kernel releases.  (There are a few
subsystems that have the similar lockdep pattern involving with
mm->mmap_sem, too.)

But, as Andrew pointed, this should be irrelevant with the IRQ
lockup.  At most it could be a dead-locking of PA and other
sound-related processes, but I've never heard the real issue.


thanks,

Takashi