Re: 2.6.14-mm1

[Andrew Morton <akpm@osdl.org>]

Michal Piotrowski <michal.k.k.piotrowski@gmail.com> wrote:
>
> On 12/11/05, Andrew Morton <akpm@osdl.org> wrote:
> > Crap.  This is one of those crashes where the sound people, the PCI people,
> > the ACPI people and the PM people all earnestly hope that it's the other
> > guy's bug and you and I are left with a mess on our hands.
> >
> > Possibly the card didn't get powered up.  I know there's a way to get all
> > those snd_printk()'s to print something, but I never have much success
> > finding the right value for the right /proc file to make it happen.
> >
> > So can you add this please?
> >
> > --- devel/sound/pci/intel8x0.c~a        2005-11-11 16:47:00.000000000 -0800
> > +++ devel-akpm/sound/pci/intel8x0.c     2005-11-11 16:48:13.000000000 -0800
> > @@ -779,6 +779,7 @@ static irqreturn_t snd_intel8x0_interrup
> >         unsigned int i;
> >
> >         status = igetdword(chip, chip->int_sta_reg);
> > +       printk("status: 0x%8x\n", status);
> >         if (status == 0xffffffff)       /* we are not yet resumed */
> >                 return IRQ_NONE;
> >
> > _
> >
> > and let us know what it says?
> >
> > Also, it would be useful if you could disable the sound driver in config
> > and see if you can get it booted.  If so, then generate the `dmesg -s
> > 1000000' output for good and bad kernels and let's see what they look like.
> >
> > Thanks.
> >
> 
> I will try to reproduce it, but according to
> http://klive.cpushare.com/2.6.14-mm1/?order_by=kernel_group&where_machine=all&branch=mm&scheduler=all&smp=all&live=all&ip=all
> I have been using 2.6.14-mm1 about 48 hours with 12 reboots and this
> problem appeared only once.

ah-hah.  This sounds rather like Reuben Farrelly's e100 failure - something
seems to be making PCI initialisation go stupid if CONFIG_PREEMPT is
enabled.

It would be interesting if you could reboot sufficiently often to work out
whether disabling CONFIG_PREEMPT fixes things up.