Messages in this thread |  | | | Date | Sun, 18 May 2008 09:55:21 +0200 | | From | Takashi Iwai <> | | Subject | Re: [Bugme-new] [Bug 10701] New: snd_pcsp lockdep warning | |
At Sat, 17 May 2008 17:24:22 +0200 (CEST),
Thomas Gleixner wrote:
> > On Sat, 17 May 2008, Takashi Iwai wrote:> > At Fri, 16 May 2008 21:32:49 +0200 (CEST),> > Thomas Gleixner wrote:> > > > > > On Fri, 16 May 2008, Andrew Morton wrote:> > > > > So the author was well aware of locking problem and the whole code is
> > > > > just a stupid hack around the problem without solving it. This code is
> > > > > not at all suited for HRTIMER_CB_IRQSAFE.> > > > > > > > You sound impressed! So what's the fix? HRTIMER_CB_SOFTIRQ?> > > > > > Yeah, impressed by creativity. HRTIMER_CB_SOFTIRQ should be the right
> > > thing.> > > > Well, it'd be basically a similar way like snd-pcsp currently does ("a
> > stupid hack" :) But, it's good to have a fix, anyway, since this
> > sounds like a generic problem with a callback in a spinlock. If the
> > callback requires another own lock, this can easily lead to a AB/BA
> > deadlock. Actually, ALSA PCM core had sometimes similar problems,
> > too.> > HRTIMER_CB_SOFTIRQ is not a stupid hack :) It's the default for
> hrtimers and it does not hold any locks when calling the callback. We
> really want to avoid tons of callbacks in the timer interrupt itself.
Yes, this would be an easy workaround for snd-pcsp problem, I guess.
A better fix would be to make snd-pcsp hrtimer calback just change the
pc-speaker port (0x61 and 0x42) and update the position. These should
be really fast path. Then, occasionally invoke a tasklet that will
call snd_pcm_period_elapsed(), which could be a (relatively) slow
path and lead to a spinlock mess.
thanks,
Takashi
|  |