[alsa-devel] snd_array_free - snd: bad kfree when HDA reconfig

[Takashi Iwai]

At Thu, 15 Sep 2011 22:23:12 +0800,
Raymond Yau wrote:
> 
> 2011/9/14 Takashi Iwai <tiwai at suse.de>:
> >
> > Try the patch below in addition to the previous one.
> >
> >
> > thanks,
> >
> > Takashi
> >
> > ---
> > From: Takashi Iwai <tiwai at suse.de>
> > Subject: [PATCH] Add missing export of snd_hidden_ksize()
> >
> > Also fix the ifdef of krealloc() wrapper for some corner cases.
> >
> > Signed-off-by: Takashi Iwai <tiwai at suse.de>
> > ---
> >  acore/sound.inc   |    1 +
> >  include/adriver.h |    3 ++-
> >  2 files changed, 3 insertions(+), 1 deletions(-)
> >
> > diff --git a/acore/sound.inc b/acore/sound.inc
> > index a79755e..2e63c98 100644
> > --- a/acore/sound.inc
> > +++ b/acore/sound.inc
> > @@ -3,6 +3,7 @@
> >  EXPORT_SYMBOL(snd_hidden_kmalloc);
> >  EXPORT_SYMBOL(snd_hidden_kzalloc);
> >  EXPORT_SYMBOL(snd_hidden_kcalloc);
> > +EXPORT_SYMBOL(snd_hidden_ksize);
> >  EXPORT_SYMBOL(snd_hidden_kfree);
> >  EXPORT_SYMBOL(snd_hidden_kstrdup);
> >  EXPORT_SYMBOL(snd_hidden_kstrndup);
> > diff --git a/include/adriver.h b/include/adriver.h
> > index fe70442..848bc17 100644
> > --- a/include/adriver.h
> > +++ b/include/adriver.h
> > @@ -2002,7 +2002,8 @@ static inline bool flush_delayed_work_sync(struct delayed_work *dwork)
> >  #endif
> >
> >  /* krealloc() wrapper */
> > -#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 22)) || defined(CONFIG_SND_DEBUG_MEMORY)
> > +#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 0) && LINUX_VERSION_CODE < KERNEL_VERSION(2, 6, 22)) || \
> > +       (defined(CONFIG_SND_DEBUG_MEMORY) && !defined(SKIP_HIDDEN_MALLOCS))
> >  #include <linux/slab.h>
> >  static inline void *snd_compat_krealloc(const void *p, size_t new_size, gfp_t flags)
> >  {
> > --
> > 1.7.6.1
> >
> >
> 
> your patch fix the problem
> 
> BTW, there is some strange behaviour of the non-sticky pcm
> implementation on ad1988
> 
> Refer to commit 0e7adbe263f89ea2ef15b5af5e80a812b2a85025
> 
> Disable sticky PCM stream assignment for AD codecs
> 
> why hda_codec_cleanup_stream is called twice  and this also occur with
> capture and playback

Isn't it just the power-save?  The suspend calls
hda_cleanup_all_streams().

That is, the first call is the normal one, and it doesn't clean up the
stream actually because of stickiness.  The second call is the real
cleanup at power-saving.


Takashi