[alsa-devel] [RFC][PATCH] ASoC: soc-core: verify Sound Card normality

[Takashi Iwai]

On Mon, 03 Apr 2017 08:29:34 +0200,
Kuninori Morimoto wrote:
> 
> 
> Hi Takashi-san
> 
> > > I think this is a good direction to at least start to mitigate these
> > > problems (which we really should be doing) and hopefully make it easier
> > > to do further improvements in future.  There's obviously more places
> > > where we should be checking the flag (controls for example) but they can
> > > be added later.  One thing I would like to see is instead of setting the
> > > flag directly when we see a problem call a function to do it.  That way
> > > if we want to improve things in the future we can do that without having
> > > to update the callers again.
> > 
> > BTW, ALSA core has snd_card_disconnect() that does this kind of
> > shut-up from user-space.  It was introduced for hot-unplug, but
> > basically unbinding is the software hot-unplug.  So, if ASoC won't
> > rebind a once-unbound component, you can simply call
> > snd_card_disconnect() at the component unbinding time to assure that
> > no further user actions can be done.
> 
> Thanks. I checked about snd_card_disconnect(), and it will be called
> from snd_card_free(). And it will be called from snd_soc_unregister_card()

Yes, snd_card_free() assures the disconnection at first, syncs the all
settled down, then releases the resources.

> So, we can call snd_soc_unregister_card() whenever CPU/Codec/Platform
> were unregsiterd.

In theory yes, but you should be careful to do so, e.g. make sure that
it won't be called again by the removal/unbind of other components /
drivers.

I suggested snd_card_disconnect() because it doesn't release resources
by itself, but it just disconnects from the further accesses.  So,
double-free won't happen in this case.  It makes the hotunplug safer
as long as the drivers manage the resource releases properly.


Takashi

> This method also solve random unbind/bind Oops too.
> Here, random unbind/bind example is that
> expected correct operation is unbind all CPU/Codec/Platfrom/Card,
> and then, bind all CPU/Codec/Platfrom/Card again.
> (here unbind order can be random)
> But this case, we will get Oops if unbind Codec -> bind Codec -> unbind Card.
> Using snd_soc_unregister_card() can solve this issue too.
> 
> Best regards
> ---
> Kuninori Morimoto
>