[pulseaudio-discuss] [PATCH v9 0/8] Bluetooth A2DP codecs

[Pali Rohár]

On Friday 26 April 2019 13:09:00 Luiz Augusto von Dentz wrote:
> Hi Pali,
> 
> On Fri, Apr 26, 2019 at 11:34 AM Pali Rohár <pali.rohar at gmail.com> wrote:
> >
> > On Friday 26 April 2019 11:20:26 Luiz Augusto von Dentz wrote:
> > > I have an assert when I use a different headset which does support apt-X HD:
> > >
> > > https://gist.github.com/Vudentz/0d6b6f2ad08524db69a3e223e26bc80d
> >
> > I have not tested aptX HD, nor aptX Low Latency. I tested only (classic)
> > aptX. I looked at code and there is incorrect calculation of buffer
> > block size for aptX HD. Try following:
> >
> > static void get_buffer_size_hd(void *codec_info, size_t link_mtu, size_t *decoded_buffer_size, size_t *encoded_buffer_size) {
> >     /* aptX HD compression ratio is 4:1 and we need to process one aptX HD sample (6 bytes) at once */
> >     *encoded_buffer_size = (link_mtu/6) * 6;
> >     *decoded_buffer_size = *encoded_buffer_size * 4;
> > }
> 
> Will try, btw if I disable aptX HD it attempts to pick up UHQ2 but it
> asserts as well:
> 
> https://gist.github.com/Vudentz/0e91f3ba260211cab38822bcf04edd1f
> 
> It seems that fill_preferred_configuration and can_accept_capabilities
> are not agreeing with one another

When fill_preferred_configuration or can_accept_capabilities fails it
prints debug message about it. But I do not see any of them in your
output. So I think problem can be somewhere else.

I have not got this assert for uhq2, so this looks like some race
condition.

> so we got a transport but no profile.

In your log is:
bluez activated sbc_uhq2, then module-bluez5-device started to be
loading, card.c chose a2dp_sink_aptx as initial profile and due to bug
in module-bluetooth-policy.c aptx was not changed to sbc_uhq2. Next
sbc_uhq2 was released and prepared for switching to initial profile
aptx. And then assertion failed.

My guess is: because card.c was not fully initialized yet (it called
hook for choosing initial profile), which called code for switching
profile, then u->card->profiles was not fully initialized and crashed.
Profile switching should be probably allowed only after full card
initialization to prevent such race conditions...

-- 
Pali Rohár
pali.rohar at gmail.com