[pulseaudio-discuss] Bluetooth regression: a2dp not selectable

[David Henningsson]

On 11/27/2012 06:39 PM, Mikel Astiz wrote:
> Hi David,
>
> On Mon, Nov 26, 2012 at 6:27 PM, Mikel Astiz <mikel.astiz.oss at gmail.com> wrote:
>> Hi David,
>>
>> On Mon, Nov 26, 2012 at 10:15 AM, David Henningsson
>> <david.henningsson at canonical.com> wrote:
>>> On 11/23/2012 05:27 PM, Tanu Kaskinen wrote:
>>>>
>>>> On Fri, 2012-11-23 at 16:40 +0100, David Henningsson wrote:
>>>>>
>>>>> Under PulseAudio 2.1 I can select a2dp without problem.
>>>>>
>>>>> Under PulseAudio 2.99.2, when I execute "pactl set-card-profile 1 a2dp"
>>>>> I get "Failure: Input/Output Error" back (as the output from pactl), and
>>>>> I see the following in PulseAudio's log:
>>>>>
>>>>> module-bluetooth-device.c: Profile has no transport
>>>>>
>>>>> ...and the profile is not changed to a2dp. Anyone want to debug with me?
>>>>
>>>>
>>>> I can try to help, if it's not too late? You seem to have left IRC.
>>>
>>>
>>> Thanks, I'll be around today.
>>>
>>>
>>>> I'm not able reproduce this with my headset.
>>>
>>>
>>> Ok. I've done some minor debugging myself, but haven't got much. The
>>> pulseaudio debug log is here: http://pastebin.se/s3CIWSl0
>>> At the same time I get these messages in syslog: http://pastebin.se/bilI1XGX
>>>
>>> It looks like bluetoothd does find the A2DP Sink (and source?!), but then
>>> there is no callback to PulseAudio to register the transport, because this
>>> part is only shown for HSP, not for A2DP:
>>>
>>> D: [lt-pulseaudio] bluetooth-util.c: dbus:
>>> interface=org.bluez.MediaEndpoint, path=/MediaEndpoint/HFPAG,
>>> member=SetConfiguration
>>> D: [lt-pulseaudio] bluetooth-util.c: dbus:
>>> interface=org.bluez.MediaEndpoint, path=/MediaEndpoint/HFPAG,
>>> member=SetConfiguration
>>> D: [lt-pulseaudio] bluetooth-util.c: Transport
>>> /org/bluez/634/hci0/dev_00_18_91_3A_B6_EC/fd3 profile 2 available
>>
>> Just to make sure... can you confirm that the device was paired before
>> PA was started? There is a potential issue immediately after pairing,
>> but this would be no regression and nobody has complaint so far.
>>
>> Assuming the device was already paired, can you check if A2DP profile
>> is connected before and after this problem? You can see this in
>> BlueZ's AudioSink.GetProperties(), property "State".
>>
>> I can't see anything strange in the log, so my best guess is that A2DP
>> is actually disconnected? In this case the behavior would be the
>> expected one. This could typically happen if you shut PA down and then
>> restart it: HSP would still be connected but not A2DP. In this case
>> you can't use Audio.Connect() to connect the device, since it'll
>> complain with AlreadyConnected (since Audio.State == "connected"
>> represents that at least one of the profiles is connected). Therefore
>> AudioSink.Connect() would solve the issue.
>>
>> It's a long shot but anyway...
>
> After checking your traces today, the hypothesis above is now discarded.
>
> The issue seems to be related to the socket-IPC vs media-API. You're
> using BlueZ 4.101 but probably you have a modified version of
> audio.conf which enables the socket interface (a line like
> Enable=Socket), since it's otherwise disabled by default.
>
> This would explain what's happening assuming that BlueZ 4.101 has a
> bug that is causing this. You could confirm this by testing with PA
> 2.1 + BlueZ 4.101 without the socket interface (just remove "Socket"
> from the Enable line), which I would expect to fail.
>
> In a similar way, please check if the issue can be reproduced with PA
> 4.99 + BlueZ 4.101 without socket IPC.

After much and long research by Mikel and myself:

This was traced down to the combination of Ubuntu enabling the socket 
interface of bluez by default, and that bluez cannot handle the d-bus 
API correctly when the socket interface is enabled.

So in short; if you want bluetooth audio to work under 2.99, don't 
enable bluez's socket API.


-- 
David Henningsson, Canonical Ltd.
https://launchpad.net/~diwic