[pulseaudio-discuss] [PATCH v4 0/3] use bluetooth HFP in pulseaudio when available

[James Bottomley]

On Thu, 2017-09-21 at 06:27 +0200, Georg Chini wrote:
> On 20.09.2017 23:12, James Bottomley wrote:
> > 
> > On Wed, 2017-09-20 at 20:41 +0200, Georg Chini wrote:
> > > 
> > > On 20.09.2017 20:10, James Bottomley wrote:
> > > > 
> > > > On Wed, 2017-09-20 at 18:11 +0200, Georg Chini wrote:
> > > > > 
> > > > > On 20.09.2017 01:27, James Bottomley wrote:
> > > > > > 
> > > > > > This is round 4 of the initial bluetooth: separate HSP and
> > > > > > HFP patch.
> > > > > >     It includes the review feedback and a global on/off
> > > > > > switch just in case there's a problem headset with dual
> > > > > > HFP/HSP but non-working HFP.   This one now includes a
> > > > > > proper rfcomm negotiation (see patch 3).  I've finally
> > > > > > figured out a bug in the rfcomm negotiation that was
> > > > > > causing issues with my LG 900 headset, so I think it's now
> > > > > > working for everything (but testing would be welcome).
> > > > > > 
> > > > > > James Bottomley (3):
> > > > > >      bluetooth: use consistent profile names
> > > > > >      bluetooth: separate HSP and HFP
> > > > > >      bluetooth: add correct HFP rfcomm negotiation
> > > > > > 
> > > > > >     src/modules/bluetooth/backend-native.c          | 168
> > > > > > +++++++++++++++++++++---
> > > > > >     src/modules/bluetooth/backend-ofono.c           |   4
> > > > > > +-
> > > > > >     src/modules/bluetooth/bluez5-util.c             |  46
> > > > > > +++++--
> > > > > >     src/modules/bluetooth/bluez5-util.h             |  10
> > > > > > +-
> > > > > >     src/modules/bluetooth/module-bluetooth-policy.c |   3
> > > > > > +-
> > > > > >     src/modules/bluetooth/module-bluez5-device.c    | 102
> > > > > > ++++++++++----
> > > > > >     src/modules/bluetooth/module-bluez5-discover.c  |   6
> > > > > > +-
> > > > > >     7 files changed, 274 insertions(+), 65 deletions(-)
> > > > > > 
> > > > > Hello James,
> > > > > 
> > > > > thank you for continuing your work. Unfortunately your patch
> > > > > set collides with running ofono. Currently, with the default
> > > > > of "headest=auto", the native and the ofono backends are
> > > > > active in parallel. This is possible because ofono only
> > > > > supports HFP while PA only supports HSP.
> > > > > 
> > > > > If PA starts supporting HFP headsets, this breaks above
> > > > > assumption and ofono and PA both try to register the
> > > > > corresponding HFP UUID.
> > > > > 
> > > > > To work around the problem, I suggest to disable native HFP
> > > > > support if headset_backend == HEADSET_BACKEND_AUTO, unless
> > > > > configured otherwise on the command line.
> > > > Actually, Pulseaudio already includes an ofono is running
> > > > check, so the enable should be set to no for backend ofono or
> > > > backend auto and ofono running, which would enable it in the
> > > > widest possible set of scenarios.
> > > > 
> > > > I can cook up a patch for that, hang on.
> > > > 
> > > > James
> > > This does only work for the special case when PA acts in the
> > > HSP headset role. In this case, no duplicate UUIDs are registered
> > > and disabling the profile only needs to be done because otherwise
> > > PA and ofono would both be listening for incoming SCO
> > > connections.
> > It seems to me that pulse checks the dbus connection to ofono
> > before deciding on the backend (and therefore deciding what to
> > register), so as long as pulse finds ofono running, there won't be
> > any duplicate registrations.
> > 
> > > 
> > > The case here is different in that a duplicate UUID is
> > > registered. This means, by the time PA recognizes that ofono is
> > > running, ofono already tried to register the UUID and failed.
> > > That's why you have to disable HFP even if ofono is currently not
> > > running. 
> > I don't think that can be true if ofono is already running, can it?
> >   Either ofono registered the HFP UUIDs way earlier or pulse sees
> > ofono isn't running and registers them.
> > 
> > I don't think requiring the ofono daemon to be running before
> > pulseaudio is insurmountable.  On my openSUSE system, ofono is
> > started from systemd and pulse from user login, so this is
> > automatically satisfied.  I think where pulse is used in system
> > mode, you just have to tell systemd to start ofono first, which is
> > certainly doable.
> 
> I don't think it is a good idea to rely on the order ofono and pulse
> are started. Also. if pulse is running and has already registered
> the UUID, what will happen if ofono starts? ofono has no checks
> for pulse and will simply try to register the UUID. I fear that this
> happens before pulse recognizes that ofono is running.

This is an initialisation issue for how you want everything to work.
 The current distro setup is ofono from systemd and pulse from login,
so I can't actually see what the problem is ... unless there's some
reason why what the distros are doing is wrong?

If pulse is already running when you start ofono, that's caveat emptor
... plus, as far as I can tell, it still all works because ofono
doesn't seem to get the uuid.

The point is to get a working default configuration the distros can
use.  That means that we need a good way of distinguishing the non-
ofono case for backend = auto.  Ofono not running seems to be the
definitive test unless there's another you can propose?

> > The current fly in the ointment is that for a dual HFP/HSP device,
> > we need to fall back to the HSP profile not simply disable the HFP
> > one, which requires extra jiggerypokery.
> 
> No, you don't need to fall back to HSP. The HFP connection will
> be handled by ofono in that case.

Well, yes you do.  There's no current easy way to use HFP_HF with
ofono, so you want the HSP profile to be exposed if it exists, so
there's an easy connection to the headset.

James