[pulseaudio-discuss] Audio APP (sink-input) bind to the sink with only unplugged hdmi-audio ports on it

[Hui Wang]

On 2018年10月02日 18:50, Georg Chini wrote:
> On 02.10.2018 11:49, Tanu Kaskinen wrote:
>> On Mon, 2018-10-01 at 10:18 +0800, Hui Wang wrote:
>>> On 2018年09月30日 18:30, Tanu Kaskinen wrote:
>>>> On Sun, 2018-09-30 at 15:03 +0800, Hui Wang wrote:
>>>>> This issue is also reported to:
>>>>> https://gitlab.freedesktop.org/pulseaudio/pulseaudio/issues/579
>>>>>
>>>>> Recently we found a weird issue on many laptops with the ubuntu 
>>>>> 18.04,
>>>>> it uses the pulseaudio-11.1 (I guess the PA of the latest version 
>>>>> also
>>>>> has this problem). The issue is like this:
>>>>>
>>>>>    1. boot the system up without plugging a hdmi monitor
>>>>>    2. run an audio app to play sound (e.g. $speaker-test)
>>>>>    3. the sound outputs from analog-speaker
>>>>>    4. plug a monitor with audio capability (through DP or HDMI port)
>>>>>    5. the sound still outputs from analog-speaker
>>>>>    6. open sound-setting (gnome-control-center --> choose sound), 
>>>>> you will
>>>>>       see two output devies: speaker and HDMI audio
>>>>>    7. choose HDMI audio, the sound will switch to HDIM audio from 
>>>>> speaker
>>>>>       (pa will remember speaker-test prefer to use hdmi-audio sink)
>>>>>    8. unplug the monitor, the default-sink is switching to 
>>>>> analog-speaker,
>>>>>       but the sound of speaker-test still route to hdmi-audio sink
>>>>>    9. run other sound apps, they all route sound to default sink
>>>>>       (analog-speaker), but speaker-test always routes to 
>>>>> hdmi-audio sink,
>>>>>       as a result, speaker-test can't output sound anymore unless we
>>>>>       replug a monitor with audio capability then the speaker-test 
>>>>> output
>>>>>       from hdmi-audio again.
>>>>> 10. if we want the speaker-test to route to analog-speaker, two ways:
>>>>>       run pacmd move-sink-input or plug a monitor, after two audio 
>>>>> devices
>>>>>       (hdmi audio and speaker) show up in the sound-setting, select
>>>>>       analog-speaker manually.
>>>>>
>>>>> This issue only happens on the laptops with 2 audio cards, analog
>>>>> devices on one card, hdmi audio on the other card. This kind of 
>>>>> laptops
>>>>> are very common, like I+A (Intel graphic + Amd Graphic), I+N(Intel +
>>>>> Nvidia), and A AMD.
>>>>>
>>>>> This issue will not happen on the laptops with only Intel graphic 
>>>>> card,
>>>>> since both analog and hdmi audio belong to one sound card. When hdmi
>>>>> monitor is unplugged, the hdmi sink will be removed from PA, then all
>>>>> sink-inputs will route to the only left sink: analog-sink.
>>>>>
>>>>> This issue will not happen on BT or USB audio. Unlike hdmi audio, 
>>>>> BT and
>>>>> USB audio cards will be removed totally from PA when they are
>>>>> unpluged/unconnected, so they don't have this issue as well.
>>>>>
>>>>> The root cause of this issue is although the hdmi monitor is 
>>>>> unplugged,
>>>>> the hdmi-sink still exists, and sink-input is selected by user to 
>>>>> bind
>>>>> to this sink, so the pa doesn't care about if this sink has valid 
>>>>> port
>>>>> or not, it bind the sink-input to this sink unconditionally.
>>>>>
>>>>> Maybe we could improve it like this: if the user selected sink 
>>>>> only has
>>>>> available_no ports, the pa will switch all sink-inputs of this 
>>>>> sink to
>>>>> other sinks (like default_sink) temporarily, once the selected 
>>>>> sink has
>>>>> availble ports, all sink-inputs switch back to this sink.
>>>>>
>>>>> Any good ideas?
>>>> The plan has been to do the following (but I haven't found the time to
>>>> implement it):
>>>>
>>>> Streams currently have a "save_sink" flag that tells whether the
>>>> current sink should be remembered and restored by module-stream-
>>>> restore. That flag should be replaced with a "preferred_sink" string
>>>> that is set when the user moves the stream. module-stream-restore
>>>> should remember and restore that string.
>>>>
>>>> When the user moves a stream to the current default sink, the
>>>> "preferred_sink" string should be set to NULL and 
>>>> module-stream-restore
>>>> should clear the routing for that stream in the stream database. From
>>>> that point on the stream will be always routed to the default sink.
>>>>
>>>> When a stream is created, module-stream-restore should set the
>>>> "preferred_sink" string and if that sink exists, the core should set
>>>> that sink as the initial routing for the new stream.
>>>>
>>>> When the default sink changes, the streams from the old default sink
>>>> should be moved to the new default sink, unless the "preferred_sink"
>>>> string is set to the old default sink and the active port of the old
>>>> default sink is not unavailable. This should be done by the core.
>>>>
>>>> When the active port of a sink becomes unavailable, all streams from
>>>> that sink should be moved to the default sink. This should be done by
>>>> the core.
>>>>
>>>> When a new sink appears or the active port of an existing sink changes
>>>> state from unavailable, all streams that have their "preferred_sink"
>>>> set to the new sink should be moved to the new sink. This should be
>>>> done by the core.
>>>>
>>>> As a result, module-stream-restore doesn't need to move streams any
>>>> more, it only needs to manage the "preferred_sink" variables. Some
>>>> other modules can perhaps be simplified as well.
>>>>
>>>> The same logic should of course be implemented for capture streams as
>>>> well.
>>>>
>>>> Would you be willing to implement this?
>>>>
>>> Got it, I will take some time to understand your plan first, then 
>>> have a
>>> try to implement it.
>> Maybe this rephrasing will help with understanding, maybe not:
>>
>> The idea is to store the preferred routing in the core pa_sink_input
>> struct instead of hiding that information in module-stream-restore. The
>> core should then take care to ensure that streams are always routed to
>> their preferred sinks, or if the preferred sink is not set or not
>> available, then to the default sink.
>>
>> There are some exceptions:
>>
>> If a sink's active port is unavailable, no streams should be routed to
>> it, unless it's the default sink (and it shouldn't be the default sink
>> unless all sinks are similarly unavailable). [1]
>>
>> An application may request a particular sink when it creates a stream.
>> The request should be honored, but the preferred_sink variable should
>> not be affected by the application request. The requested initial
>> routing should be treated as a temporary override.
>>
>> Some policy modules may move streams temporarily to a sink that is
>> neither the default sink nor the stream's preferred sink. For example,
>> module-allow-passthrough creates a null sink for streams that are moved
>> away from the "main sink" when a passthrough stream needs to access the
>> main sink. Those policy modules need to continue working.
>>
>>
>> [1] I'm afraid this requires us to provide a way to disable jack
>> detection, because on some hardware the jack detection doesn't work
>> properly, and ports are marked as unavailable when they shouldn't be.
>> If a port is incorrectly marked as unavailable, the proposed logic
>> makes it impossible to play anything to it. So this is some extra
>> work... Disabling jack detection could be implemented as an argument to
>> module-udev-detect (and module-alsa-card) that would affect all ports
>> on all alsa cards, but it would be better to have a configuration file
>> (I'm thinking /etc/pulse/alsa.conf) where the jack detection could be
>> disabled on per-card, per-port and/or per-jack-element basis.
Yes, configurable enable/disable detection on per-card, per-port and/or 
per-jack-element basis is a better idea.
>>
> There are already some patches to enable/disable jack detection,
> but they are using the messaging interface which is not yet in the
> tree.
>
>