Implicit feedback on BOSS GT-1, the saga continues...

Thu Apr 8 22:14:20 CEST 2021

Em Qui, 8 de abr de 2021 16:42, Mike Oliphant <oliphant at nostatic.org>
escreveu:

> Hi Geraldo - yes. In fact, that is how I've been operating for the past
> year or so before the recent implicit feedback changes. I use the GT-1 most
> every day as a guitar interface and it works perfectly.
>

Thanks Mike, but I'm not sure I understand. You have been operating with
implicit feedback sync on the playback for the past year? I thought that
was a new idea?

The key point here is that with a stock kernel you have to comment out the
IMPLICIT_FB_SKIP_DEV() entry for the GT-1 for the Roland code inside
audioformat_implicit_fb_quirk properly kick in.

I was suggesting you try your workaround without adding a
IMPLICIT_FB_GENERIC_DEV() for your device.

> The GT-1 (and I suspect many of the other Roland/BOSS devices) doesn't
> seem to be quirky at all with regard to implicit feedback setup - it works
> exactly as the endpoint definitions say it should.
>
> Where the GT-1 *does* seem to be quirky is that is doesn't seem to want to
> send capture data until it receives some playback data - hence my
> endpoint.c patch.
>

If that is true then there's definitely a need to rewrite endpoint.c. But
it could be something else, Mike. One thing is certain: the present kernel
behaviour is far from the ideal.

I insist: engage implicit feedback sync for the playback, disable it for
the captures and inspect your dyndbg logs for the failure cases in starting
JACK that happen when you don't patch endpoint.c.

If you see something related to sync_ep then the problem lies elsewhere,
specifically inside pcm.c

Although whether this behavior is quirky or typical I don't know. I do know
> that the GT-1 does not claim to be USB class compliant. It requires custom
> drivers on Windows and macOS and does not work at all with iOS, so it
> definitely seems to be somewhat non-standard.
>

Non-standard to say the least. I really hope that's not the case or it'll
give Takashi Iwai plenty of headache ;-)

> Mike
>
> On Thu, Apr 8, 2021 at 12:13 PM Geraldo Nascimento <
> geraldogabriel at gmail.com> wrote:
>
>> Hey Mike, got another question for you, sorry about that.
>>
>> If you comment out your GT-1 IMPLICIT_FB_GENERIC_DEV() entry and instead
>> let audioformat_implicit_fb_quirk() call add_roland_implicit_fb() with
>> everything else customized your way, your custom endpoint.c etc., does it
>> still work? And does it work crystal clear?
>>
>> I'm sorry about this test this, test that, provide the dyndbg logs
>> attitude but I'm honestly just trying to help you write the best patch
>> possible.
>>
>> Thank you,
>> Geraldo
>>
>> Em Qui, 8 de abr de 2021 02:05, Geraldo Nascimento <
>> geraldogabriel at gmail.com> escreveu:
>>
>>> Mike, I just realize I missed reply-to-all once again.
>>>
>>> In case you were wondering, I was looking for a sync_ep event right in
>>> the start of the loading of snd-usb-audio. It happens inside pcm.c. If you
>>> see one of those specifically let me know.
>>>
>>> With my Pioneer device, the DDJ-SR2, snd-usb-audio recognized the
>>> device's implicit feedback EP as sync_ep and that prevented JACK from
>>> starting.
>>>
>>> Em Qua, 7 de abr de 2021 23:47, Mike Oliphant <oliphant at nostatic.org>
>>>> escreveu:
>>>>
>>>>> Hi Geraldo - I had the same thought and I have tried enabling playback
>>>>> feedback while turning off capture feedback without my endpoint patch. It
>>>>> doesn't work - at least not on the GT-1. Playback/capture fails to start.
>>>>>
>>>>
>>>> Would you mind sharing the dyndbg logs for the failure case from the
>>>> point you connect the device, and trying to start JACK?
>>>>
>>>> Sometimes failures elucidate more than success cases.
>>>>
>>>>
>>>>> Speaking of testing, if you (or anyone else following this discussion)
>>>>> have one of these Roland/BOSS devices, it would be great if you could try
>>>>> testing with playback feedback enabled and my endpoint.c workaround.
>>>>>
>>>>
>>>> I don't have the hardware, but I'm sure it works. The problem is that
>>>> endpoint.c code really shouldn't be disabled unless all the other options
>>>> fail, and even then, I'm not so sure the maintainers will be so willing to
>>>> disable it. So let's keep some other options on the table by debugging a
>>>> little further.
>>>>
>>>> Thank you,
>>>> Geraldo
>>>>
>>>>
>>>>> Mike
>>>>>
>>>>> On Wed, Apr 7, 2021 at 7:21 PM Geraldo Nascimento <
>>>>> geraldogabriel at gmail.com> wrote:
>>>>>
>>>>>> Thanks, Mike, one thing I can tell you right off the bat is that if
>>>>>> you turn implicit feedback sync on for both endpoints at the same time like
>>>>>> you mentioned, without your workaround inside endpoint.c, they most likely
>>>>>> will stall because they would both be waiting for each other to start.
>>>>>>
>>>>>> Have you tried engaging the generic implicit feedback sync just for
>>>>>> the playback, thereby commenting out the capture implicit feedback entries,
>>>>>> and re-enabling snd_usb_endpoint_implicit_feedback_sink(ep) inside
>>>>>> endpoint.c? Does playback still fail to start?
>>>>>>
>>>>>> If you choose to perform this simple test please share with us the
>>>>>> test's dyndbg logs right from the moment you turn on the device, not just
>>>>>> after you start jackd.
>>>>>>
>>>>>> Sorry for the endless do-this, do-that. Such is the pace of
>>>>>> teledebugging
>>>>>>
>>>>>>
>>>>>> Em Qua, 7 de abr de 2021 21:38, Mike Oliphant <oliphant at nostatic.org>
>>>>>> escreveu:
>>>>>>
>>>>>>> Geraldo - here is the dmesg output when starting up jackd for
>>>>>>> capture/playback:
>>>>>>>
>>>>>>> With my patch:
>>>>>>>
>>>>>>> [  136.734081] usb 3-1.2: Open EP 0x8e, iface=2:1, idx=0
>>>>>>> [  136.734088] usb 3-1.2:   channels=2, rate=44100, format=S32_LE,
>>>>>>> period_bytes=256, periods=2, implicit_fb=0
>>>>>>> [  136.734092] usb 3-1.2: Setting usb interface 2:0 for EP 0x8e
>>>>>>> [  136.734233] usb 3-1.2: Setting usb interface 2:1 for EP 0x8e
>>>>>>> [  136.734381] usb 3-1.2: 2:1 Set sample rate 44100, clock 0
>>>>>>> [  136.734388] usb 3-1.2: Setting params for data EP 0x8e, pipe
>>>>>>> 0x70480
>>>>>>> [  136.734394] usb 3-1.2: Set up 12 URBS, ret=0
>>>>>>> [  136.734466] usb 3-1.2: Open EP 0xd, iface=1:1, idx=0
>>>>>>> [  136.734468] usb 3-1.2:   channels=2, rate=44100, format=S32_LE,
>>>>>>> period_bytes=256, periods=2, implicit_fb=1
>>>>>>> [  136.734471] usb 3-1.2: Reopened EP 0x8e (count 1)
>>>>>>> [  136.734473] usb 3-1.2: Setting usb interface 1:0 for EP 0xd
>>>>>>> [  136.734605] usb 3-1.2: Setting usb interface 1:1 for EP 0xd
>>>>>>> [  136.734762] usb 3-1.2: 1:1 Set sample rate 44100, clock 0
>>>>>>> [  136.734774] usb 3-1.2: Setting params for data EP 0xd, pipe
>>>>>>> 0x68400
>>>>>>> [  136.734781] usb 3-1.2: Set up 12 URBS, ret=0
>>>>>>> [  136.734830] usb 3-1.2: Starting data EP 0xd (running 0)
>>>>>>> [  136.734873] usb 3-1.2: 12 URBs submitted for EP 0xd
>>>>>>> [  136.734876] usb 3-1.2: Starting data EP 0x8e (running 0)
>>>>>>> [  136.734906] usb 3-1.2: 12 URBs submitted for EP 0x8e
>>>>>>> [  136.740627] usb 3-1.2: Starting data EP 0x8e (running 1)
>>>>>>> [  136.740634] usb 3-1.2: 2:1 Start Capture PCM
>>>>>>> [  136.740635] usb 3-1.2: 1:1 Start Playback PCM
>>>>>>>
>>>>>>>
>>>>>>> With the unpatched kernel:
>>>>>>>
>>>>>>> [  344.813203] usb 3-1.2: Open EP 0x8e, iface=2:1, idx=0
>>>>>>> [  344.813210] usb 3-1.2:   channels=2, rate=44100, format=S32_LE,
>>>>>>> period_bytes=256, periods=2, implicit_fb=1
>>>>>>> [  344.813213] usb 3-1.2: Open EP 0xd, iface=1:1, idx=0
>>>>>>> [  344.813215] usb 3-1.2:   channels=2, rate=44100, format=S32_LE,
>>>>>>> period_bytes=256, periods=2, implicit_fb=0
>>>>>>> [  344.813217] usb 3-1.2: Setting usb interface 2:0 for EP 0x8e
>>>>>>> [  344.813367] usb 3-1.2: Setting usb interface 2:1 for EP 0x8e
>>>>>>> [  344.813494] usb 3-1.2: 2:1 Set sample rate 44100, clock 0
>>>>>>> [  344.813503] usb 3-1.2: Setting params for data EP 0x8e, pipe
>>>>>>> 0x70580
>>>>>>> [  344.813509] usb 3-1.2: Set up 12 URBS, ret=0
>>>>>>> [  344.813513] usb 3-1.2: Setting usb interface 1:0 for EP 0xd
>>>>>>> [  344.813587] usb 3-1.2: Setting usb interface 1:1 for EP 0xd
>>>>>>> [  344.813758] usb 3-1.2: 1:1 Set sample rate 44100, clock 0
>>>>>>> [  344.813765] usb 3-1.2: Setting params for data EP 0xd, pipe
>>>>>>> 0x68500
>>>>>>> [  344.813770] usb 3-1.2: Set up 2 URBS, ret=0
>>>>>>> [  344.813858] usb 3-1.2: Reopened EP 0xd (count 1)
>>>>>>> [  344.813883] usb 3-1.2: Starting data EP 0xd (running 0)
>>>>>>> [  344.813899] usb 3-1.2: 2 URBs submitted for EP 0xd
>>>>>>> [  344.819499] usb 3-1.2: Starting data EP 0x8e (running 0)
>>>>>>> [  344.819532] usb 3-1.2: 12 URBs submitted for EP 0x8e
>>>>>>> [  344.819535] usb 3-1.2: Starting data EP 0xd (running 1)
>>>>>>> [  344.819538] usb 3-1.2: 2:1 Start Capture PCM
>>>>>>> [  344.819540] usb 3-1.2: 1:1 Start Playback PCM
>>>>>>> [  344.820663] usb 3-1.2: Stopping data EP 0xd (running 2)
>>>>>>> [  344.820688] usb 3-1.2: Stopping data EP 0x8e (running 1)
>>>>>>> [  344.820693] usb 3-1.2: 2:1 Stop Capture PCM
>>>>>>> [  344.820695] usb 3-1.2: Stopping data EP 0xd (running 1)
>>>>>>> [  344.820697] usb 3-1.2: 1:1 Stop Playback PCM
>>>>>>> [  344.833799] usb 3-1.2: Starting data EP 0xd (running 0)
>>>>>>> [  344.833852] usb 3-1.2: 2 URBs submitted for EP 0xd
>>>>>>> [  344.833867] usb 3-1.2: Starting data EP 0x8e (running 0)
>>>>>>> [  344.833888] usb 3-1.2: 12 URBs submitted for EP 0x8e
>>>>>>> [  344.833890] usb 3-1.2: Starting data EP 0xd (running 1)
>>>>>>> [  344.833891] usb 3-1.2: 2:1 Start Capture PCM
>>>>>>> [  344.833893] usb 3-1.2: 1:1 Start Playback PCM
>>>>>>>
>>>>>>> On Wed, Apr 7, 2021 at 1:30 PM Geraldo Nascimento <
>>>>>>> geraldogabriel at gmail.com> wrote:
>>>>>>>
>>>>>>>> Em Qua, 7 de abr de 2021 17:16, Mike Oliphant <
>>>>>>>> oliphant at nostatic.org> escreveu:
>>>>>>>>
>>>>>>>>> Hi Geraldo - I don't have that patch applied, but it shouldn't
>>>>>>>>> make any behavioral difference - it just seems to be simplifying the code.
>>>>>>>>>
>>>>>>>>
>>>>>>>> Point taken.
>>>>>>>>
>>>>>>>>
>>>>>>>>> The issue is that the BOSS GT-1 *does* need implicit feedback on
>>>>>>>>> playback to avoid clock timing issues, and the current behavior is
>>>>>>>>> disabling that feedback.
>>>>>>>>>
>>>>>>>>
>>>>>>>> Mike, would you mind posting some dyndbg logs for both the stock
>>>>>>>> behaviour and your endpoint.c/generic quirk modified behaviour?
>>>>>>>>
>>>>>>>> Just add snd_usb_audio.dyndbg=+p to your kernel options and please
>>>>>>>> share the logs.
>>>>>>>>
>>>>>>>>
>>>>>>>>> Mike
>>>>>>>>>
>>>>>>>>> On Wed, Apr 7, 2021 at 1:04 PM Geraldo Nascimento <
>>>>>>>>> geraldogabriel at gmail.com> wrote:
>>>>>>>>>
>>>>>>>>>> Hey Mike, did you catch the latest patch by Takashi Iwai for
>>>>>>>>>> capture quirky devices?
>>>>>>>>>>
>>>>>>>>>> You can find it here:
>>>>>>>>>> https://patchwork.kernel.org/project/alsa-devel/patch/20210406113837.32041-1-tiwai@suse.de/
>>>>>>>>>>
>>>>>>>>>> Em Qua, 7 de abr de 2021 16:55, Mike Oliphant <
>>>>>>>>>> oliphant at nostatic.org> escreveu:
>>>>>>>>>>
>>>>>>>>>>> I had thought that the recent implicit feedback changes were
>>>>>>>>>>> fully working
>>>>>>>>>>> on the BOSS GT-1, but it turns out that I just hadn't tested
>>>>>>>>>>> well enough.
>>>>>>>>>>>
>>>>>>>>>>> Audio playback and capture works, but with periodic dropouts. I
>>>>>>>>>>> get the
>>>>>>>>>>> exact same behavior as I did with the quirk to completely
>>>>>>>>>>> disable implicit
>>>>>>>>>>> feedback. Without the implicit feedback, you get dropouts from
>>>>>>>>>>> clock drift
>>>>>>>>>>> - how bad probably varies from card to card. On mine it is every
>>>>>>>>>>> second or
>>>>>>>>>>> so.
>>>>>>>>>>>
>>>>>>>>>>> If I switch playback feedback for the GT-1 to generic by doing
>>>>>>>>>>> "IMPLICIT_FB_GENERIC_DEV(0x0582, 0x01d6)", I get the previous
>>>>>>>>>>> old behavior,
>>>>>>>>>>> which is that playback completely fails to start.
>>>>>>>>>>>
>>>>>>>>>>> With generic playback feedback, and using my previous patch to
>>>>>>>>>>> endpoint.c
>>>>>>>>>>> to avoid playback waiting on capture mentioned here:
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> https://mailman.alsa-project.org/pipermail/alsa-devel/2020-January/161951.html
>>>>>>>>>>>
>>>>>>>>>>> playback and capture work perfectly for me.
>>>>>>>>>>>
>>>>>>>>>>