[REGRESSION 6.12.y] Please backport 70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")

[REGRESSION 6.12.y] Please backport 70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")

[The Darkbird]

Hello,

First of all: I noticed this regression on my LMDE 7 install and let 
Claude analyze the problem. I thoroughly read its analyses, read the kernel 
code diff involved, and I deemed it probable enough to mail it. However, I 
have no technical experience in this matter, nor with kernel-related coding, 
nor reporting regressions like these, so keep that in mind. I sincerely hope 
I won't be wasting your time with AI-generated nonsense.

-----

Please backport the following upstream commit to 6.12.y:

  70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")

It fixes ef5749ef8b30 ("ALSA: usb-audio: Prevent excessive number of
frames"), which went into 6.12.114 as commit 81e4b0009f2a. The fix
itself did not follow: it is not in 6.12.114 or 6.12.115, and it is not
in queue - 6.12 as of 2026-11-03. It has been in mainline since
v6.19-rc7, and 6.18.y already has both commits (6.18.10).

On 6.12.114 and later, USB audio playback is completely silent on
devices that use implicit feedback. In data_ep_set_params(),
ep->max_urb_frames is only assigned for playback endpoints without
implicit feedback, so for implicit-feedback playback it stays 0. The
check added to prepare_playback_urb()

    if (counts < 0 || frames + counts >= ep->max_urb_frames)

is then always true. The loop breaks before the first packet, frames
stays 0, and the function returns -EAGAIN for every URB, so no audio
data is ever sent. Players that sync video to the audio clock (mpv,
Firefox) show a black picture.

Hardware: MOTU UltraLite-mk5 (07fd:000c), which runs in implicit
feedback mode according to /proc/asound/cardN/stream0.

  6.12.113 (Debian 6.12.113-1): works
  6.12.115 (Debian 6.12.115-2): no sound at all

The code above is identical in upstream v6.12.115, so this is not
specific to Debian.

Note that 70b4db7d2581 does not apply cleanly to 6.12.y, because the
loop in prepare_playback_urb() is one indentation level shallower
there. The change itself is the same single line, and it applies with
"patch -l". I have not built and tested a 6.12 kernel with the fix.

The original report of the regression on mainline:
https://bugzilla.kernel.org/show_bug.cgi?id=220997

#regzbot introduced: 81e4b0009f2a

Thanks,
D.

Re: [REGRESSION 6.12.y] Please backport 70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")

[Sasha Levin]

On 2026-11-03, The Darkbird <thedarkbird@protonmail.com> wrote:
> Please backport the following upstream commit to 6.12.y:
>
>   70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")

Queued for 6.12, thanks.

-- 
Thanks,
Sasha

Re: [REGRESSION 6.12.y] Please backport 70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")

[Salvatore Bonaccorso]

Hi Sasha,

On Wed, Nov 04, 2026 at 10:03:14AM -0500, Sasha Levin wrote:
> On 2026-11-03, The Darkbird <thedarkbird@protonmail.com> wrote:
> > Please backport the following upstream commit to 6.12.y:
> >
> >   70b4db7d2581 ("ALSA: usb-audio: Use the right limit for PCM OOB check")
> 
> Queued for 6.12, thanks.

The commit ef5749ef8b30 ("ALSA: usb-audio: Prevent excessive number
of frames") introducing the issue was as well backported down to
6.1.191 and 6.6.160 additionall yto the found 6.12.114.

Should thus 70b4db7d2581 ("ALSA: usb-audio: Use the right limit for
PCM OOB check") be queued as well for 6.6.y and 6.1.y?

Regards,
Salvatore