Re: [PATCH v2] ALSA: usb-audio: add QUIRK_FLAG_ALWAYS_SET_RATE for Mackie DLZ Creator XS

On Sat, 08 Aug 2026 19:27:26 +0200,
JJ Macalinao wrote:
> 
> set_sample_rate_v2v3() returns early when the clock already reports the
> requested rate:
> 
> 	prev_rate = get_sample_rate_v2v3(chip, fmt->iface,
> 					 fmt->altsetting, clock);
> 	if (prev_rate == rate)
> 		goto validation;
> 
> A device advertising exactly one sample rate always takes this branch, so
> it never receives a SET_CUR for CS_SAM_FREQ_CONTROL at all.
> 
> The Mackie DLZ Creator XS (0a73:003a, 14 in / 4 out, 48 kHz only) requires
> that write.  Without it the device drops off the USB bus roughly 0.2-1.8 s
> into any stream, clearing its port CONNECTION bit; captured audio is
> byte-correct until the instant it vanishes.
> 
> USBPcap traces of a cold-booted device on Windows show SET_CUR 48000 issued
> unconditionally on every stream start, followed by clean streaming.  The
> device is otherwise driven with plain class-compliant UAC2 - it also works
> on iOS, which cannot load a vendor driver - so no vendor-specific
> initialization is involved.
> 
> The device is self-powered, so the resulting state survives a USB replug:
> initializing it on any host that issues the write leaves it working on
> Linux until it is power-cycled, which made the failure look intermittent.
> 
> Add a quirk flag rather than dropping the early exit, since the opposite
> requirement also exists in-tree: QUIRK_FLAG_FIXED_RATE suppresses rate
> setting for single-rate devices (JBL Quantum610/810).  The two behaviors
> are device-dependent and cannot both be the default.
> 
> A/B on identically cold-booted hardware, same kernel, same port, repeated
> twice:
> 
>   without the flag  device dropped after 5-6 s, then again after 3-4 s
>   with the flag     20 s playback followed by 20 s of 14-channel capture,
>                     960000 frames, zero re-enumerations
> 
> This change was developed with an AI coding assistant.  The assistant did
> the trace analysis that located the bug and wrote the patch and this
> changelog; the hardware testing, the cold-boot cycles and the decision to
> submit were the author's.  Several earlier hypotheses it proposed - URB
> queue depth, isochronous packet under-allocation, endpoint start ordering -
> were disproven by measurement before this one.
> 
> The bug was located with usbmon on Linux and USBPcap on Windows, by
> diffing an enumeration capture of a cold-booted device on each host.
> Verified on physical hardware by the A/B above.
> 
> Assisted-by: Claude-Code:claude-opus-5
> Signed-off-by: JJ Macalinao <[email protected]>
> ---
> v2: skip the GET_CUR entirely when the flag is set, per Takashi's review,
>     rather than issuing it and discarding the result.
> 
>     prev_rate is initialized to 0 because it is still read below - the
>     !cur_rate fallback and the TEAC 0x0644 block. That is not a new state:
>     get_sample_rate_v2v3() already returns 0 when the GET fails, so both
>     paths handle it today.
> 
>     Retested on cold-booted hardware, same unit and port: 20 s playback
>     followed by 20 s of 14-channel capture, 960000 frames, no
>     re-enumeration.

Applied to for-next branch now.  Thanks.


Takashi