Index | Thread | Search

From:
Alexandre Ratchov <alex@caoua.org>
Subject:
Re: uaudio(4): dynamic feature unit control size for UAC2
To:
Jonathan Armani <jonathan@armani.tech>
Cc:
Andrew Kloet <andrew@kloet.net>, tech@openbsd.org
Date:
Mon, 16 Nov 2026 21:49:24 +0100

Download raw body.

Thread
On Sun, Nov 15, 2026 at 04:24:52PM +0100, Jonathan Armani wrote:
> 
> Hello Andrew, Alex,
> 
> Perfect occupation for a rainy day :)
> 
> From what I read in your USB dump, the feature units are 18 bytes
> long, i.e. master + 2 per-channel bitmaps, but their sources have
> 8 and 4 channels so we run out of descriptor in the per-channel
> loop.
> 
> If I understand it correctly, your diff works by accident : on UAC2
> it reads the low byte of bmaControls(0) as bControlSize, which is 0
> here, so size = 0 and uaudio_getnum() never fails.
> 
> Could you try the diff below instead ? It stops the loop when the
> descriptor is exhausted, which also covers the existing "no
> per-channel descriptors" special case.
> 
> Also, could you send a raw dump of the descriptor ?  I'd expect
> 1 byte left at the failure point, not 28.
> 
> --- a/sys/dev/usb/uaudio.c
> +++ b/sys/dev/usb/uaudio.c
> @@ -1494,13 +1494,13 @@ uaudio_process_unit(struct uaudio_softc
>  			ctl >>= 2;
>  		}
>  
> -		/*
> -		 * certain devices provide no per-channel control descriptors
> -		 */
> -		if (p.wptr - p.rptr == 1)
> -			break;
> -
>  		for (j = 0; j < u->nch; j++) {
> +			/*
> +			 * certain devices provide fewer per-channel
> +			 * control descriptors than channels (or none)
> +			 */
> +			if (p.wptr - p.rptr < size)
> +				break;
>  			if (!uaudio_getnum(&p, size, &ctl))
>  				return 0;
>  			ctl = uaudio_feature_fixup(sc, ctl);
> 

This device has no usb-exposed controls (all FEATURE bitmaps are
empty), which makes perfect sense for a mixer with pleny of knobs!  So
I guess the 2 enties are just "forgotten" by the hardware designers or
are there on purpose, ex. to workaround bugs in the driver of a
popular operating system I won't name.

The above diff may help other devices as well and this is exactly the
approach of the driver: be flexible about the uac spec, rather than
using kirks.

OK ratchov