Re: [PATCH 3/3] ASoC: SOF: Intel: hda: duplicate _ADR entries share one amp index

[Pierre-Louis Bossart <pierre-louis.bossart@linux.dev>]

On 8/5/26 00:59, Sergey Lebedev wrote:
> find_acpi_adr_device() assigns each amplifier a name prefix carrying an
> index ("rt1320-1", "rt1320-2", or Left/Right) and advances that index
> once per _ADR entry. Firmware that describes one physical part with two
> _ADR entries therefore consumes two indices for one device.
> 
> The Microsoft Surface Pro 11 (Intel) does exactly that. Link 0 carries:
> 
>   SWRA  _ADR 0x000030025D132000   SDCA class 0
>   SWRB  _ADR 0x000030025D132001   SDCA class 1
> 
> identical but for the class id: same link, same manufacturer, part and
> version, same unique id 0. The part reports class 1, so SWRB is the one
> that enumerates and SWRA never attaches on any boot or firmware version
> tested.
> 
> Both still reach this function, so SWRA takes index 1 and the real
> amplifier is named "rt1320-2". Its controls appear as "rt1320-2 OT23 L/R
> Switch". The stock sof-soundwire UCM profile expects the first
> amplifier, so it enables switches on a device that is not present, and
> the speakers stay silent while everything else reports success.
> 
> Compare entries that differ only in class id and give the later one the
> earlier one's name prefix, jumping past the amplifier-index increment so
> a repeated description consumes one index rather than several.
> 
> Testing the peripheral's attach status instead does not work here, and
> was tried: at machine-select time neither entry has attached yet, so a
> status test finds both unattached, no amplifier is matched at all, and
> the card falls back to the HDMI-only HDA machine driver. Comparing
> addresses needs no runtime state.

yeah but that means two methods to detect the presence of this ghost
device...

> 
> Signed-off-by: Sergey Lebedev <lsa.uz@pm.me>
> ---
>  sound/soc/sof/intel/hda.c | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/sound/soc/sof/intel/hda.c b/sound/soc/sof/intel/hda.c
> index 4dbba9186..1d60a8caa 100644
> --- a/sound/soc/sof/intel/hda.c
> +++ b/sound/soc/sof/intel/hda.c
> @@ -1245,6 +1245,22 @@ static struct snd_soc_acpi_adr_device *find_acpi_adr_device(struct device *dev,
>  			((u64)(sdw_device->id.sdw_version & 0xF) << 44) |
>  			((u64)(sdw_device->bus->link_id & 0xF) << 48);
>  
> +	/*
> +	 * Firmware may describe a single physical part with more than one _ADR
> +	 * entry, differing only in SDCA class id. Those entries are the same
> +	 * device: they must share a name prefix, and only the first of them may
> +	 * consume an amp index. Otherwise the part that actually enumerates is
> +	 * named as though it were the second amplifier, and UCM profiles
> +	 * written for the first one address a device that is not there.
> +	 */
> +	for (j = 0; j < index; j++) {
> +		if ((adr_dev[j].adr & ~SDW_CLASS_ID_MASK) ==
> +		    (adr_dev[index].adr & ~SDW_CLASS_ID_MASK)) {
> +			adr_dev[index].name_prefix = adr_dev[j].name_prefix;
> +			goto done_name_prefix;
> +		}
> +	}
> +

The DMI quirk removes the need for the two patches that detect the
presence of the ghost device through two different mechanisms.

Can you try with a DMI quirk and let us know if this is good enough?