[BUG] SoundWire bus-wide IO timeouts on CS42L45+CS35L57 after enabling SDCA_CLASS - Dell XPS 16 DA16260 (follow-up to #5720)

[CodeBleu]

Related to #5720 (closed) — same hardware/subsystem (10280dba, Dell XPS 16
DA16260, SKU 0DBA), same root cause identified there
(CONFIG_SND_SOC_SDCA_CLASS missing), confirmed independently here on
Gentoo/OpenRC rather than Arch. However, after applying that fix I hit a
different, unresolved failure — #5720 ended in a fully working card, this
does not. Opening separately since it's a distinct symptom from a different
starting point in the boot sequence.

System information

• Machine: Dell XPS 16 DA16260 (2026), BIOS 1.9.0 (2026-07-10)
• CPU: Intel Core Ultra 7 358H (Panther Lake)
• System SKU: 0DBA (dmidecode -s system-sku-number)
• Audio hardware: Cirrus Logic CS42L45 (jack codec, link 0) + 4x Cirrus
  Logic CS35L57 Rev B2 OTP1 (amps, links 2/3)
• Distro: Gentoo Linux (OpenRC)
• Kernel: 7.1.5 (gentoo-sources, genpatches applied)
• sof-firmware: 2025.12.2 (community build, sof-ptl.ri)
• linux-firmware: 20260622

Steps already taken

1. CONFIG_SND_SOC_SOF_HDA_LINK was disabled by default — enabling it was
   required to get hdac_bus_eml_get_mutex() to return a real mutex instead
   of the always-NULL stub in include/sound/hda-mlink.h, which caused an
   immediate -ENODEV at hda_sdw_probe() before any SoundWire activity.
2. CONFIG_PINCTRL_INTEL_PLATFORM was not enabled — required for Panther
   Lake GPIO/pinctrl. Without it, amp probing failed with Failed to add gpio mapping to AF01 (-ENXIO), matching the documented failure mode in
   Documentation/sound/codecs/cs35l56.rst.
3. Added a local DMI quirk table entry for SKU 0DBA (mirroring the existing
   0DD6 entry in sof_sdw.c) — confirmed via A/B test this has no
   effect on the timeout issue below (identical dmesg with/without it).
4. Updated to the latest available kernel point release (7.1.5) and latest
   linux-firmware (20260622) — no change in behavior at that point.
5. Per No audio on Dell XPS 16 2026 (Panther Lake) - CS42L45 + CS35L57 - no driver for CS42L45, missing firmware #5720: enabled CONFIG_SND_SOC_SDCA_CLASS=m (module, since
   SND_SOC/SOUNDWIRE are also =m on this config — Kconfig won't allow
   it built-in when dependencies are modular). This was the missing piece
   to get the CS42L45 driver binding and loading firmware correctly:

[ 17.563180] sdca_class sdw:0:0:01fa:4245:01: SWFT not available
[ 17.573463] Loading firmware: sdca/1fa/1028/dba/2.bin
[ 17.574342] snd_soc_sdca_class_function snd_soc_sdca.SmartMic.0: loading SWF: 1fa-2-1
[ 17.578944] Loading firmware: sdca/1fa/1028/dba/42450005.bin
[ 17.579372] snd_soc_sdca_class_function snd_soc_sdca.SmartMic.0: loading SWF: 1fa-42450005-1301
[ 18.148725] snd_soc_sdca_class_function snd_soc_sdca.UAJ.1: loading SWF: 1fa-3-1
[ 18.276777] input: HID sdw:0:0:01fa:4245:01 Consumer Control as ...

This confirms the #5720 root cause is real and not distro-specific.

Current issue: bus-wide IO timeouts ~20s after enumeration

With item 5 applied, every active SoundWire link — CS42L45 on link 0 and
all four CS35L57 amps on links 2/3 — starts failing with IO/SCP timeouts
roughly 20 seconds after successful enumeration, each link failing within
~500ms of the others:

[ 21.092274] soundwire_intel soundwire_intel.link.0: SCP Msg trf timed out
[ 21.092289] soundwire sdw-master-0-0: trf on Slave 6 failed:-5 write addr 8088 count 0
[ 21.092302] snd_soc_sdca_class_function snd_soc_sdca.SmartMic.0: failed to restore register cache: -5
[ 21.596244] soundwire_intel soundwire_intel.link.0: IO transfer timed out, cmd 3 device 6 addr 5d len 1
[ 21.596300] sdca_class sdw:0:0:01fa:4245:01: Failed to sync masks in 5d
...
[ 26.131463] soundwire_intel soundwire_intel.link.2: IO transfer timed out, cmd 3 device 2 addr c1 len 1
[ 26.131500] soundwire_intel soundwire_intel.link.2: RX AVAIL 63 too long
...
[ 29.155608] soundwire_intel soundwire_intel.link.3: IO transfer timed out, cmd 3 device 2 addr c1 len 1
[ 29.155677] soundwire_intel soundwire_intel.link.3: RX AVAIL 63 too long

Result: no card registers.

$ cat /proc/asound/cards
--- no soundcards ---
$ aplay -l
aplay: device_list:279: no soundcards found...

The synchronized, sequential-by-link failure pattern (link 0 first, then 2,
then 3, each ~500ms apart) suggests a bus-wide event — possibly clock-stop
or power-state related — rather than a per-device problem. soundwire_intel
and snd_sof don't expose any module parameters I could use to test that on
this kernel (modinfo soundwire_intel | grep -i clock and
modinfo snd_sof | grep -i quirk both return nothing).

Expected behavior

SoundWire bus transactions should succeed after enumeration, and a
sof-soundwire card should register with working speakers/headphones/mic,
matching the result reported in #5720.

Additional notes

• .config diffs and full DMI quirk patch available on request.
• Happy to test patches or pull additional debug output (soundwire debugfs,
  ACPI dump) — fully reproducible every boot.