[LNL][X1C13] RT1318 UNATTACHED: empty Cadence PING (no Device0) + continuous CTRL_CLASH after Speaker STOP

[aramsdale]

Describe the bug

ThinkPad X1 Carbon Gen 13 21NS0012US (Lunar Lake, 8086:a828, sof-audio-pci-intel-lnl). Topology sof-lnl-rt713-l0-rt1318-l1-2ch.tplg, ADSPFW 2.14.1.1, kernel 7.1.9-arch1-2 (Omarchy 4.0.2).

Speakers worked for several days on a brand-new machine. On 2026-09-05 21:26:36 PDT, Speaker STOP on an already-running stream after s2idle/fingerprint USB activity produced link-1 DATA_CLASH + CTRL_CLASH, then DPN_PortCtrl -61, then clock-stop -110. Since then:

• sdw:0:0:025d:0713:01 RT713 Attached (jack works)
• sdw:0:1:025d:1318:01 RT1318 UNATTACHED
• Cadence MCP_SLAVE_STAT link 1 = 0 (no device 0)
• Cadence MCP_INTSTAT link 1 = 0x80000400 (IRQ + CTRL_CLASH)
• Clash on every start_bus; every frame while the clock runs
• Speakers also missing in Windows (FUNC_0713 only)

Not #3442 (PING agrees with sysfs). Not #5495 re-enum (needs Device0). Not #5914 (amps stayed Attached). Not #4535 (they had Device0). Not sof#10755 (card existed; speakers played).

Documentation/driver-api/soundwire/error_handling.rst says clash is log-only and some slave faults are likely unrecoverable. Asking whether any host recovery exists that is not another sdw_cdns_exit_reset(). If a clash during STOP can leave a soldered amp hung across OS boundaries, that is a fleet risk, not a mute/UCM ticket.

Distro companion: omacom/omarchy#10916

To Reproduce

Cannot reproduce the first death on demand. Hung state is persistent:

1. Boot this X1C13 on current kernel/SOF.
2. cat /sys/bus/soundwire/devices/sdw:0:1:025d:1318:01/status → UNATTACHED
3. RPM-wake the link and read master-0-1/intel-sdw/cdns-registers: MCP_SLAVE_STAT=0, INTSTAT=0x80000400
4. dmesg shows soundwire_intel.link.1: Bus clash for control word on first start_bus

Reproduction Rate

Hung state: 10/10 boots since 2026-09-05, including pack-off and Windows SST OED+BUS disable/enable.

Expected behavior

RT1318 answers PING as device 0 (or its sticky device number), enumerates, Speakers endpoint works. A clash during STOP should not leave the amp driving Col0 with no Device0 across reboot and OS switch.

Impact

Internal speakers dead on a new X1C13. Jack/USB-C/HDMI still work. If clash-without-Device0 is unrecoverable in software, this is a hardware-RMA class event that Linux can cause and cannot undo.

Environment

• Kernel: 7.1.9-arch1-2 (LNL IRQ-before-exit_reset present)
• SOF: ADSPFW 2.14.1.1
• Topology: sof-lnl-rt713-l0-rt1318-l1-2ch.tplg
• Platform: Lenovo ThinkPad X1 Carbon Gen 13 21NS0012US, BIOS N4BET77W (1.47)
• Live clock-stop quirk: 2 on snd-sof-intel-hda-generic (experiment; default 8 also failed to attach). sdw_md_flags=0

Tried / please do not recommend repeating

cdns-hw-reset, PCI remove/rescan of 0000:00:1f.3 (#4809), pack-off, Windows SST OED+BUS disable/enable, sof#10755 2.11 overlay, RPM pin on.

Still unused (confirm-first, not expected to create Device0)

Cadence IP MCP_CONFIG dump at ip_offset 0x4000 (read-only). sof_debug=128 / skip_imr_boot. sdw_md_flags=0x700 on link 1 (looks like RPM pin on: continuous clash). Quirk=1 (same).

Questions

1. Is there a host-side recovery for empty PING including device 0 + continuous CTRL_CLASH that is not another sdw_cdns_exit_reset() / start_bus()?
2. Does live link-1 50×8 / 19.2 MHz vs ACPI 50×4 / 9.6 MHz matter once the slave is already fighting Col0, or is that a red herring?
3. Should a manager that sees continuous CTRL_CLASH stop the clock rather than keep generating frames?
4. If this is the unrecoverable electrical class ([MTL][SDW] rt1316-sdca sdw:2:025d:1316:01: SDW_SCP_INT1 (BUS_CLASH) read failed:-61 #4535 won't-fix, except they had Device0), please say so. Remaining path is Lenovo / system-board (RT1318 is soldered, HMM FRU Introducing virtual FE  #8).
5. Is there anything userspace/distros should stop doing on LNL SoundWire after a clash (RPM hammering start_bus, PCI-level recover) so the next unit does not deepen a hung link?

Please do not suggest F10 UEFI audio (this SKU has no audio module), sof#10755 overlay, or sdw_link_mask (#4353 is a ghost link; SWD1 is real).

Console / register dump

sdw:0:0:025d:0713:01 Attached
sdw:0:1:025d:1318:01 UNATTACHED

link 0 MCP_SLAVE_STAT=0x1000  MCP_INTSTAT=0  FRAME ~50x4  CLK div 3
link 1 MCP_SLAVE_STAT=0       MCP_INTSTAT=0x80000400  FRAME ~50x8  CLK div 1
       MCP_CONTROL=0  CONFIG_UPDATE=0  INTMASK=0x8000f704 (Device0 unmasked)