Skip to content

Panther Lake + TAS2783 speaker support for HP OmniBook Ultra 8EB4 #5732

Open

Description

@minhtet-htoon

Summary


The built-in speakers on the HP OmniBook Ultra Laptop 14-kd0xxx do not produce any sound. The headphone jack (Realtek RT712 SDCA on SoundWire link 3) and HDMI/DisplayPort audio (Intel Panther Lake HDMI HDA codec) both function correctly. No Speaker playback device is created; only "Jack Out", "Deepbuffer Jack Out", and HDMI outputs appear in aplay -l.

The root cause is that the TI TAS2783 SoundWire speaker amplifiers on SoundWire links 1 and 2 fail to initialize because the linux-firmware package does not include firmware/calibration data for HP board ID 8EB4.

Steps to Reproduce

  1. Boot Fedora 43 on an HP OmniBook Ultra Laptop 14-kd0xxx (DMI Board Name: 8EB4, System SKU: D1WB3UA#ABA).
  2. Observe that no Speaker playback device is listed in aplay -l.
  3. Attempt to play audio through the built-in speakers — no sound is produced.

Actual Results

The TAS2783 driver fails to load firmware for the speaker amplifiers. Two classes of failure have been observed:

Failure 1 — Missing firmware files (default state): The kernel requests firmware files named after the board ID (8EB4) but they do not exist:

slave-tas2783 sdw:0:1:0102:0000:01:d: Failed to read fw binary 8EB4-1-D.bin
slave-tas2783 sdw:0:2:0102:0000:01:c: Direct firmware load for 8EB4-2-C.bin failed with error -2

This also causes a suspend/resume timeout since the amplifiers never initialized:

slave-tas2783 sdw:0:1:0102:0000:01:a: intel_resume_child_device: pm_runtime_resume failed: -110
PM: suspend devices took 5.566 seconds!!

No Speaker PCM device is registered. Only the following playback devices appear:

  • device 0: Jack Out (rt712 headphone)
    
  • device 5: HDMI1
    
  • device 6: HDMI2
    
  • device 7: HDMI3
    
  • device 31: Deepbuffer Jack Out
    

Failure 2 — Symlinks to generic firmware (attempted workaround): Creating symlinks from 8EB4-{1,2}-{9,A,C,D}.bin to the generic ti/audio/tas2783/8EXX-X-*.bin files (which exist in linux-firmware for other HP boards) results in the firmware loading but containing no matching calibration profile for 8EB4:

slave-tas2783 sdw:0:2:0102:0000:01:9: fw with no files
soundwire_intel soundwire_intel.link.2: Program SCP Ack not received
soundwire sdw-master-0-2: trf on Slave 2 failed:-5 write addr 8800 count 0
slave-tas2783 sdw:0:2:0102:0000:01:9: init writes failed, err=-5
slave-tas2783 sdw:0:2:0102:0000:01:9: Slave 2 initialization failed: -61
slave-tas2783 sdw:0:2:0102:0000:01:9: PARITY error detected before INT mask is enabled
slave-tas2783 sdw:0:2:0102:0000:01:9: FW download failed: -5
slave-tas2783 sdw:0:2:0102:0000:01:9: sw reset failed, err=-5

The generic 8EXX-X-*.bin container files do not include a calibration entry for board ID 8EB4. The SoundWire bus then enters a cascading failure state (I/O errors, parity errors, SCP Ack failures) because the TAS2783 chip is left uninitialized.

Copying firmware binaries from the Windows driver package also does not work (same "fw with no files" result), as the binary format differs from what the Linux driver expects.

Expected Result

The linux-firmware package should include TAS2783 firmware/calibration data for HP board ID 8EB4, either as board-specific symlinks pointing to compatible generic firmware (if the generic profiles are updated to include 8EB4) or as dedicated firmware files. The built-in speakers should appear as a Speaker playback device and produce audio.

System Information

Component Detail
Laptop HP OmniBook Ultra Laptop 14-kd0xxx
DMI Board Name 8EB4
System SKU D1WB3UA#ABA
BIOS Version F.02
PCI Audio Device Intel Panther Lake SST (8086:e328), Subsystem: 103c:8eb4
Platform Intel Panther Lake
Kernel 7.0.0-0.rc6.260331gd0c3bcd5b897.50.fc45.x86_64
OS Fedora 43 (Workstation Edition)
alsa-sof-firmware 2025.12.2-1.fc43.noarch
alsa-utils 1.2.15.2-1.fc43.x86_64
alsa-lib 1.2.15.3-1.fc43.x86_64
SOF Firmware intel/sof-ipc4/ptl/sof-ptl.ri, ADSPFW version 2.14.1.1
DSP Driver Override snd_intel_dspcfg.dsp_driver=3

SoundWire Topology (from ACPI):

Link Device Mfg ID Part ID Role Status
1 TAS2783 0x0102 0x0000 Speaker amp (unique IDs: A, D) BROKEN — no firmware
2 TAS2783 0x0102 0x0000 Speaker amp (unique IDs: 9, C) BROKEN — no firmware
3 RT712 SDCA 0x025d 0x0712 Headphone/mic codec Working

ACPI SoundWire device IDs:

  • TI 0x0002390102000001
  • TI 0x00023c0102000001
  • TI 0x00013a0102000001
  • TI 0x00013d0102000001
  • Realtek 0x000330025d071201

Analysis of linux-firmware gap:

The current linux-firmware TAS2783 submission includes generic firmware files and symlinks for the following HP boards: 8E86, 8E87, 8EF5, 8EF6. Board 8EB4 is not included. Additionally, 8EF6 has its own unique firmware files (8EF6-X-9.bin, 8EF6-X-C.bin) separate from the generic 8EXX set, indicating that not all boards can share the same calibration data.

Additional Info

alsa-info.sh output:
http://alsa-project.org/db/?f=ad78c4fd6e1b98ccc169f2812b41c85c21a4b2db

Activity

  1. plbossart commented on Apr 16, 2026

    @plbossart
    Member

    snd_intel_dspcfg.dsp_driver=3 shouldn't be necessary. If it is there's an additional missing conf. Can you retry @minhtet-htoon ?

  2. minhtet-htoon commented on Apr 17, 2026

    @minhtet-htoon
    Author

    Hi, I'm still getting errors after I removed that kernel option. The errors are roughly the same still. I have tried copying the .bin files from the Windows Driver package that I downloaded from HP and it still gives the error. Stumbled on this PR in the project and I think it may be relevant.

    #5569

    Alsa Info After Removing option: https://alsa-project.org/db/?f=d0726a0c1f1b5b9b460e081393fcc3b701e9e330

  3. minhtet-htoon commented on May 8, 2026

    @minhtet-htoon
    Author

    I'm still having issues with this. Any insight would be appreciated. Please let me know if there is any additional testing I can do on my end. Thanks in advance!

  4. skriefal commented on May 14, 2026

    @skriefal

    I'm encountering the same behavior with an HP Omnibook X Laptop AI (14-ka0xxx). The same five devices are listed by aplay -l (jacks, HDMI, but no speaker output), even after linking the generic tas2783 firmware files to 8EA6-2-9.bin.zst and 8EA6-2-C.bin.zst.

    DMI board name: 8EA6
    System SKU: D18T4UA#ABA
    Platform: Intel Panther Lake (Core Ultra 7 356H)
    OS: CachyOS (latest)
    Kernel: 7.0.6

    Alsa Info (after linking firmware files):
    alsa-info.txt

  5. D3rPaPaH0d3n commented on Jun 7, 2026

    @D3rPaPaH0d3n

    I have the exact same laptop — HP OmniBook Ultra Laptop 14-kd0xxx, DMI board 8EB4, 4× TAS2783 on SoundWire links 1 & 2, RT712 on link 3, Intel Panther Lake. I opened #5802 before I found this issue; it's the same machine and the same failure (Msg NACKed for Slave 1, SDW_SCP_BUS_CLOCK_BASE write failed: -61, fw with no files). Happy to treat #5802 as a duplicate of this one.

    On Fedora 44 (kernel 7.0.11) the built-in speakers still don't work, even though a lot of the tas2783A work seems to have landed upstream already (port prepare handling, fw loading without pci bus, reading slave properties from ACPI, etc.).

    Does PR #5569 ("Tas2783a 4ch newbin1") cover the Panther Lake 8EB4 configuration, or only the MTL 4-channel ACPI match? If there's a branch/patch set I should try for PTL, I'm glad to build a kernel and test it — I can provide full dmesg, acpidump, SoundWire sysfs dumps and alsa-info. Just point me at what to test. Thanks!

  6. D3rPaPaH0d3n commented on Jun 7, 2026

    @D3rPaPaH0d3n

    Update: I upgraded the same HP OmniBook Ultra 14-kd0xxx / 8EB4 machine from Fedora 44 KDE to Fedora 45 Rawhide.

    Rawhide still does not expose an internal speaker PCM.

    Tested:

    • Fedora release 45 (Rawhide)
    • kernel 7.1.0-0.rc6.260604gba3e43a9e601.43.fc45.x86_64
    • linux-firmware-20260519-1.fc45.noarch
    • alsa-sof-firmware-2025.12.2-1.fc45.noarch
    • alsa-ucm-1.2.16-2.fc45.noarch

    aplay -l still shows only:

    • Jack Out
    • HDMI1
    • HDMI2
    • HDMI3
    • Deepbuffer Jack Out

    No internal Speaker PCM is created.

    The kernel log still shows the same failure area:

    • No SoundWire machine driver found
    • Use SoundWire default machine driver with function topologies
    • TAS2783 FW download failed: -61
    • fw with no files
    • SoundWire Program SCP Ack not received
    • init writes failed, err=-5
    • SDW_SCP_BUS_CLOCK_BASE write failed:-61
    • Slave initialization failed: -61

    So the issue is still reproducible on Fedora Rawhide / kernel 7.1-rc6. I have the full Rawhide kernel audio log available if needed.

  7. bardliao commented on Jun 16, 2026

    @bardliao
    Collaborator

    It seems be duplicated to #5760 @niranjanhyti Could you take a look?

  8. niranjanhyti commented on Jul 21, 2026

    @niranjanhyti
    Collaborator
  9. kasemeyer commented on Aug 29, 2026

    @kasemeyer

    Same machine as @minhtet-htoon and @D3rPaPaH0d3n — HP OmniBook Ultra 14-kd0xxx, board 8EB4, 4x TAS2783 on links 1/2, RT712 on link 3 — on CachyOS.

    The firmware this issue asks for now ships, and the speakers still do not work. I think that retires the premise the report is built on.

    linux-firmware 20260810 contains real, board-specific blobs for 8EB4 — not symlinks to the generic 8EXX containers, which is what Failure 2 above was testing:

    $ ls -lL /lib/firmware/ti/audio/tas2783/8EB4*
    -rw-r--r-- 1 root root 7230  8EB4-1-0xA.bin.zst
    -rw-r--r-- 1 root root 7230  8EB4-1-0xD.bin.zst
    -rw-r--r-- 1 root root 7280  8EB4-2-0x9.bin.zst
    -rw-r--r-- 1 root root 7298  8EB4-2-0xC.bin.zst
    

    All four decompress cleanly (zstd -t passes, 40746 bytes each), and the names line up with the board ID plus the link number and the enumerated unique IDs a, d, 9, c. So neither Failed to read fw binary (Failure 1) nor the symlink workaround (Failure 2) applies here — this is the shipped-firmware case, and it still fails.

    Still no speaker PCM

    aplay -l is unchanged from the report — five devices, none of them speakers:

    device 0: Jack Out       device 5/6/7: HDMI1/2/3       device 31: Deepbuffer Jack Out
    

    The card reports zero amplifiers, and no amp topology is ever requested even though all four are present on disk:

    alsa.components = "HDA:80862822,80860101,00100000 cfg-amp:0 iec61937-pcm:7,6,5 hs:rt712 mic:rt712"
    
    loading topology 0: intel/sof-ipc4-tplg/sof-sdca-jack-id0.tplg
    loading topology 1: intel/sof-ipc4-tplg/sof-sdca-mic-id4.tplg
    loading topology 2: intel/sof-ipc4-tplg/sof-hdmi-pcm5-id5.tplg
    

    Versions, which supersede everything tested in this thread

    Here Newest previously reported
    Kernel 7.2.2 7.1.0-rc6 (Rawhide)
    linux-firmware 20260810 20260519
    sof-firmware 2025.12.2 2025.12.2
    BIOS F.02 F.02

    Cold boot tested — does not help

    #5877 (same board, also CachyOS) ends by offering a full cold boot as the one untested item. I ran it: shutdown -h now, fully powered off ~1 minute, cold start. No change — same three topologies, same cfg-amp:0, no Speaker port.

    It does change the bus behaviour substantially, though. On the cold boot two of the four amps reach Attached, one per link, rather than all four staying UNATTACHED:

    sdw:0:1:0102:0000:01:a   UNATTACHED
    sdw:0:1:0102:0000:01:d   Attached     <-
    sdw:0:2:0102:0000:01:9   UNATTACHED
    sdw:0:2:0102:0000:01:c   Attached     <-
    sdw:0:3:025d:0712:01     Attached
    

    And most of the cascade in Failure 2 is gone. PARITY error, init writes failed, FW download failed: -61 and trf ... write addr 8800 all drop to zero on the cold boot (they do occur on the preceding boot of the same install). Bus clash for control word, ClockStopNow, Clock stop failed -110 and resume: initialization timed out do not appear on 7.2.2 at all, on either boot. Total SoundWire error volume over a matched 90-second window falls from 135 lines to 63.

    So the cascading-failure description still fits older kernels, but on 7.2.2 the noise largely clears and the speakers are exactly as dead.

    What actually fails now

    The failure is symmetric across both links and splits by device number. Slave 1 on each link attaches, then fails in firmware download:

    soundwire_intel soundwire_intel.link.1: Msg NACKed for Slave 1
    soundwire sdw-master-0-1: trf on Slave 1 failed:-5 write addr 8088 count 32632
    slave-tas2783 sdw:0:1:0102:0000:01:d: FW download failed: -5
    slave-tas2783 sdw:0:1:0102:0000:01:d: fw with no files
    

    Slave 2 on each link never initializes:

    slave-tas2783 sdw:0:1:0102:0000:01:a: SDW_SCP_BUS_CLOCK_BASE write failed:-61
    slave-tas2783 sdw:0:1:0102:0000:01:a: Slave 2 initialization failed: -61
    soundwire sdw-master-0-1: trf on Slave 2 failed:-5 write addr 8001 count 0
    slave-tas2783 sdw:0:1:0102:0000:01:a: sw reset failed, err=-5
    

    Link 2 reproduces this line for line with :c as Slave 1 and :9 as Slave 2.

    Note the write address differs from Failure 2 above: addr 8088 here, versus addr 8800 with the generic-symlink firmware. addr 8800 does not appear anywhere on the cold boot.

    fw with no files still shows up for the Slave 1 amps, but it is emitted after the bus transfer fails with -5, so with correct blobs on disk I read it as downstream of the bus error rather than a firmware-content problem. I also can't account for count 32632 when the blob decompresses to 40746 bytes — flagging it as an observation, not a claim.

    What holds across every boot is that no amp ever gets a device number:

    soundwire sdw-master-0-2: Assign dev_num failed:-61
    soundwire sdw-master-0-2: Program device_num 2 failed: -61
    

    Without enumeration there is no SDCA amp function, so the generic fallback path never selects an amp topology. ACPI still reports the amps as part_id 0x0000, so no machine driver matches:

    No SoundWire machine driver found for the ACPI-reported configuration:
      link 1 mfg_id 0x0102 part_id 0x0000 version 0x3   (x2)
      link 2 mfg_id 0x0102 part_id 0x0000 version 0x3   (x2)
      link 3 mfg_id 0x025d part_id 0x0712 version 0x3
    Use SoundWire default machine driver with function topologies
    

    Summary

    With the 8EB4 blobs now shipped in linux-firmware, the remaining gap looks like dev_num assignment / machine-driver match for this board rather than missing calibration data. Given that, is a snd_soc_acpi_mach entry for 8EB4 (4x TAS2783 on links 1-2 plus RT712 on link 3) the right fix, or is something still expected from the firmware side?

    Happy to test patches, run with dyndbg on the SoundWire and SOF modules, or collect acpidump / sysfs dumps / alsa-info. Daily driver, and I can rebuild kernels on it.

    Minor: Driver needs fixing: Unsorted reg_defaults, sorting the copy still fires for all four amps on 7.2.2.

  10. kasemeyer commented on Aug 31, 2026

    @kasemeyer

    Root cause found, fix tested, patches attached. Follow-up to my comment above.

    Two independent problems on 8EB4:

    1. SDCA function type. The four TAS2783 nodes (SSDT35, _ADR 0x0002390102000001 etc.) declare control 0x05 (Function Topology) with mipi-sdca-control-access-mode = 3 and no mipi-sdca-control-dc-value, so find_sdca_function() bails with function type only supported as DisCo constant and no SmartAmp function is ever created. This is the same firmware defect as HP Elitebook X (Panther Lake) - no sound. #5760, where the conclusion was a BIOS fix. The newest BIOS for 8EB4 (F.06, 2026-06-25) still has it. @niranjanhyti — is a BIOS with a DC value for control 0x05 planned for 8EB4 (and the other 8Exx TAS2783 boards: 8EA6 in this thread, 8EA5 in Ubuntu bug 2143870)? Until then I have a minimal fallback (TAS2783 → SmartAmp) as patch 1.

    2. ACTMCTL DOAIS/DOAISE2. This is why the amps drop off the bus even once the function exists. The static DSDT declares intel-sdw-doais = 1, intel-sdw-doaise2 = 1 on every link, but SNDW._INI runs DCFG(LNKn, DAIn, DAEn, DAXn, …) and overwrites them from NVS; the driver actually reads doais = 3, doaise2 = 0. With that timing one amp per link survives enumeration and the rest drop UNATTACHED ~100–250 ms after attaching (the Assign dev_num failed: -61 I reported earlier is downstream of that). Bisecting the ACTMCTL fields live: DOAIS=1 + DOAISE2=1 is necessary and sufficient; DODS/CLDS/CLDE don't matter; DOAISE2 gates DOAIS. Patch 2 is a DMI-keyed override in intel_auxdevice.c for links 1 and 2 on HP/8EB4.

    This took 28 numbered single-change experiments (most a full instrumented reboot each) ending in a live bisect of the ACTMCTL fields; the last five boots reverted everything else to prove the two patches alone suffice — the experiment index is in the repo README below. The work was AI-assisted (Claude Code); both commits carry Assisted-by: trailers per Documentation/process/coding-assistants.rst, and I reviewed and boot-tested it all myself.

    Result on 7.2.2 with stock soundwire-bus and stock snd-soc-tas2783-sdw, only these two patches: all four amps attach and stay attached, firmware loads first try, PARITY 0 at boot and after streaming, hw:0,2 Speaker PCM, and each of the four speakers plays when driven alone. Bus enumerates at the ACPI-default 9.6 MHz; no clock change needed.

    Patches (also as a PR against topic/sof-dev: #5904): https://github.com/kaseyprichard/hp-omnibook-tas2783-sdw/tree/main/upstream — ACPI dumps and the Windows-side comparison are in the same repo.

    Open questions for the maintainers:

    • Is a DMI table in intel_auxdevice.c the right home for an ACTMCTL override, or would you rather see it in dmi-quirks.c or as a machine-driver quirk?
    • Should the SDCA fallback go in at all given HP Elitebook X (Panther Lake) - no sound. #5760, or is this strictly a BIOS matter?

    Also for UCM: the 7.2.2 machine driver names the amp mute controls Left Spk Switch / Right Spk Switch / Left Spk2 Switch / Right Spk2 Switch; the alsa-ucm-conf tas2783 profile with cset "name='Left Spk' …" fails to load the HiFi verb. Happy to send that to alsa-ucm-conf separately.

  11. skriefal commented on Aug 31, 2026

    @skriefal

    and the other 8Exx TAS2783 boards: 8EA6 in this thread, 8EA5 in Ubuntu bug 2143870

    As a general Linux user who hasn't delved into deep kernel or driver behavior since the 1990s, I can comment only at a high level on what I've experienced. On the 8EA6 board mentioned above I now have the internal amp/speakers working. This started working after the F.07 BIOS update from HP, while using kernel 7.1.x. I did not re-test with pre-7.1.x kernels and I've since wiped that Cachyos install.

    I'm currently using Fedora 44 on the same device and again have working amp/speakers with kernel 7.1.10. I did need to add symlinks for the 8EA6 firmware files to allow them to be found and loaded, but otherwise changed nothing from the stock Fedora 44 install (+updates) to get this working.

    I don't know how to check SDCA and the other things that you've mentioned. But it does seem that the BIOS update resolved the issue for my device, other than the minor annoyance of adding the firmware symlinks.

  12. PatrickH1999 commented on Sep 2, 2026

    @PatrickH1999

    Same issue on an HP OmniBook X Laptop 14-ka0xxx (Intel Panther Lake), DMI board name 8EA6.

    The TAS2783 driver fails to find board-specific firmware, and the fallback also fails — no Speaker device is created. Internal speakers produce no sound; Bluetooth audio works.

    slave-tas2783 sdw:0:2:0102:0000:01:9: Direct firmware load for 8EA6-2-0x9.bin failed with error -2
    slave-tas2783 sdw:0:2:0102:0000:01:9: Failed to read preferred fw binary: 8EA6-2-0x9.bin, attempting fallback binary load
    slave-tas2783 sdw:0:2:0102:0000:01:c: Direct firmware load for 8EA6-2-0xC.bin failed with error -2
    slave-tas2783 sdw:0:2:0102:0000:01:c: Failed to read preferred fw binary: 8EA6-2-0xC.bin, attempting fallback binary load
    

    System: Arch Linux, kernel 7.2.2-arch1-1, linux-firmware 20260810-2

    Could firmware for board ID 8EA6 please be included in a future linux-firmware TAS2783 submission?

  13. CrazyCows commented on Oct 4, 2026

    @CrazyCows

    Another data point for this issue: HP OmniBook Ultra 14 (board 8EB4), same symptoms as the OP. I can confirm the two patches from PR #5904 fix amplifier enumeration and restore the Speaker PCM on my machine — full details below. Testing was done with AI assistance (Codex), but all results are from real hardware runs on my unit.

    HP OmniBook Ultra 14 / 8EB4: PR #5904 reproduced on CachyOS 7.2.9

    The two patches from #5904
    restore amplifier enumeration and Speaker PCM on this machine, with the
    stock TAS2783 codec and stock firmware. Adding the missing upstream UCM
    profile then restores the desktop Speaker sink. Only out-of-tree audio
    modules were built; no kernel build was needed.

    System

    HP OmniBook Ultra Laptop 14-kd0xxx, board 8EB4, BIOS F.06 (2026-06-25)
    CachyOS, running kernel 7.2.9-1-cachyos
    linux-firmware / linux-firmware-other 1:20260916-1
    sof-firmware 2026.09.1-1; DSP firmware ADSPFW 2.15.0.1
    alsa-ucm-conf 1.2.16.1-1; ALSA utils 1.2.16-1.1
    PipeWire 1.6.9; WirePlumber 0.5.18
    Matching headers and base-devel were already installed.

    Original boot failure and firmware mapping

    The four board-specific firmware files are present and decompress to 40,746
    bytes each. This is not the old generic-firmware symlink workaround.

    First observed failure:

    soundwire sdw-master-0-2: trf on Slave 1 failed:-5 write addr 8088 count 32632
    slave-tas2783 sdw:0:2:0102:0000:01:c: FW download failed: -5
    slave-tas2783 sdw:0:2:0102:0000:01:c: fw with no files

    The C amplifier's requested file is 8EB4-2-0xC.bin, resolved to the shipped
    ti/audio/tas2783/8EB4-2-0xC.bin.zst. An instrumented module confirmed the
    filename on subsequent reloads. Parsing the local binary maps the original
    boot failure to zero-based block 0/file_id 0, length 32632, destination
    0x00c60088 (PRAM), exclusive end 0x00c68000. The bus reports the truncated
    paged address 0x8088, not the full codec address. The logged count is the
    transfer length; it does not identify an individual byte that NACKed.

    There is also a later boot failure on that same amp at addr 8108/count 6904.
    This maps to block 1/file_id 1, destination 0x00800108 (YRAM), exclusive end
    0x00801c00. Thus failures also occur below the proposed 0xc60000 skip cutoff.
    The mapping uses boot logs + decoded firmware + matching driver source;
    the original failing boot was not rerun with the instrumented module.

    The stock boot also reports four instances of:
    function type only supported as DisCo constant

    Before the platform fix, :c and :d are Attached, while :9 and :a are
    UNATTACHED. aplay lists Jack Out, Deepbuffer Jack Out, and HDMI only.

    Controlled experiments

    Each row uses a controller/module reload and its own saved logs. No cold
    power cycle was performed. Changes are tested individually against the
    logging baseline, with stock restored between codec tests.

    Test Result
    Logging-only codec :c/:d write all 52 blocks; :9/:a unattached; no Speaker.
    Skip dest >= 0xc60000 PRAM skipped on :c/:d; remaining writes complete;
    still 2/4 amps, no Speaker.
    2 ms delay per block Writes complete on :c/:d; still 2/4, no Speaker.
    One retry after 2 ms No write error occurred, so retry was not exercised;
    still 2/4, no Speaker.
    Status-register reads No write failure reproduced; snapshots below;
    still 2/4, no Speaker.
    PR #5904 timing only All 4 amps Attached, all 52 blocks per amp reached;
    zero FW/init/parity errors, but no Speaker PCM.
    PR #5904 timing + SDCA All 4 amps Attached; Speaker hw:0,2 and Deepbuffer
    Speaker hw:0,35; zero FW/init/parity errors.
    Add upstream ALSA UCM HiFi profile, default Speaker sink, 5 sinks/2 sources.

    Important limits: the warm logging-only baseline already downloads firmware
    successfully to the two attached amps. The delay/retry tests therefore do
    not demonstrate a cure for the original boot NACK. No after-failure status
    snapshot was captured, because no write failed during the status-read test.

    Before PRAM and YRAM writes on both reachable amps, direct uncached reads:
    SCP_STAT (0x44) = 0x20
    SCP_DEVNUMBER (0x46) = 1
    TAS register 0x800007 = 0x20
    4-byte read at 0x80103c succeeds: e0 00 20 00
    No interrupt latch was read/cleared, and the snapshot code writes no registers.

    The unconditional skip experiment is NOT the complete TI v1 patch. TI v1
    uses !first_hw_init || dest_addr < 0xc60000, so it still downloads PRAM on
    the first initialization. That distinction matters for first-boot failures.

    Working platform fix

    The timing patch logs exactly:
    sdw-master-0-1: ACTMCTL quirk: doais 0x3 -> 0x1, doaise2 0x0 -> 0x1
    sdw-master-0-2: ACTMCTL quirk: doais 0x3 -> 0x1, doaise2 0x0 -> 0x1

    The SDCA patch then creates all four SmartAmp functions. The card components
    include cfg-amp:4 and spk:tas2783. The two changes address separate failures:
    timing restores communication/enumeration, and function discovery permits
    the speaker topology/PCM to be created.

    Exact sources: CachyOS/linux tag cachyos-7.2.9-1. The PR patch was applied
    without semantic changes. Kbuild used installed headers/Module.symvers,
    LLVM=1 and Clang 23.1.1. Only soundwire-intel and snd-soc-sdca remain patched;
    snd-soc-tas2783-sdw is back to its original packaged module.

    UCM follow-up

    The installed alsa-ucm-conf lacked sof-soundwire/tas2783.conf, causing
    snd_use_case_mgr_open to fail with -2 after Speaker PCM became available.
    The current upstream file uses Left/Right Spk [2] Switch controls and
    PlaybackPCM hw:${CardId},2. Staging that one file made alsaucm import HiFi;
    installing it and restarting PipeWire/WirePlumber produced the Speaker sink.
    Source commit: 182d0122a7d98cfa5f2d42d3cb8e9e1b8697b69a, alsa-project/alsa-ucm-conf.

    Validation and limitations

    Stereo playback through the PipeWire Speaker sink completed with exit 0.
    Hearing confirmation is recorded in the appended listening result below.
    No new audio kernel error was observed during that playback.
    All three stock module files and backups still match their original SHA-256.
    The original kernel, LTS kernel, firmware and initramfs were not modified.
    Module taint is 12288 (out-of-tree/unsigned module flags); no kernel fault
    was observed. This test does not validate cold boot or system suspend/resume.
    The installed overrides are specific to 7.2.9-1-cachyos and require rebuilding
    for a different kernel. Nothing has been posted automatically to GitHub.

    References

    #5732
    #5904
    https://bugzilla.kernel.org/show_bug.cgi?id=221782
    https://lkml.iu.edu/hypermail/linux/kernel/2608.1/07099.html
    alsa-project/alsa-ucm-conf@182d012

    Listening result

    The user confirmed both left and right test tones were audible through the
    laptop speakers on replay. Both pw-play invocations returned 0. Individual
    amplifiers were not acoustically isolated in this session.

  14. DieEneSchrodinger commented on Oct 15, 2026

    @DieEneSchrodinger

    HP released BIOS version F.07 for the HP OmniBook Ultra 8EB4. After updating to this version, the issue still persists (on kernel 7.2.10) with the same dmesg errors as on F.06.

  15. swheel33 commented on Nov 4, 2026

    @swheel33

    Data point for 8EB4 on a newer stack, plus two fixes for problems that showed up once the amps attached.

    Setup: HP OmniBook Ultra 14-kd0xxx (8EB4), BIOS F.06, Omarchy's linux-omarchy 7.2.15 (7.2.15 plus a large set of sound/ASoC/SoundWire backports), linux-firmware 20260916, sof-firmware 2026.09.2. Applied #5904's two patches plus Omarchy's former 8EB4 stereo map (omarchy-pkgs#1214) and built them as module overrides.

    kasemeyer's patches work on 7.2.15: ACTMCTL quirk fires on links 1 and 2, the SmartAmp fallback is taken for all four amps, all five peripherals attach, and all four firmware files download (8EB4-{1-0xA,1-0xD,2-0x9,2-0xC}.bin, 52 files each). With the 8EB4 channel map, speaker-test -s 1 and -s 2 play from the correct sides.

    Problem 1: race between the last amp's firmware download and the first hw_params. On the first boot, amp 01:c attached ~2 s after the others. PipeWire opened the Speaker PCM ~1.2 s later, before that amp's download had finished. tas_sdw_hw_params() returns -EINVAL straight away when !fw_dl_success:

    [100.876] slave-tas2783 sdw:0:2:0102:0000:01:c: error playback without fw download
    [100.876] ASoC error (-22): at snd_soc_dai_hw_params() on tas2783-codec
       (x7 within 10 ms, each followed by)
    sdw_deprepare_stream: subdevice #0-Playback: inconsistent state state 1
    

    Problem 2: that inconsistent stream then oopsed the kernel about 40 s later, in pipewire's context:

    Oops: divide error: 0000 [#1] SMP NOPTI
    RIP: 0010:sdw_compute_params+0xe83/0x159f [soundwire_generic_allocation]
     sdw_prepare_stream+0x194/0x870 [soundwire_bus]
     snd_soc_link_prepare+0x30/0x60 [snd_soc_core]
     dpcm_be_dai_prepare+0x22e/0x5a0 [snd_soc_core]
     dpcm_fe_dai_prepare+0x89/0x270 [snd_soc_core]
     snd_pcm_do_prepare+0x40/0x190 [snd_pcm]
     snd_pcm_prepare+0x508/0x690 [snd_pcm]
    

    The faulting instruction is the DIV_ROUND_UP by params[i].full_bw in sdw_compute_group_params(); full_bw = curr_dr_freq / rate was 0.

    Fixes (both attached below):

    • ASoC: tas2783-sdw: wait for an in-flight firmware download in hw_params: wait_event_timeout(fw_wait, fw_dl_success, 2*TIMEOUT_FW_DL_MS) before giving up, rate-limit the error, and log each successful download.
    • soundwire: generic_bandwidth_allocation: reject a rate the bus cannot carry: return -EINVAL when !rate || curr_dr_freq < rate instead of dividing by zero.

    With both applied I've had clean boots with no errors and no oops. To be clear about what's tested: on those boots all four downloads happened to finish before PipeWire opened the PCM (11.0–12.2 s versus 13.0 s), so the wait path itself hasn't been exercised yet. The guard and the wait are both defensive, and the race is real, as the first boot showed.

    Happy to send these to alsa-devel properly if they look right.

    0004-ASoC-tas2783-sdw-wait-for-an-in-flight-firmware-down.patch
    diff --git a/sound/soc/codecs/tas2783-sdw.c b/sound/soc/codecs/tas2783-sdw.c
    index e06e414..b168019 100644
    --- a/sound/soc/codecs/tas2783-sdw.c
    +++ b/sound/soc/codecs/tas2783-sdw.c
    @@ -855,6 +855,9 @@ static void tas2783_fw_ready(const struct firmware *fmw, void *context)
     		ret = -EINVAL;
     	} else {
     		tas2783_update_calibdata(tas_dev);
    +		if (!ret)
    +			dev_info(tas_dev->dev, "firmware %s downloaded (%d files)\n",
    +				 tas_dev->rca_binaryname, cur_file);
     	}
     
     out:
    @@ -1075,8 +1078,18 @@ static s32 tas_sdw_hw_params(struct snd_pcm_substream *substream,
     	struct sdw_slave *sdw_peripheral = tas_dev->sdw_peripheral;
     	s32 ret, retry = 3;
     
    -	if (!tas_dev->fw_dl_success) {
    -		dev_err(tas_dev->dev, "error playback without fw download");
    +	/*
    +	 * The amps attach and download their firmware one after another; on
    +	 * the HP OmniBook Ultra 14 (8EB4) the last of four finishes a couple of
    +	 * seconds after the card is registered, and PipeWire opens the speaker
    +	 * PCM before that. Failing hw_params here makes userspace retry in a
    +	 * tight loop, which leaves the SoundWire stream half torn down. Give an
    +	 * in-flight download time to finish instead.
    +	 */
    +	if (!tas_dev->fw_dl_success &&
    +	    !wait_event_timeout(tas_dev->fw_wait, tas_dev->fw_dl_success,
    +				msecs_to_jiffies(2 * TIMEOUT_FW_DL_MS))) {
    +		dev_err_ratelimited(tas_dev->dev, "error playback without fw download");
     		return -EINVAL;
     	}
     
    -- 
    2.56.0
    0005-soundwire-generic_bandwidth_allocation-reject-a-rate.patch
    diff --git a/drivers/soundwire/generic_bandwidth_allocation.c b/drivers/soundwire/generic_bandwidth_allocation.c
    index 30a495a..0acbf53 100644
    --- a/drivers/soundwire/generic_bandwidth_allocation.c
    +++ b/drivers/soundwire/generic_bandwidth_allocation.c
    @@ -236,6 +236,17 @@ static int sdw_compute_group_params(struct sdw_bus *bus,
     	for (i = 0; i < group->count; i++) {
     		params[i].rate = group->rates[i];
     		params[i].lane = group->lanes[i];
    +		/*
    +		 * A stream left half-configured by repeated hw_params failures
    +		 * can reach here with no usable bus frequency; full_bw is the
    +		 * divisor below, so refuse instead of dividing by zero.
    +		 */
    +		if (!params[i].rate ||
    +		    bus->params.curr_dr_freq < params[i].rate) {
    +			dev_err(bus->dev, "invalid rate %u for bus frequency %u\n",
    +				params[i].rate, bus->params.curr_dr_freq);
    +			return -EINVAL;
    +		}
     		params[i].full_bw = bus->params.curr_dr_freq / params[i].rate;
     	}
     
    -- 
    2.56.0
  16. swheel33 commented on Nov 6, 2026

    @swheel33

    Follow-up for 8EB4: a second problem, separate from the hw_params race. It shows up on s2idle resume.

    Symptom: after opening the lid, keyboard and touchpad input was lost for ~10 s and the lock screen looked hung. The compositor stayed responsive, but userspace was still frozen because the kernel hadn't finished resuming.

    Cause: the four amps lose power in s2idle and re-attach on resume. tas_update_status() resets each one and downloads its firmware inline (tas_io_init() → tas_fw_load() waits for the download to finish), about 2.5 s per amp. The bus handles peripherals one at a time and only signals initialization_complete after update_status returns, and each amp's resume waits on that. So the whole system resume waits for all four downloads, one after another:

    357.564 ACPI: EC: interrupt unblocked
    357.719 nvme nvme0: 16/0/0 default/read/poll queues
    360.488 slave-tas2783 sdw:0:1:0102:0000:01:d: firmware 8EB4-1-0xD.bin downloaded (52 files)
    362.938 slave-tas2783 sdw:0:1:0102:0000:01:a: firmware 8EB4-1-0xA.bin downloaded (52 files)
    365.470 slave-tas2783 sdw:0:2:0102:0000:01:c: firmware 8EB4-2-0xC.bin downloaded (52 files)
    368.010 slave-tas2783 sdw:0:2:0102:0000:01:9: firmware 8EB4-2-0x9.bin downloaded (52 files)
    368.126 Restarting tasks: Starting
    

    (kernel timestamps; the -110 failed to resume errors some people see on these amps after resume come from the same wait.)

    Fix (attached): run the reset and download from a work item instead of inline in update_status:

    • The work runs without holding sdw_dev_lock, so a multi-second download can't block the bus from delivering that peripheral's status changes.
    • fw_dl_success is cleared when the amp detaches or re-attaches uninitialized, so playback can't use an amp whose firmware is gone.
    • hw_params waits up to 15 s for the deferred init to complete: init writes done, not just the download. This replaces the 2× timeout in my earlier 0004.
    • The work is cancelled on system suspend and on remove.
    • The download wait in tas_fw_load() goes from 3 s to 15 s. After resume the firmware loader can't read the file until userspace is thawed, and then all four amps download concurrently (~4.5 s total). My first version kept the 3 s wait, gave up before the downloads finished, and left the amps without their init writes.

    Result on linux-omarchy 7.2.15:

     97.409 ACPI: EC: interrupt unblocked
     97.568 nvme nvme0: 16/0/0 default/read/poll queues
     98.027 Restarting tasks: Done
    102.573 slave-tas2783 sdw:0:1:0102:0000:01:d: firmware 8EB4-1-0xD.bin downloaded (52 files)
    102.577 slave-tas2783 sdw:0:2:0102:0000:01:c: firmware 8EB4-2-0xC.bin downloaded (52 files)
    

    Resume goes from ~10.5 s to ~0.6 s; the slowest device left is xe at ~0.4 s. The speakers come back ~4.5 s after wake, with no errors, and playback works after resume. What's tested: so far one boot and one resume on the final version, so treat it as early. The concurrent download over a shared link worked here, but it's worth a look from someone who knows the SoundWire side, in case the amps should be serialized instead.

    0006-ASoC-tas2783-sdw-defer-amp-reset-and-firmware-download.patch
    diff --git a/sound/soc/codecs/tas2783-sdw.c b/sound/soc/codecs/tas2783-sdw.c
    index b168019..7797c9a 100644
    --- a/sound/soc/codecs/tas2783-sdw.c
    +++ b/sound/soc/codecs/tas2783-sdw.c
    @@ -26,6 +26,7 @@
     #include <linux/pm_runtime.h>
     #include <linux/regmap.h>
     #include <linux/wait.h>
    +#include <linux/workqueue.h>
     #include <linux/soundwire/sdw.h>
     #include <linux/soundwire/sdw_registers.h>
     #include <linux/soundwire/sdw_type.h>
    @@ -106,6 +107,8 @@ struct tas2783_prv {
     	bool fw_use_fallback;
     	/* HP 8EB4 channel map logged once */
     	bool hp8eb4_logged;
    +	/* reset + firmware download, deferred out of update_status */
    +	struct work_struct init_work;
     };
     
     static const struct reg_default tas2783_reg_default[] = {
    @@ -1086,9 +1089,10 @@ static s32 tas_sdw_hw_params(struct snd_pcm_substream *substream,
     	 * tight loop, which leaves the SoundWire stream half torn down. Give an
     	 * in-flight download time to finish instead.
     	 */
    -	if (!tas_dev->fw_dl_success &&
    -	    !wait_event_timeout(tas_dev->fw_wait, tas_dev->fw_dl_success,
    -				msecs_to_jiffies(2 * TIMEOUT_FW_DL_MS))) {
    +	if (!(tas_dev->hw_init && tas_dev->fw_dl_success) &&
    +	    !wait_event_timeout(tas_dev->fw_wait,
    +				tas_dev->hw_init && tas_dev->fw_dl_success,
    +				msecs_to_jiffies(5 * TIMEOUT_FW_DL_MS))) {
     		dev_err_ratelimited(tas_dev->dev, "error playback without fw download");
     		return -EINVAL;
     	}
    @@ -1284,6 +1288,10 @@ static s32 tas2783_sdca_dev_suspend(struct device *dev)
     
     static s32 tas2783_sdca_dev_system_suspend(struct device *dev)
     {
    +	struct tas2783_prv *tas_dev = dev_get_drvdata(dev);
    +
    +	/* Do not suspend with a firmware download still in flight. */
    +	cancel_work_sync(&tas_dev->init_work);
     	return tas2783_sdca_dev_suspend(dev);
     }
     
    @@ -1362,8 +1370,13 @@ static s32 tas_fw_load(struct tas2783_prv *tas_dev, struct sdw_slave *slave)
     		return ret;
     	}
     
    +	/*
    +	 * After a system resume the firmware loader cannot read the file until
    +	 * userspace is thawed, and the amps then download concurrently over a
    +	 * shared link, so allow for all of them rather than one.
    +	 */
     	ret = wait_event_timeout(tas_dev->fw_wait, tas_dev->fw_dl_task_done,
    -				 msecs_to_jiffies(TIMEOUT_FW_DL_MS));
    +				 msecs_to_jiffies(5 * TIMEOUT_FW_DL_MS));
     	if (!ret) {
     		dev_err(tas_dev->dev, "fw request, wait_event timeout\n");
     		return -EAGAIN;
    @@ -1412,6 +1425,50 @@ static s32 tas_io_init(struct device *dev, struct sdw_slave *slave)
     	return ret;
     }
     
    +/*
    + * Resetting the amp and downloading its firmware takes ~2.5 s per amp. The
    + * bus handles peripheral status one device after another and system resume
    + * waits for every peripheral to finish initializing, so doing this inline in
    + * update_status held the whole system in resume (userspace frozen) for ~10 s
    + * on the HP OmniBook Ultra 14 (8EB4), which has four amps. Do it from a work
    + * item instead; hw_params waits for it to complete.
    + */
    +static void tas_init_work(struct work_struct *work)
    +{
    +	struct tas2783_prv *tas_dev =
    +		container_of(work, struct tas2783_prv, init_work);
    +	struct sdw_slave *slave = tas_dev->sdw_peripheral;
    +	struct device *dev = &slave->dev;
    +	bool rpm;
    +	s32 ret;
    +
    +	/* -EACCES: runtime PM disabled during a system transition, still active */
    +	ret = pm_runtime_resume_and_get(dev);
    +	if (ret < 0 && ret != -EACCES) {
    +		dev_err(dev, "init work: runtime resume failed: %d\n", ret);
    +		return;
    +	}
    +	rpm = ret >= 0;
    +
    +	/*
    +	 * Not under sdw_dev_lock: the download takes seconds and would block the
    +	 * bus from delivering this peripheral's status changes meanwhile. remove
    +	 * cancels this work before tearing anything down.
    +	 */
    +	if (!tas_dev->hw_init && tas_dev->status == SDW_SLAVE_ATTACHED) {
    +		ret = tas_io_init(dev, slave);
    +		if (ret)
    +			dev_err(dev, "deferred init failed: %d\n", ret);
    +	}
    +
    +	wake_up(&tas_dev->fw_wait);
    +
    +	if (rpm) {
    +		pm_runtime_mark_last_busy(dev);
    +		pm_runtime_put_autosuspend(dev);
    +	}
    +}
    +
     static s32 tas_update_status(struct sdw_slave *slave,
     			     enum sdw_slave_status status)
     {
    @@ -1423,8 +1480,10 @@ static s32 tas_update_status(struct sdw_slave *slave,
     		 status == SDW_SLAVE_ATTACHED ? "attached" : "alert");
     
     	tas_dev->status = status;
    -	if (status == SDW_SLAVE_UNATTACHED)
    +	if (status == SDW_SLAVE_UNATTACHED) {
     		tas_dev->hw_init = false;
    +		tas_dev->fw_dl_success = false;
    +	}
     
     	/* Perform initialization only if slave status
     	 * is present and hw_init flag is false
    @@ -1450,8 +1509,15 @@ static s32 tas_update_status(struct sdw_slave *slave,
     	 */
     	regcache_drop_region(tas_dev->regmap, 0, UINT_MAX);
     
    +	/*
    +	 * The amp lost its firmware; keep hw_params from using it before the
    +	 * deferred init has downloaded it again.
    +	 */
    +	tas_dev->fw_dl_success = false;
    +
     	/* perform I/O transfers required for Slave initialization */
    -	return tas_io_init(&slave->dev, slave);
    +	queue_work(system_long_wq, &tas_dev->init_work);
    +	return 0;
     }
     
     /*
    @@ -1585,6 +1651,7 @@ static s32 tas_sdw_probe(struct sdw_slave *peripheral,
     	mutex_init(&tas_dev->pde_lock);
     
     	init_waitqueue_head(&tas_dev->fw_wait);
    +	INIT_WORK(&tas_dev->init_work, tas_init_work);
     	dev_set_drvdata(dev, tas_dev);
     	regmap = devm_regmap_init_sdw_mbq_cfg(&peripheral->dev,
     					      peripheral,
    @@ -1604,6 +1671,7 @@ static void tas_sdw_remove(struct sdw_slave *peripheral)
     {
     	struct tas2783_prv *tas_dev = dev_get_drvdata(&peripheral->dev);
     
    +	cancel_work_sync(&tas_dev->init_work);
     	pm_runtime_disable(tas_dev->dev);
     	tas_remove(tas_dev);
     	mutex_destroy(&tas_dev->calib_lock);
    -- 
    2.56.0
    
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions