Skip to content

soundwire/SDCA: enable the four TAS2783 speakers on HP OmniBook Ultra 14 (8EB4) - #5904

Open
kasemeyer wants to merge 2 commits into
thesofproject:topic/sof-devfrom
kasemeyer:hp-omnibook-8eb4-tas2783
Open

kasemeyer wants to merge 2 commits into
thesofproject:topic/sof-devfrom
kasemeyer:hp-omnibook-8eb4-tas2783

Conversation

@kasemeyer

Copy link
Copy Markdown

Fixes the silent internal speakers on the HP OmniBook Ultra 14 (board 8EB4, Panther Lake, 4x TI TAS2783 on SoundWire links 1/2 + RT712 on link 3). Tracked in #5732 (also #5877, #5802); the SDCA half is the same defect as #5760.

Two independent problems, one patch each — see the commit messages for details:

  1. ASoC: SDCA: fall back to SmartAmp function type for TI TAS2783 — the BIOS declares the Function Topology control with access mode 3 and no DC value; upstream deliberately requires a DC value (HP Elitebook X (Panther Lake) - no sound. #5760) and HP fixed the EliteBook X via BIOS, but 8EB4 F.06 (latest) still lacks it. Minimal known-peripheral fallback; I'm fine dropping it if HP commits to a BIOS fix.
  2. soundwire: intel: override ACTMCTL DOAIS/DOAISE2 on HP OmniBook Ultra 14 — SNDW._INI overwrites the static intel-sdw-doais/doaise2 = 1 with NVS values 3/0; with those, three of four amps drop off the bus after attaching. Runtime bisect: DOAIS=1 + DOAISE2=1 is necessary and sufficient. DMI-keyed override for links 1 and 2.

Tested on 7.2.2 with stock soundwire-bus and snd-soc-tas2783-sdw, only these two changes: 4/4 amps attach and stay attached, firmware first try, PARITY 0 at boot and after streaming, Speaker PCM present, all four speakers verified individually. checkpatch --strict clean.

How this was found: 28 numbered single-change experiments on the machine (most a full instrumented reboot with a captured boot log), finishing with a live runtime bisect of the ACTMCTL fields; experiments 0024–0028 then reverted everything else to prove these two patches alone suffice on pristine v7.2.2 files. The work was AI-assisted (Claude Code); both commits carry Assisted-by: trailers per Documentation/process/coding-assistants.rst, and I have reviewed and boot-tested everything here myself. The experiment index is in the repo README linked below.

Questions: is intel_auxdevice.c the right home for an ACTMCTL override (vs dmi-quirks.c or a machine-driver quirk)? And should the SDCA fallback go in at all, or is 8EB4 strictly a BIOS matter?

Supporting material (ACPI dumps, Windows-side comparison, UCM files): https://github.com/kaseyprichard/hp-omnibook-tas2783-sdw

On the HP OmniBook Ultra 14 (board 8EB4, Panther Lake) the four TI
TAS2783 SmartAmp peripherals declare the SDCA Function Topology control
(0x05) with access mode 0x03 and no mipi-sdca-control-dc-value, so
find_sdca_function() rejects them ("function type only supported as
DisCo constant"). No SmartAmp function is created, no speaker DAI
exists and the internal speakers are silent. The RT712 on link 3 does
provide a DC value and is unaffected.

The same firmware defect was reported for the HP EliteBook X in
thesofproject#5760, where the conclusion was that the BIOS should
supply a DC value, and HP fixed that machine with a BIOS update. The
newest BIOS for 8EB4 (F.06, 2026-06-25) still ships access mode 0x03
with no DC value for the amplifiers, so machines in the field need a
fallback. Keep it minimal: a table of known peripherals (TAS2783,
mfg 0x0102 part 0x0000, is always a SmartAmp) rather than reading the
control from hardware, so functions can still be parsed before the
device is available. Happy to drop this if HP commits to a BIOS fix
for 8EB4.

Link: thesofproject#5732
Link: thesofproject#5760
Assisted-by: Claude:claude-fable-5
Signed-off-by: Kasey Prichard <kasey.prichard@gmail.com>
The HP OmniBook Ultra 14 (board 8EB4, Panther Lake, BIOS F.06) has four
TI TAS2783 SmartAmp peripherals, two each on SoundWire links 1 and 2.
The static DSDT declares intel-sdw-doais = 1 and intel-sdw-doaise2 = 1
for every link, but SNDW._INI then runs DCFG() on each LNKn package and
overwrites those entries from NVS bytes (DAIn, DAXn, ...), and the
values the driver actually reads are doais = 3, doaise2 = 0. With that
ACTMCTL timing only one of the four amplifiers survives enumeration:
the other three attach, answer DEVID reads, then drop off the bus
within a few hundred ms and never come back, so there is no speaker
output on Linux. Windows drives all four amplifiers on the same
firmware.

Bisecting the ACTMCTL fields at runtime shows that DOAIS = 1 together
with DOAISE2 = 1 is both necessary and sufficient: with that pair all
four amplifiers attach, download firmware on the first attempt and stay
attached with no parity errors, and all four speakers play. DODS, CLDS
and CLDE make no difference. DOAISE2 gates whether DOAIS is honoured,
so neither field alone helps.

Add a DMI-keyed override for the ACTMCTL fields, applied after the
_DSD properties are read, and use it to force DOAIS = 1 / DOAISE2 = 1
on links 1 and 2 of this board. Other links and other boards are
untouched.

Link: thesofproject#5732
Assisted-by: Claude:claude-fable-5
Signed-off-by: Kasey Prichard <kasey.prichard@gmail.com>
@sofci

sofci commented Aug 31, 2026

Copy link
Copy Markdown
Collaborator

Can one of the admins verify this patch?

reply test this please to run this test once

@bardliao

Copy link
Copy Markdown
Collaborator

I am pretty sure this will be fixed in the BIOS. @niranjanhyti @jim-monkey FYI

@niranjanhyti

Copy link
Copy Markdown
Collaborator

@jimxbj-dot (@jim-monkey) now supporting tas2783a - Jim please comment.

@CrazyCows

Copy link
Copy Markdown

Tested on HP OmniBook Ultra 14 / 8EB4, fixes enumeration and Speaker PCM, details in #5732

@kasemeyer

Copy link
Copy Markdown
Author

Tested on HP OmniBook Ultra 14 / 8EB4, fixes enumeration and Speaker PCM, details in #5732

I spent days on that fix. Happy someone is getting a benefit from it!

@lgirdwood

Copy link
Copy Markdown
Member

@bardliao any update on BIOS fix, if not then probably best to upstream as a quirk ?

@bardliao

Copy link
Copy Markdown
Collaborator

@bardliao any update on BIOS fix, if not then probably best to upstream as a quirk ?

The BIOS fix is already shared with HP. Not sure when will be the official release. @niranjanhyti @jimxbj-dot (@jim-monkey) any comment?

@leepaulmann

Copy link
Copy Markdown

Data point from a second vendor, with a caveat for 7.3.

Machine: ASUS ProArt PX13 HN7306EA (BIOS HN7306EA.310, AMD ACP), 2x TAS2783 on SoundWire link 1 (sdw:0:1:0102:0000:01:8 and :01:b). Same firmware defect: on a stock kernel both amps log function type only supported as DisCo constant.

Tested: the sdca_functions.c patch from this PR (only that hunk) on 7.3-rc6, built as an out-of-tree snd-soc-sdca. The parse works: SDCA function SmartAmp (type 1) at 0x1 for both amps, tas2783 reports Found Smart Amp function at index 0, and the ACPI init table is applied. But with the stock 7.3-rc6 tas2783 driver, the right speaker goes silent.

Checked by ear with speaker-test -c2 -t wav:

snd-soc-sdca tas2783 result
stock 7.3-rc6 stock 7.3-rc6 stereo, correct sides, also after s2idle
this PR stock 7.3-rc6 Front Left from the left speaker, Front Right silent; ~1600 regmap-sdw-mbq: Defer on undeferrable control warnings during playback (none without the patch)
this PR out-of-tree tas2783 that keeps the full playback ch_mask and selects the channel via a cluster-index control stereo

Cause, as far as I can tell: the PX13's ACPI init table sets the UDMPU ClusterIndex (SDW_SDCA_CTL(1, PPU21, 0x10, 0) = 0x40480800) to 0x01 on amp 0x8 and 0x04 on amp 0xb. Read back from the regmap they hold exactly that. Since 7.3, tas2783_sdw_hw_params() narrows each amp's playback ch_mask to a single channel by name prefix (the "mirror mode" comment). With tas2783_init_seq leaving ClusterIndex 0x01 on both amps, that gives stereo. With the ACPI table applied instead, amp 0xb receives a one-channel payload but is told to render Right (0x04), so it is silent.

For 8EB4: your test was on 7.2.2. As far as I can see, the ch_mask change is not in 7.2.x (I checked 7.2.9) and is in 7.3-rc6. So it may be worth re-testing 8EB4 on 7.3, depending on what its init table writes to ClusterIndex for each of the four amps.

So on the PX13, the fallback as posted would turn working stereo into left-only on a stock 7.3 kernel. If it goes in as a quirk, limiting it to the boards it was tested on (DMI, like the ACTMCTL override) would avoid that. Alternatively, the ch_mask narrowing and the ACPI ClusterIndex need to agree when sa_func_data is present. I'm happy to test any revision on the PX13.

Disclosure: I worked through this with AI assistance (Claude Code). The results above are from my machine, checked by ear and in the kernel log.

@jonnyjonjon

jonnyjonjon commented Nov 7, 2026 •

Copy link
Copy Markdown

Issue persists for board 8EB4 with BIOS F.07 (early October) installed, so HP has not fixed yet.
Using Fedora Rawhide and Linux Kernel 7.4.0-RC1.

@jonnyjonjon

Copy link
Copy Markdown

@bardliao @niranjanhyti @jimxbj-dot (@jim-monkey) since the F.07 BIOS was released without resolving the issue, is there any way to follow up with HP? unclear where bugs are reported to them or tracked.

@bardliao

bardliao commented Nov 9, 2026

Copy link
Copy Markdown
Collaborator

@bardliao @niranjanhyti @jimxbj-dot (@jim-monkey) since the F.07 BIOS was released without resolving the issue, is there any way to follow up with HP? unclear where bugs are reported to them or tracked.

@jonnyjonjon What laptop module do you use? The HP BIOS F.11 fixes this on the OmniBook X Flip 14 (8EA4). #5981 (comment)

@jonnyjonjon

jonnyjonjon commented Nov 9, 2026 •

Copy link
Copy Markdown

@bardliao @niranjanhyti @jimxbj-dot (@jim-monkey) since the F.07 BIOS was released without resolving the issue, is there any way to follow up with HP? unclear where bugs are reported to them or tracked.

@jonnyjonjon What laptop module do you use? The HP BIOS F.11 fixes this on the OmniBook X Flip 14 (8EA4). #5981 (comment)

Omnibook Ultra 14 (board 8EB4, Panther Lake, 4x TI TAS2783 amps), same as the original submitter of this PR. F.07 is the latest BIOS. Was a bug submitted to HP for this model as well?

@leepaulmann

Copy link
Copy Markdown

Update to my comment above: on 7.4 my objection no longer applies, and this fallback is now what the PX13 needs.

7.4-rc1 includes f42fbd5 ("ASoC: tas2783-sdw: add firmware download status check"). It removes the 7.3 stereo split from tas2783_sdw_hw_params() and leaves channel selection to the DisCo init table. On the PX13 (ASUS ProArt HN7306EA, 2x TAS2783 on SoundWire link 1), that table is never applied without this PR, so on 7.4 the right speaker is silent.

Tested on 7.3.3 by swapping modules, checked by ear with speaker-test -c2 -t wav:

snd-soc-sdca tas2783 result
stock v7.4-rc1 Front Left from left speaker, Front Right not heard, right speaker silent
this PR (sdca_functions.c hunk only) v7.4-rc1 stereo, correct sides; no Defer on undeferrable control warnings

With this PR the log shows function type not a DisCo constant, using fallback for known peripheral and Found Smart Amp function at index 0 for both amps.

Good to see HP F.11 fixes 8EA4. The PX13 has the same firmware defect (function type only supported as DisCo constant on both amps) on BIOS HN7306EA.310 and on the latest, HN7306EA.314 (09/18/2026), so ASUS has not fixed it yet. @bardliao, do you know whether ASUS has been told, as HP was?

Until ASUS ships a fix, please disregard my earlier suggestion to limit the fallback by DMI: on 7.4 that would leave the PX13 without a right speaker. The mfg 0x0102 / part 0x0000 match as posted covers it. @jonnyjonjon, since you're on 7.4-rc1, this combination may also be worth checking on 8EB4.

I've reported the 7.4 regression on linux-sound in reply to the TI patch: https://lore.kernel.org/all/20261107212552.10990-1-leepaulmann@gmail.com/

Disclosure: I worked through this with AI assistance (Claude Code). The results above are from my machine, checked by ear and in the kernel log.

@jonnyjonjon

jonnyjonjon commented Nov 9, 2026 •

Copy link
Copy Markdown

@leepaulmann The Asus PX13 is a separate laptop from a different manufacturer. It likely has its own bug tracker and candidate PRs. Please search for those, and comment there, not here.

From the title, this PR was opened specifically or the HP Omnibook Ultra 14 (8EB4). I joined to add details that the latest bios update for this exact model (F.07, early Oct) did not resolve its issues, since I saw that @bardliao indicated that a bug had been submitted to HP.

dmesg | grep -iE "sof" still shows:

[ 889.573615] sof-audio-pci-intel-ptl 0000:00:1f.3: No SoundWire machine driver found for the ACPI-reported configuration:
[ 889.573626] sof-audio-pci-intel-ptl 0000:00:1f.3: link 1 mfg_id 0x0102 part_id 0x0000 version 0x3
[ 889.573634] sof-audio-pci-intel-ptl 0000:00:1f.3: link 1 mfg_id 0x0102 part_id 0x0000 version 0x3
[ 889.573637] sof-audio-pci-intel-ptl 0000:00:1f.3: link 2 mfg_id 0x0102 part_id 0x0000 version 0x3
[ 889.573639] sof-audio-pci-intel-ptl 0000:00:1f.3: link 2 mfg_id 0x0102 part_id 0x0000 version 0x3

Happy to provide additional info or logs, as needed.

@leepaulmann

Copy link
Copy Markdown

Fair enough. I'd only add that the SDCA hunk matches the TAS2783 by mfg/part ID rather than by DMI, so it changes behaviour on the PX13 too. That's why I tested it here. I'll keep further PX13 follow-up on linux-sound.

@bardliao

Copy link
Copy Markdown
Collaborator

@leepaulmann The Asus PX13 is a separate laptop from a different manufacturer. It likely has its own bug tracker and candidate PRs. Please search for those, and comment there, not here.

From the title, this PR was opened specifically or the HP Omnibook Ultra 14 (8EB4). I joined to add details that the latest bios update for this exact model (F.07, early Oct) did not resolve its issues, since I saw that @bardliao indicated that a bug had been submitted to HP.

dmesg | grep -iE "sof" still shows:

[ 889.573615] sof-audio-pci-intel-ptl 0000:00:1f.3: No SoundWire machine driver found for the ACPI-reported configuration: [ 889.573626] sof-audio-pci-intel-ptl 0000:00:1f.3: link 1 mfg_id 0x0102 part_id 0x0000 version 0x3 [ 889.573634] sof-audio-pci-intel-ptl 0000:00:1f.3: link 1 mfg_id 0x0102 part_id 0x0000 version 0x3 [ 889.573637] sof-audio-pci-intel-ptl 0000:00:1f.3: link 2 mfg_id 0x0102 part_id 0x0000 version 0x3 [ 889.573639] sof-audio-pci-intel-ptl 0000:00:1f.3: link 2 mfg_id 0x0102 part_id 0x0000 version 0x3

Happy to provide additional info or logs, as needed.

Yes, please share the full dmesg log. No SoundWire machine driver found for the ACPI-reported configuration: is expected for PTL+ platforms.

@jonnyjonjon

jonnyjonjon commented Nov 11, 2026 •

Copy link
Copy Markdown

@bardliao - Ran the following commands:

journalctl -k -b > dmesg_log.txt
sudo acpidump > acpidump.txt
ls -l /sys/bus/soundwire/devices/ > sdw_devices.txt
alsa-info.sh --no-upload --output alsa-info.txt

Output here: https://github.com/jonnyjonjon/hp-omnibook-ultra-14-kd0000-8eb4-tas2783

Log is spammed with tas2783 communication errors.

Let me know if you need anything else. Also, if I should add anything to the associated issue: #5732

jonnyjonjon added a commit to jonnyjonjon/linux that referenced this pull request Nov 17, 2026
intel_shim_vs_init() takes every ACTMCTL timing field from the _DSD
properties, except bit 0 (DACTQE), which it always sets. That dates from
the first ACE2.x code, which also hard-coded DOAIS = 1.
commit 8096248 ("soundwire: intel_ace2x: cleanup DOAIS/DODS
settings") changed the DOAIS default to the hardware default (3), and
commit 3b0b441 ("soundwire: intel_ace2x: use DOAIS and DODS settings
from firmware") then took DOAIS from the _DSD properties, but DACTQE is
still forced on.

On the HP OmniBook Ultra 14 (board 8EB4, Panther Lake, BIOS F.07) the
firmware gives every link DOAIS = 3 and DOAISE2 = 0. Links 1 and 2 each
carry two TI TAS2783 amplifiers. Linux programs ACTMCTL = 0x117D, and one
amplifier per link keeps failing: "Program SCP Ack not received", parity
errors and firmware download failures, tens of thousands of log lines per
boot. Link 3, with a single RT712, works with the same value.

ACTMCTL reads 0x003A before Linux writes it, so DACTQE resets to 0.
Leaving it at that value gives ACTMCTL = 0x117C with the firmware's own
DOAIS and DOAISE2, and all four amplifiers attach and play.

An alternative proposed for this board in thesofproject/linux PR thesofproject#5904
forces DOAIS = 1 and DOAISE2 = 1 through a DMI quirk, which restores the
original DOAIS = 1 / DACTQE = 1 pairing. Not forcing DACTQE avoids a
board quirk and leaves every ACTMCTL field to the firmware properties or
the hardware reset value.

This changes ACTMCTL on every ACE2.x and later platform. It has only been
tested on the board above, and the meaning of DACTQE is not publicly
documented, so it needs testing on other platforms.

On 8EB4 the speakers also need a separate SDCA fix (the firmware lacks
the function type constant for the amplifiers); without it the
amplifiers enumerate but no speaker endpoint is created.

Fixes: 8096248 ("soundwire: intel_ace2x: cleanup DOAIS/DODS settings")
Link: thesofproject#5732
Link: thesofproject#5904
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jonnyjonjon added a commit to jonnyjonjon/linux that referenced this pull request Nov 19, 2026
intel_shim_vs_init() takes every ACTMCTL timing field from the _DSD
properties, except bit 0 (DACTQE), which it always sets. That dates from
the first ACE2.x code, which also hard-coded DOAIS = 1.
commit 8096248 ("soundwire: intel_ace2x: cleanup DOAIS/DODS
settings") changed the DOAIS default to the hardware default (3), and
commit 3b0b441 ("soundwire: intel_ace2x: use DOAIS and DODS settings
from firmware") then took DOAIS from the _DSD properties, but DACTQE is
still forced on.

On the HP OmniBook Ultra 14 (board 8EB4, Panther Lake, BIOS F.07) the
firmware gives every link DOAIS = 3 and DOAISE2 = 0. Links 1 and 2 each
carry two TI TAS2783 amplifiers. Linux programs ACTMCTL = 0x117D, and one
amplifier per link keeps failing: "Program SCP Ack not received", parity
errors and firmware download failures, tens of thousands of log lines per
boot. Link 3, with a single RT712, works with the same value.

ACTMCTL reads 0x003A before Linux writes it, so DACTQE resets to 0.
Leaving it at that value gives ACTMCTL = 0x117C with the firmware's own
DOAIS and DOAISE2, and all four amplifiers attach and play.

The meaning of DACTQE is not publicly documented and leaving it clear has
only been tested on this board, so add a DMI quirk instead of changing
every ACE2.x platform: on matching systems the driver no longer sets
DACTQE and leaves it at its hardware reset value. The quirk is evaluated
once per link when the Intel properties are read. Other platforms keep
the current behaviour.

An alternative proposed for this board in thesofproject/linux PR thesofproject#5904
forces DOAIS = 1 and DOAISE2 = 1 through a DMI quirk, which restores the
original DOAIS = 1 / DACTQE = 1 pairing. This quirk instead overrides
nothing the firmware specifies; it only stops forcing a bit for which
the firmware has no property.

On 8EB4 the speakers also need a separate SDCA fix (the firmware lacks
the function type constant for the amplifiers); without it the
amplifiers enumerate but no speaker endpoint is created.

Link: thesofproject#5732
Link: thesofproject#5904
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jonnyjonjon added a commit to jonnyjonjon/linux that referenced this pull request Nov 19, 2026
…a 14

On the HP OmniBook Ultra 14 (board 8EB4, Panther Lake; BIOS F.06 and
F.07) the firmware declares the Function Topology control (0x05) of the
four TI TAS2783 amplifiers with access mode 3 and no
mipi-sdca-control-dc-value. find_sdca_function() then fails with
"function type only supported as DisCo constant", no SmartAmp function
is created, there is no speaker endpoint, and the internal speakers are
silent. The RT712 on the same machine provides the constant and is
unaffected.

This is a firmware defect. The same one was fixed by a BIOS update on the
HP EliteBook X (thesofproject#5760) and on the OmniBook X Flip 14 (8EA4, BIOS F.11),
and a fix has reportedly been shared with HP for this board too. Until
it ships, add a quirk: when the constant is absent on this board, treat
a TAS2783 (mfg 0x0102, part 0x0000) as a SmartAmp, which is the only
function that part implements. On the OmniBook X Flip 14 (8EA4), whose
fixed BIOS (F.11) declares the constant, the same part is reported as a
SmartAmp (type 1).

The quirk is limited to this vendor and board name, and only applies
when the property is absent, so a fixed BIOS bypasses it and no other
machine changes behaviour. The fallback is logged with FW_BUG.

Based on the unconditional fallback proposed by Kasey Prichard in
thesofproject/linux PR thesofproject#5904.

Link: thesofproject#5732
Link: thesofproject#5904
Link: thesofproject#5760
Link: thesofproject#5981
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

8 participants