Skip to content

j293 (MacBook Pro 13" M1): built-in mics record silence, AOP decimator gets no PDM input #355

Open

Description

@zipos

System

  • MacBook Pro 13" M1 2020 (J293, t8103)
  • Omarchy 4.5.0, fresh install (2026-11-02)
  • Kernel: linux-aurora 7.2.11.aurora14-1
  • asahi,os-fw-version: 13.5 (iBoot-8422.141.2); AOP firmware reports aop-audio v240.10 (built Jun 30 2023)
  • aurora-audio 4.1.aurora3-3, triforce-lv2 0.3.2, pipewire 1.6.9, wireplumber 0.5.18

What's wrong

The built-in mic array records silence under Linux. The mic works fine in macOS on the same machine, so the hardware is OK.

The PipeWire / asahi-audio chain is fine (RawMics → j293-mic triforce filter → "MacBook Microphone (Mono)" → Omarchy's stereo remap; all linked and running). The problem shows up at the ALSA level, with PipeWire out of the way:

arecord -D hw:AppleJ293HPAI,0 -f FLOAT_LE -c 3 -r 48000 -d 5 f.wav

while playing a 1 kHz tone through the built-in speakers (audible), or while talking:

  • The first ~50 ms of every capture contain small real-looking float values (peak ≈ 7e-4), which then decay smoothly to zero (3e-6 at 50 ms, ~2e-38 after that). That looks like the decimator's response settling with no PDM input, as if the mics are never powered or clocked.
  • After that the stream is all zeros or denormals. The tone or voice never appears in any channel (1 kHz magnitude ~1e-40).
  • Channels 0 and 1 are bit-identical in ~100% of samples.
  • The kernel logs no errors on open/close (no "Unable to power mic on"). The AOP syslog only logs the PDM state change once, at probe (PDMDev<pdm0> off ->xi0 , 0->2400000 AP state 1), and nothing when a capture stream is opened.
  • Reproduced on several cold boots (full power-off) with zero suspends, the last one after updating to Omarchy 4.5.0. Lid open (logind LidClosed=false).

I checked:

  • sound/soc/apple/aop_audio.rs in linux-aurora is identical to AsahiLinux/linux asahi.
  • The t8103-j293 DTB's AOP/audio nodes match linux-asahi 7.2.12.asahi3 apart from phandle numbering.

Side note: driver rebind fails

modprobe -r snd_soc_aop && modprobe snd_soc_aop does not recover the mic. The re-probe fails because the AOP still has pdm0 attached, and the card stays gone until reboot:

apple_aop 24ac00000.aop: RTKit: syslog message: Controller.cpp:386: cond "!(pNode->isAttached())" fail, _commandNodeAttach() ln 386, stat 0xe00002d
apple_aop 24ac00000.aop: RTKit: syslog message: Controller.cpp:393: Failed to attach node 'pdm0'
snd_soc_apple_aop audio.1.auto: Unable to attach device 1885629744, return code 3758097109
snd_soc_apple_aop audio.1.auto: probe with driver snd_soc_apple_aop failed with error -5

AOP boot log

apple_aop 24ac00000.aop: RTKit: syslog message: ClientInfo.cpp:184: exp "inCallbackAction(it.key(), inRef)" fail, RunActionOnClientManagers(), ln 1
apple_aop 24ac00000.aop: RTKit: syslog message: udioPDMNodeBase.cpp:150: PDMDev<pdm0> wake notification. Prev AP state 1
apple_aop 24ac00000.aop: RTKit: syslog message: iggerController.cpp:129: VoiceTrig ready!! <AppleAOPVoiceTriggerController v240.2 ...>
apple_aop 24ac00000.aop: RTKit: syslog message: Controller.cpp:160: RTKitAudioFramework v240.4 (built Jun 30 2023 17:51:05) ready!! {12 nodes}
apple_aop 24ac00000.aop: RTKit: syslog message: Controller.cpp:251: aop-audio v240.10 (built Jun 30 2023 17:56:11, debug 0x0) ready!!
apple_aop 24ac00000.aop: RTKit: syslog message: SPU_init.cpp:320: ===( AOP IS READY )===
apple_aop 24ac00000.aop: RTKit: syslog message: udioPDMNodeBase.cpp:115: PDMDev<pdm0> LPClkCfg=Src:pll ,Freq:2400000
apple_aop 24ac00000.aop: RTKit: syslog message: udioPDMNodeBase.cpp:554: PDMDev<pdm0> off ->xi0 , 0->2400000 AP state 1

Happy to run any test builds or extra diagnostics.

Activity

  1. gianmegantara commented on Nov 6, 2026

    @gianmegantara

    Same symptom on J293 (MacBook Pro 13" M1 2020), and it looks like the same root cause as #141 (J313): the AOP capture buffer is a fixed loop, not live samples. The mic works in macOS on this machine, so the hardware is fine.

    System:

    • MacBook Pro 13" M1 2020 (J293, t8103), Omarchy/Aurora, fresh install
    • linux-aurora 7.2.11.aurora14-1, aurora-audio 4.1.aurora3-3
    • asahi,os-fw-version 13.5; AOP reports aop-audio v240.10 (built Jun 30 2023)

    Details:

    • FLOAT_LE / 3 ch / 48 kHz: the stream decays to zeros/denormals, same as reported here.
    • With the AOP PCM forced to S32_LE (WirePlumber audio.format = "S32LE"), the samples are non-zero and the loop becomes visible:
    • The loop is unaffected by loud speech, by tapping the lid next to the mic, or by a 440 Hz tone played through the built-in speakers (arecord -D hw:AppleJ293HPAI,0 -f S32_LE -c 3 -r 48000). RMS and peak stay constant to 3 significant figures across a 25 s capture.
    • No kernel errors on open/close; the AOP logs the PDM state only once at probe (PDMDev<pdm0> off ->xi0 , 0->2400000 AP state 1), and nothing when a capture stream is opened.

    So this is not the WirePlumber/asahi-audio chain (bypassing it with the raw hw: device reproduces it) and not hardware. It looks like the t8103 AOP/ADMAC/DART capture path never receives/refreshes PDM samples, consistent with the #141 analysis.

    Happy to run test kernels or extra diagnostics. I can also retest after any ADMAC/DART/AOP changes.

  2. Natetgmaxwell commented on Nov 6, 2026

    @Natetgmaxwell

    Confirming on another J293 (MacBook Pro 13" M1 2020, MacBookPro17,1).

    System:

    • linux-aurora 7.2.11.aurora15-1 (kernel 7.2.11-15-1-ARCH); aurora14 is the same per the other reports
    • asahi-audio 4.1-2, alsa-ucm-conf-asahi 10-1, PipeWire 1.6.10, WirePlumber 0.5.18
    • asahi,os-fw-version 13.5; AOP boots normally, and the PDM clock starts at probe (PDMDev<pdm0> off ->xi0 , 0->2400000). Nothing is logged when a capture stream opens, and there are no kernel errors.

    Observations (PipeWire bypassed with raw arecord -D hw:2,0 -r 48000 -c 3, speaking during capture):

    • Only 3 channels are accepted (-c 1 and -c 2 fail with "Channels count non available").
    • S32_LE: all three channels are almost identical. A frame looks like [a, a, b], with channels 0 and 1 equal. The values are a large slow-drifting DC offset (about 7.7e8 rising to 9.7e8), and the min/max are identical between separate captures.
    • After removing the DC offset, the residual level is constant over time (10th and 90th percentile within about 3%), which means no speech dynamics.
    • Through the PipeWire chain (effect_output.j293-mic), peak is about 140 out of 32767, which is a silent noise floor, so whisper-style dictation reports "No speech detected".
    • hw:1,0 (the headset path, jack unplugged) is exact zeros, as expected.

    This matches the fixed-loop explanation above (same symptom as #141 / #17602). It is not a userspace problem, and I couldn't find any config-level workaround. Happy to test patches or a 7.1.x kernel build on this machine.

    Diagnosed with Claude Code.

  3. DanirHW commented on Nov 6, 2026

    @DanirHW

    Same symptom on J293, and an exact fixed-loop confirmation on the S32_LE path.

    Mac and installed build

    • Model / board / SoC: MacBook Pro 13" M1 2020 / apple,j293 / apple,t8103
    • Omarchy 4.5.0, fresh install (2026-11-05), later fully reinstalled again: same result
    • uname -r: 7.2.11-14-1-ARCH (linux-aurora 7.2.11.aurora14-1)
    • aurora-audio 4.1.aurora3-3, PipeWire 1.6.9, WirePlumber 0.5.18
    • No local patches.

    Both firmware versions and provisioning

    • asahi,os-fw-version: 13.5
    • AOP syslog reports aop-audio v240.10 (built Jun 30 2023)
    • Provisioning: not relevant.

    Reproduction, expected result, and actual result

    Raw capture with PipeWire out of the way (arecord -D hw:AppleJ293HPAI,0 -f S32_LE -c 3 -r 48000 -t raw), while playing a clearly audible 1 kHz tone through the built-in speakers:

    • The stream repeats exactly every 40,647 frames (100 % sample-for-sample match across consecutive periods, checked over multiple captures).
    • 55.5 % of 32-bit words are exactly 0; channels 0 and 1 are bit-identical in 100 % of samples.
    • A large per-capture DC offset is present (latest capture: mean ≈ +4.9e6); its value differs between capture sessions, matching the slow-drift description above.
    • The tone produces no response at all: RMS, peak and the loop contents are unchanged while sound is playing. Expected: independent live mic channels; actual: the fixed pattern only.
    • Frequency: every capture, every boot, after several full power cycles and two independent OS installs.
    • Last known good (owner report): the built-in mic worked on the previous Omarchy install in September; it stopped after a system update. Build of that install not recorded.

    Relevant error lines

    None. No kernel errors on PCM open/close; the only PDM-related AOP line is at probe:

    udioPDMNodeBase.cpp:554: PDMDev<pdm0> off ->xi0 , 0->2400000 AP state 1
    

    and nothing is logged when a capture stream opens (same as the reports above).

    Through the aurora-audio DSP graph the PipeWire chain is intact and linked (RawMics → j293-mic → effect_output.j293-mic → Omarchy's stereo remap, default source); apps record only the fixed pattern's noise floor, so dictation sees "no speech".

    Related issues, pull requests, or code examples

    Matches the fixed-loop analysis in #141 and the parent report here.

    What was not tested

    • Upstream linux-asahi kernel on this Mac (owner not swapping kernels for now).
    • Bluetooth / USB microphone workaround (no headset at hand yet).
    • macOS cross-check (macOS not installed on this unit).
  4. PastaPastaPasta commented on Nov 9, 2026

    @PastaPastaPasta

    Confirming this on a J293 (MacBook Pro 13" M1, apple,t8103) with linux-aurora 7.2.11.aurora14-1 and omarchy-mac 0.2.3-2, and confirming the workaround from omacom/omarchy-mac-pkgs#90 fixes it on this model.

    Observed before the workaround:

    • ALSA capture (AppleJ293HPAI, 3ch float 48 kHz) is RUNNING with the hardware pointer advancing; PipeWire shows effect_output.j293-mic and omarchy_asahi_mic running, unmuted, actively linked.
    • pw-record from the default source (and directly from effect_output.j293-mic) captures only a ~65 ms filter transient, then exact digital zeros for the rest of the take. Restarting the full audio stack did not help (the post-restart take was 100% zeros).
    • apple_sep was not loaded; asahi,os-fw-version is 13.5, so the older firmware-version issue doesn't apply.

    After loading the enclave driver read-only, exactly as in omacom/omarchy-mac-pkgs#90:

    sudo modprobe apple-sep xart_writes=0 provision_keybag=0
    

    the kernel logs writes disabled for xART, the same pw-record test immediately captures normal speech (RMS −25.7 dBFS, peaks −7.6 dBFS over 6 s), and hold-to-talk dictation works end-to-end. This was loaded mid-session; no PipeWire restart was needed afterwards.

    So the root cause/fix path in #141 / #180 also covers J293. Happy to attach pw-dump/journal excerpts if useful.


    Disclosure: This report was drafted with the help of an AI coding assistant (Muse Code, powered by muse-wonder-full-1.0) during a live debugging session. The measurements and workaround result above were reproduced and verified on my machine.

  5. adamcooper commented on Nov 11, 2026

    @adamcooper

    Another J293 (MacBook Pro 13" M1) confirmation, fresh install 2026-11-05.

    Versions: linux-aurora 7.2.11.aurora14-1, omarchy-mac 0.2.3-2, aurora-audio 4.1.aurora3-3, triforce-lv2 0.3.2-1, pipewire 1.6.9, wireplumber 0.5.18. macOS stub firmware 13.5 (aop-audio v240.10).

    Symptom: the built-in mics have never worked on this install. hw:AppleJ293HPAI,0 capture runs (hw_ptr advances), but all 3 channels are exact 0.0 (−752 dBFS), even with a 1 kHz tone playing through the speakers. I dumped the raw input with debug.wav-path on audio_effect.j293-mic, so this is upstream of the DSP chain. Reboots and PipeWire/WirePlumber restarts made no difference. apple_sep is not loaded at boot (no module alias since 36ebdee).

    Workaround confirmed: sudo modprobe apple-sep xart_writes=0 provision_keybag=0 makes the mic live immediately on the same boot, with no PipeWire restart. dmesg shows xART: ... writes disabled, TZ0 accepted, IMG4 acknowledged. The same tone test then gives real signal (2% zero samples instead of 100%), and dictation works end to end. As expected, it's lost on reboot until loaded again.

    +1 for landing #180 / omarchy-mac-pkgs#90.

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