J313: tas2770 TDM TX edge (TDM_CFG4) misaligned; speakersafetyd panics on garbage IV sense

[paganotoni]

Summary

On MacBook Air M1 (J313), speakersafetyd panics immediately on garbage I/V sense data, so snd-soc-macaudio locks the speakers silent. Playback TDM is fine; capture (SDOUT / IV sense) is bit-shifted because tas2770 leaves TDM_CFG4 TX edge at the chip reset default (falling). Setting TX edge to rising makes IV sense valid and the daemon stays up.

Hardware / software

• Machine: apple,j313 (MacBookAir10,1, M1 Air 2020)
• Kernel: linux-asahi 7.1.13.asahi3-1 (7.1.13-3-1-ARCH)
• speakersafetyd 2.0.1-1, asahi-audio 4.1-1
• Codecs: ti,tas5770l / ti,tas2770 at i2c 1-0031 (Left) and 2-0034 (Right), bound to snd_soc_tas2770

DT sense slots (as expected):

• Left: ti,imon-slot-no = 0, ti,vmon-slot-no = 2
• Right: ti,imon-slot-no = 4, ti,vmon-slot-no = 6

Symptoms

speakersafetyd never stays running after boot. Typical panics:

Left: Negative power, bad ivsense data? (-4.68)
Left: Coil temperature limit exceeded (135.00 > 120)

Kernel then logs Speaker volumes locked: Main control not locked and analog output is silent even though PipeWire is playing to the j313 convolver / RawSpeakers.

The 135 °C trip happens within ~1s of a cold start (thermal model starts near 99 °C). That is not real coil heating; it is the model integrating garbage power.

What is not wrong

• Sense slots are programmed correctly (TDM_CFG5/TDM_CFG6).
• Amps are active with ISENSE/VSENSE on (PWR_CTRL = 0x00).
• Die temp via hwmon is ~31 °C, no latched faults.
• Playback path works: unlocking speakers briefly produced sound.
• j313.conf channel map (is_chan/vs_chan 0/1 and 2/3, 4ch S16) is correct once SDOUT is aligned.

Register dump (while playing, speakers locked)

Reg	Left 1-0031	Right 2-0034	Meaning
0x02 PWR_CTRL	0x00	0x00	Active, I/V sense not powered down
0x0A TDM_CFG0	0x08	0x08	96 kHz
0x0C TDM_CFG2	0x1e	0x1e	32-bit slots
0x0F TDM_CFG5	0x42	0x46	VSNS enable, slots 2 / 6
0x10 TDM_CFG6	0x40	0x44	ISNS enable, slots 0 / 4
0x0E TDM_CFG4	0x13	0x13	chip default: TX falling, offset 1, FILL

tas2770_set_fmt() configures SDIN (RX). tas2770_codec_probe() sets IV TX slots in CFG5/CFG6 but does not program CFG4 TX edge, so SDOUT keeps the silicon reset value 0x13.

IV capture vs TX edge

arecord -D hw:AppleJ313,2 -c 4 -r 96000 -f S16_LE with speakers locked/silent:

TDM_CFG4	Result
0x13 (default, TX falling)	Full-scale uncorrelated noise (RMS ~16k on S16), corr(ch0,ch1) ≈ 0, speakersafetyd-style power ≈ -0.14 W
0x12 (same but TX rising)	Real noise floor (RMS ~15–20), power ≈ 0 W
0x03 (falling, no FILL)	Still garbage / large negative power
0x10 / 0x02 / 0x00 (rising)	Quiet, like 0x12

So the failing bit is TDM_CFG4 bit 0 (TX_EDGE_FALLING). RX can use falling edge and still play; capture wants rising on this machine.

Workaround (confirmed)

i2cset -y -f 1 0x31 0x0E 0x12
i2cset -y -f 2 0x34 0x0E 0x12
systemctl restart speakersafetyd

After that, speakersafetyd stays running, volumes unlock, speakers work. CFG4 stays 0x12 across PCM open.

This is a userspace band-aid; it needs repeating after codec reset/probe (e.g. ExecStartPre on speakersafetyd.service).

Suggested kernel fix

In sound/soc/codecs/tas2770.c, program TDM TX edge/offset in tas2770_set_fmt() / probe so SDOUT matches Apple MCA capture (rising edge on J313), instead of leaving CFG4 at reset default. A DT property would be cleaner if other machines need falling.

Extra noise (probably unrelated)

First probe logs tas2770_i2c_probe: Failed to parse devicetree because SDZ supply is missing (supply SDZ not found, using dummy regulator); deferred probe then succeeds. IV slots are still set correctly after that.

[chadmed]

Cannot reproduce on Fedora Asahi Remix or Gentoo. Consider using a stable and supported distro.

https://asahilinux.org/llm-policy/