[erikwb]

On the J713 (MacBook Air 13" M4) the speaker amps' current and voltage sense goes bad for a moment at the end of every playback stream. This makes the SN012776 amps wait for the speaker bus to stop before they go to software shutdown. It's opt-in per board through a new device tree property; only the J713 sets it.

Depends on #439 (the J713 speakers). The first commits here are #439's; the last four are new.

What happens

The AOP keeps the speaker TDM bus running for a while after a playback stream stops, and the speaker protection daemon keeps reading the amps' I/V sense words in that time. The TAS2764 driver mutes each amp at the stream's end, waits for the ramp-down, then puts it in software shutdown, one amp after another. Those shutdowns land in the bus's tail.

The right tweeter's voltage sense slot then reads full scale for about 20 samples at every stream stop. It happens for a stream of digital silence played with the speaker volumes locked, so it isn't sound. speakersafetyd and speakerguardd both charge those samples as power.

The change

1. dt-bindings: apple,deferred-shutdown on ti,tas27xx: the amp's bus outlives its playback stream and its sense slots are still sampled, so it must not enter software shutdown until the bus has stopped.
2. tas2764: with the property, the amp is muted at the stream's end as before and goes to software shutdown 150 ms later. On the J713 the bus ran on for up to 3520 frames (73 ms) after a stream's end.
   • A stream that starts in between cancels the pending shutdown.
   • Suspend and removal shut the amp down at once.
   • A muted amp sees the bus clock stop while it waits, and can see the clock fault latched again after its shutdown. While the amp is stopped, that clock-only fault is expected and no longer logged. A clock fault during a stream is still reported.
   • Amps without the property take exactly the same paths as before.
3. t8132-j713: set it on the four amps.
4. tas2764 (new): drop the clock fault latched while stopped at unmute.
   • The problem: a stream that started soon after the last one ended often logged "bad TDM clock" on the first amp the link unmutes (the left woofer). The fault wasn't the new stream's. The muted amp latches the stopped clock about every 10 ms while it waits. The four amps share one fault line, and each handler reads six latch registers over I2C, so a latch can reach its handler up to about 10 ms late. By then the unmute has already cleared the stopped state.
   • The fix: at unmute, a latch holding only the clock fault is cleared before the amp counts as running. The handler holds a lock while it reads and judges the latches, so the unmute can't fall between the two. Any other latched fault is left for the handler.

Testing (J713)

• Sense tails: I captured the speaker sense stream directly while playing three silent streams and a chord with the volumes locked.

  • Before: about 90 full-scale words per run, all on the right tweeter's voltage slot, at every stream stop.
  • After: none.
  • No fault logged, and every amp still ends in software shutdown (PWR_CTRL 0x2).

• Restarts: pairs of short streams 90, 120 and 150 ms apart, so the second starts while the first's shutdown is pending. All four amps play, the sense words stay valid, and there are no pops.

• Faults at restarts (commit 4): silent streams of 0.3 s on the speakers, with gaps of 40 ms to 1 s before each start. Faults are counted from the journal.

  	Starts that logged a fault
  Without commit 4	9 of 64, all on the left woofer: 9 of 32 with gaps up to 140 ms, 0 of 32 with gaps of 170 ms or more
  With commit 4	0 of 96 (and 0 of 96 again with tracing, where 11 stale latches were cleared at unmute)

  An earlier version of this description said the restart pairs logged no fault. That was wrong: they hit this race.

• Under speakerguardd: a recording of 13 streams, including those restarts, has no full-scale sense words, and its replay certifies.

• Checks:

  • checkpatch --strict is clean;
  • every commit builds with W=1;
  • dt_binding_check passes for ti,tas27xx.yaml;
  • CHECK_DTBS on the J713 DTB gives the same diagnostics as ASoC: apple: J713 (M4 MacBook Air) speakers on the AOP (depends on #437) #439's head.

Untested elsewhere. The J616s has the same AOP-fed speaker stream and may want the property too.

[erikwb]

Superseded by #820, where these four commits are 13–16, unchanged.