[joshuaswarren]

Summary

The audio side of the Apple M6 (T8152) Mac mini after the plumbing of #194: the DART owned-streams change, the per-channel ADMAC apertures, the CIO reset driver, the AOP audio driver variant and the macaudio card for the J873, the mailbox support the AOP and ACIO need, and the device-tree nodes. Twelve commits.

This series now enables the internal speaker on the J873 (commit 30). It stays a draft until the review of commit 30 is done. The T8152 speaker link carries 24-bit samples, and the earlier limiter and integer output commits assumed 2^30. Commit 27 fixes the scale. Commits 28 and 29 add a 100 Hz high-pass and a measured ceiling. The messages of the limiter commit (43f2664) and the plain-PCM commit (68c6401) still carry the old 2^30 wording. The scale commit says so. Before this PR is merged, ask the maintainers to squash, or rewrite those messages.

Commit 30 sets enable_speakers = true and safe_vol = 0 in macaudio_j873_cfg. Before it, the speaker link was skipped unless the please_blow_up_my_speakers module parameter was set.

The first twelve commits of this branch are #191 (the M6 base) and #194 (audio plumbing). Only the last seventeen are new here: the diff of this PR shows the other twelve too until they merge, then I rebase. Review the last seventeen.

It is not needed for the release candidate, and there is no rush to review or merge it. Sound does not reach a codec or speaker from this branch yet (see Blocked).

Commits

#	Commit	Notes
13	iommu/apple-dart: add owned-streams and inherited-roots for T8152	a DART shared with firmware: Linux owns only the listed streams, never resets or acknowledges the others; the exception handler and the probe drain mask pending exceptions with the owned bitmap
14	dma: apple-admac: T8152 channels have per-channel 16 KiB apertures	TX n at 0x40000 + n * 0x4000, RX n at 0x80000 + n * 0x4000, selected by a flag set only for T8152
15	reset: apple-cio: add apple,t8152-cio-reset	request at reg + 0, done at reg + 4, one bit per cluster, 4 resets, child of the PMGR syscon
16	ASoC: apple: t8140-aop-audio: add the T8152 variant	Rust; no lpai/hpai/tap, no source ring, no fabric raise; the speaker front-end and back-end accept one or two channels
17	ASoC: apple: macaudio: add apple,j873-macaudio	j873 configuration, Speaker prefix for the mono amplifier; speakers stay disabled in the configuration
18	soc: apple: mailbox: add the T8152 ACIO counted-FIFO mailbox	identical driver patch to the commit in #197 (same patch-id). The apple,t8152-acio-mailbox binding text is only in #197, so this layer needs #197 first for that compatible; the second copy of the driver patch becomes empty when both land
19	soc: apple: mailbox: add the T8152 ASCWrap v8 AOP mailbox compatible	apple,t8152-ascwrap-v8-mailbox on the same hardware data
20	dt-bindings: sound: cirrus,cs42l84: add the DAI common properties	the codec node carries sound-name-prefix
21	arm64: dts: apple: t8152-j873g: add the audio I2C codecs
22	dt-bindings: soc: apple: aop: no power domain on T8152	required only for T8140
23	arm64: dts: apple: t8152-j873g: add the AOP audio chain	AOP, its mailbox, ADMAC nodes, the shared DART
24	ASoC: apple: t8140-aop-audio: limit the T8152 speaker with a thermal model	in-kernel limiter in the speaker copy path, KUnit suite of four tests; see the section below
25	ASoC: apple: t8140-aop-audio: give the T8152 speaker plain PCM	the T8152 amplifier takes the limited word as a plain 32-bit integer, not float32 bits (3 lines); found by a microphone run on the M6 with the previous commit
26	ASoC: apple: macaudio: give the J873 speaker link a rising RX edge	the speaker amplifier stays in software shutdown with the default falling receive edge; a platform field sets the speaker link format only (10 lines, macaudio.c); inert until enable_speakers is true
27	ASoC: apple: t8140-aop-audio: scale the T8152 speaker words to 24 bits	arithmetic shift right by 7 of the limited word, so full scale is 2^23 as the amplifier link wraps; code only in t8140_aop_audio.rs; measured, see the next section
28	ASoC: apple: t8140-aop-audio: high-pass the T8152 speaker at 100 Hz	a filter in the speaker copy path, KUnit tests including no latch of the clamped state
29	ASoC: apple: t8140-aop-audio: set the T8152 speaker ceiling	CEILING_Q30 = 83,800,000 with its measured basis
30	ASoC: apple: macaudio: enable the J873 speaker	enable_speakers = true, safe_vol = 0, comments updated

Measured, on an M6 Mac mini (out-of-tree builds of the same changes, before this series was rebased)

• The AOP boots, the audio service has speaker, jack output and jack input devices and no lpai/hpai/tap.
• With the T8152 ADMAC apertures the jack stream (S24, 48 kHz stereo) is consumed in real time: a DMA report every 125.00 ms for 48000-byte periods.
• The speaker stream drains at half the frame rate in stereo and at 48 kHz in mono (24000-byte periods, one report per 124.67 ms), with no hang and no DART fault. aplay -c 1 on the speaker PCM was refused before the back-end change and played after it.
• The AOP serves audio with a node that has no power-domains.

Speaker series (commits 27 to 30), measured on the M6 (25 Oct)

MEASURED:

• With the scale fix, a low-level tone came out 17.0 dB above the previous low-level anchor. The shift predicts 17.1 dB. So the 128x shift and the 2^23 scale hold at low level.
• Tones at 440 Hz, 1, 2 and 4 kHz were clean. There was no sign of buzz or rub from 440 Hz to 4000 Hz.
• At the ceiling, 440 Hz was 26.7 dB over the room. Its second, third and fourth harmonics were 33, 41 and 45 dB down.
• With the high-pass in the path, 440 Hz was 1.2 dB higher and 200 Hz was 0.35 dB lower than without it. That is inside the 1.9 dB repeatability of this setup. The filter added no distortion at 440 Hz. The limiter state was unity at both stream starts.

NOT measured:

• The absolute output level and the amplifier output voltage. Both come from the model.
• The speaker cone excursion at low frequencies. Nothing below 200 Hz was played at the ceiling. The 100 Hz high-pass is insurance, reasoned and not measured. 100 Hz and 60 Hz were not played.
• Music. Only tones and one noise burst were played.
• A bad TDM clock line appears once per stream start. It is benign, and not understood further.

Limits of the run: there was no earlier multi-frequency reference, and a dead band is excluded only at the four tones tested.

Not measured

• Sound reaching the jack codec, the headset or the speaker; capture; suspend and resume.
• The CIO reset driver on hardware, and that ACIO n maps to reset cluster n (inferred).
• A foreign DART exception: none was seen, so the handler masking is read, not exercised. A foreign stream that latches an exception keeps a level interrupt asserted; the handler returns IRQ_NONE and the line could be disabled as spurious.
• This branch was not booted.

Blocked

• Commit 30 enables the speaker. It needs one review before this PR leaves draft, because it sets speaker levels on hardware.

Compile and check receipt

arm64 clang, W=1, CONFIG_RUST=y, 16K pages, integration/aurora at 00e601f plus the 26 commits of this branch: iommu/apple-dart.o, dma/apple-admac.o, reset/reset-apple-cio.o, soc/apple/aop.o, soc/apple/mailbox.o, apple/t8152-j873g.dtb, sound/soc/apple/t8140_aop_audio.o and sound/soc/apple/macaudio.o all rc 0 with 0 warnings; the only warning in the Rust build is an unused method in rust/kernel/dma_fence.rs, which these commits do not touch. dtcheck (dtschema 2026.9): the board file gives 0 messages; the 72 binding examples give only the existing apple,dcp-typec-routes message.

Known edges

• A foreign DART stream that latches an exception keeps a shared level interrupt asserted. The handler returns IRQ_NONE, so after enough unhandled counts the kernel could disable the line, and real faults of owned streams with it. No foreign exception was seen on the M6.
• apple,inherited-roots and apple,owned-streams are meant for the T8152 DART only; the schema does not enforce it.
• The CIO reset deassert does not clear a done bit that firmware may have left set before the request, so a stale bit would end the poll at once. Not measured, and there is no consumer yet.
• checkpatch warns that device-tree binding docs should be a separate patch on five commits that carry a driver and its binding together.

Review

Two independent reviews, both GREEN on the first eleven commits at 4b2beadf5d49b05a6d2ef8b704d4f1c4bf12815e, and both GREEN on the twelfth (the limiter) at 43f266453804e1843362c697c2dc4cec5ee09e5b, both GREEN on the thirteenth (T8152 speaker plain PCM) at 68c6401203afb55e7a60706fbca7c55f57350422, and both GREEN on the fourteenth (J873 speaker link rising RX edge) at 444f88a8eff09c01aa345b41654005b57f48307e, the head of this PR. The first review found one defect in the DART exception handler, which is fixed in this head. The second review recomputed dtcheck (0 messages) and the admac, PSD and DMA-channel arithmetic against the device tree.

Speaker thermal limiter (commit 43f2664, t8140_aop_audio.rs only)

Correction (25 Oct): all figures in this section are in the model's units, which assume a full scale of 2^30. The amplifier's real full scale is 2^23, so they are 128 times too optimistic for the words this series sends. See the hold note at the top.

In-kernel feed-forward limiter in the T8152 speaker copy path (the speaker front-end has no mmap flags, so every sample passes through it). The T8152 AOP has no speaker sense tap, so the sense-based userspace daemon cannot run; this limiter uses these thermal model numbers (11.9 V per full scale, 4.6 ohm, coil 40 C/W with 3.4 s, magnet 60 C/W with 210 s, ambient 50 C, limit 140 C): sustained limit 0.9 W (full-scale square wave 30.78 W, so minimum gain -15.43 dB at 2 % margin), immediate attack per 128-word block, 20 s release, idle cooling between streams, fixed point (i64 only). A hard ceiling 20 dB below the sustained level is a const, removed by a rebuild only, for the first audible tests.

Receipts: host reference model plus integer model, 66 checks pass (worst-case sine, square, noise and DC step from four start temperatures stay below 140 C; closed-loop equilibrium 139.89 C); KUnit suite snd_soc_t8140_aop_audio_thermal 4/4 on an arm64 qemu (TCG) kernel at this head, 0 fail 0 skip; compile CONFIG_RUST+KUNIT 16K pages rc 0, 0 warnings; two independent GREEN reviews by hash.

NOT measured: the real amplifier output voltage per full scale, the real speaker resistance and thermal constants (the model uses these numbers as given), the loudest speaker step (distorted on the M6, unexplained) and absolute sound level (the speaker is still disabled in the configuration), the plant 25 % worse than these numbers with the amplifier above the card's gain cap reaches 229.7 C in simulation (no margin by design beyond the 3 dB of the amplifier gain cap and the 2 % power margin). enable_speakers stays false in this series. Not needed for the release candidate.

MEASURED on the M6 with the plain-PCM commit (limiter build, microphone 12 to 18 inches away): a 440 Hz tone at the quietest step is detected 38 dB above the room at 440 Hz, with harmonics below -30 dB relative to the fundamental; the next step, 10 dB up, measures 9 dB louder. Before that commit the same tone gave a steady sound whose strongest component was 880 Hz, at the same level for two steps 10 dB apart. The amplifier itself still needs its configuration and a clock-before-unmute fix in a separate driver change. With the macaudio commit (26) and no register writes from user space the amplifier leaves software shutdown: the quietest step is 39 dB above the room with harmonics below -29 dB relative to the fundamental, and the step 10 dB up is 7 dB louder. NOT measured: absolute level; the loudest step is distorted and unexplained.

[joshuaswarren]

GREEN at 4b2bead

Scope: the eleven commits above the base layer (7d52ab5), read by diff: the shared T8152 DART (owned streams and inherited roots), the per-channel ADMAC apertures, the CIO reset driver, the T8152 AOP audio variant, the J873 macaudio card, the counted-FIFO mailbox and the ASCWrap v8 AOP mailbox compatible, the binding changes and the device-tree nodes. Findings and how they were closed: (1) the runtime Gen3 exception handler acknowledged and logged bits of streams the OS does not own; the head masks the pending word with the owned bitmap through one helper, before the report and before the write-1-to-clear, and the probe drain uses the same helper. (2) With apple,inherited-roots the DART is treated as locked, and every full-reset path (probe, remove, resume) and the attach and detach paths are guarded by that flag, so no existing translation or TTBR is cleared. (3) The counted-FIFO mailbox commit has the same patch-id as the one in the ACIO series that was reviewed separately. (4) The device-tree windows, interrupts, power-domain slot windows, DMA channel numbers and DAI indices were compared with the firmware device tree of the machine and match; the owned stream numbers and the AOP mailbox interrupt order were not independently checked. Residual points the commit text already states: a foreign stream that latches an exception keeps a level interrupt asserted and the handler returns IRQ_NONE, so the line may eventually be disabled as spurious; no such exception has been seen. The speaker front-end and back-end accept one channel on T8152 (the serializer takes one word per frame); the back-end half has a hardware observation, the front-end half is measured only through it. Not covered: any build, device-tree schema check or hardware run by this reviewer; Rust error paths were read for the T8152 probe, not traced in full. Merge needs a second non-author GREEN and CI.

[joshuaswarren]

GREEN at 4b2bead

Review of the 11 commits above 7d52ab5f (DART owned streams, ADMAC apertures, CIO reset, AOP audio T8152, macaudio J873, counted-FIFO mailbox, ASCWrap v8 mailbox, three binding changes, two board-file commits). Four notes below, none blocking.

Checked

• Cumulative patch against the previous head of this branch. Diffing the two cumulative patches (7d52ab5f..ef54e0e0 against ..4b2beadf, index lines ignored) shows hunks only in drivers/iommu/apple-dart.c: a new apple_dart_owned_mask() helper, and both the Gen3 runtime handler and the probe drain now do pending &= apple_dart_owned_mask(dart, i) before logging and before the write-one-to-clear acknowledge. With no apple,owned-streams, the helper returns U32_MAX, so every other DART behaves as before.
• DART parse. An owned stream id at or above num_streams fails the probe with -ERANGE; an empty property fails it; of_xlate refuses a stream that is not owned; apple,inherited-roots makes the DART count as locked and skips the tunables.
• dtschema (2026.9) and dtc 1.7.2 on this head. t8152-j873g.dts preprocesses, compiles with 0 dtc warnings and validates with 0 messages, the new audio, AOP, DART and I2C codec nodes included. The examples of the 7 touched binding files give 0 messages. Of the 72 apple,*.yaml example files, the only message is the older apple,dcp-typec-routes one (typec0-dpin0).
• Mailbox. The drivers/soc/apple/mailbox.c change of the counted-FIFO commit is identical, added and removed lines, to the mailbox.c section of Apple M6 (T8152) USB4/Thunderbolt host router, rear port ACIO3 (on top of #191) #197.
• Board-file numbers. The ADMAC register sizes equal the aperture formula 0x40000 / 0x80000 + (ch / 2) * 0x4000: base-ns, 13 channels, the last RX aperture ends at 0xb4000, which is the reg size; leap-ns, 9 channels, 0xa4000. The dmas entries 2, 10, 11 and 3 are TX1, TX5, RX5 and RX1 under the channel 2n / 2n+1 rule. The PSD windows 0x303a10000 + (15 << 8) and + (17 << 8) are 0x303a10f00 and 0x303a11100. apple,owned-streams = <0 1 40 42> covers every iommus user (AOP 0, base-ns 42, leap-ns 40). sound-dai indexes 0 and 4 equal the T8140 J700 board. The speakers stay disabled in the card configuration.
• AOP audio without lpai. snd_card_free is guarded by !card.is_null(), and the lid handler and microphone jack are created only inside has_lpai.
• checkpatch (--no-tree). 0 errors in all 11 commits. Warnings: "DT binding docs and includes should be a separate patch" on five driver-plus-binding commits, and line-length warnings in the Rust file and the AOP-chain commit.
• Commit text. No host names, home paths or internal names in the commit messages.

Notes

1. Shared interrupt line. The runtime handler now returns IRQ_NONE for a latched exception of a stream that firmware owns. On a shared level-triggered line the kernel disables the interrupt after about 99,900 of 100,000 unhandled counts, which would also drop real faults of owned streams. The DART commit says the line stays asserted and that no foreign exception was seen. It is disclosed, not measured; keep it in the description.
2. Binding dependency. The counted-FIFO commit adds the apple,t8152-acio-mailbox compatible to the driver, but the binding text for it is in Apple M6 (T8152) USB4/Thunderbolt host router, rear port ACIO3 (on top of #191) #197 and not in this branch (this branch documents only apple,t8152-ascwrap-v8-mailbox). State that this layer needs Apple M6 (T8152) USB4/Thunderbolt host router, rear port ACIO3 (on top of #191) #197 first, or add the hunk.
3. CIO done bit. t8152_cio_deassert does not clear a done bit that firmware may have left set before it writes the request, so a stale bit would end the poll at once. Not measured, no consumer yet; clear it first if the vendor sequence allows.
4. Binding wording. The descriptions of apple,inherited-roots and apple,owned-streams say "apple,t8152-dart only"; the schema does not enforce that.

Not checked

A kernel build (I rely on the author's clang W=1 CONFIG_RUST=y 16K-page results), a boot, audio on hardware, the CIO reset consumers (ACIO n to cluster n is inferred), and the single-channel speaker claim, which is taken from the commit text.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The changes since the previous review appear safe to merge, with no new actionable defects found.

Summary

The PR adds T8152 audio support through DART, ADMAC, mailbox, AOP audio, and the J873 sound card.

Changes since the previous review:

• The J873 speaker plays mono audio through a limited 24-bit path.
• The J873 USB2 ports start in their fixed host or device roles.
• T8152 firmware and DMA blocks can share resources with Linux.
• T8152 Mac minis can use the Apple NVMe controller.

No new actionable issues were found. The checks covered sample bounds, filter resets, mmap restrictions, and amplifier gain limits. No build or hardware tests were run during this review.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A["Speaker PCM write"] --> B["Volume gain"]
  B --> C["100 Hz high-pass"]
  C --> D["Peak clamp"]
  D --> E["Ceiling and thermal gain"]
  E --> F["Shift right by 7"]
  F --> G["ADMAC / AOP"]
  G --> H["Speaker amplifier"]

Loading

_{Reviews (5) · Last reviewed commit: "ASoC: apple: macaudio: enable the J873 s..." · Reviewed by Greptile}

[joshuaswarren]

New head 43f266453804e1843362c697c2dc4cec5ee09e5b: one commit added on top, ASoC: apple: t8140-aop-audio: limit the T8152 speaker with a thermal model (only sound/soc/apple/t8140_aop_audio.rs, +294/-5, with a KUnit suite of four tests). The earlier eleven commits are unchanged. Receipts: arm64 clang W=1, CONFIG_RUST=y, CONFIG_KUNIT=y, 16K pages: t8140_aop_audio.o, aop.o, macaudio.o, apple-dart.o, apple-admac.o, reset-apple-cio.o, mailbox.o and t8152-j873g.dtb rc 0, 0 warnings (the existing rust/kernel/dma_fence.rs dead-code warning only); the KUnit test symbols are in the object. NOT run: the KUnit suite (it needs a booted arm64 kernel with this tree), a boot, audio on hardware.

[joshuaswarren]

GREEN at 43f2664

Review of the one commit above 4b2bead (speaker thermal limiter, sound/soc/apple/t8140_aop_audio.rs only, +294/-5). The branch above that head is that commit alone; its patch-id equals the patch I reviewed earlier by hash, so what follows is that review. Algorithm: the model uses the vendor numbers recorded in the commit text (11.9 V per full scale, 4.6 ohm, coil 40 C/W with 3.4 s, magnet 60 C/W with 210 s, ambient 50 C, limit 140 C); I re-derived the constants: full-scale square power 11.9^2/4.6 = 30.78 W, sustainable (140-50)/(40+60) = 0.9 W, minimum gain 10 log10(0.98 * 0.9 / 30.78) = -15.43 dB, and the first-test ceiling 20 dB lower, amplitude 0.01693 (CEILING_M20_Q30 = 18174652 / 2^30), 0.2 V peak and about 9 mW as a square. I ran the author host test (66 checks) on the CT: all pass, including the ceiling check, the plant-equal and 25-percent-worse cases at the amplifier card cap, and the printed case that has no margin by design (a 25-percent-worse plant with the amplifier above the card cap). Integration: the limiter runs in the speaker copy op, the speaker front-end is created without MMAP flags (INTERLEAVED only), so every sample passes through the copy op and no stream can bypass it; the idle cooling runs in prepare; the state sits behind a spin lock; the arithmetic is fixed point and the commit says the kernel integer code reproduces the host model bit for bit as a plain Rust test crate with overflow checks. The 4 KUnit tests were not run (as the PR comment states). Not covered: the real speaker (resistance, thermal constants and the amplifier output per full scale are not measured, as the commit text says), a kernel boot, any audible test (the first audible test is a separate, human-present step with the build ceiling in place), and the Rust error paths beyond what the compile receipt shows. Merge needs a second non-author GREEN and CI.

[joshuaswarren]

GREEN at 43f2664

Review of the one commit above the previously reviewed head 4b2beadf ("limit the T8152 speaker with a thermal model", sound/soc/apple/t8140_aop_audio.rs only, +294 -5). The earlier 11 commits are unchanged; the compare against 4b2beadf shows this one commit. Four notes, none blocking.

Checked

• The integer model, recomputed. I wrote a Python model of spk_thermal that follows the Rust line by line (i64 semantics, arithmetic shifts, an assertion on every intermediate against the i64 range) and ran the four KUnit scenarios: ten seconds of a full-scale square wave then two seconds of silence, six times ten seconds, the 60 s idle cooling from 139 C, and the first block at the ceiling. Every expected constant in the tests is reproduced exactly (152358436004836, 102381555896261, 152358436004836, 102381555896261, 206396213; the post-silence state 130024328395229, 101932305864922, 135521886534109, 102381555896261, 226457860; the cooled 99824966751432 and 99097783336276; the ceiling word 18174652), and no intermediate leaves the i64 range: the largest is 2^61, with full 32-bit input words at both the production ceiling and the 0 dB ceiling. The six-times-ten-seconds square wave ends at 139.10 C, below 140 C.
• The constants. KCQ and KMQ equal 40 and 60 C/W times (11.9 V)^2 / 4.6 ohm in the model's units (315236 and 472854); AC and AM equal 1 - exp(-1 / (48000 * 3.4 s)) and 1 - exp(-1 / (48000 * 210 s)) in Q32 (26317 and 426); AC10 and AM10 are the same for 10 ms; REL is one 128-word block over 20 s; the 65 gain-table entries are 10^(-15.43 dB * k / 64 / 20) in Q30 (last entry 181746525); the temperature constants are 50, 120, 20 and 5 C times 2^40. The ceiling 18174652 is -20 dB below the sustained level (about 9 mW).
• Placement. The speaker front-end is opened without mmap (hw.info = SNDRV_PCM_INFO_INTERLEAVED for FE_SPKR), so every word reaches the DMA buffer through asoc_pcm_copy; the copy chunk is 512 bytes, so the 128-word array and words[..count] cannot overrun; the limiter state is behind a spinlock taken for at most 128 words; the T8140 path (has_lpai) is unchanged.
• Build and text. The commit text and the code comments carry no host names, home paths or internal names. The author's W=1 CONFIG_RUST=y CONFIG_KUNIT=y result is taken as stated; I did not build.

Notes

1. With the -20 dB ceiling the thermal part never acts. In my model a full-scale square wave at the production ceiling heats the coil to 74 C after two minutes and the gain stays at 1.0. The first audible tests therefore exercise only the ceiling; the thermal model is checked by the KUnit and host numbers until the ceiling is removed.
2. Any prepare resets the hysteresis. cool_idle runs on the prepare of every stream when there is no lpai, jack streams included, and cool() sets the hysteresis temperatures to the current ones. A jack stream opened while the speaker plays gives the speaker a small release step. Negligible at this ceiling.
3. Idle cooling is a loop. It runs up to 120,000 steps (20 minutes) in prepare with the spinlock held; that is well under a millisecond, but a closed form would remove the loop.
4. The model assumes one word per 1/48000 s. That is what the front-end enforces (48 kHz only, one word per frame as measured); a change of that rule changes the time constants.

Not checked

A kernel build, a KUnit run (the numbers above are reproduced by my model instead), the speaker, and anything audible.

[joshuaswarren]

Status at head 43f266453804e1843362c697c2dc4cec5ee09e5b: two independent reviews GREEN at this exact head, Greptile pass with no comments on the earlier head, arm64 clang W=1 CONFIG_RUST=y CONFIG_KUNIT=y compile of the touched objects rc 0 with 0 warnings, device-tree schema check 0 messages. Known edges and the other reviewer notes are in the PR text and the review comments. Not done: a boot on hardware, running the KUnit suite, CI. Needs a maintainer's CI approval and merge. This PR needs #197 first for the apple,t8152-acio-mailbox binding text. Not needed for the release candidate.

[joshuaswarren]

KUnit run at head 43f266453804e1843362c697c2dc4cec5ee09e5b: arm64 Image (CONFIG_RUST=y, CONFIG_KUNIT=y, 16K pages, audio driver built in) booted under qemu-system-aarch64 (virt, TCG, no root filesystem). All T8140 audio suites pass: names 2/2, parser 5/5, thermal 4/4, speaker 5/5, lid 7/7; also apple_aop_ring 17/17. 0 fail, 0 skip. Image sha256 354ffb7cb0437b6b416cae11f955c34f7ba234b2afe73ef9e3324cc9ad0ab778. This is emulation: it checks the test logic and constants, not behaviour on the M6 hardware.

[joshuaswarren]

New head 68c6401203afb55e7a60706fbca7c55f57350422: one commit on top of the limiter, ASoC: apple: t8140-aop-audio: give the T8152 speaker plain PCM.

A microphone run on the M6 found that the limiter's output was still written as float32 bit patterns, which the T8152 amplifier plays as 32-bit integers. The commit writes the limited word as a plain integer on the T8152. The T8140 path is unchanged. It is 3 lines in one file.

Two reviews are GREEN at this head. The whole stack compiles with W=1 and Rust on, and the audio objects have 0 warnings. I did not rerun the KUnit suite, since no tested function changed. I have NOT measured the sound with this change; that needs the next run on the Mac mini.

[joshuaswarren]

New head 444f88a8eff09c01aa345b41654005b57f48307e: one more commit, ASoC: apple: macaudio: give the J873 speaker link a rising RX edge.

The J873 speaker amplifier stays in software shutdown with the card's default serial format. The commit sets a rising receive edge on the speaker link only. It adds 10 lines to macaudio.c, and the headphone link and every other machine keep the default.

Two reviews are GREEN at this head. The whole stack compiles with W=1 and Rust on, and the audio objects have 0 warnings. On the M6, with no register writes from user space, the quietest step is 39 dB above the room and the next step up is 7 dB louder. I have NOT measured the absolute level, and the loudest step is distorted and unexplained. The change does nothing until enable_speakers is true, and it stays false in this series.

[joshuaswarren]

This series does not turn the internal speaker on. The speaker stays off in the J873 config, and the link is skipped unless the module parameter please_blow_up_my_speakers is set. The speaker commits only matter once it is turned on. That waits for open items: a gain cut at high levels that is unexplained, and the absolute level, which is not measured.

[joshuaswarren]

Hold: the T8152 speaker output scale is wrong. The link carries 24-bit samples, so words must be shifted to a full scale of 2^23. The thermal limiter and the integer output in this series assume 2^30, so the words are 128 times too large. Do not merge until the fix lands and is re-measured. The speaker stays disabled in the configuration.

[joshuaswarren]

Review at 444f88a: my GREENs on the limiter commits (24, 25) do not stand as a safety statement.

A hardware peak test on the T8152 speaker path found that the amplifier path wraps at a word value of 2^23, not at 2^30. The limiter model in these commits assumes the I2S full scale is the word scale; every word written in the earlier tone tests was therefore about 128 times (42 dB) larger than the model believed. My GREENs checked the code against its stated model (the arithmetic, the tests, the bounds), not that scale, which was never measured, and the PR text already says the speaker stays disabled. Until a corrected scale lands (the author is preparing it), nothing in this PR should be read as "protects the speaker" or "safe level", the first audible test must not be run with the current module, and the speaker stays off for users. I will review the corrected scale by its exact head.

[joshuaswarren]

GREEN at 394a040

Commit 27 (394a040, the only commit over 444f88a): one file, one hunk. After the limiter runs on the block, the copy loop writes the limited i32 word shifted right by 7 (arithmetic, sign kept), so the limiter's 2^30 full scale becomes 2^23 on the wire and the amplifier's 24-bit wrap cannot be reached by a word the limiter lets through. The limiter still runs before the shift. The code comment and the commit message say what was measured (wrap at 2^23 with the unscaled words, the three peaks where the tone fell) and what was not (the sound with this change, the absolute level, the amplifier output voltage), and that the speaker stays disabled. This does not remove the hold in the PR body: nothing here is a statement that the speaker is safe; the shifted module still has to be measured at the wire.

Not built by me (the author reports 27 of 27 check-commits and a clean clang W=1 build of the two files); not run.

[joshuaswarren]

GREEN at 394a040, as a code review of commit 27 only. This is not a statement that the speaker is safe to run.

Checked:

• The delta from 444f88a is one commit, one file, one hunk (t8140_aop_audio.rs, +6 -3). Its added and removed lines equal the reviewed patch exactly.
• The change: after the limiter has processed the block, the copy path writes *w >> 7 of each limited word. The words are i32, so the shift is arithmetic and the sign is kept, and the limiter's 2^30 full scale becomes 2^23 on the wire, where the amplifier's 24-bit field wraps. The limiter is still applied before the shift.
• I also compared the built modules (before and after): exactly one function differs, asoc_pcm_copy, by one added arithmetic shift of seven.
• The commit message states what was measured (the wrap at 2^23 with the unscaled words, from room-microphone levels) and what was not (the sound with this change, the absolute level, the amplifier output voltage), and that the speaker stays disabled in the J873 configuration. It has no safety wording.

What this GREEN does not cover: the scale is inferred from the observed wrap, not measured at the speaker or at the wire. The earlier commits 24 and 25 still carry the 2^30 wording in their messages (the PR body says so). The hold stays until the shifted module is measured, and nothing in this series should be read as protecting the speaker.

Not checked: a build of this head, and any run on hardware.

[joshuaswarren]

GREEN at e9d9aa0

The three commits over 394a040 (8108707 high-pass, df7f942 ceiling, e9d9aa0 enable) change exactly the lines of the reviewed patches: I compared the changed lines of each commit with its patch and they are identical (same checksum for all three). What I reviewed in those patches: the Q28 Butterworth coefficients recomputed independently and matching, no 64-bit overflow in the worst case, the clamped-output and unclamped-feedback fix (the latch I reproduced with the old code is gone), the 83.8M ceiling as a scale of full scale, and for the enable commit: safe_vol 0 only drops the volume-lock controls, the amplifier gain stays capped at 15 by macaudio_set_speaker(), and the speaker front end has no mmap so the limiter cannot be bypassed. Not checked: a build or a run.

[joshuaswarren]

GREEN at e9d9aa0, as a code review of the series. This is not a statement that the speaker is safe to run.

Checked:

• Commits up to 394a040 are the head I reviewed before (GREEN at 394a040).
• The three new commits (high-pass at 100 Hz, speaker ceiling, enable the J873 speaker) have code bodies identical, line for line, to the three patch files I read earlier (sha256 prefixes 8a4b3cda, fa207604, 89b11b74): 143, 24 and 23 changed lines, 0 differences.
• The last commit sets enable_speakers = true and safe_vol = 0, plus comments in the dtsi and in macaudio.c. It adds no other path.

Not checked: any run on the device. The absolute output level and the cone excursion are unmeasured, so the first run must use a low level with a person at the machine.

[joshuaswarren]

Note on the red checks of this PR (ci / package, ci / host tests (tools) and the ci / result that sums them). The audio commits are not the cause.

• ci / package fails in modpost: __tracepoint_dwc3_readl, __traceiter_dwc3_readl and the two dwc3_writel symbols are undefined for drivers/usb/dwc3/dwc3-apple.ko. The T8152 USB2 controller commit (48cbf4a3, in the lower layers of this branch) makes the Apple glue driver call dwc3_readl and dwc3_writel from io.h. Those inline functions fire the dwc3 core tracepoints, and the core module does not export them.
• ci / host tests (tools) fails on apple-dcp/run.sh, case 31 dwc3-teardown. That case reads the teardown functions of dwc3-apple.c, so it is the same driver. It passes on the base branch.

The commits of this PR that I added (27 to 30) change t8140_aop_audio.rs, macaudio.c and the J873 audio device tree include. The checks for compile, device trees, KUnit and commit format pass.

I think the module fix is one line each for the two tracepoints in drivers/usb/dwc3/trace.c (EXPORT_TRACEPOINT_SYMBOL_GPL). I have not built it. If you want, I will open it as a separate PR against the lower layer, with the build result.

[joshuaswarren]

Note

🤖 GPT-6.1 Terra responding on behalf of joshuaswarren

Replaced by #359, which adds the missing Signed-off-by trailers without rewriting this branch.