Skip to content

arm64: dts: apple: t8142-j704: PMP always-on, trackpad actuator, MCC - #1847

Draft
dmitzsaz wants to merge 3 commits into
aurora-silicon:integration/aurorafrom
dmitzsaz:pr/j704-dt-followup
Draft

dmitzsaz wants to merge 3 commits into
aurora-silicon:integration/aurorafrom
dmitzsaz:pr/j704-dt-followup

Conversation

@dmitzsaz

Copy link
Copy Markdown
Contributor

Three DT-only follow-ups to #1231 that bring the J704 in line with what the M5 Airs got this week. The drivers are already in integration/aurora; each change mirrors the Air commit for the same die.

Tested on: the PMP change on J704 with the same property on an earlier tree (one clean s2idle and wake, display, Wi-Fi and Bluetooth back). The actuator and the MCC are not tested on the J704 yet; the PR stays a draft until they are.

scripts/dtc/apple-dt-check.sh against a34d26c: 137 DTBs, no new warnings and none removed. The J704 keeps only the disp0_piodma alias warning it already had.

Prepared with Claude Opus 5.5.

The PMP firmware keeps running DCS and fabric DVFS while the system is
up. The J704 ps_pmp domain is not marked always-on, so genpd gates it
at s2idle and the PMP takes an SError in its own firmware. The J813
got the same fix in ceaab64 ("arm64: dts: apple: t8142-j813: Keep
the PMP power domain on"); T8132 and T8140 mark ps_pmp always-on too.

Tested on J704 with the same property on an earlier tree: the first
s2idle no longer crashes the PMP, and the machine wakes with the
display, Wi-Fi and Bluetooth back.

Signed-off-by: Dmitriy <noc@pimk.in>
Assisted-by: LLM
The J704's MTP firmware offers the trackpad's haptic engine as
interface 4 "actuator", as on the M5 MacBook Airs:

  Initializing comm interface <4- "actuator">

Without a node for it, dockchannel-hid ignores the interface and the
click stays firmware-driven, with no Force Touch stage. Describe it as
d2db3f8 ("arm64: dts: apple: t8142-j813-j815: Add the trackpad
actuator") does for the Airs.

Signed-off-by: Dmitriy <noc@pimk.in>
Assisted-by: LLM
The J704 ADT's mcc,t8142 node matches the J815's: one AMCC
(amcc-count 1, amcc-reg-idx 1), amcc-plane-enable-mask 0xf, reg[1] at
0x220000000 spanning the four planes 0x40000 apart. Enable the
t8142.dtsi node from 167e403 ("arm64: dts: apple: t8142: Add the
MCC (system level cache) node"). The boot loader turns allocation on;
the driver checks it and flushes the cache on shutdown.

Signed-off-by: Dmitriy <noc@pimk.in>
Assisted-by: LLM
@RyanTheTide

Copy link
Copy Markdown
Contributor

Blockers: 2.

  1. Commits e30a7ce, b484d69, a54463e: each ends with an Assisted-by: LLM trailer. Commits should carry only the human Signed-off-by; drop the trailer from the commits (amend onto a new branch, no force push) and put one Assisted-by: LLM (<Vendor>:<Model>) line as the last line of the PR body instead of "Prepared with Claude Opus 5.5."
  2. No J704 run on this head. The PMP change ran with "the same property on an earlier tree". The actuator (t8142-j704-peripherals.dtsi:262) and the MCC enable (t8142-j704.dts:318) have not run on a J704 at all. The DT itself matches what the J813/J815 carry on integration/aurora (actuator { }; in t8142-j813-j815.dtsi:342, &mcc { status = "okay"; } in t8142-j813.dts:303), and the ps_pmp apple,always-on mirrors ceaab64. So the open question is only whether each one behaves on this machine. The MCC enable matters most, because it switches on cache maintenance in a driver that has only run on the Airs.

HW row: J704, integration/aurora head + these 3 commits on the default command line: the MCC driver binds and reports the loader-enabled ways (or refuses cleanly), dockchannel-hid brings up interface 4 "actuator" and a Force Touch press reports the second stage, 3 s2idle cycles with no PMP SError and display/Wi-Fi/BT back, a clean shutdown (MCC flush path); 0 AVC, 0 failed units, no new kernel errors versus the same head without the series.

Assisted-by: LLM (Claude:Opus 5.5)

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.

2 participants