Skip to content

soc: apple: report T6050 wake-controller snapshots - #2137

Draft
DjDeveloperr wants to merge 1 commit into
integration/aurorafrom
djdev/t6050-wake-register-reporting
Draft

DjDeveloperr wants to merge 1 commit into
integration/aurorafrom
djdev/t6050-wake-register-reporting

Conversation

@DjDeveloperr

@DjDeveloperr DjDeveloperr commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

T6050 owns its wake handling, so the common pending-register read is skipped and the common nine-word buffer cannot represent AP290/GFX321. Retain ten AP pending/mask words and one GFX word at the first region wake callback, before PMU/SPMI acknowledgement, then report the saved data after cleanup.

Keep this observation separate from common wake-policy flags. Invalidate it at every begin/capture, require a real returned reset before reporting, and mark missing/unarmed mappings unavailable. No wake-source, mask, reset, GPIO, firmware-command or cleanup-decision change.

Validation:

  • GCC/Clang host runner compiles actual source callback slices. All 352 positions, mask interpretation, invalid/refused lifecycle, acknowledgement ordering, parser refusal and nine compiled behavioral controls per compiler pass.
  • Full Image/modules/DT build and a complete returning-device/recovery test passed in controlled native candidate cb5a0c9541e6, containing this identical wake header plus separate cached PD/PHY logs. The exact public-head native build is not claimed.
  • The first physical RTC row is now recovered. PM write returned successfully on the same boot; the native callback reported AP76 only and GFX0, with masks retained. The boot-object resume counter advanced 0→1 and fresh MBSW8 was zero. RTC/monotonic brackets estimate approximately 41.6 seconds asleep (approximately ±1 second). This supplies one real native early-GFX/callback execution witness.
  • The overall row FAILED N1 recovery: deferred teardown refused with DMA resources retained (-16, original 0); the owner observed a responsive desktop without Wi-Fi. The N1 follow-up is wifi: centauri: recover after DMA was released before physical reset #2144. Device recovery and long-hold readiness are not established by the wake report.

The owner authorized one Left Front cable reboot after reporting the desktop state. Original intg9/prev intg8, all system/ESP/layout fingerprints and default GPU checks were then verified restored; ownership released at 09:59:18 EST. The original failed result and unsafe evidence remain preserved. Manual recovery does not turn the row into a clean qualification.

Draft pending review. Snapshots are sequential early-wake levels/readback, not atomic final-commit state or exclusive wake-event cause. This adds no slpw/SSH wire timing or raw GPIO observation and does not enable ordinary suspend. All 7 public-head checks passed on c50eeb27f319c7037a750cbebfc87117c66d1e58: Host tests run44040894867 and SEP source checks run44040894825 succeeded. No source change was made for this evidence update.

Retain the AP/GFX pending and mask words before PMU acknowledgement,
then report only a real reset return. Keep observations separate from
common wake policy and invalidate them at each new attempt.

Exercise actual source callbacks and all352 positions in host tests.
@RyanTheTide

Copy link
Copy Markdown
Contributor

Blockers: 2. I recommend closing this PR.

  1. Superseded. soc: apple: sleep: T6050: name the wake source after a wake #2120 (7e63437, "soc: apple: sleep: T6050: name the wake source after a wake") is on integration/aurora. t6050_wake_source_read() (drivers/soc/apple/sleep-t6050.h at a84637d, called at the start of the wake) already latches the same ten AP status words at +0x34 and the GFX word at +0x10 before anything is acknowledged. t6050_wake_source_report() and t6050_wake_events_report() then print them by source name, and the raw words go to the debug log. This PR adds a second reader of the same registers, with its own validity flag, through t6050_wake_snapshot()/t6050_report_wake_snapshot(). The branch also conflicts with its base. The only thing soc: apple: sleep: T6050: name the wake source after a wake #2120 does not cover is the mask words at return. If those are still wanted, add them to t6050_wake_source_read() as a small follow-up on current integration/aurora.
  2. Commit c50eeb2 has no Signed-off-by. Any follow-up needs a human Signed-off-by.

HW row: none for this PR. A mask-word follow-up needs one J714s or J716c s2idle/deep wake on its head: the events line still matches, the mask words are logged once, and there are 0 failed units and no new kernel errors.

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