Skip to content

soc: apple: rtkit: don't warn twice about a crash on the crashlog endpoint - #2496

Merged
DjDeveloperr merged 1 commit into
integration/aurorafrom
fix/rtkit-crashlog-report-wording
Nov 20, 2026
Merged

DjDeveloperr merged 1 commit into
integration/aurorafrom
fix/rtkit-crashlog-report-wording

Conversation

@RyanTheTide

Copy link
Copy Markdown
Contributor

Problem. The AOP leaves the RTKit crashlog endpoint (0x01) out of its endpoint map, so it is never started. When the AOP firmware crashes it still reports on that endpoint, and the receive path first warns RTKit: Message to undiscovered endpoint 0x01, then apple_rtkit_crashlog_rx() reports the crash. The warning says "to" for a message from the co-processor. On the J700 (Neo) this line was read as the AP sending to an endpoint before it was discovered across sleep, when it is the AOP's own crash report.

Change. The generic undiscovered-endpoint warning now skips the crashlog endpoint, because apple_rtkit_crashlog_rx() prints a line for every message there ("co-processor has crashed: ...", "crashed without a usable crashlog buffer" or "Unknown crashlog message"). For the other endpoints it now says "from". The log changes; nothing else does.

Testing. Builds clean with W=1 (drivers/soc/apple/rtkit.o, fleet arm64 config, LLVM). checkpatch is clean. No tool in the tree greps for the old string.

Still to run. This can't be checked on hardware on demand: the AOP crash is intermittent (0 of 17 boots / 51 deep sleeps on the Neo in a dedicated repro run).

Context: the J700 AOP crash this came from
  • Seen twice on 11-20, both at the first deep sleep of a boot: on integration 26294e8, then on a 0abc2f3-based test kernel. The sequence is apple-sleep: 10 coprocessors up and stopped -> apple_aop: RTKit: Message to undiscovered endpoint 0x01 -> co-processor has crashed: crash log buffer at 0x10040054000 not mapped (-2) -> AOP firmware crashed. Later sleeps print firmware crashed: sleeping without it.
  • Interrupts are off from syscore_suspend to syscore_resume, so the report is delivered when they come back at the wake. The AOP died during the sleep or its exit, not on a message the AP sent.
  • Dedicated repro (11-20 10:30-11:21): 13 boots of 26294e8 and 4 of production 7.3.0-40, 3 deep S2R each with an apple-pmu-rtc wake: 0 crashes in 51 sleeps. Overall that day: 2 of 28 boots, on post-6f618bcc kernels only (2/21 vs 0/7). That split is not significant (Fisher p ~ 0.56).
  • No AOP code in 6f618bc..26294e8 runs on the J700. aop_las (iio: aop_las: stop the lid angle reports for system sleep #2223) is not bound there; the J700 AOP binds only apple_aop, iio_aop_als and snd_soc_apple_aop. The 27.2 __EDATA/ED0p change (60f279c) leaves the bootargs as before when there is no extra region, which is the case on the J700.

Assisted-by: LLM (Claude:Opus 5.5)

…point

The AOP leaves the crashlog endpoint out of its endpoint map, so it is
never started ("crashlog endpoint not in the endpoint map" at every
boot). When the AOP firmware crashes it still reports on that endpoint,
and the receive path first warns

  RTKit: Message to undiscovered endpoint 0x01

before apple_rtkit_crashlog_rx() reports the crash itself. The warning
says "to", but it is a message from the co-processor, and read next to a
sleep commit it looks like the AP sent something to an endpoint it never
discovered. On the J700 it led to exactly that misreading of an
intermittent AOP crash during deep sleep.

Leave the crashlog endpoint out of that warning, as its own handler
names every message there, and say "from" for the endpoints that remain.

Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>

@DjDeveloperr DjDeveloperr left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks right. The crashlog endpoint still goes through apple_rtkit_crashlog_rx(), which already reports the crash for firmware that leaves the endpoint out of its map, so dropping the second warning loses nothing, and "from" is the accurate wording for the remaining case. CI 7/7 on this head.

@DjDeveloperr
DjDeveloperr merged commit 2c35f50 into integration/aurora Nov 20, 2026
7 checks passed
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