Repository navigation
Conversation
Both modules (the watchdog through s2idle and the PMIC wake read) are now in a not-for-merge commit on the owner's kernel fork, as asked on derekross#244.
The one-second sleeps were dmesg deltas, and the kernel log's clock stops while the SoC is suspended. The RTC shows full-length sleeps, so the core power-down hook (derekross#255) is withdrawn, and S1's wake counts stand.
|
I've added a correction to this PR. The "sleeps were short" paragraphs that merged with #244 were my measurement error: I took each sleep's length from dmesg, whose clock stops while the SoC is suspended. The RTC shows full-length sleeps (details on #255, which I've closed). The paragraphs are replaced by one correction paragraph that says what was wrong, what the PMIC readings really show, and how to measure a sleep (the RTC). So S1's wake counts and the AGENTS gotcha stand, and no AGENTS PR is needed. |
…on, all woke, timed by the RTC
|
Thank you for this, and for the correction. Linking the modules' source was what I asked for on #244. I read e761ba793: both modules carry their SPDX line, and the README says the watchdog module isn't a way to catch the hang and needs a runbook entry before it runs again. The correction is the more useful part. A dmesg delta across s2idle leaves out the sleep, because the kernel's clock stops while the SoC is suspended. I took the one-second figure at face value too, so thank you for checking it against the RTC and saying plainly what changed. It's right that no AGENTS change is needed. Two small things in the new paragraph on the thirty 120 s sleeps:
Once those two are fixed, this is ready to merge. |
…h, and three woke early on the PMIC line The count said 28 and added up to 31. IRQ 56 is the S2MPG14's line, shared with other PMIC events, and the series kept no alarm count, so the 64, 106 and 112 s sleeps are early PMIC wakes, not alarm wakes.
|
Both fixed in 497085e. The count was off by one: 27 sleeps lasted 118-123 s, not 28, so with the 106, 112 and 64 s ones it's thirty. The series recorded only the wake IRQ, not the |
|
Landed on main as dab156a (squashed, authored by you). My review notes from before the merge, for the record: Thanks, that's exactly it. 27 + 3 is thirty now, and saying plainly that the series recorded only the wake IRQ, so the three short sleeps count as early PMIC wakes rather than alarm wakes, is the honest reading. I've nothing else on this one and will merge it. (The red "Tests" job on this run is a timing-sensitive check in the audio package's speaker watchdog, which this PR doesn't touch. I'll re-run it before merging.) |
#315) The test set the "measured" 33 C 0.2 s after the watchdog started and expected its baseline at 0.6 s. On a loaded machine the step sometimes landed after the baseline was taken, so the watchdog saw 25 -> 41 C instead of 33 -> 41 C and the test failed for timing alone. It failed three times on 2026-11-01 (pull requests #256, #313, and main), and it aborted every nightly development publish, which builds and tests every package first: pkgs.omamobile.org/development stayed at omamobile-shell 0.2.45. The step now lands at once, the baseline is taken at 1.0 s and the rise comes at 2.0 s, a second's margin either side. No change to the watchdog itself. Claude-Session: https://claude.ai/code/session_01BeEcdVeSfyLCbijF8XQF1d Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…rect the one-second sleeps (derekross#256) Links the source of the S1 bring-up modules (tegu-wdt-s2idle and tegu-pmic-wakeread, one commit on the contributor's kernel fork with a README saying the watchdog module needs a runbook entry before it runs again), and corrects the "one-second sleeps": those were dmesg time differences, and the kernel log's clock stops while the SoC is suspended. Thirty 120 s sleeps timed by the RTC: 27 lasted 118-123 s and three woke early on the PMIC line (IRQ 56, shared with the alarm), counted as early PMIC wakes. Closes derekross#256 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
derekross#315) The test set the "measured" 33 C 0.2 s after the watchdog started and expected its baseline at 0.6 s. On a loaded machine the step sometimes landed after the baseline was taken, so the watchdog saw 25 -> 41 C instead of 33 -> 41 C and the test failed for timing alone. It failed three times on 2026-11-01 (pull requests derekross#256, derekross#313, and main), and it aborted every nightly development publish, which builds and tests every package first: pkgs.omamobile.org/development stayed at omamobile-shell 0.2.45. The step now lands at once, the baseline is taken at 1.0 s and the rise comes at 2.0 s, a second's margin either side. No change to the watchdog itself. Claude-Session: https://claude.ai/code/session_01BeEcdVeSfyLCbijF8XQF1d Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
As you asked on #244: the source of
tegu-wdt-s2idleandtegu-pmic-wakereadis in one commit on my kernel fork, e761ba793 ontegu-sleep-bringup-modules, underbringup/tegu-sleep/with a README. It says the watchdog module isn't a way to catch the hang, and that anything like it needs a runbook entry first. Both SLEEP-PLAN paragraphs now link it. Neither module is on the palmtop now; they were loaded from/tmpand are gone after the resets.The AGENTS correction about the short sleeps will come as its own PR once I've measured the real length of the earlier 30-minute sleeps from the palmtop's journal.
tools/pr-checkpasses.