Skip to content

device/tegu NOTES: every testing row on a published-style kernel, and the bring-up leftovers - #244

Closed
W4HO-ham wants to merge 10 commits into
derekross:mainfrom
W4HO-ham:tegu-notes-20261101
Closed

W4HO-ham wants to merge 10 commits into
derekross:mainfrom
W4HO-ham:tegu-notes-20261101

Conversation

@W4HO-ham

@W4HO-ham W4HO-ham commented Nov 1, 2026

Copy link
Copy Markdown
Contributor

Two dated entries in the 9a's notes, for the record behind today's pull requests.

Every testing row on a published-style kernel. Each was a boot of the named head, built with defconfig + omamobile.config (+ #227) + olddefconfig, with the daily kernel pushed back afterwards:

It also covers the config gaps #227 fixes, and the two driver findings: the clock-out optional-parent check, and the zuma PMU allowlist entry the camera PHY needs.

Leftovers found on the palmtop from bring-up.

Docs only. No identifiers (checked for MAC addresses and the network's name); tools/pr-check passes.

@derekross

Copy link
Copy Markdown
Owner

Thank you for this. Running every testing row against a build from the shared fragment is exactly what those rows needed, and the config gaps it turned up (#227) are a real catch. Two things before it merges:

  1. Each row names a pull request but not the build that ran. Could you add the uname -r for each boot, or the kernel commit for each row? The device table will cite these lines, and a pull request's head moves.
  2. The last bullet records slot B being marked again. I've asked about that devinfo write on AGENTS: on the Pixel 9a a plain fastboot flash of slot B clears its successful flag too #224. Please also put the evidence from that PR here (the date, the commands, slot-successful:b and slot-retry-count:b before and after), so AGENTS can link to it.

Small things: a line on how NetworkManager replaced iwd would help the next person who finds it that way. This and #180 and #202 all add their section at the top of NOTES; I'm merging #180 now, so this will need a quick rebase.

… the bring-up leftovers found on the palmtop

Each testing row was booted on the palmtop from its pull request's head with the project's config, and the entry links the results. It also records the kernel config gaps, the stale plugin links and masks left from bring-up, and that a plain fastboot flash of slot B clears its successful flag.
…id not

The PERIC0 GPIO clock gate that hung the SoC under modem traffic is not the sleep hang.
…lease, how NetworkManager replaced iwd, and the slot B flash and devinfo evidence

Each row names the kernel release of its boot, the NetworkManager line says how the switch was made, and a new entry records the five fastboot flashes that cleared slot B's mark and the one-byte devinfo write that set it again.
@W4HO-ham

W4HO-ham commented Nov 1, 2026

Copy link
Copy Markdown
Contributor Author

Done, rebased onto main with #180 in (these two entries above the 10-30 morning):

  1. Each testing row now gives its boot's uname -r (7.3.0-rc6-omamobile-push1-g<commit>; the hash is the built tree's commit).
  2. A new entry, "2026-10-31: a fastboot flash of slot B cleared its successful mark, and B was marked again", has the five commands, the getvar readings before and after, and the devinfo write in full. AGENTS: on the Pixel 9a a plain fastboot flash of slot B clears its successful flag too #224's AGENTS line links to it. The details of the write are in my answer on AGENTS: on the Pixel 9a a plain fastboot flash of slot B clears its successful flag too #224.

The NetworkManager line now says how the switch was made: the mask removed, iwd disabled, NetworkManager enabled, all under a two-minute dead-man.

The reset by the buttons kept no ramoops, and the journal ends at the suspend entry, so the next run needs a watchdog reset to keep a record.
@W4HO-ham
W4HO-ham force-pushed the tegu-notes-20261101 branch from a7d3f18 to 80f787b Compare November 1, 2026 14:23
…, so it can't record the hang

Armed, three sleeps of three reset with ACPM no longer answering after resume; disarmed, two of two woke.
pm_wakeup_irq is the S2MPG14's line, raised by a source with no mapped child 0.6 s after entry; the RTC alarm itself fires on time. The wake counts so far may describe sleeps of about a second.
…top sleeps

With an alarm set, alarm 0 latches within 60 ms whatever its due time; with none, a power key release does. So every sleep has probably lasted about a second.
…alarm

One 30 s sleep with cpuidle state1 disabled on all CPUs woke on the real alarm 29 s in, so the false PMIC events come with the CPUs' power-down.
@derekross

Copy link
Copy Markdown
Owner

Thanks for this round. Every testing row now names its boot's release, the NetworkManager line says exactly how the switch was made (and under a dead-man), and the flash-and-mark entry gives the getvar readings and the devinfo write in full. I compared each tested commit with its kernel pull request's current head, and for #161, #162, #105, #164 and #165 the trees are identical, so the rows still describe what will merge. For #168 the only difference is the brcmfmac test commit, which the row already names.

A few things, most of which I can do at merge:

  1. The approval's date. The new entry says the write was "made under the 10-21 approval", but the approval on record is 2026-10-22 (the "Approved B-success operation after layout review, 2026-10-22" section, and AGENTS). I'll correct it to 10-22 unless you'd rather.
  2. The title. Since my last comment the PR has gained four SLEEP-PLAN commits (S1 on the clock-fix kernel, the empty pstore, the watchdog-through-s2idle try, the one-second PMIC wakes). They're good results, but the title only says NOTES. I'll retitle it to "device/tegu NOTES, SLEEP-PLAN: …" and write the merged message by hand, since the squash would otherwise carry all six commit messages.
  3. The watchdog module. The SLEEP-PLAN paragraph describes tegu-wdt-s2idle, a bring-up module that re-armed the cluster 0 watchdog and its PMU reset path in the noirq phase, "in the owner's logs". It ended in three resets, and it wrote PMU state from a module nobody else can read. Could you push its source to a branch on your kernel fork and link the commit from that paragraph? The plan already says it isn't a way to catch the hang. If anything like it runs again, it should get a runbook entry first, with the Stop and the UFS check after each reset.
  4. The one-second wakes. The finding that S1's woken sleeps may have lasted about a second matters beyond this plan. AGENTS' 9a sleep gotcha counts "35 short and seven longer systemd sleeps woke", and some of those may be the same short sleeps. No change needed here, but once you've checked the 30-minute sleeps' real lengths, please send the AGENTS correction as its own small PR.

On the order: this goes first. #224 and #243 link to its anchors, #248 cites its rows, and #202 also adds its section at the top of the notes. I'll resolve that one when I merge #202, keeping both sections newest first.

@derekross derekross closed this in bb9fb69 Nov 1, 2026
derekross pushed a commit that referenced this pull request Nov 1, 2026
…y default, fingerprint opt-in, GNSS to testing (#248)

With omamobile-device-tegu 0.2.3 the charge limit, the AoC and NFC start
at boot from the package, so the Pixel 9a's Charging, AoC, Speakers/mics,
Sensors and NFC cells move to "works" (on by default), and Fingerprint to
"works (opt-in)", as on the 7 Pro. Each was verified on a boot of its
kernel pull request's head built from the shared config fragment; the
record is #244's entry, which gives each boot's uname -r. The Boot cell
records the 10-31 unmark and re-mark, and GNSS moves to testing with the
modem started by hand (#114, #202, kernel #172). The kernel parts are in
review, as each cell says.

Closes #248

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
… kernel, the slot B flash and devinfo records, and S1 on the clock-fix kernel (derekross#244)

The Pixel 9a notes give every testing row its kernel release and record the fastboot flash that cleared slot B's successful mark and the devinfo write made during the hold; SLEEP-PLAN adds S1 on the clock-fix kernel, the empty pstore, the watchdog-through-s2idle try and the one-second PMIC wakes (W4HO-ham).

Closes derekross#244

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…y default, fingerprint opt-in, GNSS to testing (derekross#248)

With omamobile-device-tegu 0.2.3 the charge limit, the AoC and NFC start
at boot from the package, so the Pixel 9a's Charging, AoC, Speakers/mics,
Sensors and NFC cells move to "works" (on by default), and Fingerprint to
"works (opt-in)", as on the 7 Pro. Each was verified on a boot of its
kernel pull request's head built from the shared config fragment; the
record is derekross#244's entry, which gives each boot's uname -r. The Boot cell
records the 10-31 unmark and re-mark, and GNSS moves to testing with the
modem started by hand (derekross#114, derekross#202, kernel derekross#172). The kernel parts are in
review, as each cell says.

Closes derekross#248

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gitColinX added a commit to gitColinX/omarchy-mobile that referenced this pull request Nov 16, 2026
…ekross#243 and derekross#244

- NOTES no longer calls SUSPEND-PLAN a draft (derekross#802).
- The ROADMAP Suspend row says mem_sleep read "s2idle [deep]" before
  kernel derekross#244 and reads "[s2idle] deep" on the published kernel (derekross#802).
- The ROADMAP Time row no longer says "No RTC": the S2MPG14's RTC is
  rtc0 since kernel derekross#243 (asked for on that PR).
- The 2026-11-15 NOTES entry records /proc/cmdline's
  mem_sleep_default=s2idle with mem_sleep (asked for on kernel derekross#244),
  and the RTC's probe lines, wakealarm and idle interrupt count (asked
  for on kernel derekross#243).

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
derekross pushed a commit that referenced this pull request Nov 18, 2026
…efore H2; NOTES and ROADMAP leftovers

H0b (the Wi-Fi and PCIe unload and reload, no sleep) and H1 (`pm_test` freezer, devices and platform through `systemctl suspend`: three successes, no failures) ran attended on Colin's 8 Pro on the published `omamobile` at `82f31d3a7` with device-husky 0.1.13; touch, the display and a played haptic effect after H1 are carried into H2, and H2 to H4 have not run. The plan now says that the UFS `dme-set: attr-id 0x4021` line at resume is expected and not a stop, lists the four brcmfmac noise lines, names the maintainer's go-ahead per stage and device-husky 0.1.14 or later for H2 to H4, and records the sleep configuration in effect. NOTES and ROADMAP pick up the leftovers from #802 and the RTC and `mem_sleep_default=s2idle` facts from kernel #243 and #244.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
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