Repository navigation
AGENTS, device/tegu: a fallback to slot A on the 9a costs the Linux root - #125
Conversation
The 9a lost its root twice: after an omamobile flash, slot B stays unmarked, every boot (and the initramfs USB safety net) burns an ABL retry, and at zero ABL boots stock Android, which reformats userdata. Mark B successful on the first good boot after a flash.
|
Thanks for writing this up, @W4HO-ham, and I'm sorry the palmtop lost its root twice. You're right that this belongs in AGENTS.md: the rest of our docs treat "slot B runs out of retries, ABL boots stock on slot A" as the safe way back (CONTRIBUTING §3.4, "Slot fallback"), and on the 9a it hasn't been safe. The mechanism you describe matches tegu's notes: the retry byte goes down by one on every ordinary unmarked boot (the 0x34 table in NOTES), I can't merge the line as it is yet, though. A safety rule in AGENTS.md is something every agent and contributor follows word for word, so it has to point at things they can read, and it has to agree with the approvals and records we already have. Four things: 1. Both references point at private files. 2. The two losses need a dated record in tegu's notes. At the moment nothing on 3. "Stock reformats and encrypts userdata" disagrees with your own records. #100's MODEM-PLAN correction (2026-10-27, night) says that after the 10-26 fallback "stock Android on slot A can't mount its data and stops in recovery", and that "the factory reset it would need wipes the Arch install". That is the behaviour I'd expect: Android's recovery offers "Try again" or "Factory data reset" and doesn't wipe on its own. The difference changes the rule. If stock stops at a recovery prompt, the gotcha should say never choose Factory data reset; power off, get to fastboot with Volume Down + Power, and 4. The fix it prescribes is a One more thing, about wording. "Never boot an untested image while B is unmarked" can't be followed together with "after every Longer term, I think the proper fix is a boot-success marker in tegu's device package, the same mechanism cheetah uses and the one write CONTRIBUTING already allows, now that the 3.15 layout is agreed. Then nobody has to remember a manual step after every flash. That's a separate PR and my call to make, so please don't fold it into this one. I'll think about it once the notes are in. Once 1-4 are addressed (most of it is a notes entry plus a reworded line), I'm happy to merge this. Thanks again for raising it rather than just restoring and moving on. |
…he approved marking and the records
|
Thanks for the careful read. Pushed 8c66e8d:
The line is reworded as you suggested: |
L-279 to L-293: the 9a's slot B hold and what made devinfo writable (#125), per-device flags checked by building both command lines (#101), env defaults in the shared build library and CI waiting for approval (#79), GLib retry paths and radio stops checked at the controller (#99), silent re-pushes and stacks with several tops (the tegu stack), and two of our own: merge only after every check has finished (#126), and an empty diff is checked for empty inputs (kernel #123). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks, @W4HO-ham. This is exactly the record I was hoping for, and it changes my mind on point 3: I'd assumed stock only stops at a recovery prompt, and your notes show A few things before I merge: 1. Two places still say stock "stops in recovery". 2. What did the 10-26 boot run? The entry says the failing boot was "the Google-kernel RAM comparison". Nothing in tegu's notes describes that boot, and it matters for the cause. Was Google's kernel written to slot B with our initramfs (storage guard, USB safety net), or with stock's vendor ramdisk? If stock's first-stage init ran on slot B, it could have touched 3. A few facts the entry implies but doesn't state:
4. The marking itself. Thanks for the full account. Both markings were fine: they're the operation I approved on #24, under its written conditions (the same tool, a fresh backup checked against live storage, one bit at 0x35, the bytes recorded), and I'm happy with how you did them. What I want to settle before the next one is the laptop script around the tool, which makes Merge order: this PR and #100 both add a section at the top of Thanks again for turning a bad day into a rule the next 9a owner can follow. |
…wenty-sixth rounds (#127) * docs/review: 19 lessons and 3 corrections from the twenty-fifth round The confirmed candidates from round 25 (L-260 to L-278): the 9a slot B marking as a devinfo question, cross-repository links, squash-merged stacks and version files, stale hand-built prebuilts, and before-merge Pixel 7 Pro runs on omamobile-int. C-16 to C-18 correct our own earlier comments on kernel #98, kernel #80 and #74; L-241 is corrected in place. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * docs/review: 15 lessons from the twenty-sixth round L-279 to L-293: the 9a's slot B hold and what made devinfo writable (#125), per-device flags checked by building both command lines (#101), env defaults in the shared build library and CI waiting for approval (#79), GLib retry paths and radio stops checked at the controller (#99), silent re-pushes and stacks with several tops (the tegu stack), and two of our own: merge only after every check has finished (#126), and an empty diff is checked for empty inputs (kernel #123). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
|
Done in bb5acda:
|
|
Thanks, @W4HO-ham, for working through all of that so quickly. I checked each point against bb5acda:
Policy and tests pass on this head, and it still merges cleanly with One small thing before I merge, and it came from the rewrite. The 10-26 paragraph used to start with our kernel's release, and the 10-28 paragraph still begins "The same kernel, with a new DTB". Now that 10-26 names only Google's 6.1 kernel, "the same kernel" reads as Google's kernel on 10-28 too. 10-28 is the day that carries the rule, because it ran our own kernel and initramfs, so please name the release there ( One optional question, if you remember: on 10-26 Google's kernel ran without our safety net, so what used up B's three tries? Watchdog resets, or restarts by the buttons? It would round off the 10-26 account, but I won't hold the merge for it. When I merge, I'll write the squash message myself. The default would carry your first commit's text ("stock reformats After that, this goes in first and #100 follows. |
…was masked Review: the status says it works on the owner's integration kernel, with LHBM open as derekross/linux-omarchy-mobile#126 and the DSI reads as derekross#125, and the fprintd mask came from the palmtop's private bring-up root. The G3 next step is on the charger.
… H5 scripts, the meter rule and rollbacks The maintainer's review of the runbook: H2 now copies and enables the early read (and H6 removes it), H1 checks the I2C device nodes and that the DTB dead-man is unarmed and disables it, the meter rule covers any reset with the drive attached, and H4 and H5 have rollback lines and a word on pstore. H5 says plainly that no SSH is guaranteed after the drive-for-laptop swap, what can bring the link back, and the forced restart that boots like H3. The scripts that run while the laptop is out are published next to the early read and started detached with systemd-run: H4's unbinds the charger 20 s after the drive enumerates and logs VBUS and CHG_CNFG_00; H5's resets without a shutdown, and only when the early read is armed and a drive has enumerated. The early read loads i2c-dev itself, since it runs before systemd-modules-load, and waits for the root to be read-write. The DTB paragraph records the gn3 build's sha256 and that it isn't derekross#127 byte for byte; the node comparison against a derekross#127 build is the open step. "settled on derekross#125" is "being settled".
…fter, the code by commit The runbook stops if slot B isn't marked successful and doesn't mark it (a devinfo write on the 9a, derekross#125). S1 needs the NFC stop-poll hook active or polling stopped by hand, with the kernel's derekross#131 state and the rule that stays, a Rollback line and pstore after a cycle that doesn't come back. S2 snapshots regulator_summary before and diffs it after, stops if more than the four rails changed, records stock's consumers of them (none in its DT: no phandle on any of the four, the GNSS three on the PMIC's channel mux, LDO15S a spare), keeps the modem off in that boot and stays away from a G3 session. The push wording says what bin/omamobile push writes and that only vendor_kernel_boot differs. The module, the bring-up DT and the S2 change are linked by commit on the fork's tegu-sleep-bringup branch, and the CPU_INFORM hints of 10-28 are credited to the bring-up module that wrote them over the PMU SMC. The measurements link derekross#100's NOTES entry.
…oot, S1's command and column S2 says it turns LDO15S off on purpose for one boot against FINGERPRINT-PLAN's rule (that plan gets the matching note once derekross#138 is on main), and step 3 keeps omamobile-fp, the trusted app and storageproxyd stopped before the push and until the rollback, since F6c still writes secure objects. S1 names its command, power-ab with derekross#157's systemd suspend mode (rtcwake -m no plus systemctl suspend, so the NFC hook runs), reads charge_now, column 7, because the 9a's gauge has no charge_counter, and waits for derekross#157. Also: derekross#125's rules are being settled; the four rails read disabled with or without the unused-rail cleanup; stock's source grepped for the four rail names (DT sources and the register header only, no module asks by name); a Polling read before each run; nothing starts the modem at boot, so nothing to mask.
|
A gentle ping on my 00:53Z ask, since #100, #131 and #136 all wait on this one: in the 10-28 paragraph of NOTES (around line 46), could you replace "The same kernel" with the kernel release that boot ran ( |
* device/tegu: SLEEP-PLAN, an audit of the s2idle drain The 9a sleeps and wakes, but its drain is high. This audits why from stock's device tree and a snapshot of the palmtop: no cluster or SICD idle states, the GNSS and fingerprint rails always on in the bring-up DTB, and MIF, with a measurement runbook. * device/tegu SLEEP-PLAN: the firmware's PSCI has no OS-initiated mode So cluster states on zumapro would be platform-coordinated, with a composite StateID; and the core state's rejection rate, read from cpuidle. * device/tegu SLEEP-PLAN: start from what's measured The cable-out s2idle figure (about 140 mA), MIF in sleep (about 57 mA saved), and the cluster power-down tries of 10-28 (cluster 0 hangs; clusters 1-2 save nothing measurable) were already measured; the candidates are re-ranked on them. * device/tegu SLEEP-PLAN: S2's image, its expectation and rollback * device/tegu SLEEP-PLAN: the review's preconditions, S2's before and after, the code by commit The runbook stops if slot B isn't marked successful and doesn't mark it (a devinfo write on the 9a, #125). S1 needs the NFC stop-poll hook active or polling stopped by hand, with the kernel's #131 state and the rule that stays, a Rollback line and pstore after a cycle that doesn't come back. S2 snapshots regulator_summary before and diffs it after, stops if more than the four rails changed, records stock's consumers of them (none in its DT: no phandle on any of the four, the GNSS three on the PMIC's channel mux, LDO15S a spare), keeps the modem off in that boot and stays away from a G3 session. The push wording says what bin/omamobile push writes and that only vendor_kernel_boot differs. The module, the bring-up DT and the S2 change are linked by commit on the fork's tegu-sleep-bringup branch, and the CPU_INFORM hints of 10-28 are credited to the bring-up module that wrote them over the PMU SMC. The measurements link #100's NOTES entry. * device/tegu SLEEP-PLAN: the fingerprint stack stays stopped in S2's boot, S1's command and column S2 says it turns LDO15S off on purpose for one boot against FINGERPRINT-PLAN's rule (that plan gets the matching note once #138 is on main), and step 3 keeps omamobile-fp, the trusted app and storageproxyd stopped before the push and until the rollback, since F6c still writes secure objects. S1 names its command, power-ab with #157's systemd suspend mode (rtcwake -m no plus systemctl suspend, so the NFC hook runs), reads charge_now, column 7, because the 9a's gauge has no charge_counter, and waits for #157. Also: #125's rules are being settled; the four rails read disabled with or without the unused-rail cleanup; stock's source grepped for the four rail names (DT sources and the register header only, no module asks by name); a Polling read before each run; nothing starts the modem at boot, so nothing to mask.
…he initramfs "The same kernel" is now the release string the day's run kit records, with the note that it names the tree rather than one build, and the initramfs that boot used: ours, packed by the kit.
|
Sorry for the wait on that one; it's in at 6890213. The 10-28 paragraph now names the release that boot ran, |
|
Thanks, @W4HO-ham, and no need to apologise. The 10-28 paragraph now names Before I merge, there's one sentence I want to change, and it's my own earlier wording I'm revisiting, not something you missed. Marking slot B is on hold until I've decided how it should work. The AGENTS line still presents the marking as the step after a
The two links stay, with the second one labelled "the operation approved on 2026-10-22, now on hold". When I come back with a decision on marking, the line changes again with it. While I'm in there, one word in NOTES: "Marking B both days was the approved operation above". That section is below the new entry now, so "above" should be "below". Since you've allowed maintainer edits, I'll push both changes to your branch myself and merge, unless you'd rather make them. If you do, just say so here. I'll write the squash message myself, as I said, so your first commit's wording doesn't land. Then #100, #131 and #136 follow in that order. |
|
Please push both changes yourself and merge; your wording for the AGENTS line says what we agreed, and "below" is right for the NOTES section now. Thanks. |
…nterrupt only, with its way back From Derek's review of 2198e47. The H2b overlay no longer gives the boost its enable line: SGPIO7 is the RT9470's charge enable, and driving it with nothing managing the charger could start it charging beside the MAX77779; that line stays with kernel derekross#127, which must clear CHG_EN before taking it. H2b now reads the boost's FUNCTION and ICSTAT after the modprobe, as the H4 script does, and stops if ICSTAT shows charging. Its rollback says what to do when SSH doesn't come back after a forced restart: fastboot by the buttons and stop, no flash without Derek (derekross#125), so the step is risk medium; the header and the DTB paragraph say that H2b's push writes slot B's images through the normal path. The status line says H2b boots gn3 plus the overlay, step 1 is marked done by H2, the H5 script traps TERM so the swap timer's reboot disarms the early read too, the overlay's header says bring-up only, and the watchdog sentences record the reworked boost driver (6c5d97e18) as the follow-up asked for on derekross#131.
…efore a flash The AGENTS line presented the devinfo marking as the step after a flash. The hold agreed in derekross#125 lived only in the thread, so say it in the line, and label the approved operation as on hold. NOTES: the approved section is now below the new entry. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # device/tegu/NOTES.md
Pixel 9a records from 2026-10-27 and 2026-10-28 in tegu NOTES: cameras, NFC, USB-C, storage power, fingerprint and drain on 10-27; camera switching, pstore, idle hints and sleep memory rate on 10-28; and what the device table's claims rest on. MODEM-PLAN, AGENTS and the DEVICES row follow those records. NOTES keeps newest first: the 10-29 section, then the slot A fallbacks from #125, then these three. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks, @W4HO-ham. I've pushed the AGENTS sentence we agreed (marking B on hold, |
…ing (#307) The 9a's row after today's merges, with a NOTES entry for the evening's boots that the new cells link to. - **Wi-Fi, testing to works.** `omamobile-device-tegu` 0.2.4 (#238) starts it at boot now, through the shared gs201 host and zuma PHY (kernel #178 on #156, both in review) and brcmfmac (kernel #56, in review), with only the extractor's firmware. The NOTES entry has the boot, and the three rebinds with Bluetooth up after #178's FLR commit. The kernel parts are still in review, and the cell says so, as the display cells did. - **Suspend, WIP to testing.** The power key now lights the lock screen on the first press (#268, #275). Thirty attended sleeps and a day of real use with auto-sleep on all woke (SLEEP-PLAN, #284). The cell still says that three test sleeps on 10-28 to 10-30 never woke and that sleep tests stay attended. If you'd rather keep WIP until that hang is understood, I'll revert just the word. - **Display:** the display DT is merged, and brightness, DPMS off and on and the 60/120 Hz switch now have a dated record (kernel #125's boot, which also checks the panel ID). - **Charging, Haptics, Torch, NFC:** kernel #161, #96, #104 and #105 are merged. - **USB net:** the rescue also handles a replug the gadget didn't see (#280). The NOTES entry lists each boot with its release: Wi-Fi from the package, the power-key wake, the replug, storage idle (kernel #99), the panel ID (kernel #125), the RTC without the PMIC interrupt (kernel #55), and the auto-sleep day. GNSS and Modem stay as they are until the SIM test on Tuesday. Docs only; `tools/pr-check` passes, and the diff has no identifiers. Closes #307 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* device/tegu: SLEEP-PLAN, an audit of the s2idle drain The 9a sleeps and wakes, but its drain is high. This audits why from stock's device tree and a snapshot of the palmtop: no cluster or SICD idle states, the GNSS and fingerprint rails always on in the bring-up DTB, and MIF, with a measurement runbook. * device/tegu SLEEP-PLAN: the firmware's PSCI has no OS-initiated mode So cluster states on zumapro would be platform-coordinated, with a composite StateID; and the core state's rejection rate, read from cpuidle. * device/tegu SLEEP-PLAN: start from what's measured The cable-out s2idle figure (about 140 mA), MIF in sleep (about 57 mA saved), and the cluster power-down tries of 10-28 (cluster 0 hangs; clusters 1-2 save nothing measurable) were already measured; the candidates are re-ranked on them. * device/tegu SLEEP-PLAN: S2's image, its expectation and rollback * device/tegu SLEEP-PLAN: the review's preconditions, S2's before and after, the code by commit The runbook stops if slot B isn't marked successful and doesn't mark it (a devinfo write on the 9a, derekross#125). S1 needs the NFC stop-poll hook active or polling stopped by hand, with the kernel's derekross#131 state and the rule that stays, a Rollback line and pstore after a cycle that doesn't come back. S2 snapshots regulator_summary before and diffs it after, stops if more than the four rails changed, records stock's consumers of them (none in its DT: no phandle on any of the four, the GNSS three on the PMIC's channel mux, LDO15S a spare), keeps the modem off in that boot and stays away from a G3 session. The push wording says what bin/omamobile push writes and that only vendor_kernel_boot differs. The module, the bring-up DT and the S2 change are linked by commit on the fork's tegu-sleep-bringup branch, and the CPU_INFORM hints of 10-28 are credited to the bring-up module that wrote them over the PMU SMC. The measurements link derekross#100's NOTES entry. * device/tegu SLEEP-PLAN: the fingerprint stack stays stopped in S2's boot, S1's command and column S2 says it turns LDO15S off on purpose for one boot against FINGERPRINT-PLAN's rule (that plan gets the matching note once derekross#138 is on main), and step 3 keeps omamobile-fp, the trusted app and storageproxyd stopped before the push and until the rollback, since F6c still writes secure objects. S1 names its command, power-ab with derekross#157's systemd suspend mode (rtcwake -m no plus systemctl suspend, so the NFC hook runs), reads charge_now, column 7, because the 9a's gauge has no charge_counter, and waits for derekross#157. Also: derekross#125's rules are being settled; the four rails read disabled with or without the unused-rail cleanup; stock's source grepped for the four rail names (DT sources and the register header only, no module asks by name); a Polling read before each run; nothing starts the modem at boot, so nothing to mask.
…oot (derekross#125) On 2026-10-28, and most likely on 2026-10-26, the Pixel 9a fell back to stock on slot A after slot B, unmarked, used its three ABL tries, and stock overwrote userdata before recovery could offer anything. tegu NOTES records both days: what each boot ran, how B was recovered and marked, and the bytes the approved devinfo write changed. AGENTS gains the rule: flash only a known-good set and ask the maintainer before any flash on the 9a; marking B is on hold until the maintainer decides how it's done; with B marked, try new images with push only. The MODEM-PLAN correction now says stock overwrites userdata before recovery and links the NOTES entry. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ss#100) Pixel 9a records from 2026-10-27 and 2026-10-28 in tegu NOTES: cameras, NFC, USB-C, storage power, fingerprint and drain on 10-27; camera switching, pstore, idle hints and sleep memory rate on 10-28; and what the device table's claims rest on. MODEM-PLAN, AGENTS and the DEVICES row follow those records. NOTES keeps newest first: the 10-29 section, then the slot A fallbacks from derekross#125, then these three. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ing (derekross#307) The 9a's row after today's merges, with a NOTES entry for the evening's boots that the new cells link to. - **Wi-Fi, testing to works.** `omamobile-device-tegu` 0.2.4 (derekross#238) starts it at boot now, through the shared gs201 host and zuma PHY (kernel derekross#178 on derekross#156, both in review) and brcmfmac (kernel derekross#56, in review), with only the extractor's firmware. The NOTES entry has the boot, and the three rebinds with Bluetooth up after derekross#178's FLR commit. The kernel parts are still in review, and the cell says so, as the display cells did. - **Suspend, WIP to testing.** The power key now lights the lock screen on the first press (derekross#268, derekross#275). Thirty attended sleeps and a day of real use with auto-sleep on all woke (SLEEP-PLAN, derekross#284). The cell still says that three test sleeps on 10-28 to 10-30 never woke and that sleep tests stay attended. If you'd rather keep WIP until that hang is understood, I'll revert just the word. - **Display:** the display DT is merged, and brightness, DPMS off and on and the 60/120 Hz switch now have a dated record (kernel derekross#125's boot, which also checks the panel ID). - **Charging, Haptics, Torch, NFC:** kernel derekross#161, derekross#96, derekross#104 and derekross#105 are merged. - **USB net:** the rescue also handles a replug the gadget didn't see (derekross#280). The NOTES entry lists each boot with its release: Wi-Fi from the package, the power-key wake, the replug, storage idle (kernel derekross#99), the panel ID (kernel derekross#125), the RTC without the PMIC interrupt (kernel derekross#55), and the auto-sleep day. GNSS and Modem stay as they are until the SIM test on Tuesday. Docs only; `tools/pr-check` passes, and the diff has no identifiers. Closes derekross#307 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 9a lost its Linux root twice (10-26 and today). After
omamobile flash(fastboot --set-active=b), slot B is unmarked, and nothing on the 9a marks it again. Every boot burns an ABL retry, and the initramfs USB safety net burns them fast. At zero, ABL boots stock Android, which reformats and encrypts userdata. This adds the rule to the boot-chain gotchas: mark B successful on the first good boot after a flash, and never boot an untested image while it's unmarked.