Repository navigation
device/tegu: GNSS G3's fifth to forty-fourth tries, the CP link hang, and GNSS parked - #153
Conversation
The engine tracks satellites and gnssd's watchdog does the rest. The link garbles single bytes at the 6.24 MHz the SPI really runs at, and stock's OSCCLK clock, 32-bit words and auto CS break the framing.
Stock's OSCCLK clock breaks framing here. Reading in mode 1 lines every frame up on the as-ported clock, but the first bytes after each chip select still arrive damaged, and Kepler resets can take the SoC down.
Kepler resends frames, so the copies can be compared. In mode 1, long frames lose the same bytes at every 64-byte read boundary.
Holding chip select between Kepler's reads ends long frames in zeros, so Kepler wants it released between reads, as stock does.
With feedback delay 3 Kepler's frames are intact but one bit late. A per-read realignment gives clean short frames and satellites; long frames still lose data at each chip-select release.
At stock's clock Kepler's frames are intact but its bit offset drifts with every transfer, more as the clock slows. The CP link hang came back with the slower watcher, so it remains open.
The doorbell writes into the CP's BAR with no lock held across its link check, and the PCIe power-off clears pcie_powered_on outside the transmit lock, so a doorbell can land while the PHY goes into isolation.
The doorbell path is serialized with the PCIe power-off after all; no unguarded BAR access found by reading.
Two kernel changes give the GNSS SPI stock's 3.072 MHz and one uncut 64-byte read per block. Mode 0 reads stay marginal, mode 1 reads lose one byte per back-to-back block; a logic analyzer is the next tool.
…or the night, the link-hang trace gated The paragraph on the twenty-fourth try was in twice; the second copy goes. The small hours' section held tries 24-34 and the CP link analysis under the "fifteenth to twenty-third tries" heading, so they get their own. The trace at each BAR access in CPIF, written as a hands-free next step, is now G3-L1: attended, on the charger, with Expect, Stop, a cap on resets, Rollback and risk, and the tracing change on a kernel branch before it boots. It only runs if GNSS is resumed; the status line now says GNSS is parked. Placeholders mark where the bring-up branch and the partition check go.
They came before any GNSS step on the night of 10-29, so they belong in NOTES rather than inside MODEM-PLAN's GNSS section, where the maintainer found them hard to find.
The ten commits with G3's fifth to forty-fourth tries, the CP link hang and GNSS parked are reviewed on derekross#153 now, so this pull request goes back to fingerprint F6c-F10 and G3's third and fourth tries: MODEM-PLAN is the file as of 6fe8313 again.
M0.1's checksum check, run after the night's modem sessions and resets: all six protected partitions match the 10-28 backup, read-only and unmounted. Also how the test DTBs reached slot B.
…ile#138 now Kernel derekross#135 went up without a sign-off and is replaced by derekross#138 with the same two commits; the parked paragraph links the live one.
Every variant that ran during G3's tries is now a commit on the owner's fork of the kernel (linux-omarchy-mobile#137, a draft), linked by commit from the plan so each try can be read next to its code.
|
Both of the points carried over from #138 are in now. The partition check is recorded (all six protected partitions match the 10-28 backup, read-only, unmounted), and the test code behind G3's tries is public as derekross/linux-omarchy-mobile#137 (draft, branch tegu-gnss-spi-bringup on my fork, head 8d77391d9), on top of the two kernel fixes (now #138 there): the gnssif_spi driver as ported, then one commit per variant in the order of the tries, the pdma1 probe and memcpy modules, the four test overlays, and the g7 scripts with the offline CRC checks. The plan links each by commit (cf9ff1c). One correction to what I wrote before: there was no separate USI3 mux-and-divider module; that code is the osc and div parameters inside gnss_spi, and the commit messages say so. Each test DTB went onto slot B through the kit's omamobile images and omamobile push; the palmtop is back on the gn3 DTB with the as-ported modules. |
…derekross#113) * device-raven, tools: haptics through feedbackd on the CS40L25A The kernel side is derekross/linux-omarchy-mobile#114 (cs40l25-vibra, its binding and raven's node). This is the rest: - device/raven/kernel.config builds cs40l25-vibra as a module. - extract-vendor-firmware -d raven stages the CS40L25A's files from the owner's factory image: the CLAB RAM firmware and the default wavetable, which the driver loads, and the CLAB configuration (braking), not loaded yet. - device-raven's 72-omamobile-feedbackd.rules makes the motor a feedbackd vibra, as cheetah's CS40L26. The motor's input device appears only once its firmware has loaded, about 5 s into the boot, and feedbackd (D-Bus activated) only looks for it when it starts, so on a boot where something asked for feedback first there were no haptics until feedbackd was restarted. The motor's arrival now starts omamobile-feedbackd-rescan.service, which stops a running feedbackd; the next request starts it again with the motor. HAPTICS-TORCH-PLAN records H5 and the boot race. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * device-raven 0.1.27: the haptics version bump Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Derek Ross <derekross@gmail.com>
|
Thank you for this. It's a remarkable night's work, written up so the next person can follow every step. Finding that the "hardfault" is gnssd's own watchdog, then tracing the link problems to USI3's clock and the 63 + 1 cut, turns GNSS on the 9a from a mystery into one well-defined SPI question. Parking it there, with the two kernel fixes posted, is a sensible call. All the round 29 points carried over from #138 are in:
The policy check and CI pass. I found no identifiers and no vendor files, and nothing here writes outside slot B. A few small things before merge:
Not blocking: the description still mentions the two placeholders as unfilled. It doesn't land on main, but an update would help anyone reading the PR later. |
Twenty-six lessons (L-318 to L-343) from the round 29 omamobile-reviewer runs, and three corrections: C-20, our kernel derekross#109 advice that the Pixel 7's SPI controller DMAs the touch report (it polls; gs201's spi_0 has no dmas); C-21, asking for FT3658=m in panther's device kernel.config against RELEASES; C-22, a Bluetooth address in kernel derekross#98's thread that the first review missed. Five candidates were dropped as already covered. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…rary and the CP state it waits for (derekross#141) * omamobile-modem 0.1.8, extract-vendor-chroot: gnssd's custom GNSS library and the CP state it waits for On the 9a gps.cfg sets GlueLayer_IsUseCustomGnssLib, and gnssd dlopen()s libcustomgnss.so, which waits for ONLINE at /sys/bus/platform/devices/cpif/modem_state and then opens RIL's GPS channel through libsitril-gps and libsitril-client. The extractor left all three out, since nothing links them, and the chroot's /sys had only the device path, not its bus link. Neither fixes Kepler's engine-start fault yet. * omamobile-modem 0.1.8: the extractor re-run, and the shared /sys bind The extractor was run once more with this change on the owner's BP3A factory image: 195 files against the previous 188, nothing removed, every library resolved, nothing new to stub. The seven new paths are the three GNSS libraries and their closure (libbinder, libbinder_ndk, libapexsupport and the Samsung radioExternal NDK library). MODEM-PLAN's G3 section records that, and that an existing 9a chroot gets the libraries only when its owner re-extracts. The second /sys bind is shared by every phone's modem profile. It is harmless on the 7 Pro and the 6 Pro: cbd never looks there.
The clock and the cut are in linux-omarchy-mobile#138, in review, not yet in the kernel; the status line and the morning's section say so. The morning's tries ran images with the night's release string, since the build fixes LOCALVERSION, so the plan names the Image files and their sha256 prefixes instead. The partition check at 10:20 came after the morning's tries, so it covers them. The private paths that were left (the realign tree on maclab, the g7 script, the drafted clock change, the offline tools on the laptop) now point at the public commits of linux-omarchy-mobile#137 and derekross#138. The tool that wrote slot B is named once, plainly: our bin/omamobile push, run by a local wrapper after bin/omamobile images.
|
Done in 9ec37d7, points 2-6; the rebase (1) waits for #138 to land and I'll resolve MODEM-PLAN with this branch's file then, or you resolve it the same way at merge, whichever comes first.
The description is updated too: the placeholders are filled and the kernel fixes are #138 in review. |
…th tries (derekross#138) FINGERPRINT-PLAN F6c-F10: on the Pixel 9a the under-display sensor captures GOOD frames once touches reach the HAL in panel pixels and the circle is lit before the capture; the enrolled finger matches and other fingers are refused. MODEM-PLAN G3: the third and fourth tries at the GNSS SPI link. The later GNSS tries moved to derekross#153. Co-authored-by: W4HO-ham <87742609+W4HO-ham@users.noreply.github.com>
…unter (derekross#157) --suspend systemd runs the A/B drain cycles through systemctl suspend, so sleep hooks (the 9a's NFC stop-poll) run; it refuses unless s2idle is the mode that will actually run, and rejects unknown --suspend values, so a phone without mem_sleep_default=s2idle can't enter deep with the cable out. A gauge without charge_counter (the 9a's) is read through charge_now. The default rtcwake mode is unchanged. Co-authored-by: W4HO-ham <87742609+W4HO-ham@users.noreply.github.com>
…s hangs Read offline from the ramoops consoles the palmtop keeps: the seven watchdog resets all end on the cp2ap_wakeup handler's toggle with the link cycling twice a second, one shows an RCU stall of three CPUs right as gnssd starts, one a doorbell parked because the link was down. That puts the hang in the window between the wake request and the link coming back, from a path that reads the modem's BAR while the PHY is isolated. G3-L1 gets a cheaper first step (hold the link up through a cpif parameter) and named trace points.
|
One more commit, df8d568, from reading the morning's consoles offline (the palmtop keeps every ramoops console, and each one ends with the next bootloader's |
device/tegu/MODEM-PLAN.md conflicted on the status line: this branch's file already carries everything derekross#138 added. It is resolved to this branch's file, plus the "gnssd's custom GNSS library (2026-10-28, omamobile-modem 0.1.8)" bullet that derekross#141 added on main, which this branch never had and a plain checkout of its file would have dropped.
|
Thank you for 9ec37d7 and for the extra hours on the consoles. Points 2 to 6 from my last comment are all in: the status line and the morning section name derekross/linux-omarchy-mobile#138 as in review, the morning's images are told apart by their sha256 since the release string can't do it, the 10:20 check covers the morning's resets, the four private paths now point at the #137 commits, and the plan names The rebase. #138 is on main now, and I've resolved the conflict as we agreed, with one change. When I suggested df8d568 is excellent detective work. Sorting the 24 consoles by
Not blocking:
After those two, this merges, and #154 follows right after it. Thanks again for a record this thorough. |
…from the sources Stock's node (swap-mode, 3.125 MHz, feedback delay 0, no chip-select-mode property) and Google's driver default of automatic chip select mean stock drives nCS from the controller with a two-clock gap, in PIO, with 32-bit words and the controller's swap; our port uses manual native chip select and feedback delay 2, and auto chip select was tried only at the wrong clock with 32-bit words. The next boot when GNSS resumes is that combination at stock's clock, 8-bit first.
…L1's frame The count was wrong by one (eight watchdog resets, the eight times listed). G3-L1 now says its first step, the link held up through cpif's keep_link_up, runs under the same frame and counts toward the same cap; the watcher is named as derekross#148 in review, and the partition check's span is the consoles' (07:34 to 09:19).
|
Thanks for the merge commit; I pulled it before adding anything, and the #141 bullet is intact. Both points are in at d46d6de: eight hangs (the eight times listed; the count was the typo), and G3-L1 now opens with the hold-the-link step under the same frame (M0.1 before and after, attended on the charger, kmsg with --watchdog, the same Stop and Rollback), its sessions and any hang counting toward the same caps. The two small ones too: the watcher is named as #148 in review, and the partition check's span is the consoles' (07:34 to 09:19). One more paragraph went in just before (4f1e95a): stock's SPI configuration for this link read from the sources, which turns out to be automatic chip select, feedback delay 0 and 32-bit words with the controller's swap, in PIO; our port runs manual chip select and delay 2, and auto chip select was tried only at the wrong clock, so that is the first boot when GNSS resumes (G3-A). The cpif keep_link_up build for G3-L1's first step is on my fork (tegu-cpif-keep-link, 3f76f123b) and staged on the palmtop; nothing boots until the owner is at it. |
The reset boot's last 34 s match the surviving boots around it to the tenth of a second and end right after bluetoothd and hostnamed start; the 120 s Bluetooth dead-man was not it and the audio script had not acted yet. The journal has nothing more, and the console of that reset is gone.
|
Thank you, both points are in at d46d6de, and thanks for pulling the merge commit first.
The stock SPI paragraph (4f1e95a) is a useful read. Automatic chip select, feedback delay 0 and the controller's swap explain why no earlier try matched stock, and G3-A is a sensible first boot when GNSS resumes. The journal read of the 03:03 reset in NOTES is careful too. One optional wording point, for whenever you next touch the file. The hold-the-link step runs three sessions, and those count toward a cap of three, so on paper the trace has nothing left. And "waits for the next day" reads differently from the Cap line's "then the step is closed". If you meant the cap per day, saying so in the Cap line would settle it. Nothing runs while GNSS is parked, so this doesn't hold the merge. CI is green on 1600fc6, and it merges cleanly with main. I'm merging this now, with a squash message I've written by hand, so the early commit lines ("the clock and the cut fixed") don't land as if the fixes were in the kernel. #154 follows straight after. Thanks again for such a thorough record. |
kernel derekross#156's Wi-Fi needs CONFIG_PHY_ZUMA_PCIE, a zuma-only PHY, so it goes in husky's fragment. On a kernel without derekross#156, olddefconfig drops the unknown symbol and nothing else changes (checked on omamobile with derekross#153). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
… rounds L-348 to L-361: shared drivers judged against every int-line user (#150), board pins against every stock user (#151), ownership flags under the consumer's lock (#152), /chosen children and Google's panel constants (#154), same-day notes PRs on one device (#208/#209), merge simulations for DEVICES rows, ours included (#165), stacked PRs that merged main (#115/#117), titles behind renumbered versions, a contributor merge that dropped the PR (#99), smoke package lists as provenance and "owner-authorized" writes (#200), promised fixes and named tags checked on the remote (#153/#157/#158), and file swaps into old packages (#193). C-23 (usbc-deadman's device scope, kernel #107) and C-24 (BNEP in a lynx-only fragment, #98). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s on the 7 Pro (#254) Kernel #153 (husky touch), #166 and #133 and omarchy-mobile #149 (panther touch) merged on 2026-10-31 and 11-01, and omamobile-audio 0.5.8 raised the 7 Pro's default speaker gains under verified protection. The website's device pages are built from this file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…th tries (derekross#138) FINGERPRINT-PLAN F6c-F10: on the Pixel 9a the under-display sensor captures GOOD frames once touches reach the HAL in panel pixels and the circle is lit before the capture; the enrolled finger matches and other fingers are refused. MODEM-PLAN G3: the third and fourth tries at the GNSS SPI link. The later GNSS tries moved to derekross#153. Co-authored-by: W4HO-ham <87742609+W4HO-ham@users.noreply.github.com>
… and GNSS parked (derekross#153) MODEM-PLAN's GNSS section from the night of 10-28 into 10-29. The "KEP M7 HARDFAULT" at every engine start is gnssd's soft watchdog resetting Kepler because the CHPP link over the GNSS SPI loses bytes. Two kernel causes were found, USI3 stuck on the 399.36 MHz parent and spi-s3c64xx cutting a full-FIFO transfer into 63 + 1 bytes; both are in derekross/linux-omarchy-mobile#138, in review. With them, mode 0 is marginal and mode 1 loses the first byte of each back-to-back block. Stock's SPI setup, read from its device tree and Google's driver, is automatic chip select, feedback delay 0 and the controller's byte swap, in PIO; no try has run that at stock's clock, so it is the first boot (G3-A) when GNSS resumes. GNSS is parked at the owner's decision. Every try links its commit on derekross/linux-omarchy-mobile#137. The CP link hang stays open. The morning's 24 ramoops consoles hold eight watchdog hangs, each ending between the cp2ap_wakeup edge and the link coming back, while the link cycles quickly. G3-L1 (attended, on the charger, capped, M0.1 before and after) first holds the link up with a bring-up cpif parameter, then traces CPIF's BAR accesses. The six protected partitions matched the 10-28 backup after the night's and morning's boots. NOTES records two resets at about 34 s after Bluetooth started, the 03:03 one read back from the journal. Test images reached slot B through bin/omamobile push only. Co-authored-by: W4HO-ham <87742609+W4HO-ham@users.noreply.github.com>
…ross#208) Power, volume keys and the touchscreen run on the Pixel 8 Pro with the zuma ALIVE pin controller, gpio-keys and the touch stack from linux-omarchy-mobile derekross#148 and derekross#153 (test tags b, b2, c5, c6 on the author's fork). DEVICES marks them testing.
… rounds L-348 to L-361: shared drivers judged against every int-line user (derekross#150), board pins against every stock user (derekross#151), ownership flags under the consumer's lock (derekross#152), /chosen children and Google's panel constants (derekross#154), same-day notes PRs on one device (derekross#208/derekross#209), merge simulations for DEVICES rows, ours included (derekross#165), stacked PRs that merged main (derekross#115/derekross#117), titles behind renumbered versions, a contributor merge that dropped the PR (derekross#99), smoke package lists as provenance and "owner-authorized" writes (derekross#200), promised fixes and named tags checked on the remote (derekross#153/derekross#157/derekross#158), and file swaps into old packages (derekross#193). C-23 (usbc-deadman's device scope, kernel derekross#107) and C-24 (BNEP in a lynx-only fragment, derekross#98). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s on the 7 Pro (derekross#254) Kernel derekross#153 (husky touch), derekross#166 and derekross#133 and omarchy-mobile derekross#149 (panther touch) merged on 2026-10-31 and 11-01, and omamobile-audio 0.5.8 raised the 7 Pro's default speaker gains under verified protection. The website's device pages are built from this file. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What and why
Split out of #138 at the maintainer's suggestion, so that one merges as fingerprint F6c-F10 plus G3's third and fourth tries, and the GNSS night gets its own review. This is MODEM-PLAN's GNSS section from the night of 10-28 into the morning of 10-29: G3's fifth to forty-fourth tries, the CP link hang, and the decision to park GNSS. Docs only. The ten commits of the night come from #138's branch and sit on its G3 third and fourth try text, so the diff shows those thirty lines too until #138 merges; after that they are identical to main.
What the night found. The "KEP M7 HARDFAULT" at every engine start is not Kepler's own fault. With gnssd's debug log on, the engine tracks satellites within about ten seconds of the start, and the fault that follows is gnssd's soft watchdog resetting Kepler and asking it for a hard fault, because the CHPP link over the GNSS SPI loses packets. Two causes were in the kernel: USI3's clock was stuck on the 399.36 MHz parent, so the SPI ran at 6.24 MHz instead of stock's 3.07 MHz, and mainline spi-s3c64xx cut every 64-byte transfer into 63 + 1 bytes. Both are in derekross/linux-omarchy-mobile#138, in review (the signed-off replacement of #135, which is closed). With those in, mode 0 is marginal at every feedback delay (the reads flip between on time and one bit late within a session), and mode 1 reads are intact but lose the first byte of every block after the first in a back-to-back message. That is a question for a logic analyzer on SPI21, not for more module-level guessing, so GNSS is parked at my decision, without a hardware probe. The two kernel changes stay and the as-ported modules run on the palmtop.
Still open: the CP link hang. After Kepler's watchdog resets, gnssd reloads the image over the CP's PCIe link, and the SoC reset within a second three times (02:43, 03:15, 05:02). Reading CPIF's source found no unguarded BAR access. The trace that would find it is now a gated step, G3-L1: attended, on the charger, with Expect, Stop (two resets or any
ufshcd_err_handlerline), a cap of two resets per session and three sessions, Rollback, risk medium-high, and the tracing change on a kernel branch before it boots. It runs only if GNSS is resumed.The maintainer's points from #138, as they stand here:
bin/omamobile push, run by a local wrapper afterbin/omamobile images.Device and build
uname -r): 7.3.0-rc6-omamobile-tegu-s2mpg15-m1-g991042e2eff3-dirty for the night's boots; the morning's tries ran boot images built from the same source with the two changes of spi: s3c64xx, clk: zumapro: the Pixel 9a's GNSS SPI at stock's clock, without the 63+1 cut linux-omarchy-mobile#138, in review (the signed-off replacement of omamobile-fingerprint 0.1.11: touches in panel pixels, the circle lit before the capture, the refresh rate the HAL parses #135); the release string is the same as the night's, because the build fixes LOCALVERSION, so the plan names the Image files and their sha256 prefixesChecks (no phone needed)
tools/pr-check origin/mainpassestests/allpassestests/packagespasses (or the packages I touched: none touched, docs only)versionbumped, and the new version is in the commit subject (no package changed)Co-Authored-By:orAssisted-by:trailer, never aSigned-off-by:On the phone
userdata, weakens the storage guard, or reads registers from userspace (AGENTS.md §1). The soft-watchdog RAM dump andkepler.binstay private.tools/phone-smokereportTo be added before merge.