diff --git a/device/tegu/NOTES.md b/device/tegu/NOTES.md index fcd51c864..292ac9254 100644 --- a/device/tegu/NOTES.md +++ b/device/tegu/NOTES.md @@ -35,6 +35,83 @@ resets) and are in [MODEM-PLAN](MODEM-PLAN.md) under "G3, twenty-fourth to thirty-fourth tries" and "The CP link hang". Not a GNSS matter: if a boot with nothing but Bluetooth starting resets again at that time, start here. +## 2026-10-28, evening: USB-C host mode through TCPM + +Kernel `7.3.0-rc6-omamobile-tegu-s2mpg15-m1-g991042e2eff3-dirty` (the +integration tree), DTB with the TCPC as a dual-role port. + +- **USB-C host mode works end to end** [V]. TCPM runs the port as dual-role + (try sink, source PDO 5 V 900 mA), with `vbus-supply` = the MAX77779 + charger's `chgin-otg` regulator (kernel #124) over the RT9470's OTG boost, + and a graph to dwc3's role switch. A USB flash drive attaches as source + (SRC_READY), VBUS holds at 5.2 V, the data role goes to host, and the + drive enumerates at high speed (`sde: sde1 sde2`). With the drive out and + the laptop back: sink, a PD contract, the gadget configured. Again after a + fresh boot. Two parts of that setup aren't in a public tree yet: the DT + fragment (the TCPC as dual-role with `vbus-supply`, the graph to dwc3) and + the RT9470 bring-up module that writes OTG_EN, JEITA_EN and the watchdog + bits. They are on the kernel bring-up branch + derekross/linux-omarchy-mobile#127, with the plan and runbook in #136. + This supersedes #100's "the eUSB2 repeater doesn't take the host role + yet". +- **How the cable swaps were covered** [V]: attended, with W4HO at the + palmtop for every swap. Taking the laptop out cuts the only link, so each + sampling script carried its own dead-man: with no USB gadget configured 6 + to 8 minutes after the cable came out, it unloaded `tcpci_maxim` and set + the role switch back to device, so the laptop would come back as a sink. + It never fired, because the laptop came back every time. The DTB boot + itself ran under a one-boot DTB dead-man. The next run uses a reboot + timer armed before the unplug and stopped over a fresh login (#136's + runbook). `tools/usbc-deadman` is written for the 7 Pro's MAX77759 and + doesn't apply here. +- **Why VBUS collapsed at first** [V]: the RT9470's JEITA temperature + control (0x0c bit 7) was on with STAT2 reading 0x80, and its I2C + watchdog (0x01 bits 1:0) had expired (STAT3 0x20). That JEITA read "hot" + is an inference [I]: the TS pin may have no thermistor on it, which + hasn't been checked. With either set, the + boost reported OTG (ICSTAT 0x0f) while VBUS fell from 5.2 V within ~60 ms, + and TCPM retried every second. Google's `rt9471_charger` turns JEITA off + on the 9a (`en-jeita` is unset in `zumapro-buck-charge-rt9471.dtsi`) and + only runs the watchdog while it kicks it. With both off, VBUS holds under + Rd and Rp, with and without AutoDischargeDisconnect. The boost's driver + has to clear both. +- **What that leaves open** [I]: with the watchdog off, nothing turns the + boost off after a crash. What a warm reboot with host mode on leaves + behind (the RT9470's OTG bit and ICSTAT before any driver loads) hasn't + been measured, so whether VBUS stays out across one is unknown. With + JEITA off the RT9470 has no temperature cut-off of its own, which + matters only if it charges. Its CE comes from the MAX77779's sGPIO 7, + which hasn't been read. On the charger with the battery full, FUNCTION + read 0x41 (`CHG_EN` set) and ICSTAT 0xf1 (VBUS ready, not charging), so + whether it charges under our kernel isn't known yet. +- **The eUSB2 side** [V]: the PHY has to come out of reset with the device + attached. A drive plugged into a port that's already host isn't seen + (the TUSB2E11's attach-detect interrupt never latched). TCPM's flow + (attach on CC, then the role switch, then `phy-snps-eusb2` resetting the + PHY on the role change) does it in that order. +- **NFC after s2idle** [V]: the ST54L loses its NCI state across a + suspend. The first command after resume gets no I2C ACK (`NFC: Error -6 + on I2C send`), and every NCI request times out after it. Powering the + adapter off and on through neard (a fresh NCI init) brings it back, so + the sleep hook has to power the adapter off before the sleep and on + after it, not only stop polling (10-29: not needed with + https://github.com/derekross/linux-omarchy-mobile/pull/131, and a + power-off hook failed to wake once; see #137). +- **Fingerprint calibration** [V]: the HAL injects it itself + (`injectMtCali`), reading a 112-byte header at offset 0 of + `/dev/block/by-name/mfg_data`. The file `fastboot oem udfps get-cal-url` + names has exactly that layout (magic `0xca11ca11`, version 1, and the + payload length at 0x6c). Written into the chroot's private `mfg_data.img` + copy (the partition itself untouched), it's accepted: + `vendor.gf.cali.state = 1`, and the "No SA calibration present" line is + gone. Later that evening enrollment completed, but matching doesn't work + yet (#135; 10-29: it does now, see FINGERPRINT-PLAN F9/F10 in #138). +- **GNSS** [V]: telling the AoC the GNSS type (`gnss-type = <2>`, as + Google's `zumapro-phone-ignss.dtsi`) doesn't change the fault. Kepler + loads and goes OPERATIONAL, and the HAL starts a session, but every engine + start still ends in "KEP M7 HARDFAULT ... LR,PC 600B5B39 0000062C" + (20 in 4 minutes). + ## 2026-10-26 and 2026-10-28: slot A fallbacks that cost userdata Both times slot B was active but **not marked successful**, so each boot