Skip to content

arm64: dts: exynos: google: Add the Pixel 9 Pro Fold (comet) - #273

Open
kdegeek wants to merge 2 commits into
derekross:omamobilefrom
kdegeek:comet-board-dts
Open

kdegeek wants to merge 2 commits into
derekross:omamobilefrom
kdegeek:comet-board-dts

Conversation

@kdegeek

@kdegeek kdegeek commented Nov 18, 2026 •

Copy link
Copy Markdown

This is the kernel side of omarchy-mobile#203, as asked there on 11-12: the dt-bindings: commit first, new nodes disabled, no other board touched. USB uses the Pixel 9a's plain snps,dwc3 node, so it needs no glue (#271 is closed). It sits on the zumapro SoC description that came in with the Pixel 9a stack (#187 and its base) and changes neither zumapro.dtsi nor zumapro-reserved-memory.dtsi.

Evidence

Every value comes from the DT the phone runs every day: kernel 7.3.0-rc6-omamobile-comet-rc2-cand21, and its packed DT compared node by node with this file's DTB.

  • Enabled here, and the same on the phone: UFS with its supply, the keys, both watchdogs, the clock controllers, pinctrl, sysregs, ACPM with the main PMIC, ramoops, and the CPU OPP tables and thermal trips.
  • Enabled here, the 9a's way: USB (plain snps,dwc3, the four HSI0 clocks, peripheral mode at high speed) and both PHYs. The phone's daily DT still uses the glue node.
  • Described here but disabled, while the phone runs each of them enabled: the MAX77779 charger, gauge and TCPC, the MAX17201 second gauge, the haptics, torch, NFC, the Hall switch and the GPU at up to 940 MHz.

Cold boot (2026-11-18)

ABL booted this board DT, unmodified, on 7.3.0-rc6-omamobile-comet-rc2-cand21. That was v2's DTB. v3 differs from it only by dropping the three reserved-memory nodes ABL inserts itself, which had only logged overlaps. The trial vendor_kernel_boot_b restored the daily image from its own init, so a lost USB link would have recovered on the next boot.

  • 11210000.usb bound to dwc3 directly. The gadget was up 3.17 s into boot, and the host configured it (configured driver=dwc3).
  • Both watchdogs: with no tmr_atboot in the bootargs, the driver stops them at probe. Userspace then opened and fed them.
  • tools/phone-smoke:
Check Result
boot (required) ok: slot b, 7.3.0-rc6-omamobile-comet-rc2-cand21
storage guard (required) ok: only userdata and slot B's boot images writable
root (required) ok: ext4 on userdata
kernel health (required) ok: no oops or BUG
USB gadget (required) ok: bound to 11210000.usb
buttons ok: gpio-keys
display, GPU, touch, session not ok: not in this DT
Wi-Fi, Bluetooth, battery, sound, sensors skip: not in this DT
  • New dmesg lines against the daily tree:
    • the three reserved-memory overlaps, fixed in v3;
    • five samsung-pinctrl …: irq number not available, from zumapro.dtsi's pinctrl nodes having no interrupts. That file isn't touched here, and this board uses no GPIO interrupts on those banks.

Differences from the phone's DT, on purpose

  • GSA protected region: stays at the SoC's 0x3ff000. The phone's DT asks for 4 MiB, which overlaps gsa-log@905ff000.
  • The second gauge's 3 mOhm shunt: the max1720x binding can't express it yet, so the gauge stays disabled.
  • Charger: the step-charge table, thermal-mitigation levels and the charger/TCPC links are left out. The MAX77779 binding here has no such properties yet.
  • Stock's thermal trips: not used. The trips are comet's lower values (60/65/85 °C), the ones the phone runs.

Left out: no driver or binding here, or no SoC node in zumapro.dtsi

  • Cover display (CT3D on DSIM0) and inner display (CT3B on DSIM1): no panel drivers or bindings, no dsi1 node, and the comet DECON variants are missing. TE comes with them.
  • Touch, cover (Synaptics TCM on SPI19) and inner (Goodix GT9966 on SPI20): no bindings, and no spi19.
  • Cameras (S5K3K1 ×2, S5K3J1, S5K3LU, IMX787), their lenses and the SLG51002/RT6160 module PMICs: the CSIS and D/C-PHY bindings are gs201-only. So the ultrawide link's google,lrte-packet-delimiters comes with the camera series. The IMX787 binding is in media: i2c: imx386: IMX787 (Pixel 9 Pro Fold main camera); ak7375: focus after standby and sensor-owned power #272.
  • AoC sensors (USF), audio (CS35L41 ×2 on SPI13, AoC sound card), Bluetooth over AoC, the GSA variant comet uses, and Trusty's IRQ routing: no comet AoC description yet, and no spi13.
  • Wi-Fi (BCM4390 on PCIe1), the modem (S5400 on PCIe0) and CP power/SPMI: no bindings for the endpoints or comet's PCIe variant.
  • Fingerprint (FPC1540), UWB (QM35), eSE (ST54 SPI), the eUSB repeater (TUSB2E11), wireless charging (P9222, PCA9468), the S2MPG14 RTC, S2MPG15 and its board thermistors, and BCL: no bindings here.

Checks

On v3: checkpatch --strict gives the same single FILE_PATH_CHANGES warning, and the DTB builds with dtc, identical to v2's apart from the three dropped nodes.

On v2, with tools/kernel-pr-check --build OUT --merge origin/omamobile HEAD:

  • Merges cleanly.
  • Subjects have prefixes, policy passes, and dt_binding_check passes on google.yaml.
  • checkpatch --strict: no errors, one warning (FILE_PATH_CHANGES). exynos/google/ is already in MAINTAINERS.
  • 11 DTBs build with W=1 and dtbs_check, with no new warnings. zumapro-comet.dtb on its own builds with zero.
  • Every commit is signed off by its author.

Merges after

Nothing.

🤖 Generated with Claude Code

@derekross

Copy link
Copy Markdown
Owner

Thanks, Nate. For a first board file this is the shape I'd hoped for: the binding first, the SoC description untouched, and everything beyond storage, keys, watchdogs and clocks described but disabled. I ran the gate on it even though it's a draft, since it is cheap for a board: it merges cleanly, all 11 device trees build with W=1 with no new warnings, and with dtschema run by hand your DTB validates with zero messages (and is rejected on its compatible by the base's schema, so the check is real). The board id and revision match Google's zumapro-comet-mp.dts, and the Hall switch matches their zuma-comet-hall-sensor.dtsi. CI is waiting for my approval; I'll approve it.

What has to change before it leaves draft:

  1. It can't be booted as it stands. USB is disabled, and USB is the phone's only link: our init panics when no USB host shows up, so this DTB would reset in a loop, and with your slot B marked there is no fallback, only the buttons. Please don't flash it as it is. The way through is the boot I asked for on usb: dwc3: Tensor G4 PHY-before-core glue, with system sleep (Pixel 9 Pro Fold) #271, with &usb as the 9a has it (plain snps,dwc3, four clocks, both PHYs on). If that works, enable USB here the same way: a node proven on a same-SoC phone may be added enabled. Then boot this exact DTB and put the tools/phone-smoke report and uname -r in the description. That boot is what turns "built and validated" into something I can merge.
  2. One comment is copied from the 9a and is wrong there too: "the init's storage guard …, which runs before the disks appear". UFS is built in, so the disks attach before the init runs; the guard marks the partitions read-only as soon as it starts. Please reword it (I'll fix the 9a's copy).
  3. The Link: in the second commit sits above a blank line, so git doesn't count it as a trailer. Removing the blank line fixes it.

What I did not check in this round, and will when it is ready: the CPU tables against stock's limits, the four extra reserved-memory regions, and the battery, charger and GPU properties (all disabled here). Two things to keep in mind for later: the Hall switch is a lid switch to logind, so its enabling commit goes out together with your device package's lid drop-in; and nothing that enables the MAX17201 may write its nonvolatile memory.

@derekross derekross left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

CI has run on this head and passes: clean merge onto today's omamobile, sign-offs, prefixes, checkpatch (only the "does MAINTAINERS need updating?" note for the new file, which it doesn't), all 11 Google device trees building with W=1 and passing dtbs_check with no new warning, and dt_binding_check on the binding. The one skipped line is "no C files touched". So the build side of the draft is confirmed by CI as well as by my hand run.

What it can't show is the boot, and that is still what this draft waits for: please don't flash it with USB disabled, and the three points in my comment above stand.

Add the Google Pixel 9 Pro Fold, a Tensor G4 board, as
google,zumapro-comet with the google,zumapro fallback, next to the
Pixel 9a. The bootloader's selector properties are the SoC's and are
already described.

Assisted-by: LLM
Signed-off-by: Nate Crawford <nate@kdegeek.com>
Add the Google Pixel 9 Pro Fold (comet), a Tensor G4 board, on the
zumapro SoC description the Pixel 9a uses, with what it needs to boot
from UFS with its keys, both watchdogs and the USB gadget:

 - the bootloader's selectors for comet's FOPLP and IPOP parts, its
   bootargs and the firmware carve-outs its own DT reserves, less the
   three it inserts itself;
 - the UFS supply on gpp0-1 and the key banks' interrupts;
 - comet's CPU levels from its own ACPM table (the inherited levels it
   doesn't have are deleted, those above what has run on it are
   turbo-mode) and its lower thermal trips;
 - the ramoops layout stock uses;
 - the Pixel 9a's plain DWC3 configuration for USB 2.0 peripheral mode
   with four clocks and both PHYs enabled.

Described but disabled, each to be enabled once it has its own review:

 - the MAX77779 charger, gauge and USB-C port controller, and the
   second cell's MAX17201;
 - the CS40L26A haptics, the LM3644 torch and the ST54L NFC controller;
 - the fold's Hall switch;
 - the GPU's ACPM clock and its levels up to 940 MHz.

The cover and inner displays, touch, cameras, AoC sensors and audio,
Wi-Fi, Bluetooth, the modem, fingerprint, UWB, the board thermistors
and wireless charging are left out: their drivers or bindings are not
in this tree yet, or need SoC nodes the zumapro description doesn't
have.

Tested on comet with a cold boot through the bootloader of this DT
(with the three inserted carve-outs still listed, which only logged
overlaps) on the v7.3-rc6 based 7.3.0-rc6-omamobile-comet-rc2-cand21
kernel: the board booted from UFS to the root file system, both
watchdogs were taken over, and dwc3 bound 11210000.usb directly; the
gadget was up 3.2 s into boot and the host configured it.

Link: https://android.googlesource.com/kernel/devices/google/comet/+/c826c913562277d8d7ee4c089d4a0aab54507637/dts/zumapro-comet-mp.dts
Assisted-by: LLM
Signed-off-by: Nate Crawford <nate@kdegeek.com>
@kdegeek

kdegeek commented Nov 18, 2026

Copy link
Copy Markdown
Author

Derek,

v3 is pushed. The binding commit is unchanged. Since v2, this board DT has had its cold boot on comet, and the board commit carries the result.

Changes in the board commit:

  1. USB is enabled the 9a's way: plain snps,dwc3, the four HSI0 clocks (bus_early, ref, suspend, subctl), peripheral mode at high speed, the same two quirks, and both PHYs on with the eUSB2 supplies.
  2. The three carve-outs ABL inserts itself are gone: pkvm-guest-firmware@b5000000, secure-dram@b6200000 and secure-page-table@e0000000. As zuma.dtsi already says, ABL adds pkvm_guest_firmware, sec_dram and sec_pt at the same ranges. With both present, the kernel logged OF: reserved mem: OVERLAP DETECTED! three times, once per pair. abl@f8800000 and the other carve-outs from comet's own DT stay.
  3. The UFS comment now reads: built-in UFS attaches the disks before the init runs, and the guard marks the partitions read-only as soon as it starts.
  4. Link: sits in the trailer block, and the message now gives the test.

The cold boot (2026-11-18):

  • Setup: ABL booted vendor_kernel_boot_b holding v2's DTB, rebuilt here byte for byte from c1425be8f, with the running kernel 7.3.0-rc6-omamobile-comet-rc2-cand21 and no other change. v3 differs from that DTB only by the three removed nodes.
  • Recovery: I wasn't at the phone, so the trial image's init wrote the daily vendor_kernel_boot_b back and checked its hash before switching to the root file system. If the DTB lost USB, the next boot would come back on its own.
  • USB: 11210000.usb bound to dwc3 directly, with no glue module. The gadget was up 3.17 s into boot, and the UDC read configured driver=dwc3. The host enumerated it, and the run was reached over that link.
  • Watchdogs: the board's /chosen/bootargs has no tmr_atboot, so s3c2410_wdt stops both watchdogs at probe (watchdog inactive, reset disabled). The phone then has no reset until userspace opens them. For this run, the trial init opened both right after its modules loaded and fed them until switch_root, and a feeder on the root took them over within the same second. The two watchdog did not stop! lines at 3.42 s are the init closing them under nowayout, not a fault. Whether comet's bootargs should carry tmr_atboot, so the phone is covered from probe, is a separate question.
  • tools/phone-smoke: boot, storage guard, root, kernel health and USB gadget were ok, all on slot B. Display, GPU, touch and session were not ok, and Wi-Fi, Bluetooth, battery, sound and sensors were skipped. All of these are disabled or left out in this DT, as the message says.
  • New dmesg lines against the daily tree:
    • The three overlaps, fixed above.
    • Five samsung-pinctrl …: irq number not available, for PERIC0, PERIC1, HSI1, HSI2 and HSI2UFS. These come from zumapro.dtsi on omamobile, whose five pinctrl nodes have no interrupts. This PR doesn't touch that file, and none of comet's nodes here uses a GPIO interrupt on those banks: the keys and the Hall switch are on ALIVE (gpa2, gpa4, gpa6). I'll send that separately.
  • Afterwards: the phone came back to the daily image on its own. vendor_kernel_boot_b matched the daily hash, and devinfo matched its 11-08 hash.

I'll keep in mind the lid drop-in that has to come with the Hall switch, and that nothing that enables the MAX17201 may write its NVM.

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