Skip to content

Pixel 8a (akita): touchscreen, its supplies, and the panel size - #241

Merged
derekross merged 4 commits into
derekross:omamobilefrom
aninmucek:omamobile-zuma-akita-touch
Nov 12, 2026
Merged

derekross merged 4 commits into
derekross:omamobilefrom
aninmucek:omamobile-zuma-akita-touch

Conversation

@aninmucek

@aninmucek aninmucek commented Nov 11, 2026 •

Copy link
Copy Markdown
Contributor

The Pixel 8a's touchscreen, its supplies, and the panel's physical size. This finishes akita's M3 (input) after #238's buttons.

Stacked on #240 (the S2MPG14 supply setter): until it merges, the diff also shows its commit, 7d143b515. The three commits that are this PR:

  • zuma-akita: Add the touch supplies. The S2MPG14 on ACPM, with LDO4M at stock's 3.0 V and LDO25M at 1.8 V, both fixed. The values come from the stock phone's live device tree. This is zuma-husky's node, except for avdd: the 8 Pro's is 3.3 V. It needs regulator: s2mps11: Let the S2MPG14 touch supplies set their voltage #240, because without a setter the core can't move ABL's 3.3 V and the probe fails with -EINVAL.
  • zuma-akita: Add the touchscreen. A Goodix GT9895, goodix,gt9895 with 14-byte contact records, as on the 7a.
    • Why a GT9895: both of stock's firmware files, goodix_firmware_ak3a.bin and _ak3b.bin, carry 9895 in their Berlin header.
    • Wiring: zuma-husky's bus and lines: USI2_HSI0's SPI at 10 MHz, interrupt on gpn0-0 (falling edge, stock's goodix,irq-flags = 2), reset on gpp21-2. Pin groups come from zuma-pinctrl.dtsi.
    • No firmware: the driver loads none.
  • zuma-akita: Add the panel's physical size. zuma-husky's disabled panel0 node, with 64×141 mm, the lit area measured with a ruler.
    • Without it, simpledrm guessed 285×635 mm, and the on-screen keyboard came out tiny.
    • The source: 64×141 is what a 6.1-inch 20:9 panel gives (63.6×141.3), and the height Google's touch node names (goodix,panel-height-mm = <141>).
    • Stock's own figure is larger than the glass. The ak3b's width_mm/height_mm in Google's panel-google-ak3b.c are 70×149, and stock's DPI follows from them (xDpi 391.885, yDpi 409.127). The comment names both. (Updated after your review; the first version used 70×149.)

Tested on a Pixel 8a from slot B (7.3.0-rc6, fc7086a68 + #240 + these three; the last boot was release …g3933790300a1), one change per boot, as you asked on #238:

  1. The supplies with no consumer: touch-avdd: Bringing 3300000uV into 3000000-3000000uV. The rails read back 3.0 V and 1.8 V, disabled, with 0 users (the evidence in regulator: s2mps11: Let the S2MPG14 touch supplies set their voltage #240).

  2. The touchscreen, in a later boot: with goodix_berlin_spi held back by a modprobe blacklist, the rails read 3.0 V and 1.8 V and off. Then the driver was loaded by hand:

    • it enabled the rails and registered "Goodix Berlin Capacitive TouchScreen";
    • touches in the four corners came in at (110,121), (1004,118), (118,2311) and (971,2298), so no axis swap or inversion is needed;
    • a swipe down the middle ran from (528,824) to (564,1892);
    • 842 interrupts gave no checksum errors.

    The blacklist was then removed.

  3. The next boots, with the driver loading by itself at 3.6 s:

    • the PIN pad unlocks by touch;
    • apps open and the shell swipes;
    • the on-screen keyboard types;
    • Power locks and wakes.
  4. The panel size: first with stock's 70×149 mm. Then, after your review, with the measured 64×141 mm: Hyprland reports 64×141, and the keyboard is at its normal height. That's the tree here: zuma-akita.dts is blob 3714d9502168.

What isn't covered:

  • The session is slow: there's no GPU and no cpufreq on akita yet, so everything renders in software at ABL's CPU rates.
  • Once, closing an app from the app switcher made Hyprland drop the shell's Wayland connection ("error in client communication", then "Invalid argument"), and the shell relaunched. It didn't happen in the next try. That's userspace, and it'll be tracked in omarchy-mobile if it comes back.
  • Suspend isn't tried, as before.

tools/kernel-pr-check --build OUT --merge origin/omamobile, with the dtschema validator, passes everything over the four commits (#240's included):

  • a clean merge onto fc7086a68;
  • the sign-offs: 4 commits, each signed off by its author, none by an AI;
  • checkpatch --strict with no errors and no warnings;
  • the policy checks;
  • drivers/regulator/s2mps11.c builds with W=1;
  • the 10 Google device trees build with W=1 and pass dtbs_check, with no new warnings.
tools/phone-smoke report (the 64×141 boot, omarchy-mobile main)

phone-smoke: akita

Kernel 7.3.0-rc6-omamobile-push1-g3933790300a1, tools at 2db3f2b, 2026-11-11T16:33Z.

Check Result Detail
boot (required) ok kernel 7.3.0-rc6-omamobile-push1-g3933790300a1, slot b, up 30s
storage guard (required) ok 62 partitions; only userdata and slot B's boot images writable
root (required) ok ext4 on userdata, 98G free
kernel health (required) ok no oops or BUG; 0 WARN, 2 error lines in dmesg (2 known, 0 new)
systemd units ok none failed
USB gadget (required) ok bound to 11210000.usb
display ok card0-Unknown-1 connected, 1080x2400
GPU not ok no render node
session (required) ok Hyprland and the Omarchy shell running
lock ok PAM file present, the shell can lock (passwordPam true, locked true)
touch ok Goodix Berlin Capacitive TouchScreen
buttons ok gpio-keys
Wi-Fi skip no wireless interface
Bluetooth skip no controller
battery skip no battery power supply
sound cards skip none
sensors skip no sensor proxy
cameras skip no libcamera tools

Packages: hyprland 0.57.0-3.2, libcamera 0.7.2-4.23, mesa 1:26.2.4-1.4, omacam 0.3.3-1, omamessages 0.2.2-1, omamobile-apps 0.2.2-1, omamobile-device-akita 0.1.1-1, omamobile-firstboot 0.1.3-1, omamobile-keyring 20261026-1, omamobile-mirror 0.1.1-1, omamobile-omarchy 0.1.38-1, omamobile-owl-widgets 0.1.2-1, omamobile-shell 0.2.108-1, omamobile-voice 0.1.0-1, omaphone 0.2.3-1, omarchy 4.0.6-1, omatel-agent 0.2.4-2, quickshell 0.3.2.r7.g2b4501a-1

🤖 Generated with Claude Code

ABL can leave LDO4M at another voltage than the touch controller needs:
on a Pixel 8a it leaves 3.3 V, read back through the driver's own
get_voltage_sel, while the stock phone runs the Goodix GT9895 there at
3.0 V (its live device tree). With voltage selectors firmware-owned, a
board that fixes 3.0 V fails to probe ("failed to apply
3000000-3000000uV constraint: -EINVAL"), and one without a constraint
would power the controller at ABL's 3.3 V.

Give LDO4M and LDO25M a set_voltage_sel through the regmap, which goes
to the PMIC through ACPM as enable does. The selector tables are
Google's: s2mpg14-regulator.c puts LDO4M in voltage group 5 (1.8 V
minimum) and LDO25M in group 4 (0.7 V), each with 64 steps of 25 mV
(s2mpg14-register.h), and writes the selector into the low six bits of
the control register (0x2e and 0x43), as the descriptors here already
read it. The other S2MPG14 outputs keep firmware-owned selectors.

The core calls the setter only to bring a voltage into the board's
constraints, so a board without voltage constraints (zumapro-tegu), or
whose constraint already matches what ABL left (zuma-husky's fixed
3.3 V, which probes today without a setter), sees no write.

Tested on a Pixel 8a from slot B (7.3.0-rc6), with LDO4M fixed at
3.0 V and LDO25M at 1.8 V in the board file and nothing consuming them:
the core logs "touch-avdd: Bringing 3300000uV into 3000000-3000000uV";
both rails then read back 3000000 and 1800000 uV (the PMIC regmap has
no cache, so that is the register through ACPM), disabled, with no
users, and stay so past the unused-regulator cleanup. Nothing enabled
them in that boot.

Link: https://android.googlesource.com/kernel/google-modules/soc/gs/+/204078758737776959ecc85c527a00fd2cabb62f/drivers/regulator/s2mpg14-regulator.c
Link: https://android.googlesource.com/kernel/google-modules/soc/gs/+/204078758737776959ecc85c527a00fd2cabb62f/include/linux/mfd/samsung/s2mpg14-register.h
Assisted-by: Claude Opus 5.5
Signed-off-by: DarekM <149845949+aninmucek@users.noreply.github.com>
The Pixel 8a's touch controller takes vdd from the S2MPG14's LDO25M at
1.8 V and avdd from its LDO4M at 3.0 V (the stock phone's live device
tree; Google's zuma-akita-touch.dtsi fixes both and drops always-on).
Describe the main PMIC on ACPM with those two regulators, as zuma-husky
does, but with akita's 3.0 V avdd rather than the Pixel 8 Pro's 3.3 V.

Nothing uses them yet: the touchscreen node follows in its own commit.

Tested on a Pixel 8a from slot B (7.3.0-rc6, with the S2MPG14 setter):
the core logs "touch-avdd: Bringing 3300000uV into 3000000-3000000uV"
and both rails read back 3000000 and 1800000 uV, disabled, with no
users, in a boot with no consumer.

Assisted-by: Claude Opus 5.5
Signed-off-by: DarekM <149845949+aninmucek@users.noreply.github.com>
The Pixel 8a's touchscreen is a Goodix GT9895 (Berlin) on USI2_HSI0's
SPI, the bus, interrupt and reset lines of the Pixel 8 Pro's: 10 MHz,
interrupt on gpn0-0 (falling edge, as stock's goodix,irq-flags = 2),
reset on gpp21-2 (Google's zuma-akita-touch.dtsi; the stock phone's
/proc/interrupts lists goodix_ts on gpn0 0 Edge). Both of stock's
firmware files, goodix_firmware_ak3a.bin and goodix_firmware_ak3b.bin,
carry "9895" in their Berlin header, so it takes the gt9895 match data
(14-byte contact records), as the Pixel 7a does. The driver runs the
firmware already on the chip and loads none.

The supplies are the LDO4M and LDO25M described with the PMIC, at
stock's 3.0 V and 1.8 V.

Tested on a Pixel 8a from slot B (7.3.0-rc6), in a later boot than the
one that first set the voltage: with the driver held back, the rails
read 3.0 V and 1.8 V and off; loading goodix_berlin_spi enabled them
and registered "Goodix Berlin Capacitive TouchScreen". Touches in the
four corners came in at (110,121), (1004,118), (118,2311) and
(971,2298), so no axis swap or inversion is needed, and 842 interrupts
gave no checksum errors. In the next boots the driver loads by itself,
and the PIN pad, the shell and the on-screen keyboard work by touch.

Assisted-by: Claude Opus 5.5
Signed-off-by: DarekM <149845949+aninmucek@users.noreply.github.com>
@derekross

Copy link
Copy Markdown
Owner

Thank you, @aninmucek. This went the way I hoped after #238: the supplies read back at 3.0 V and 1.8 V with nothing connected before the controller was ever powered, the driver held back for its first load and released afterwards, and four corners plus 842 interrupts with no checksum errors to show the 14-byte records are right. The 8a is usable by hand now, which is a real step.

What I checked:

  • Our gate passes on today's omamobile (fc7086a68) over the four commits: a clean merge, the sign-offs, checkpatch --strict with nothing to read, the W=1 build, and the ten device trees with no new warnings. CI is green, including dtbs_check.
  • The touch node against Google's zuma-akita-touch.dtsi: 10 MHz, the interrupt on gpn0-0 falling, the reset on gpp21-2, 1080x2400, and the two ak3a/ak3b firmware names. Stock doesn't override the two pin groups, so the shared ones are right.
  • usi2_hsi0 and spi20 are disabled in zuma.dtsi and you enable both.
  • The supplies are stock's values with min = max, as on the 8 Pro apart from avdd.

One thing before I merge, and one question.

1. The panel size. Your citation is right: panel-google-ak3b.c says 70 and 149, and stock's DPI follows from it. But that isn't the size of the glass. A 6.1-inch 1080x2400 panel is about 64 × 141 mm, the same Google tree's touch node says goodix,panel-height-mm = <141>, and akita's NOTES said until now that 70 × 149 "is larger than a 6.1-inch 20:9 panel (about 64 × 141 mm), so no size goes into the board file yet". width-mm and height-mm describe the hardware, and with 70 × 149 everything sized in millimetres comes out about 9 % small. Could you measure the lit area with a ruler and use that, with a comment that names Google's 70 × 149 as stock's figure? If you'd rather keep stock's scale, tell me why and say in the comment that the value is stock's and larger than the glass. If you want touch in sooner, you can also drop the third commit and send the size on its own.

2. Which tree booted? The smoke report is from 01:14Z on g7b13e4166249, and the three commits here are dated 01:23Z, so I assume you amended the messages with the results afterwards. Could you confirm that zuma-akita.dts in the tested tree is blob 5d7c30781b80 (git rev-parse <tested>:arch/arm64/boot/dts/exynos/google/zuma-akita.dts)?

On merging: this PR carries #240's commit, so I'll rebase-merge this one, which lands #240's commit first, and close #240 as landed. No rebase needed from you for that.

Without a panel node, simpledrm estimates the display's size from its
pixels at 96 dpi, so the compositor sees a 285x635 mm screen and the
on-screen keyboard comes out a fraction of its height. Add a disabled
panel node with width-mm and height-mm and point the framebuffer at it,
as zuma-husky does.

64x141 mm is the lit area, measured with a ruler on a Pixel 8a: what a
6.1-inch 20:9 panel gives (63.6x141.3 mm), and the height Google's
touch node names (goodix,panel-height-mm = <141>). Stock's own figure is
larger than the glass: Google's panel-google-ak3b.c gives the ak3b
70x149 mm, and the stock phone's DPI follows from it (dumpsys display:
xDpi 391.885, yDpi 409.127 for 1080x2400), so stock sizes everything
about 9 % small. The comment names both.

Tested on a Pixel 8a from slot B (7.3.0-rc6): Hyprland reports the
monitor as 64x141 mm (285x635 mm without the node), and the on-screen
keyboard comes out at its normal height. An earlier boot with stock's
70x149 mm also worked, with everything about 9 % smaller.

Assisted-by: Claude Opus 5.5
Signed-off-by: DarekM <149845949+aninmucek@users.noreply.github.com>
@aninmucek
aninmucek force-pushed the omamobile-zuma-akita-touch branch from 0d0f8ec to 9613d79 Compare November 11, 2026 16:53
@aninmucek

Copy link
Copy Markdown
Contributor Author

Thank you, and you're right about the panel size. I took stock's figure without weighing it against our own NOTES.

  1. The panel size. The lit area measures 64 × 141 mm with a ruler, so the third commit (now 9613d7934) carries 64 × 141. Its comment and message say that's the measured glass, matching a 6.1-inch 20:9 panel and the touch node's 141. They also name Google's 70 × 149 as stock's figure, larger than the glass, with stock's DPI following from it. I booted it on the 8a before re-signing:

    • Hyprland reports 64 × 141 mm, and the keyboard is at its normal height;
    • running, with no failed units;
    • phone-smoke (main 2db3f2b): every required row ok, touch and buttons ok, 0 new error lines.

    The description and its smoke report are updated.

  2. Which tree booted. Confirmed: the tested commit for the earlier version was 7b13e4166 (release …g7b13e4166249), and git rev-parse 7b13e4166:arch/arm64/boot/dts/exynos/google/zuma-akita.dts gives 5d7c30781b80, the PR's blob at the time. I'd added the "Tested on" paragraphs to the messages afterwards, and nothing else changed. For this push, the booted tree is 393379030 (release …g3933790300a1), with blob 3714d9502168, the same as the new head 9613d7934.

The other three commits are unchanged, so #240's commit is still 7d143b515. The gate passes again on fc7086a68: a clean merge, 4 sign-offs, checkpatch with nothing to read, the W=1 build, and the ten device trees with no new warnings.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @aninmucek. Both points from last time are done.

  • The panel size. panel0 now carries the measured 64 × 141 mm, and the comment says what I hoped it would: the lit area by ruler, what a 6.1-inch 20:9 panel gives, the 141 of Google's touch node, and Google's 70 × 149 named as stock's figure, larger than the glass. Thanks for booting it again before re-signing.
  • Which tree booted. Confirmed from my side too: the head's zuma-akita.dts is blob 3714d9502168, the one you booted.

The other three commits are unchanged, and #240's commit is still 7d143b515. Our gate passes on today's omamobile (fc7086a68): a clean merge, the sign-offs, checkpatch --strict with nothing to read, the W=1 build and the ten device trees with no new warnings.

One thing on my side, not yours: the CI job for this push was cancelled while it was still installing its tools, so it never reached the checks. I'll re-run it, and once it's green I'll rebase-merge this, which lands #240's commit first, and close #240 as landed. omarchy-mobile#650 follows right after.

The 8a has buttons and touch now. Nice work.

@derekross
derekross merged commit a7356e9 into derekross:omamobile Nov 12, 2026
1 of 2 checks passed
@aninmucek
aninmucek deleted the omamobile-zuma-akita-touch branch November 12, 2026 06:22
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