Skip to content

arm64: dts: exynos: google: zumapro-tegu: the main PMIC's interrupt, so the RTC alarm wakes the Pixel 9a - #236

Merged
derekross merged 3 commits into
derekross:omamobilefrom
W4HO-ham:tegu-dt-pmic-irq
Nov 10, 2026
Merged

derekross merged 3 commits into
derekross:omamobilefrom
W4HO-ham:tegu-dt-pmic-irq

Conversation

@W4HO-ham

Copy link
Copy Markdown
Contributor

On the published tree the Pixel 9a's main PMIC has no interrupt: sec-pmic-acpm firmware:power-management:pmic-1: no interrupt, so no RTC alarm at boot, rtc0 without wakealarm, and a suspend armed with rtcwake -m no -s 60 slept until a key was pressed (2026-11-09, 7.3.0-rc6-omamobile-push1-gf693b36e62ce, the published zumapro-tegu.dtb; rtcwake: set rtc wake alarm failed: Invalid argument). The owner's integration tree has had the interrupt since October, which is what every RTC-woken series in omarchy-mobile#409 woke on (pm_wakeup_irq 56).

Three commits, the way the gs201 boards describe theirs: the google,zumapro-vgpio2ap-sysreg compatible in the sysreg binding (no clock, as gs201's), the VGPIO2AP syscon at 0x15090000 in zumapro.dtsi, and interrupts = <GIC_SPI 74 IRQ_TYPE_LEVEL_HIGH>, google,vgpio2ap-sysreg and wakeup-source on pmic-1 in zumapro-tegu.dts. The samsung,s2mpg14-pmic binding already allows all three. Nothing else changes; the Pixel 7 Pro's files aren't touched.

Checks: tools/kernel-pr-check passes; zumapro-tegu.dtb builds and CHECK_DTBS=y reports nothing on it.

On the palmtop (2026-11-09): omamobile at abcd6aac51cf with these three commits and #224's two, 7.3.0-rc6-omamobile-push1-g6c9420bfa77f, built with defconfig + kernel/configs/omamobile.config, its own modules installed. The boot has no "no RTC alarm" line and rtc0 has wakealarm; 36 of 36 systemctl suspend sleeps armed with rtcwake -m no -s 60 woke on the alarm after 59.7 to 60.2 s, pm_wakeup_irq 52 (the PMIC's interrupt) every time, no UFS error line. The same series is running again with Wi-Fi up (the build with omarchy-mobile#274's PHY option); details are on #224 and in SLEEP-PLAN on omarchy-mobile#409.

Tensor G4 (zumapro) latches the main PMIC's in-band interrupt in a
VGPIO2AP system register block at 0x15090000, as gs201 does at
0x182f0000; the PMIC driver reads it through the google,vgpio2ap-sysreg
phandle of the samsung,s2mpg14-pmic binding. Add the compatible, with no
clock, as gs201's.

Signed-off-by: W4HO <87742609+W4HO-ham@users.noreply.github.com>
The system registers that latch the main PMIC's in-band interrupt (GIC
SPI 74 on this SoC), the way gs201.dtsi describes gs201's at 0x182f0000.
The PMIC node points at it with google,vgpio2ap-sysreg.

Signed-off-by: W4HO <87742609+W4HO-ham@users.noreply.github.com>
…rrupt

Without an interrupt the S2MPG14 driver supplies its rails but none of
the PMIC's events reach the host ("no interrupt, so no RTC alarm" at
boot), so rtc0 has no wake alarm: on the Pixel 9a a suspend with an RTC
alarm armed slept until a key was pressed (2026-11-09, on the published
tree). The interrupt is GIC SPI 74, latched by the VGPIO2AP sysreg, as
the gs201 boards' nodes have it; wakeup-source makes the alarm a wake
source, the first one for suspend.

Signed-off-by: W4HO <87742609+W4HO-ham@users.noreply.github.com>
@derekross

Copy link
Copy Markdown
Owner

Thank you, @W4HO-ham. This is the missing piece for timed wakes on the published tree, and it's put together the way I like to see it: binding first, then the SoC node, then the board.

I checked the three values against Google's GPL tree rather than taking them from the description: SYSREG_VGPIO2AP is 0x15090000 and INTC0_IPEND is 0x290 in soc/gs's s2mpg14-register.h, their IRQ thread reads and writes back that register as our sec-irq.c does, and IRQ_INTCOMB_VGPIO2AP_ALIVE is 74 in the zumapro interrupt header. The PMIC binding already allows the three properties, the sysreg schema applies clocks: false to the new compatible, and zumapro.dtsi is only used by the 9a. Our gate passes on today's omamobile (clean merge, all three signed off, checkpatch --strict clean, no new device-tree warnings), CI is green, and it merges cleanly with #237 in either order. 36 of 36 on the alarm is a good record.

Two small things, only if you push again; I won't hold the merge for them:

  • in zumapro.dtsi, syscon@15090000 would sit by address between pinctrl@15060000 and mailbox@15110000;
  • the last commit says wakeup-source makes the alarm "the first" wake source for suspend, but the power key has had wakeup-source all along. "The first timed one" would be right.

I'll merge this one first, then #237.

@derekross
derekross merged commit 8af10a2 into derekross:omamobile Nov 10, 2026
1 check passed
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