Switch the Pocket EVO to real suspend to RAM

[jesherman]

Important

Do not merge before armada-packages#23. On a kernel without those patches, mem is the configuration that hangs the EVO on resume with a Steam session running — recoverable only by holding the power button.

What changed

• ayaneo-pocket-evo.conf: ARMADA_SUSPEND_MODE=mem
• New sleep.conf.d drop-in pinning SuspendState=mem — systemd's default mem standby freeze falls through to s2idle on a failed deep suspend, which wedges SM8550 hard enough to need a power-button reset (we hit this during bring-up). ROCKNIX pins the same for this SoC.
• README: sleep section still described fake suspend as universal, stale since the Odin 3 moved to mem

Results

Pocket EVO, on battery, Steam running: 27.8 mA vs 161.8 mA standby — ~2 days to ~12. 1202s of uninterrupted deep suspend woken only by the RTC alarm, success=3 fail=0, session intact on resume.

Gaps

Headline numbers are from a 7.0.11 build; packages has since moved to 7.1.5, confirmed working but only over a ~4 minute run. No multi-hour soak, single drain sample per mode, one device. The 5.8× gap is well beyond sampling error, but holding this until there's a longer soak on 7.1.5 is a reasonable call.

🤖 Generated with Claude Code

[jesherman]

Marking this draft — found a blocker, and it's one my testing should have caught much earlier.

The power button does not wake the device from deep suspend. Pressing power suspends fine, then pressing it again does nothing; the device is genuinely asleep and unwakeable, which presents as a hang and needs a hold-power reset.

Every "successful" result I reported was woken by an RTC alarm (rtcwake or a wakealarm write). I never tested the actual user-facing wake path. That's a significant gap in the validation behind the numbers above.

Evidence, from a suspend with a 120s RTC backstop and a power press partway through:

pm_wakeup_irq=202                → pmic_arb RTC alarm
rtc@6100   wakeup count 1 → 2    → RTC fired
pwrkey     wakeup count 0 → 0    → power key registered nothing

pmic_pwrkey shows wakeup = enabled and pon@1300:pwrkey is registered as a wakeup source, so it is armed in principle — but it never fires from deep. The RTC proves PMIC-side IRQs can wake the SoC, so this looks specific to the pwrkey path rather than PMIC wake generally.

The kernel work in armada-packages#23 is unaffected — deep suspend itself holds fine and the drain numbers stand. It is only this PR, flipping the EVO over to mem, that must wait until power-key wake works.

Next step is finding why the pwrkey doesn't arm a PDC wake on SM8550. Leaving this in draft until then.

[jesherman]

replaced by #228