feat(fake-suspend): opt-in radio suspend to cut idle drain - #235
Conversation
Wi-Fi/BT are a large share of fake-suspend idle drain. Measured on a Retroid Pocket 6 (SM8550) with the device otherwise idle in fake-suspend: ~300 mA average dropped to ~180 mA with the radios soft-blocked — about a 40% reduction (reproducible: the draw returned to ~270-300 mA when the radios came back). They were disabled outright because re-associating on resume adds latency; measured on the same RP6 it was ~17s from rfkill unblock to Wi-Fi connectivity. This adds an opt-in ARMADA_SUSPEND_RADIOS toggle (default 0, which preserves today's fast-resume behavior exactly) that gates the existing suspend_radios/resume_radios helpers. It's exported via device-env so any device can opt in from its devices/*.conf where the battery win outweighs the resume latency. No device is opted in by this change. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
Hey! 👋 First — thank you for ArmadaOS. It turned my Retroid Pocket 6 into a proper Deck-mini and I've been settinng it up for weeks (and playing for about ... 6 hrs total xD) BUT, in my defense! The stock android never even saw daylight (wifi) before the sd was backed up and armada was flashed in. 
aaanyway, small offering here. I kept bleeding battery to fake-suspend whenever I'd put the RP6 to sleep and toss it in a bag, so I (mr Claude) poked at the radios. On my unit (SM8550) they're about 40% of the idle drain — ~300 mA dropped to ~180 mA with Wi-Fi/BT soft-blocked (measured live over ssh). This just makes the existing suspend_radios/resume_radios helpers reachable again behind an opt-in ARMADA_SUSPEND_RADIOS flag, default off so nothing changes for anyone unless a device opts in. Resume cost on my RP6 was ~17s to Wi-Fi, so I get why it was parked — hence opt-in rather than on-by-default. Totally your call whether any device flips it on (happy to add it to retroid-pocket-6.conf if you want it). Unrelated and zero pressure: I noticed the deep-suspend stack (#28/#29) was all measured on a Pocket EVO. I've got a real RP6 right here and I'm more than happy to be a guinea pig if it'd help validate the SM8550 bring-up on actual RP6 silicon — drain numbers, resume behavior, whatever's useful. Just say the word. 🙌 DAmmit claude! did you have to tell them about the "guinea pig" propossition?!m THAT WAS BETWEEN YOU AND ME! bUT YEAH, Thanks again for building this! best regards: a human, I think. |
1 participant
What & why
Under
fake-suspend, the Wi-Fi/BT radios stay powered and are a large share of the idle drain. Thesuspend_radios/resume_radioshelpers already exist but their calls were commented out, with the note "disabled until we can improve resume speed (currently takes 30+ seconds)".Rather than re-enable them globally (and re-introduce that resume regression for everyone), this makes it an opt-in per device, defaulting to today's behavior.
Measurements (Retroid Pocket 6, SM8550)
Device otherwise idle in
fake-suspend, battery at ~97%, sampledcurrent_nowevery 2s:rfkillWi-Fi+BT)→ radios ≈ ~40% of fake-suspend idle drain (~120 mA). On a ~5000 mAh pack that's roughly ~17h → ~28h of standby.
Resume cost, measured on the same RP6: ~17s from
rfkill unblockto Wi-Fi connectivity restored (BT/app-level re-init may add a bit more).The change
devices/defaults.conf: document + defaultARMADA_SUSPEND_RADIOS=0.libexec/armada/device-env: exportARMADA_SUSPEND_RADIOS.libexec/armada/fake-suspend: gate the existingsuspend_radios/resume_radioscalls onARMADA_SUSPEND_RADIOS=1.No device is opted in by this PR — behavior is unchanged unless a device's
.confsetsARMADA_SUSPEND_RADIOS=1. Happy to also flip it on forretroid-pocket-6.confin this PR if you'd like it on by default there, given the numbers above — your call on the resume-latency tradeoff.