Repository navigation
omamobile-device-tegu 0.2.15: one sleep state, so a refused sleep isn't tried twice (#619) - #654
Conversation
…'t tried twice systemd's default SuspendState is "mem standby freeze", written to /sys/power/state one after the other until a write succeeds. With mem_sleep at s2idle the second write is the same sleep again, so on the Pixel 9a a sleep the modem's link refused was tried a second time half a second later, and in that second try brcmfmac failed to suspend and Wi-Fi was lost (derekross#619). The sleep drop-in now says SuspendState=freeze: one state, one try, and suspend-to-idle whatever mem_sleep says. The package test compared the drop-in byte for byte with the 7 Pro's, which gets the line in a change of its own; it now checks the five settings. Tested on the palmtop on 2026-11-10 with the modem off and the line in a runtime drop-in (7.3.0-rc6-omamobile-push1-dbg-gff4fdf58035e): cat-config showed SuspendState=freeze and MemorySleepMode=s2idle, one systemctl suspend with an RTC alarm logged "PM: suspend entry (s2idle)" and woke after 30 s, suspend_stats success 0 to 1, fail 0 to 0, no UFS error lines. The result is in MODEM-PLAN, "Suspend with the modem on, with the sleep fixes".
|
Thank you, this is what I asked for on #619, and it is as small as it should be. I checked it against today's main: only the 9a's package changes, the 7 Pro's drop-in is untouched, and the two packages that still compare their copy with the 7 Pro's are not affected. Replacing the byte-for-byte comparison with a check of the file's own five settings is right, since the 7 Pro's copy gets the line after its own test. CI is green and The test is the one I described: one I'm merging this as 0.2.15, and #590 takes 0.2.16. One thing for the palmtop: it runs your hand-built 0.2.17, so it won't pick this up by itself. When the tegu PRs are in, please install |
…'t tried twice (derekross#619) (derekross#654) systemd's default `SuspendState` is "mem standby freeze", written to `/sys/power/state` in turn until one is accepted; with `mem_sleep` at s2idle the second write is the same sleep again, so a sleep the modem's link refused was tried a second time half a second later. The 9a's sleep drop-in now sets `SuspendState=freeze`, and the package test checks the drop-in's own settings in place of comparing it with the 7 Pro's. Tested with one hook-run sleep on the palmtop (runtime drop-in, modem off); a refused sleep isn't shown yet.
…n't tried twice (#619) The Pixel 8 Pro's part of #619, as #654 did for the Pixel 9a: `SuspendState=freeze` in the sleep.conf drop-in, so systemd writes one state to `/sys/power/state` and a sleep a driver refuses isn't tried a second time; it stays suspend-to-idle whatever `mem_sleep` says. The test now checks the drop-in's own five settings instead of comparing with the Pixel 7 Pro's copy, which is unchanged. Tested on one 8 Pro (`7.3.0-rc6-omamobile-push1-g596d44e5bd5e`, the package built from the branch): one `systemctl suspend` woken by the RTC alarm, `suspend_stats` success 0 to 1, fail 0, Wi-Fi loaded and back after the wake.
The Pixel 9a's part of #619:
SuspendState=freezeinomamobile-device-tegu'ssleep.confdrop-in, so systemd writes/sys/power/stateonce per sleep and a sleep a driver refused isn't tried a second time half a second later.vendor/systemd/sleep.conf.d/10-omamobile.conf: the line, and a comment saying whyfreezeand notmem.test/all: the drop-in was held byte for byte to the 7 Pro's. That comparison is replaced by a check of the five settings, since the 7 Pro's copy gets the line in a PR of its own.device/tegu/MODEM-PLAN.md("Suspend with the modem on, with the sleep fixes") say what is set and record the test.Test
One sleep through the hooks on the palmtop, 2026-11-10, modem off for the boot, on
7.3.0-rc6-omamobile-push1-dbg-gff4fdf58035e(the published tip plus kernel #224, the build of that evening's series). The line came from a drop-in in/run/systemd/sleep.conf.dwith the sameSuspendState=freeze, not from the installed package (the palmtop runs a hand-built 0.2.17 from the bring-up branches). The drop-in was removed after the test.cat-configissystemd-analyze cat-config systemd/sleep.confwithout its comment lines; the wake was the 25 s RTC alarm (irq 53 is the PMIC's line on that build).What it doesn't show: a refused sleep. That needs the modem up, so it waits for an attended run, as #619 says, and the Wi-Fi loss there also happened with one state, so this line alone isn't claimed to fix it.
packages/omamobile-device-tegu/test/allandtools/pr-check origin/mainpass here.tests/packagespasses exceptomamobile-omarchy, which this doesn't touch and which fails on this laptop on main too.