Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
34 changes: 28 additions & 6 deletions device/tegu/SLEEP-PLAN.md
Original file line number Diff line number Diff line change
@@ -1,12 +1,13 @@
<!-- SPDX-License-Identifier: MIT -->
# Pixel 9a sleep: why s2idle drains so much

Status (2026-11-01): s2idle usually works (RTC and power-key wakes; ten of
ten 30-second cycles on 10-31 with the console kept through the freeze),
but not always: four sleeps have never woken (two direct freezes, one
30-minute systemd sleep with `tegu-mif-sleep` loaded, and the sixth
30-second cycle of a boot on 10-30, which left the USB bus and needed a
forced restart), and one more woke with UFS dead (`link startup failed` x5,
Status (2026-11-02): s2idle usually works (RTC and power-key wakes; ten of
ten 30-second cycles on 10-31 with the console kept through the freeze;
126 of 127 120-second sleeps overnight on 11-01 to 11-02), but not always:
six sleeps have never woken: two direct freezes, one 30-minute systemd
sleep with `tegu-mif-sleep` loaded, the sixth 30-second cycle of a boot
on 10-30 (it left the USB bus and needed a forced restart), the 21st of
11-01's thirty cycles, and the overnight series' 127th; one more woke with UFS dead (`link startup failed` x5,
then I/O errors and `emergency_ro`, 10-30 afternoon). The drain in sleep is
high. Sleep tests are attended. The shell's auto-sleep (the `omamobile-sleep`
toggle, which the package ships off) has been on on the palmtop since
Expand Down Expand Up @@ -527,6 +528,27 @@ with the cable in.
1.2 s to about 0.25 s; before that, a second press often looked
necessary. So none of the day's sleeps hung. This adds to the evidence
but doesn't retire the hang of 10-28 to 10-30 either.

**An overnight series, 2026-11-01 to 11-02 (same release, unattended
through the night).** Two hundred 120 s sleeps were planned through
`systemctl suspend` with an RTC alarm, auto-sleep off for the run, the
console kept through the freeze (`printk.console_suspend=N`), cable in,
modem never loaded, NFC polling stopped by the sleep hook, Wi-Fi kept
loaded. 126 sleeps woke: `suspend_stats` success 126 and fail 0, wake IRQ
56 every time, no UFS or ACPM lines, and Wi-Fi came back after each
(brcmfmac re-probes after every wake: "device lost its state in sleep").
The 127th didn't wake (about 22:42 by the palmtop's clock): no SSH over
USB or Tailscale for nine hours, and the owner's Power + Volume Up in the
morning restarted it. It entered that sleep like the 126 before it, with
the same NFC notifications, hooks and timings, and the journal stops at
`Reached target Sleep` and the user slice's freeze. Nothing was kept:
`/var/lib/systemd/pstore` held only an older entry, so with the console
kept through the freeze, a Power + Volume Up reset doesn't keep the
ramoops region either (the console never reached it, or the reset
cleared it). That makes it one hang in 127 here, against one in 21 on
11-01 and about one in 40 before; the rate varies too much to compare
conditions with short runs. Logs: `logs/tegu-sleep-overnight-20261102/`
(the series script, `after-hang.txt`).
- **What the five clean cycles rule out, and what they don't.** Five in a
row with the storage counters at zero says the UFS resume path is not
broken per se; the afternoon failure and this hang are both the sixth or
Expand Down
Loading