No log channel survives s2idle here: netconsole is gagged and its netdev is down

scope: device:google-taimen · severity: trap · confidence: proven · subsystem: debug

taimen 7.2.2 #76, 2026-09-20. ph-capture.sh armed netconsole on BOTH transports and verified them end to end (‘netconsole VERIFIED end to end’, usb0 6666 and wlan0 6667 both live), with watchdog gov=panic pretimeout=10 timeout=30. Across three watchdog resets in s2idle it delivered NOTHING after ‘printk: Suspending console(s)’. Setting /sys/module/printk/parameters/console_suspend=N removed that line and still delivered nothing, so the transport is the blocker, not the console gag. pstore has never worked on 7.2.2 and usb0 drops off the bus in suspend (tk_device_state reads ABSENT). Also: /sys/class/watchdog/watchdog0/timeout is -r–r–r– and cannot be written at all – the handoff recipe ‘echo 60 > …/timeout’ fails with ‘Permission denied’ even as root, because the watchdog core only settles the timeout through the WDIOC_SETTIMEOUT ioctl, and systemd owns the device at RuntimeWatchdogUSec=30s

Symptom — you arm every channel, the tool says the channel is verified end to end, the phone takes a watchdog reset inside s2idle, and you get nothing:

12:23:52 [nc:6666] [  687.981885] printk: Suspending console(s) (use no_console_suspend to debug)
<nothing, ever>

Cause, in two layers. The first is the obvious one and it is not the real one. netconsole is a console, so suspend_console() gags it on the way down. That is fixable at runtime, no rebuild:

echo N > /sys/module/printk/parameters/console_suspend

Do that and the Suspending console(s) line disappears – and the capture is still empty. The second layer is the transport: netpoll has to transmit through a netdev that suspend has taken down. wlan0 still receives while suspended, which is why a magic packet wakes this phone, but that is the firmware’s doing and it does not give netpoll a working TX path.

So on this device, for a failure inside s2idle:

channel	why it cannot witness
netconsole / usb0	the gadget drops off the bus in suspend – `tk_device_state` reads ABSENT
netconsole / wlan0	receives, cannot transmit; console gag is only the first half
pstore / ramoops	has never produced a record on 7.2.2 – pstore-has-never-worked-on-7-2
`dmesg` / journal	die with the reset
watchdog pretimeout panic	`gov=panic pretimeout=10` is armed and still reaches no transport

And the disarm recipe that would have rescued it does not exist. /sys/class/watchdog/watchdog0/timeout is -r--r--r--:

/tmp/ph-prep.sh: line 10: can't create /sys/class/watchdog/watchdog0/timeout: Permission denied

Root does not help. The watchdog core exposes timeout read-only and settles it only through the WDIOC_SETTIMEOUT ioctl, and systemd already owns the device at RuntimeWatchdogUSec=30s. Changing it means RuntimeWatchdogSec= plus a re-exec of PID 1, which is itself a known way to freeze this phone. A run that believed it had disarmed the watchdog was really just another armed run, and its prep-hook error was the only tell.

What to do instead — stop buying a witness and make the oracle binary. ph-suspend-cycle.sh fsyncs every line before the suspend write, so btime - SUSPEND_ENTER survives, and a reset is a new boot_id. Then bisect by toggle: one variable per cycle, interleaved, and count resets. That is how the armed-tap hang was pinned in under an hour with no kernel log at all – a-tap-on-an-armed-suspended-screen-hangs-the-resume.

pm_test is the one instrument that does keep a log, because the machine never goes down – but it cannot reach the real sleep state, and not even all of its own rungs exist here: pm-test-cannot-go-above-platform-on-s2idle.