FW 13 AMD Ryzen AI 300 (HX 370): hibernate enters S4, then machine wakes itself within ~1 minute (wake event 0x0) — BIOS 3.0.16.5

[bezark]

Summary

Framework Laptop 13 (AMD Ryzen AI 300 / AI 9 HX 370): systemctl hibernate completes a full S4 cycle — image written, platform powers off — but the machine wakes itself within ~1 minute, with no wake event recorded (Controller resume with wake event 0x0). The user finds a warm laptop and a dead battery in the bag.

This matches what staff (Matt_Hartley) described investigating in community thread 75092 ("an OS level wake event I am seeing on AI 300 that I did not see on the 7040 series for hibernate (powers off, not suspend)").

Device Information

• System model / SKU: Framework Laptop 13, AMD Ryzen AI 300 Series (AI 9 HX 370)
• BIOS version: 3.0.16.5 (fwupd reports this as the latest available)
• Memory: 32 GiB
• Storage: NVMe PCIe SSD, hibernation image written to a dedicated swap partition (resume= on the kernel command line, not a swapfile)
• WiFi: MediaTek MT7925 (mt7925e driver)
• OS / kernel: NixOS 26.11 unstable, kernel 7.2.0, amdgpu, s2idle default
• Expansion cards: USB-C + HDMI

Evidence 1 — 2026-08-29 19:10:05 EDT, complete 52-second cycle (entry → S4 → self-wake)

19:10:05 systemd-logind: hibernate requested from client PID 70968 ('systemctl')
19:10:06 kernel: PM: hibernation: hibernation entry
19:10:58 kernel: PM: hibernation: Allocated 12635040 kbytes in 16.36 seconds (772.31 MB/s)
19:10:58 kernel: ACPI: PM: Preparing to enter system sleep state S4
19:10:58 kernel: ACPI: PM: Saving platform NVS memory
19:10:58 kernel: PM: hibernation: Normal pages needed: 3464240 + 1024, available pages: 4737061
19:10:58 kernel: ACPI: PM: Restoring platform NVS memory
19:10:58 bluetoothd: Controller resume with wake event 0x0
19:10:58 kernel: ACPI: PM: Waking up from system sleep state S4
19:10:58 kernel: PM: hibernation: hibernation exit
19:10:58 systemd-sleep: System returned from sleep operation 'hibernate.'

No power button press, no lid open, no RTC wakealarm (/sys/class/rtc/rtc0/wakealarm was empty).

Evidence 2 — same day, 14:16:55 EDT (independent cycle, same outcome)

16:24:14 kernel: PM: hibernation: Allocated 12627512 kbytes in 8.58 seconds (1471.73 MB/s)
16:24:14 kernel: ACPI: PM: Preparing to enter system sleep state S4
16:24:14 kernel: ACPI: PM: Saving platform NVS memory
16:24:14 kernel: ACPI: PM: Waking up from system sleep state S4
16:24:14 systemd-sleep: System returned from sleep operation 'hibernate.'
16:24:14 kernel: PM: hibernation: hibernation exit

Evidence 3 — contrast: a genuinely successful overnight cycle, 2026-08-20 19:48 → 2026-08-21 06:29 EDT

Aug 20 19:48:42 kernel: PM: hibernation: hibernation entry
[real poweroff — machine off overnight]
Aug 21 06:29:13 kernel: ACPI: PM: Waking up from system sleep state S4
Aug 21 06:29:13 systemd-sleep: System returned from sleep operation 'hibernate.'

So S4 can complete on this machine — but note this resume was followed 77 seconds later by an unexpected full reset (the boot ends in "crash" in last(1)), consistent with the sync-flood instant-reset variant described in community thread 83040.

Notes

• pm_async is at the kernel default (1). The pm_async=0 workaround from thread 83040 was tested earlier this summer against the resume-reset variant — different failure mode from the one reported here.
• Real-world impact of the instant-wake: the machine fell back into s2idle with the lid closed and drained ~12%/hr until it hard-powered-off (32% → 3% over 2.5 h); the session was lost.

Additional context

• community.frame.work/t/75092 — Hibernation problems (HX 370), staff investigation of an OS-level wake event
• community.frame.work/t/83040 — pm_async=0 A/B workaround (resume-reset variant)
• community.frame.work/t/73314 — sleep-mode support on Ryzen AI 300