Hard system freeze when suspend starts ~15–30 s after s2idle resume — RTX 4090 Laptop (Alienware m18 R2), reproduced on 580.173.02, 610.43.02

nvidia-bug-report.gz (813.0 KB)
Summary

Intermittent hard freeze (no panic, no oops, power-cycle required) on a laptop using s2idle suspend. Across five incidents and four driver builds, the fatal pattern is always the same: a suspend initiated within ~15–30 seconds of a resume. The journal ends abruptly either the instant nvidia-suspend.service starts or right at PM: suspend entry (s2idle) — nothing from the suspend path is ever logged after that point. Normal suspend/resume cycles (minutes/hours apart) almost always succeed; the freeze rate is roughly once per week of normal lid use.

No Xid errors, no MCE, nothing in pstore. The machine is unresponsive (display off/frozen, no input reaction) and must be power-cycled.

System

  • Alienware m18 R2, BIOS 1.22.0 (all firmware current per fwupd)
  • Intel Raptor Lake (i9), NVIDIA GeForce RTX 4090 Laptop GPU (GN21-X11)
  • Ubuntu 24.04, kernel 7.0.0-30-generic (also reproduced on 6.17 / 7.0.0-28)
  • /sys/power/mem_sleep: [s2idle] (platform is s2idle-only, no deep)
  • Currently: driver 580.173.02 proprietary (Ubuntu nvidia-driver-580)
  • nvidia-suspend.service / nvidia-resume.service enabled
  • Module parameters (verified in /proc/driver/nvidia/params):
    • EnableS0ixPowerManagement=1
    • PreserveVideoMemoryAllocations=0
    • DynamicPowerManagement=0

Ruled out

The freeze has survived all of the following, so it does not appear tied to a driver branch, kernel module flavor, or power-management mode:

  • Driver versions: 595.71.05 → 595.84 → 610.43.02 → 580.173.02 (LTS)
  • Kernel module flavor: open (610-open) and proprietary (610, 580)
  • Kernels: 6.17 → 7.0.0-28 → 7.0.0-30
  • NVreg_DynamicPowerManagement: 3 and 0
  • S0ix power management: off and on (EnableS0ixPowerManagement=1)
  • BIOS/firmware: everything current (fwupd)

Note: on 595.x/610.x, fatal resumes logged NVRM: RmHandleDNotifierEvent ... status=0x11 pairs — but the 2026-08-25 fatal boot (610.43.02 proprietary) had zero D-Notifier errors, so that error is not required for the freeze.

Incident 1 — 2026-08-24, driver 610.43.02 (open)

Resume completes, suspend starts 25 s later, journal ends at PM: suspend entry:

Aug 24 14:00:08 systemd[1]: Finished nvidia-resume.service - NVIDIA system resume actions.
Aug 24 14:00:33 systemd-logind[1591]: Suspending...
Aug 24 14:00:33 systemd[1]: Finished nvidia-suspend.service - NVIDIA system suspend actions.
Aug 24 14:00:34 systemd-sleep[15892]: Performing sleep operation 'suspend'...
Aug 24 14:00:34 kernel: PM: suspend entry (s2idle)
<journal ends — machine found hard-frozen, power-cycled>

Incident 2 — 2026-08-25, driver 610.43.02 (proprietary)

A 4-second suspend/resume cycle; ~15 s after the resume the GPU is visibly wedged (X server lock timeout, Mutter DisplayConfig D-Bus timeouts, Chrome cannot create GPU command buffers). A lid-close suspend then hangs the machine the instant nvidia-suspend.service starts:

Aug 25 16:52:55 kernel: PM: suspend entry (s2idle)
Aug 25 16:52:59 kernel: PM: suspend exit                      <- 4-second cycle
Aug 25 16:53:00 systemd[1]: Finished nvidia-resume.service - NVIDIA system resume actions.
Aug 25 16:53:14 gdm-x-session[8514]: nvLock: client timed out, taking the lock
Aug 25 16:53:23 systemd-logind[1230]: Suspending...
Aug 25 16:53:41 gsd-power[8845]: Error setting property 'PowerSaveMode' on interface org.gnome.Mutter.DisplayConfig: Timeout was reached
Aug 25 16:53:45 google-chrome[10487]: ContextResult::kTransientFailure: Failed to send GpuControl.CreateCommandBuffer.
Aug 25 16:53:54 systemd[1]: Starting nvidia-suspend.service - NVIDIA system suspend actions...
<journal ends — hard freeze, power-cycled>

This incident suggests the GPU/driver deadlocks shortly after the rapid resume, and the subsequent nvidia-suspend hangs against the already-dead GPU.

Incident 3 — 2026-08-28, driver 580.173.02 (proprietary, LTS)

Same signature on the LTS branch. A USB mouse wake with the lid closed, logind re-suspends 24 s later:

Aug 28 03:27:27 kernel: PM: suspend exit
Aug 28 03:27:28 systemd[1]: Finished nvidia-resume.service - NVIDIA system resume actions.
Aug 28 03:27:52 systemd-logind[1230]: Suspending...
Aug 28 03:27:52 systemd[1]: Reached target sleep.target - Sleep.
Aug 28 03:27:52 systemd[1]: Starting nvidia-suspend.service - NVIDIA system suspend actions...
<journal ends — hard freeze, power-cycled>

The whole 2-day boot had zero NVRM errors and zero Xid before this. Four earlier suspend/resume cycles in the same boot (spaced hours apart) were clean.

Reproduction

Not 100% deterministic, but every fatal incident matches: resume from s2idle, then initiate another suspend within ~15–30 seconds. Suspends spaced normally (minutes+) survive. Frequency ~1 freeze/week under normal lid use before we identified the pattern.