- Shipped
- September 29, 2026 at 12:40 AM UTC
- Author
- Kamo
- Commit
- 5bb1174
Measured on a throwaway -ai4 workstation (SP98 Task 16), the wake stalled because Xorg does not survive a resume: with no KMS driver it drives the console through VESA, and on every resume it segfaults re-entering its VT (VBESetVBEMode -> xf86ExecX86int10, "Segmentation fault at address 0x0"). Chrome, left without its display, then spun on both vCPUs while the restored kernel's scheduler clock stood still (a 3 s busy loop read 0 ns of run time): 7 to 16 minutes to AWAKE, or AWAKE in a minute and no usable desktop for 15. Minimizing the windows first changed nothing; closing Chrome first removed the stall but left about 4 minutes before the screen guard signed the AI in. - before_sleep stops LightDM, so no X server goes into the sleep image; the minimize pass is gone. - The desktop starts again (systemctl start --no-block lightdm) on the first resume check more than 30 s after the stop: the wake's token push, a watcher tick or the screen guard. A sleep that never happened gets it back too. - Chrome keeps --force-prefers-reduced-motion: Kamo's sign-in page no longer holds a vCPU on an idle workstation.