The following is AI generated, but I've reviewed it for accuracy:
====
Adding data from a GTX 1060 (Pascal) desktop on the same driver generation -
this bug family appears wider than the X11 black screen reported above, and I
have ~30 instrumented suspend/resume cycles across three configurations.
System: Ubuntu 24.04, GNOME/GDM, kernel 6.8.0-139-generic, proprietary
nvidia-driver-580 580.178.04, GTX 1060 (GP106), monitor on DisplayPort,
nvidia-suspend/nvidia-resume.service active, NVreg_TemporaryFilePath=/var/tmp.
Configurations tested:
a) X11 session, NVreg_PreserveVideoMemoryAllocations=1
b) Wayland session, PreserveVideoMemoryAllocations=0
c) Wayland session, PreserveVideoMemoryAllocations=1 (current)
The pattern is consistent across all three: a suspend resumes, and in some
fraction of cycles the driver comes back with poisoned channel state; the
NEXT suspend then fails. The visible symptom depends on the stack, which may
explain why different reporters describe different failures:
a) X11 + P=1: resume leaves the screen dark (DP link never retrains) or the
session wedged (display renders, clock updates, input dead). Terminal
state after several incidents: the old Xorg holds DRM master and GDM
crash-loops with
"(EE) systemd-logind: failed to take device /dev/dri/card1:
Device or resource busy" -> "Session never registered, failing"
i.e. even the greeter cannot start. Only recovery: power off.
b) Wayland + P=0: resume produces an Xid 13 storm (Graphics Exception,
SAVE_RESTORE_ADDR_OOB) submitted by gnome-shell itself; the screen shows
psychedelic garbage while the compositor stays partly alive. Recoverable
via VT switch + systemctl restart gdm.
c) Wayland + P=1 (current): the suspend itself sleeps fine (multi-hour S3,
VRAM save/restore completes in seconds), and the resume returns with the
display healthy - but an application GPU channel dies silently:
kernel: NVRM: Xid 69, pid=8468, name=Discord, Class Error: channel
0x00000038, Class 0000902d, ErrorCode 00000004
HOURS LATER the next suspend fails at the freezer stage:
kernel: Freezing user space processes failed after 20.001 seconds
(2 tasks refusing to freeze)
kernel: task:gnome-shell state:R running task
kernel: ? os_acquire_spinlock+0x12/0x30 [nvidia]
kernel: ? _nv059470rm+0x6c/0x90 [nvidia] (deep _nv*rm spinloop)
kernel: NVRM: Xid 158, pid=2877, name=gnome-shell, - timeout error
waiting for NV_UFLUSH_FB_FLUSH
-> the suspend ABORTS ("PM: suspend exit" ~20 s after entry): the machine
never sleeps. fbcon takes over the console, the session wedges, and
the left-over suspend transaction then even refuses a clean reboot:
systemd[1]: Failed to enqueue replace-irreversibly job for
ctrl-alt-del.target: Transaction for reboot.target/start is destructive
This happened on two independent evenings (2026-09-21 and 2026-09-22)
with identical signatures. Under P=0 the same sequence instead shows the
Xid 13 storm at resume time; under P=1 the damage is silent at resume and
surfaces on the following suspend. PreserveVideoMemoryAllocations=1
therefore fixes the VRAM restore but not the channel/RM-state problem.
Observation method: a detection-only hook runs from a drop-in on
nvidia-resume.service (transient unit) on every resume. It logs
nvidia-smi -q "Display Active", counts "NVRM: Xid" lines in the kernel
journal since resume, and watches for "Freezing user space processes failed".
Detection is reliable and cheap: on the P=0/Wayland incident NVML reported
"Display Active: Enabled" while the framebuffer showed garbage - the Xid
counter is the signal that catches it.
What recovers / what does not:
- monitor power-cycle (HPD retrain) revives a dark display (variant a)
- VT switch + "systemctl restart gdm" recovers the Wayland session but
NOT the kernel driver state; suspend remains broken until a reboot
- "systemctl reboot --force" from a VT always works; a plain reboot can be
refused while the suspend transaction is wedged
- system sysfs (card*/status, enabled) is NOT a reliable liveness signal
under the proprietary Xorg driver, but is accurate under Wayland/KMS
Dates of instrumented incidents: 2026-09-11, 09-13, 09-14,
09-20 (x3), 09-21 (x2), 09-22.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2139645
Title:
Black screen on resume since NVIDIA 580.126.09 update
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-570/+bug/2139645/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
