Same symptom on Ubuntu 24.04 with an HDMI monitor, and an HDMI connector
that flaps while DPMS is off. Adding data because several controls here
narrow it down. All times are local (America/New_York).
SYSTEM
- Ubuntu 24.04.5, kernel 7.0.0-43-generic, Wayland
- gnome-shell 46.0-0ubuntu6~24.04.17
- libmutter-14-0 46.2-1ubuntu0.24.04.18
- Mesa 25.2.8-0ubuntu0.24.04.4
- Lenovo laptop, Intel i5-1145G7 (i915), 31 GiB RAM
- eDP-1 plus an external ASUS PB278 on HDMI-A-1, extended display, lid open.
Same result through the laptop's HDMI port and through the HDMI port of a
USB dock. Monitor mode: [1920x1200 @ 60Hz]
- idle-delay is 20 s. /etc/environment also sets
MUTTER_DEBUG_FORCE_KMS_MODE=simple (added during earlier troubleshooting;
the leak predates it, see "Earlier occurrence" below).
WHAT HAPPENS
When the screen blanks, gnome-shell's GPU-backed memory grows linearly at
about 6.5 MB/s (~23 GB/h) until the display wakes. It is freed within ~3 s of
waking. gnome-shell is the only client that grows (per-process fdinfo,
drm-resident-system0), and its growth matches system Shmem to within a few MB.
TIMELINE (one run, 1 s sampling, 2026-11-13)
- 14:33:41 ScreenSaver active, org.gnome.Mutter.DisplayConfig
PowerSaveMode set to 3 (off)
- 14:33:51 first kernel HOTPLUG=1 pair on connector 596 (HDMI-A-1),
followed by MonitorsChanged signals
- then a pair of HOTPLUG=1 uevents about every 11.1 s, each followed by 2x
MonitorsChanged. 24 pairs (udev SEQNUM 9149-9196) until the wake.
- HDMI-A-1 reports "disconnected" for about 1 s per cycle while "enabled"
stays "enabled" and dpms stays Off
- gnome-shell DRM-resident memory rises by about 62 MiB per cycle, in step
with Shmem, and does not change between cycles
- 14:38:10 WakeUpScreen, PowerSaveMode set to 0; Shmem falls from 2.72 GB
to 0.97 GB by 14:38:13
CONTROLS
- All GNOME extensions disabled: same leak, same rate
- Auto-lock off (screen unlocked on wake): same leak
- Laptop panel alone (monitor unplugged): no leak
- idle-delay 0 (screen never blanks): no leak
- DDC/CI disabled in the monitor's OSD: no change
- Powering the monitor off with its power button during a blank: the
hotplug events and the growth stop (as in comment #7)
EARLIER OCCURRENCE
On 2026-08-29 (kernel 7.0.0-30) the same monitor was connected. Shmem
reached ~24 GB and gnome-shell was OOM-killed. The OOM call trace showed
gnome-shell allocating through i915_gem_set_domain_ioctl ->
shmem_get_pages. I have no per-process data from that day, so I can only
say the rate and trigger match.
NOTES
- Every other report here is DisplayPort. This is HDMI, so flapping is
not specific to DP.
- This does not look like the "animating window while locked" case in
bug 2143073: it needs no background app, and the laptop panel alone does
not leak.
- Attached: fine.log (1 s Shmem + gnome-shell DRM + connector status),
udev.log (drm uevents), mutter.log (DisplayConfig signals).
Warning: Claude helped me do this debugging and form this post response!
If I am asked to supply more info, I'll try to best to help even though
I'm not familiar with some of these subsystems or operations.
--
You received this bug notification because you are a member of Ubuntu
Desktop Bugs, which is subscribed to mutter in Ubuntu.
https://bugs.launchpad.net/bugs/2170511
Title:
Shmem leak (~25 GB/h) leading to OOM when a DisplayPort monitor
repeatedly hotplugs while DPMS is off (regression in 50.1-0ubuntu2.4)
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/mutter/+bug/2170511/+subscriptions
--
desktop-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/desktop-bugs