I think your suspicion about the 46.2-1ubuntu0.24.04.13 update (the LP
#1971434 / monitor-manager backports) is well founded. Cross-refs: LP
#2143073 (same symptoms, and a version bisect there found .12 clean over
20 days while .13/.15 leak), my i915 report LP #2148793, and LP #2156739
where most of the evidence below is accumulating.

Setup: ThinkPad T16 Gen 4, Arrow Lake-U, Ubuntu 24.04, kernel
6.17.0-1028-oem, mutter 46.2-1ubuntu0.24.04.15, GNOME 46 Wayland,
clamshell with one external HDMI 4K@60 display. Your "OOM while locked,
never during use" pattern matches mine, and I can fill in what happens
during the locked period:

1. Locking blanks/DPMS-offs the display ~15s later regardless of idle
settings (gsd-power hardcodes this after screensaver activation) — which
is why the problem tracks LOCKED time specifically.

2. While the display is in DPMS-off, Mutter (gnome-shell) emits
MonitorsChanged in pairs every ~8.5 seconds, continuously, captured with
dbus-monitor. gsd-xsettings, gsd-power and xdg-desktop-portal re-query
the display config every cycle.

3. Each cycle strands ~2 framebuffer-sized GPU buffers (32 MB each at
4K) inside gnome-shell. On my machine (xe driver) they are visible as
exported DMA-BUF fds in /proc/<gnome-shell-pid>/fdinfo (exp_name/size
fields): ~12 fds/min, ~0.5 GB/min. In one 45-minute locked period the
count grew 12 -> 396 (~12.3 GB) and ended in a global OOM that killed
the session. On i915 the same class of leak surfaces as
Shmem/ShmemHugePages instead (LP #2148793, LP #2143073) — GPU buffer
objects either way, which is why no process RSS ever accounts for the
lost memory.

4. The buffers are never freed on wake; only a gnome-shell restart
releases them.

5. A live event source during power-save appears necessary to sustain
the ~8.5 s loop: e.g. audio streaming over HDMI to the monitor's
speakers; in LP #2143073 a monitor with flaky input-autodetect
(disconnect/reconnect cycles) plays the same role. If your monitor drops
off the bus when it enters power-save (yours reportedly started truly
powering off after the .13 update), that itself may generate the
repeated connector events.

-- 
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/2146851

Title:
  OOM during screen lock after mutter DPMS change

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/mutter/+bug/2146851/+subscriptions


-- 
desktop-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/desktop-bugs

Reply via email to