Please consider nominating this for Noble (24.04 LTS) as well. There is currently no Noble task, but the same leak affects mutter 46.2 on noble, and the fix appears to backport cleanly.
Affected system: - ThinkPad T16 Gen 4, Intel Arrow Lake-U (8086:7D41), xe driver - Ubuntu 24.04, kernel 6.17.0-1032-oem - mutter 46.2-1ubuntu0.24.04.16, GNOME 46 Wayland - Clamshell mode (lid closed), single external HDMI monitor at 3840x2160@60 Symptoms, which match the !5285 description: - While the external display is in DPMS-off, Mutter emits MonitorsChanged in pairs about every 8.5 seconds, for as long as the display stays off. - Each cycle leaves roughly two framebuffer-sized (32 MB) exported DMA-BUFs in gnome-shell. That is about 0.5 GB/min, measured from /proc/<gnome-shell-pid>/fdinfo. - The memory is not attributed to any process's RSS, so the OOM killer can't reclaim it. Left alone, the session dies in a global OOM about 45-50 minutes after the display blanks. This has happened more than ten times on this machine. - A genuine display wake (PowerSaveMode back to 0) releases the accumulated buffers without a gnome-shell restart. In one test, fds dropped from 180 to 32 shortly after a wake. That fits the explanation in !5285: detached onscreens are only cleaned up after a modeset, and no modeset happens during power saving. Why a noble backport looks low-risk: - Comment #25 on LP #2143073 reports that af926cff applied to the noble 46.2-1ubuntu0.24.04.14 source with zero fuzz (line offsets only). With the rebuilt package, GPU memory stayed flat for a 30-minute locked test on amdgpu, where it had been growing at ~15 GiB/hour before. - So far the leak has been reproduced on amdgpu, i915, xe and NVIDIA hybrid (upstream #4101 and this bug). It is not specific to one driver. My current workaround is a user script that forces org.gnome.Mutter.DisplayConfig PowerSaveMode back to 0 whenever it goes non-zero. That keeps gnome-shell flat, but it means the monitor never sleeps, so it is not a real fix. I'm happy to test a package from noble-proposed. The leak reproduces within minutes of a screen blank here, and I have baseline numbers on the current .16 package, so verification would be quick. -- 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/2167374 Title: gnome-shell leaks scanout dma-bufs while locked (fixed in mutter 50.5) To manage notifications about this bug go to: https://bugs.launchpad.net/mutter/+bug/2167374/+subscriptions -- desktop-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/desktop-bugs
