Public bug reported:
[Summary] Since upgrading to the HWE 6.17 kernel, KDE X11 desktop daemons hang
inside Mesa's DRI3/Present swap path several times per day. Two manifestations
of what appears to be one defect:
- kwin_x11: main loop blocks indefinitely (0% CPU, D-Bus unresponsive,
compositor stops presenting; Xorg itself stays healthy). 9 occurrences in 4
days, previously zero in months on the 6.8 kernel.
- plasmashell: a QSGRenderThread (created for a notification popup window)
spins at 100% CPU inside libGLX_mesa.so.0 in the glXSwapBuffers path and never
returns; the main thread blocks in QQuickWindow scene-graph sync waiting for it.
Panels/shell freeze. Captured live with gdb (attached).
[Environment] Ubuntu 24.04.4 LTS, X11, KDE Plasma 5.27.12 / kwin 5.27.11 ·
mesa 25.2.8-0ubuntu0.24.04.2 (unchanged since 2026-06-15) · xserver-xorg-core
2:21.1.12-1ubuntu1.6 (modesetting) · kernel 6.17.0-40-generic HWE (regression
window opened with 6.17.0-35 on 2026-07-11; first hang 2026-07-13) · Dell
Precision 5690, Intel Core Ultra (Meteor Lake), Intel Arc Graphics —
compositing on the iGPU via GLX (hybrid NVIDIA unused for desktop).
[Timeline/correlation] Months on 6.8.x: zero hangs. 2026-07-11: HWE rolls to
6.17.0-35; Mesa and Xorg unchanged. 2026-07-13 → 07-17: 10 kwin_x11 hangs + 1
plasmashell hang, all on 6.17 boots. dmesg at hang time is clean — no
i915/GuC/DRM errors — pointing to present/vblank event timing exposing a
userspace race, not a logged GPU fault.
[Evidence — plasmashell, spinning variant] Thread 23 (LWP 2156701)
"QSGRenderThread": frames #0–#2 in libGLX_mesa.so.0, #3
libqxcb-glx-integration.so, #4–#5 libQt5Quick.so.5 (glXSwapBuffers path; 100%
CPU 11+ min). Main thread:
QWaitCondition::wait ← libQt5Quick ← QQuickWindow::event (blocked forever on
the render thread). All other threads parked normally.
[Evidence — kwin_x11, blocking variant] Process alive, main loop blocked:
org.kde.KWin D-Bus times out 25 s+, ~3.5 min total CPU over 9 h (idle-blocked,
not spinning), Xorg/xrandr responsive throughout; recovery = SIGKILL + restart.
Repeatedly triggered around notification-popup churn and screen-lock
transitions — rapid create/swap/destroy of GLX windows.
[Reproducer] A storm of desktop notifications (15 rapid notify-send
popups with 0.5–3 s expiries every ~8 s) wedged plasmashell within ~10
minutes; trigger is short-lived popup windows destroyed with swaps in
flight.
[Fix candidate] Mesa 26.1.5 (2026-07-15): "dri3: Increment draw->send_sbc
after waiting for last presentation" (Michel Dänzer), fixing "OpenGL app
deadlocks in loader_dri3_swap_buffers_msc → xcb_wait_for_special_event (Telegram
Desktop)" and "Qt application TrenchBroom hangs in glXSwapBuffers" — the same
two signatures we observe. Those were reported against Mesa 26.1; the
loader_dri3 sbc logic is long-lived and our 25.2.8 hangs with the same
signature once
6.17 changes present timing, so we ask that this fix be evaluated for
SRU/backport to noble's 25.2.x.
[Related reports reviewed — NOT duplicates] LP#2127687 is a hard full-system
lockup (SSH dies, journal stops) — ours keeps Xorg/SSH/ttys alive. Arch "Meteor
Lake GSC proxy errors" reports log explicit i915/GuC errors — our dmesg is
silent. Mesa 26.1.5 relnotes (docs.mesa3d.org/relnotes/26.1.5.html) carry the
matching upstream issues.
[Workaround — verified] LIBGL_DRI3_DISABLE=1 for both daemons: the
same reproducer then completes with both healthy, compositing still
hardware OpenGL — further isolating the defect to the DRI3 present path.
[Asks] (1) Evaluate/backport the 26.1.5 send_sbc fix to noble. (2) We
can reproduce on demand and capture more traces (dbgsym gdb,
LIBGL_DEBUG, present traces) — the machine reproduces several times
daily without the workaround.
** Affects: mesa (Ubuntu)
Importance: Undecided
Status: New
** Attachment added: "gdb-backtrace-all.txt"
https://bugs.launchpad.net/bugs/2161079/+attachment/5982894/+files/gdb-backtrace-all.txt
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2161079
Title:
kwin_x11 and plasmashell hang forever in DRI3 glXSwapBuffers after HWE
kernel 6.17 update (Meteor Lake, X11) — fix candidate in Mesa 26.1.5
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/2161079/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
