** Description changed:

  [ Impact ]
  
   * On systems with more than one monitor, the lock screen wallpaper
     disappears (turns black) on one monitor, leaving only the lock screen
     text on a black background. To a user this looks like the session has
     crashed or hung, and it hits the common laptop docking/undocking
     workflow (locking a laptop before unplugging the dock, or plugging in
     a monitor while the screen is locked).
   * There are four reports over five years: first observed in 2021
     (LP: #1922340), then LP: #2013042 (2023) with duplicates LP: #2044190
     and #2128888. Affects noble, resolute and stonking.
   * The cause is in mutter's clutter: whenever painting is redirected
     into a private offscreen framebuffer (effect caches such as the lock
     screen blur via Shell.BlurEffect and ClutterLayerNode,
     ClutterOffscreenEffect, or actors with offscreen-redirect), the
     on-screen redraw clip was still applied to the offscreen capture.
     During a repaint of one monitor, the blur cache for another monitor
     could be captured with an unrelated (empty) clip and end up blank.
     Because the wrong-clip capture depends on paint scheduling, the bug
     reproduces intermittently rather than on every attempt.
   * The fix is 4 lines of code plus an explanatory comment: return NULL
     ("paint everything") from clutter_paint_context_get_redraw_clip()
     when painting into a non-base framebuffer. Upstream as mutter merge
     request !5307, written by this bug's mutter (Ubuntu) assignee; the
     same commit is used in all three Ubuntu uploads. Upstream review
     status: open (see Other Info).
  
  [ Test Plan ]
  
   * The failure is an intermittent race, so verification is
     baseline-anchored: on the same machine and session, first reproduce
     the bug on the unpatched package and record how often it occurs
     (target: reproduces at least 3 times within the 20 cycles below; if
     it never reproduces, the environment is not hitting the race and is
     not a valid verification environment). The clean run on the fixed
     package is only meaningful against that reproduction rate.
   * Setup: two monitors (real hardware, or a VM with two hot-pluggable
     virtual displays such as virtio-gpu with two heads), Wayland session;
     note the GPU and driver. Disable automatic screen blanking before
     locking (blanking forces repaints and muddies the result).
     Use a physical cable or the VM's hotplug mechanism - display
     settings are not reachable while locked.
   * Steps:
     1. Connect both monitors.
     2. Lock the screen (Super+L). Confirm the blurred wallpaper is shown
        on the lock screen.
     3. While the screen stays locked, plug in/unplug the second monitor
        (both directions) 20 times.
     4. After each cycle, check every connected monitor.
     5. Expected with the fix: the wallpaper is never black in any cycle.
        Without the fix, one monitor's wallpaper intermittently turns
        black, leaving only the lock screen text.
   * Also lock/unlock the screen several times with the second monitor
     connected, to check the blur cache stays valid across repaints.
   * Regression checks (the changed path affects every offscreen capture,
     which the shell uses routinely): lock/unlock blur, overview, panel /
     dash / dialog opacity animations (they use offscreen-redirect
     ALWAYS), the magnifier colour effects, the close-dialog
     frozen-window effect, screen recording/screencast, and any shell
     extension that puts effects over the window group. Follow the
     desktop team Mutter test plan:
     
https://canonical-ubuntu-projectubuntu.readthedocs-hosted.com/maintainers/test-plans/desktop-team/Mutter/
   * Verify resolute (mutter 50.1) and stonking (mutter 51.0) separately:
     51.0 already contains other offscreen-path fixes (!5220, !5233) that
     50.1 does not, so the two releases are not interchangeable for
     testing this fix.
   * autopkgtests for mutter run automatically against the upload in
     -proposed. No new autopkgtest was added: the failure needs multiple
     monitors, a lock screen and paint-scheduling control, which the
     autopkgtest environment cannot provide; the existing clutter
     conformance tests shipped in mutter-18-tests (including the
     offscreen-redirect tests) exercise the affected code paths.
  
  [ Where problems could occur ]
  
   * The change only affects painting that is redirected into a non-base
     framebuffer. On-screen painting is untouched: the stage's root paint
     node pushes the same framebuffer object as the paint context's base,
     so normal partial redraws, window-group culling and their redraw
     clips behave exactly as before.
   * Within redirected painting, only three consumers change, and each
     already had a NULL fallback: MetaWindowGroup paints without
     occlusion culling (the same fallback it already uses for clone
     paints), MetaBackgroundContent paints its full area, and MetaStage
     watchers are notified with a full-stage clip (relevant to screencast
     damage tracking).
   * The real risk is paint cost: offscreen captures now repaint their
     whole content instead of an intersected region. The shell captures
     offscreen routinely (panel, dash and dialogs use offscreen-redirect
     ALWAYS and re-capture whenever their content changes; the lock
     screen blur re-captures on lock/unlock and monitor changes), so
     increased CPU/GPU usage is possible in sessions with such effects,
     and a shell extension or accessibility tool that puts an effect over
     the window group would lose occlusion culling on every frame.
   * A latent bug in offscreen cache invalidation could surface as visual
     artifacts (e.g. flickering or stale blur textures) on the lock
     screen or other effect users; the regression checks above cover
     lock/unlock, monitor changes and recording.
  
  [ Other Info ]
  
   * mutter is not covered by the GNOME micro release exception, so this
     is a standard bug-fix SRU with ordinary SRU team review.
   * The same patch is applied to:
+    - tenacious (devel): mutter 51.0-1ubuntu5 (development release fixed
+      first, per SRU policy)
+    - stonking: mutter 51.0-1ubuntu3.2
+    - resolute: mutter 50.1-0ubuntu2.8
+  * The same uploads also fix LP: #2175469 (mouse pointer movement rotated
+    after displays wake on portrait monitors), which carries its own SRU
+    documentation on that bug.
+  * The resolute upload stacks on 50.1-0ubuntu2.6 (LP: #2169221, still
+    unverified in resolute-proposed) and includes it; that bug will be
+    verified against 2.8, and both fixes age and release together
+    (coordinated with the 2.6 uploader, Mitchell Augustin).
+  * noble (24.04 LTS) is also affected but is a prior release relative to
+    resolute; fixing resolute while skipping an older release does not
+    create an upgrade regression. It can be considered separately if
+    there is demand.
+  * Upstream merge request !5307 is open and under review; the Ubuntu
+    uploads use the exact same commit. If upstream lands an amended
+    version, the Ubuntu patch will be updated to match in a follow-up
+    SRU.
+ 
+ ---- Original description ----
+ [ Impact ]
+ 
+  * On systems with more than one monitor, the lock screen wallpaper
+    disappears (turns black) on one monitor, leaving only the lock screen
+    text on a black background. To a user this looks like the session has
+    crashed or hung, and it hits the common laptop docking/undocking
+    workflow (locking a laptop before unplugging the dock, or plugging in
+    a monitor while the screen is locked).
+  * There are four reports over five years: first observed in 2021
+    (LP: #1922340), then LP: #2013042 (2023) with duplicates LP: #2044190
+    and #2128888. Affects noble, resolute and stonking.
+  * The cause is in mutter's clutter: whenever painting is redirected
+    into a private offscreen framebuffer (effect caches such as the lock
+    screen blur via Shell.BlurEffect and ClutterLayerNode,
+    ClutterOffscreenEffect, or actors with offscreen-redirect), the
+    on-screen redraw clip was still applied to the offscreen capture.
+    During a repaint of one monitor, the blur cache for another monitor
+    could be captured with an unrelated (empty) clip and end up blank.
+    Because the wrong-clip capture depends on paint scheduling, the bug
+    reproduces intermittently rather than on every attempt.
+  * The fix is 4 lines of code plus an explanatory comment: return NULL
+    ("paint everything") from clutter_paint_context_get_redraw_clip()
+    when painting into a non-base framebuffer. Upstream as mutter merge
+    request !5307, written by this bug's mutter (Ubuntu) assignee; the
+    same commit is used in all three Ubuntu uploads. Upstream review
+    status: open (see Other Info).
+ 
+ [ Test Plan ]
+ 
+  * The failure is an intermittent race, so verification is
+    baseline-anchored: on the same machine and session, first reproduce
+    the bug on the unpatched package and record how often it occurs
+    (target: reproduces at least 3 times within the 20 cycles below; if
+    it never reproduces, the environment is not hitting the race and is
+    not a valid verification environment). The clean run on the fixed
+    package is only meaningful against that reproduction rate.
+  * Setup: two monitors (real hardware, or a VM with two hot-pluggable
+    virtual displays such as virtio-gpu with two heads), Wayland session;
+    note the GPU and driver. Disable automatic screen blanking before
+    locking (blanking forces repaints and muddies the result).
+    Use a physical cable or the VM's hotplug mechanism - display
+    settings are not reachable while locked.
+  * Steps:
+    1. Connect both monitors.
+    2. Lock the screen (Super+L). Confirm the blurred wallpaper is shown
+       on the lock screen.
+    3. While the screen stays locked, plug in/unplug the second monitor
+       (both directions) 20 times.
+    4. After each cycle, check every connected monitor.
+    5. Expected with the fix: the wallpaper is never black in any cycle.
+       Without the fix, one monitor's wallpaper intermittently turns
+       black, leaving only the lock screen text.
+  * Also lock/unlock the screen several times with the second monitor
+    connected, to check the blur cache stays valid across repaints.
+  * Regression checks (the changed path affects every offscreen capture,
+    which the shell uses routinely): lock/unlock blur, overview, panel /
+    dash / dialog opacity animations (they use offscreen-redirect
+    ALWAYS), the magnifier colour effects, the close-dialog
+    frozen-window effect, screen recording/screencast, and any shell
+    extension that puts effects over the window group. Follow the
+    desktop team Mutter test plan:
+    
https://canonical-ubuntu-projectubuntu.readthedocs-hosted.com/maintainers/test-plans/desktop-team/Mutter/
+  * Verify resolute (mutter 50.1) and stonking (mutter 51.0) separately:
+    51.0 already contains other offscreen-path fixes (!5220, !5233) that
+    50.1 does not, so the two releases are not interchangeable for
+    testing this fix.
+  * autopkgtests for mutter run automatically against the upload in
+    -proposed. No new autopkgtest was added: the failure needs multiple
+    monitors, a lock screen and paint-scheduling control, which the
+    autopkgtest environment cannot provide; the existing clutter
+    conformance tests shipped in mutter-18-tests (including the
+    offscreen-redirect tests) exercise the affected code paths.
+ 
+ [ Where problems could occur ]
+ 
+  * The change only affects painting that is redirected into a non-base
+    framebuffer. On-screen painting is untouched: the stage's root paint
+    node pushes the same framebuffer object as the paint context's base,
+    so normal partial redraws, window-group culling and their redraw
+    clips behave exactly as before.
+  * Within redirected painting, only three consumers change, and each
+    already had a NULL fallback: MetaWindowGroup paints without
+    occlusion culling (the same fallback it already uses for clone
+    paints), MetaBackgroundContent paints its full area, and MetaStage
+    watchers are notified with a full-stage clip (relevant to screencast
+    damage tracking).
+  * The real risk is paint cost: offscreen captures now repaint their
+    whole content instead of an intersected region. The shell captures
+    offscreen routinely (panel, dash and dialogs use offscreen-redirect
+    ALWAYS and re-capture whenever their content changes; the lock
+    screen blur re-captures on lock/unlock and monitor changes), so
+    increased CPU/GPU usage is possible in sessions with such effects,
+    and a shell extension or accessibility tool that puts an effect over
+    the window group would lose occlusion culling on every frame.
+  * A latent bug in offscreen cache invalidation could surface as visual
+    artifacts (e.g. flickering or stale blur textures) on the lock
+    screen or other effect users; the regression checks above cover
+    lock/unlock, monitor changes and recording.
+ 
+ [ Other Info ]
+ 
+  * mutter is not covered by the GNOME micro release exception, so this
+    is a standard bug-fix SRU with ordinary SRU team review.
+  * The same patch is applied to:
     - tenacious (devel): mutter 51.0-1ubuntu4 (development release fixed
       first, per SRU policy)
     - stonking: mutter 51.0-1ubuntu3.1
     - resolute: mutter 50.1-0ubuntu2.7
   * The resolute upload stacks on 50.1-0ubuntu2.6 (LP: #2169221, still
     unverified in resolute-proposed) and includes it; that bug will be
     verified against 2.7, and both fixes age and release together
     (coordinated with the 2.6 uploader, Mitchell Augustin).
   * noble (24.04 LTS) is also affected but is a prior release relative to
     resolute; fixing resolute while skipping an older release does not
     create an upgrade regression. It can be considered separately if
     there is demand.
   * Upstream merge request !5307 is open and under review; the Ubuntu
     uploads use the exact same commit. If upstream lands an amended
     version, the Ubuntu patch will be updated to match in a follow-up
     SRU.
  
  ---- Original description ----
  Lock screen's blurry wallpaper vanishes on the primary monitor if a second 
monitor is plugged in/out. I'm left with just the lock screen text on a black 
background.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2013042

Title:
  Lock screen wallpaper vanishes (goes black) if a second monitor is
  plugged in/out

To manage notifications about this bug go to:
https://bugs.launchpad.net/gnome-shell/+bug/2013042/+subscriptions


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

Reply via email to