I can also confirm this issue. Running top after the slowdown happens, I see the resource usage of GNOME Shell is increased (~190 MB resident memory, > 20% CPU usage on my computer). There is a noticeable latency while performing actions like moving windows and switching windows and workspaces.

I also used the perf tool to generate these reports that might help:

- Before slowdown (i.e. fresh restart of GNOME Shell):

# Events: 2K cycles
#
# Overhead      Command                       Shared Object                            Symbol
# ........  ...........  ..................................  ................................
#
    15.01%  threaded-ml  [ppdev]                             [k] 0xc102ce0e
    12.03%  threaded-ml  libgobject-2.0.so.0.2990.0          [.] 0x337f2 
     8.96%  threaded-ml  libcogl.so.2.2.0                    [.] 0x278a9 
     8.77%  threaded-ml  libc-2.13.so                        [.] 0x6d8b1 
     8.75%  threaded-ml  libglib-2.0.so.0.2990.0             [.] 0x6d04b 
     8.14%  threaded-ml  libclutter-glx-1.0.so.0.714.0       [.] 0x3f220 
     6.68%  threaded-ml  libmozjs185.so.1.0.0                [.] 0xa9891 
     4.65%  threaded-ml  libdricore.so                       [.] 0xe6c32 
     2.90%  threaded-ml  libgnome-shell.so                   [.] 0x3b46f 
     2.66%  threaded-ml  libgirepository-1.0.so.1.0.0        [.] 0xaa46  
     2.15%  threaded-ml  libcairo.so.2.11000.2               [.] 0x436e1 
     2.08%  threaded-ml  libpthread-2.13.so                  [.] pthread_mutex_lock
     1.87%  threaded-ml  libgjs.so.0.0.0                     [.] 0x1d901 
     1.81%  threaded-ml  libpthread-2.13.so                  [.] __pthread_mutex_unlock_usercnt
     1.79%  threaded-ml  libmutter.so.0.0.0                  [.] 0x2365b 
     1.38%  threaded-ml  i915_dri.so                         [.] 0x427d7 
     1.33%  threaded-ml  libgdk-3.so.0.118.0                 [.] 0x494a3 
     0.96%  threaded-ml  ld-2.13.so                          [.] 0xe502  
     0.63%  threaded-ml  libpixman-1.so.0.20.2               [.] 0x3bcc1 
     0.60%  threaded-ml  libdrm_intel.so.1.0.0               [.] 0x69e7  
     0.56%  threaded-ml  libX11.so.6.3.0                     [.] 0x37ce3 
     0.54%  threaded-ml  libpango-1.0.so.0.2903.0            [.] 0x10f81 


- After slowdown (usually after resuming from suspend, or unlocking screen after the system's been idle):

# Events: 5K cycles
#
# Overhead      Command                                     Shared Object                           Symbol
# ........  ...........  ................................................  ...............................
#
    67.43%  threaded-ml  libgdk-3.so.0.118.0                               [.] 0x3d08d 
     4.80%  threaded-ml  [ppdev]                                           [k] 0xc102ce0e
     3.46%  threaded-ml  libmozjs185.so.1.0.0                              [.] 0x11fa32
     3.41%  threaded-ml  libgobject-2.0.so.0.2990.0                        [.] 0x336e1 
     2.94%  threaded-ml  libglib-2.0.so.0.2990.0                           [.] 0x132a0 
     2.85%  threaded-ml  libcogl.so.2.2.0                                  [.] 0x17361 
     2.83%  threaded-ml  libclutter-glx-1.0.so.0.714.0                     [.] 0xbb4a1 
     2.42%  threaded-ml  libc-2.13.so                                      [.] 0x7042b 
     1.91%  threaded-ml  libdricore.so                                     [.] 0x71b01 
     1.74%  threaded-ml  libgnome-shell.so                                 [.] 0x5cda1 
     0.79%  threaded-ml  i915_dri.so                                       [.] 0x2cede 
     0.72%  threaded-ml  libgirepository-1.0.so.1.0.0                      [.] 0x8c50  
     0.54%  threaded-ml  libpthread-2.13.so                                [.] pthread_mutex_lock
     0.51%  threaded-ml  libgjs.so.0.0.0                                   [.] 0x1ae21

I use the nvidia driver.

Also occurs with extremetuxracer (3D game).

<magcius> so yes, it's not suspend/resume, it's unredirected windows

Created attachment 196326 [details] [review]
MetaWindowActor: Make sure to always pop an error trap

Here's the mismatched error trap...

Created attachment 196327 [details] [review]
MetaWindowActor: Don't bother creating pixmaps for unredirected windows

... but why are we creating pixmaps for unredirected windows? Seems silly...

Created attachment 196329 [details] [review]
MetaWindowActor: Delete useless error trap

Actually, why not this? It doesn't seem like there's any X usage outside of that first trap (Clutter has its own traps, right?)

Created attachment 196407 [details] [review]
meta-window-actor: Skip necessary steps in pre_paint

We don't have to call check_needs_pixmap, check_needs_reshape
and check_needs_shadow on unredirected windows, so skip that.

Review of attachment 196407 [details] [review]:

... Should have read the other patches before submitting anything.

Review of attachment 196407 [details] [review]:

Sure.

Review of attachment 196326 [details] [review]:

I like this approach better - knowing what exact error traps we need is *really* black art - depends on the exact details of GLX, refcounting inside cogl, etc. If we have something that works, leave it that way - the error traps aren't expensive if we nest them properly.

Review of attachment 196329 [details] [review]:

As noted, don't feel this is safe

Review of attachment 196407 [details] [review]:

Looks good to me too

Modified "Skip necessary steps" to "Skip unnecessary steps", otherwise pushes as is

Pushed the correct commits this time.

I tested this patch as a fix for https://bugzilla.gnome.org/show_bug.cgi?id=657869 . it semi-fixes that bug for me - it causes the screensaver/lock screen window to display correctly (instead of showing me my stale, pre-idle desktop), but doesn't cause the unlock dialog to appear. However, it *does* seem to cause Shell to crash, I think whenever I drag a window across the border between my two displays.

filed https://bugzilla.gnome.org/show_bug.cgi?id=658997 for the above crasher.

*** Bug 659297 has been marked as a duplicate of this bug. ***