Log at https://pantheon.corp.google.com/storage/browser/chromiumos-test-logs/bugfiles/cr/655770/

Consistent issue

Is it related with VT2  issue 640082  ? 

No, switching to VT2 (several times) works. This is Gfx issue with coming back from VT2.

Sometimes this same issue happens on static screen graphics, even on login screen.

Since this doesn't affect end users, this is not a blocker... We have way more critical things to fix.

Most likely Chrome flips to a buffer with tiling on before doing a modeset.
We were just discussing something similar affecting WebGL.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/547665586e0d17adc9d8ecf03b41c9448fe7719c

commit 547665586e0d17adc9d8ecf03b41c9448fe7719c
Author: dnicoara <dnicoara@chromium.org>
Date: Tue Nov 15 14:51:33 2016

Fix potential GPU crash caused by VT switching

When display control is released Frecon may change the configuration at
will. Since Chrome never stops page flipping, there is a race condition
when Chrome regains control where the Chrome hasn't polled for the
configuration but is still page flipping which leads to a failed page
flip due to invalid display configuration (which causes the GPU process
to be killed).

Turning the displays off unparents the windows from the displays
allowing Chrome to continue page flipping while the display
configuration is renewed.

BUG= 655770 
TEST=Updated display_unittests

Review-Url: https://codereview.chromium.org/2498103002
Cr-Commit-Position: refs/heads/master@{#432177}

[modify] https://crrev.com/547665586e0d17adc9d8ecf03b41c9448fe7719c/ui/display/chromeos/display_configurator.cc
[modify] https://crrev.com/547665586e0d17adc9d8ecf03b41c9448fe7719c/ui/display/chromeos/display_configurator.h
[modify] https://crrev.com/547665586e0d17adc9d8ecf03b41c9448fe7719c/ui/display/chromeos/display_configurator_unittest.cc

verified in 8872.54.0, 55.0.2883.54

Moving old issues out of Internal>Graphics to delete this obsolete component ( crbug.com/685425  for details)

FWIW, this fix probably broke crouton (https://github.com/dnschneid/crouton/issues/2923#issuecomment-279973524). We managed to work around it by calling xrandr --auto after crouton's X server has started.

Still unclear about the root cause.