#5114 HIGH Update.: 339 display problem

[Zarro Boogs per Child]

#5114: 339 display problem
------------------------------+---------------------------------------------
  Reporter:  walter           |       Owner:  bernie         
      Type:  defect           |      Status:  new            
  Priority:  high             |   Milestone:  Update.1       
 Component:  x window system  |     Version:                 
Resolution:                   |    Keywords:  dcon, Update.1?
  Verified:  0                |  
------------------------------+---------------------------------------------
Changes (by dilinger):

 * cc: JordanCrouse (added)
  * owner:  ApprovalForUpdate => bernie


Comment:

 We managed to reproduce this on a machine that Richard was using.  First
 off, we're calling this the double vision problem; the display looks as
 though you're seeing double.  Assuming that's what we're talking about:

 The dcon is *not* frozen when it happens.  If you trigger it on a webpage
 w/ javascript, you can see the page actively updating while displaying the
 double vision screen (eventually, the screen completely decays).

 'xset dpms force off' will trigger the bug, and switching DPMS back will
 fix the bug.  That's why we see it when we're idle after a little while,
 and hitting a key makes it go away.

 'echo 1 > /sys/devices/platform/dcon/sleep' works properly even when we've
 hit this bug; the DCON turns off, and even after we turn the DCON back on,
 we can still trigger the bug by toggling DPMS.

 We log EC commands that are run, and when we 'xset dpms force off', the EC
 cmd that turns off power to the DCON is never run.

 Toggling DPMS *should* switch the DCON off and back on.. However, when we
 trigger this bug, the DCON remains awake.  Thus, my current theory is that
 this is a bug in the xorg amd driver; we never actually make it to putting
 the DCON to sleep.  We hit the DPMS path, we attempt to mess w/ LX GPU or
 other video stuff, that presumably fails (leaving the GPU in some state w/
 clocks disabled), and we never get as far as echo'ing 1 to
 /sys/devices/platform/dcon/sleep.

 The other possibility is that we're disabling something in the GPU from
 the xorg driver, and when  echo'ing 1 to /sys/devices/platform/dcon/sleep
 is ignored because we refuse to put the DCON to sleep on machines that are
 not C2 or greater; however, Richard reproduced this problem on a C2.

-- 
Ticket URL: <http://dev.laptop.org/ticket/5114#comment:8>
One Laptop Per Child <http://dev.laptop.org>
OLPC bug tracking system