xpra icon
Bug tracker and wiki

Changes between Initial Version and Version 1 of Ticket #475


Ignore:
Timestamp:
12/17/13 07:28:20 (7 years ago)
Author:
Antoine Martin
Comment:

This is a regression, but not a recent one:

  • 0.9.x does not seem to be affected (could just be harder to hit..)
  • 0.10.x is affected (maybe a little harder to trigger than 0.11.x? races in resize code can be like that..)


(and it could still be related to lib injection from buggy OpenCL / Nvidia bits.)

Let the bisection "fun" begin (using rgb encoding for testing):

So this crash is caused by r3613 which enables XShm (#345) by default. This can be confirmed by running normally crashing versions (including trunk) without problems by using:

XPRA_XSHM=0 xpra start ...

Problem is that XShm provides a major performance boost, so ideally we want to find a fix for the crash without turning it off completely..

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #475

    • Property Owner changed from Antoine Martin to Antoine Martin
    • Property Status changed from new to assigned
    • Property Version changed from to 0.10.x
    • Property Summary changed from server deadlocks when resizing glxgears to server deadlocks when resizing fast updating window (ie: glxgears or mplayer)
  • Ticket #475 – Description

    initial v1  
    1 This is a regression (0.10.x is not affected) which needs to be fixed before release.
    2 
    3 [[BR]]
    4 
    51It doesn't crash it just seems to be stuck:
    62{{{
     
    4642}}}
    4743
    48 So it looks like we're stuck waiting for a reply from the X11 server, and none is coming...
     44And here is another one (with 0.10.x r4000), showing up in a different place:
     45{{{
     46wait_for_reply()
     47XGetGeometry()
     48gdk_window_get_geometry()
     49resize_corral_window()
     50xpra.x11.gtk_x11.error.call_synced()
     51xpra.x11.gtk_x11.window.may_resize_corral_window()
     52}}}
     53With the X11 server just waiting for something:
     54{{{
     55(gdb) bt
     56#0  0x00000034528ed533 in __select_nocancel () at ../sysdeps/unix/syscall-template.S:81
     57#1  0x000000000046692c in WaitForSomething ()
     58#2  0x0000000000436eb1 in Dispatch ()
     59#3  0x00000000004266da in main ()
     60}}}
     61
     62Sometimes it also just kills the X11 server (no xpra backtrace in that case - no X11 backtrace yet..).
     63
     64
     65In the xpra backtrace case, it looks like we're stuck waiting for a reply from the X11 server, and none is coming...
    4966Possibly because we accessed the server from a non-UI thread somewhere. (Ouch!)
    5067Could also be some library problems (#422: the AMD {{{OpenCL}}} icd is known to cause weird problems)