https://bugzilla.novell.com/show_bug.cgi?id=473302 User matz@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=473302#c6 --- Comment #6 from Michael Matz 2009-02-06 09:53:27 MST --- I can reproduce it now. From a remote machine I simply do xset dpms force standby KDE3 will notice this, and starts kdesktop_lock and the kss module (the program that paints something nice), immediately SIGSTOPs that kss module (because due to DPMS it wouldn't be visible anyway), and then the countdown to the bug starts. stracing X then reveals that the select() loop starts with a 870 something seconds timeout, which is reduced with each call to select, by the amount it took for the select to return plus 20 milli-seconds. So, for the first ~870 seconds after the screensaver becomes active all is well, as there's still a time-out so select only returns seldomly. After these ~870 seconds the timeout value is 0, and _doesn't_ get reset to something larger again. From then on X will take 100% CPU. I can reproduce this reliably, but it takes some time to show up (exactly the ~870 seconds). I've tried looking at the patches that went in after beta4. The only thing which barely is related to screen-saving seems to be dpms_screensaver.diff, from 2008-11-28. I think this was shortly after beta4 was released. The patch itself looks innocent (just introduces a call to dixSaveScreens()), but perhaps it triggers a bug somewhere else (in the timeout calculations for select). [probably irrelevant: grepping for "select" in all diffs of xorg-x11-server reveals some occurences in xorg-server-xf4vnc.patch and xorg-server-xf4vnc-busyloop.diff, where at least some in xf4vnc.patch also explicitely use a {0,0} timeout. That at least looks dubious, as that usually is not what one wants. But as this is vnc, it's most probably unrelated.] -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=473302 Stefan Dirsch changed: What |Removed |Added ---------------------------------------------------------------------------- CC|sndirsch@novell.com |lverhaegen@novell.com AssignedTo|bnc-team-xorg-bugs@forge.pr |eich@novell.com |ovo.novell.com | -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=473302 User matz@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=473302#c9 --- Comment #9 from Michael Matz 2009-02-09 03:41:52 MST --- So that means your initial select timeout is 1 second (where it is about 870 for me)? That's fully possible, but still would hint at the same problem. X has a list of timers (with an expire time). The select timeout is calculated in such a way that it times out at or before the next timer has to fire. The list of timers depend on the applications running on the display, so slight changes there are expected. But the general theme is still the same. My assumption is that this list of timers is not cleaned up in the bug case, so that even already expired timers stay in the list. Due to compensation code for negative offsets this would lead exactly to the picture of zero-timeout selects. (the other possibility for that to happen is with the SMART_SCHEDULE code as mentioned above) Btw. Ludwig: You can find out which application is connected to the socket corresponding to FD 40 in Xorg in the following way: # lsof -p $(pidof X) | grep '40u.*unix' X 6191 root 40u unix 0xf7c94280 0t0 20618 /tmp/.X11-unix/X0 # lsof | grep 20617 kopete 6598 matz 4u unix 0xf7c94080 0t0 20617 socket What you need to look out for is a socket filedescriptor with an inode of one less than the inode number of the socket on the X11 side (above 20618 on X side, 20617 on kopete side). This obviously depends on how the kernel invents these inode numbers, but for my kernel (32bit kernel-pae-2.6.27.7-9.1) it uses the above rules. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=473302 User matz@novell.com added comment https://bugzilla.novell.com/show_bug.cgi?id=473302#c11 --- Comment #11 from Michael Matz 2009-02-09 07:43:37 MST --- And you can't reproduce it anymore when kde{d,init}4 isn't running anymore? Then this one triggers the bug in Xorg, maybe a nice hint. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=473302 Stefan Dirsch changed: What |Removed |Added ---------------------------------------------------------------------------- Priority|P5 - None |P3 - Medium CC| |sndirsch@novell.com -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.