Bug 10351

Created attachment 9233 [details]
Xorg logfile

Any news on this? I'll be testing this on RC4 of the driver, but so far it's still affecting me.

I tried disabling XVideo in the config file, and the server continued to work fine (just with no XVideo).  This may require you to run X under the debugger from another machine to pinpoint where the failure is happening.

Okay, I've checked it on my end; I logged in as root, set ulimit to allow core files, and made it dump core.

Running gdb on the core gives me the hint that it's complaining at LinearAlloc. I had forgotten that I put that in (sorry), and apparently, with the LinearAlloc option removed (and XVideo set to on), the X server runs.

HOWEVER, with LinearAlloc set and XVideo set off, the server does crash.

Created attachment 9592 [details]
Xorg core file

Xorg core file

Here's the back trace:

(gdb) bt
#0  0xffffe410 in __kernel_vsyscall ()
#1  0xb7d6ddf0 in raise () from /lib/tls/i686/cmov/libc.so.6
#2  0xb7d6f641 in abort () from /lib/tls/i686/cmov/libc.so.6
#3  0x080a6525 in ddxGiveUp () at xf86Init.c:1129
#4  0x081b9d08 in AbortServer () at log.c:409
#5  0x081ba256 in FatalError (f=0x81c47cc "Caught signal %d.  Server aborting\n")
    at log.c:555
#6  0x080c506d in xf86SigHandler (signo=11) at xf86Events.c:771
#7  <signal handler called>
#8  0xb7bb9048 in I830ScreenInit (scrnIndex=0, pScreen=0x8214b18, argc=2, 
    argv=0xbfd41034) at i830_driver.c:2548
#9  0x0807264e in AddScreen (pfnInit=0xb7bb7dd0 <I830ScreenInit>, argc=2, 
    argv=0xbfd41034) at main.c:784
#10 0x080a6c1e in InitOutput (pScreenInfo=0x81f3d60, argc=2, argv=0xbfd41034)
    at xf86Init.c:909
#11 0x08072df1 in main (argc=2, argv=0xbfd41034, envp=Cannot access memory at address 0x8
) at main.c:370

More info: continuing with the core file, gdb tells me the value of pI830->xaa_linear is null.

Apparently, in i830_memory.c:654, there's a check for XV; if the LinearAlloc option is set *and* XV is disabled, pI830->xaa_linear isn't malloc'd.

Created attachment 9594 [details]
Also check that XV is enabled before calling xf86InitFBManagerLinear

This patch adds a check to ->XvEnabled; having LinearAlloc in the driver options when XV is disabled is *wrong*, I believe, and adding this check prevents the whole thing coming down in that case.

Fixed in the 2.0.0 release.