[Bug 41653] New: Xorg crashes on exit [bisected]

[Michal Suchanek]

> Looks like Xorg crashes on exit after 463dd87
>
> To reproduce:
>
> export DISPLAY=:1 ; X :1 & sleep 2 ; dwm &  sleep 1 ; xrandr --output DVI-0
> --off ; xrandr --output HDMI-0 --auto --rotate left ; xrandr --output DVI-0
> --auto --left-of HDMI-0 ; sleep 3 ; kill %1
>
> (gdb) bt full
> #0  DamageUnregister (pDrawable=0x0, pDamage=0x22da320) at
> ../../../miext/damage/damage.c:1923
>        pScreen = <value optimized out>
>        pScrPriv = 0x0
> #1  0x00000000004b04cb in xf86RotateDestroy (crtc=0x2036890) at
> ../../../../hw/xfree86/modes/xf86Rotate.c:308
>        pScrn = <value optimized out>
>        pScreen = 0x2040c80
>        xf86_config = 0x2036700
>        c = <value optimized out>
> #2  0x00000000004b05d5 in xf86RotateCloseScreen (screen=<value optimized out>)
>    at ../../../../hw/xfree86/modes/xf86Rotate.c:344
>        scrn = <value optimized out>
>        xf86_config = 0x2036700
>        c = <value optimized out>
> #3  0x00000000004a2182 in xf86CrtcCloseScreen (index=0, screen=0x2040c80) at
> ../../../../hw/xfree86/modes/xf86Crtc.c:745
>        scrn = <value optimized out>
>        config = 0x2036700
>        o = <value optimized out>
>        c = <value optimized out>
> #4  0x00000000004c07fb in CursorCloseScreen (index=0, pScreen=0x2040c80) at
> ../../xfixes/cursor.c:207
>        ret = <value optimized out>
>        close_proc = <value optimized out>
>        display_proc = <value optimized out>
>        constrain_proc = <value optimized out>
> #5  0x00000000004f498c in AnimCurCloseScreen (index=<value optimized out>,
> pScreen=<value optimized out>)
>    at ../../render/animcur.c:106
>        as = 0x2070d60
>        ret = <value optimized out>
> #6  0x000000000042719f in main (argc=2, argv=<value optimized out>, envp=<value
> optimized out>) at ../../dix/main.c:327
>        i = 0
>        alwaysCheckForInput = {0, 1}
>

I worked around the crash by not calling DamageUnregister if drawable is 0.

        /* Free damage structure */
        if (xf86_config->rotation_damage_registered)
        {
            DrawablePtr screenDrawable = &pScreen->root->drawable;
            if (screenDrawable) DamageUnregister (screenDrawable,
                              xf86_config->rotation_damage);


This causes a double-free of the damage, likely on the next line.
  DamageDestroy (xf86_config->rotation_damage);


I suspect the issue is that  (xf86_config->rotation_damage_registered)
is independent of the damage being actually valid.

In FreeAllResources it is already destroyed by the point X gets to
CursorCloseScreen.


==7817== Invalid read of size 8
==7817==    at 0x4F74F0: DamageDestroy (damage.c:1962)
==7817==    by 0x4B065E: xf86RotateDestroy (xf86Rotate.c:314)
==7817==    by 0x4B0774: xf86RotateCloseScreen (xf86Rotate.c:345)
==7817==    by 0x4A2311: xf86CrtcCloseScreen (xf86Crtc.c:745)
==7817==    by 0x4C093A: CursorCloseScreen (cursor.c:207)
==7817==    by 0x4F483B: AnimCurCloseScreen (animcur.c:106)
==7817==    by 0x4272EE: main (main.c:327)
==7817==  Address 0xb666e28 is 136 bytes inside a block of size 152 free'd
==7817==    at 0x4C268FE: free (in
/usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==7817==    by 0x4F7655: damageDestroyWindow (damage.c:1741)
==7817==    by 0x7CB43D9: XvDestroyWindow (xvmain.c:439)
==7817==    by 0x489F32: xf86XVDestroyWindow (xf86xv.c:1153)
==7817==    by 0x45D5E3: FreeWindowResources (window.c:941)
==7817==    by 0x460347: DeleteWindow (window.c:1019)
==7817==    by 0x456F41: doFreeResource (resource.c:571)
==7817==    by 0x457C79: FreeClientResources (resource.c:853)
==7817==    by 0x457D26: FreeAllResources (resource.c:869)
==7817==    by 0x42726D: main (main.c:308)
==7817==
==7817== Invalid read of size 8
==7817==    at 0x4F7520: DamageDestroy (damage.c:1965)
==7817==    by 0x4B065E: xf86RotateDestroy (xf86Rotate.c:314)
==7817==    by 0x4B0774: xf86RotateCloseScreen (xf86Rotate.c:345)
==7817==    by 0x4A2311: xf86CrtcCloseScreen (xf86Crtc.c:745)
==7817==    by 0x4C093A: CursorCloseScreen (cursor.c:207)
==7817==    by 0x4F483B: AnimCurCloseScreen (animcur.c:106)
==7817==    by 0x4272EE: main (main.c:327)
==7817==  Address 0xb666df0 is 80 bytes inside a block of size 152 free'd
==7817==    at 0x4C268FE: free (in
/usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==7817==    by 0x4F7655: damageDestroyWindow (damage.c:1741)
==7817==    by 0x7CB43D9: XvDestroyWindow (xvmain.c:439)
==7817==    by 0x489F32: xf86XVDestroyWindow (xf86xv.c:1153)
==7817==    by 0x45D5E3: FreeWindowResources (window.c:941)
==7817==    by 0x460347: DeleteWindow (window.c:1019)
==7817==    by 0x456F41: doFreeResource (resource.c:571)
==7817==    by 0x457C79: FreeClientResources (resource.c:853)
==7817==    by 0x457D26: FreeAllResources (resource.c:869)
==7817==    by 0x42726D: main (main.c:308)
==7817==

Doing the check around the whole damage handling workarounds both the
crash and the double free.

I ma not sure this is a very robust solution, though. I wonder if the
damage could be unregistered from the screen when it is destroyed.

Thanks

Michal
-------------- next part --------------
A non-text attachment was scrubbed...
Name: workaround-crash-on-close.patch
Type: text/x-patch
Size: 1069 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg-devel/attachments/20111011/da4c4ad8/attachment.bin>