[Mesa-dev] segfault in dri2 state tracker with TFP

[Jakob Bornecrantz]

On Sun, May 30, 2010 at 1:38 PM, Dave Airlie <airlied at gmail.com> wrote:
> On Sun, May 30, 2010 at 5:52 PM, Dave Airlie <airlied at gmail.com> wrote:
>> Running texture_from_pixmap in xdemos against r300g
>>
>> Program received signal SIGSEGV, Segmentation fault.
>> dri2_invalidate_drawable (dPriv=0x8066990) at dri2.c:57
>> 57         struct dri_context *ctx = dri_context(dPriv->driContextPriv);
>> Missing separate debuginfos, use: debuginfo-install
>> expat-2.0.1-8.fc12.i686 glew-1.5.4-1.fc14.i686 glibc-2.11.1-6.i686
>> libX11-1.3-1.fc12.i686 libXdamage-1.1.2-1.fc12.i686
>> libXext-1.1-2.fc12.i686 libXfixes-4.0.4-1.fc12.i686
>> libXxf86vm-1.1.0-1.fc12.i686 libdrm-2.4.21-0.1.fc14.i686
>> libgcc-4.4.3-4.fc12.i686 libselinux-2.0.90-5.fc12.i686
>> libstdc++-4.4.3-4.fc12.i686 libxcb-1.5-1.fc12.i686
>> mesa-libGLU-7.7-4.fc12.i686
>> (gdb) bt
>> #0  dri2_invalidate_drawable (dPriv=0x8066990) at dri2.c:57
>> #1  0xb7f9ee73 in dri2_bind_tex_image (dpy=0x804b008, drawable=75497476,
>>    buffer=8414, attrib_list=0x0) at dri2_glx.c:581
>> #2  0xb7f7dab6 in __glXBindTexImageEXT (dpy=0x804b008, drawable=75497476,
>>    buffer=8414, attrib_list=0x0) at glxcmds.c:3098
>> #3  0x0804978c in BindPixmapTexture (argc=1, argv=0xbffff304)
>>    at texture_from_pixmap.c:257
>> #4  main (argc=1, argv=0xbffff304) at texture_from_pixmap.c:391
>>
>> dPriv->driContextPriv is NULL and dri_context macro derferences it.
>>
>> attached patch "fixes" it but not sure if we want drI_context macro to
>> catch this or whether something else is doing something wrong and
>> causing this.
>
> I've no idea what the code in dri2_invalidate_drawable is for, the
> intel + nouveau legacy drivers don't seem to need this extra codepath
> and just call the generic invalidate.
>
> This patch + a patch to dri_util.c to always initialise
> pdp->driContextPriv to NULL in driCreateNewDrawable works for me, but
> it would be nice to know why the dri2 st cares about context in this
> place.

Its probably better to just check for null in the dri_context macro.
Thanks for looking into and fixing this.

To answer your question: st/dri pushes the invalidate notify into the
the API state tracker (st/mesa) so that it set a invalidate flag, this
is instead of it calling back into st/dri all the time. st/mesa
implements the flag itself and private from st/dri and due to various
reasons it needs to keep the st_framebuffer_iface <-> gl_framebuffer
binding on the context.

Cheers Jakob.