I have the following card (SandyBridge):
00:02.0 VGA compatible controller [0300]: Intel Corporation Device [8086:0116] (rev 09)

With X 1.9.3 and Mesa 7.10 I get a segfault if I play extremetuxracer. You have to play for a few minutes in fullscreen to get the segfault (I'm not sure non-fullscreen won't segfault).

x11-driver-video-intel-2.14.0-1mdv2010.2
x11-server-xorg-1.9.3-3mdv2010.2
libdrm2-2.4.23-1mdv2010.2

Here is the backtrace:


Program received signal SIGSEGV, Segmentation fault.
prepare_wm_surfaces (brw=0x8e78270) at brw_wm_surface_state.c:602
602              brw_add_validated_bo(brw, region->buffer);
(gdb) bt
#0  prepare_wm_surfaces (brw=0x8e78270) at brw_wm_surface_state.c:602
#1  0xb6f01e38 in brw_validate_state (brw=0x8e78270) at brw_state_upload.c:397
#2  0xb6ef0f8f in brw_try_draw_prims (ctx=0x8e78270, arrays=0x9202438, 
    prim=0x9200e0c, nr_prims=1, ib=0x0, index_bounds_valid=1 '\001', 
    min_index=0, max_index=3) at brw_draw.c:362
#3  brw_draw_prims (ctx=0x8e78270, arrays=0x9202438, prim=0x9200e0c, 
    nr_prims=1, ib=0x0, index_bounds_valid=1 '\001', min_index=0, max_index=3)
    at brw_draw.c:447
#4  0xb6fe6ae9 in vbo_exec_vtx_flush (exec=0x9200c98, unmap=1 '\001')
    at vbo/vbo_exec_draw.c:381
#5  0xb6fddbc9 in vbo_exec_FlushVertices_internal (ctx=0x8ea4788, 
    unmap=0 '\000') at vbo/vbo_exec_api.c:911
#6  0xb6fddc68 in vbo_exec_FlushVertices (ctx=0x8ea4788, flags=1)
    at vbo/vbo_exec_api.c:945
#7  0xb70cb7a1 in _mesa_PopAttrib () at main/attrib.c:858
#8  0xb733f0de in __glXDisp_PopAttrib (pc=0x9bcccbc "\004")
    at indirect_dispatch.c:1443
#9  0xb7367d29 in __glXDisp_Render (cl=0x8f22f18, pc=0x9bcccb8 "\004")
    at glxcmds.c:1847
#10 0xb736c870 in __glXDispatch (client=0x8f22e40) at glxext.c:600
#11 0x08070fff in Dispatch () at dispatch.c:432
#12 0x080625ba in main (argc=8, argv=0xbfa489f4, envp=0xbfa48a18) at main.c:291
(gdb) display region
1: region = (struct intel_region *) 0x0



And here is a piece of the code that crashes:

   if (ctx->DrawBuffer->_NumColorDrawBuffers >= 1) {
      for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
         struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i];
         struct intel_renderbuffer *irb = intel_renderbuffer(rb);
         struct intel_region *region = irb ? irb->region : NULL;

         brw_add_validated_bo(brw, region->buffer);
         nr_surfaces = SURF_INDEX_DRAW(i) + 1;
      }
   }


In our case, "irb" is null, so we assign NULL to "region". On the next line, we try to access region->buffer, which doesn't make sense. We really shouldn't assign NULL to region and then access region->buffer...


So I tried a very simple patch:

diff -Nrup Mesa-7.10//src/mesa/drivers/dri/i965/brw_wm_surface_state.c patched/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
--- Mesa-7.10//src/mesa/drivers/dri/i965/brw_wm_surface_state.c 2011-01-02 20:58:35.000000000 -0200
+++ patched/src/mesa/drivers/dri/i965/brw_wm_surface_state.c    2011-01-24 16:35:47.063197793 -0200
@@ -599,6 +599,9 @@ prepare_wm_surfaces(struct brw_context *
         struct intel_renderbuffer *irb = intel_renderbuffer(rb);
         struct intel_region *region = irb ? irb->region : NULL;
 
+        if (!region)
+            continue;
+
         brw_add_validated_bo(brw, region->buffer);
         nr_surfaces = SURF_INDEX_DRAW(i) + 1;
       }


Then I compiled/rebooted and tried to run etracer again. This time I got a different segfault (but still a NULL region):


Program received signal SIGSEGV, Segmentation fault.
intel_region_buffer (intel=0xa5aaf28, region=0x0, flag=2)
    at intel_regions.c:514
514        if (region->pbo) {
(gdb) bt
#0  intel_region_buffer (intel=0xa5aaf28, region=0x0, flag=2)
    at intel_regions.c:514
#1  0xb6e78d21 in intelClearWithBlit (ctx=0xa5aaf28, mask=2)
    at intel_blit.c:262
#2  0xb6e7bccd in intelClear (ctx=0xa5aaf28, mask=2) at intel_clear.c:177
#3  0xb7083958 in _mesa_Clear (mask=0) at main/clear.c:241
#4  0xb72f2e07 in __glXDisp_Clear (pc=0xb1ea3040 "")
    at indirect_dispatch.c:1335
#5  0xb731bd29 in __glXDisp_Render (cl=0xa206eb0, pc=0xb1ea303c "\b")
    at glxcmds.c:1847
#6  0xb7320870 in __glXDispatch (client=0xa206dd8) at glxext.c:600
#7  0x08070fff in Dispatch () at dispatch.c:432
#8  0x080625ba in main (argc=8, argv=0xbfbd3464, envp=0xbfbd3488) at main.c:291
(gdb) display region
1: region = (struct intel_region *) 0x0


But this backtrace is the one PCPA reported for Salome:
https://bugs.freedesktop.org/show_bug.cgi?id=27333 (see backtrace on last comment).

So I tried PCPA's patch:

diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c
index 2c85ad3..e369783 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.c
+++ b/src/mesa/drivers/dri/intel/intel_blit.c
@@ -263,6 +263,9 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask)

       /* OK, clear this renderbuffer */
       irb = intel_get_renderbuffer(fb, buf);
+      if (irb->region == NULL)
+         goto clear_bit;
+
       write_buffer = intel_region_buffer(intel, irb->region,
                                         all ? INTEL_WRITE_FULL :
                                         INTEL_WRITE_PART);
@@ -370,6 +373,7 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
       if (intel->always_flush_cache)
         intel_batchbuffer_emit_mi_flush(intel->batch);

+  clear_bit:
       if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL)
         mask &= ~(BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL);
       else

Then I tried to test... After playing for a while (without crashing, the game going fine), I stopped etracer and went to the desktop. It was in a very inconsistent state: the screen was not refreshing correctly, I could still see parts of the game in my desktop, maximizing/minimizing applications had a weird behavior/flickr... When redrawing parts of the screen, sometimes X draw not the top window, but the content of the window below it. I'm using kwin.

Then, even without being able to use konsole correctly (I was not seeing what I was typing), I launched etracer again. The screen was really flashing, many times per second. Really annoying to watch.

So I came here report the bug =)

Any hints on what could be going on? Something I could investigate? A breakpoint to add? Do you need any "printf"s inside the code?

Thanks,
Paulo