Bug 79421 - [llvmpipe] SIGSEGV src/gallium/drivers/llvmpipe/lp_rast_priv.h:218
Summary: [llvmpipe] SIGSEGV src/gallium/drivers/llvmpipe/lp_rast_priv.h:218
Status: RESOLVED FIXED
Alias: None
Product: Mesa
Classification: Unclassified
Component: Other (show other bugs)
Version: git
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: mesa-dev
QA Contact:
URL:
Whiteboard:
Keywords: have-backtrace
Depends on:
Blocks:
 
Reported: 2014-05-29 17:04 UTC by Vinson Lee
Modified: 2014-05-31 20:50 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Vinson Lee 2014-05-29 17:04:10 UTC
mesa: c0bd206a148b8fcda253359efe8c95871f89bd05 (master 10.3.0-devel)

Run piglit test fbo-drawbuffers-none glBlendFunci on llvmpipe.

$ ./bin/fbo-drawbuffers-none glBlendFunci -auto
Testing glBlendFunci.
Segmentation fault (core dumped)

(gdb) bt
#0  0x00007f49681c56a7 in lp_rast_get_unswizzled_color_block_pointer (task=0x1e27838, buf=0, x=0, y=0, layer=0)
    at src/gallium/drivers/llvmpipe/lp_rast_priv.h:218
#1  0x00007f49681c6328 in lp_rast_shade_tile (task=0x1e27838, arg=...) at src/gallium/drivers/llvmpipe/lp_rast.c:316
#2  0x00007f49681c6d5b in do_rasterize_bin (task=0x1e27838, bin=0x7f496b95b1e8, x=0, y=0)
    at src/gallium/drivers/llvmpipe/lp_rast.c:589
#3  0x00007f49681c6dc8 in rasterize_bin (task=0x1e27838, bin=0x7f496b95b1e8, x=0, y=0)
    at src/gallium/drivers/llvmpipe/lp_rast.c:608
#4  0x00007f49681c6ed8 in rasterize_scene (task=0x1e27838, scene=0x7f496b95b010)
    at src/gallium/drivers/llvmpipe/lp_rast.c:657
#5  0x00007f49681c71a4 in thread_function (init_data=0x1e27838) at src/gallium/drivers/llvmpipe/lp_rast.c:778
#6  0x00007f49681c4ce9 in impl_thrd_routine (p=0x1e18c40) at include/c11/threads_posix.h:87
#7  0x00007f496716c182 in start_thread (arg=0x7f4966b0f700) at pthread_create.c:312
#8  0x00007f496afb330d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111
(gdb) frame 0
#0  0x00007f49681c56a7 in lp_rast_get_unswizzled_color_block_pointer (task=0x1e27838, buf=0, x=0, y=0, layer=0)
    at src/gallium/drivers/llvmpipe/lp_rast_priv.h:218
218	   format_bytes = util_format_get_blocksize(task->scene->fb.cbufs[buf]->format);
(gdb) print task->scene->fb.cbufs[buf]
$1 = (struct pipe_surface *) 0x0
Comment 1 Roland Scheidegger 2014-05-29 22:14:24 UTC
I can't reproduce this, and valgrind didn't show anything suspicious neither.
Is this a regression?
I see the problem though, we're missing a test for non-zero cbuf in lp_rast_shade_tile (which we already do in lp_rast_shade_quads_mask).
For some reason though when I run the test it never touches lp_rast_shade_tile just lp_rast_shade_quads_mask which certainly explains why it doesn't crash - from a quick glance with the geometry I get lp_rast_shade_tile should be impossible. In any case I increased the FB_SIZE in the piglit test which made it crash indeed. I'm going to post a fix...
Comment 2 Roland Scheidegger 2014-05-31 20:50:28 UTC
I still have no idea why this could happen with the fb size used by this piglit test but fixed with 576868140bbb1abd177e7fd122720883d773137e.


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct.