[Mesa-dev] [Bug 99527] Provide option for llvmpipe JIT code to run cleanly under valgrind

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Wed Jan 25 01:37:31 UTC 2017


https://bugs.freedesktop.org/show_bug.cgi?id=99527

            Bug ID: 99527
           Summary: Provide option for llvmpipe JIT code to run cleanly
                    under valgrind
           Product: Mesa
           Version: 13.0
          Hardware: x86-64 (AMD64)
                OS: Linux (All)
            Status: NEW
          Severity: normal
          Priority: medium
         Component: Other
          Assignee: mesa-dev at lists.freedesktop.org
          Reporter: john.firebaugh at gmail.com
        QA Contact: mesa-dev at lists.freedesktop.org

Currently llvmpipe JIT code is known to trigger errors when run under valgrind.
For example, bug #29922 reports the following, which I also observe:

==17795== Conditional jump or move depends on uninitialised value(s)
==17795==    at 0x573F792: ???
==17795==    by 0x4171342: lp_rast_shade_quads_mask (lp_rast.c:473)
==17795==    by 0x4173EE9: do_block_4_3 (lp_rast_tri_tmp.h:61)
==17795==    by 0x4178087: lp_rast_triangle_3_16 (lp_rast_tri.c:229)
==17795==    by 0x4171913: rasterize_bin (lp_rast.c:667)
==17795==    by 0x4171ACE: rasterize_scene (lp_rast.c:766)
==17795==    by 0x4171BA4: lp_rast_queue_scene (lp_rast.c:791)
==17795==    by 0x4178EB4: lp_scene_rasterize (lp_scene.c:405)
==17795==    by 0x4179DF4: lp_setup_rasterize_scene (lp_setup.c:158)
==17795==    by 0x417A296: set_scene_state (lp_setup.c:260)
==17795==    by 0x417A39C: lp_setup_flush (lp_setup.c:295)
==17795==    by 0x416E756: llvmpipe_flush (lp_flush.c:56)

That bug is closed as RESOLVED WONTFIX but I would like to ask that this be
reconsidered. Conscientious downstream developers want to make sure their code
runs cleanly under valgrind. If libraries they use trigger lots of errors, it
makes this task more difficult. For instance, I first had to determine whether
or not this error represented a misuse of OpenGL by my own code. In this case,
it's possible to search for "valgrind lp_rast_shade_quads_mask" and find the
above bug report, so I was able to reasonably conclude that this was not a bug
I was responsible for. In many of the other errors in JIT code that valgrind
reports, that's not the case, and I'm still not 100% sure of the status --
whether it's a bug in my code, a bug in llvm, a supposedly harmless use of an
uninitialized value, or a true false positive.

I'm not the only one dissatisfied with the status quo. For a more strongly
worded opinion, see
http://www.americanteeth.org/2013/08/14/valgrind-is-not-optional/.

If you believe that fixing these errors would harm performance of production
builds, please consider using the `--enable-valgrind` configure flag as an
explicit opt-in mechanism.

For reference, here are some of the other errors I have received:

==9337== Conditional jump or move depends on uninitialised value(s)
==9337==    at 0x402E63D: ???
==9337==    by 0xD32C84D: lp_rast_shade_quads_all (lp_rast_priv.h:271)
==9337==    by 0xD32C368: block_full_4 (lp_rast_tri.c:46)
==9337==    by 0xD329222: do_block_16_32_3 (lp_rast_tri_tmp.h:167)
==9337==    by 0xD328E52: lp_rast_triangle_32_3 (lp_rast_tri_tmp.h:305)
==9337==    by 0xD32073C: do_rasterize_bin (lp_rast.c:609)
==9337==    by 0xD3203EB: rasterize_bin (lp_rast.c:628)
==9337==    by 0xD31FBD1: rasterize_scene (lp_rast.c:688)
==9337==    by 0xD321823: thread_function (lp_rast.c:828)
==9337==    by 0xD321A61: impl_thrd_routine (threads_posix.h:87)
==9337==    by 0x4E42183: start_thread (pthread_create.c:312)
==9337==    by 0x6A6E37C: clone (clone.S:111)
==9337==  Uninitialised value was created by a heap allocation
==9337==    at 0x4C2B221: operator new(unsigned long) (in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/valgrind/3.12.0/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==9337==    by 0xDB14217: llvm::User::operator new(unsigned long, unsigned int)
(in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/mesa/13.0.3/lib/dri/swrast_dri.so)
==9337==    by 0xDA60CDA: llvm::ConstantFP::get(llvm::LLVMContext&,
llvm::APFloat const&) (in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/mesa/13.0.3/lib/dri/swrast_dri.so)
==9337==    by 0xDA629BD: llvm::ConstantFP::get(llvm::Type*, double) (in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/mesa/13.0.3/lib/dri/swrast_dri.so)
==9337==    by 0xD29993E: lp_build_const_elem (lp_bld_const.c:309)
==9337==    by 0xD2999F0: lp_build_const_vec (lp_bld_const.c:333)
==9337==    by 0xD29B902: lp_build_conv (lp_bld_conv.c:654)
==9337==    by 0xD29B08E: lp_build_conv_auto (lp_bld_conv.c:491)
==9337==    by 0xD344C3C: generate_unswizzled_blend (lp_state_fs.c:1884)
==9337==    by 0xD342505: generate_fragment (lp_state_fs.c:2452)
==9337==    by 0xD340947: generate_variant (lp_state_fs.c:2637)
==9337==    by 0xD33FC79: llvmpipe_update_fs (lp_state_fs.c:3204)
==9337== 

==9337== Thread 3 llvmpipe-1:
==9337== Use of uninitialised value of size 8
==9337==    at 0x4035AEE: ???
==9337==    by 0x40354D4: ???
==9337==    by 0xD31F6A6: lp_rast_shade_quads_mask (lp_rast.c:457)
==9337==    by 0xD32D069: do_block_4_32_2 (lp_rast_tri_tmp.h:67)
==9337==    by 0xD328943: do_block_16_32_2 (lp_rast_tri_tmp.h:152)
==9337==    by 0xD328610: lp_rast_triangle_32_2 (lp_rast_tri_tmp.h:305)
==9337==    by 0xD32073C: do_rasterize_bin (lp_rast.c:609)
==9337==    by 0xD3203EB: rasterize_bin (lp_rast.c:628)
==9337==    by 0xD31FBD1: rasterize_scene (lp_rast.c:688)
==9337==    by 0xD321823: thread_function (lp_rast.c:828)
==9337==    by 0xD321A61: impl_thrd_routine (threads_posix.h:87)
==9337==    by 0x4E42183: start_thread (pthread_create.c:312)
==9337==  Uninitialised value was created by a heap allocation
==9337==    at 0x4C2B221: operator new(unsigned long) (in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/valgrind/3.12.0/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==9337==    by 0xDB14217: llvm::User::operator new(unsigned long, unsigned int)
(in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/mesa/13.0.3/lib/dri/swrast_dri.so)
==9337==    by 0xDA60CDA: llvm::ConstantFP::get(llvm::LLVMContext&,
llvm::APFloat const&) (in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/mesa/13.0.3/lib/dri/swrast_dri.so)
==9337==    by 0xDA629BD: llvm::ConstantFP::get(llvm::Type*, double) (in
/home/travis/build/mapbox/mapbox-gl-native/mason_packages/linux-x86_64/mesa/13.0.3/lib/dri/swrast_dri.so)
==9337==    by 0xD29993E: lp_build_const_elem (lp_bld_const.c:309)
==9337==    by 0xD2999F0: lp_build_const_vec (lp_bld_const.c:333)
==9337==    by 0xD296F04: lp_build_sin_or_cos (lp_bld_arit.c:2914)
==9337==    by 0xD2965EE: lp_build_sin (lp_bld_arit.c:2930)
==9337==    by 0xD2C9A63: sin_emit_cpu (lp_bld_tgsi_action.c:2028)
==9337==    by 0xD2CC759: lp_build_tgsi_inst_llvm (lp_bld_tgsi.c:306)
==9337==    by 0xD2CCD79: lp_build_tgsi_llvm (lp_bld_tgsi.c:523)
==9337==    by 0xD2CF475: lp_build_tgsi_soa (lp_bld_tgsi_soa.c:4058)
==9337==

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170125/97e73903/attachment.html>


More information about the mesa-dev mailing list