Package: mesa-vulkan-drivers
Version: 26.1.5-1
Severity: important
X-Debbugs-Cc: debian-riscv@lists.debian.org
Tags: patch
User: debian-riscv@lists.debian.org
Usertags: riscv64
Control: forwarded -1
https://gitlab.freedesktop.org/mesa/mesa/-/work_items/15946
Control: affects -1 + src:dawn

Dear maintainer,

I am using Mesa's software rasterizer (lavapipe) on a riscv64 SBC
(StarFive VisionFive 2, JH7110 SoC)
via Dawn/WebGPU in a normal desktop session.

lavapipe works, but any Vulkan program that
creates a VkDevice via lavapipe and lets that device be destroyed from
an atexit handler
or C++ file-scope destructor crashes with exit code 139 (SIGSEGV), 134
(SIGABRT), or just hangs for hours (Debian CI).

dawn's riscv64 autopkgtest reproduces this exactly:
https://ci.debian.net/data/autopkgtest/unstable/riscv64/d/dawn/73952498/log.gz

run-simple-vulkan-program (hung, killed by the 20000s autopkgtest timeout):
  1263s Building simple-vulkan-program...
  1263s Running simple-vulkan-program (lavapipe)...
  21263s autopkgtest [21:26:07]: ERROR: timed out on command ...
run-simple-vulkan-program
  21264s run-simple-vulkan-program FAIL timed out

run-simple-dawn-program (segfault):
  21758s Building simple-dawn-program...
  21763s Running simple-dawn-program (lavapipe)...
  21763s Segmentation fault
  21764s run-simple-dawn-program FAIL non-zero exit status 139

The same 2 tiny test programs work fine on amd64 and arm64.

I added a reproducer and patch to the upstream issue:
https://gitlab.freedesktop.org/mesa/mesa/-/work_items/15946

Can Debian's Mesa carry the patch or should I wait until upstream fixes it?

Best
-Dominique