Using wgpu in a test causes a segfault in WSL

[randomPoison]

Description

Creating a wgpu::Instance in a test causes the test to segfault when the test exits when running in Ubuntu under WSL (Windows subsystem for linux).

Repro steps

Running the following test under WSL triggers the crash:

#[test]
fn inline_test() {
    let instance = wgpu::Instance::new(&wgpu::InstanceDescriptor::default());
}

Expected vs observed behavior

I expected I'd be able to run wgpu logic in a regular Rust test, but instead it crashes with a segfault. This doesn't seem to happen under Windows, but it fails when run in WSL.

Extra materials

Output when running cargo test:

     Running unittests src/main.rs (target/debug/deps/main-07cb0cf27c3418f0)

running 1 test
libEGL warning: DRI3: Screen seems not DRI3 capable
libEGL warning: DRI3: Screen seems not DRI3 capable
MESA: error: ZINK: failed to choose pdev
libEGL warning: egl: failed to create dri2 screen
error: test failed, to rerun pass `--bin main`

Caused by:
  process didn't exit successfully: `/home/random/dev/wgpu-gol/target/debug/deps/main-07cb0cf27c3418f0` (signal: 11, SIGSEGV: invalid memory reference)

If I run under GDB and try to get the stacktrace it looks like it's dying under start_thread.

Thread 2 "inline_test" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7ffff7c8a6c0 (LWP 8174)]
0x00007fffe1aaab10 in ?? ()
(gdb) bt
#0  0x00007fffe1aaab10 in ?? ()
#1  0x00007ffff7d26340 in __GI___nptl_deallocate_tsd () at ./nptl/nptl_deallocate_tsd.c:73
#2  __GI___nptl_deallocate_tsd () at ./nptl/nptl_deallocate_tsd.c:22
#3  0x00007ffff7d29890 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:455
#4  0x00007ffff7db6c3c in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:78

I tried stepping through it manually to see if I could find out what's happening, seems like the test finishes successfully, i.e. it's not dying in the Instance destructor. Looks like after the test exits we then die in __GI___nptl_deallocate_tsd which is called as part of the cleanup in start_thread.

Platform

• Dev Environment: Ubuntu 24.024.2 LTS (running under WSL2)
• Host Environment: Windows 10
• System Architecture: x86-64
• wgpu Version: 25.0.0

[randomPoison]

Oh, another minor note: The segfault occurs when running the tests directly or running it under GDB, but it doesn't crash when run under rr. I'm not sure what that really tells us, though.

[Vecvec]

Have you tried other graphics applications (not using wgpu)? I found that even when using something like vlc or nautilus and closing the window caused a seg-fault with the default drivers.

[randomPoison]

I haven't tried other GPU apps, but the wgpu program I'm writing runs normally and exits without a segfault. This only happens when running tests with Rust's test infra, so I suspect it's not a system-wide issue.

[randomPoison]

Well, actually, now that I say that, I realize that I'm not spawning any threads in my app. So if this only happens in the presence of threads I wouldn't see it normally.

[cwfitzgerald]

I feel like there's some other issue going on here that wgpu is the one which is triggering the problem, possibly because we pull in a lot of dynamic libraries, gpu driver libraries, etc.

The rust test harness will actually make at least one thread to run the tests, so turning wgpu on is triggering some kind of issue.

I'm going to close this for now as I suspect there's not anything we can do about this, but if you or anyone else finds new information pointing to a more direct problem, do post and I can re-open it.