STATUS_ACCESS_VIOLATION in CorePipelinePlugin

[Weihnachtsbaum]

Bevy version

0.13.2

System information

• cargo 1.80.0-nightly (84dc5dc11 2024-05-20)
• Windows 10

`AdapterInfo { name: "Intel(R) HD Graphics 5500", vendor: 32902, device: 5654, device_type: IntegratedGpu, driver: "", driver_info: "", backend: Dx12 }`

What you did

Initially I did this:

use bevy::prelude::*;
fn main() {
    App::new()
        .add_plugins(DefaultPlugins)
        .run()
}

Then I tried narrowing the problem down and ended up with this:

fn main() {
    bevy_app::App::new()
        // MinimalPlugins
        .add_plugins(bevy_core::TaskPoolPlugin::default())
        .add_plugins(bevy_core::TypeRegistrationPlugin)
        .add_plugins(bevy_core::FrameCountPlugin)
        .add_plugins(bevy_time::TimePlugin)
        .add_plugins(bevy_app::ScheduleRunnerPlugin::default())

        // required for CorePipelinePlugin to run
        .add_plugins(bevy_window::WindowPlugin::default())
        .add_plugins(bevy_asset::AssetPlugin::default())
        .add_plugins(bevy_render::RenderPlugin::default())
        .add_plugins(bevy_render::texture::ImagePlugin::default())

        // causes crash
        .add_plugins(bevy_core_pipeline::CorePipelinePlugin)

        .run();
}

What went wrong

The program crashes after a few seconds with the error:

error: process didn't exit successfully: `target\debug\bevy_crash.exe` (exit code: 0xc0000005, STATUS_ACCESS_VIOLATION)

[alice-i-cecile]

Are you able to run the examples of 'wgpu' correctly? It looks like this is likely to be a hardware or driver issue.

[Weihnachtsbaum]

Some examples work fine.
Many of the examples generate these warnings but still work:

[2024-05-28T18:30:04Z WARN  wgpu_core::instance] Missing downlevel flags: DownlevelFlags(VERTEX_AND_INSTANCE_INDEX_RESPECTS_RESPECTIVE_FIRST_VALUE_IN_INDIRECT_DRAW)
    The underlying API or device in use does not support enough features to be a fully compliant implementation of WebGPU. A subset of the features can still be used. If you are running this program on native and not in a browser and wish to limit the features you use to the supported subset, call Adapter::downlevel_properties or Device::downlevel_properties to get a listing of the features the current platform supports.
[2024-05-28T18:30:04Z WARN  wgpu_core::instance] DownlevelCapabilities {
        flags: DownlevelFlags(
            COMPUTE_SHADERS | FRAGMENT_WRITABLE_STORAGE | INDIRECT_EXECUTION | BASE_VERTEX | READ_ONLY_DEPTH_STENCIL | NON_POWER_OF_TWO_MIPMAPPED_TEXTURES | CUBE_ARRAY_TEXTURES | COMPARISON_SAMPLERS | INDEPENDENT_BLEND | VERTEX_STORAGE | ANISOTROPIC_FILTERING | FRAGMENT_STORAGE | MULTISAMPLED_SHADING | DEPTH_TEXTURE_AND_BUFFER_COPIES | WEBGPU_TEXTURE_FORMAT_SUPPORT | BUFFER_BINDINGS_NOT_16_BYTE_ALIGNED | UNRESTRICTED_INDEX_BUFFER | FULL_DRAW_INDEX_UINT32 | DEPTH_BIAS_CLAMP | VIEW_FORMATS | UNRESTRICTED_EXTERNAL_TEXTURE_COPIES | SURFACE_VIEW_FORMATS | NONBLOCKING_QUERY_RESOLVE,
        ),
        limits: DownlevelLimits,
        shader_model: Sm5,
    }

The repeated_compute example also generates these warnings:

[2024-05-28T18:37:55.955682900Z WARN  wgpu_hal::auxil::dxgi::exception] Process is terminating. Using simple reporting. Please call ReportLiveObjects() at runtime for standard reporting.
[2024-05-28T18:37:55.958813500Z WARN  wgpu_hal::auxil::dxgi::exception] Live Producer at 0x0000029E2618D9C8, Refcount: 44.
[2024-05-28T18:37:55.959932900Z WARN  wgpu_hal::auxil::dxgi::exception]         Live Object at 0x0000029E260D8E90, Refcount: 0.
-- snip --
[2024-05-28T18:37:56.049361800Z WARN  wgpu_hal::auxil::dxgi::exception] Live                         Object :     62

Only the conservative_raster example dosn't work. It generates this error:

error: process didn't exit successfully: `target\debug\wgpu-examples.exe conservative_raster` (exit code: 101)

[alice-i-cecile]

Awesome, thanks a ton for the investigation. I think the problem here is likely us escalating the warnings to a panic. The next step would be figuring out where that's going on, and then seeing if we can fail more gracefully.

[Weihnachtsbaum]

I tried to find the specific part of CorePipelinePlugin's build function that causes the error by removing everything that works fine. I ended up with this:

        load_internal_asset!(
            app,
            FULLSCREEN_SHADER_HANDLE,
            "fullscreen_vertex_shader/fullscreen.wgsl",
            Shader::from_wgsl
        );
        app.add_plugins(CopyDeferredLightingIdPlugin);

If I either remove the load_internal_asset! call or not add CopyDeferredLightingIdPlugin, the error disappears, so the problem is somewhere in there.

[xangelix]

@Weihnachtsbaum could you kindly check if you experience this same issue on the latest stable version of rust? I'm tracing an issue that may be related to yours.

[Weihnachtsbaum]

@xangelix it works fine on the latest stable version (1.78.0)

[xangelix]

@Weihnachtsbaum great! Could you also check nightly-2024-04-17 and nightly-2024-04-18?
I believe nightly-2024-04-18 is where things go wrong.

[Weihnachtsbaum]

@xangelix that's correct, nightly-2024-04-17 works and nightly-2024-04-18 doesn't

[xangelix]

Thanks! I unfortunately don't know enough about bevy/wgpu/rust/windows internals and don't have enough time to probably go any further, but nightly-2024-04-17 seems to work for now. Hopefully this helps someone else narrow the issue down quicker.

It appears this issue only concerns Windows builds (including cross-compilation with mingw) on nightly-2024-04-18 or greater. I have the same issue on bevy 0.13.2 and the current main branch.

Notably, I have the logs you described on my side in BOTH the working and non-working nightly versions:

[2024-05-30T16:26:34Z WARN  wgpu_hal::dx12::shader_compilation::dxc] Failed to load dxil.dll. Defaulting to FXC instead: Failed to load library "dxil.dll": LoadLibraryExW { source: Os { code: 126, kind: Uncategorized, message: "The specified module could not be found." } }
[2024-05-30T16:26:34Z WARN  wgpu_hal::gles::adapter] Returned GL context is 1.1, when 3.3+ is needed
[2024-05-30T16:26:34Z INFO  wgpu_core::instance] Adapter Dx12 AdapterInfo { name: "Microsoft Basic Render Driver", vendor: 5140, device: 140, device_type: Cpu, driver: "", driver_info: "", backend: Dx12 }
[2024-05-30T16:26:34Z WARN  wgpu_core::instance] Missing downlevel flags: DownlevelFlags(VERTEX_AND_INSTANCE_INDEX_RESPECTS_RESPECTIVE_FIRST_VALUE_IN_INDIRECT_DRAW)
    The underlying API or device in use does not support enough features to be a fully compliant implementation of WebGPU. A subset of the features can still be used. If you are running this program on native and not in a browser and wish to limit the features you use to the supported subset, call Adapter::downlevel_properties or Device::downlevel_properties to get a listing of the features the current platform supports.
[2024-05-30T16:26:34Z WARN  wgpu_core::instance] DownlevelCapabilities {
        flags: DownlevelFlags(
            COMPUTE_SHADERS | FRAGMENT_WRITABLE_STORAGE | INDIRECT_EXECUTION | BASE_VERTEX | READ_ONLY_DEPTH_STENCIL | NON_POWER_OF_TWO_MIPMAPPED_TEXTURES | CUBE_ARRAY_TEXTURES | COMPARISON_SAMPLERS | INDEPENDENT_BLEND | VERTEX_STORAGE | ANISOTROPIC_FILTERING | FRAGMENT_STORAGE | MULTISAMPLED_SHADING | DEPTH_TEXTURE_AND_BUFFER_COPIES | WEBGPU_TEXTURE_FORMAT_SUPPORT | BUFFER_BINDINGS_NOT_16_BYTE_ALIGNED | UNRESTRICTED_INDEX_BUFFER | FULL_DRAW_INDEX_UINT32 | DEPTH_BIAS_CLAMP | VIEW_FORMATS | UNRESTRICTED_EXTERNAL_TEXTURE_COPIES | SURFACE_VIEW_FORMATS | NONBLOCKING_QUERY_RESOLVE,
        ),
        limits: DownlevelLimits,
        shader_model: Sm5,
    }

I also never get the

error: process didn't exit successfully: `target\debug\bevy_crash.exe` (exit code: 0xc0000005, STATUS_ACCESS_VIOLATION)

error you described. In fact, I don't get any error at all. The program just closes after a few successful ticks, never drawing more than a black screen.

I would be glad to collect any additional logs/environment details if it helps.

[Weihnachtsbaum]

Are we sure that the problem lies within bevy or could it be in rustc?

[xangelix]

It could definitely be rustc afaik, but I would expect behavior changing in an unstable feature used by a bevy dependency (like wgpu) to also be likely. rustc by itself might actually be unlikely given the issue is still present 6 weeks later.

If you have time, a few ideas come to mind, but I'm really not the best person to ask.

• Are these "Missing downlevel flags" messages actually related to the issue? When do they appear? My windows VM uses a CPU renderer, is this specific to the problem? It does appear specific to Windows, which tells me GPU/dx12 seems likely? (is bevy vulkan compatible with windows? I don't think I can test bevy vulkan in windows with my CPU driver)
• What changes on nightly features were made between nightly-2024-04-17 and nightly-2024-04-18 (be careful with exact branch/compile times)? Does this issue exist on the 1.79.0 beta? (I believe the beta disables all nightly features, so you could then know if it was a stable change or an unstable feature change made between those two times).
• I can get log messages out of the first few game update ticks, and then it crashes. What system would crash/close here?

I'm also not especially familiar with rust or c development on windows, it's quite possible I didn't get any crash error because I don't have some windows debugging options that I don't know about enabled outside of rust.

The bevy devs will certainly have better intuition here.

[Weihnachtsbaum]

I just tested it out, it also crashes in 1.79.0-beta.7 (9ca20fadf 2024-05-22). Using nightly-2024-04-17 or prior works for now but as the readme says:

Bevy relies heavily on improvements in the Rust language and compiler. As a result, the Minimum Supported Rust Version (MSRV) is generally close to "the latest stable release" of Rust.

so in the future this won't be possible anymore.

Unfortunately, I am a Rust and Bevy beginner and don't think I will be very helpful in fixing this issue.

[xangelix]

thought I'd give it another go after the winit v0.30 merge, unfortunately the problem still exists on the latest nightly and bb51635

[xangelix]

@Weihnachtsbaum looks like things are happening! :) rust-lang/rust#126442
gfx-rs/wgpu@d309bba should fix this

[Weihnachtsbaum]

@xangelix Great! Do you know when we can expect this to be included in bevy?

[alice-i-cecile]

The fix is merged by wgpu here: gfx-rs/wgpu#5812

That should be going out in a new release shortly, and #13186 will grab it for 0.14 for sure. I believe they're also backporting it, which would be nice.

[djeedai]

Hanabi CI is hitting those "Missing downlevel flags" too: https://github.com/djeedai/bevy_hanabi/actions/runs/9609820286/job/26505417917, as well as example crashes:

error: process didn't exit successfully: `target\debug\examples\activate.exe` (exit code: 0xc000041d)

[alice-i-cecile]

This is not yet fixed: https://github.com/gfx-rs/wgpu/releases/tag/v0.20.1 does not include the relevant PR.

[alice-i-cecile]

https://github.com/gfx-rs/wgpu/releases/tag/v0.20.2 and wgpu is upgraded. Closing as fixed, thanks y'all!