Intermittent SIGILL on build, restore and/or publish

[ksitarek]

Hi All,

Describe the bug

Repro repo: https://github.com/ksitarek/sigill_dotnet_docker (plain console app with hello world)

Intermittent Illegal Instruction on various commands such as dotnet restore, dotnet build, dotnet publish when ran on official docker sdk image.

Seem to be working outside of docker (aka plain build via Rider).

To Reproduce

1. Clone repo https://github.com/ksitarek/sigill_dotnet_docker
2. Attempt DOCKER_BUILDKIT=0 docker build . --no-cache (might require couple of tries, as sometimes it "just works")

Exceptions (if any)

Illegal instruction

Further technical details

• Happens regularly on Apple M4 Max
• Does not happen on Apple M2

❯ DOCKER_BUILDKIT=0 docker build . --no-cache
DEPRECATED: The legacy builder is deprecated and will be removed in a future release.
            BuildKit is currently disabled; enable it by removing the DOCKER_BUILDKIT=0
            environment-variable.

Sending build context to Docker daemon  9.216kB
Step 1/18 : FROM mcr.microsoft.com/dotnet/runtime:9.0 AS base
 ---> 4fb8b01c8a0d
Step 2/18 : USER $APP_UID
 ---> Running in 4c226e3d8f88
 ---> Removed intermediate container 4c226e3d8f88
 ---> d2d42b51fb96
Step 3/18 : WORKDIR /app
 ---> Running in 911b3e8c95c1
 ---> Removed intermediate container 911b3e8c95c1
 ---> faf61b2ce781
Step 4/18 : FROM mcr.microsoft.com/dotnet/sdk:9.0 AS build
 ---> 3fcf6f1e809c
Step 5/18 : ARG BUILD_CONFIGURATION=Release
 ---> Running in 0b2927713c0c
 ---> Removed intermediate container 0b2927713c0c
 ---> 1b95917bc1bd
Step 6/18 : WORKDIR /src
 ---> Running in 2e3304a49655
 ---> Removed intermediate container 2e3304a49655
 ---> 89c273297771
Step 7/18 : COPY ["ConsoleApp1/ConsoleApp1.csproj", "ConsoleApp1/"]
 ---> de94c9a92034
Step 8/18 : RUN dotnet restore "ConsoleApp1/ConsoleApp1.csproj"
 ---> Running in ecafd1770e36
  Determining projects to restore...
  Restored /src/ConsoleApp1/ConsoleApp1.csproj (in 26 ms).
 ---> Removed intermediate container ecafd1770e36
 ---> b90d6820a3d0
Step 9/18 : COPY . .
 ---> fbef895cd601
Step 10/18 : WORKDIR "/src/ConsoleApp1"
 ---> Running in 080297ca64c3
 ---> Removed intermediate container 080297ca64c3
 ---> 90eccb1786f0
Step 11/18 : RUN dotnet build "ConsoleApp1.csproj" -c $BUILD_CONFIGURATION -o /app/build
 ---> Running in 2d569a8b64c4
  Determining projects to restore...
  All projects are up-to-date for restore.
Illegal instruction
The command '/bin/sh -c dotnet build "ConsoleApp1.csproj" -c $BUILD_CONFIGURATION -o /app/build' returned a non-zero code: 132

[ksitarek]

On another (much more complex project, but same setup with pretty much same Dockerfile) I was somehow able to catch following error:
/usr/share/dotnet/sdk/9.0.101/Roslyn/Microsoft.CSharp.Core.targets(89,5): error MSB6006: "csc.dll" exited with code 132.

This might be red herring, but 132 seem to match SIGILL.

[gingters]

One colleague of mine has this issue with his M4 too (building a project in docker). On my Apple M3 Max this does not happen.

[alexaka1]

I am not crazy after all. I am having massive issues on my M4 Pro with dotnet and docker. To the point where I have literally setup a while ! build; do sleep 1; done as a build script in my repos. I have no idea how to fix it, other than brute forcing it (running multiple times without changing anything) works. This was not an issue on an M3 Max for me. I am using buildkit however, and this issue comes up for both native arm64 and emulated amd64 builds as well.

I can pretty much reliable reproduce the issue, so if anyone has some insight what I need to enable to provide debug logs, core dumps whatever, let me know. I'm happy to help, I just have no idea what I need to do.

[sebaspalmyt]

I also can reproduce it most of the time. Is really rare that it works. Is exactly as described. The issue started a couple of weeks ago. In Mac M1 it works without problems. Mac M4 is the problem.

[denis-nikolayev-warby]

having the same issue with m4, just getting Illegal instruction
the same project works fine on M1 and on M3 with no issues

[baronfel]

@rainersigwald do we have any known problems with MSBuild on m4 Macs, or should be bring the runtime folks in to consult on this one?

[rainersigwald]

Nothing that I know of, and I don't think MSBuild (or SDK) could easily cause this, I think JIT folks need to look.

[alexaka1]

Something changed with 9.0.200/mcr.microsoft.com/dotnet/sdk:9.0@sha256:7f8e8b1514a2eeccb025f1e9dd554e191b21afa7f43f8321b7bd2009cdd59a1d, as using it for 3 days I have not experienced the problem. Fingers crossed. 🤞