ci: add riscv64-linux build via cross-compilation on ubuntu-24.04

Rework from native RISE runner to standard ubuntu-24.04 runner with
CMake cross-compilation, following alexcrichton's preference.

Changes:
- ci/docker/Dockerfile.riscv64-linux: Ubuntu 24.04 with
  crossbuild-essential-riscv64; sets CC/CXX to riscv64-linux-gnu-gcc
  so CMake detects cross-compilation and LLVM builds a native tablegen
  before cross-compiling the rest of the toolchain.
- ci/docker-build.sh: use artifact-specific Dockerfile if one exists
  (ci/docker/Dockerfile.<artifact>), fall back to default; make the
  wasmtime volume mount conditional on WASI_SDK_CI_SKIP_SYSROOT.
- .github/workflows/main.yml: new riscv64-linux matrix entry on
  ubuntu-24.04 with cross_cmake_args (-DCMAKE_SYSTEM_NAME=Linux
  -DCMAKE_SYSTEM_PROCESSOR=riscv64 -DWASI_SDK_LLDB=OFF) and
  WASI_SDK_CI_SKIP_SYSROOT=1; handle cross_cmake_args in the cmake
  flags step.

WASI_SDK_CI_SKIP_SYSROOT=1: the cross-compiled clang runs on riscv64,
not on the x86_64 build host, so the wasm sysroot step is skipped.
WASI_SDK_LLDB=OFF: avoids cross-compiling libedit/libxml2 in this
first iteration; can be re-enabled as a follow-up.

Closes #607

Signed-off-by: Bruno Verachten <gounthar@gmail.com>

.github/workflows/main.yml

@@ -28,6 +28,16 @@ jobs:
           - artifact: arm64-linux
             os: ubuntu-22.04-arm
 
+          - artifact: riscv64-linux
+            os: ubuntu-24.04
+            rust_target: riscv64-unknown-linux-gnu
+            cross_cmake_args: >-
+              -DCMAKE_SYSTEM_NAME=Linux
+              -DCMAKE_SYSTEM_PROCESSOR=riscv64
+              -DWASI_SDK_LLDB=OFF
+            env:
+              WASI_SDK_CI_SKIP_SYSROOT: 1
+
           - artifact: arm64-macos
             os: macos-14
             rust_target: aarch64-apple-darwin
@@ -85,6 +95,9 @@ jobs:
             rustup target add ${{ matrix.rust_target }}
             cmake_args="$cmake_args -DRUST_TARGET=${{ matrix.rust_target }}"
           fi
+          if [ "${{ matrix.cross_cmake_args }}" != "" ]; then
+            cmake_args="$cmake_args ${{ matrix.cross_cmake_args }}"
+          fi
           echo WASI_SDK_CI_TOOLCHAIN_CMAKE_ARGS="$cmake_args" >> $GITHUB_ENV
         shell: bash
 

ci/docker-build.sh

@@ -15,8 +15,13 @@ fi
 
 set -x
 
-# Build the Docker imager
-docker build --tag wasi-sdk-builder ci/docker
+# Build the Docker image. Use an artifact-specific Dockerfile if one exists
+# (e.g. ci/docker/Dockerfile.riscv64-linux), otherwise use the default.
+dockerfile=ci/docker/Dockerfile
+if [ -f "ci/docker/Dockerfile.$1" ]; then
+  dockerfile="ci/docker/Dockerfile.$1"
+fi
+docker build --tag wasi-sdk-builder --file "$dockerfile" ci/docker
 
 # Perform the build in `/src`. The current directory is mounted read-write at
 # this location as well. To ensure that container-created files are reasonable
@@ -34,9 +39,11 @@ args="$args --volume $ccache_dir:/ccache:Z --env CCACHE_DIR=/ccache"
 
 # Inherit some tools from the host into this container. This ensures that the
 # decision made on CI of what versions to use is the canonical source of truth
-# for theset ools
+# for these tools.
 args="$args --volume `rustc --print sysroot`:/rustc:ro"
-args="$args --volume $(dirname $(which wasmtime)):/wasmtime:ro"
+if [ "${WASI_SDK_CI_SKIP_SYSROOT:-}" != "1" ]; then
+  args="$args --volume $(dirname $(command -v wasmtime)):/wasmtime:ro"
+fi
 
 # Pass through some env vars that `build.sh` reads
 args="$args --env WASI_SDK_CI_TOOLCHAIN_CMAKE_ARGS"

ci/docker/Dockerfile.riscv64-linux

@@ -0,0 +1,30 @@
+# Ubuntu 24.04 is used here (rather than AlmaLinux 8) because it has
+# riscv64 cross-compilation packages in its repositories.
+FROM ubuntu:24.04
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+  curl \
+  ca-certificates \
+  crossbuild-essential-riscv64 \
+  clang \
+  lld \
+  python3 \
+  git \
+  unzip \
+  cmake \
+  ninja-build \
+  ccache \
+  && rm -rf /var/lib/apt/lists/*
+
+# Set the cross-compiler as the default so that cmake picks it up via CC/CXX
+# and sets CMAKE_CROSSCOMPILING=TRUE, which causes LLVM to build a native
+# tablegen before cross-compiling the rest of the toolchain.
+ENV CC=riscv64-linux-gnu-gcc
+ENV CXX=riscv64-linux-gnu-g++
+
+# Cargo needs an explicit linker when cross-compiling for riscv64.
+ENV CARGO_TARGET_RISCV64_UNKNOWN_LINUX_GNU_LINKER=riscv64-linux-gnu-gcc
+
+# Tell programs to cache in a location that both isn't a `--volume` mounted root
+# and isn't `/root` in the container as that won't be writable during the build.
+ENV XDG_CACHE_HOME=/tmp/cache