Skip to content

agoessling/bazel_arm_toolchains

Repository files navigation

bazel_arm_toolchains

Provides hermetic Bazel C/C++ toolchains from ARM.

Usage

WORKSPACE

To incorporate bazel_arm_toolchains toolchains into your project, copy the following into your WORKSPACE file.

load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

http_archive(
    name = "bazel_arm_toolchains",
    # See release page for latest version url and sha.
)

load("@bazel_arm_toolchains//toolchains:toolchains.bzl", "all_toolchain_deps")

all_toolchain_deps()

This will bring all of the available toolchains (and their associated platform definitions) into your project. The toolchains will only be downloaded when actually utilized, but if you prefer, you can only import a specific toolchain:

load("@bazel_arm_toolchains//toolchains:toolchains.bzl", "toolchain_deps")

toolchain_deps(
    host_arch = "x86_64",
    host_os = "linux",
    target = "arm-none-eabi",
    version = "12.3.rel1",
)

host_arch, host_os, target, and version should match the naming convention of the ARM pre-built toolchains.

NOTE: The linux host OS is omitted from the ARM toolchain name.

Available Toolchains and CPUs

Currently bazel_arm_toolchains provides the following toolchains:

Host Architecture Host OS Target Version
x86_64 linux arm-none-eabi 12.3.rel1

and the following CPUs:

CPU
cortex-m0
cortex-m0plus
cortex-m1
cortex-m3
cortex-m4
cortex-m7
cortex-m7+nofp.dp
cortex-a8
cortex-a9

Available options match the -mpu= GCC flags.

These lists are easily expanded. If a toolchain or CPU of interest isn't available feel free to submit and issue, or alternatively take a look at _AVAILABLE_TOOLCHAINS in setup_toolchains.py and create a pull request. Don't forget to actually run setup_toolchains.py after adding a toolchain and before submitting a PR:

bazel run //:setup_toolchains

Platforms

bazel_arm_toolchains defines a different platform for each toolchain, CPU combo:

native.platform(
    name = {cpu}-{target_os}-{version},
    constraint_values = [
        "//platforms/cpu:{cpu}",
        "@platforms//os:{target_os}",
        "//platforms/toolchain_version:{version}",
    ],
    visibility = ["//visibility:public"],
)

To see all available platforms run:

bazel query @bazel_arm_toolchains//platforms:*

Building With Toolchain

In order to enable toolchain selection via platforms, Bazel requires a special flag along with the target platform:

bazel build --incompatible_enable_cc_toolchain_resolution --platforms=@bazel_arm_toolchains//platforms:{cpu}-{target_os}-{version} //...

The host architecture and OS will be derived automatically and the correct toolchain used.

The ergonomics can be improved by placing the flags in a .bazelrc file:

build --incompatible_enable_cc_toolchain_resolution
build --platforms=@bazel_arm_toolchains//platforms:{cpu}-{target_os}-{version}

Then a simple bazel build //... will utilize the desired toolchain.

Direct Tool Access

Direct access to the various tools (gdb, objdump, readelf, etc.) can be achieved through the various wrapper targets in toolchains/tool_wrappers. For example:

bazel run @bazel_arm_toolchains//toolchains/tool_wrappers/x86_64/linux/arm-none-eabi/12.3.rel1:readelf

Creating alias targets in your project may aid ergonomics:

alias(
    name = "objdump",
    actual = "@bazel_arm_toolchains//toolchains/tool_wrappers/x86_64/linux/arm-none-eabi/12.3.rel1:objdump",
)

These tools can also be utilized indirectly via genrules:

genrule(
    name = "objdump_output",
    srcs = [":binary"],
    outs = ["objdump_output.txt"],
    cmd = "$(location :objdump) -x $< > $@",
    tools = [":objdump"],
)