Building Toolchains with Guix
In order to deploy embedded software using Guix we first need to teach Guix how to cross-compile it. Since Guix builds everything from source, this means we must teach Guix how to build our cross-compilation toolchain.
The Zephyr Project uses its own fork of GCC with custom configs for the architectures supported by the project. In this article, we describe the cross-compilation toolchain we defined for Zephyr; it is implemented as a Guix channel.
Zephyr is a real-time operating system from the Linux Foundation. It aims to provide a common environment which can target even the most resource constrained devices.
Zephyr introduces a module system which allows third parties to share code in a uniform way. Zephyr uses CMake to perform physical component composition of these modules. It searches the filesystem and generates scripts which the toolchain will use to successfully combine those components into a firmware image.
The fact that Zephyr provides this mechanism is one reason I chose to target it in the first place.
This separation of modules in an embedded context is a really great thing. It brings many of the advantages that it brings to the Linux world such as code re-use, smaller binaries, more efficient cache/RAM usage, etc. It also allows us to work as independent groups and compose contributions from many teams.
It also brings all of the complexity. Suddenly most of the problems that plague traditional deployment now apply to our embedded system. The fact that the libraries are statically linked at compile time instead of dynamically at runtime is simply an implementation detail. I say most because everything is statically linked so there is no runtime component discovery that needs to be accounted for.
Anatomy of a Toolchain
Toolchains are responsible for taking high level descriptions of programs and lowering them down to a series of equivalent machine instructions. This process involves more than just a compiler. The compiler uses the GNU Binutils to manipulate its internal representation down to a given architecture. It also needs the use of the C standard library as well as a few other libraries needed for some compiler optimizations.
The C library provides the interface to the underlying kernel. System
read are provided by GNU C Library
(glibc) on most distributions.
In embedded systems, smaller implementations like RedHat's newlib and newlib-nano are used.
Bootstrapping a Toolchain
In order to compile GCC we need a C library that's been compiled for our target architecture. How can we cross compile our C library if we need our C library to build a cross compiler? The solution is to build a simpler compiler that doesn't require the C library to function. It will not be capable of as many optimizations and it will be very slow, however it will be able to build the C libraries as well as the complete version of GCC.
In order to build the simpler compiler we need to compile the Binutils to work with our target architecture. Binutils can be bootstrapped with our host GCC and have no target dependencies. More information is available in this article.
Doesn't sound so bad right? It isn't... in theory. However internet forums since time immemorial have been littered with the laments of those who came before. From incorrect versions of ISL to the wrong C library being linked or the host linker being used, etc. The one commonality between all of these issues is the environment. Building GCC is difficult because isolating build environments is hard.
In fact as of
v0.14.2, the Zephyr “software development kit” (SDK) repository took down the build
instructions and posted a sign that read
"Building this is too complicated, don't worry about it."
(I'm paraphrasing, but
not by much.)
We will neatly sidestep all of these problems and not risk destroying or polluting our host system with garbage by using Guix to manage our environments for us.
Our toolchain only requires the first pass compiler because newlib(-nano) is statically linked and introduced to the toolchain by normal package composition.
Defining the Packages
All of the base packages are defined in
Zephyr modules (coming soon!) are defined in
following the pattern of other module systems implemented by Guix.
First thing we need to build is the
This is very easy in Guix.
(define-public arm-zephyr-eabi-binutils (let ((xbinutils (cross-binutils "arm-zephyr-eabi"))) (package (inherit xbinutils) (name "arm-zephyr-eabi-binutils") (version "2.38") (source (origin (method git-fetch) (uri (git-reference (url "https://github.com/zephyrproject-rtos/binutils-gdb") (commit "6a1be1a6a571957fea8b130e4ca2dcc65e753469"))) (file-name (git-file-name name version)) (sha256 (base32 "0ylnl48jj5jk3jrmvfx5zf8byvwg7g7my7jwwyqw3a95qcyh0isr")))) (arguments `(#:tests? #f ,@(substitute-keyword-arguments (package-arguments xbinutils) ((#:configure-flags flags) `(cons "--program-prefix=arm-zephyr-eabi-" ,flags))))) (native-inputs (modify-inputs (package-native-inputs xbinutils) (prepend texinfo bison flex gmp dejagnu))) (home-page "https://zephyrproject.org") (synopsis "Binutils for the Zephyr RTOS"))))
returns a package which has been configured for the given GNU triplet.
We simply inherit that package and replace the source. The Zephyr build
system expects the binutils to be prefixed with
is accomplished by adding another flag to the
We can test our package definition using the
-L flag with
to add our packages.
$ guix build -L guix-zephyr zephyr-binutils /gnu/store/...-zephyr-binutils-2.38
This directory contains the results of
GCC sans libc
This one is a bit more involved. Don't be afraid! This version of GCC wants ISL version 0.15. It's easy enough to make that happen. Inherit the current version of ISL and swap out the source and update the version. For most packages the build process doesn't change that much between versions.
(define-public isl-0.15 (package (inherit isl) (version "0.15") (source (origin (method url-fetch) (uri (list (string-append "mirror://sourceforge/libisl/isl-" version ".tar.gz"))) (sha256 (base32 "11vrpznpdh7w8jp4wm4i8zqhzq2h7nix71xfdddp8xnzhz26gyq2"))))))
Like the binutils, there is a
for creating cross-GCC packages. This one accepts keywords specifying
which binutils and libc to use. If libc isn't given (like here), gcc is
configured with many options disabled to facilitate being built without
libc. Therefore we need to add the extra options we want (I got them
from the SDK configuration scripts in the
Git repository as well as the
commits to use for each of the tools).
(define-public gcc-arm-zephyr-eabi-12 (let ((xgcc (cross-gcc "arm-zephyr-eabi" #:xbinutils zephyr-binutils))) (package (inherit xgcc) (version "12.1.0") (source (origin (method git-fetch) (uri (git-reference (url "https://github.com/zephyrproject-rtos/gcc") (commit "0218469df050c33479a1d5be3e5239ac0eb351bf"))) (file-name (git-file-name (package-name xgcc) version)) (sha256 (base32 "1s409qmidlvzaw1ns6jaanigh3azcxisjplzwn7j2n3s33b76zjk")) (patches (search-patches "gcc-12-cross-environment-variables.patch" "gcc-cross-gxx-include-dir.patch")))) (native-inputs (modify-inputs (package-native-inputs xgcc) ;; Get rid of stock ISL (delete "isl") ;; Add additional dependencies that xgcc doesn't have ;; including our special ISL (prepend flex isl-0.15))) (arguments (substitute-keyword-arguments (package-arguments xgcc) ((#:phases phases) `(modify-phases ,phases (add-after 'unpack 'fix-genmultilib (lambda _ (patch-shebang "gcc/genmultilib"))) (add-after 'set-paths 'augment-CPLUS_INCLUDE_PATH (lambda* (#:key inputs #:allow-other-keys) (let ((gcc (assoc-ref inputs "gcc"))) ;; Remove the default compiler from CPLUS_INCLUDE_PATH to ;; prevent header conflict with the GCC from native-inputs. (setenv "CPLUS_INCLUDE_PATH" (string-join (delete (string-append gcc "/include/c++") (string-split (getenv "CPLUS_INCLUDE_PATH") #\:)) ":")) (format #t "environment variable `CPLUS_INCLUDE_PATH' changed to `a`%" (getenv "CPLUS_INCLUDE_PATH"))))))) ((#:configure-flags flags) ;; The configure flags are largely identical to the flags used by the ;; "GCC ARM embedded" project. `(append (list "--enable-multilib" "--with-newlib" "--with-multilib-list=rmprofile" "--with-host-libstdcxx=-static-libgcc -Wl,-Bstatic,-lstdc++,-Bdynamic -lm" "--enable-plugins" "--disable-decimal-float" "--disable-libffi" "--disable-libgomp" "--disable-libmudflap" "--disable-libquadmath" "--disable-libssp" "--disable-libstdcxx-pch" "--disable-nls" "--disable-shared" "--disable-threads" "--disable-tls" "--with-gnu-ld" "--with-gnu-as" "--enable-initfini-array") (delete "--disable-multilib" ,flags))))) (native-search-paths (list (search-path-specification (variable "CROSS_C_INCLUDE_PATH") (files '("arm-zephyr-eabi/include"))) (search-path-specification (variable "CROSS_CPLUS_INCLUDE_PATH") (files '("arm-zephyr-eabi/include" "arm-zephyr-eabi/c++" "arm-zephyr-eabi/c++/arm-zephyr-eabi"))) (search-path-specification (variable "CROSS_LIBRARY_PATH") (files '("arm-zephyr-eabi/lib"))))) (home-page "https://zephyrproject.org") (synopsis "GCC for the Zephyr RTOS"))))
This GCC can be built like so.
$ guix build -L guix-zephyr gcc-cross-sans-libc-arm-zephyr-eabi /gnu/store/...-gcc-cross-sans-libc-arm-zephyr-eabi-12.1.0-lib /gnu/store/...-gcc-cross-sans-libc-arm-zephyr-eabi-12.1.0
Great! We now have our stage-1 compiler.
The newlib package package is quite straight forward (relatively).
It is mostly adding in the relevent configuration flags and patching
the files the
patch-shebangs phase missed.
(define-public zephyr-newlib (package (name "zephyr-newlib") (version "3.3") (source (origin (method git-fetch) (uri (git-reference (url "https://github.com/zephyrproject-rtos/newlib-cygwin") (commit "4e150303bcc1e44f4d90f3489a4417433980d5ff"))) (sha256 (base32 "08qwjpj5jhpc3p7a5mbl7n6z7rav5yqlydqanm6nny42qpa8kxij")))) (build-system gnu-build-system) (arguments `(#:out-of-source? #t #:configure-flags '("--target=arm-zephyr-eabi" "--enable-newlib-io-long-long" "--enable-newlib-io-float" "--enable-newlib-io-c99-formats" "--enable-newlib-retargetable-locking" "--enable-newlib-lite-exit" "--enable-newlib-multithread" "--enable-newlib-register-fini" "--enable-newlib-extra-sections" "--disable-newlib-wide-orient" "--disable-newlib-fseek-optimization" "--disable-newlib-supplied-syscalls" "--disable-newlib-target-optspace" "--disable-nls") #:phases (modify-phases %standard-phases (add-after 'unpack 'fix-references-to-/bin/sh (lambda _ (substitute# '("libgloss/arm/cpu-init/Makefile.in" "libgloss/arm/Makefile.in" "libgloss/libnosys/Makefile.in" "libgloss/Makefile.in") (("/bin/sh") (which "sh"))) #t))))) (native-inputs `(("xbinutils" ,zephyr-binutils) ("xgcc" ,gcc-arm-zephyr-eabi-12) ("texinfo" ,texinfo))) (home-page "https://www.sourceware.org/newlib/") (synopsis "C library for use on embedded systems") (description "Newlib is a C library intended for use on embedded systems. It is a conglomeration of several library parts that are easily usable on embedded products.") (license (license:non-copyleft "https://www.sourceware.org/newlib/COPYING.NEWLIB"))))
And the build.
$ guix build -L guix-zephyr zephyr-newlib /gnu/store/...-zephyr-newlib-3.3
Mostly complete. libstdc++ does not build because
arm-zephyr-eabi is not
arm-none-eabi so a dynamic link check is
performed/failed. I cannot figure out how crosstool-ng handles this.
Now that we've got the individual tools it's time to create our complete toolchain. For this we need to do some package transformations. Because these transformations are going to have to be done for every combination of binutils/gcc/newlib it is best to create a function which we can reuse for every version of the SDK.
(define (arm-zephyr-eabi-toolchain xgcc newlib version) "Produce a cross-compiler zephyr toolchain package with the compiler XGCC and the C\n library variant NEWLIB." (let ((newlib-with-xgcc (package (inherit newlib) (native-inputs (modify-inputs (package-native-inputs newlib) (replace "xgcc" xgcc)))))) (package (name (string-append "arm-zephyr-eabi" (if (string=? (package-name newlib-with-xgcc) "newlib-nano") "-nano" "") "-toolchain")) (version version) (source #f) (build-system trivial-build-system) (arguments '(#:modules ((guix build union) (guix build utils)) #:builder (begin (use-modules (ice-9 match) (guix build union) (guix build utils)) (let ((out (assoc-ref %outputs "out"))) (mkdir-p out) (match %build-inputs (((names . directories) ...) (union-build (string-append out "/arm-zephyr-eabi") directories))))))) (inputs `(("binutils" ,zephyr-binutils) ("gcc" ,xgcc) ("newlib" ,newlib-with-xgcc))) (synopsis "Complete GCC tool chain for ARM zephyrRTOS development") (description "This package provides a complete GCC tool chain for ARM bare metal development with zephyr rtos. This includes the GCC arm-zephyr-eabi cross compiler and newlib (or newlib-nano) as the C library. The supported programming language is C.") (home-page (package-home-page xgcc)) (license (package-license xgcc)))))
This function creates a special package which consists of the toolchain
in a special directory hierarchy, i.e
Our complete toolchain definition looks like this.
(define-public arm-zephyr-eabi-toolchain-0.15.0 (arm-zephyr-eabi-toolchain gcc-arm-zephyr-eabi-12 zephyr-newlib "0.15.0"))
$ guix build -L guix-zephyr arm-zephyr-eabi-toolchain /gnu/store/...-arm-zephyr-eabi-toolchain-0.15.0
Note: Guix now includes a mechanism to describe platforms at a high level, and which the
--targetbuild options build upon. It is not used here but could be a way to better integrate Zephyr support in the future.
Integrating with Zephyr Build System
Zephyr uses CMake as its build system. It contains numerous CMake files in both the so-called
the zephyr source code repository, as well as a handful in the SDK which help select the correct toolchain
for a given board.
There are standard locations the build system will look for the SDK. We are not using any of them.
Our SDK lives in the store, immutable forever.
According to the Zephyr documentation, the variable
ZEPHYR_SDK_INSTALL_DIR needs to point to our custom spot.
We also need to grab the CMake files from the
and create a file,
contains the version string
ZEPHYR_BASE uses to find a compatible SDK.
Along with the SDK proper we need to include a number of python packages required by the build system.
(define-public zephyr-sdk (package (name "zephyr-sdk") (version "0.15.0") (home-page "https://zephyrproject.org") (source (origin (method git-fetch) (uri (git-reference (url "https://github.com/zephyrproject-rtos/sdk-ng") (commit "v0.15.0"))) (file-name (git-file-name name version)) (sha256 (base32 "04gsvh20y820dkv5lrwppbj7w3wdqvd8hcanm8hl4wi907lwlmwi")))) (build-system trivial-build-system) (arguments `(#:modules ((guix build union) (guix build utils)) #:builder (begin (use-modules (guix build union) (ice-9 match) (guix build utils)) (let ((out (assoc-ref %outputs "out")) (cmake-scripts (string-append (assoc-ref %build-inputs "source") "/cmake")) (sdk-out (string-append out "/zephyr-sdk-0.15.0"))) (mkdir-p out) (match (assoc-remove! %build-inputs "source") (((names . directories) ...) (union-build sdk-out directories))) (copy-recursively cmake-scripts (string-append sdk-out "/cmake")) (with-directory-excursion sdk-out (call-with-output-file "sdk_version" (lambda (p) (format p "0.15.0")))))))) (propagated-inputs (list arm-zephyr-eabi-toolchain-0.15.0 zephyr-binutils dtc python-3 python-pyelftools python-pykwalify python-pyyaml python-packaging)) (native-search-paths (list (search-path-specification (variable "ZEPHYR_SDK_INSTALL_DIR") (separator #f) (files '(""))))) (synopsis "Zephyr SDK") (description "zephyr-sdk contains bundles a complete gcc toolchain as well as host tools like dtc, openocd, qemu, and required python packages.") (license license:apsl2)))
In order to test we will need an environment with the SDK installed.
We can take advantage of
guix shell to avoid installing test packages into
our home environment. This way if it causes problems we can just exit the shell
and try again.
guix shell -L guix-zephyr zephyr-sdk cmake ninja git
ZEPHYR_BASE can be cloned into a temporary workspace to test our toolchain functionality.
(For now. Eventually we will need to create a package for
zephyr-build-system can use.)
mkdir /tmp/zephyr-project cd /tmp/zephyr-project git clone https://github.com/zephyrproject-rtos/zephyr export ZEPHYR_BASE=/tmp/zephyr-project/zephyr
In order to build for the test board (k64f in this case) we need to get a hold of the vendor Hardware Abstraction Layers and CMSIS. (These will also need to become Guix packages to allow the build system to compose modules).
git clone https://github.com/zephyrproject-rtos/hal_nxp && \ git clone https://github.com/zephyrproject-rtos/cmsis
To inform the build system about this module we pass it in with
-DZEPHYR_MODULES= which is
a semicolon separated list of paths containing a module.yml file.
To build the hello world sample we use the following incantation.
cmake -Bbuild $ZEPHYR_BASE/samples/hello_world \ -GNinja \ -DBOARD=frdm_k64f \ -DBUILD_VERSION=3.1.0 \ -DZEPHYR_MODULES="/tmp/zephyr-project/hal_nxp;/tmp/zephyr-project/cmsis" \ && ninja -Cbuild
If everything is set up correctly we will end up with a
directory with all our build artifacts. The SDK is correctly installed!
A customized cross toolchain is one of the most difficult pieces of software to build. Using Guix, we do not need to be afraid of the complexity! We can fiddle with settings, swap out components, and do the most brain dead things to our environments without a care in the world. Just exit the environment and it's like it never happened at all.
It highlights one of my favorite aspects of Guix, every package is a working reference design for you to modify and learn from.
About GNU Guix
GNU Guix is a transactional package manager and an advanced distribution of the GNU system that respects user freedom. Guix can be used on top of any system running the Hurd or the Linux kernel, or it can be used as a standalone operating system distribution for i686, x86_64, ARMv7, AArch64 and POWER9 machines.
In addition to standard package management features, Guix supports transactional upgrades and roll-backs, unprivileged package management, per-user profiles, and garbage collection. When used as a standalone GNU/Linux distribution, Guix offers a declarative, stateless approach to operating system configuration management. Guix is highly customizable and hackable through Guile programming interfaces and extensions to the Scheme language.
Unless otherwise stated, blog posts on this site are copyrighted by their respective authors and published under the terms of the CC-BY-SA 4.0 license and those of the GNU Free Documentation License (version 1.3 or later, with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts).