Note: The Fuchsia source includes Zircon. See Fuchsia's Getting Started documentation.
This guide assumes that the Fuchsia project is checked out into $FUCHSIA_DIR, and
fx has been configured.
fx command wraps the various tools used to configure, build and interact with Fuchsia. The
fx set is used to specify the “bringup” product and the board architecture. The “bringup” product identifies the Zircon components and excludes all other Fuchsia components from the build. For example, to build Zircon for arm64:
fx set bringup.arm64
Fuchsia uses the concept of products to create a collection of build targets. The “bringup” product is the smallest product, other product configurations can be listed by running:
The defined board architectures are listed by running:
To execute the build:
The build results are saved in $FUCHSIA_DIR/out/default.zircon.
By default Fuchsia uses the
clang toolchain. This can be set to
gcc by using the
variants argument with
fx set bringup.x64 --variant gcc
You can also enable asan by using the variant flag.
You can build for all targets with
fx multi and using a file that contains all the specifications to build. The output for each target is found in $FUCHSIA_DIR/out/..variant. An example of a multi build spec is bringup-cq which approximates what is built for a CQ test.
Please build for all targets before submitting to ensure builds work on all architectures.
You can skip this if you're only testing on actual hardware, but the emulator is handy for quick local tests and generally worth having around.
See QEMU for information on building and using QEMU with zircon.
If the prebuilt toolchain binaries do not work for you, you can build your own from vanilla upstream sources.
variants = [ "clang" ]or
variants = [ "asan" ].
Build one or the other or both, as needed for how you want build Zircon.
We use GNU
binutils 2.30[^1] and GCC 8.2[^2], configured with
--enable-initfini-array --enable-gold, and with
--target=x86_64-elf --enable-targets=x86_64-pep for x86-64 or
--target=aarch64-elf for ARM64.
binutils, we recommend
--enable-deterministic-archives but that switch is not necessary to get a working build.
For GCC, it's necessary to pass
MAKEOVERRIDES=USE_GCC_STDINT=provide on the
make command line. This should ensure that the
stdint.h GCC installs is one that works standalone (
stdint-gcc.h in the source) rather than one that uses
#include_next and expects another
stdint.h file installed elsewhere.
Only the C and C++ language support is required and no target libraries other than
libgcc are required, so you can use various
configure switches to disable other things and make your build of GCC itself go more quickly and use less storage, e.g.
--enable-languages=c,c++ --disable-libstdcxx --disable-libssp --disable-libquadmath. See the GCC installation documentation for more details.
You may need various other
configure switches or other prerequisites to build on your particular host system. See the GNU documentation.
binutils 2.30 release has some harmless
make check failures in the
x86_64-elf configurations. These are fixed on the upstream
binutils-2_30-branch git branch, which is what we actually build. But the 2.30 release version works fine for building Zircon; it just has some spurious failures in its own test suite.
[^2]: As of 2008-6-15, GCC 8.2 has not been released yet. There is no released version of GCC that works for building Zircon without backporting some fixes. What we actually use is the upstream
gcc-8-branch git branch.
We use a trunk snapshot of Clang and update to new snapshots frequently. Any build of recent-enough Clang with support for
aarch64 compiled in should work. You'll need a toolchain that also includes the runtime libraries. We normally also use the same build of Clang for the host as well as for the
*-fuchsia targets. See here for details on how we build Clang.
If you're using the prebuilt toolchains, you can skip this step, since the build will find them automatically.
Set the build argument that points to where you installed the toolchains:
fx set bringup.x64 --variant clang --args clang_tool_dir = "<absolute path to>/clang-install/bin/"
or for GCC:
fx set bringup.x64 --variant gcc --args gcc_tool_dir = "<absolute path to>/gcc-install/bin/"
*_tool_dir should have a trailing slash. If the
gcc in your
PATH works for Zircon, you can just use empty prefixes.
With local link IPv6 configured you can use
fx cp to copy files to and from the device.
The Zircon build creates a bootfs image containing necessary userspace components for the system to boot (the device manager, some device drivers, etc). The kernel is capable of including a second bootfs image which is provided by QEMU or the bootloader as a ramdisk image.
To create such a bootfs image, use the zbi tool that's generated as part of the build. It can assemble a bootfs image for either source directories (in which case every file in the specified directory and its subdirectories are included) or via a manifest file which specifies on a file-by-file basis which files to include.
$BUILDDIR/tools/zbi -o extra.bootfs @/path/to/directory echo "issue.txt=/etc/issue" > manifest echo "etc/hosts=/etc/hosts" >> manifest $BUILDDIR/tools/zbi -o extra.bootfs manifest
On the booted Zircon system, the files in the bootfs will appear under /boot, so in the above manifest example, the “hosts” file would appear at /boot/etc/hosts.
Network booting is supported via two mechanisms: Gigaboot and Zirconboot. Gigaboot is an EFI based bootloader whereas zirconboot is a mechanism that allows a minimal zircon system to serve as a bootloader for zircon.
On systems that boot via EFI (such as Acer and NUC), either option is viable. On other systems, zirconboot may be the only option for network booting.
The GigaBoot20x6 bootloader speaks a simple network boot protocol (over IPV6 UDP) which does not require any special host configuration or privileged access to use.
It does this by taking advantage of IPV6 Link Local Addressing and Multicast, allowing the device being booted to advertise its bootability and the host to find it and send a system image to it.
If you have a device (for example a Broadwell or Skylake Intel NUC) running GigaBoot20x6, first create a USB drive.
$BUILDDIR/tools/bootserver $BUILDDIR/zircon.bin # if you have an extra bootfs image (see above): $BUILDDIR/tools/bootserver $BUILDDIR/zircon.bin /path/to/extra.bootfs
By default bootserver will continue to run and every time it observes a netboot beacon it will send the kernel (and bootfs if provided) to that device. If you pass the -1 option, bootserver will exit after a successful boot instead.
Zirconboot is a mechanism that allows a zircon system to serve as the bootloader for zircon itself. Zirconboot speaks the same boot protocol as Gigaboot described above.
To use zirconboot, pass the
netsvc.netboot=true argument to zircon via the kernel command line. When zirconboot starts, it will attempt to fetch and boot into a zircon system from a bootserver running on the attached host.
The default build of Zircon includes a network log service that multicasts the system log over the link local IPv6 UDP. Please note that this is a quick hack and the protocol will certainly change at some point.
For now, if you're running Zircon on QEMU with the -N flag or running on hardware with a supported ethernet interface (ASIX USB Dongle or Intel Ethernet on NUC), the loglistener tool will observe logs broadcast over the local link:
For random tips on debugging in the zircon environment see debugging.