To start using musl, you have three choices.
Using the musl-gcc wrapper
This allows you to test and use musl on a glibc/uclibc system with no great effort. You cannot, however, use C++ with it.
When building musl, there are 3 important flags to pass to configure:
- Where to install musl. A goood choice is something like $HOME/musl .
- This is where the musl-gcc wrapper gets installed to. It should point to somewhere in your PATH. A good choice is something like $HOME/bin .
- This is where the dynamic linker gets installed into. The default is /lib; this path will be baked into binaries built by musl-gcc, so you should not alter it if you want your dynamic binaries to be portable. Installing to /lib on most systems requires root privileges, so a typical choice for non-root users would be $HOME/musl/lib .
you can omit shared library support (static linking only) and cut musl's build time in half using --disable-shared.
The configure run will generate a config.mak file, which contains your settings.
Now run make && make install
Now you can use musl-gcc instead of gcc to compile things against musl. Use -static to build static binaries. For example to compile a software package that uses autoconf statically against musl:
CC="musl-gcc -static" ./configure --prefix=$HOME/musl && make
Building a cross compiler targeting musl libc
Use the prebuilt crosscompilers at http://bitbucket.org/GregorR/musl-cross/downloads or build it yourself using the supplied scripts (git mirror). This gives you a full musl toolchain including C++ support. Note that although the scripts are capable of building compilers targeting other architectures, a cross compiler targeting your host architecture but musl libc is also useful, as the whole toolchain is aware of its target libc.
Notes on ARM Float Mode
There are three float mode available on modern ARM SoC's:
- soft -- floating point is completly emulated, very slow
- softfp -- uses hardware floating point, but is ABI compatible with soft
- hard -- complete hardware floating point, incompatible ABI with soft and softfp
On modern armv6 and armv7 chips, hardware floating point is usually implemented on chip. If someone is planing to compile musl-cross with hardware floating point, add the following to your config.sh:
GCC_BOOTSTRAP_CONFFLAGS="--with-arch=armv7-a --with-float=hard --with-fpu=vfpv3-d16"
GCC_CONFFLAGS="--with-arch=armv7-a --with-float=hard --with-fpu=vfpv3-d16"
This should produce a cross-toolchain that is compatible at least with: Marvell Dove, Freescale i.MX5x, TI OMAP3+4, Qualcomm Snapdragon, nVidia Tegra2+3 and probably all other modern Cortex-A8, Cortex-A9 and Cortex-A15 SoC's on the market. If you plan to compile for older armv6 SoC's, like the one found on the RasperryPi, use --with-arch=armv6-a --with-float=hard --with-fpu=vfpv2. VFPv3 contains the VFPv2 subset, so you can also use the armv6 binaries on a more modern armv7 system, but losing some performance. On contrary, for Cortex-A15 SoC's like the new Samsung Exynos 5, you can activate the even more powerfull VFPv4.
Using a distro targeting musl
If your distro uses musl natively, then naturally, anything compiled on that distro will use musl. Several distros using musl, such as sabotage, are listed on the distributions page of this wiki.