Use musl-libc instead of system libc

[osresearch]

Building with alpine linux didn't work well (issue #54), but it appears that it is easier to do a hybrid build of linux and coreboot with the regular gcc, and the initrd tools with the musl-gcc wrapper.

[rofl0r]

you might as well try sabotage linux https://github.com/sabotage-linux/sabotage .
our current initrd is 1.3 MB with full busybox, cryptsetup and lvm (everything static linked).

[Kokokokoka]

And what about buildroot? It is used by jeltka, coreboot payload, one can choose between uclibc-ng, musl,
there are external projects like https://github.com/xSmurf/coreboot_buildroot_project I'm trying to add zfs
package here as zfs will soon have native encryption so I'll try to use it with TPM and my payload will
be busybox+zfs-tools+xen, then I'll install Qubes. Now I'm using buildroot with patches from this
repo, and it works fine.

[osresearch]

Buildroot looks promising, especially since it includes (experimental) support for reproducible builds.One concern is that it doesn't seem to fit -- xSmurf says "Installing on the X220 requires hardware modification as the flash has to be upgraded to 16MiB." With the Heads bundler the entire runtime fits into the 4MB of the top SPI flash (and my experimental musl-libc branch has 400 KB free).

[Kokokokoka]

Buildroot looks promising, especially since it includes (experimental) support for reproducible builds.One concern is that it doesn't seem to fit -- xSmurf says "Installing on the X220 requires hardware modification as the flash has to be upgraded to 16MiB." With the Heads bundler the entire runtime fits into the 4MB of the top SPI flash (and my experimental musl-libc branch has 400 KB free).

This assumption is wrong, I'm using buildroot with uclibc-ng, and I've got about 1.6 mb payload (don't remember for sure), with wpa-supplicant and dropbear and me 1.5mb. (I'll tell what is the real size of the payload later and I'll remove wpa-supplicant as I took it from jeltka and it's not really needed.)
I'll share my config if it's needed, I don't remember for sure, but there was plenty of the free space; with
bloated config there was about 200 kb free.
Also there is reproducible build option. And a rather good strip option.
So, I've removed the wpa-supplicant and dropbear and I've switched again to musl libc, now my payload is
1.3 mb also I've got libnl compiled and lspci, the only thing that I had troubles with was the nvramtool:
accessors/cmos-hw-unix.o: In function cmos_hal_write':
cmos-hw-unix.c:(.text+0x51): undefined reference to outb' accessors/cmos-hw-unix.o: In function cmos_hal_read':
cmos-hw-unix.c:(.text+0xb6): undefined reference to outb' cmos-hw-unix.c:(.text+0xc0): undefined reference to inb'
accessors/cmos-hw-unix.o: In function cmos_hal_write': cmos-hw-unix.c:(.text+0x63): undefined reference to outb'
`

[osresearch]

Thanks for the review on the musl-libc branch merge. sabotage looks intriguing as a future platform; I had run across it earlier as a way to build busybox with musl-libc and we're using its kernel headers package to host the gcc wrapper.

The changes are now in master and appear to build from a clean checkout (on ubuntu and fedora). This doesn't satisfy the reproducible build goal yet (issue #2), but it does close out dependency on the system libc (issue #7).