This version is actually known bad, but it should definitely include the
magic string that the test code is looking for to indicate what port gdb
is listening on.

Need to see the output to know why things are failing.

Need to test the build with a version that's actually sane.

The new version should not eat errors in the Makefile, which means the
build of this PR should actually fail because pexpect is still missing
from the travis file.

List libusb-1.0 Ubuntu dependency

Let the build fail when debug-check fails.

Bump PK, to fix a boot race condition related to filtering device trees

Bump riscv-tests to pull in mi-csr test fix.

bump riscv-gnu-toolchain

fesvr: Bump to pick up multi-core DTM fixes

fesvr: bump to pick up fixes and not dying on error

Disable -Werror for the rv32ima script as well

Add more dependencies for OS X

fesvr: Bump to pick up change to make all harts start at _start

Add inittab option to busybox

Without this option, busybox will try to find /etc/rcS instead of inittab, and the initramfs of this README won't work.

Bump gnu-toolchain and openocd, to do a release

Note, this might not be a complete or even minimal list, but
it's a start and might take out a tiny bit of frustration.

Disable -Werror when building OpenOCD

It appears that newer GCC versions don't like some of our macros.  Since
OpenOCD defaults to --enable-werror, it'll blow up on those platforms.