Alexey:
1. Significantly rework cache-related functionality.
In particular that fixes coherency problems in some corner-cases,
allows us to enable and disable caches in run-time and still
have properly running system, finally support execution from
real flash (before we used to run from DDR from the very beginning).
2. Remove string routines implemented in assembly.
That allows us to build and run U-Boot on wide range of ARC cores
with different configurations. I.e. whatever tuning is used on GCC's
command-line we'll get code for desired flavor of ARC.
Otherwise for each and every corner-case we would need to add ifdefs
in assembly code to accommodate missing instructions etc.
3. Get use of GCC's garbage collector which helps to slim-down resulting image
quite a bit.
4. Also now we may disable U-Boot self-relocation for ARC if needed either
by platform or for debugging purposes.