First cut at NetBSD/powerpc support. Copied from Linux mostly as-is either directly or by applying Linux 386->power changes to NetBSD 386.
The only semantic change w.r.t. Linux code is to lib9/setfcr-NetBSD-power.S - don't use with-update variant to store FPSCR as the code never reverts the update and hence the stack pointer becomes out of whack. This looks like a genuine bug in the Linux code.
Only minimally tested:
emu -c does not work
Compiling with -O2 in CFLAGS causes host's limbo to dump core when generating runt.h
Graphics work, but the colors are wrong. Looks like wrong RGB endianness.
The crash with -O2 is taken care of with bc94d01 that add explicit -fno-strict-aliasing to be on the safer side of things. This code was written well before modern compilers became opinionated, so making that explicit doesn't hurt (even if default -O doesn't enable it).