Did some more debugging on the real hardware initrd loading problem I was having earlier; Stefan walked me through setting up the GDB stub (had to build a CVS version of GDB to get working breakpoints, and for some reason remotebaud defaults to the ridiculous 4294967295 rather than something sensible, like the 115200 that gPXE uses). Didn't figure anything new out with this, although it looks like the GDB stub could be very useful in the future. After that, tried building with DEBUG=bzimage, and uncovered some interesting differences in the length of the initrd segment and other bzimage debug output; sent a mail to etherboot-developers with details.