Loader: Add load offset to powerpc kernel entry point

[jhibbits]

Summary

There is logic in ELF loadimage() to relocate kernels, but currently
only type ET_EXEC. PowerPC kernels are ET_DYN, and can be relocated anywhere.
Add the load offset to kernel entry points on this platform.

[imp]

stand/common/load_elf.c

488–489

and why not adjust it here?

523

This should move up to where I tagged it above (though maybe not in that if) since we already have a powerpc ifdef there.

[jhibbits]

stand/common/load_elf.c

488–489

Look above and you see this block is for type ET_EXEC. PowerPC kernels are ET_DYN now. This block looks really weird, and I would love to remove it. Not sure if we can, though, because I don't know if any "supported" older release kernels are being booted with a modern loader. Otherwise, yes, I could muck with the condition above and muck with this as well. Or completely remove this block for PowerPC, and do the work below where I added, only.

[• ian]

stand/common/load_elf.c

523

Okay, that's two of us who thought this needed to move up. We were wrong, but clearly the current code structure is going to fool people.

Maybe it would be better to move this block up so that the overall structure is:

  if (ehdr->e_type == ET_DYN) {
#ifdef powerpc
    // new stuff
#endif
  } else if (ehdr->e_type == ET_EXEC) {
    // existing stuff
  }

[jhibbits]

Address feedback. I think this is better overall, simplifies things.

[jhibbits]

stand/common/load_elf.c

480

I don't think this block should really be here, but I'm not sure enough to remove it. Since the current idea is that we typically load in the first 256MB, I don't see a problem enough to fix it.

[imp]

stand/common/load_elf.c

523

or just make it an else if here for the new stuff.

but why then is this powerpc specific if other platforms aren't currently doing it? Will there be harm to them?

[• ian]

Yep, this is much cleaner than the old code.