Can you test on misaligned buffers?

Looks good! Sorry for the delay in reviewing.

This looks good. Thanks for doing the work, and sorry for the delay in reviewing!

Hm, this is close, but I think it might read nicer if the *_vsx.S contained the VSX implementations of "phase 2", while the plain files contained the fallback, something like:

I just ran a test building llvm with ninja (3701 jobs):

Fix the CPU domain binding. Only cpu0's domain was being bound, for all CPUs. Apply NUMA to HPT.

Will take this to completion.

Neither em(4) nor iavf(4) are built as modules for powerpc, so cannot be used as examples. I don't have a good answer for you on this, maybe installkernel needs to be aware of LINKS and instead of creating links create either symlinks or copies? Are symlinks supported on msdosfs?

Correct. It didn't logically belong with D19352, and probably other chunks of D19352 can be broken out, too, into more consumable pieces. The booke part I added for completness/consistency sake.

I'm on the fence between interrupt.c and trap.c. I put it in interrupt.c for the following reasons:

Patch created by Brandon Bergren, who also pointed out the problem.

Add pmap_private.h to the diff, fix more style issues in mmu_radix.c.

Fix HPT. ZFS has been confirmed working, when loaded as a module.

Hi @kib, thanks for your review. There's still a bit of work to do, but this definitely helps.

Still looking into it, it's probably something in vmparam.h that needs to be switchable. The problem seen is init crashing with a bus error. I haven't debugged that further yet.

Add a dummy freezer, to keep the code path easier to follow.

I have two alternative paths to go: if (freeze... != NULL), or have a default empty freeze function. Doesn't matter to me which way we go with, but, yes, we need a fallback to do it the lame way.

@jeff this wasn't just on a 32-bit system; I experienced the same problem on a 64-bit PowerPC Book-E system, which currently does not support UMA_MD_SMALL_ALLOC (I tried turning it on with @markj's suggestion, and it failed quite spectacularly, so more work is needed on that front). So this helps more than just 32-bit platforms.

This looks like a correctness fix, too. I don't think we should be calling that on an error anyway.

Looks fine to me now.