This allows us to clear the area around the null pointer, and so trap errors.

The switch is tricky wrt setting up memory, since we need the memory to
be mapped in order to free it, but we need free memory to map (for the page
tables).  What we do is set up a temporart 1:1 map at ffff800000000000
that mirrors the first 1G map we already have from bootstreap, use that
to free all of memory under 1G, then map and free everything else.

Initial memory is physical; the mmu converts it to virtual addresses, and
then it can be added to the memory pool.  Right now there is not much
difference, but the 1:1 mapping is moving soon.

mmap() did not consider that a large page may be used to map an area that
it is overwriting.  This can happen if the primordial 1:1 area is remapped
(as when the payload is creating some guard pages).  The result is that
random memory is overwritten with a pte.

Fix by detecting large pages and splitting them up.

Reserves a vma range for later use.  Can be useful to implement mmap() hints
with a two-step algorithm - reserve a range, then overwrite it with the real
mmap.

This is simpler than [addr, addr+size) for splitting and merging.

When loading a small library (libverify.so) this happens.  Since it's legal,
allow it.

This fixes vm initialization failing after system_reset because vm_created
was left set to 1.

This allows generic code to include arch specific code, by eliding the
arch/x64/ prefix.

We will never build a single program containing code from two architectures,
so segregating arch-specific code into a separate namespace is not solving
any problem.  However it makes using arch-specific code from generic code
impossible.

Remove the 86 namespace to fix this.

much more efficient.

Should we warn instead?

These sections have memsz > filesz.  This means we need separate
map_file() and file_anon() with caller-specified addresses.