amd64: replace tailq pointers with iterators

[dougm]

Summary

Change architecture-specific code to use iterators rather than tailq pointers.

[markj]

sys/amd64/amd64/pmap.c

7711

Why can't we just assign m = m_start?

7942

The vm_object_populate() call above guarantees that the object has resident pages for pindices [pindex, pindex+size). I don't think we can have p == NULL here.

7944

Why is it ok to drop the physical address check?

[dougm]

sys/amd64/amd64/pmap.c

7711

That won't set pages.index to m_start->pindex, so iter_step() and iter_jump() will be adding to an incorrect or uninitialized value to update the index value.

7944

Because line 7940 is
p = vm_radix_iter_next(&pages);
and not
p = vm_radix_iter_step(&pages);

The latter would be equivalent to "p = TAILQ_NEXT(p, listq). The former gets the page with the next address, and gets NULL if there is no such page.

[markj]

sys/amd64/amd64/pmap.c

7944

Contiguity in the pindex space doesn't imply contiguity in the physical address space. The vm_object_populate() call ensures that a page is resident at all pindices in the range, so the old code could have equivalently used vm_page_next().

[dougm]

Drop null check. Restore physical address check.