- Page faults are handled in an (obviously) architecture-dependent way...
- ...So, the entry point for page faults is somewhere in arch/*...
- See, for example, arch/x86/mm/fault.c::do_page_fault()
  + After some architecture-dependent stuff, do_page_fault() ends up in
    calling do_kern_addr_fault() or do_user_addr_fault()
    + do_user_addr_fault() -> handle_mm_fault() (architecture-independent code)
      + mm/memory.c::handle_mm_fault() -> handle_pte_fault()
      + For anonymous pages, it invokes do_anonymous_page() -> alloc_pages() ->
        -> alloc_zeroed_user_highpage_movable() -> alloc_page_vma() ->
        -> alloc_pages_vma() with order 0 -> alloc_pages() with order 0
           (for no NUMA)