This is a patch against 2.6.9 that is an update to an earlier patch
I posted on 10/13/2004 ("NUMA policy support for file mappings").

The patch allows NUMA policies for file mappings. Page cache pages are
allocated using a policy located in a shared_policy Red-Black tree attached
to the mapping object (address_space). This involves:

1. add a shared_policy tree to the address_space object in fs.h.

2. modify page_cache_alloc() in pagemap.h to take a page index in
   addition to a mapping object, and use those to locate the correct policy
   in the mapping->policy tree when allocating the page.

3. modify filemap.c to pass the additional page offset to 
page_cache_alloc().

4. Also in filemap.c, implement generic file {set|get}_policy() methods and
   add those to generic_file_vm_ops.

5. Init the file's shared policy in alloc_inode(), and free the shared 
policy in
  destroy_inode().

In addition, the patch adds a new flag to the mbind() syscall, MPOL_MF_MOVE.
If this flag is set, any existing mapped anonymous or filemap pagecache 
pages
that are/can be used for the given virtual memory region, that do not 
satisfy the
NUMA policy, are moved to a new page that satisfies the policy. Here's how
the new flag works with the existing MPOL_MF_STRICT flag (in the following
discussion, "invalid page" means a page that does not satisfy the NUMA 
policy):

MOVE and STRICT both set:
attempt to move invalid pages, and if any move fails, return mbind()
syscall failure.

MOVE set:
attempt to move invalid pages, but do not return error if any move fails.

STRICT set:
do not attempt to move invalid pages, returning mbind() failure (same
behavior as before).

neither MOVE or STRICT set:
ignore invalid pages.


Steve