Chris Wright wrote:
> * Andy Whitcroft (apw@shadowen.org) wrote:
> 
>>I think a concensus is forming that the checks for merging across
>>zones were removed from the buddy allocator without anyone noticing.
>>So I propose that the configuration option UNALIGNED_ZONE_BOUNDARIES
>>default to on, and those architectures which have been auditied
>>for alignment may turn it off.
> 
> 
> So what's the final outcome here for -stable?  The only
> relevant patch upstream appears to be Bob Picco's patch
> <http://kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=e984bb43f7450312ba66fe0e67a99efa6be3b246>

I am not sure we necessarily need to make any changes for stable.  The
lack of alignment checks has been in the mainline tree for a number of
months.  I believe that i386 in the simple cases should be aligned
correctly and that covers the majority of users.

If we are going to make any changes then I'd say we want two patches.
The node_mem_map alignment patch from Bob Picco (as cited above) and the
attached patch.  This is a simplification of the patches currently in
-mm, it should be functionally equivalent to the changes in -mm without
the exclusions and configuration options.

I've just run a regression suite over this one on the machines I have
here without any problems.

Comments?

-apw