On Thu, Mar 19, 2015 at 06:29:47PM -0700, Linus Torvalds wrote:

And the VM_WRITE test should be stable and not have any subtle
interaction with the other changes that the numa pte things
introduced. It would be good to see if the profiles then pop something
*else* up as the performance difference (which I'm sure will remain,
since the 7m50s was so far off).

As a side-note, I did test a patch that checked pte_write and preserved
it across both faults and setting the protections. It did not alter
migration activity much but there was a  drop in minor faults - 20% drop in
autonumabench, 58% drop in xfsrepair workload. I'm assuming this is due to
refaults to mark pages writable.  The patch looks and is hacky so I won't
post it to save people bleaching their eyes. I'll spend some time soon
(hopefully today) at a smooth way of falling through to WP checks after
trapping a NUMA fault.

-- 
Mel Gorman
SUSE Labs