> > > attached patch shrinks pv_entry by eliminating pv_pmap member.
> > > while this increases cost to get pmap pointer a little,
> > > i think size of pv_entry is more important.
> > >
> > > any comments?
> > 
> > Have you benchmarked operation with and without pv_pmap?
> 
> not yet.

do you have any idea about how to benchmark?

YAMAMOTO Takashi