> Now that I am planning on doing is putting the PT_NC attribute into
> pg->mdpage.pvh_attrs, rather than having to traverse the pv_entry list
> to determine the caching status for a page (IIRC, it must always be
> true that if one mapping of a page is NC, *all* mappings of that page
> must be NC).
> 

No, this isn't true.  Just because one process has to make a page NC doesn't necessarily mean that all processes must do so.

See the table I added describing the conditions (before pmap_vac_me_harder)

R.