On Wed, Jan 30, 2002 at 12:08:05PM +0000, Richard Earnshaw wrote:

 > Given your need to recalculate the cacheable bits here, the above approach 
 > may no-longer be the best.  It might be the case that we should now make 
 > the cacheability of a page based on the current PTE writable attribute 
 > rather than the UVM attributes for the page.

Hm.  Then perhaps the best approach is to have a "pte_prot_proto" array,
one for cacheable, one for non-cacheable, that has the AP_* bits as well
as the PT_B and PT_C bits.

Actually, now that I think about it, this does make more sense in the
long run, since the use of some extended caching modes available on
XScale overlap with some of the AP_* bits, I believe... hm, I'll have
to look at the manual to double-check that (but after I get back from
the dentist .. eek).

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>