On Fri, Feb 09, 2001 at 10:50:11PM -0500, Bill Sommerfeld wrote:

 > > The problem is still multiple calls to pmap_remove however
 > > - one for each mapped segment in the pmap AFAIR. 
 > 
 > To help batch up TLB shootdown traffic Jason and I have discussed
 > changing the pmap API to allow it to defer/batch up shootdowns/cache
 > invalidations/etc. until a subsequent "sync"-like call.  That sort of
 > change would clearly help here as well...

Yah, in Mach, this is what pmap_update() was for -- unfortunately,
it seemed to have gotten ... well, ignored when Mach VM was ported
to Berkeley UNIX.

I can't help but wonder that happened is that Mike noticed bugs, and
just littered the hp300 pmap with TLB invalidations, rather than
finding the missing pmap_update() calls in the VM system.  As it is
now, there's only one call to pmap_update(), and it's at the end of
uvm_swapout_threads() ... and, as far as I know, it's never taken
an argument (assumed all the defered actions must have global scope).

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