On Thu, May 11, 2000 at 12:13:20PM +0200, Ignatios Souvatzis wrote:

 > Are the pages to be zeroed by the idle loop zeroing mechanism guaranteed to
 > be otherwise unmapped?
 > 
 > At least some revisions of 68060 and 68040 require that no page is mapped
 > at the same time cachable and non-cachable, as a workaround for some CPU
 > bug.

Yes.  The pages are taken from the free list, passed to the pmap's zeroing
hook, and then placed back on the free list's "zero'd queue".  There are
supposed to be no valid mappings for such pages.

I suppose there is one problem -- device pager mappings for /dev/mem.  In
that case, you'd want to traverse the PV table and make such mappings
uncached when you do the uncached mapping for the pmap zeroing hook.

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