On Wed, Jan 03, 2001 at 08:29:02PM -0800, Jason R Thorpe wrote:
> On Wed, Jan 03, 2001 at 10:07:58PM -0500, Chuck Cranor wrote:
> I can possibly see an argument for PV entries... but not for PT pages,
> since kernel PT pages are always pre-allocated.
 
i had thought that the PV allocator used PTs as a backup memory
allocation mechanism, but either that code got removed or i forgot
to put it in?


> ...but there's still a problem -- once you've run out of mappings to
> steal, you STILL panic... so, you can "recover" for a while, but not
> forever.  I guess for those cases, you want the caller to decide to
> return NULL or panic or whatever.

my thinking at the time was that you had to have some valid mappings
in order to make forward progress... so there should be a victim out
there.   one thing that could make it a bit iffy is handling of resources
associated with wired mappings.


chuck