On Mon, Jan 07, 2008 at 09:53:34PM +0100, Martin Husemann wrote:
On Mon, Jan 07, 2008 at 09:15:55PM +0100, Tobias Nygren wrote:
The problem is that the pmap refcount is set to zero in
pmap_remove_all(). No other ports I checked do this.
Nice catch.
pmap_remove_all() is an optimization to avoid a lot of work updating
mapping only to throw them all away later. So when we know the pmap
is going away (process exiting), pmap_remove_all() invalidated the
context, and any further pmap_remove()/pmap_page_protect()/pmap_update()
calls will avoid heavy work.
A working solution is to cast the return value from atomic_dec_uint_nv
in pmap_destroy into a signed value. This restores the previous
(broken) behaviour. Maybe you guys have a better idea? :-)
Since pm_refs is an int I think this is the correct solution and the
previous behaviour is not broken at all. Andrew, what do you think?
pmap_remove_all() will only be called by the last LWP in a process
so the refcount trick is fine, casting the return to (int) seems the
best thing to do. I will make the change shortly.