> Not from within the pmap code, the pmap asks uvm for a 16k, 16k-aligned 
> chunk of memory, it's the uvm system that asks the paging system to 
> reclaim the memory.  The paging system wakes up; decides that enough pages 
> are already available to fulfil the request and goes back to sleep again.

Looks like the uvm <=> paging system interface is broken!
A 'physcontig' and 'phyaligned' flag might be useful.
Maybe not too hard to implement?

I haven't looked at how this code is synchronised between the
requester and pageout code.  But if the paging system thinks there
are enough pages doesn't it wake the sleeper?
In which case I'd expect a rather violent loop...

	David

-- 
David Laight: david@l8s.co.uk