On Mon, 2008-10-20 at 13:58 +0200, Ingo Molnar wrote:

> yes but note that by caching the whole mapping on 64-bit we get 
> everything we want: trivially lockless, works from any CPU, can be 
> preempted at will, and there are no ugly INVLPG flushes anywhere.

I was assuming that on 64-bit, the map would be created at driver init
time and be left in place until the driver closed; if that's what you
mean by 'caching', then yes, we should cache the map.

> 32-bit we should handle as well but not design for it.

As long as we get kmap_atomic-like performance, and we get to simplify
our code, I'm up for it.

-- 
keith.packard@intel.com