On Wed, Sep 04, 2002 at 10:28:55PM +0200, Manuel Bouyer wrote:

 > Hum, and if we have a variable, per-device MAXPHYS, the per-device minphys
 > can probably die, isn't it ?
 > This would avoid several function pointer dereference in the I/O path,
 > to only end up calling the global minphys() for most of the drivers.

Actually... you could eliminate minphys() completely, if you could
guarantee that:

	1. A per-device limit was never larger than the global limit.

	2. Nothing ever generated anything larger than the per-device
	   limit.

If you make these two things invariants, then you can eliminate minphys()
completely.

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