It should be also considered to move most copy_*_user()/get/put_user() calls
out-of-line to save code bloat: there is evidence(1) that most of the 2.1
binary image bloat (my 2.1 kernel is over 100k bigger than the equivalent 2.0
one) comes from these functions.
Maybe on the Intel, but on sparc64 at least the cost of get/put_user
is 2 instructions always (and on sparc thats the cost of calling some
routine so...). We already do the copy/clear user stuff out of line
since we use the thick VIS code to make it go like smoke.
Later,
David S. Miller
davem@dm.cobaltmicro.com