[0/4] RFC: SLB rewrite

[David Gibson]

Here, at long last, the much awaited, SLB rewrite!

I have this in four patches at the moment:

1/4: slbrewrite2     - The guts of the rewrite, miss path in asm
2/4: noprolog        - Unify do_slb_bolted with full SLB miss path
3/4: customentryexit - Optimise SLB exception entry/exit path
4/4: newvsid         - Replace VSID algorithm with a better one

Overall these seem to improve the (user address) SLB miss time by
around 25% (200ns to 150ns on a G5).  Test program and resulting
graphs are at http://www.ozlabs.org/people/dgibson/slbtest/ the most
interesting graph is probably:
	http://www.ozlabs.org/people/dgibson/slbtest/stagger.png

These patches have been tested (though not extensively) on an Apple G5
(PowerPC 970), a p630 LPAR (Power4+), an RS/6000 270 (POWER3) and an
RS64 iSeries partition (the latter two obviously aren't expected to
show any improvement, the point is just to check I haven't broken
things dramatically for STAB machines).  Anton has also tested some
earlier versions of the rewrite on a Power5 (bare metal Linux).

If there are no serious problems, I'm thinking of pushing at least the
first three of these to Linus/akpm in the next week or so.

--
David Gibson			| For every complex problem there is a
david AT gibson.dropbear.id.au	| solution which is simple, neat and
				| wrong.
http://www.ozlabs.org/people/dgibson

** Sent via the linuxppc64-dev mail list. See http://lists.linuxppc.org/