On Tue, 10 May 2011 01:02:55 +0200 Andi Kleen <andi@xxxxxxxxxxxxxx> wrote: > On Mon, May 09, 2011 at 03:28:41PM -0700, Andrew Morton wrote: > > On Mon, 09 May 2011 15:23:03 -0700 > > Andi Kleen <ak@xxxxxxxxxxxxxxx> wrote: > > > > > > After fixing that and doing an allnoconfig x86_64 build, the patchset > > > > takes rmap.o's .text from 6167 bytes to 6551. This is likely to be a > > > > regression for uniprocessor machines. What can we do about this? > > > > > > > > > > Regression in what way? > > > > It makes the code larger and probably slower, for no gain? > > It should be actually faster because there are much less atomic ops. > Atomic ops are quite expensive -- especially on some older CPUs, even when > not contended. hm, which atomic ops are those? We shouldn't need buslocked operations on UP. > > > > > I guess I can move some of the functions out of > > > line. > > > > I don't know how much that will help. Perhaps a wholesale refactoring > > and making it all SMP-only will be justified. > > Yes I don't think there were a lot of callers. > > I can take out the lockbreak. I was a bit dubious on its utility > anyways. I guess that running something like latencytop with a suitably nasty workload would permit detection of any problems. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>