On Wed, 9 Jun 2010 09:54:48 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote: > On Tue, 8 Jun 2010 16:31:29 +0900 > Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote: > > > On Tue, 8 Jun 2010 12:19:01 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote: (snip) > > > 1. Because css_get/put calls atoimic_inc()/dec, heavy call of them > > > on large smp will not scale well. > > I'm sorry if I'm asking a stupid question, the number of css_get/put > > would be: > > > > before: > > get:1 in charge > > put:1 in uncharge > > after: > > get:1, put:1 in charge > > no get/put in uncharge > > > > right ? > > No. > > before: get 1 in charge. > put 1 at charge > > after: > no get at charge in fast path (cunsume_stcok hits.) > get 1 at accssing res_counter and reclaim, put 1 after it. > no get/put in uncharge. > > > Then, isn't there any change as a whole ? > > > We get much benefit when consume_stock() works. > Ah, I missed comsume_stock(). The number of get/put would be decreased very much. Thank you for your explanation. Thanks, Daisuke Nishimura. -- 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/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>