On Thu, 15 Dec 2011 11:04:43 +0100 Johannes Weiner <hannes@xxxxxxxxxxx> wrote: > On Thu, Dec 15, 2011 at 03:05:22PM +0900, KAMEZAWA Hiroyuki wrote: > > >From 528f5f2667da17c26e40d271b24691412e1cbe81 Mon Sep 17 00:00:00 2001 > > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> > > Date: Thu, 15 Dec 2011 11:41:18 +0900 > > Subject: [PATCH 1/5] memcg: simplify account moving check > > > > Now, percpu variable MEM_CGROUP_ON_MOVE is used for indicating that > > a memcg is under move_account() and pc->mem_cgroup under it may be > > overwritten. > > > > But this value is almost read only and not worth to be percpu. > > Using atomic_t instread. > > I like this, but I think you can go one further. The only place I see > where the per-cpu counter is actually read is to avoid taking the > lock, but if you make that counter an atomic anyway - why bother? > > Couldn't you remove the counter completely and just take move_lock > unconditionally in the page stat updating? > Hmm, I (and Greg Thelen) warned that 'please _never_ add atomic ops to this path' by Peter Zilstra. That 'moving_account' condition checking was for avoiding atomic ops in this path (for most of cases.) We'll need to gather enough performance data after implementing per-memcg dirty ratio accounting. So, could you wait for a while ? Anyway, my patch '[PATCH 5/5] memcg: remove PCG_MOVE_LOCK' may not work enough well without the moving_account check. I'll consider more. Thank you for review. -Kame -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. 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>