(2012/03/31 23:09), Andi Kleen wrote: > From: Andi Kleen <ak@xxxxxxxxxxxxxxx> > > We found that the mm struct anon page counter cache line is much hotter > with transparent huge pages compared to small pages. > > Small pages use a special fast counter mechanism in task_struct, but huge pages > didn't. The huge pages are larger than the normal 64 entry threshold for the > fast counter, so it cannot be directly used. Use a new special counter for huge > pages to handle them efficiently. > > Any users just calculate the correct total. > > The only special case is transferring the large page count to small pages > when splitting. I put it somewhat arbitarily into the tricky split > sequence. Some review on this part is appreciated. > > [An alternative would be to not do that, but that could lead to > negative counters. These should still give the correct result] > > Contains a fix for a problem found by Andrea in review. > > Cc: aarcange@xxxxxxxxxx > Signed-off-by: Andi Kleen <ak@xxxxxxxxxxxxxxx> Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> -- 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>