On Thu, Aug 04, 2011 at 02:07:22PM -0700, Michel Lespinasse wrote: > This change introduces a new get_page_unless_zero() function, to be > used for idle page tracking in a a future patch series. It also > illustrates why I care about introducing the page count lock discussed > in the previous commit. > > To explain the context: for idle page tracking, I am scanning pages > at a known rate based on their physical address. I want to find out > if pages have been referenced since the last scan using page_referenced(), > but before that I must acquire a reference on the page and to basic > checks about the page type. Before THP, it was safe to acquire references > using get_page_unless_zero(), but this won't work with in THP enabled kernel > due to the possible race with __split_huge_page_refcount(). Thus, the new > proposed get_first_page_unless_zero() function: > > - must act like get_page_unless_zero() if the page is not a tail page; > - returns 0 for tail pages. > > Without the page count lock I'm proposing, other approaches don't work > as well to provide mutual exclusion with __split_huge_page_refcount(): > > - using the zone LRU lock would work, but has a low granularity and > exhibits contention under some of our workloads I thougt this but it seems your concern is LRU lock contention. This patch doesn't include any use case(Sometime it hurts reviewers) but I expect it's in idle tracking patch set. But we don't conclude yet idle page tracking patchset is reasonable or not to merge mainline. So, I think it's rather rash idea. (But I admit [1,2/3] is enough to discuss regardless of idle page tracking) What I suggestion is as follows, 1. Replace naked page->_count accesses with accessor functions 2. page count lock 3. idle page tracking with simple lock(ex, zone->lru_lock) 4. get_first_page_unless_zero to optimize lock overhead. -- Kind regards, Minchan Kim -- 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>