On 07/23/2013 04:21 AM, Rik van Riel wrote:
On 07/19/2013 04:55 PM, Johannes Weiner wrote:@@ -1984,7 +1992,8 @@ this_zone_full: goto zonelist_scan; } - if (page) + if (page) { + atomic_sub(1U << order, &zone->alloc_batch); /* * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was * necessary to allocate the page. The expectation isCould this be moved into the slow path in buffered_rmqueue and rmqueue_bulk, or would the effect of ignoring the pcp buffers be too detrimental to keeping the balance between zones? It would be kind of nice to not have this atomic operation on every page allocation...
atomic operation will lock cache line or memory bus? And cmpxchg will lock cache line or memory bus? ;-)
As a side benefit, higher-order buffered_rmqueue and rmqueue_bulk both happen under the zone->lock, so moving this accounting down to that layer might allow you to get rid of the atomics alltogether. I like the overall approach though. This is something Linux has needed for a long time, and could be extremely useful to automatic NUMA balancing as well...
-- 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/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>