- Improve drain pages performance on large systems

Heavy swapping within a cpuset causes frequent calls to drain_all_pages()

Quoted 25 lines

Index: linux/mm/page_alloc.c
===================================================================

mm/page_alloc.c A A A 2011-02-15 16:29:43.085502487 -0600 renamed from mm/page_alloc.c A 2011-02-15 16:28:36.165921713 -0600 +13 -0

@@ -592,10 +592,24 @@ static void free_pcppages_bulk(struct zo
 A  A  A  A int batch_free = 0;
 A  A  A  A int to_free = count;
 
+ A  A  A  /*
+ A  A  A  A * Quick scan of zones. If all are empty, there is nothing to do.
+ A  A  A  A */
+ A  A  A  for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) {
+ A  A  A  A  A  A  A  struct list_head *list;
+
+ A  A  A  A  A  A  A  list = &pcp->lists[migratetype];
+ A  A  A  A  A  A  A  if (!list_empty(list))
+ A  A  A  A  A  A  A  A  A  A  A  break;
+ A  A  A  }
+ A  A  A  if (migratetype == MIGRATE_PCPTYPES)
+ A  A  A  A  A  A  A  return;
+
 A  A  A  A spin_lock(&zone->lock);
 A  A  A  A zone->all_unreclaimable = 0;
 A  A  A  A zone->pages_scanned = 0;
 

mm/page_alloc.c A A A 2011-02-15 16:29:43.085502487 -0600 static void free_pcppages_bulk(struct zo · line 615

Quoted 4 lines

mm/page_alloc.c A A A 2011-02-15 16:29:43.085502487 -0600 renamed from mm/page_alloc.c A 2011-02-15 16:28:36.165921713 -0600 +1 -0

+ A  A  A  migratetype = 0;
 A  A  A  A while (to_free) {
 A  A  A  A  A  A  A  A struct page *page;
 A  A  A  A  A  A  A  A struct list_head *list;

It does make sense to me.
Although new code looks to be rather costly in small box, anyway we use the same logic in while loop so cache would be hot. so cost would be little.

But how about this? This one never affect fast-critical path.

mm/page_alloc.c +0 -0

index ff7e158..2dfb61a 100644
@@ -1095,8 +1095,10 @@ static void drain_pages(unsigned int cpu)
                pset = per_cpu_ptr(zone->pageset, cpu);
 

Quoted 52 lines · Jack Steiner

Heavy swapping within a cpuset causes frequent calls to drain_all_pages(). This sends IPIs to all cpus to free PCP pages. In most cases, there are no pages to be freed on cpus outside of the swapping cpuset.

Add checks to minimize locking and updates to potentially hot cachelines. Before acquiring locks, do a quick check to see if any pages are in the PCP queues. Exit if none.

On a 128 node SGI UV system, this reduced the IPI overhead to cpus outside of the swapping cpuset by 38% and improved time to run a pass of the swaping test from 98 sec to 51 sec. These times are obviously test & configuration dependent but the improvements are significant.

Signed-off-by Jack Steiner <[email protected]>

---
 mm/page_alloc.c |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

Index: linux/mm/page_alloc.c
===================================================================

mm/page_alloc.c 2011-02-15 16:29:43.085502487 -0600 renamed from mm/page_alloc.c 2011-02-15 16:28:36.165921713 -0600 +14 -1

@@ -592,10 +592,24 @@ static void free_pcppages_bulk(struct zo
       int batch_free = 0;
       int to_free = count;
 
+       /*
+        * Quick scan of zones. If all are empty, there is nothing to do.
+        */
+       for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) {
+               struct list_head *list;
+
+               list = &pcp->lists[migratetype];
+               if (!list_empty(list))
+                       break;
+       }
+       if (migratetype == MIGRATE_PCPTYPES)
+               return;
+
       spin_lock(&zone->lock);
       zone->all_unreclaimable = 0;
       zone->pages_scanned = 0;
 
+       migratetype = 0;
       while (to_free) {
               struct page *page;
               struct list_head *list;
 
--

Quoted 6 lines

To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected]. 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:"[email protected]"> [email protected] </a>

It does make sense to me.
Although new code looks to be rather costly in small box, anyway we use the same logic in while loop so cache would be hot. so cost would be little.

But how about this? This one never affect fast-critical path.

mm/page_alloc.c +4 -2

index ff7e158..2dfb61a 100644
@@ -1095,8 +1095,10 @@ static void drain_pages(unsigned int cpu)
                pset = per_cpu_ptr(zone->pageset, cpu);
 
                pcp = &pset->pcp;
-               free_pcppages_bulk(zone, pcp->count, pcp);
-               pcp->count = 0;
+               if (pcp->count > 0) {
+                       free_pcppages_bulk(zone, pcp->count, pcp);
+                       pcp->count = 0;
+               }
                local_irq_restore(flags);
        }
 }

Signature · 3 lines

-- 
Kind regards,
Minchan Kim

Heavy swapping within a cpuset causes frequent calls to drain_all_pages().