[PATCH v2 1/3] mm/slab: Fix drain freelist excessively

[Wanpeng Li]

The drain_freelist is called to drain slabs_free lists for cache reap,
cache shrink, memory hotplug callback etc. The tofree parameter is the
number of slab objects to free instead of the number of slabs to free.
The parameter transfered from callers is n->free_objects or n->freelimit
+ 5 * (searchp->num - 1) / (5 * searchp->num), and both of them mean
the number of slabs objects. I add printk to dump drain information:

[  122.864255] tofree is 2, actually free is 52, cache size is 26

The number of objects which caller prefer to drain is 2, however, actually
52 objects are drained, this destroy the cache locality. This patch fix it
by compare the number of slabs objects which already drained instead of
compare the number of slabs to the number of slab objects prefer to drain.
After this patch applied, there is no excessive drain as dump information 
mentioned above.

Signed-off-by: Wanpeng Li <liwanp@xxxxxxxxxxxxxxxxxx>
---
 mm/slab.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/slab.c b/mm/slab.c
index 8ccd296..f58eb0b 100644
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -2479,7 +2479,7 @@ static void drain_cpu_caches(struct kmem_cache *cachep)
 
 /*
  * Remove slabs from the list of free slabs.
- * Specify the number of slabs to drain in tofree.
+ * Specify the number of slab objects to drain in tofree.
  *
  * Returns the actual number of slabs released.
  */
@@ -2491,7 +2491,7 @@ static int drain_freelist(struct kmem_cache *cache,
 	struct slab *slabp;
 
 	nr_freed = 0;
-	while (nr_freed < tofree && !list_empty(&n->slabs_free)) {
+	while (nr_freed * cache->num < tofree && !list_empty(&n->slabs_free)) {
 
 		spin_lock_irq(&n->list_lock);
 		p = n->slabs_free.prev;
-- 
1.8.1.2

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/