memcgroup: move memory controller allocations to their own slabs
Balbir Singh [Tue, 29 Apr 2008 08:00:19 +0000 (01:00 -0700)]
Move the memory controller data structure page_cgroup to its own slab cache.
It saves space on the system, allocations are not necessarily pushed to order
of 2 and should provide performance benefits.  Users who disable the memory
controller can also double check that the memory controller is not allocating
page_cgroup's.

NOTE: Hugh Dickins brought up the issue of whether we want to mark page_cgroup
as __GFP_MOVABLE or __GFP_RECLAIMABLE.  I don't think there is an easy answer
at the moment.  page_cgroup's are associated with user pages, they can be
reclaimed once the user page has been reclaimed, so it might make sense to
mark them as __GFP_RECLAIMABLE.  For now, I am leaving the marking to default
values that the slab allocator uses.

Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Sudhir Kumar <skumar@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c

index 350a14d..f407969 100644 (file)
@@ -26,6 +26,7 @@
 #include <linux/backing-dev.h>
 #include <linux/bit_spinlock.h>
 #include <linux/rcupdate.h>
+#include <linux/slab.h>
 #include <linux/swap.h>
 #include <linux/spinlock.h>
 #include <linux/fs.h>
@@ -35,6 +36,7 @@
 
 struct cgroup_subsys mem_cgroup_subsys;
 static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
+static struct kmem_cache *page_cgroup_cache;
 
 /*
  * Statistics for memory cgroup.
@@ -547,7 +549,7 @@ retry:
        }
        unlock_page_cgroup(page);
 
-       pc = kzalloc(sizeof(struct page_cgroup), gfp_mask);
+       pc = kmem_cache_zalloc(page_cgroup_cache, gfp_mask);
        if (pc == NULL)
                goto err;
 
@@ -609,7 +611,7 @@ retry:
                 */
                res_counter_uncharge(&mem->res, PAGE_SIZE);
                css_put(&mem->css);
-               kfree(pc);
+               kmem_cache_free(page_cgroup_cache, pc);
                goto retry;
        }
        page_assign_page_cgroup(page, pc);
@@ -624,7 +626,7 @@ done:
        return 0;
 out:
        css_put(&mem->css);
-       kfree(pc);
+       kmem_cache_free(page_cgroup_cache, pc);
 err:
        return -ENOMEM;
 }
@@ -682,7 +684,7 @@ void mem_cgroup_uncharge_page(struct page *page)
                res_counter_uncharge(&mem->res, PAGE_SIZE);
                css_put(&mem->css);
 
-               kfree(pc);
+               kmem_cache_free(page_cgroup_cache, pc);
                return;
        }
 
@@ -989,10 +991,12 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
        struct mem_cgroup *mem;
        int node;
 
-       if (unlikely((cont->parent) == NULL))
+       if (unlikely((cont->parent) == NULL)) {
                mem = &init_mem_cgroup;
-       else
+               page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
+       } else {
                mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL);
+       }
 
        if (mem == NULL)
                return ERR_PTR(-ENOMEM);