slub: Allow removal of slab caches during boot
authorChristoph Lameter <cl@linux-foundation.org>
Mon, 19 Jul 2010 16:39:11 +0000 (11:39 -0500)
committerPekka Enberg <penberg@cs.helsinki.fi>
Tue, 3 Aug 2010 04:28:32 +0000 (07:28 +0300)
Serialize kmem_cache_create and kmem_cache_destroy using the slub_lock. Only
possible after the use of the slub_lock during dynamic dma creation has been
removed.

Then make sure that the setup of the slab sysfs entries does not race
with kmem_cache_create and kmem_cache destroy.

If a slab cache is removed before we have setup sysfs then simply skip over
the sysfs handling.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Roland Dreier <rdreier@cisco.com>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
mm/slub.c

index f0f403693bd9621180d3f969e75048dc9fe2b42c..fb6518efe1ed4ebfa511158229dc37ec45b9ee58 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2491,7 +2491,6 @@ void kmem_cache_destroy(struct kmem_cache *s)
        s->refcount--;
        if (!s->refcount) {
                list_del(&s->list);
-               up_write(&slub_lock);
                if (kmem_cache_close(s)) {
                        printk(KERN_ERR "SLUB %s: %s called for cache that "
                                "still has objects.\n", s->name, __func__);
@@ -2500,8 +2499,8 @@ void kmem_cache_destroy(struct kmem_cache *s)
                if (s->flags & SLAB_DESTROY_BY_RCU)
                        rcu_barrier();
                sysfs_slab_remove(s);
-       } else
-               up_write(&slub_lock);
+       }
+       up_write(&slub_lock);
 }
 EXPORT_SYMBOL(kmem_cache_destroy);
 
@@ -3227,14 +3226,12 @@ struct kmem_cache *kmem_cache_create(const char *name, size_t size,
                 */
                s->objsize = max(s->objsize, (int)size);
                s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
-               up_write(&slub_lock);
 
                if (sysfs_slab_alias(s, name)) {
-                       down_write(&slub_lock);
                        s->refcount--;
-                       up_write(&slub_lock);
                        goto err;
                }
+               up_write(&slub_lock);
                return s;
        }
 
@@ -3243,14 +3240,12 @@ struct kmem_cache *kmem_cache_create(const char *name, size_t size,
                if (kmem_cache_open(s, GFP_KERNEL, name,
                                size, align, flags, ctor)) {
                        list_add(&s->list, &slab_caches);
-                       up_write(&slub_lock);
                        if (sysfs_slab_add(s)) {
-                               down_write(&slub_lock);
                                list_del(&s->list);
-                               up_write(&slub_lock);
                                kfree(s);
                                goto err;
                        }
+                       up_write(&slub_lock);
                        return s;
                }
                kfree(s);
@@ -4498,6 +4493,13 @@ static int sysfs_slab_add(struct kmem_cache *s)
 
 static void sysfs_slab_remove(struct kmem_cache *s)
 {
+       if (slab_state < SYSFS)
+               /*
+                * Sysfs has not been setup yet so no need to remove the
+                * cache from sysfs.
+                */
+               return;
+
        kobject_uevent(&s->kobj, KOBJ_REMOVE);
        kobject_del(&s->kobj);
        kobject_put(&s->kobj);
@@ -4543,8 +4545,11 @@ static int __init slab_sysfs_init(void)
        struct kmem_cache *s;
        int err;
 
+       down_write(&slub_lock);
+
        slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
        if (!slab_kset) {
+               up_write(&slub_lock);
                printk(KERN_ERR "Cannot register slab subsystem.\n");
                return -ENOSYS;
        }
@@ -4569,6 +4574,7 @@ static int __init slab_sysfs_init(void)
                kfree(al);
        }
 
+       up_write(&slub_lock);
        resiliency_test();
        return 0;
 }