From 781b2ba6eb5f22440afac9c79a89ebd6e3674a60 Mon Sep 17 00:00:00 2001 From: Pekka Enberg Date: Wed, 10 Jun 2009 18:50:32 +0300 Subject: SLUB: Out-of-memory diagnostics As suggested by Mel Gorman, add out-of-memory diagnostics to the SLUB allocator to make debugging OOM conditions easier. This patch helped hunt down a nasty OOM issue that popped up every now that was caused by SLUB debugging code which forced 4096 byte allocations to use order 1 pages even in the fallback case. An example print out looks like this: SLUB: Unable to allocate memory on node -1 (gfp=20) cache: kmalloc-4096, object size: 4096, buffer size: 4168, default order: 3, min order: 1 node 0: slabs: 95, objs: 665, free: 0 Acked-by: Christoph Lameter Acked-by: Mel Gorman Tested-by: Larry Finger Signed-off-by: Pekka Enberg --- mm/slub.c | 70 ++++++++++++++++++++++++++++++++++++++++++++++----------------- 1 file changed, 51 insertions(+), 19 deletions(-) (limited to 'mm/slub.c') diff --git a/mm/slub.c b/mm/slub.c index 65ffda5934b0..a5a4ecf7e393 100644 --- a/mm/slub.c +++ b/mm/slub.c @@ -1484,6 +1484,56 @@ static inline int node_match(struct kmem_cache_cpu *c, int node) return 1; } +static int count_free(struct page *page) +{ + return page->objects - page->inuse; +} + +static unsigned long count_partial(struct kmem_cache_node *n, + int (*get_count)(struct page *)) +{ + unsigned long flags; + unsigned long x = 0; + struct page *page; + + spin_lock_irqsave(&n->list_lock, flags); + list_for_each_entry(page, &n->partial, lru) + x += get_count(page); + spin_unlock_irqrestore(&n->list_lock, flags); + return x; +} + +static noinline void +slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) +{ + int node; + + printk(KERN_WARNING + "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", + nid, gfpflags); + printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " + "default order: %d, min order: %d\n", s->name, s->objsize, + s->size, oo_order(s->oo), oo_order(s->min)); + + for_each_online_node(node) { + struct kmem_cache_node *n = get_node(s, node); + unsigned long nr_slabs; + unsigned long nr_objs; + unsigned long nr_free; + + if (!n) + continue; + + nr_slabs = atomic_long_read(&n->nr_slabs); + nr_objs = atomic_long_read(&n->total_objects); + nr_free = count_partial(n, count_free); + + printk(KERN_WARNING + " node %d: slabs: %ld, objs: %ld, free: %ld\n", + node, nr_slabs, nr_objs, nr_free); + } +} + /* * Slow path. The lockless freelist is empty or we need to perform * debugging duties. @@ -1565,6 +1615,7 @@ new_slab: c->page = new; goto load_freelist; } + slab_out_of_memory(s, gfpflags, node); return NULL; debug: if (!alloc_debug_processing(s, c->page, object, addr)) @@ -3318,20 +3369,6 @@ void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, } #ifdef CONFIG_SLUB_DEBUG -static unsigned long count_partial(struct kmem_cache_node *n, - int (*get_count)(struct page *)) -{ - unsigned long flags; - unsigned long x = 0; - struct page *page; - - spin_lock_irqsave(&n->list_lock, flags); - list_for_each_entry(page, &n->partial, lru) - x += get_count(page); - spin_unlock_irqrestore(&n->list_lock, flags); - return x; -} - static int count_inuse(struct page *page) { return page->inuse; @@ -3342,11 +3379,6 @@ static int count_total(struct page *page) return page->objects; } -static int count_free(struct page *page) -{ - return page->objects - page->inuse; -} - static int validate_slab(struct kmem_cache *s, struct page *page, unsigned long *map) { -- cgit v1.2.3 From 26c02cf05ddadfee3952e829b841583794bf46f6 Mon Sep 17 00:00:00 2001 From: Alexander Beregalov Date: Thu, 11 Jun 2009 14:08:48 +0400 Subject: SLUB: fix build when !SLUB_DEBUG Fix this build error when CONFIG_SLUB_DEBUG is not set: mm/slub.c: In function 'slab_out_of_memory': mm/slub.c:1551: error: 'struct kmem_cache_node' has no member named 'nr_slabs' mm/slub.c:1552: error: 'struct kmem_cache_node' has no member named 'total_objects' [ penberg@cs.helsinki.fi: cleanups ] Signed-off-by: Alexander Beregalov Signed-off-by: Pekka Enberg --- mm/slub.c | 22 +++++++++++++++++++--- 1 file changed, 19 insertions(+), 3 deletions(-) (limited to 'mm/slub.c') diff --git a/mm/slub.c b/mm/slub.c index a5a4ecf7e393..9fb892b6afe3 100644 --- a/mm/slub.c +++ b/mm/slub.c @@ -832,6 +832,11 @@ static inline unsigned long slabs_node(struct kmem_cache *s, int node) return atomic_long_read(&n->nr_slabs); } +static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) +{ + return atomic_long_read(&n->nr_slabs); +} + static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) { struct kmem_cache_node *n = get_node(s, node); @@ -1050,6 +1055,8 @@ static inline unsigned long kmem_cache_flags(unsigned long objsize, static inline unsigned long slabs_node(struct kmem_cache *s, int node) { return 0; } +static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) + { return 0; } static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) {} static inline void dec_slabs_node(struct kmem_cache *s, int node, @@ -1503,6 +1510,15 @@ static unsigned long count_partial(struct kmem_cache_node *n, return x; } +static inline unsigned long node_nr_objs(struct kmem_cache_node *n) +{ +#ifdef CONFIG_SLUB_DEBUG + return atomic_long_read(&n->total_objects); +#else + return 0; +#endif +} + static noinline void slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) { @@ -1524,9 +1540,9 @@ slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) if (!n) continue; - nr_slabs = atomic_long_read(&n->nr_slabs); - nr_objs = atomic_long_read(&n->total_objects); - nr_free = count_partial(n, count_free); + nr_free = count_partial(n, count_free); + nr_slabs = node_nr_slabs(n); + nr_objs = node_nr_objs(n); printk(KERN_WARNING " node %d: slabs: %ld, objs: %ld, free: %ld\n", -- cgit v1.2.3 From 95f8598931bd86a5775073db2fa2004b892dd3d0 Mon Sep 17 00:00:00 2001 From: Pekka Enberg Date: Thu, 11 Jun 2009 16:18:09 +0300 Subject: SLUB: Don't print out OOM warning for __GFP_NOFAIL We must check for __GFP_NOFAIL like the page allocator does; otherwise we end up with false positives. While at it, add the printk_ratelimit() check in SLUB as well. Cc: Alexander Beregalov Signed-off-by: Pekka Enberg --- mm/slub.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'mm/slub.c') diff --git a/mm/slub.c b/mm/slub.c index 9fb892b6afe3..0996b3be751a 100644 --- a/mm/slub.c +++ b/mm/slub.c @@ -1631,7 +1631,8 @@ new_slab: c->page = new; goto load_freelist; } - slab_out_of_memory(s, gfpflags, node); + if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) + slab_out_of_memory(s, gfpflags, node); return NULL; debug: if (!alloc_debug_processing(s, c->page, object, addr)) -- cgit v1.2.3