blob: ba2ca53f6c3aafdd3165c3ae5115fb719d0c17bd [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
Christoph Lameter881db7f2011-06-01 12:25:53 -05005 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07007 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter881db7f2011-06-01 12:25:53 -05009 * (C) 2011 Linux Foundation, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -070010 */
11
12#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100013#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070014#include <linux/module.h>
15#include <linux/bit_spinlock.h>
16#include <linux/interrupt.h>
17#include <linux/bitops.h>
18#include <linux/slab.h>
Christoph Lameter97d06602012-07-06 15:25:11 -050019#include "slab.h"
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040020#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070021#include <linux/seq_file.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020022#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070023#include <linux/cpu.h>
24#include <linux/cpuset.h>
25#include <linux/mempolicy.h>
26#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070027#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070028#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070029#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070030#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090031#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030032#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060033#include <linux/prefetch.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080034#include <linux/memcontrol.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070035
Richard Kennedy4a923792010-10-21 10:29:19 +010036#include <trace/events/kmem.h>
37
Mel Gorman072bb0a2012-07-31 16:43:58 -070038#include "internal.h"
39
Christoph Lameter81819f02007-05-06 14:49:36 -070040/*
41 * Lock order:
Christoph Lameter18004c52012-07-06 15:25:12 -050042 * 1. slab_mutex (Global Mutex)
Christoph Lameter881db7f2011-06-01 12:25:53 -050043 * 2. node->list_lock
44 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070045 *
Christoph Lameter18004c52012-07-06 15:25:12 -050046 * slab_mutex
Christoph Lameter881db7f2011-06-01 12:25:53 -050047 *
Christoph Lameter18004c52012-07-06 15:25:12 -050048 * The role of the slab_mutex is to protect the list of all the slabs
Christoph Lameter881db7f2011-06-01 12:25:53 -050049 * and to synchronize major metadata changes to slab cache structures.
50 *
51 * The slab_lock is only used for debugging and on arches that do not
52 * have the ability to do a cmpxchg_double. It only protects the second
53 * double word in the page struct. Meaning
54 * A. page->freelist -> List of object free in a page
55 * B. page->counters -> Counters of objects
56 * C. page->frozen -> frozen state
57 *
58 * If a slab is frozen then it is exempt from list management. It is not
59 * on any list. The processor that froze the slab is the one who can
60 * perform list operations on the page. Other processors may put objects
61 * onto the freelist but the processor that froze the slab is the only
62 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070063 *
64 * The list_lock protects the partial and full list on each node and
65 * the partial slab counter. If taken then no new slabs may be added or
66 * removed from the lists nor make the number of partial slabs be modified.
67 * (Note that the total number of slabs is an atomic value that may be
68 * modified without taking the list lock).
69 *
70 * The list_lock is a centralized lock and thus we avoid taking it as
71 * much as possible. As long as SLUB does not have to handle partial
72 * slabs, operations can continue without any centralized lock. F.e.
73 * allocating a long series of objects that fill up slabs does not require
74 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070075 * Interrupts are disabled during allocation and deallocation in order to
76 * make the slab allocator safe to use in the context of an irq. In addition
77 * interrupts are disabled to ensure that the processor does not change
78 * while handling per_cpu slabs, due to kernel preemption.
79 *
80 * SLUB assigns one slab for allocation to each processor.
81 * Allocations only occur from these slabs called cpu slabs.
82 *
Christoph Lameter672bba32007-05-09 02:32:39 -070083 * Slabs with free elements are kept on a partial list and during regular
84 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070085 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070086 * We track full slabs for debugging purposes though because otherwise we
87 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070088 *
89 * Slabs are freed when they become empty. Teardown and setup is
90 * minimal so we rely on the page allocators per cpu caches for
91 * fast frees and allocs.
92 *
93 * Overloading of page flags that are otherwise used for LRU management.
94 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070095 * PageActive The slab is frozen and exempt from list processing.
96 * This means that the slab is dedicated to a purpose
97 * such as satisfying allocations for a specific
98 * processor. Objects may be freed in the slab while
99 * it is frozen but slab_free will then skip the usual
100 * list operations. It is up to the processor holding
101 * the slab to integrate the slab into the slab lists
102 * when the slab is no longer needed.
103 *
104 * One use of this flag is to mark slabs that are
105 * used for allocations. Then such a slab becomes a cpu
106 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700107 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700108 * free objects in addition to the regular freelist
109 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700110 *
111 * PageError Slab requires special handling due to debug
112 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700113 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700114 */
115
Christoph Lameteraf537b02010-07-09 14:07:14 -0500116static inline int kmem_cache_debug(struct kmem_cache *s)
117{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700118#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500119 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500121 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700122#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500123}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700124
Christoph Lameter81819f02007-05-06 14:49:36 -0700125/*
126 * Issues still to be resolved:
127 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700128 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
129 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700130 * - Variable sizing of the per node arrays
131 */
132
133/* Enable to test recovery from slab corruption on boot */
134#undef SLUB_RESILIENCY_TEST
135
Christoph Lameterb789ef52011-06-01 12:25:49 -0500136/* Enable to log cmpxchg failures */
137#undef SLUB_DEBUG_CMPXCHG
138
Christoph Lameter81819f02007-05-06 14:49:36 -0700139/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700140 * Mininum number of partial slabs. These will be left on the partial
141 * lists even if they are empty. kmem_cache_shrink may reclaim them.
142 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800143#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700144
Christoph Lameter2086d262007-05-06 14:49:46 -0700145/*
146 * Maximum number of desirable partial slabs.
147 * The existence of more partial slabs makes kmem_cache_shrink
148 * sort the partial list by the number of objects in the.
149 */
150#define MAX_PARTIAL 10
151
Christoph Lameter81819f02007-05-06 14:49:36 -0700152#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
153 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700154
Christoph Lameter81819f02007-05-06 14:49:36 -0700155/*
David Rientjes3de47212009-07-27 18:30:35 -0700156 * Debugging flags that require metadata to be stored in the slab. These get
157 * disabled when slub_debug=O is used and a cache's min order increases with
158 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700159 */
David Rientjes3de47212009-07-27 18:30:35 -0700160#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700161
162/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700163 * Set of flags that will prevent slab merging
164 */
165#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300166 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
167 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700168
169#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200170 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700171
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400172#define OO_SHIFT 16
173#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500174#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400175
Christoph Lameter81819f02007-05-06 14:49:36 -0700176/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500177#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500178#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700179
Christoph Lameter81819f02007-05-06 14:49:36 -0700180#ifdef CONFIG_SMP
181static struct notifier_block slab_notifier;
182#endif
183
Christoph Lameter02cbc872007-05-09 02:32:43 -0700184/*
185 * Tracking user of a slab.
186 */
Ben Greeard6543e32011-07-07 11:36:36 -0700187#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700188struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300189 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700190#ifdef CONFIG_STACKTRACE
191 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
192#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700193 int cpu; /* Was running on cpu */
194 int pid; /* Pid context */
195 unsigned long when; /* When did the operation occur */
196};
197
198enum track_item { TRACK_ALLOC, TRACK_FREE };
199
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500200#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700201static int sysfs_slab_add(struct kmem_cache *);
202static int sysfs_slab_alias(struct kmem_cache *, const char *);
203static void sysfs_slab_remove(struct kmem_cache *);
Glauber Costa107dab52012-12-18 14:23:05 -0800204static void memcg_propagate_slab_attrs(struct kmem_cache *s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700205#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700206static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
207static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
208 { return 0; }
Christoph Lameterdb265ec2012-09-04 23:18:33 +0000209static inline void sysfs_slab_remove(struct kmem_cache *s) { }
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800210
Glauber Costa107dab52012-12-18 14:23:05 -0800211static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700212#endif
213
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500214static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800215{
216#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e62009-12-18 16:26:23 -0600217 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800218#endif
219}
220
Christoph Lameter81819f02007-05-06 14:49:36 -0700221/********************************************************************
222 * Core slab cache functions
223 *******************************************************************/
224
Christoph Lameter81819f02007-05-06 14:49:36 -0700225static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
226{
Christoph Lameter81819f02007-05-06 14:49:36 -0700227 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700228}
229
Christoph Lameter6446faa2008-02-15 23:45:26 -0800230/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700231static inline int check_valid_pointer(struct kmem_cache *s,
232 struct page *page, const void *object)
233{
234 void *base;
235
Christoph Lametera973e9d2008-03-01 13:40:44 -0800236 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700237 return 1;
238
Christoph Lametera973e9d2008-03-01 13:40:44 -0800239 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300240 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700241 (object - base) % s->size) {
242 return 0;
243 }
244
245 return 1;
246}
247
Christoph Lameter7656c722007-05-09 02:32:40 -0700248static inline void *get_freepointer(struct kmem_cache *s, void *object)
249{
250 return *(void **)(object + s->offset);
251}
252
Eric Dumazet0ad95002011-12-16 16:25:34 +0100253static void prefetch_freepointer(const struct kmem_cache *s, void *object)
254{
255 prefetch(object + s->offset);
256}
257
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500258static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
259{
260 void *p;
261
262#ifdef CONFIG_DEBUG_PAGEALLOC
263 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
264#else
265 p = get_freepointer(s, object);
266#endif
267 return p;
268}
269
Christoph Lameter7656c722007-05-09 02:32:40 -0700270static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
271{
272 *(void **)(object + s->offset) = fp;
273}
274
275/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300276#define for_each_object(__p, __s, __addr, __objects) \
277 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700278 __p += (__s)->size)
279
Christoph Lameter7656c722007-05-09 02:32:40 -0700280/* Determine object index from a given position */
281static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
282{
283 return (p - addr) / s->size;
284}
285
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100286static inline size_t slab_ksize(const struct kmem_cache *s)
287{
288#ifdef CONFIG_SLUB_DEBUG
289 /*
290 * Debugging requires use of the padding between object
291 * and whatever may come after it.
292 */
293 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500294 return s->object_size;
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100295
296#endif
297 /*
298 * If we have the need to store the freelist pointer
299 * back there or track user information then we can
300 * only use the space before that information.
301 */
302 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
303 return s->inuse;
304 /*
305 * Else we can use all the padding etc for the allocation
306 */
307 return s->size;
308}
309
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800310static inline int order_objects(int order, unsigned long size, int reserved)
311{
312 return ((PAGE_SIZE << order) - reserved) / size;
313}
314
Christoph Lameter834f3d12008-04-14 19:11:31 +0300315static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800316 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300317{
318 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800319 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300320 };
321
322 return x;
323}
324
325static inline int oo_order(struct kmem_cache_order_objects x)
326{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400327 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300328}
329
330static inline int oo_objects(struct kmem_cache_order_objects x)
331{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400332 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300333}
334
Christoph Lameter881db7f2011-06-01 12:25:53 -0500335/*
336 * Per slab locking using the pagelock
337 */
338static __always_inline void slab_lock(struct page *page)
339{
340 bit_spin_lock(PG_locked, &page->flags);
341}
342
343static __always_inline void slab_unlock(struct page *page)
344{
345 __bit_spin_unlock(PG_locked, &page->flags);
346}
347
Christoph Lameter1d071712011-07-14 12:49:12 -0500348/* Interrupts must be disabled (for the fallback code to work right) */
349static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500350 void *freelist_old, unsigned long counters_old,
351 void *freelist_new, unsigned long counters_new,
352 const char *n)
353{
Christoph Lameter1d071712011-07-14 12:49:12 -0500354 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800355#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
356 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500357 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000358 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500359 freelist_old, counters_old,
360 freelist_new, counters_new))
361 return 1;
362 } else
363#endif
364 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500365 slab_lock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500366 if (page->freelist == freelist_old && page->counters == counters_old) {
367 page->freelist = freelist_new;
368 page->counters = counters_new;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500369 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500370 return 1;
371 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500372 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500373 }
374
375 cpu_relax();
376 stat(s, CMPXCHG_DOUBLE_FAIL);
377
378#ifdef SLUB_DEBUG_CMPXCHG
379 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
380#endif
381
382 return 0;
383}
384
Christoph Lameter1d071712011-07-14 12:49:12 -0500385static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
386 void *freelist_old, unsigned long counters_old,
387 void *freelist_new, unsigned long counters_new,
388 const char *n)
389{
Heiko Carstens25654092012-01-12 17:17:33 -0800390#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
391 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500392 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000393 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameter1d071712011-07-14 12:49:12 -0500394 freelist_old, counters_old,
395 freelist_new, counters_new))
396 return 1;
397 } else
398#endif
399 {
400 unsigned long flags;
401
402 local_irq_save(flags);
403 slab_lock(page);
404 if (page->freelist == freelist_old && page->counters == counters_old) {
405 page->freelist = freelist_new;
406 page->counters = counters_new;
407 slab_unlock(page);
408 local_irq_restore(flags);
409 return 1;
410 }
411 slab_unlock(page);
412 local_irq_restore(flags);
413 }
414
415 cpu_relax();
416 stat(s, CMPXCHG_DOUBLE_FAIL);
417
418#ifdef SLUB_DEBUG_CMPXCHG
419 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
420#endif
421
422 return 0;
423}
424
Christoph Lameter41ecc552007-05-09 02:32:44 -0700425#ifdef CONFIG_SLUB_DEBUG
426/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500427 * Determine a map of object in use on a page.
428 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500429 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500430 * not vanish from under us.
431 */
432static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
433{
434 void *p;
435 void *addr = page_address(page);
436
437 for (p = page->freelist; p; p = get_freepointer(s, p))
438 set_bit(slab_index(p, s, addr), map);
439}
440
Christoph Lameter41ecc552007-05-09 02:32:44 -0700441/*
442 * Debug settings:
443 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700444#ifdef CONFIG_SLUB_DEBUG_ON
445static int slub_debug = DEBUG_DEFAULT_FLAGS;
446#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700447static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700448#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700449
450static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700451static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700452
Christoph Lameter7656c722007-05-09 02:32:40 -0700453/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700454 * Object debugging
455 */
456static void print_section(char *text, u8 *addr, unsigned int length)
457{
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200458 print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
459 length, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700460}
461
Christoph Lameter81819f02007-05-06 14:49:36 -0700462static struct track *get_track(struct kmem_cache *s, void *object,
463 enum track_item alloc)
464{
465 struct track *p;
466
467 if (s->offset)
468 p = object + s->offset + sizeof(void *);
469 else
470 p = object + s->inuse;
471
472 return p + alloc;
473}
474
475static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300476 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700477{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900478 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700479
Christoph Lameter81819f02007-05-06 14:49:36 -0700480 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700481#ifdef CONFIG_STACKTRACE
482 struct stack_trace trace;
483 int i;
484
485 trace.nr_entries = 0;
486 trace.max_entries = TRACK_ADDRS_COUNT;
487 trace.entries = p->addrs;
488 trace.skip = 3;
489 save_stack_trace(&trace);
490
491 /* See rant in lockdep.c */
492 if (trace.nr_entries != 0 &&
493 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
494 trace.nr_entries--;
495
496 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
497 p->addrs[i] = 0;
498#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700499 p->addr = addr;
500 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400501 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700502 p->when = jiffies;
503 } else
504 memset(p, 0, sizeof(struct track));
505}
506
Christoph Lameter81819f02007-05-06 14:49:36 -0700507static void init_tracking(struct kmem_cache *s, void *object)
508{
Christoph Lameter24922682007-07-17 04:03:18 -0700509 if (!(s->flags & SLAB_STORE_USER))
510 return;
511
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300512 set_track(s, object, TRACK_FREE, 0UL);
513 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700514}
515
516static void print_track(const char *s, struct track *t)
517{
518 if (!t->addr)
519 return;
520
Linus Torvalds7daf7052008-07-14 12:12:53 -0700521 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300522 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700523#ifdef CONFIG_STACKTRACE
524 {
525 int i;
526 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
527 if (t->addrs[i])
528 printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
529 else
530 break;
531 }
532#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700533}
534
Christoph Lameter24922682007-07-17 04:03:18 -0700535static void print_tracking(struct kmem_cache *s, void *object)
536{
537 if (!(s->flags & SLAB_STORE_USER))
538 return;
539
540 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
541 print_track("Freed", get_track(s, object, TRACK_FREE));
542}
543
544static void print_page_info(struct page *page)
545{
Christoph Lameter39b26462008-04-14 19:11:30 +0300546 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
547 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700548
549}
550
551static void slab_bug(struct kmem_cache *s, char *fmt, ...)
552{
553 va_list args;
554 char buf[100];
555
556 va_start(args, fmt);
557 vsnprintf(buf, sizeof(buf), fmt, args);
558 va_end(args);
559 printk(KERN_ERR "========================================"
560 "=====================================\n");
Dave Jones265d47e2011-11-15 15:04:00 -0800561 printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700562 printk(KERN_ERR "----------------------------------------"
563 "-------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400564
565 add_taint(TAINT_BAD_PAGE);
Christoph Lameter24922682007-07-17 04:03:18 -0700566}
567
568static void slab_fix(struct kmem_cache *s, char *fmt, ...)
569{
570 va_list args;
571 char buf[100];
572
573 va_start(args, fmt);
574 vsnprintf(buf, sizeof(buf), fmt, args);
575 va_end(args);
576 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
577}
578
579static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700580{
581 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800582 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700583
584 print_tracking(s, p);
585
586 print_page_info(page);
587
588 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
589 p, p - addr, get_freepointer(s, p));
590
591 if (p > addr + 16)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200592 print_section("Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700593
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500594 print_section("Object ", p, min_t(unsigned long, s->object_size,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200595 PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700596 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500597 print_section("Redzone ", p + s->object_size,
598 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700599
Christoph Lameter81819f02007-05-06 14:49:36 -0700600 if (s->offset)
601 off = s->offset + sizeof(void *);
602 else
603 off = s->inuse;
604
Christoph Lameter24922682007-07-17 04:03:18 -0700605 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700606 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700607
608 if (off != s->size)
609 /* Beginning of the filler is the free pointer */
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200610 print_section("Padding ", p + off, s->size - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700611
612 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700613}
614
615static void object_err(struct kmem_cache *s, struct page *page,
616 u8 *object, char *reason)
617{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700618 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700619 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700620}
621
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000622static void slab_err(struct kmem_cache *s, struct page *page, const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700623{
624 va_list args;
625 char buf[100];
626
Christoph Lameter24922682007-07-17 04:03:18 -0700627 va_start(args, fmt);
628 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700629 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700630 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700631 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700632 dump_stack();
633}
634
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500635static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700636{
637 u8 *p = object;
638
639 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500640 memset(p, POISON_FREE, s->object_size - 1);
641 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700642 }
643
644 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500645 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700646}
647
Christoph Lameter24922682007-07-17 04:03:18 -0700648static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
649 void *from, void *to)
650{
651 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
652 memset(from, data, to - from);
653}
654
655static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
656 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800657 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700658{
659 u8 *fault;
660 u8 *end;
661
Akinobu Mita798248202011-10-31 17:08:07 -0700662 fault = memchr_inv(start, value, bytes);
Christoph Lameter24922682007-07-17 04:03:18 -0700663 if (!fault)
664 return 1;
665
666 end = start + bytes;
667 while (end > fault && end[-1] == value)
668 end--;
669
670 slab_bug(s, "%s overwritten", what);
671 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
672 fault, end - 1, fault[0], value);
673 print_trailer(s, page, object);
674
675 restore_bytes(s, what, value, fault, end);
676 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700677}
678
Christoph Lameter81819f02007-05-06 14:49:36 -0700679/*
680 * Object layout:
681 *
682 * object address
683 * Bytes of the object to be managed.
684 * If the freepointer may overlay the object then the free
685 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700686 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700687 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
688 * 0xa5 (POISON_END)
689 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500690 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700691 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700692 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500693 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700694 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700695 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
696 * 0xcc (RED_ACTIVE) for objects in use.
697 *
698 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700699 * Meta data starts here.
700 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700701 * A. Free pointer (if we cannot overwrite object on free)
702 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700703 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800704 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700705 * before the word boundary.
706 *
707 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700708 *
709 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700710 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700711 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500712 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700713 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700714 * may be used with merged slabcaches.
715 */
716
Christoph Lameter81819f02007-05-06 14:49:36 -0700717static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
718{
719 unsigned long off = s->inuse; /* The end of info */
720
721 if (s->offset)
722 /* Freepointer is placed after the object. */
723 off += sizeof(void *);
724
725 if (s->flags & SLAB_STORE_USER)
726 /* We also have user information there */
727 off += 2 * sizeof(struct track);
728
729 if (s->size == off)
730 return 1;
731
Christoph Lameter24922682007-07-17 04:03:18 -0700732 return check_bytes_and_report(s, page, p, "Object padding",
733 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700734}
735
Christoph Lameter39b26462008-04-14 19:11:30 +0300736/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700737static int slab_pad_check(struct kmem_cache *s, struct page *page)
738{
Christoph Lameter24922682007-07-17 04:03:18 -0700739 u8 *start;
740 u8 *fault;
741 u8 *end;
742 int length;
743 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700744
745 if (!(s->flags & SLAB_POISON))
746 return 1;
747
Christoph Lametera973e9d2008-03-01 13:40:44 -0800748 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800749 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300750 end = start + length;
751 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700752 if (!remainder)
753 return 1;
754
Akinobu Mita798248202011-10-31 17:08:07 -0700755 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700756 if (!fault)
757 return 1;
758 while (end > fault && end[-1] == POISON_INUSE)
759 end--;
760
761 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200762 print_section("Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700763
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200764 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700765 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700766}
767
768static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500769 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700770{
771 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500772 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700773
774 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700775 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500776 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700777 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700778 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500779 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800780 check_bytes_and_report(s, page, p, "Alignment padding",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500781 endobject, POISON_INUSE, s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800782 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700783 }
784
785 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500786 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700787 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500788 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700789 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500790 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700791 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700792 /*
793 * check_pad_bytes cleans up on its own.
794 */
795 check_pad_bytes(s, page, p);
796 }
797
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500798 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700799 /*
800 * Object and freepointer overlap. Cannot check
801 * freepointer while object is allocated.
802 */
803 return 1;
804
805 /* Check free pointer validity */
806 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
807 object_err(s, page, p, "Freepointer corrupt");
808 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100809 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700810 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700811 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700812 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800813 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700814 return 0;
815 }
816 return 1;
817}
818
819static int check_slab(struct kmem_cache *s, struct page *page)
820{
Christoph Lameter39b26462008-04-14 19:11:30 +0300821 int maxobj;
822
Christoph Lameter81819f02007-05-06 14:49:36 -0700823 VM_BUG_ON(!irqs_disabled());
824
825 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700826 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700827 return 0;
828 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300829
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800830 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300831 if (page->objects > maxobj) {
832 slab_err(s, page, "objects %u > max %u",
833 s->name, page->objects, maxobj);
834 return 0;
835 }
836 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700837 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300838 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700839 return 0;
840 }
841 /* Slab_pad_check fixes things up after itself */
842 slab_pad_check(s, page);
843 return 1;
844}
845
846/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700847 * Determine if a certain object on a page is on the freelist. Must hold the
848 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700849 */
850static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
851{
852 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500853 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700854 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300855 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700856
Christoph Lameter881db7f2011-06-01 12:25:53 -0500857 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300858 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700859 if (fp == search)
860 return 1;
861 if (!check_valid_pointer(s, page, fp)) {
862 if (object) {
863 object_err(s, page, object,
864 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800865 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700866 break;
867 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700868 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800869 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300870 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700871 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700872 return 0;
873 }
874 break;
875 }
876 object = fp;
877 fp = get_freepointer(s, object);
878 nr++;
879 }
880
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800881 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400882 if (max_objects > MAX_OBJS_PER_PAGE)
883 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300884
885 if (page->objects != max_objects) {
886 slab_err(s, page, "Wrong number of objects. Found %d but "
887 "should be %d", page->objects, max_objects);
888 page->objects = max_objects;
889 slab_fix(s, "Number of objects adjusted.");
890 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300891 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700892 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300893 "counted were %d", page->inuse, page->objects - nr);
894 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700895 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700896 }
897 return search == NULL;
898}
899
Christoph Lameter0121c6192008-04-29 16:11:12 -0700900static void trace(struct kmem_cache *s, struct page *page, void *object,
901 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700902{
903 if (s->flags & SLAB_TRACE) {
904 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
905 s->name,
906 alloc ? "alloc" : "free",
907 object, page->inuse,
908 page->freelist);
909
910 if (!alloc)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500911 print_section("Object ", (void *)object, s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700912
913 dump_stack();
914 }
915}
916
Christoph Lameter643b1132007-05-06 14:49:42 -0700917/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500918 * Hooks for other subsystems that check memory allocations. In a typical
919 * production configuration these hooks all should produce no code at all.
920 */
921static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
922{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500923 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500924 lockdep_trace_alloc(flags);
925 might_sleep_if(flags & __GFP_WAIT);
926
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500927 return should_failslab(s->object_size, flags, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500928}
929
930static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
931{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500932 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100933 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500934 kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500935}
936
937static inline void slab_free_hook(struct kmem_cache *s, void *x)
938{
939 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500940
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600941 /*
942 * Trouble is that we may no longer disable interupts in the fast path
943 * So in order to make the debug calls that expect irqs to be
944 * disabled we need to disable interrupts temporarily.
945 */
946#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
947 {
948 unsigned long flags;
949
950 local_irq_save(flags);
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500951 kmemcheck_slab_free(s, x, s->object_size);
952 debug_check_no_locks_freed(x, s->object_size);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600953 local_irq_restore(flags);
954 }
955#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200956 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500957 debug_check_no_obj_freed(x, s->object_size);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500958}
959
960/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700961 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500962 *
963 * list_lock must be held.
Christoph Lameter643b1132007-05-06 14:49:42 -0700964 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500965static void add_full(struct kmem_cache *s,
966 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700967{
Christoph Lameter643b1132007-05-06 14:49:42 -0700968 if (!(s->flags & SLAB_STORE_USER))
969 return;
970
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500971 list_add(&page->lru, &n->full);
972}
Christoph Lameter643b1132007-05-06 14:49:42 -0700973
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500974/*
975 * list_lock must be held.
976 */
977static void remove_full(struct kmem_cache *s, struct page *page)
978{
979 if (!(s->flags & SLAB_STORE_USER))
980 return;
981
Christoph Lameter643b1132007-05-06 14:49:42 -0700982 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -0700983}
984
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300985/* Tracking of the number of slabs for debugging purposes */
986static inline unsigned long slabs_node(struct kmem_cache *s, int node)
987{
988 struct kmem_cache_node *n = get_node(s, node);
989
990 return atomic_long_read(&n->nr_slabs);
991}
992
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400993static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
994{
995 return atomic_long_read(&n->nr_slabs);
996}
997
Christoph Lameter205ab992008-04-14 19:11:40 +0300998static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300999{
1000 struct kmem_cache_node *n = get_node(s, node);
1001
1002 /*
1003 * May be called early in order to allocate a slab for the
1004 * kmem_cache_node structure. Solve the chicken-egg
1005 * dilemma by deferring the increment of the count during
1006 * bootstrap (see early_kmem_cache_node_alloc).
1007 */
Christoph Lameter7340cc82010-09-28 08:10:26 -05001008 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001009 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001010 atomic_long_add(objects, &n->total_objects);
1011 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001012}
Christoph Lameter205ab992008-04-14 19:11:40 +03001013static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001014{
1015 struct kmem_cache_node *n = get_node(s, node);
1016
1017 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001018 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001019}
1020
1021/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001022static void setup_object_debug(struct kmem_cache *s, struct page *page,
1023 void *object)
1024{
1025 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1026 return;
1027
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001028 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001029 init_tracking(s, object);
1030}
1031
Christoph Lameter15370662010-08-20 12:37:12 -05001032static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001033 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001034{
1035 if (!check_slab(s, page))
1036 goto bad;
1037
Christoph Lameter81819f02007-05-06 14:49:36 -07001038 if (!check_valid_pointer(s, page, object)) {
1039 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001040 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001041 }
1042
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001043 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001044 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001045
Christoph Lameter3ec09742007-05-16 22:11:00 -07001046 /* Success perform special debug activities for allocs */
1047 if (s->flags & SLAB_STORE_USER)
1048 set_track(s, object, TRACK_ALLOC, addr);
1049 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001050 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001051 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001052
Christoph Lameter81819f02007-05-06 14:49:36 -07001053bad:
1054 if (PageSlab(page)) {
1055 /*
1056 * If this is a slab page then lets do the best we can
1057 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001058 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001059 */
Christoph Lameter24922682007-07-17 04:03:18 -07001060 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001061 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001062 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001063 }
1064 return 0;
1065}
1066
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001067static noinline struct kmem_cache_node *free_debug_processing(
1068 struct kmem_cache *s, struct page *page, void *object,
1069 unsigned long addr, unsigned long *flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001070{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001071 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001072
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001073 spin_lock_irqsave(&n->list_lock, *flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001074 slab_lock(page);
1075
Christoph Lameter81819f02007-05-06 14:49:36 -07001076 if (!check_slab(s, page))
1077 goto fail;
1078
1079 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001080 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001081 goto fail;
1082 }
1083
1084 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001085 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001086 goto fail;
1087 }
1088
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001089 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001090 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001091
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001092 if (unlikely(s != page->slab_cache)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001093 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001094 slab_err(s, page, "Attempt to free object(0x%p) "
1095 "outside of slab", object);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001096 } else if (!page->slab_cache) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001097 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001098 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001099 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001100 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001101 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001102 object_err(s, page, object,
1103 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001104 goto fail;
1105 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001106
Christoph Lameter3ec09742007-05-16 22:11:00 -07001107 if (s->flags & SLAB_STORE_USER)
1108 set_track(s, object, TRACK_FREE, addr);
1109 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001110 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001111out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001112 slab_unlock(page);
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001113 /*
1114 * Keep node_lock to preserve integrity
1115 * until the object is actually freed
1116 */
1117 return n;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001118
Christoph Lameter81819f02007-05-06 14:49:36 -07001119fail:
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001120 slab_unlock(page);
1121 spin_unlock_irqrestore(&n->list_lock, *flags);
Christoph Lameter24922682007-07-17 04:03:18 -07001122 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001123 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001124}
1125
Christoph Lameter41ecc552007-05-09 02:32:44 -07001126static int __init setup_slub_debug(char *str)
1127{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001128 slub_debug = DEBUG_DEFAULT_FLAGS;
1129 if (*str++ != '=' || !*str)
1130 /*
1131 * No options specified. Switch on full debugging.
1132 */
1133 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001134
1135 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001136 /*
1137 * No options but restriction on slabs. This means full
1138 * debugging for slabs matching a pattern.
1139 */
1140 goto check_slabs;
1141
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001142 if (tolower(*str) == 'o') {
1143 /*
1144 * Avoid enabling debugging on caches if its minimum order
1145 * would increase as a result.
1146 */
1147 disable_higher_order_debug = 1;
1148 goto out;
1149 }
1150
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001151 slub_debug = 0;
1152 if (*str == '-')
1153 /*
1154 * Switch off all debugging measures.
1155 */
1156 goto out;
1157
1158 /*
1159 * Determine which debug features should be switched on
1160 */
Pekka Enberg06428782008-01-07 23:20:27 -08001161 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001162 switch (tolower(*str)) {
1163 case 'f':
1164 slub_debug |= SLAB_DEBUG_FREE;
1165 break;
1166 case 'z':
1167 slub_debug |= SLAB_RED_ZONE;
1168 break;
1169 case 'p':
1170 slub_debug |= SLAB_POISON;
1171 break;
1172 case 'u':
1173 slub_debug |= SLAB_STORE_USER;
1174 break;
1175 case 't':
1176 slub_debug |= SLAB_TRACE;
1177 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001178 case 'a':
1179 slub_debug |= SLAB_FAILSLAB;
1180 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001181 default:
1182 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001183 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001184 }
1185 }
1186
1187check_slabs:
1188 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001189 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001190out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001191 return 1;
1192}
1193
1194__setup("slub_debug", setup_slub_debug);
1195
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001196static unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001197 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001198 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001199{
1200 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001201 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001202 */
Christoph Lametere1533622008-02-15 23:45:24 -08001203 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001204 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1205 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001206
1207 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001208}
1209#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001210static inline void setup_object_debug(struct kmem_cache *s,
1211 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001212
Christoph Lameter3ec09742007-05-16 22:11:00 -07001213static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001214 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001215
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001216static inline struct kmem_cache_node *free_debug_processing(
1217 struct kmem_cache *s, struct page *page, void *object,
1218 unsigned long addr, unsigned long *flags) { return NULL; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001219
Christoph Lameter41ecc552007-05-09 02:32:44 -07001220static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1221 { return 1; }
1222static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001223 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001224static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1225 struct page *page) {}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001226static inline void remove_full(struct kmem_cache *s, struct page *page) {}
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001227static inline unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001228 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001229 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001230{
1231 return flags;
1232}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001233#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001234
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001235#define disable_higher_order_debug 0
1236
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001237static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1238 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001239static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1240 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001241static inline void inc_slabs_node(struct kmem_cache *s, int node,
1242 int objects) {}
1243static inline void dec_slabs_node(struct kmem_cache *s, int node,
1244 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001245
1246static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1247 { return 0; }
1248
1249static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1250 void *object) {}
1251
1252static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1253
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001254#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001255
Christoph Lameter81819f02007-05-06 14:49:36 -07001256/*
1257 * Slab allocation and freeing
1258 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001259static inline struct page *alloc_slab_page(gfp_t flags, int node,
1260 struct kmem_cache_order_objects oo)
1261{
1262 int order = oo_order(oo);
1263
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001264 flags |= __GFP_NOTRACK;
1265
Christoph Lameter2154a332010-07-09 14:07:10 -05001266 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001267 return alloc_pages(flags, order);
1268 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001269 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001270}
1271
Christoph Lameter81819f02007-05-06 14:49:36 -07001272static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1273{
Pekka Enberg06428782008-01-07 23:20:27 -08001274 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001275 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001276 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001277
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001278 flags &= gfp_allowed_mask;
1279
1280 if (flags & __GFP_WAIT)
1281 local_irq_enable();
1282
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001283 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001284
Pekka Enbergba522702009-06-24 21:59:51 +03001285 /*
1286 * Let the initial higher-order allocation fail under memory pressure
1287 * so we fall-back to the minimum order allocation.
1288 */
1289 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1290
1291 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001292 if (unlikely(!page)) {
1293 oo = s->min;
1294 /*
1295 * Allocation may have failed due to fragmentation.
1296 * Try a lower order alloc if possible
1297 */
1298 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001299
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001300 if (page)
1301 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001302 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001303
David Rientjes737b7192012-07-09 14:00:38 -07001304 if (kmemcheck_enabled && page
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001305 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001306 int pages = 1 << oo_order(oo);
1307
1308 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1309
1310 /*
1311 * Objects from caches that have a constructor don't get
1312 * cleared when they're allocated, so we need to do it here.
1313 */
1314 if (s->ctor)
1315 kmemcheck_mark_uninitialized_pages(page, pages);
1316 else
1317 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001318 }
1319
David Rientjes737b7192012-07-09 14:00:38 -07001320 if (flags & __GFP_WAIT)
1321 local_irq_disable();
1322 if (!page)
1323 return NULL;
1324
Christoph Lameter834f3d12008-04-14 19:11:31 +03001325 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001326 mod_zone_page_state(page_zone(page),
1327 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1328 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001329 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001330
1331 return page;
1332}
1333
1334static void setup_object(struct kmem_cache *s, struct page *page,
1335 void *object)
1336{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001337 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001338 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001339 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001340}
1341
1342static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1343{
1344 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001345 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001346 void *last;
1347 void *p;
Glauber Costa1f458cb2012-12-18 14:22:50 -08001348 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001349
Christoph Lameter6cb06222007-10-16 01:25:41 -07001350 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001351
Christoph Lameter6cb06222007-10-16 01:25:41 -07001352 page = allocate_slab(s,
1353 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001354 if (!page)
1355 goto out;
1356
Glauber Costa1f458cb2012-12-18 14:22:50 -08001357 order = compound_order(page);
Christoph Lameter205ab992008-04-14 19:11:40 +03001358 inc_slabs_node(s, page_to_nid(page), page->objects);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001359 memcg_bind_pages(s, order);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001360 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001361 __SetPageSlab(page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07001362 if (page->pfmemalloc)
1363 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001364
1365 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001366
1367 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001368 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001369
1370 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001371 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001372 setup_object(s, page, last);
1373 set_freepointer(s, last, p);
1374 last = p;
1375 }
1376 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001377 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001378
1379 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001380 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001381 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001382out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001383 return page;
1384}
1385
1386static void __free_slab(struct kmem_cache *s, struct page *page)
1387{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001388 int order = compound_order(page);
1389 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001390
Christoph Lameteraf537b02010-07-09 14:07:14 -05001391 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001392 void *p;
1393
1394 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001395 for_each_object(p, s, page_address(page),
1396 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001397 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001398 }
1399
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001400 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001401
Christoph Lameter81819f02007-05-06 14:49:36 -07001402 mod_zone_page_state(page_zone(page),
1403 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1404 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001405 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001406
Mel Gorman072bb0a2012-07-31 16:43:58 -07001407 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001408 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001409
1410 memcg_release_pages(s, order);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001411 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001412 if (current->reclaim_state)
1413 current->reclaim_state->reclaimed_slab += pages;
Glauber Costad79923f2012-12-18 14:22:48 -08001414 __free_memcg_kmem_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001415}
1416
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001417#define need_reserve_slab_rcu \
1418 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1419
Christoph Lameter81819f02007-05-06 14:49:36 -07001420static void rcu_free_slab(struct rcu_head *h)
1421{
1422 struct page *page;
1423
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001424 if (need_reserve_slab_rcu)
1425 page = virt_to_head_page(h);
1426 else
1427 page = container_of((struct list_head *)h, struct page, lru);
1428
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001429 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001430}
1431
1432static void free_slab(struct kmem_cache *s, struct page *page)
1433{
1434 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001435 struct rcu_head *head;
1436
1437 if (need_reserve_slab_rcu) {
1438 int order = compound_order(page);
1439 int offset = (PAGE_SIZE << order) - s->reserved;
1440
1441 VM_BUG_ON(s->reserved != sizeof(*head));
1442 head = page_address(page) + offset;
1443 } else {
1444 /*
1445 * RCU free overloads the RCU head over the LRU
1446 */
1447 head = (void *)&page->lru;
1448 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001449
1450 call_rcu(head, rcu_free_slab);
1451 } else
1452 __free_slab(s, page);
1453}
1454
1455static void discard_slab(struct kmem_cache *s, struct page *page)
1456{
Christoph Lameter205ab992008-04-14 19:11:40 +03001457 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001458 free_slab(s, page);
1459}
1460
1461/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001462 * Management of partially allocated slabs.
1463 *
1464 * list_lock must be held.
Christoph Lameter81819f02007-05-06 14:49:36 -07001465 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001466static inline void add_partial(struct kmem_cache_node *n,
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001467 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001468{
Christoph Lametere95eed52007-05-06 14:49:44 -07001469 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001470 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001471 list_add_tail(&page->lru, &n->partial);
1472 else
1473 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001474}
1475
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001476/*
1477 * list_lock must be held.
1478 */
1479static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001480 struct page *page)
1481{
1482 list_del(&page->lru);
1483 n->nr_partial--;
1484}
1485
Christoph Lameter81819f02007-05-06 14:49:36 -07001486/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001487 * Remove slab from the partial list, freeze it and
1488 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001489 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001490 * Returns a list of objects or NULL if it fails.
1491 *
Christoph Lameter7ced3712012-05-09 10:09:53 -05001492 * Must hold list_lock since we modify the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001493 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001494static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001495 struct kmem_cache_node *n, struct page *page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001496 int mode)
Christoph Lameter81819f02007-05-06 14:49:36 -07001497{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001498 void *freelist;
1499 unsigned long counters;
1500 struct page new;
1501
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001502 /*
1503 * Zap the freelist and set the frozen bit.
1504 * The old freelist is the list of objects for the
1505 * per cpu allocation list.
1506 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001507 freelist = page->freelist;
1508 counters = page->counters;
1509 new.counters = counters;
Pekka Enberg23910c52012-06-04 10:14:58 +03001510 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001511 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001512 new.freelist = NULL;
1513 } else {
1514 new.freelist = freelist;
1515 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001516
Christoph Lameter7ced3712012-05-09 10:09:53 -05001517 VM_BUG_ON(new.frozen);
1518 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001519
Christoph Lameter7ced3712012-05-09 10:09:53 -05001520 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001521 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001522 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001523 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001524 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001525
1526 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001527 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001528 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001529}
1530
Christoph Lameter49e22582011-08-09 16:12:27 -05001531static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001532static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001533
Christoph Lameter81819f02007-05-06 14:49:36 -07001534/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001535 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001536 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001537static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1538 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001539{
Christoph Lameter49e22582011-08-09 16:12:27 -05001540 struct page *page, *page2;
1541 void *object = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001542
1543 /*
1544 * Racy check. If we mistakenly see no partial slabs then we
1545 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001546 * partial slab and there is none available then get_partials()
1547 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001548 */
1549 if (!n || !n->nr_partial)
1550 return NULL;
1551
1552 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001553 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001554 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001555 int available;
1556
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001557 if (!pfmemalloc_match(page, flags))
1558 continue;
1559
1560 t = acquire_slab(s, n, page, object == NULL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001561 if (!t)
1562 break;
1563
Alex,Shi12d79632011-09-07 10:26:36 +08001564 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001565 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001566 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001567 object = t;
1568 available = page->objects - page->inuse;
1569 } else {
Christoph Lameter49e22582011-08-09 16:12:27 -05001570 available = put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001571 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001572 }
1573 if (kmem_cache_debug(s) || available > s->cpu_partial / 2)
1574 break;
1575
Christoph Lameter497b66f2011-08-09 16:12:26 -05001576 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001577 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001578 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001579}
1580
1581/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001582 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001583 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001584static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001585 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001586{
1587#ifdef CONFIG_NUMA
1588 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001589 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001590 struct zone *zone;
1591 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001592 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001593 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001594
1595 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001596 * The defrag ratio allows a configuration of the tradeoffs between
1597 * inter node defragmentation and node local allocations. A lower
1598 * defrag_ratio increases the tendency to do local allocations
1599 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001600 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001601 * If the defrag_ratio is set to 0 then kmalloc() always
1602 * returns node local objects. If the ratio is higher then kmalloc()
1603 * may return off node objects because partial slabs are obtained
1604 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001605 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001606 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001607 * defrag_ratio = 1000) then every (well almost) allocation will
1608 * first attempt to defrag slab caches on other nodes. This means
1609 * scanning over all nodes to look for partial slabs which may be
1610 * expensive if we do it every time we are trying to find a slab
1611 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001612 */
Christoph Lameter98246012008-01-07 23:20:26 -08001613 if (!s->remote_node_defrag_ratio ||
1614 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001615 return NULL;
1616
Mel Gormancc9a6c82012-03-21 16:34:11 -07001617 do {
1618 cpuset_mems_cookie = get_mems_allowed();
Andi Kleene7b691b2012-06-09 02:40:03 -07001619 zonelist = node_zonelist(slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001620 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1621 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001622
Mel Gormancc9a6c82012-03-21 16:34:11 -07001623 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001624
Mel Gormancc9a6c82012-03-21 16:34:11 -07001625 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1626 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001627 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001628 if (object) {
1629 /*
1630 * Return the object even if
1631 * put_mems_allowed indicated that
1632 * the cpuset mems_allowed was
1633 * updated in parallel. It's a
1634 * harmless race between the alloc
1635 * and the cpuset update.
1636 */
1637 put_mems_allowed(cpuset_mems_cookie);
1638 return object;
1639 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001640 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001641 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07001642 } while (!put_mems_allowed(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001643#endif
1644 return NULL;
1645}
1646
1647/*
1648 * Get a partial page, lock it and return it.
1649 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001650static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001651 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001652{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001653 void *object;
Christoph Lameter2154a332010-07-09 14:07:10 -05001654 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001655
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001656 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001657 if (object || node != NUMA_NO_NODE)
1658 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001659
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001660 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001661}
1662
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001663#ifdef CONFIG_PREEMPT
1664/*
1665 * Calculate the next globally unique transaction for disambiguiation
1666 * during cmpxchg. The transactions start with the cpu number and are then
1667 * incremented by CONFIG_NR_CPUS.
1668 */
1669#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1670#else
1671/*
1672 * No preemption supported therefore also no need to check for
1673 * different cpus.
1674 */
1675#define TID_STEP 1
1676#endif
1677
1678static inline unsigned long next_tid(unsigned long tid)
1679{
1680 return tid + TID_STEP;
1681}
1682
1683static inline unsigned int tid_to_cpu(unsigned long tid)
1684{
1685 return tid % TID_STEP;
1686}
1687
1688static inline unsigned long tid_to_event(unsigned long tid)
1689{
1690 return tid / TID_STEP;
1691}
1692
1693static inline unsigned int init_tid(int cpu)
1694{
1695 return cpu;
1696}
1697
1698static inline void note_cmpxchg_failure(const char *n,
1699 const struct kmem_cache *s, unsigned long tid)
1700{
1701#ifdef SLUB_DEBUG_CMPXCHG
1702 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1703
1704 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1705
1706#ifdef CONFIG_PREEMPT
1707 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1708 printk("due to cpu change %d -> %d\n",
1709 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1710 else
1711#endif
1712 if (tid_to_event(tid) != tid_to_event(actual_tid))
1713 printk("due to cpu running other code. Event %ld->%ld\n",
1714 tid_to_event(tid), tid_to_event(actual_tid));
1715 else
1716 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1717 actual_tid, tid, next_tid(tid));
1718#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001719 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001720}
1721
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001722static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001723{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001724 int cpu;
1725
1726 for_each_possible_cpu(cpu)
1727 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001728}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001729
1730/*
1731 * Remove the cpu slab
1732 */
Christoph Lameterc17dda42012-05-09 10:09:57 -05001733static void deactivate_slab(struct kmem_cache *s, struct page *page, void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07001734{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001735 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001736 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1737 int lock = 0;
1738 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001739 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001740 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001741 struct page new;
1742 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001743
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001744 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001745 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001746 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001747 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001748
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001749 /*
1750 * Stage one: Free all available per cpu objects back
1751 * to the page freelist while it is still frozen. Leave the
1752 * last one.
1753 *
1754 * There is no need to take the list->lock because the page
1755 * is still frozen.
1756 */
1757 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1758 void *prior;
1759 unsigned long counters;
1760
1761 do {
1762 prior = page->freelist;
1763 counters = page->counters;
1764 set_freepointer(s, freelist, prior);
1765 new.counters = counters;
1766 new.inuse--;
1767 VM_BUG_ON(!new.frozen);
1768
Christoph Lameter1d071712011-07-14 12:49:12 -05001769 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001770 prior, counters,
1771 freelist, new.counters,
1772 "drain percpu freelist"));
1773
1774 freelist = nextfree;
1775 }
1776
1777 /*
1778 * Stage two: Ensure that the page is unfrozen while the
1779 * list presence reflects the actual number of objects
1780 * during unfreeze.
1781 *
1782 * We setup the list membership and then perform a cmpxchg
1783 * with the count. If there is a mismatch then the page
1784 * is not unfrozen but the page is on the wrong list.
1785 *
1786 * Then we restart the process which may have to remove
1787 * the page from the list that we just put it on again
1788 * because the number of objects in the slab may have
1789 * changed.
1790 */
1791redo:
1792
1793 old.freelist = page->freelist;
1794 old.counters = page->counters;
1795 VM_BUG_ON(!old.frozen);
1796
1797 /* Determine target state of the slab */
1798 new.counters = old.counters;
1799 if (freelist) {
1800 new.inuse--;
1801 set_freepointer(s, freelist, old.freelist);
1802 new.freelist = freelist;
1803 } else
1804 new.freelist = old.freelist;
1805
1806 new.frozen = 0;
1807
Christoph Lameter81107182011-08-09 13:01:32 -05001808 if (!new.inuse && n->nr_partial > s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001809 m = M_FREE;
1810 else if (new.freelist) {
1811 m = M_PARTIAL;
1812 if (!lock) {
1813 lock = 1;
1814 /*
1815 * Taking the spinlock removes the possiblity
1816 * that acquire_slab() will see a slab page that
1817 * is frozen
1818 */
1819 spin_lock(&n->list_lock);
1820 }
1821 } else {
1822 m = M_FULL;
1823 if (kmem_cache_debug(s) && !lock) {
1824 lock = 1;
1825 /*
1826 * This also ensures that the scanning of full
1827 * slabs from diagnostic functions will not see
1828 * any frozen slabs.
1829 */
1830 spin_lock(&n->list_lock);
1831 }
1832 }
1833
1834 if (l != m) {
1835
1836 if (l == M_PARTIAL)
1837
1838 remove_partial(n, page);
1839
1840 else if (l == M_FULL)
1841
1842 remove_full(s, page);
1843
1844 if (m == M_PARTIAL) {
1845
1846 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001847 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001848
1849 } else if (m == M_FULL) {
1850
1851 stat(s, DEACTIVATE_FULL);
1852 add_full(s, n, page);
1853
1854 }
1855 }
1856
1857 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001858 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001859 old.freelist, old.counters,
1860 new.freelist, new.counters,
1861 "unfreezing slab"))
1862 goto redo;
1863
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001864 if (lock)
1865 spin_unlock(&n->list_lock);
1866
1867 if (m == M_FREE) {
1868 stat(s, DEACTIVATE_EMPTY);
1869 discard_slab(s, page);
1870 stat(s, FREE_SLAB);
1871 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001872}
1873
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001874/*
1875 * Unfreeze all the cpu partial slabs.
1876 *
Christoph Lameter59a09912012-11-28 16:23:00 +00001877 * This function must be called with interrupts disabled
1878 * for the cpu using c (or some other guarantee must be there
1879 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001880 */
Christoph Lameter59a09912012-11-28 16:23:00 +00001881static void unfreeze_partials(struct kmem_cache *s,
1882 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05001883{
Joonsoo Kim43d77862012-06-09 02:23:16 +09001884 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08001885 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001886
1887 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001888 struct page new;
1889 struct page old;
1890
1891 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09001892
1893 n2 = get_node(s, page_to_nid(page));
1894 if (n != n2) {
1895 if (n)
1896 spin_unlock(&n->list_lock);
1897
1898 n = n2;
1899 spin_lock(&n->list_lock);
1900 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001901
1902 do {
1903
1904 old.freelist = page->freelist;
1905 old.counters = page->counters;
1906 VM_BUG_ON(!old.frozen);
1907
1908 new.counters = old.counters;
1909 new.freelist = old.freelist;
1910
1911 new.frozen = 0;
1912
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001913 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001914 old.freelist, old.counters,
1915 new.freelist, new.counters,
1916 "unfreezing slab"));
1917
Joonsoo Kim43d77862012-06-09 02:23:16 +09001918 if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08001919 page->next = discard_page;
1920 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09001921 } else {
1922 add_partial(n, page, DEACTIVATE_TO_TAIL);
1923 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001924 }
1925 }
1926
1927 if (n)
1928 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08001929
1930 while (discard_page) {
1931 page = discard_page;
1932 discard_page = discard_page->next;
1933
1934 stat(s, DEACTIVATE_EMPTY);
1935 discard_slab(s, page);
1936 stat(s, FREE_SLAB);
1937 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001938}
1939
1940/*
1941 * Put a page that was just frozen (in __slab_free) into a partial page
1942 * slot if available. This is done without interrupts disabled and without
1943 * preemption disabled. The cmpxchg is racy and may put the partial page
1944 * onto a random cpus partial slot.
1945 *
1946 * If we did not find a slot then simply move all the partials to the
1947 * per node partial list.
1948 */
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001949static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05001950{
1951 struct page *oldpage;
1952 int pages;
1953 int pobjects;
1954
1955 do {
1956 pages = 0;
1957 pobjects = 0;
1958 oldpage = this_cpu_read(s->cpu_slab->partial);
1959
1960 if (oldpage) {
1961 pobjects = oldpage->pobjects;
1962 pages = oldpage->pages;
1963 if (drain && pobjects > s->cpu_partial) {
1964 unsigned long flags;
1965 /*
1966 * partial array is full. Move the existing
1967 * set to the per node partial list.
1968 */
1969 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00001970 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05001971 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09001972 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001973 pobjects = 0;
1974 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08001975 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05001976 }
1977 }
1978
1979 pages++;
1980 pobjects += page->objects - page->inuse;
1981
1982 page->pages = pages;
1983 page->pobjects = pobjects;
1984 page->next = oldpage;
1985
Christoph Lameter933393f2011-12-22 11:58:51 -06001986 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage);
Christoph Lameter49e22582011-08-09 16:12:27 -05001987 return pobjects;
1988}
1989
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001990static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001991{
Christoph Lameter84e554e62009-12-18 16:26:23 -06001992 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05001993 deactivate_slab(s, c->page, c->freelist);
1994
1995 c->tid = next_tid(c->tid);
1996 c->page = NULL;
1997 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001998}
1999
2000/*
2001 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002002 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002003 * Called from IPI handler with interrupts disabled.
2004 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002005static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002006{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002007 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002008
Christoph Lameter49e22582011-08-09 16:12:27 -05002009 if (likely(c)) {
2010 if (c->page)
2011 flush_slab(s, c);
2012
Christoph Lameter59a09912012-11-28 16:23:00 +00002013 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002014 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002015}
2016
2017static void flush_cpu_slab(void *d)
2018{
2019 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002020
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002021 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002022}
2023
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002024static bool has_cpu_slab(int cpu, void *info)
2025{
2026 struct kmem_cache *s = info;
2027 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2028
majianpeng02e1a9c2012-05-17 17:03:26 -07002029 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002030}
2031
Christoph Lameter81819f02007-05-06 14:49:36 -07002032static void flush_all(struct kmem_cache *s)
2033{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002034 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002035}
2036
2037/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002038 * Check if the objects in a per cpu structure fit numa
2039 * locality expectations.
2040 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002041static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002042{
2043#ifdef CONFIG_NUMA
Christoph Lameter57d437d2012-05-09 10:09:59 -05002044 if (node != NUMA_NO_NODE && page_to_nid(page) != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002045 return 0;
2046#endif
2047 return 1;
2048}
2049
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002050static int count_free(struct page *page)
2051{
2052 return page->objects - page->inuse;
2053}
2054
2055static unsigned long count_partial(struct kmem_cache_node *n,
2056 int (*get_count)(struct page *))
2057{
2058 unsigned long flags;
2059 unsigned long x = 0;
2060 struct page *page;
2061
2062 spin_lock_irqsave(&n->list_lock, flags);
2063 list_for_each_entry(page, &n->partial, lru)
2064 x += get_count(page);
2065 spin_unlock_irqrestore(&n->list_lock, flags);
2066 return x;
2067}
2068
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002069static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2070{
2071#ifdef CONFIG_SLUB_DEBUG
2072 return atomic_long_read(&n->total_objects);
2073#else
2074 return 0;
2075#endif
2076}
2077
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002078static noinline void
2079slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2080{
2081 int node;
2082
2083 printk(KERN_WARNING
2084 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2085 nid, gfpflags);
2086 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002087 "default order: %d, min order: %d\n", s->name, s->object_size,
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002088 s->size, oo_order(s->oo), oo_order(s->min));
2089
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002090 if (oo_order(s->min) > get_order(s->object_size))
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002091 printk(KERN_WARNING " %s debugging increased min order, use "
2092 "slub_debug=O to disable.\n", s->name);
2093
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002094 for_each_online_node(node) {
2095 struct kmem_cache_node *n = get_node(s, node);
2096 unsigned long nr_slabs;
2097 unsigned long nr_objs;
2098 unsigned long nr_free;
2099
2100 if (!n)
2101 continue;
2102
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002103 nr_free = count_partial(n, count_free);
2104 nr_slabs = node_nr_slabs(n);
2105 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002106
2107 printk(KERN_WARNING
2108 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2109 node, nr_slabs, nr_objs, nr_free);
2110 }
2111}
2112
Christoph Lameter497b66f2011-08-09 16:12:26 -05002113static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2114 int node, struct kmem_cache_cpu **pc)
2115{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002116 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002117 struct kmem_cache_cpu *c = *pc;
2118 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002119
Christoph Lameter188fd062012-05-09 10:09:55 -05002120 freelist = get_partial(s, flags, node, c);
2121
2122 if (freelist)
2123 return freelist;
2124
2125 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002126 if (page) {
2127 c = __this_cpu_ptr(s->cpu_slab);
2128 if (c->page)
2129 flush_slab(s, c);
2130
2131 /*
2132 * No other reference to the page yet so we can
2133 * muck around with it freely without cmpxchg
2134 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002135 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002136 page->freelist = NULL;
2137
2138 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002139 c->page = page;
2140 *pc = c;
2141 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002142 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002143
Christoph Lameter6faa6832012-05-09 10:09:51 -05002144 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002145}
2146
Mel Gorman072bb0a2012-07-31 16:43:58 -07002147static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2148{
2149 if (unlikely(PageSlabPfmemalloc(page)))
2150 return gfp_pfmemalloc_allowed(gfpflags);
2151
2152 return true;
2153}
2154
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002155/*
Christoph Lameter213eeb92011-11-11 14:07:14 -06002156 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2157 * or deactivate the page.
2158 *
2159 * The page is still frozen if the return value is not NULL.
2160 *
2161 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002162 *
2163 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002164 */
2165static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2166{
2167 struct page new;
2168 unsigned long counters;
2169 void *freelist;
2170
2171 do {
2172 freelist = page->freelist;
2173 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002174
Christoph Lameter213eeb92011-11-11 14:07:14 -06002175 new.counters = counters;
2176 VM_BUG_ON(!new.frozen);
2177
2178 new.inuse = page->objects;
2179 new.frozen = freelist != NULL;
2180
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002181 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002182 freelist, counters,
2183 NULL, new.counters,
2184 "get_freelist"));
2185
2186 return freelist;
2187}
2188
2189/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002190 * Slow path. The lockless freelist is empty or we need to perform
2191 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002192 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002193 * Processing is still very fast if new objects have been freed to the
2194 * regular freelist. In that case we simply take over the regular freelist
2195 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002196 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002197 * If that is not working then we fall back to the partial lists. We take the
2198 * first element of the freelist as the object to allocate now and move the
2199 * rest of the freelist to the lockless freelist.
2200 *
2201 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002202 * we need to allocate a new slab. This is the slowest path since it involves
2203 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002204 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002205static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2206 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002207{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002208 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002209 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002210 unsigned long flags;
2211
2212 local_irq_save(flags);
2213#ifdef CONFIG_PREEMPT
2214 /*
2215 * We may have been preempted and rescheduled on a different
2216 * cpu before disabling interrupts. Need to reload cpu area
2217 * pointer.
2218 */
2219 c = this_cpu_ptr(s->cpu_slab);
2220#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002221
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002222 page = c->page;
2223 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002224 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002225redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002226
Christoph Lameter57d437d2012-05-09 10:09:59 -05002227 if (unlikely(!node_match(page, node))) {
Christoph Lametere36a2652011-06-01 12:25:57 -05002228 stat(s, ALLOC_NODE_MISMATCH);
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002229 deactivate_slab(s, page, c->freelist);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002230 c->page = NULL;
2231 c->freelist = NULL;
Christoph Lameterfc59c052011-06-01 12:25:56 -05002232 goto new_slab;
2233 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002234
Mel Gorman072bb0a2012-07-31 16:43:58 -07002235 /*
2236 * By rights, we should be searching for a slab page that was
2237 * PFMEMALLOC but right now, we are losing the pfmemalloc
2238 * information when the page leaves the per-cpu allocator
2239 */
2240 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2241 deactivate_slab(s, page, c->freelist);
2242 c->page = NULL;
2243 c->freelist = NULL;
2244 goto new_slab;
2245 }
2246
Eric Dumazet73736e02011-12-13 04:57:06 +01002247 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002248 freelist = c->freelist;
2249 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002250 goto load_freelist;
2251
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002252 stat(s, ALLOC_SLOWPATH);
2253
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002254 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002255
Christoph Lameter6faa6832012-05-09 10:09:51 -05002256 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002257 c->page = NULL;
2258 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002259 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002260 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002261
Christoph Lameter81819f02007-05-06 14:49:36 -07002262 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002263
Christoph Lameter894b8782007-05-10 03:15:16 -07002264load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002265 /*
2266 * freelist is pointing to the list of objects to be used.
2267 * page is pointing to the page from which the objects are obtained.
2268 * That page must be frozen for per cpu allocations to work.
2269 */
2270 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002271 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002272 c->tid = next_tid(c->tid);
2273 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002274 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002275
Christoph Lameter81819f02007-05-06 14:49:36 -07002276new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002277
Christoph Lameter49e22582011-08-09 16:12:27 -05002278 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002279 page = c->page = c->partial;
2280 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002281 stat(s, CPU_PARTIAL_ALLOC);
2282 c->freelist = NULL;
2283 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002284 }
2285
Christoph Lameter188fd062012-05-09 10:09:55 -05002286 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002287
Christoph Lameterf46974362012-05-09 10:09:54 -05002288 if (unlikely(!freelist)) {
2289 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
2290 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002291
Christoph Lameterf46974362012-05-09 10:09:54 -05002292 local_irq_restore(flags);
2293 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002294 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002295
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002296 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002297 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002298 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002299
Christoph Lameter497b66f2011-08-09 16:12:26 -05002300 /* Only entered in the debug case */
Christoph Lameter5091b742012-07-31 16:44:00 -07002301 if (kmem_cache_debug(s) && !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002302 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002303
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002304 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002305 c->page = NULL;
2306 c->freelist = NULL;
Christoph Lametera71ae472011-05-25 09:47:43 -05002307 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002308 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002309}
2310
2311/*
2312 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2313 * have the fastpath folded into their functions. So no function call
2314 * overhead for requests that can be satisfied on the fastpath.
2315 *
2316 * The fastpath works by first checking if the lockless freelist can be used.
2317 * If not then __slab_alloc is called for slow processing.
2318 *
2319 * Otherwise we can simply pick the next object from the lockless free list.
2320 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002321static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002322 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002323{
Christoph Lameter894b8782007-05-10 03:15:16 -07002324 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002325 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002326 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002327 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002328
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002329 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09002330 return NULL;
2331
Glauber Costad79923f2012-12-18 14:22:48 -08002332 s = memcg_kmem_get_cache(s, gfpflags);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002333redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002334
2335 /*
2336 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2337 * enabled. We may switch back and forth between cpus while
2338 * reading from one cpu area. That does not matter as long
2339 * as we end up on the original cpu again when doing the cmpxchg.
2340 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002341 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002342
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002343 /*
2344 * The transaction ids are globally unique per cpu and per operation on
2345 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2346 * occurs on the right processor and that there was no operation on the
2347 * linked list in between.
2348 */
2349 tid = c->tid;
2350 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002351
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002352 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002353 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002354 if (unlikely(!object || !node_match(page, node)))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002355 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07002356
2357 else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002358 void *next_object = get_freepointer_safe(s, object);
2359
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002360 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002361 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002362 * operation and if we are on the right processor.
2363 *
2364 * The cmpxchg does the following atomically (without lock semantics!)
2365 * 1. Relocate first pointer to the current per cpu area.
2366 * 2. Verify that tid and freelist have not been changed
2367 * 3. If they were not changed replace tid and freelist
2368 *
2369 * Since this is without lock semantics the protection is only against
2370 * code executing on this cpu *not* from access by other cpus.
2371 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002372 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002373 s->cpu_slab->freelist, s->cpu_slab->tid,
2374 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002375 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002376
2377 note_cmpxchg_failure("slab_alloc", s, tid);
2378 goto redo;
2379 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002380 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002381 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002382 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002383
Pekka Enberg74e21342009-11-25 20:14:48 +02002384 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002385 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002386
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002387 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002388
Christoph Lameter894b8782007-05-10 03:15:16 -07002389 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002390}
2391
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002392static __always_inline void *slab_alloc(struct kmem_cache *s,
2393 gfp_t gfpflags, unsigned long addr)
2394{
2395 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2396}
2397
Christoph Lameter81819f02007-05-06 14:49:36 -07002398void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2399{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002400 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002401
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002402 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002403
2404 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002405}
2406EXPORT_SYMBOL(kmem_cache_alloc);
2407
Li Zefan0f24f122009-12-11 15:45:30 +08002408#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002409void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002410{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002411 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002412 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2413 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002414}
Richard Kennedy4a923792010-10-21 10:29:19 +01002415EXPORT_SYMBOL(kmem_cache_alloc_trace);
2416
2417void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2418{
2419 void *ret = kmalloc_order(size, flags, order);
2420 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2421 return ret;
2422}
2423EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002424#endif
2425
Christoph Lameter81819f02007-05-06 14:49:36 -07002426#ifdef CONFIG_NUMA
2427void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2428{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002429 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002430
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002431 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002432 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002433
2434 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002435}
2436EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002437
Li Zefan0f24f122009-12-11 15:45:30 +08002438#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002439void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002440 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002441 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002442{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002443 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002444
2445 trace_kmalloc_node(_RET_IP_, ret,
2446 size, s->size, gfpflags, node);
2447 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002448}
Richard Kennedy4a923792010-10-21 10:29:19 +01002449EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002450#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002451#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002452
Christoph Lameter81819f02007-05-06 14:49:36 -07002453/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002454 * Slow patch handling. This may still be called frequently since objects
2455 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002456 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002457 * So we still attempt to reduce cache line usage. Just take the slab
2458 * lock and free the item. If there is no additional partial page
2459 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002460 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002461static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002462 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002463{
2464 void *prior;
2465 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002466 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002467 struct page new;
2468 unsigned long counters;
2469 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002470 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002471
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002472 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002473
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002474 if (kmem_cache_debug(s) &&
2475 !(n = free_debug_processing(s, page, x, addr, &flags)))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002476 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002477
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002478 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002479 if (unlikely(n)) {
2480 spin_unlock_irqrestore(&n->list_lock, flags);
2481 n = NULL;
2482 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002483 prior = page->freelist;
2484 counters = page->counters;
2485 set_freepointer(s, object, prior);
2486 new.counters = counters;
2487 was_frozen = new.frozen;
2488 new.inuse--;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002489 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002490
2491 if (!kmem_cache_debug(s) && !prior)
2492
2493 /*
2494 * Slab was on no list before and will be partially empty
2495 * We can defer the list move and instead freeze it.
2496 */
2497 new.frozen = 1;
2498
2499 else { /* Needs to be taken off a list */
2500
2501 n = get_node(s, page_to_nid(page));
2502 /*
2503 * Speculatively acquire the list_lock.
2504 * If the cmpxchg does not succeed then we may
2505 * drop the list_lock without any processing.
2506 *
2507 * Otherwise the list_lock will synchronize with
2508 * other processors updating the list of slabs.
2509 */
2510 spin_lock_irqsave(&n->list_lock, flags);
2511
2512 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002513 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002514
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002515 } while (!cmpxchg_double_slab(s, page,
2516 prior, counters,
2517 object, new.counters,
2518 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002519
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002520 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002521
2522 /*
2523 * If we just froze the page then put it onto the
2524 * per cpu partial list.
2525 */
Alex Shi8028dce2012-02-03 23:34:56 +08002526 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002527 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002528 stat(s, CPU_PARTIAL_FREE);
2529 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002530 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002531 * The list lock was not taken therefore no list
2532 * activity can be necessary.
2533 */
2534 if (was_frozen)
2535 stat(s, FREE_FROZEN);
Christoph Lameter80f08c12011-06-01 12:25:55 -05002536 return;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002537 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002538
Joonsoo Kim837d6782012-08-16 00:02:40 +09002539 if (unlikely(!new.inuse && n->nr_partial > s->min_partial))
2540 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002541
Joonsoo Kim837d6782012-08-16 00:02:40 +09002542 /*
2543 * Objects left in the slab. If it was not on the partial list before
2544 * then add it.
2545 */
2546 if (kmem_cache_debug(s) && unlikely(!prior)) {
2547 remove_full(s, page);
2548 add_partial(n, page, DEACTIVATE_TO_TAIL);
2549 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002550 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002551 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002552 return;
2553
2554slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002555 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002556 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002557 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002558 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002559 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002560 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002561 } else
2562 /* Slab must be on the full list */
2563 remove_full(s, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002564
Christoph Lameter80f08c12011-06-01 12:25:55 -05002565 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002566 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002567 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002568}
2569
Christoph Lameter894b8782007-05-10 03:15:16 -07002570/*
2571 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2572 * can perform fastpath freeing without additional function calls.
2573 *
2574 * The fastpath is only possible if we are freeing to the current cpu slab
2575 * of this processor. This typically the case if we have just allocated
2576 * the item before.
2577 *
2578 * If fastpath is not possible then fall back to __slab_free where we deal
2579 * with all sorts of special processing.
2580 */
Pekka Enberg06428782008-01-07 23:20:27 -08002581static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002582 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002583{
2584 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002585 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002586 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002587
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002588 slab_free_hook(s, x);
2589
Christoph Lametera24c5a02011-03-15 12:45:21 -05002590redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002591 /*
2592 * Determine the currently cpus per cpu slab.
2593 * The cpu may change afterward. However that does not matter since
2594 * data is retrieved via this pointer. If we are on the same cpu
2595 * during the cmpxchg then the free will succedd.
2596 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002597 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002598
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002599 tid = c->tid;
2600 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002601
Christoph Lameter442b06b2011-05-17 16:29:31 -05002602 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002603 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002604
Christoph Lameter933393f2011-12-22 11:58:51 -06002605 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002606 s->cpu_slab->freelist, s->cpu_slab->tid,
2607 c->freelist, tid,
2608 object, next_tid(tid)))) {
2609
2610 note_cmpxchg_failure("slab_free", s, tid);
2611 goto redo;
2612 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002613 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002614 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002615 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002616
Christoph Lameter894b8782007-05-10 03:15:16 -07002617}
2618
Christoph Lameter81819f02007-05-06 14:49:36 -07002619void kmem_cache_free(struct kmem_cache *s, void *x)
2620{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002621 s = cache_from_obj(s, x);
2622 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00002623 return;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002624 slab_free(s, virt_to_head_page(x), x, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002625 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002626}
2627EXPORT_SYMBOL(kmem_cache_free);
2628
Christoph Lameter81819f02007-05-06 14:49:36 -07002629/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002630 * Object placement in a slab is made very easy because we always start at
2631 * offset 0. If we tune the size of the object to the alignment then we can
2632 * get the required alignment by putting one properly sized object after
2633 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002634 *
2635 * Notice that the allocation order determines the sizes of the per cpu
2636 * caches. Each processor has always one slab available for allocations.
2637 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002638 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002639 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002640 */
2641
2642/*
2643 * Mininum / Maximum order of slab pages. This influences locking overhead
2644 * and slab fragmentation. A higher order reduces the number of partial slabs
2645 * and increases the number of allocations possible without having to
2646 * take the list_lock.
2647 */
2648static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002649static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002650static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002651
2652/*
2653 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002654 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002655 */
2656static int slub_nomerge;
2657
2658/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002659 * Calculate the order of allocation given an slab object size.
2660 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002661 * The order of allocation has significant impact on performance and other
2662 * system components. Generally order 0 allocations should be preferred since
2663 * order 0 does not cause fragmentation in the page allocator. Larger objects
2664 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002665 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002666 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002667 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002668 * In order to reach satisfactory performance we must ensure that a minimum
2669 * number of objects is in one slab. Otherwise we may generate too much
2670 * activity on the partial lists which requires taking the list_lock. This is
2671 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002672 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002673 * slub_max_order specifies the order where we begin to stop considering the
2674 * number of objects in a slab as critical. If we reach slub_max_order then
2675 * we try to keep the page order as low as possible. So we accept more waste
2676 * of space in favor of a small page order.
2677 *
2678 * Higher order allocations also allow the placement of more objects in a
2679 * slab and thereby reduce object handling overhead. If the user has
2680 * requested a higher mininum order then we start with that one instead of
2681 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002682 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002683static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002684 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002685{
2686 int order;
2687 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002688 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002689
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002690 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002691 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002692
Christoph Lameter6300ea72007-07-17 04:03:20 -07002693 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002694 fls(min_objects * size - 1) - PAGE_SHIFT);
2695 order <= max_order; order++) {
2696
Christoph Lameter81819f02007-05-06 14:49:36 -07002697 unsigned long slab_size = PAGE_SIZE << order;
2698
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002699 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002700 continue;
2701
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002702 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002703
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002704 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002705 break;
2706
2707 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002708
Christoph Lameter81819f02007-05-06 14:49:36 -07002709 return order;
2710}
2711
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002712static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002713{
2714 int order;
2715 int min_objects;
2716 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002717 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002718
2719 /*
2720 * Attempt to find best configuration for a slab. This
2721 * works by first attempting to generate a layout with
2722 * the best configuration and backing off gradually.
2723 *
2724 * First we reduce the acceptable waste in a slab. Then
2725 * we reduce the minimum objects required in a slab.
2726 */
2727 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002728 if (!min_objects)
2729 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002730 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002731 min_objects = min(min_objects, max_objects);
2732
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002733 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002734 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002735 while (fraction >= 4) {
2736 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002737 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002738 if (order <= slub_max_order)
2739 return order;
2740 fraction /= 2;
2741 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002742 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002743 }
2744
2745 /*
2746 * We were unable to place multiple objects in a slab. Now
2747 * lets see if we can place a single object there.
2748 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002749 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002750 if (order <= slub_max_order)
2751 return order;
2752
2753 /*
2754 * Doh this slab cannot be placed using slub_max_order.
2755 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002756 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002757 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002758 return order;
2759 return -ENOSYS;
2760}
2761
Pekka Enberg5595cff2008-08-05 09:28:47 +03002762static void
Joonsoo Kim40534972012-05-11 00:50:47 +09002763init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002764{
2765 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002766 spin_lock_init(&n->list_lock);
2767 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002768#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002769 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002770 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002771 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002772#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002773}
2774
Christoph Lameter55136592010-08-20 12:37:13 -05002775static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002776{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002777 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2778 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002779
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002780 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002781 * Must align to double word boundary for the double cmpxchg
2782 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002783 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002784 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2785 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002786
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002787 if (!s->cpu_slab)
2788 return 0;
2789
2790 init_kmem_cache_cpus(s);
2791
2792 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002793}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002794
Christoph Lameter51df1142010-08-20 12:37:15 -05002795static struct kmem_cache *kmem_cache_node;
2796
Christoph Lameter81819f02007-05-06 14:49:36 -07002797/*
2798 * No kmalloc_node yet so do it by hand. We know that this is the first
2799 * slab on the node for this slabcache. There are no concurrent accesses
2800 * possible.
2801 *
2802 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002803 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2804 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002805 */
Christoph Lameter55136592010-08-20 12:37:13 -05002806static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002807{
2808 struct page *page;
2809 struct kmem_cache_node *n;
2810
Christoph Lameter51df1142010-08-20 12:37:15 -05002811 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002812
Christoph Lameter51df1142010-08-20 12:37:15 -05002813 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002814
2815 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002816 if (page_to_nid(page) != node) {
2817 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2818 "node %d\n", node);
2819 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2820 "in order to be able to continue\n");
2821 }
2822
Christoph Lameter81819f02007-05-06 14:49:36 -07002823 n = page->freelist;
2824 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002825 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05002826 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002827 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002828 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002829#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002830 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002831 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002832#endif
Joonsoo Kim40534972012-05-11 00:50:47 +09002833 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002834 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002835
Shaohua Li136333d2011-08-24 08:57:52 +08002836 add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002837}
2838
2839static void free_kmem_cache_nodes(struct kmem_cache *s)
2840{
2841 int node;
2842
Christoph Lameterf64dc582007-10-16 01:25:33 -07002843 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002844 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002845
Alexander Duyck73367bd2010-05-21 14:41:35 -07002846 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002847 kmem_cache_free(kmem_cache_node, n);
2848
Christoph Lameter81819f02007-05-06 14:49:36 -07002849 s->node[node] = NULL;
2850 }
2851}
2852
Christoph Lameter55136592010-08-20 12:37:13 -05002853static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002854{
2855 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002856
Christoph Lameterf64dc582007-10-16 01:25:33 -07002857 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002858 struct kmem_cache_node *n;
2859
Alexander Duyck73367bd2010-05-21 14:41:35 -07002860 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002861 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002862 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002863 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002864 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002865 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002866
2867 if (!n) {
2868 free_kmem_cache_nodes(s);
2869 return 0;
2870 }
2871
Christoph Lameter81819f02007-05-06 14:49:36 -07002872 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09002873 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002874 }
2875 return 1;
2876}
Christoph Lameter81819f02007-05-06 14:49:36 -07002877
David Rientjesc0bdb232009-02-25 09:16:35 +02002878static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002879{
2880 if (min < MIN_PARTIAL)
2881 min = MIN_PARTIAL;
2882 else if (min > MAX_PARTIAL)
2883 min = MAX_PARTIAL;
2884 s->min_partial = min;
2885}
2886
Christoph Lameter81819f02007-05-06 14:49:36 -07002887/*
2888 * calculate_sizes() determines the order and the distribution of data within
2889 * a slab object.
2890 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002891static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002892{
2893 unsigned long flags = s->flags;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002894 unsigned long size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002895 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002896
2897 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002898 * Round up object size to the next word boundary. We can only
2899 * place the free pointer at word boundaries and this determines
2900 * the possible location of the free pointer.
2901 */
2902 size = ALIGN(size, sizeof(void *));
2903
2904#ifdef CONFIG_SLUB_DEBUG
2905 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002906 * Determine if we can poison the object itself. If the user of
2907 * the slab may touch the object after free or before allocation
2908 * then we should never poison the object itself.
2909 */
2910 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002911 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002912 s->flags |= __OBJECT_POISON;
2913 else
2914 s->flags &= ~__OBJECT_POISON;
2915
Christoph Lameter81819f02007-05-06 14:49:36 -07002916
2917 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002918 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002919 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002920 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002921 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002922 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07002923 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002924#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002925
2926 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002927 * With that we have determined the number of bytes in actual use
2928 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002929 */
2930 s->inuse = size;
2931
2932 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002933 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002934 /*
2935 * Relocate free pointer after the object if it is not
2936 * permitted to overwrite the first word of the object on
2937 * kmem_cache_free.
2938 *
2939 * This is the case if we do RCU, have a constructor or
2940 * destructor or are poisoning the objects.
2941 */
2942 s->offset = size;
2943 size += sizeof(void *);
2944 }
2945
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002946#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002947 if (flags & SLAB_STORE_USER)
2948 /*
2949 * Need to store information about allocs and frees after
2950 * the object.
2951 */
2952 size += 2 * sizeof(struct track);
2953
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002954 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002955 /*
2956 * Add some empty padding so that we can catch
2957 * overwrites from earlier objects rather than let
2958 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002959 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002960 * of the object.
2961 */
2962 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002963#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002964
Christoph Lameter81819f02007-05-06 14:49:36 -07002965 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002966 * SLUB stores one object immediately after another beginning from
2967 * offset 0. In order to align the objects we have to simply size
2968 * each object to conform to the alignment.
2969 */
Christoph Lameter45906852012-11-28 16:23:16 +00002970 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07002971 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002972 if (forced_order >= 0)
2973 order = forced_order;
2974 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002975 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002976
Christoph Lameter834f3d12008-04-14 19:11:31 +03002977 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002978 return 0;
2979
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002980 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002981 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002982 s->allocflags |= __GFP_COMP;
2983
2984 if (s->flags & SLAB_CACHE_DMA)
2985 s->allocflags |= SLUB_DMA;
2986
2987 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2988 s->allocflags |= __GFP_RECLAIMABLE;
2989
Christoph Lameter81819f02007-05-06 14:49:36 -07002990 /*
2991 * Determine the number of objects per slab
2992 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002993 s->oo = oo_make(order, size, s->reserved);
2994 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002995 if (oo_objects(s->oo) > oo_objects(s->max))
2996 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002997
Christoph Lameter834f3d12008-04-14 19:11:31 +03002998 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002999}
3000
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003001static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003002{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003003 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003004 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003005
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003006 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3007 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003008
Christoph Lameter06b285d2008-04-14 19:11:41 +03003009 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003010 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003011 if (disable_higher_order_debug) {
3012 /*
3013 * Disable debugging flags that store metadata if the min slab
3014 * order increased.
3015 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003016 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003017 s->flags &= ~DEBUG_METADATA_FLAGS;
3018 s->offset = 0;
3019 if (!calculate_sizes(s, -1))
3020 goto error;
3021 }
3022 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003023
Heiko Carstens25654092012-01-12 17:17:33 -08003024#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3025 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003026 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
3027 /* Enable fast mode */
3028 s->flags |= __CMPXCHG_DOUBLE;
3029#endif
3030
David Rientjes3b89d7d2009-02-22 17:40:07 -08003031 /*
3032 * The larger the object size is, the more pages we want on the partial
3033 * list to avoid pounding the page allocator excessively.
3034 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003035 set_min_partial(s, ilog2(s->size) / 2);
3036
3037 /*
3038 * cpu_partial determined the maximum number of objects kept in the
3039 * per cpu partial lists of a processor.
3040 *
3041 * Per cpu partial lists mainly contain slabs that just have one
3042 * object freed. If they are used for allocation then they can be
3043 * filled up again with minimal effort. The slab will never hit the
3044 * per node partial lists and therefore no locking will be required.
3045 *
3046 * This setting also determines
3047 *
3048 * A) The number of objects from per cpu partial slabs dumped to the
3049 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003050 * B) The number of objects in cpu partial slabs to extract from the
Christoph Lameter49e22582011-08-09 16:12:27 -05003051 * per node list when we run out of per cpu objects. We only fetch 50%
3052 * to keep some capacity around for frees.
3053 */
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003054 if (kmem_cache_debug(s))
3055 s->cpu_partial = 0;
3056 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003057 s->cpu_partial = 2;
3058 else if (s->size >= 1024)
3059 s->cpu_partial = 6;
3060 else if (s->size >= 256)
3061 s->cpu_partial = 13;
3062 else
3063 s->cpu_partial = 30;
3064
Christoph Lameter81819f02007-05-06 14:49:36 -07003065#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003066 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003067#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003068 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003069 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003070
Christoph Lameter55136592010-08-20 12:37:13 -05003071 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003072 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003073
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003074 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003075error:
3076 if (flags & SLAB_PANIC)
3077 panic("Cannot create slab %s size=%lu realsize=%u "
3078 "order=%u offset=%u flags=%lx\n",
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003079 s->name, (unsigned long)s->size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07003080 s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003081 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003082}
Christoph Lameter81819f02007-05-06 14:49:36 -07003083
Christoph Lameter33b12c32008-04-25 12:22:43 -07003084static void list_slab_objects(struct kmem_cache *s, struct page *page,
3085 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003086{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003087#ifdef CONFIG_SLUB_DEBUG
3088 void *addr = page_address(page);
3089 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003090 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3091 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003092 if (!map)
3093 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003094 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003095 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003096
Christoph Lameter5f80b132011-04-15 14:48:13 -05003097 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003098 for_each_object(p, s, addr, page->objects) {
3099
3100 if (!test_bit(slab_index(p, s, addr), map)) {
3101 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
3102 p, p - addr);
3103 print_tracking(s, p);
3104 }
3105 }
3106 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003107 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003108#endif
3109}
3110
Christoph Lameter81819f02007-05-06 14:49:36 -07003111/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003112 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003113 * This is called from kmem_cache_close(). We must be the last thread
3114 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003115 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003116static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003117{
Christoph Lameter81819f02007-05-06 14:49:36 -07003118 struct page *page, *h;
3119
Christoph Lameter33b12c32008-04-25 12:22:43 -07003120 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003121 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003122 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003123 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003124 } else {
3125 list_slab_objects(s, page,
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003126 "Objects remaining in %s on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003127 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003128 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003129}
3130
3131/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003132 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003133 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003134static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003135{
3136 int node;
3137
3138 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003139 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07003140 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003141 struct kmem_cache_node *n = get_node(s, node);
3142
Christoph Lameter599870b2008-04-23 12:36:52 -07003143 free_partial(s, n);
3144 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003145 return 1;
3146 }
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003147 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003148 free_kmem_cache_nodes(s);
3149 return 0;
3150}
3151
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003152int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003153{
Christoph Lameter12c36672012-09-04 23:38:33 +00003154 int rc = kmem_cache_close(s);
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003155
Glauber Costa5413dfb2012-12-18 14:23:13 -08003156 if (!rc) {
3157 /*
3158 * We do the same lock strategy around sysfs_slab_add, see
3159 * __kmem_cache_create. Because this is pretty much the last
3160 * operation we do and the lock will be released shortly after
3161 * that in slab_common.c, we could just move sysfs_slab_remove
3162 * to a later point in common code. We should do that when we
3163 * have a common sysfs framework for all allocators.
3164 */
3165 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07003166 sysfs_slab_remove(s);
Glauber Costa5413dfb2012-12-18 14:23:13 -08003167 mutex_lock(&slab_mutex);
3168 }
Christoph Lameter12c36672012-09-04 23:38:33 +00003169
3170 return rc;
Christoph Lameter81819f02007-05-06 14:49:36 -07003171}
Christoph Lameter81819f02007-05-06 14:49:36 -07003172
3173/********************************************************************
3174 * Kmalloc subsystem
3175 *******************************************************************/
3176
Christoph Lameter51df1142010-08-20 12:37:15 -05003177struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07003178EXPORT_SYMBOL(kmalloc_caches);
3179
Christoph Lameter55136592010-08-20 12:37:13 -05003180#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003181static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05003182#endif
3183
Christoph Lameter81819f02007-05-06 14:49:36 -07003184static int __init setup_slub_min_order(char *str)
3185{
Pekka Enberg06428782008-01-07 23:20:27 -08003186 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003187
3188 return 1;
3189}
3190
3191__setup("slub_min_order=", setup_slub_min_order);
3192
3193static int __init setup_slub_max_order(char *str)
3194{
Pekka Enberg06428782008-01-07 23:20:27 -08003195 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003196 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003197
3198 return 1;
3199}
3200
3201__setup("slub_max_order=", setup_slub_max_order);
3202
3203static int __init setup_slub_min_objects(char *str)
3204{
Pekka Enberg06428782008-01-07 23:20:27 -08003205 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003206
3207 return 1;
3208}
3209
3210__setup("slub_min_objects=", setup_slub_min_objects);
3211
3212static int __init setup_slub_nomerge(char *str)
3213{
3214 slub_nomerge = 1;
3215 return 1;
3216}
3217
3218__setup("slub_nomerge", setup_slub_nomerge);
3219
Christoph Lameterf1b26332007-07-17 04:03:26 -07003220/*
3221 * Conversion table for small slabs sizes / 8 to the index in the
3222 * kmalloc array. This is necessary for slabs < 192 since we have non power
3223 * of two cache sizes there. The size of larger slabs can be determined using
3224 * fls.
3225 */
3226static s8 size_index[24] = {
3227 3, /* 8 */
3228 4, /* 16 */
3229 5, /* 24 */
3230 5, /* 32 */
3231 6, /* 40 */
3232 6, /* 48 */
3233 6, /* 56 */
3234 6, /* 64 */
3235 1, /* 72 */
3236 1, /* 80 */
3237 1, /* 88 */
3238 1, /* 96 */
3239 7, /* 104 */
3240 7, /* 112 */
3241 7, /* 120 */
3242 7, /* 128 */
3243 2, /* 136 */
3244 2, /* 144 */
3245 2, /* 152 */
3246 2, /* 160 */
3247 2, /* 168 */
3248 2, /* 176 */
3249 2, /* 184 */
3250 2 /* 192 */
3251};
3252
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003253static inline int size_index_elem(size_t bytes)
3254{
3255 return (bytes - 1) / 8;
3256}
3257
Christoph Lameter81819f02007-05-06 14:49:36 -07003258static struct kmem_cache *get_slab(size_t size, gfp_t flags)
3259{
Christoph Lameterf1b26332007-07-17 04:03:26 -07003260 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07003261
Christoph Lameterf1b26332007-07-17 04:03:26 -07003262 if (size <= 192) {
3263 if (!size)
3264 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07003265
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003266 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003267 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07003268 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003269
3270#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07003271 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05003272 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07003273
Christoph Lameter81819f02007-05-06 14:49:36 -07003274#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003275 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07003276}
3277
3278void *__kmalloc(size_t size, gfp_t flags)
3279{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003280 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003281 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003282
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003283 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003284 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003285
3286 s = get_slab(size, flags);
3287
3288 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003289 return s;
3290
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003291 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003292
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003293 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003294
3295 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003296}
3297EXPORT_SYMBOL(__kmalloc);
3298
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003299#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003300static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3301{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003302 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003303 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003304
Glauber Costad79923f2012-12-18 14:22:48 -08003305 flags |= __GFP_COMP | __GFP_NOTRACK | __GFP_KMEMCG;
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003306 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003307 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003308 ptr = page_address(page);
3309
3310 kmemleak_alloc(ptr, size, 1, flags);
3311 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003312}
3313
Christoph Lameter81819f02007-05-06 14:49:36 -07003314void *__kmalloc_node(size_t size, gfp_t flags, int node)
3315{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003316 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003317 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003318
Ingo Molnar057685c2009-02-20 12:15:30 +01003319 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003320 ret = kmalloc_large_node(size, flags, node);
3321
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003322 trace_kmalloc_node(_RET_IP_, ret,
3323 size, PAGE_SIZE << get_order(size),
3324 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003325
3326 return ret;
3327 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003328
3329 s = get_slab(size, flags);
3330
3331 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003332 return s;
3333
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003334 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003335
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003336 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003337
3338 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003339}
3340EXPORT_SYMBOL(__kmalloc_node);
3341#endif
3342
3343size_t ksize(const void *object)
3344{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003345 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003346
Christoph Lameteref8b4522007-10-16 01:24:46 -07003347 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003348 return 0;
3349
Vegard Nossum294a80a2007-12-04 23:45:30 -08003350 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003351
Pekka Enberg76994412008-05-22 19:22:25 +03003352 if (unlikely(!PageSlab(page))) {
3353 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003354 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003355 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003356
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003357 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003358}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003359EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003360
Ben Greeard18a90d2011-07-07 11:36:37 -07003361#ifdef CONFIG_SLUB_DEBUG
3362bool verify_mem_not_deleted(const void *x)
3363{
3364 struct page *page;
3365 void *object = (void *)x;
3366 unsigned long flags;
3367 bool rv;
3368
3369 if (unlikely(ZERO_OR_NULL_PTR(x)))
3370 return false;
3371
3372 local_irq_save(flags);
3373
3374 page = virt_to_head_page(x);
3375 if (unlikely(!PageSlab(page))) {
3376 /* maybe it was from stack? */
3377 rv = true;
3378 goto out_unlock;
3379 }
3380
3381 slab_lock(page);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003382 if (on_freelist(page->slab_cache, page, object)) {
3383 object_err(page->slab_cache, page, object, "Object is on free-list");
Ben Greeard18a90d2011-07-07 11:36:37 -07003384 rv = false;
3385 } else {
3386 rv = true;
3387 }
3388 slab_unlock(page);
3389
3390out_unlock:
3391 local_irq_restore(flags);
3392 return rv;
3393}
3394EXPORT_SYMBOL(verify_mem_not_deleted);
3395#endif
3396
Christoph Lameter81819f02007-05-06 14:49:36 -07003397void kfree(const void *x)
3398{
Christoph Lameter81819f02007-05-06 14:49:36 -07003399 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003400 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003401
Pekka Enberg2121db72009-03-25 11:05:57 +02003402 trace_kfree(_RET_IP_, x);
3403
Satyam Sharma2408c552007-10-16 01:24:44 -07003404 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003405 return;
3406
Christoph Lameterb49af682007-05-06 14:49:41 -07003407 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003408 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003409 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003410 kmemleak_free(x);
Glauber Costad79923f2012-12-18 14:22:48 -08003411 __free_memcg_kmem_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003412 return;
3413 }
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003414 slab_free(page->slab_cache, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003415}
3416EXPORT_SYMBOL(kfree);
3417
Christoph Lameter2086d262007-05-06 14:49:46 -07003418/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003419 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3420 * the remaining slabs by the number of items in use. The slabs with the
3421 * most items in use come first. New allocations will then fill those up
3422 * and thus they can be removed from the partial lists.
3423 *
3424 * The slabs with the least items are placed last. This results in them
3425 * being allocated from last increasing the chance that the last objects
3426 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003427 */
3428int kmem_cache_shrink(struct kmem_cache *s)
3429{
3430 int node;
3431 int i;
3432 struct kmem_cache_node *n;
3433 struct page *page;
3434 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003435 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003436 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003437 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003438 unsigned long flags;
3439
3440 if (!slabs_by_inuse)
3441 return -ENOMEM;
3442
3443 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003444 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003445 n = get_node(s, node);
3446
3447 if (!n->nr_partial)
3448 continue;
3449
Christoph Lameter834f3d12008-04-14 19:11:31 +03003450 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003451 INIT_LIST_HEAD(slabs_by_inuse + i);
3452
3453 spin_lock_irqsave(&n->list_lock, flags);
3454
3455 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003456 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003457 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003458 * Note that concurrent frees may occur while we hold the
3459 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003460 */
3461 list_for_each_entry_safe(page, t, &n->partial, lru) {
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003462 list_move(&page->lru, slabs_by_inuse + page->inuse);
3463 if (!page->inuse)
3464 n->nr_partial--;
Christoph Lameter2086d262007-05-06 14:49:46 -07003465 }
3466
Christoph Lameter2086d262007-05-06 14:49:46 -07003467 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003468 * Rebuild the partial list with the slabs filled up most
3469 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003470 */
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003471 for (i = objects - 1; i > 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003472 list_splice(slabs_by_inuse + i, n->partial.prev);
3473
Christoph Lameter2086d262007-05-06 14:49:46 -07003474 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003475
3476 /* Release empty slabs */
3477 list_for_each_entry_safe(page, t, slabs_by_inuse, lru)
3478 discard_slab(s, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003479 }
3480
3481 kfree(slabs_by_inuse);
3482 return 0;
3483}
3484EXPORT_SYMBOL(kmem_cache_shrink);
3485
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003486#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003487static int slab_mem_going_offline_callback(void *arg)
3488{
3489 struct kmem_cache *s;
3490
Christoph Lameter18004c52012-07-06 15:25:12 -05003491 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003492 list_for_each_entry(s, &slab_caches, list)
3493 kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05003494 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003495
3496 return 0;
3497}
3498
3499static void slab_mem_offline_callback(void *arg)
3500{
3501 struct kmem_cache_node *n;
3502 struct kmem_cache *s;
3503 struct memory_notify *marg = arg;
3504 int offline_node;
3505
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003506 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003507
3508 /*
3509 * If the node still has available memory. we need kmem_cache_node
3510 * for it yet.
3511 */
3512 if (offline_node < 0)
3513 return;
3514
Christoph Lameter18004c52012-07-06 15:25:12 -05003515 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003516 list_for_each_entry(s, &slab_caches, list) {
3517 n = get_node(s, offline_node);
3518 if (n) {
3519 /*
3520 * if n->nr_slabs > 0, slabs still exist on the node
3521 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003522 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003523 * callback. So, we must fail.
3524 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003525 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003526
3527 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003528 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003529 }
3530 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003531 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003532}
3533
3534static int slab_mem_going_online_callback(void *arg)
3535{
3536 struct kmem_cache_node *n;
3537 struct kmem_cache *s;
3538 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003539 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003540 int ret = 0;
3541
3542 /*
3543 * If the node's memory is already available, then kmem_cache_node is
3544 * already created. Nothing to do.
3545 */
3546 if (nid < 0)
3547 return 0;
3548
3549 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003550 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003551 * allocate a kmem_cache_node structure in order to bring the node
3552 * online.
3553 */
Christoph Lameter18004c52012-07-06 15:25:12 -05003554 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003555 list_for_each_entry(s, &slab_caches, list) {
3556 /*
3557 * XXX: kmem_cache_alloc_node will fallback to other nodes
3558 * since memory is not yet available from the node that
3559 * is brought up.
3560 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003561 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003562 if (!n) {
3563 ret = -ENOMEM;
3564 goto out;
3565 }
Joonsoo Kim40534972012-05-11 00:50:47 +09003566 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003567 s->node[nid] = n;
3568 }
3569out:
Christoph Lameter18004c52012-07-06 15:25:12 -05003570 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003571 return ret;
3572}
3573
3574static int slab_memory_callback(struct notifier_block *self,
3575 unsigned long action, void *arg)
3576{
3577 int ret = 0;
3578
3579 switch (action) {
3580 case MEM_GOING_ONLINE:
3581 ret = slab_mem_going_online_callback(arg);
3582 break;
3583 case MEM_GOING_OFFLINE:
3584 ret = slab_mem_going_offline_callback(arg);
3585 break;
3586 case MEM_OFFLINE:
3587 case MEM_CANCEL_ONLINE:
3588 slab_mem_offline_callback(arg);
3589 break;
3590 case MEM_ONLINE:
3591 case MEM_CANCEL_OFFLINE:
3592 break;
3593 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003594 if (ret)
3595 ret = notifier_from_errno(ret);
3596 else
3597 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003598 return ret;
3599}
3600
3601#endif /* CONFIG_MEMORY_HOTPLUG */
3602
Christoph Lameter81819f02007-05-06 14:49:36 -07003603/********************************************************************
3604 * Basic setup of slabs
3605 *******************************************************************/
3606
Christoph Lameter51df1142010-08-20 12:37:15 -05003607/*
3608 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003609 * the page allocator. Allocate them properly then fix up the pointers
3610 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05003611 */
3612
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003613static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05003614{
3615 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003616 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameter51df1142010-08-20 12:37:15 -05003617
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003618 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003619
3620 for_each_node_state(node, N_NORMAL_MEMORY) {
3621 struct kmem_cache_node *n = get_node(s, node);
3622 struct page *p;
3623
3624 if (n) {
3625 list_for_each_entry(p, &n->partial, lru)
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003626 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003627
Li Zefan607bf322011-04-12 15:22:26 +08003628#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003629 list_for_each_entry(p, &n->full, lru)
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003630 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003631#endif
3632 }
3633 }
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003634 list_add(&s->list, &slab_caches);
3635 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003636}
3637
Christoph Lameter81819f02007-05-06 14:49:36 -07003638void __init kmem_cache_init(void)
3639{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003640 static __initdata struct kmem_cache boot_kmem_cache,
3641 boot_kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003642 int i;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003643 int caches = 2;
Christoph Lameter51df1142010-08-20 12:37:15 -05003644
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08003645 if (debug_guardpage_minorder())
3646 slub_max_order = 0;
3647
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003648 kmem_cache_node = &boot_kmem_cache_node;
3649 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05003650
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003651 create_boot_cache(kmem_cache_node, "kmem_cache_node",
3652 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003653
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003654 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003655
3656 /* Able to allocate the per node structures */
3657 slab_state = PARTIAL;
3658
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003659 create_boot_cache(kmem_cache, "kmem_cache",
3660 offsetof(struct kmem_cache, node) +
3661 nr_node_ids * sizeof(struct kmem_cache_node *),
3662 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003663
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003664 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003665
Christoph Lameter51df1142010-08-20 12:37:15 -05003666 /*
3667 * Allocate kmem_cache_node properly from the kmem_cache slab.
3668 * kmem_cache_node is separately allocated so no need to
3669 * update any list pointers.
3670 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003671 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05003672
3673 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003674
3675 /*
3676 * Patch up the size_index table if we have strange large alignment
3677 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003678 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003679 *
3680 * Largest permitted alignment is 256 bytes due to the way we
3681 * handle the index determination for the smaller caches.
3682 *
3683 * Make sure that nothing crazy happens if someone starts tinkering
3684 * around with ARCH_KMALLOC_MINALIGN
3685 */
3686 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3687 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3688
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003689 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3690 int elem = size_index_elem(i);
3691 if (elem >= ARRAY_SIZE(size_index))
3692 break;
3693 size_index[elem] = KMALLOC_SHIFT_LOW;
3694 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003695
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003696 if (KMALLOC_MIN_SIZE == 64) {
3697 /*
3698 * The 96 byte size cache is not used if the alignment
3699 * is 64 byte.
3700 */
3701 for (i = 64 + 8; i <= 96; i += 8)
3702 size_index[size_index_elem(i)] = 7;
3703 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003704 /*
3705 * The 192 byte sized cache is not used if the alignment
3706 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3707 * instead.
3708 */
3709 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003710 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003711 }
3712
Christoph Lameter51df1142010-08-20 12:37:15 -05003713 /* Caches that are not of the two-to-the-power-of size */
3714 if (KMALLOC_MIN_SIZE <= 32) {
3715 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3716 caches++;
3717 }
3718
3719 if (KMALLOC_MIN_SIZE <= 64) {
3720 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3721 caches++;
3722 }
3723
3724 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3725 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3726 caches++;
3727 }
3728
Christoph Lameter81819f02007-05-06 14:49:36 -07003729 slab_state = UP;
3730
3731 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003732 if (KMALLOC_MIN_SIZE <= 32) {
3733 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3734 BUG_ON(!kmalloc_caches[1]->name);
3735 }
3736
3737 if (KMALLOC_MIN_SIZE <= 64) {
3738 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3739 BUG_ON(!kmalloc_caches[2]->name);
3740 }
3741
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003742 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3743 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3744
3745 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003746 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003747 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003748
3749#ifdef CONFIG_SMP
3750 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003751#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003752
Christoph Lameter55136592010-08-20 12:37:13 -05003753#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003754 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3755 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003756
Christoph Lameter51df1142010-08-20 12:37:15 -05003757 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003758 char *name = kasprintf(GFP_NOWAIT,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003759 "dma-kmalloc-%d", s->object_size);
Christoph Lameter55136592010-08-20 12:37:13 -05003760
3761 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003762 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003763 s->object_size, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003764 }
3765 }
3766#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003767 printk(KERN_INFO
3768 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003769 " CPUs=%d, Nodes=%d\n",
3770 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003771 slub_min_order, slub_max_order, slub_min_objects,
3772 nr_cpu_ids, nr_node_ids);
3773}
3774
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003775void __init kmem_cache_init_late(void)
3776{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003777}
3778
Christoph Lameter81819f02007-05-06 14:49:36 -07003779/*
3780 * Find a mergeable slab cache
3781 */
3782static int slab_unmergeable(struct kmem_cache *s)
3783{
3784 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3785 return 1;
3786
Christoph Lameterc59def92007-05-16 22:10:50 -07003787 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003788 return 1;
3789
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003790 /*
3791 * We may have set a slab to be unmergeable during bootstrap.
3792 */
3793 if (s->refcount < 0)
3794 return 1;
3795
Christoph Lameter81819f02007-05-06 14:49:36 -07003796 return 0;
3797}
3798
Glauber Costa2633d7a2012-12-18 14:22:34 -08003799static struct kmem_cache *find_mergeable(struct mem_cgroup *memcg, size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003800 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003801 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003802{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003803 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003804
3805 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3806 return NULL;
3807
Christoph Lameterc59def92007-05-16 22:10:50 -07003808 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003809 return NULL;
3810
3811 size = ALIGN(size, sizeof(void *));
3812 align = calculate_alignment(flags, align, size);
3813 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003814 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003815
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003816 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003817 if (slab_unmergeable(s))
3818 continue;
3819
3820 if (size > s->size)
3821 continue;
3822
Christoph Lameterba0268a2007-09-11 15:24:11 -07003823 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003824 continue;
3825 /*
3826 * Check if alignment is compatible.
3827 * Courtesy of Adrian Drzewiecki
3828 */
Pekka Enberg06428782008-01-07 23:20:27 -08003829 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003830 continue;
3831
3832 if (s->size - size >= sizeof(void *))
3833 continue;
3834
Glauber Costa2633d7a2012-12-18 14:22:34 -08003835 if (!cache_match_memcg(s, memcg))
3836 continue;
3837
Christoph Lameter81819f02007-05-06 14:49:36 -07003838 return s;
3839 }
3840 return NULL;
3841}
3842
Glauber Costa2633d7a2012-12-18 14:22:34 -08003843struct kmem_cache *
3844__kmem_cache_alias(struct mem_cgroup *memcg, const char *name, size_t size,
3845 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003846{
3847 struct kmem_cache *s;
3848
Glauber Costa2633d7a2012-12-18 14:22:34 -08003849 s = find_mergeable(memcg, size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003850 if (s) {
3851 s->refcount++;
3852 /*
3853 * Adjust the object sizes so that we clear
3854 * the complete object on kzalloc.
3855 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003856 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003857 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003858
David Rientjes7b8f3b62008-12-17 22:09:46 -08003859 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003860 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00003861 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003862 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003863 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003864
Christoph Lametercbb79692012-09-05 00:18:32 +00003865 return s;
3866}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003867
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003868int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00003869{
Pekka Enbergaac3a162012-09-05 12:07:44 +03003870 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003871
Pekka Enbergaac3a162012-09-05 12:07:44 +03003872 err = kmem_cache_open(s, flags);
3873 if (err)
3874 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003875
Christoph Lameter45530c42012-11-28 16:23:07 +00003876 /* Mutex is not taken during early boot */
3877 if (slab_state <= UP)
3878 return 0;
3879
Glauber Costa107dab52012-12-18 14:23:05 -08003880 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003881 mutex_unlock(&slab_mutex);
3882 err = sysfs_slab_add(s);
3883 mutex_lock(&slab_mutex);
Christoph Lameter20cea962012-07-06 15:25:13 -05003884
Pekka Enbergaac3a162012-09-05 12:07:44 +03003885 if (err)
3886 kmem_cache_close(s);
3887
3888 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07003889}
Christoph Lameter81819f02007-05-06 14:49:36 -07003890
Christoph Lameter81819f02007-05-06 14:49:36 -07003891#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003892/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003893 * Use the cpu notifier to insure that the cpu slabs are flushed when
3894 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003895 */
3896static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3897 unsigned long action, void *hcpu)
3898{
3899 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003900 struct kmem_cache *s;
3901 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003902
3903 switch (action) {
3904 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003905 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003906 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003907 case CPU_DEAD_FROZEN:
Christoph Lameter18004c52012-07-06 15:25:12 -05003908 mutex_lock(&slab_mutex);
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003909 list_for_each_entry(s, &slab_caches, list) {
3910 local_irq_save(flags);
3911 __flush_cpu_slab(s, cpu);
3912 local_irq_restore(flags);
3913 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003914 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07003915 break;
3916 default:
3917 break;
3918 }
3919 return NOTIFY_OK;
3920}
3921
Pekka Enberg06428782008-01-07 23:20:27 -08003922static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003923 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003924};
Christoph Lameter81819f02007-05-06 14:49:36 -07003925
3926#endif
3927
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003928void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003929{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003930 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003931 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003932
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003933 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003934 return kmalloc_large(size, gfpflags);
3935
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003936 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003937
Satyam Sharma2408c552007-10-16 01:24:44 -07003938 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003939 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003940
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003941 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003942
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003943 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003944 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003945
3946 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003947}
3948
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003949#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003950void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003951 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003952{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003953 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003954 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003955
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003956 if (unlikely(size > SLUB_MAX_SIZE)) {
3957 ret = kmalloc_large_node(size, gfpflags, node);
3958
3959 trace_kmalloc_node(caller, ret,
3960 size, PAGE_SIZE << get_order(size),
3961 gfpflags, node);
3962
3963 return ret;
3964 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003965
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003966 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003967
Satyam Sharma2408c552007-10-16 01:24:44 -07003968 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003969 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003970
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003971 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003972
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003973 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003974 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003975
3976 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003977}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003978#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003979
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003980#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003981static int count_inuse(struct page *page)
3982{
3983 return page->inuse;
3984}
3985
3986static int count_total(struct page *page)
3987{
3988 return page->objects;
3989}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003990#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003991
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003992#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003993static int validate_slab(struct kmem_cache *s, struct page *page,
3994 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003995{
3996 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003997 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003998
3999 if (!check_slab(s, page) ||
4000 !on_freelist(s, page, NULL))
4001 return 0;
4002
4003 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004004 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004005
Christoph Lameter5f80b132011-04-15 14:48:13 -05004006 get_map(s, page, map);
4007 for_each_object(p, s, addr, page->objects) {
4008 if (test_bit(slab_index(p, s, addr), map))
4009 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4010 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004011 }
4012
Christoph Lameter224a88b2008-04-14 19:11:31 +03004013 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004014 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004015 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004016 return 0;
4017 return 1;
4018}
4019
Christoph Lameter434e2452007-07-17 04:03:30 -07004020static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4021 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004022{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004023 slab_lock(page);
4024 validate_slab(s, page, map);
4025 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004026}
4027
Christoph Lameter434e2452007-07-17 04:03:30 -07004028static int validate_slab_node(struct kmem_cache *s,
4029 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004030{
4031 unsigned long count = 0;
4032 struct page *page;
4033 unsigned long flags;
4034
4035 spin_lock_irqsave(&n->list_lock, flags);
4036
4037 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004038 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004039 count++;
4040 }
4041 if (count != n->nr_partial)
4042 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
4043 "counter=%ld\n", s->name, count, n->nr_partial);
4044
4045 if (!(s->flags & SLAB_STORE_USER))
4046 goto out;
4047
4048 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004049 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004050 count++;
4051 }
4052 if (count != atomic_long_read(&n->nr_slabs))
4053 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
4054 "counter=%ld\n", s->name, count,
4055 atomic_long_read(&n->nr_slabs));
4056
4057out:
4058 spin_unlock_irqrestore(&n->list_lock, flags);
4059 return count;
4060}
4061
Christoph Lameter434e2452007-07-17 04:03:30 -07004062static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004063{
4064 int node;
4065 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004066 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004067 sizeof(unsigned long), GFP_KERNEL);
4068
4069 if (!map)
4070 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004071
4072 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07004073 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07004074 struct kmem_cache_node *n = get_node(s, node);
4075
Christoph Lameter434e2452007-07-17 04:03:30 -07004076 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004077 }
Christoph Lameter434e2452007-07-17 04:03:30 -07004078 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004079 return count;
4080}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004081/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004082 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004083 * and freed.
4084 */
4085
4086struct location {
4087 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004088 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004089 long long sum_time;
4090 long min_time;
4091 long max_time;
4092 long min_pid;
4093 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304094 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004095 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004096};
4097
4098struct loc_track {
4099 unsigned long max;
4100 unsigned long count;
4101 struct location *loc;
4102};
4103
4104static void free_loc_track(struct loc_track *t)
4105{
4106 if (t->max)
4107 free_pages((unsigned long)t->loc,
4108 get_order(sizeof(struct location) * t->max));
4109}
4110
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004111static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004112{
4113 struct location *l;
4114 int order;
4115
Christoph Lameter88a420e2007-05-06 14:49:45 -07004116 order = get_order(sizeof(struct location) * max);
4117
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004118 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004119 if (!l)
4120 return 0;
4121
4122 if (t->count) {
4123 memcpy(l, t->loc, sizeof(struct location) * t->count);
4124 free_loc_track(t);
4125 }
4126 t->max = max;
4127 t->loc = l;
4128 return 1;
4129}
4130
4131static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004132 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004133{
4134 long start, end, pos;
4135 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004136 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004137 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004138
4139 start = -1;
4140 end = t->count;
4141
4142 for ( ; ; ) {
4143 pos = start + (end - start + 1) / 2;
4144
4145 /*
4146 * There is nothing at "end". If we end up there
4147 * we need to add something to before end.
4148 */
4149 if (pos == end)
4150 break;
4151
4152 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004153 if (track->addr == caddr) {
4154
4155 l = &t->loc[pos];
4156 l->count++;
4157 if (track->when) {
4158 l->sum_time += age;
4159 if (age < l->min_time)
4160 l->min_time = age;
4161 if (age > l->max_time)
4162 l->max_time = age;
4163
4164 if (track->pid < l->min_pid)
4165 l->min_pid = track->pid;
4166 if (track->pid > l->max_pid)
4167 l->max_pid = track->pid;
4168
Rusty Russell174596a2009-01-01 10:12:29 +10304169 cpumask_set_cpu(track->cpu,
4170 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004171 }
4172 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004173 return 1;
4174 }
4175
Christoph Lameter45edfa52007-05-09 02:32:45 -07004176 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004177 end = pos;
4178 else
4179 start = pos;
4180 }
4181
4182 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004183 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004184 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004185 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004186 return 0;
4187
4188 l = t->loc + pos;
4189 if (pos < t->count)
4190 memmove(l + 1, l,
4191 (t->count - pos) * sizeof(struct location));
4192 t->count++;
4193 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004194 l->addr = track->addr;
4195 l->sum_time = age;
4196 l->min_time = age;
4197 l->max_time = age;
4198 l->min_pid = track->pid;
4199 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304200 cpumask_clear(to_cpumask(l->cpus));
4201 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004202 nodes_clear(l->nodes);
4203 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004204 return 1;
4205}
4206
4207static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004208 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004209 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004210{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004211 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004212 void *p;
4213
Christoph Lameter39b26462008-04-14 19:11:30 +03004214 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004215 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004216
Christoph Lameter224a88b2008-04-14 19:11:31 +03004217 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004218 if (!test_bit(slab_index(p, s, addr), map))
4219 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004220}
4221
4222static int list_locations(struct kmem_cache *s, char *buf,
4223 enum track_item alloc)
4224{
Harvey Harrisone374d482008-01-31 15:20:50 -08004225 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004226 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004227 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004228 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004229 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4230 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004231
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004232 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4233 GFP_TEMPORARY)) {
4234 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004235 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004236 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004237 /* Push back cpu slabs */
4238 flush_all(s);
4239
Christoph Lameterf64dc582007-10-16 01:25:33 -07004240 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004241 struct kmem_cache_node *n = get_node(s, node);
4242 unsigned long flags;
4243 struct page *page;
4244
Christoph Lameter9e869432007-08-22 14:01:56 -07004245 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004246 continue;
4247
4248 spin_lock_irqsave(&n->list_lock, flags);
4249 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004250 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004251 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004252 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004253 spin_unlock_irqrestore(&n->list_lock, flags);
4254 }
4255
4256 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004257 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004258
Hugh Dickins9c246242008-12-09 13:14:27 -08004259 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004260 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004261 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004262
4263 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004264 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004265 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004266 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004267
4268 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004269 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004270 l->min_time,
4271 (long)div_u64(l->sum_time, l->count),
4272 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004273 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004274 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004275 l->min_time);
4276
4277 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004278 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004279 l->min_pid, l->max_pid);
4280 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004281 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004282 l->min_pid);
4283
Rusty Russell174596a2009-01-01 10:12:29 +10304284 if (num_online_cpus() > 1 &&
4285 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004286 len < PAGE_SIZE - 60) {
4287 len += sprintf(buf + len, " cpus=");
4288 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10304289 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004290 }
4291
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004292 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004293 len < PAGE_SIZE - 60) {
4294 len += sprintf(buf + len, " nodes=");
4295 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004296 l->nodes);
4297 }
4298
Harvey Harrisone374d482008-01-31 15:20:50 -08004299 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004300 }
4301
4302 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004303 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004304 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004305 len += sprintf(buf, "No data\n");
4306 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004307}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004308#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004309
Christoph Lametera5a84752010-10-05 13:57:27 -05004310#ifdef SLUB_RESILIENCY_TEST
4311static void resiliency_test(void)
4312{
4313 u8 *p;
4314
4315 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
4316
4317 printk(KERN_ERR "SLUB resiliency testing\n");
4318 printk(KERN_ERR "-----------------------\n");
4319 printk(KERN_ERR "A. Corruption after allocation\n");
4320
4321 p = kzalloc(16, GFP_KERNEL);
4322 p[16] = 0x12;
4323 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
4324 " 0x12->0x%p\n\n", p + 16);
4325
4326 validate_slab_cache(kmalloc_caches[4]);
4327
4328 /* Hmmm... The next two are dangerous */
4329 p = kzalloc(32, GFP_KERNEL);
4330 p[32 + sizeof(void *)] = 0x34;
4331 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
4332 " 0x34 -> -0x%p\n", p);
4333 printk(KERN_ERR
4334 "If allocated object is overwritten then not detectable\n\n");
4335
4336 validate_slab_cache(kmalloc_caches[5]);
4337 p = kzalloc(64, GFP_KERNEL);
4338 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4339 *p = 0x56;
4340 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4341 p);
4342 printk(KERN_ERR
4343 "If allocated object is overwritten then not detectable\n\n");
4344 validate_slab_cache(kmalloc_caches[6]);
4345
4346 printk(KERN_ERR "\nB. Corruption after free\n");
4347 p = kzalloc(128, GFP_KERNEL);
4348 kfree(p);
4349 *p = 0x78;
4350 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4351 validate_slab_cache(kmalloc_caches[7]);
4352
4353 p = kzalloc(256, GFP_KERNEL);
4354 kfree(p);
4355 p[50] = 0x9a;
4356 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4357 p);
4358 validate_slab_cache(kmalloc_caches[8]);
4359
4360 p = kzalloc(512, GFP_KERNEL);
4361 kfree(p);
4362 p[512] = 0xab;
4363 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4364 validate_slab_cache(kmalloc_caches[9]);
4365}
4366#else
4367#ifdef CONFIG_SYSFS
4368static void resiliency_test(void) {};
4369#endif
4370#endif
4371
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004372#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004373enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004374 SL_ALL, /* All slabs */
4375 SL_PARTIAL, /* Only partially allocated slabs */
4376 SL_CPU, /* Only slabs used for cpu caches */
4377 SL_OBJECTS, /* Determine allocated objects not slabs */
4378 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004379};
4380
Christoph Lameter205ab992008-04-14 19:11:40 +03004381#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004382#define SO_PARTIAL (1 << SL_PARTIAL)
4383#define SO_CPU (1 << SL_CPU)
4384#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004385#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004386
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004387static ssize_t show_slab_objects(struct kmem_cache *s,
4388 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004389{
4390 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004391 int node;
4392 int x;
4393 unsigned long *nodes;
4394 unsigned long *per_cpu;
4395
4396 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004397 if (!nodes)
4398 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004399 per_cpu = nodes + nr_node_ids;
4400
Christoph Lameter205ab992008-04-14 19:11:40 +03004401 if (flags & SO_CPU) {
4402 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004403
Christoph Lameter205ab992008-04-14 19:11:40 +03004404 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004405 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004406 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004407 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004408
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004409 page = ACCESS_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004410 if (!page)
4411 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004412
Christoph Lameterec3ab082012-05-09 10:09:56 -05004413 node = page_to_nid(page);
4414 if (flags & SO_TOTAL)
4415 x = page->objects;
4416 else if (flags & SO_OBJECTS)
4417 x = page->inuse;
4418 else
4419 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004420
Christoph Lameterec3ab082012-05-09 10:09:56 -05004421 total += x;
4422 nodes[node] += x;
4423
4424 page = ACCESS_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004425 if (page) {
4426 x = page->pobjects;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004427 total += x;
4428 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004429 }
Christoph Lameterec3ab082012-05-09 10:09:56 -05004430
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004431 per_cpu[node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004432 }
4433 }
4434
Christoph Lameter04d94872011-01-10 10:15:15 -06004435 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004436#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004437 if (flags & SO_ALL) {
4438 for_each_node_state(node, N_NORMAL_MEMORY) {
4439 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004440
Christoph Lameter205ab992008-04-14 19:11:40 +03004441 if (flags & SO_TOTAL)
4442 x = atomic_long_read(&n->total_objects);
4443 else if (flags & SO_OBJECTS)
4444 x = atomic_long_read(&n->total_objects) -
4445 count_partial(n, count_free);
4446
4447 else
4448 x = atomic_long_read(&n->nr_slabs);
4449 total += x;
4450 nodes[node] += x;
4451 }
4452
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004453 } else
4454#endif
4455 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004456 for_each_node_state(node, N_NORMAL_MEMORY) {
4457 struct kmem_cache_node *n = get_node(s, node);
4458
4459 if (flags & SO_TOTAL)
4460 x = count_partial(n, count_total);
4461 else if (flags & SO_OBJECTS)
4462 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004463 else
4464 x = n->nr_partial;
4465 total += x;
4466 nodes[node] += x;
4467 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004468 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004469 x = sprintf(buf, "%lu", total);
4470#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004471 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004472 if (nodes[node])
4473 x += sprintf(buf + x, " N%d=%lu",
4474 node, nodes[node]);
4475#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004476 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004477 kfree(nodes);
4478 return x + sprintf(buf + x, "\n");
4479}
4480
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004481#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004482static int any_slab_objects(struct kmem_cache *s)
4483{
4484 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004485
4486 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004487 struct kmem_cache_node *n = get_node(s, node);
4488
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004489 if (!n)
4490 continue;
4491
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004492 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004493 return 1;
4494 }
4495 return 0;
4496}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004497#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004498
4499#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004500#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004501
4502struct slab_attribute {
4503 struct attribute attr;
4504 ssize_t (*show)(struct kmem_cache *s, char *buf);
4505 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4506};
4507
4508#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004509 static struct slab_attribute _name##_attr = \
4510 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004511
4512#define SLAB_ATTR(_name) \
4513 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004514 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004515
Christoph Lameter81819f02007-05-06 14:49:36 -07004516static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4517{
4518 return sprintf(buf, "%d\n", s->size);
4519}
4520SLAB_ATTR_RO(slab_size);
4521
4522static ssize_t align_show(struct kmem_cache *s, char *buf)
4523{
4524 return sprintf(buf, "%d\n", s->align);
4525}
4526SLAB_ATTR_RO(align);
4527
4528static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4529{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004530 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004531}
4532SLAB_ATTR_RO(object_size);
4533
4534static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4535{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004536 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004537}
4538SLAB_ATTR_RO(objs_per_slab);
4539
Christoph Lameter06b285d2008-04-14 19:11:41 +03004540static ssize_t order_store(struct kmem_cache *s,
4541 const char *buf, size_t length)
4542{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004543 unsigned long order;
4544 int err;
4545
4546 err = strict_strtoul(buf, 10, &order);
4547 if (err)
4548 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004549
4550 if (order > slub_max_order || order < slub_min_order)
4551 return -EINVAL;
4552
4553 calculate_sizes(s, order);
4554 return length;
4555}
4556
Christoph Lameter81819f02007-05-06 14:49:36 -07004557static ssize_t order_show(struct kmem_cache *s, char *buf)
4558{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004559 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004560}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004561SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004562
David Rientjes73d342b2009-02-22 17:40:09 -08004563static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4564{
4565 return sprintf(buf, "%lu\n", s->min_partial);
4566}
4567
4568static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4569 size_t length)
4570{
4571 unsigned long min;
4572 int err;
4573
4574 err = strict_strtoul(buf, 10, &min);
4575 if (err)
4576 return err;
4577
David Rientjesc0bdb232009-02-25 09:16:35 +02004578 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004579 return length;
4580}
4581SLAB_ATTR(min_partial);
4582
Christoph Lameter49e22582011-08-09 16:12:27 -05004583static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4584{
4585 return sprintf(buf, "%u\n", s->cpu_partial);
4586}
4587
4588static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4589 size_t length)
4590{
4591 unsigned long objects;
4592 int err;
4593
4594 err = strict_strtoul(buf, 10, &objects);
4595 if (err)
4596 return err;
David Rientjes74ee4ef2012-01-09 13:19:45 -08004597 if (objects && kmem_cache_debug(s))
4598 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004599
4600 s->cpu_partial = objects;
4601 flush_all(s);
4602 return length;
4603}
4604SLAB_ATTR(cpu_partial);
4605
Christoph Lameter81819f02007-05-06 14:49:36 -07004606static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4607{
Joe Perches62c70bc2011-01-13 15:45:52 -08004608 if (!s->ctor)
4609 return 0;
4610 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004611}
4612SLAB_ATTR_RO(ctor);
4613
Christoph Lameter81819f02007-05-06 14:49:36 -07004614static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4615{
4616 return sprintf(buf, "%d\n", s->refcount - 1);
4617}
4618SLAB_ATTR_RO(aliases);
4619
Christoph Lameter81819f02007-05-06 14:49:36 -07004620static ssize_t partial_show(struct kmem_cache *s, char *buf)
4621{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004622 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004623}
4624SLAB_ATTR_RO(partial);
4625
4626static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4627{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004628 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004629}
4630SLAB_ATTR_RO(cpu_slabs);
4631
4632static ssize_t objects_show(struct kmem_cache *s, char *buf)
4633{
Christoph Lameter205ab992008-04-14 19:11:40 +03004634 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004635}
4636SLAB_ATTR_RO(objects);
4637
Christoph Lameter205ab992008-04-14 19:11:40 +03004638static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4639{
4640 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4641}
4642SLAB_ATTR_RO(objects_partial);
4643
Christoph Lameter49e22582011-08-09 16:12:27 -05004644static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4645{
4646 int objects = 0;
4647 int pages = 0;
4648 int cpu;
4649 int len;
4650
4651 for_each_online_cpu(cpu) {
4652 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4653
4654 if (page) {
4655 pages += page->pages;
4656 objects += page->pobjects;
4657 }
4658 }
4659
4660 len = sprintf(buf, "%d(%d)", objects, pages);
4661
4662#ifdef CONFIG_SMP
4663 for_each_online_cpu(cpu) {
4664 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4665
4666 if (page && len < PAGE_SIZE - 20)
4667 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4668 page->pobjects, page->pages);
4669 }
4670#endif
4671 return len + sprintf(buf + len, "\n");
4672}
4673SLAB_ATTR_RO(slabs_cpu_partial);
4674
Christoph Lameter81819f02007-05-06 14:49:36 -07004675static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4676{
4677 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4678}
4679
4680static ssize_t reclaim_account_store(struct kmem_cache *s,
4681 const char *buf, size_t length)
4682{
4683 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4684 if (buf[0] == '1')
4685 s->flags |= SLAB_RECLAIM_ACCOUNT;
4686 return length;
4687}
4688SLAB_ATTR(reclaim_account);
4689
4690static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4691{
Christoph Lameter5af60832007-05-06 14:49:56 -07004692 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004693}
4694SLAB_ATTR_RO(hwcache_align);
4695
4696#ifdef CONFIG_ZONE_DMA
4697static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4698{
4699 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4700}
4701SLAB_ATTR_RO(cache_dma);
4702#endif
4703
4704static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4705{
4706 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4707}
4708SLAB_ATTR_RO(destroy_by_rcu);
4709
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004710static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4711{
4712 return sprintf(buf, "%d\n", s->reserved);
4713}
4714SLAB_ATTR_RO(reserved);
4715
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004716#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004717static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4718{
4719 return show_slab_objects(s, buf, SO_ALL);
4720}
4721SLAB_ATTR_RO(slabs);
4722
4723static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4724{
4725 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4726}
4727SLAB_ATTR_RO(total_objects);
4728
4729static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4730{
4731 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4732}
4733
4734static ssize_t sanity_checks_store(struct kmem_cache *s,
4735 const char *buf, size_t length)
4736{
4737 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004738 if (buf[0] == '1') {
4739 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004740 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004741 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004742 return length;
4743}
4744SLAB_ATTR(sanity_checks);
4745
4746static ssize_t trace_show(struct kmem_cache *s, char *buf)
4747{
4748 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4749}
4750
4751static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4752 size_t length)
4753{
4754 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004755 if (buf[0] == '1') {
4756 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004757 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004758 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004759 return length;
4760}
4761SLAB_ATTR(trace);
4762
Christoph Lameter81819f02007-05-06 14:49:36 -07004763static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4764{
4765 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4766}
4767
4768static ssize_t red_zone_store(struct kmem_cache *s,
4769 const char *buf, size_t length)
4770{
4771 if (any_slab_objects(s))
4772 return -EBUSY;
4773
4774 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004775 if (buf[0] == '1') {
4776 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004777 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004778 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004779 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004780 return length;
4781}
4782SLAB_ATTR(red_zone);
4783
4784static ssize_t poison_show(struct kmem_cache *s, char *buf)
4785{
4786 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4787}
4788
4789static ssize_t poison_store(struct kmem_cache *s,
4790 const char *buf, size_t length)
4791{
4792 if (any_slab_objects(s))
4793 return -EBUSY;
4794
4795 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004796 if (buf[0] == '1') {
4797 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004798 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004799 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004800 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004801 return length;
4802}
4803SLAB_ATTR(poison);
4804
4805static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4806{
4807 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4808}
4809
4810static ssize_t store_user_store(struct kmem_cache *s,
4811 const char *buf, size_t length)
4812{
4813 if (any_slab_objects(s))
4814 return -EBUSY;
4815
4816 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004817 if (buf[0] == '1') {
4818 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004819 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004820 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004821 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004822 return length;
4823}
4824SLAB_ATTR(store_user);
4825
Christoph Lameter53e15af2007-05-06 14:49:43 -07004826static ssize_t validate_show(struct kmem_cache *s, char *buf)
4827{
4828 return 0;
4829}
4830
4831static ssize_t validate_store(struct kmem_cache *s,
4832 const char *buf, size_t length)
4833{
Christoph Lameter434e2452007-07-17 04:03:30 -07004834 int ret = -EINVAL;
4835
4836 if (buf[0] == '1') {
4837 ret = validate_slab_cache(s);
4838 if (ret >= 0)
4839 ret = length;
4840 }
4841 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004842}
4843SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004844
4845static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4846{
4847 if (!(s->flags & SLAB_STORE_USER))
4848 return -ENOSYS;
4849 return list_locations(s, buf, TRACK_ALLOC);
4850}
4851SLAB_ATTR_RO(alloc_calls);
4852
4853static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4854{
4855 if (!(s->flags & SLAB_STORE_USER))
4856 return -ENOSYS;
4857 return list_locations(s, buf, TRACK_FREE);
4858}
4859SLAB_ATTR_RO(free_calls);
4860#endif /* CONFIG_SLUB_DEBUG */
4861
4862#ifdef CONFIG_FAILSLAB
4863static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4864{
4865 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4866}
4867
4868static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4869 size_t length)
4870{
4871 s->flags &= ~SLAB_FAILSLAB;
4872 if (buf[0] == '1')
4873 s->flags |= SLAB_FAILSLAB;
4874 return length;
4875}
4876SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004877#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004878
Christoph Lameter2086d262007-05-06 14:49:46 -07004879static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4880{
4881 return 0;
4882}
4883
4884static ssize_t shrink_store(struct kmem_cache *s,
4885 const char *buf, size_t length)
4886{
4887 if (buf[0] == '1') {
4888 int rc = kmem_cache_shrink(s);
4889
4890 if (rc)
4891 return rc;
4892 } else
4893 return -EINVAL;
4894 return length;
4895}
4896SLAB_ATTR(shrink);
4897
Christoph Lameter81819f02007-05-06 14:49:36 -07004898#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004899static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004900{
Christoph Lameter98246012008-01-07 23:20:26 -08004901 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004902}
4903
Christoph Lameter98246012008-01-07 23:20:26 -08004904static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004905 const char *buf, size_t length)
4906{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004907 unsigned long ratio;
4908 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004909
Christoph Lameter0121c6192008-04-29 16:11:12 -07004910 err = strict_strtoul(buf, 10, &ratio);
4911 if (err)
4912 return err;
4913
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004914 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004915 s->remote_node_defrag_ratio = ratio * 10;
4916
Christoph Lameter81819f02007-05-06 14:49:36 -07004917 return length;
4918}
Christoph Lameter98246012008-01-07 23:20:26 -08004919SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004920#endif
4921
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004922#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004923static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4924{
4925 unsigned long sum = 0;
4926 int cpu;
4927 int len;
4928 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4929
4930 if (!data)
4931 return -ENOMEM;
4932
4933 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004934 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004935
4936 data[cpu] = x;
4937 sum += x;
4938 }
4939
4940 len = sprintf(buf, "%lu", sum);
4941
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004942#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004943 for_each_online_cpu(cpu) {
4944 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004945 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004946 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004947#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004948 kfree(data);
4949 return len + sprintf(buf + len, "\n");
4950}
4951
David Rientjes78eb00c2009-10-15 02:20:22 -07004952static void clear_stat(struct kmem_cache *s, enum stat_item si)
4953{
4954 int cpu;
4955
4956 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004957 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004958}
4959
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004960#define STAT_ATTR(si, text) \
4961static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4962{ \
4963 return show_stat(s, buf, si); \
4964} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004965static ssize_t text##_store(struct kmem_cache *s, \
4966 const char *buf, size_t length) \
4967{ \
4968 if (buf[0] != '0') \
4969 return -EINVAL; \
4970 clear_stat(s, si); \
4971 return length; \
4972} \
4973SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004974
4975STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4976STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4977STAT_ATTR(FREE_FASTPATH, free_fastpath);
4978STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4979STAT_ATTR(FREE_FROZEN, free_frozen);
4980STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4981STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4982STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4983STAT_ATTR(ALLOC_SLAB, alloc_slab);
4984STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05004985STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004986STAT_ATTR(FREE_SLAB, free_slab);
4987STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4988STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4989STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4990STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4991STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4992STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05004993STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03004994STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004995STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4996STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05004997STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
4998STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08004999STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5000STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005001#endif
5002
Pekka Enberg06428782008-01-07 23:20:27 -08005003static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005004 &slab_size_attr.attr,
5005 &object_size_attr.attr,
5006 &objs_per_slab_attr.attr,
5007 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005008 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005009 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005010 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005011 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005012 &partial_attr.attr,
5013 &cpu_slabs_attr.attr,
5014 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005015 &aliases_attr.attr,
5016 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005017 &hwcache_align_attr.attr,
5018 &reclaim_account_attr.attr,
5019 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005020 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005021 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005022 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005023#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005024 &total_objects_attr.attr,
5025 &slabs_attr.attr,
5026 &sanity_checks_attr.attr,
5027 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005028 &red_zone_attr.attr,
5029 &poison_attr.attr,
5030 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005031 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005032 &alloc_calls_attr.attr,
5033 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005034#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005035#ifdef CONFIG_ZONE_DMA
5036 &cache_dma_attr.attr,
5037#endif
5038#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005039 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005040#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005041#ifdef CONFIG_SLUB_STATS
5042 &alloc_fastpath_attr.attr,
5043 &alloc_slowpath_attr.attr,
5044 &free_fastpath_attr.attr,
5045 &free_slowpath_attr.attr,
5046 &free_frozen_attr.attr,
5047 &free_add_partial_attr.attr,
5048 &free_remove_partial_attr.attr,
5049 &alloc_from_partial_attr.attr,
5050 &alloc_slab_attr.attr,
5051 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005052 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005053 &free_slab_attr.attr,
5054 &cpuslab_flush_attr.attr,
5055 &deactivate_full_attr.attr,
5056 &deactivate_empty_attr.attr,
5057 &deactivate_to_head_attr.attr,
5058 &deactivate_to_tail_attr.attr,
5059 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005060 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005061 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005062 &cmpxchg_double_fail_attr.attr,
5063 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005064 &cpu_partial_alloc_attr.attr,
5065 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005066 &cpu_partial_node_attr.attr,
5067 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005068#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005069#ifdef CONFIG_FAILSLAB
5070 &failslab_attr.attr,
5071#endif
5072
Christoph Lameter81819f02007-05-06 14:49:36 -07005073 NULL
5074};
5075
5076static struct attribute_group slab_attr_group = {
5077 .attrs = slab_attrs,
5078};
5079
5080static ssize_t slab_attr_show(struct kobject *kobj,
5081 struct attribute *attr,
5082 char *buf)
5083{
5084 struct slab_attribute *attribute;
5085 struct kmem_cache *s;
5086 int err;
5087
5088 attribute = to_slab_attr(attr);
5089 s = to_slab(kobj);
5090
5091 if (!attribute->show)
5092 return -EIO;
5093
5094 err = attribute->show(s, buf);
5095
5096 return err;
5097}
5098
5099static ssize_t slab_attr_store(struct kobject *kobj,
5100 struct attribute *attr,
5101 const char *buf, size_t len)
5102{
5103 struct slab_attribute *attribute;
5104 struct kmem_cache *s;
5105 int err;
5106
5107 attribute = to_slab_attr(attr);
5108 s = to_slab(kobj);
5109
5110 if (!attribute->store)
5111 return -EIO;
5112
5113 err = attribute->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005114#ifdef CONFIG_MEMCG_KMEM
5115 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
5116 int i;
Christoph Lameter81819f02007-05-06 14:49:36 -07005117
Glauber Costa107dab52012-12-18 14:23:05 -08005118 mutex_lock(&slab_mutex);
5119 if (s->max_attr_size < len)
5120 s->max_attr_size = len;
5121
Glauber Costaebe945c2012-12-18 14:23:10 -08005122 /*
5123 * This is a best effort propagation, so this function's return
5124 * value will be determined by the parent cache only. This is
5125 * basically because not all attributes will have a well
5126 * defined semantics for rollbacks - most of the actions will
5127 * have permanent effects.
5128 *
5129 * Returning the error value of any of the children that fail
5130 * is not 100 % defined, in the sense that users seeing the
5131 * error code won't be able to know anything about the state of
5132 * the cache.
5133 *
5134 * Only returning the error code for the parent cache at least
5135 * has well defined semantics. The cache being written to
5136 * directly either failed or succeeded, in which case we loop
5137 * through the descendants with best-effort propagation.
5138 */
Glauber Costa107dab52012-12-18 14:23:05 -08005139 for_each_memcg_cache_index(i) {
5140 struct kmem_cache *c = cache_from_memcg(s, i);
Glauber Costa107dab52012-12-18 14:23:05 -08005141 if (c)
5142 attribute->store(c, buf, len);
5143 }
5144 mutex_unlock(&slab_mutex);
5145 }
5146#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005147 return err;
5148}
5149
Glauber Costa107dab52012-12-18 14:23:05 -08005150static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5151{
5152#ifdef CONFIG_MEMCG_KMEM
5153 int i;
5154 char *buffer = NULL;
5155
5156 if (!is_root_cache(s))
5157 return;
5158
5159 /*
5160 * This mean this cache had no attribute written. Therefore, no point
5161 * in copying default values around
5162 */
5163 if (!s->max_attr_size)
5164 return;
5165
5166 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5167 char mbuf[64];
5168 char *buf;
5169 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5170
5171 if (!attr || !attr->store || !attr->show)
5172 continue;
5173
5174 /*
5175 * It is really bad that we have to allocate here, so we will
5176 * do it only as a fallback. If we actually allocate, though,
5177 * we can just use the allocated buffer until the end.
5178 *
5179 * Most of the slub attributes will tend to be very small in
5180 * size, but sysfs allows buffers up to a page, so they can
5181 * theoretically happen.
5182 */
5183 if (buffer)
5184 buf = buffer;
5185 else if (s->max_attr_size < ARRAY_SIZE(mbuf))
5186 buf = mbuf;
5187 else {
5188 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5189 if (WARN_ON(!buffer))
5190 continue;
5191 buf = buffer;
5192 }
5193
5194 attr->show(s->memcg_params->root_cache, buf);
5195 attr->store(s, buf, strlen(buf));
5196 }
5197
5198 if (buffer)
5199 free_page((unsigned long)buffer);
5200#endif
5201}
5202
Emese Revfy52cf25d2010-01-19 02:58:23 +01005203static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005204 .show = slab_attr_show,
5205 .store = slab_attr_store,
5206};
5207
5208static struct kobj_type slab_ktype = {
5209 .sysfs_ops = &slab_sysfs_ops,
5210};
5211
5212static int uevent_filter(struct kset *kset, struct kobject *kobj)
5213{
5214 struct kobj_type *ktype = get_ktype(kobj);
5215
5216 if (ktype == &slab_ktype)
5217 return 1;
5218 return 0;
5219}
5220
Emese Revfy9cd43612009-12-31 14:52:51 +01005221static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005222 .filter = uevent_filter,
5223};
5224
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005225static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005226
5227#define ID_STR_LENGTH 64
5228
5229/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005230 *
5231 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005232 */
5233static char *create_unique_id(struct kmem_cache *s)
5234{
5235 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5236 char *p = name;
5237
5238 BUG_ON(!name);
5239
5240 *p++ = ':';
5241 /*
5242 * First flags affecting slabcache operations. We will only
5243 * get here for aliasable slabs so we do not need to support
5244 * too many flags. The flags here must cover all flags that
5245 * are matched during merging to guarantee that the id is
5246 * unique.
5247 */
5248 if (s->flags & SLAB_CACHE_DMA)
5249 *p++ = 'd';
5250 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5251 *p++ = 'a';
5252 if (s->flags & SLAB_DEBUG_FREE)
5253 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005254 if (!(s->flags & SLAB_NOTRACK))
5255 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005256 if (p != name + 1)
5257 *p++ = '-';
5258 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005259
5260#ifdef CONFIG_MEMCG_KMEM
5261 if (!is_root_cache(s))
5262 p += sprintf(p, "-%08d", memcg_cache_id(s->memcg_params->memcg));
5263#endif
5264
Christoph Lameter81819f02007-05-06 14:49:36 -07005265 BUG_ON(p > name + ID_STR_LENGTH - 1);
5266 return name;
5267}
5268
5269static int sysfs_slab_add(struct kmem_cache *s)
5270{
5271 int err;
5272 const char *name;
Christoph Lameter45530c42012-11-28 16:23:07 +00005273 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005274
Christoph Lameter81819f02007-05-06 14:49:36 -07005275 if (unmergeable) {
5276 /*
5277 * Slabcache can never be merged so we can use the name proper.
5278 * This is typically the case for debug situations. In that
5279 * case we can catch duplicate names easily.
5280 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005281 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005282 name = s->name;
5283 } else {
5284 /*
5285 * Create a unique name for the slab as a target
5286 * for the symlinks.
5287 */
5288 name = create_unique_id(s);
5289 }
5290
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005291 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005292 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
5293 if (err) {
5294 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005295 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005296 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005297
5298 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005299 if (err) {
5300 kobject_del(&s->kobj);
5301 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005302 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005303 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005304 kobject_uevent(&s->kobj, KOBJ_ADD);
5305 if (!unmergeable) {
5306 /* Setup first alias */
5307 sysfs_slab_alias(s, s->name);
5308 kfree(name);
5309 }
5310 return 0;
5311}
5312
5313static void sysfs_slab_remove(struct kmem_cache *s)
5314{
Christoph Lameter97d06602012-07-06 15:25:11 -05005315 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005316 /*
5317 * Sysfs has not been setup yet so no need to remove the
5318 * cache from sysfs.
5319 */
5320 return;
5321
Christoph Lameter81819f02007-05-06 14:49:36 -07005322 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5323 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005324 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005325}
5326
5327/*
5328 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005329 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005330 */
5331struct saved_alias {
5332 struct kmem_cache *s;
5333 const char *name;
5334 struct saved_alias *next;
5335};
5336
Adrian Bunk5af328a2007-07-17 04:03:27 -07005337static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005338
5339static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5340{
5341 struct saved_alias *al;
5342
Christoph Lameter97d06602012-07-06 15:25:11 -05005343 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005344 /*
5345 * If we have a leftover link then remove it.
5346 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005347 sysfs_remove_link(&slab_kset->kobj, name);
5348 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005349 }
5350
5351 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5352 if (!al)
5353 return -ENOMEM;
5354
5355 al->s = s;
5356 al->name = name;
5357 al->next = alias_list;
5358 alias_list = al;
5359 return 0;
5360}
5361
5362static int __init slab_sysfs_init(void)
5363{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005364 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005365 int err;
5366
Christoph Lameter18004c52012-07-06 15:25:12 -05005367 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005368
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005369 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005370 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005371 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005372 printk(KERN_ERR "Cannot register slab subsystem.\n");
5373 return -ENOSYS;
5374 }
5375
Christoph Lameter97d06602012-07-06 15:25:11 -05005376 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005377
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005378 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005379 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005380 if (err)
5381 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
5382 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005383 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005384
5385 while (alias_list) {
5386 struct saved_alias *al = alias_list;
5387
5388 alias_list = alias_list->next;
5389 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005390 if (err)
5391 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
Julia Lawall068ce412012-07-08 13:37:40 +02005392 " %s to sysfs\n", al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005393 kfree(al);
5394 }
5395
Christoph Lameter18004c52012-07-06 15:25:12 -05005396 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005397 resiliency_test();
5398 return 0;
5399}
5400
5401__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005402#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005403
5404/*
5405 * The /proc/slabinfo ABI
5406 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005407#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005408void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005409{
5410 unsigned long nr_partials = 0;
5411 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005412 unsigned long nr_objs = 0;
5413 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005414 int node;
5415
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005416 for_each_online_node(node) {
5417 struct kmem_cache_node *n = get_node(s, node);
5418
5419 if (!n)
5420 continue;
5421
5422 nr_partials += n->nr_partial;
5423 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03005424 nr_objs += atomic_long_read(&n->total_objects);
5425 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005426 }
5427
Glauber Costa0d7561c2012-10-19 18:20:27 +04005428 sinfo->active_objs = nr_objs - nr_free;
5429 sinfo->num_objs = nr_objs;
5430 sinfo->active_slabs = nr_slabs;
5431 sinfo->num_slabs = nr_slabs;
5432 sinfo->objects_per_slab = oo_objects(s->oo);
5433 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005434}
5435
Glauber Costa0d7561c2012-10-19 18:20:27 +04005436void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005437{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005438}
5439
Glauber Costab7454ad2012-10-19 18:20:25 +04005440ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5441 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005442{
Glauber Costab7454ad2012-10-19 18:20:25 +04005443 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005444}
Linus Torvalds158a9622008-01-02 13:04:48 -08005445#endif /* CONFIG_SLABINFO */