blob: 7d2a996c307e4306bd233f4ae340a02d6915ffb1 [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>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040019#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070020#include <linux/seq_file.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020021#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070022#include <linux/cpu.h>
23#include <linux/cpuset.h>
24#include <linux/mempolicy.h>
25#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070026#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070027#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070028#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070029#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090030#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030031#include <linux/stacktrace.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070032
Richard Kennedy4a923792010-10-21 10:29:19 +010033#include <trace/events/kmem.h>
34
Christoph Lameter81819f02007-05-06 14:49:36 -070035/*
36 * Lock order:
Christoph Lameter881db7f2011-06-01 12:25:53 -050037 * 1. slub_lock (Global Semaphore)
38 * 2. node->list_lock
39 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070040 *
Christoph Lameter881db7f2011-06-01 12:25:53 -050041 * slub_lock
42 *
43 * The role of the slub_lock is to protect the list of all the slabs
44 * and to synchronize major metadata changes to slab cache structures.
45 *
46 * The slab_lock is only used for debugging and on arches that do not
47 * have the ability to do a cmpxchg_double. It only protects the second
48 * double word in the page struct. Meaning
49 * A. page->freelist -> List of object free in a page
50 * B. page->counters -> Counters of objects
51 * C. page->frozen -> frozen state
52 *
53 * If a slab is frozen then it is exempt from list management. It is not
54 * on any list. The processor that froze the slab is the one who can
55 * perform list operations on the page. Other processors may put objects
56 * onto the freelist but the processor that froze the slab is the only
57 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070058 *
59 * The list_lock protects the partial and full list on each node and
60 * the partial slab counter. If taken then no new slabs may be added or
61 * removed from the lists nor make the number of partial slabs be modified.
62 * (Note that the total number of slabs is an atomic value that may be
63 * modified without taking the list lock).
64 *
65 * The list_lock is a centralized lock and thus we avoid taking it as
66 * much as possible. As long as SLUB does not have to handle partial
67 * slabs, operations can continue without any centralized lock. F.e.
68 * allocating a long series of objects that fill up slabs does not require
69 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070070 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
74 *
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
77 *
Christoph Lameter672bba32007-05-09 02:32:39 -070078 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070080 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070081 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070083 *
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
87 *
88 * Overloading of page flags that are otherwise used for LRU management.
89 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070090 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
98 *
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700102 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700105 *
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700108 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700109 */
110
Christoph Lameteraf537b02010-07-09 14:07:14 -0500111#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
113
114static inline int kmem_cache_debug(struct kmem_cache *s)
115{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700116#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500117 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700118#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500119 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500121}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700122
Christoph Lameter81819f02007-05-06 14:49:36 -0700123/*
124 * Issues still to be resolved:
125 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
127 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700128 * - Variable sizing of the per node arrays
129 */
130
131/* Enable to test recovery from slab corruption on boot */
132#undef SLUB_RESILIENCY_TEST
133
Christoph Lameterb789ef52011-06-01 12:25:49 -0500134/* Enable to log cmpxchg failures */
135#undef SLUB_DEBUG_CMPXCHG
136
Christoph Lameter81819f02007-05-06 14:49:36 -0700137/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700138 * Mininum number of partial slabs. These will be left on the partial
139 * lists even if they are empty. kmem_cache_shrink may reclaim them.
140 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800141#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700142
Christoph Lameter2086d262007-05-06 14:49:46 -0700143/*
144 * Maximum number of desirable partial slabs.
145 * The existence of more partial slabs makes kmem_cache_shrink
146 * sort the partial list by the number of objects in the.
147 */
148#define MAX_PARTIAL 10
149
Christoph Lameter81819f02007-05-06 14:49:36 -0700150#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
151 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700152
Christoph Lameter81819f02007-05-06 14:49:36 -0700153/*
David Rientjes3de47212009-07-27 18:30:35 -0700154 * Debugging flags that require metadata to be stored in the slab. These get
155 * disabled when slub_debug=O is used and a cache's min order increases with
156 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700157 */
David Rientjes3de47212009-07-27 18:30:35 -0700158#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700159
160/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700161 * Set of flags that will prevent slab merging
162 */
163#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300164 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
165 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166
167#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200168 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700169
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400170#define OO_SHIFT 16
171#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500172#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400173
Christoph Lameter81819f02007-05-06 14:49:36 -0700174/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500175#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500176#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700177
178static int kmem_size = sizeof(struct kmem_cache);
179
180#ifdef CONFIG_SMP
181static struct notifier_block slab_notifier;
182#endif
183
184static enum {
185 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500186 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700187 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700188 SYSFS /* Sysfs up */
189} slab_state = DOWN;
190
191/* A list of all slab caches on the system */
192static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700193static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700194
Christoph Lameter02cbc872007-05-09 02:32:43 -0700195/*
196 * Tracking user of a slab.
197 */
Ben Greeard6543e32011-07-07 11:36:36 -0700198#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700199struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300200 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700201#ifdef CONFIG_STACKTRACE
202 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
203#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700204 int cpu; /* Was running on cpu */
205 int pid; /* Pid context */
206 unsigned long when; /* When did the operation occur */
207};
208
209enum track_item { TRACK_ALLOC, TRACK_FREE };
210
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500211#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700212static int sysfs_slab_add(struct kmem_cache *);
213static int sysfs_slab_alias(struct kmem_cache *, const char *);
214static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800215
Christoph Lameter81819f02007-05-06 14:49:36 -0700216#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700217static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
218static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
219 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800220static inline void sysfs_slab_remove(struct kmem_cache *s)
221{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300222 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800223 kfree(s);
224}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800225
Christoph Lameter81819f02007-05-06 14:49:36 -0700226#endif
227
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500228static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800229{
230#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600231 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800232#endif
233}
234
Christoph Lameter81819f02007-05-06 14:49:36 -0700235/********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
238
239int slab_is_available(void)
240{
241 return slab_state >= UP;
242}
243
244static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
245{
Christoph Lameter81819f02007-05-06 14:49:36 -0700246 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700247}
248
Christoph Lameter6446faa2008-02-15 23:45:26 -0800249/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700250static inline int check_valid_pointer(struct kmem_cache *s,
251 struct page *page, const void *object)
252{
253 void *base;
254
Christoph Lametera973e9d2008-03-01 13:40:44 -0800255 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700256 return 1;
257
Christoph Lametera973e9d2008-03-01 13:40:44 -0800258 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300259 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700260 (object - base) % s->size) {
261 return 0;
262 }
263
264 return 1;
265}
266
Christoph Lameter7656c722007-05-09 02:32:40 -0700267static inline void *get_freepointer(struct kmem_cache *s, void *object)
268{
269 return *(void **)(object + s->offset);
270}
271
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500272static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
273{
274 void *p;
275
276#ifdef CONFIG_DEBUG_PAGEALLOC
277 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
278#else
279 p = get_freepointer(s, object);
280#endif
281 return p;
282}
283
Christoph Lameter7656c722007-05-09 02:32:40 -0700284static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
285{
286 *(void **)(object + s->offset) = fp;
287}
288
289/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300290#define for_each_object(__p, __s, __addr, __objects) \
291 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700292 __p += (__s)->size)
293
Christoph Lameter7656c722007-05-09 02:32:40 -0700294/* Determine object index from a given position */
295static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
296{
297 return (p - addr) / s->size;
298}
299
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100300static inline size_t slab_ksize(const struct kmem_cache *s)
301{
302#ifdef CONFIG_SLUB_DEBUG
303 /*
304 * Debugging requires use of the padding between object
305 * and whatever may come after it.
306 */
307 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
308 return s->objsize;
309
310#endif
311 /*
312 * If we have the need to store the freelist pointer
313 * back there or track user information then we can
314 * only use the space before that information.
315 */
316 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
317 return s->inuse;
318 /*
319 * Else we can use all the padding etc for the allocation
320 */
321 return s->size;
322}
323
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800324static inline int order_objects(int order, unsigned long size, int reserved)
325{
326 return ((PAGE_SIZE << order) - reserved) / size;
327}
328
Christoph Lameter834f3d12008-04-14 19:11:31 +0300329static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800330 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300331{
332 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800333 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300334 };
335
336 return x;
337}
338
339static inline int oo_order(struct kmem_cache_order_objects x)
340{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400341 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300342}
343
344static inline int oo_objects(struct kmem_cache_order_objects x)
345{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400346 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300347}
348
Christoph Lameter881db7f2011-06-01 12:25:53 -0500349/*
350 * Per slab locking using the pagelock
351 */
352static __always_inline void slab_lock(struct page *page)
353{
354 bit_spin_lock(PG_locked, &page->flags);
355}
356
357static __always_inline void slab_unlock(struct page *page)
358{
359 __bit_spin_unlock(PG_locked, &page->flags);
360}
361
Christoph Lameter1d071712011-07-14 12:49:12 -0500362/* Interrupts must be disabled (for the fallback code to work right) */
363static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500364 void *freelist_old, unsigned long counters_old,
365 void *freelist_new, unsigned long counters_new,
366 const char *n)
367{
Christoph Lameter1d071712011-07-14 12:49:12 -0500368 VM_BUG_ON(!irqs_disabled());
Christoph Lameterb789ef52011-06-01 12:25:49 -0500369#ifdef CONFIG_CMPXCHG_DOUBLE
370 if (s->flags & __CMPXCHG_DOUBLE) {
371 if (cmpxchg_double(&page->freelist,
372 freelist_old, counters_old,
373 freelist_new, counters_new))
374 return 1;
375 } else
376#endif
377 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500378 slab_lock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500379 if (page->freelist == freelist_old && page->counters == counters_old) {
380 page->freelist = freelist_new;
381 page->counters = counters_new;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500382 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500383 return 1;
384 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500385 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500386 }
387
388 cpu_relax();
389 stat(s, CMPXCHG_DOUBLE_FAIL);
390
391#ifdef SLUB_DEBUG_CMPXCHG
392 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
393#endif
394
395 return 0;
396}
397
Christoph Lameter1d071712011-07-14 12:49:12 -0500398static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
399 void *freelist_old, unsigned long counters_old,
400 void *freelist_new, unsigned long counters_new,
401 const char *n)
402{
403#ifdef CONFIG_CMPXCHG_DOUBLE
404 if (s->flags & __CMPXCHG_DOUBLE) {
405 if (cmpxchg_double(&page->freelist,
406 freelist_old, counters_old,
407 freelist_new, counters_new))
408 return 1;
409 } else
410#endif
411 {
412 unsigned long flags;
413
414 local_irq_save(flags);
415 slab_lock(page);
416 if (page->freelist == freelist_old && page->counters == counters_old) {
417 page->freelist = freelist_new;
418 page->counters = counters_new;
419 slab_unlock(page);
420 local_irq_restore(flags);
421 return 1;
422 }
423 slab_unlock(page);
424 local_irq_restore(flags);
425 }
426
427 cpu_relax();
428 stat(s, CMPXCHG_DOUBLE_FAIL);
429
430#ifdef SLUB_DEBUG_CMPXCHG
431 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
432#endif
433
434 return 0;
435}
436
Christoph Lameter41ecc552007-05-09 02:32:44 -0700437#ifdef CONFIG_SLUB_DEBUG
438/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500439 * Determine a map of object in use on a page.
440 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500441 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500442 * not vanish from under us.
443 */
444static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
445{
446 void *p;
447 void *addr = page_address(page);
448
449 for (p = page->freelist; p; p = get_freepointer(s, p))
450 set_bit(slab_index(p, s, addr), map);
451}
452
Christoph Lameter41ecc552007-05-09 02:32:44 -0700453/*
454 * Debug settings:
455 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700456#ifdef CONFIG_SLUB_DEBUG_ON
457static int slub_debug = DEBUG_DEFAULT_FLAGS;
458#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700459static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700460#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700461
462static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700463static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700464
Christoph Lameter7656c722007-05-09 02:32:40 -0700465/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700466 * Object debugging
467 */
468static void print_section(char *text, u8 *addr, unsigned int length)
469{
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200470 print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
471 length, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700472}
473
Christoph Lameter81819f02007-05-06 14:49:36 -0700474static struct track *get_track(struct kmem_cache *s, void *object,
475 enum track_item alloc)
476{
477 struct track *p;
478
479 if (s->offset)
480 p = object + s->offset + sizeof(void *);
481 else
482 p = object + s->inuse;
483
484 return p + alloc;
485}
486
487static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300488 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700489{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900490 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700491
Christoph Lameter81819f02007-05-06 14:49:36 -0700492 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700493#ifdef CONFIG_STACKTRACE
494 struct stack_trace trace;
495 int i;
496
497 trace.nr_entries = 0;
498 trace.max_entries = TRACK_ADDRS_COUNT;
499 trace.entries = p->addrs;
500 trace.skip = 3;
501 save_stack_trace(&trace);
502
503 /* See rant in lockdep.c */
504 if (trace.nr_entries != 0 &&
505 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
506 trace.nr_entries--;
507
508 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
509 p->addrs[i] = 0;
510#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700511 p->addr = addr;
512 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400513 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700514 p->when = jiffies;
515 } else
516 memset(p, 0, sizeof(struct track));
517}
518
Christoph Lameter81819f02007-05-06 14:49:36 -0700519static void init_tracking(struct kmem_cache *s, void *object)
520{
Christoph Lameter24922682007-07-17 04:03:18 -0700521 if (!(s->flags & SLAB_STORE_USER))
522 return;
523
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300524 set_track(s, object, TRACK_FREE, 0UL);
525 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700526}
527
528static void print_track(const char *s, struct track *t)
529{
530 if (!t->addr)
531 return;
532
Linus Torvalds7daf7052008-07-14 12:12:53 -0700533 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300534 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700535#ifdef CONFIG_STACKTRACE
536 {
537 int i;
538 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
539 if (t->addrs[i])
540 printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
541 else
542 break;
543 }
544#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700545}
546
Christoph Lameter24922682007-07-17 04:03:18 -0700547static void print_tracking(struct kmem_cache *s, void *object)
548{
549 if (!(s->flags & SLAB_STORE_USER))
550 return;
551
552 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
553 print_track("Freed", get_track(s, object, TRACK_FREE));
554}
555
556static void print_page_info(struct page *page)
557{
Christoph Lameter39b26462008-04-14 19:11:30 +0300558 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
559 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700560
561}
562
563static void slab_bug(struct kmem_cache *s, char *fmt, ...)
564{
565 va_list args;
566 char buf[100];
567
568 va_start(args, fmt);
569 vsnprintf(buf, sizeof(buf), fmt, args);
570 va_end(args);
571 printk(KERN_ERR "========================================"
572 "=====================================\n");
573 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
574 printk(KERN_ERR "----------------------------------------"
575 "-------------------------------------\n\n");
576}
577
578static void slab_fix(struct kmem_cache *s, char *fmt, ...)
579{
580 va_list args;
581 char buf[100];
582
583 va_start(args, fmt);
584 vsnprintf(buf, sizeof(buf), fmt, args);
585 va_end(args);
586 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
587}
588
589static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700590{
591 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800592 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700593
594 print_tracking(s, p);
595
596 print_page_info(page);
597
598 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
599 p, p - addr, get_freepointer(s, p));
600
601 if (p > addr + 16)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200602 print_section("Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700603
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200604 print_section("Object ", p, min_t(unsigned long, s->objsize,
605 PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700606 if (s->flags & SLAB_RED_ZONE)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200607 print_section("Redzone ", p + s->objsize,
Christoph Lameter81819f02007-05-06 14:49:36 -0700608 s->inuse - s->objsize);
609
Christoph Lameter81819f02007-05-06 14:49:36 -0700610 if (s->offset)
611 off = s->offset + sizeof(void *);
612 else
613 off = s->inuse;
614
Christoph Lameter24922682007-07-17 04:03:18 -0700615 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700616 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700617
618 if (off != s->size)
619 /* Beginning of the filler is the free pointer */
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200620 print_section("Padding ", p + off, s->size - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700621
622 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700623}
624
625static void object_err(struct kmem_cache *s, struct page *page,
626 u8 *object, char *reason)
627{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700628 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700629 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700630}
631
Christoph Lameter24922682007-07-17 04:03:18 -0700632static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700633{
634 va_list args;
635 char buf[100];
636
Christoph Lameter24922682007-07-17 04:03:18 -0700637 va_start(args, fmt);
638 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700639 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700640 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700641 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700642 dump_stack();
643}
644
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500645static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700646{
647 u8 *p = object;
648
649 if (s->flags & __OBJECT_POISON) {
650 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800651 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700652 }
653
654 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500655 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700656}
657
Christoph Lameter24922682007-07-17 04:03:18 -0700658static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
659 void *from, void *to)
660{
661 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
662 memset(from, data, to - from);
663}
664
665static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
666 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800667 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700668{
669 u8 *fault;
670 u8 *end;
671
Akinobu Mita79824822011-10-31 17:08:07 -0700672 fault = memchr_inv(start, value, bytes);
Christoph Lameter24922682007-07-17 04:03:18 -0700673 if (!fault)
674 return 1;
675
676 end = start + bytes;
677 while (end > fault && end[-1] == value)
678 end--;
679
680 slab_bug(s, "%s overwritten", what);
681 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
682 fault, end - 1, fault[0], value);
683 print_trailer(s, page, object);
684
685 restore_bytes(s, what, value, fault, end);
686 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700687}
688
Christoph Lameter81819f02007-05-06 14:49:36 -0700689/*
690 * Object layout:
691 *
692 * object address
693 * Bytes of the object to be managed.
694 * If the freepointer may overlay the object then the free
695 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700696 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700697 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
698 * 0xa5 (POISON_END)
699 *
700 * object + s->objsize
701 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700702 * Padding is extended by another word if Redzoning is enabled and
703 * objsize == inuse.
704 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700705 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
706 * 0xcc (RED_ACTIVE) for objects in use.
707 *
708 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700709 * Meta data starts here.
710 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700711 * A. Free pointer (if we cannot overwrite object on free)
712 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700713 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800714 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700715 * before the word boundary.
716 *
717 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700718 *
719 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700720 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700721 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700722 * If slabcaches are merged then the objsize and inuse boundaries are mostly
723 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700724 * may be used with merged slabcaches.
725 */
726
Christoph Lameter81819f02007-05-06 14:49:36 -0700727static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
728{
729 unsigned long off = s->inuse; /* The end of info */
730
731 if (s->offset)
732 /* Freepointer is placed after the object. */
733 off += sizeof(void *);
734
735 if (s->flags & SLAB_STORE_USER)
736 /* We also have user information there */
737 off += 2 * sizeof(struct track);
738
739 if (s->size == off)
740 return 1;
741
Christoph Lameter24922682007-07-17 04:03:18 -0700742 return check_bytes_and_report(s, page, p, "Object padding",
743 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700744}
745
Christoph Lameter39b26462008-04-14 19:11:30 +0300746/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700747static int slab_pad_check(struct kmem_cache *s, struct page *page)
748{
Christoph Lameter24922682007-07-17 04:03:18 -0700749 u8 *start;
750 u8 *fault;
751 u8 *end;
752 int length;
753 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700754
755 if (!(s->flags & SLAB_POISON))
756 return 1;
757
Christoph Lametera973e9d2008-03-01 13:40:44 -0800758 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800759 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300760 end = start + length;
761 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700762 if (!remainder)
763 return 1;
764
Akinobu Mita79824822011-10-31 17:08:07 -0700765 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700766 if (!fault)
767 return 1;
768 while (end > fault && end[-1] == POISON_INUSE)
769 end--;
770
771 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200772 print_section("Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700773
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200774 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700775 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700776}
777
778static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500779 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700780{
781 u8 *p = object;
782 u8 *endobject = object + s->objsize;
783
784 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700785 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500786 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700787 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700788 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800789 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
790 check_bytes_and_report(s, page, p, "Alignment padding",
791 endobject, POISON_INUSE, s->inuse - s->objsize);
792 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700793 }
794
795 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500796 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700797 (!check_bytes_and_report(s, page, p, "Poison", p,
798 POISON_FREE, s->objsize - 1) ||
799 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800800 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700801 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700802 /*
803 * check_pad_bytes cleans up on its own.
804 */
805 check_pad_bytes(s, page, p);
806 }
807
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500808 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700809 /*
810 * Object and freepointer overlap. Cannot check
811 * freepointer while object is allocated.
812 */
813 return 1;
814
815 /* Check free pointer validity */
816 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
817 object_err(s, page, p, "Freepointer corrupt");
818 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100819 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700820 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700821 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700822 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800823 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700824 return 0;
825 }
826 return 1;
827}
828
829static int check_slab(struct kmem_cache *s, struct page *page)
830{
Christoph Lameter39b26462008-04-14 19:11:30 +0300831 int maxobj;
832
Christoph Lameter81819f02007-05-06 14:49:36 -0700833 VM_BUG_ON(!irqs_disabled());
834
835 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700836 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700837 return 0;
838 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300839
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800840 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300841 if (page->objects > maxobj) {
842 slab_err(s, page, "objects %u > max %u",
843 s->name, page->objects, maxobj);
844 return 0;
845 }
846 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700847 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300848 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700849 return 0;
850 }
851 /* Slab_pad_check fixes things up after itself */
852 slab_pad_check(s, page);
853 return 1;
854}
855
856/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700857 * Determine if a certain object on a page is on the freelist. Must hold the
858 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700859 */
860static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
861{
862 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500863 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700864 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300865 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700866
Christoph Lameter881db7f2011-06-01 12:25:53 -0500867 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300868 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700869 if (fp == search)
870 return 1;
871 if (!check_valid_pointer(s, page, fp)) {
872 if (object) {
873 object_err(s, page, object,
874 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800875 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700876 break;
877 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700878 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800879 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300880 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700881 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700882 return 0;
883 }
884 break;
885 }
886 object = fp;
887 fp = get_freepointer(s, object);
888 nr++;
889 }
890
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800891 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400892 if (max_objects > MAX_OBJS_PER_PAGE)
893 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300894
895 if (page->objects != max_objects) {
896 slab_err(s, page, "Wrong number of objects. Found %d but "
897 "should be %d", page->objects, max_objects);
898 page->objects = max_objects;
899 slab_fix(s, "Number of objects adjusted.");
900 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300901 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700902 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300903 "counted were %d", page->inuse, page->objects - nr);
904 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700905 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700906 }
907 return search == NULL;
908}
909
Christoph Lameter0121c6192008-04-29 16:11:12 -0700910static void trace(struct kmem_cache *s, struct page *page, void *object,
911 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700912{
913 if (s->flags & SLAB_TRACE) {
914 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
915 s->name,
916 alloc ? "alloc" : "free",
917 object, page->inuse,
918 page->freelist);
919
920 if (!alloc)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200921 print_section("Object ", (void *)object, s->objsize);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700922
923 dump_stack();
924 }
925}
926
Christoph Lameter643b1132007-05-06 14:49:42 -0700927/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500928 * Hooks for other subsystems that check memory allocations. In a typical
929 * production configuration these hooks all should produce no code at all.
930 */
931static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
932{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500933 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500934 lockdep_trace_alloc(flags);
935 might_sleep_if(flags & __GFP_WAIT);
936
937 return should_failslab(s->objsize, flags, s->flags);
938}
939
940static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
941{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500942 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100943 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500944 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
945}
946
947static inline void slab_free_hook(struct kmem_cache *s, void *x)
948{
949 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500950
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600951 /*
952 * Trouble is that we may no longer disable interupts in the fast path
953 * So in order to make the debug calls that expect irqs to be
954 * disabled we need to disable interrupts temporarily.
955 */
956#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
957 {
958 unsigned long flags;
959
960 local_irq_save(flags);
961 kmemcheck_slab_free(s, x, s->objsize);
962 debug_check_no_locks_freed(x, s->objsize);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600963 local_irq_restore(flags);
964 }
965#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200966 if (!(s->flags & SLAB_DEBUG_OBJECTS))
967 debug_check_no_obj_freed(x, s->objsize);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500968}
969
970/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700971 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500972 *
973 * list_lock must be held.
Christoph Lameter643b1132007-05-06 14:49:42 -0700974 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500975static void add_full(struct kmem_cache *s,
976 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700977{
Christoph Lameter643b1132007-05-06 14:49:42 -0700978 if (!(s->flags & SLAB_STORE_USER))
979 return;
980
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500981 list_add(&page->lru, &n->full);
982}
Christoph Lameter643b1132007-05-06 14:49:42 -0700983
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500984/*
985 * list_lock must be held.
986 */
987static void remove_full(struct kmem_cache *s, struct page *page)
988{
989 if (!(s->flags & SLAB_STORE_USER))
990 return;
991
Christoph Lameter643b1132007-05-06 14:49:42 -0700992 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -0700993}
994
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300995/* Tracking of the number of slabs for debugging purposes */
996static inline unsigned long slabs_node(struct kmem_cache *s, int node)
997{
998 struct kmem_cache_node *n = get_node(s, node);
999
1000 return atomic_long_read(&n->nr_slabs);
1001}
1002
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001003static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1004{
1005 return atomic_long_read(&n->nr_slabs);
1006}
1007
Christoph Lameter205ab992008-04-14 19:11:40 +03001008static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001009{
1010 struct kmem_cache_node *n = get_node(s, node);
1011
1012 /*
1013 * May be called early in order to allocate a slab for the
1014 * kmem_cache_node structure. Solve the chicken-egg
1015 * dilemma by deferring the increment of the count during
1016 * bootstrap (see early_kmem_cache_node_alloc).
1017 */
Christoph Lameter7340cc82010-09-28 08:10:26 -05001018 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001019 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001020 atomic_long_add(objects, &n->total_objects);
1021 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001022}
Christoph Lameter205ab992008-04-14 19:11:40 +03001023static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001024{
1025 struct kmem_cache_node *n = get_node(s, node);
1026
1027 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001028 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001029}
1030
1031/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001032static void setup_object_debug(struct kmem_cache *s, struct page *page,
1033 void *object)
1034{
1035 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1036 return;
1037
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001038 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001039 init_tracking(s, object);
1040}
1041
Christoph Lameter15370662010-08-20 12:37:12 -05001042static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001043 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001044{
1045 if (!check_slab(s, page))
1046 goto bad;
1047
Christoph Lameter81819f02007-05-06 14:49:36 -07001048 if (!check_valid_pointer(s, page, object)) {
1049 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001050 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001051 }
1052
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001053 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001054 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001055
Christoph Lameter3ec09742007-05-16 22:11:00 -07001056 /* Success perform special debug activities for allocs */
1057 if (s->flags & SLAB_STORE_USER)
1058 set_track(s, object, TRACK_ALLOC, addr);
1059 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001060 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001061 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001062
Christoph Lameter81819f02007-05-06 14:49:36 -07001063bad:
1064 if (PageSlab(page)) {
1065 /*
1066 * If this is a slab page then lets do the best we can
1067 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001068 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001069 */
Christoph Lameter24922682007-07-17 04:03:18 -07001070 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001071 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001072 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001073 }
1074 return 0;
1075}
1076
Christoph Lameter15370662010-08-20 12:37:12 -05001077static noinline int free_debug_processing(struct kmem_cache *s,
1078 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001079{
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001080 unsigned long flags;
1081 int rc = 0;
1082
1083 local_irq_save(flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001084 slab_lock(page);
1085
Christoph Lameter81819f02007-05-06 14:49:36 -07001086 if (!check_slab(s, page))
1087 goto fail;
1088
1089 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001090 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001091 goto fail;
1092 }
1093
1094 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001095 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001096 goto fail;
1097 }
1098
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001099 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001100 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001101
1102 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001103 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001104 slab_err(s, page, "Attempt to free object(0x%p) "
1105 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001106 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001107 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001108 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001109 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001110 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001111 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001112 object_err(s, page, object,
1113 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001114 goto fail;
1115 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001116
Christoph Lameter3ec09742007-05-16 22:11:00 -07001117 if (s->flags & SLAB_STORE_USER)
1118 set_track(s, object, TRACK_FREE, addr);
1119 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001120 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001121 rc = 1;
1122out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001123 slab_unlock(page);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001124 local_irq_restore(flags);
1125 return rc;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001126
Christoph Lameter81819f02007-05-06 14:49:36 -07001127fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001128 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001129 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001130}
1131
Christoph Lameter41ecc552007-05-09 02:32:44 -07001132static int __init setup_slub_debug(char *str)
1133{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001134 slub_debug = DEBUG_DEFAULT_FLAGS;
1135 if (*str++ != '=' || !*str)
1136 /*
1137 * No options specified. Switch on full debugging.
1138 */
1139 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001140
1141 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001142 /*
1143 * No options but restriction on slabs. This means full
1144 * debugging for slabs matching a pattern.
1145 */
1146 goto check_slabs;
1147
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001148 if (tolower(*str) == 'o') {
1149 /*
1150 * Avoid enabling debugging on caches if its minimum order
1151 * would increase as a result.
1152 */
1153 disable_higher_order_debug = 1;
1154 goto out;
1155 }
1156
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001157 slub_debug = 0;
1158 if (*str == '-')
1159 /*
1160 * Switch off all debugging measures.
1161 */
1162 goto out;
1163
1164 /*
1165 * Determine which debug features should be switched on
1166 */
Pekka Enberg06428782008-01-07 23:20:27 -08001167 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001168 switch (tolower(*str)) {
1169 case 'f':
1170 slub_debug |= SLAB_DEBUG_FREE;
1171 break;
1172 case 'z':
1173 slub_debug |= SLAB_RED_ZONE;
1174 break;
1175 case 'p':
1176 slub_debug |= SLAB_POISON;
1177 break;
1178 case 'u':
1179 slub_debug |= SLAB_STORE_USER;
1180 break;
1181 case 't':
1182 slub_debug |= SLAB_TRACE;
1183 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001184 case 'a':
1185 slub_debug |= SLAB_FAILSLAB;
1186 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001187 default:
1188 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001189 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001190 }
1191 }
1192
1193check_slabs:
1194 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001195 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001196out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001197 return 1;
1198}
1199
1200__setup("slub_debug", setup_slub_debug);
1201
Christoph Lameterba0268a2007-09-11 15:24:11 -07001202static unsigned long kmem_cache_flags(unsigned long objsize,
1203 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001204 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001205{
1206 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001207 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001208 */
Christoph Lametere1533622008-02-15 23:45:24 -08001209 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001210 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1211 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001212
1213 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001214}
1215#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001216static inline void setup_object_debug(struct kmem_cache *s,
1217 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001218
Christoph Lameter3ec09742007-05-16 22:11:00 -07001219static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001220 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001221
Christoph Lameter3ec09742007-05-16 22:11:00 -07001222static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001223 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001224
Christoph Lameter41ecc552007-05-09 02:32:44 -07001225static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1226 { return 1; }
1227static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001228 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001229static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1230 struct page *page) {}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001231static inline void remove_full(struct kmem_cache *s, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001232static inline unsigned long kmem_cache_flags(unsigned long objsize,
1233 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001234 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001235{
1236 return flags;
1237}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001238#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001239
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001240#define disable_higher_order_debug 0
1241
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001242static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1243 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001244static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1245 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001246static inline void inc_slabs_node(struct kmem_cache *s, int node,
1247 int objects) {}
1248static inline void dec_slabs_node(struct kmem_cache *s, int node,
1249 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001250
1251static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1252 { return 0; }
1253
1254static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1255 void *object) {}
1256
1257static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1258
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001259#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001260
Christoph Lameter81819f02007-05-06 14:49:36 -07001261/*
1262 * Slab allocation and freeing
1263 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001264static inline struct page *alloc_slab_page(gfp_t flags, int node,
1265 struct kmem_cache_order_objects oo)
1266{
1267 int order = oo_order(oo);
1268
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001269 flags |= __GFP_NOTRACK;
1270
Christoph Lameter2154a332010-07-09 14:07:10 -05001271 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001272 return alloc_pages(flags, order);
1273 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001274 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001275}
1276
Christoph Lameter81819f02007-05-06 14:49:36 -07001277static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1278{
Pekka Enberg06428782008-01-07 23:20:27 -08001279 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001280 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001281 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001282
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001283 flags &= gfp_allowed_mask;
1284
1285 if (flags & __GFP_WAIT)
1286 local_irq_enable();
1287
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001288 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001289
Pekka Enbergba522702009-06-24 21:59:51 +03001290 /*
1291 * Let the initial higher-order allocation fail under memory pressure
1292 * so we fall-back to the minimum order allocation.
1293 */
1294 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1295
1296 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001297 if (unlikely(!page)) {
1298 oo = s->min;
1299 /*
1300 * Allocation may have failed due to fragmentation.
1301 * Try a lower order alloc if possible
1302 */
1303 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001304
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001305 if (page)
1306 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001307 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001308
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001309 if (flags & __GFP_WAIT)
1310 local_irq_disable();
1311
1312 if (!page)
1313 return NULL;
1314
Vegard Nossum5a896d92008-04-04 00:54:48 +02001315 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001316 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001317 int pages = 1 << oo_order(oo);
1318
1319 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1320
1321 /*
1322 * Objects from caches that have a constructor don't get
1323 * cleared when they're allocated, so we need to do it here.
1324 */
1325 if (s->ctor)
1326 kmemcheck_mark_uninitialized_pages(page, pages);
1327 else
1328 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001329 }
1330
Christoph Lameter834f3d12008-04-14 19:11:31 +03001331 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001332 mod_zone_page_state(page_zone(page),
1333 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1334 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001335 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001336
1337 return page;
1338}
1339
1340static void setup_object(struct kmem_cache *s, struct page *page,
1341 void *object)
1342{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001343 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001344 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001345 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001346}
1347
1348static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1349{
1350 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001351 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001352 void *last;
1353 void *p;
1354
Christoph Lameter6cb06222007-10-16 01:25:41 -07001355 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001356
Christoph Lameter6cb06222007-10-16 01:25:41 -07001357 page = allocate_slab(s,
1358 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001359 if (!page)
1360 goto out;
1361
Christoph Lameter205ab992008-04-14 19:11:40 +03001362 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001363 page->slab = s;
1364 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001365
1366 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001367
1368 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001369 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001370
1371 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001372 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001373 setup_object(s, page, last);
1374 set_freepointer(s, last, p);
1375 last = p;
1376 }
1377 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001378 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001379
1380 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001381 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001382 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001383out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001384 return page;
1385}
1386
1387static void __free_slab(struct kmem_cache *s, struct page *page)
1388{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001389 int order = compound_order(page);
1390 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001391
Christoph Lameteraf537b02010-07-09 14:07:14 -05001392 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001393 void *p;
1394
1395 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001396 for_each_object(p, s, page_address(page),
1397 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001398 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001399 }
1400
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001401 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001402
Christoph Lameter81819f02007-05-06 14:49:36 -07001403 mod_zone_page_state(page_zone(page),
1404 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1405 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001406 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001407
Christoph Lameter49bd5222008-04-14 18:52:18 +03001408 __ClearPageSlab(page);
1409 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001410 if (current->reclaim_state)
1411 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001412 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001413}
1414
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001415#define need_reserve_slab_rcu \
1416 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1417
Christoph Lameter81819f02007-05-06 14:49:36 -07001418static void rcu_free_slab(struct rcu_head *h)
1419{
1420 struct page *page;
1421
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001422 if (need_reserve_slab_rcu)
1423 page = virt_to_head_page(h);
1424 else
1425 page = container_of((struct list_head *)h, struct page, lru);
1426
Christoph Lameter81819f02007-05-06 14:49:36 -07001427 __free_slab(page->slab, page);
1428}
1429
1430static void free_slab(struct kmem_cache *s, struct page *page)
1431{
1432 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001433 struct rcu_head *head;
1434
1435 if (need_reserve_slab_rcu) {
1436 int order = compound_order(page);
1437 int offset = (PAGE_SIZE << order) - s->reserved;
1438
1439 VM_BUG_ON(s->reserved != sizeof(*head));
1440 head = page_address(page) + offset;
1441 } else {
1442 /*
1443 * RCU free overloads the RCU head over the LRU
1444 */
1445 head = (void *)&page->lru;
1446 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001447
1448 call_rcu(head, rcu_free_slab);
1449 } else
1450 __free_slab(s, page);
1451}
1452
1453static void discard_slab(struct kmem_cache *s, struct page *page)
1454{
Christoph Lameter205ab992008-04-14 19:11:40 +03001455 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001456 free_slab(s, page);
1457}
1458
1459/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001460 * Management of partially allocated slabs.
1461 *
1462 * list_lock must be held.
Christoph Lameter81819f02007-05-06 14:49:36 -07001463 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001464static inline void add_partial(struct kmem_cache_node *n,
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001465 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001466{
Christoph Lametere95eed52007-05-06 14:49:44 -07001467 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001468 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001469 list_add_tail(&page->lru, &n->partial);
1470 else
1471 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001472}
1473
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001474/*
1475 * list_lock must be held.
1476 */
1477static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001478 struct page *page)
1479{
1480 list_del(&page->lru);
1481 n->nr_partial--;
1482}
1483
Christoph Lameter81819f02007-05-06 14:49:36 -07001484/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001485 * Lock slab, remove from the partial list and put the object into the
1486 * per cpu freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001487 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001488 * Returns a list of objects or NULL if it fails.
1489 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001490 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001491 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001492static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001493 struct kmem_cache_node *n, struct page *page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001494 int mode)
Christoph Lameter81819f02007-05-06 14:49:36 -07001495{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001496 void *freelist;
1497 unsigned long counters;
1498 struct page new;
1499
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001500 /*
1501 * Zap the freelist and set the frozen bit.
1502 * The old freelist is the list of objects for the
1503 * per cpu allocation list.
1504 */
1505 do {
1506 freelist = page->freelist;
1507 counters = page->counters;
1508 new.counters = counters;
Christoph Lameter49e22582011-08-09 16:12:27 -05001509 if (mode)
1510 new.inuse = page->objects;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001511
1512 VM_BUG_ON(new.frozen);
1513 new.frozen = 1;
1514
Christoph Lameter1d071712011-07-14 12:49:12 -05001515 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001516 freelist, counters,
1517 NULL, new.counters,
1518 "lock and freeze"));
1519
1520 remove_partial(n, page);
Christoph Lameter49e22582011-08-09 16:12:27 -05001521 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001522}
1523
Christoph Lameter49e22582011-08-09 16:12:27 -05001524static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1525
Christoph Lameter81819f02007-05-06 14:49:36 -07001526/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001527 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001528 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001529static void *get_partial_node(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001530 struct kmem_cache_node *n, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001531{
Christoph Lameter49e22582011-08-09 16:12:27 -05001532 struct page *page, *page2;
1533 void *object = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001534
1535 /*
1536 * Racy check. If we mistakenly see no partial slabs then we
1537 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001538 * partial slab and there is none available then get_partials()
1539 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001540 */
1541 if (!n || !n->nr_partial)
1542 return NULL;
1543
1544 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001545 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Alex,Shi12d79632011-09-07 10:26:36 +08001546 void *t = acquire_slab(s, n, page, object == NULL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001547 int available;
1548
1549 if (!t)
1550 break;
1551
Alex,Shi12d79632011-09-07 10:26:36 +08001552 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001553 c->page = page;
1554 c->node = page_to_nid(page);
1555 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001556 object = t;
1557 available = page->objects - page->inuse;
1558 } else {
1559 page->freelist = t;
1560 available = put_cpu_partial(s, page, 0);
1561 }
1562 if (kmem_cache_debug(s) || available > s->cpu_partial / 2)
1563 break;
1564
Christoph Lameter497b66f2011-08-09 16:12:26 -05001565 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001566 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001567 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001568}
1569
1570/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001571 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001572 */
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001573static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags,
1574 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001575{
1576#ifdef CONFIG_NUMA
1577 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001578 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001579 struct zone *zone;
1580 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001581 void *object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001582
1583 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001584 * The defrag ratio allows a configuration of the tradeoffs between
1585 * inter node defragmentation and node local allocations. A lower
1586 * defrag_ratio increases the tendency to do local allocations
1587 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001588 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001589 * If the defrag_ratio is set to 0 then kmalloc() always
1590 * returns node local objects. If the ratio is higher then kmalloc()
1591 * may return off node objects because partial slabs are obtained
1592 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001593 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001594 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001595 * defrag_ratio = 1000) then every (well almost) allocation will
1596 * first attempt to defrag slab caches on other nodes. This means
1597 * scanning over all nodes to look for partial slabs which may be
1598 * expensive if we do it every time we are trying to find a slab
1599 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001600 */
Christoph Lameter98246012008-01-07 23:20:26 -08001601 if (!s->remote_node_defrag_ratio ||
1602 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001603 return NULL;
1604
Miao Xiec0ff7452010-05-24 14:32:08 -07001605 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001606 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001607 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001608 struct kmem_cache_node *n;
1609
Mel Gorman54a6eb52008-04-28 02:12:16 -07001610 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001611
Mel Gorman54a6eb52008-04-28 02:12:16 -07001612 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001613 n->nr_partial > s->min_partial) {
Christoph Lameter497b66f2011-08-09 16:12:26 -05001614 object = get_partial_node(s, n, c);
1615 if (object) {
Miao Xiec0ff7452010-05-24 14:32:08 -07001616 put_mems_allowed();
Christoph Lameter497b66f2011-08-09 16:12:26 -05001617 return object;
Miao Xiec0ff7452010-05-24 14:32:08 -07001618 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001619 }
1620 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001621 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001622#endif
1623 return NULL;
1624}
1625
1626/*
1627 * Get a partial page, lock it and return it.
1628 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001629static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001630 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001631{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001632 void *object;
Christoph Lameter2154a332010-07-09 14:07:10 -05001633 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001634
Christoph Lameter497b66f2011-08-09 16:12:26 -05001635 object = get_partial_node(s, get_node(s, searchnode), c);
1636 if (object || node != NUMA_NO_NODE)
1637 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001638
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001639 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001640}
1641
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001642#ifdef CONFIG_PREEMPT
1643/*
1644 * Calculate the next globally unique transaction for disambiguiation
1645 * during cmpxchg. The transactions start with the cpu number and are then
1646 * incremented by CONFIG_NR_CPUS.
1647 */
1648#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1649#else
1650/*
1651 * No preemption supported therefore also no need to check for
1652 * different cpus.
1653 */
1654#define TID_STEP 1
1655#endif
1656
1657static inline unsigned long next_tid(unsigned long tid)
1658{
1659 return tid + TID_STEP;
1660}
1661
1662static inline unsigned int tid_to_cpu(unsigned long tid)
1663{
1664 return tid % TID_STEP;
1665}
1666
1667static inline unsigned long tid_to_event(unsigned long tid)
1668{
1669 return tid / TID_STEP;
1670}
1671
1672static inline unsigned int init_tid(int cpu)
1673{
1674 return cpu;
1675}
1676
1677static inline void note_cmpxchg_failure(const char *n,
1678 const struct kmem_cache *s, unsigned long tid)
1679{
1680#ifdef SLUB_DEBUG_CMPXCHG
1681 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1682
1683 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1684
1685#ifdef CONFIG_PREEMPT
1686 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1687 printk("due to cpu change %d -> %d\n",
1688 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1689 else
1690#endif
1691 if (tid_to_event(tid) != tid_to_event(actual_tid))
1692 printk("due to cpu running other code. Event %ld->%ld\n",
1693 tid_to_event(tid), tid_to_event(actual_tid));
1694 else
1695 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1696 actual_tid, tid, next_tid(tid));
1697#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001698 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001699}
1700
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001701void init_kmem_cache_cpus(struct kmem_cache *s)
1702{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001703 int cpu;
1704
1705 for_each_possible_cpu(cpu)
1706 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001707}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001708
1709/*
1710 * Remove the cpu slab
1711 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001712static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001713{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001714 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001715 struct page *page = c->page;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001716 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1717 int lock = 0;
1718 enum slab_modes l = M_NONE, m = M_NONE;
1719 void *freelist;
1720 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001721 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001722 struct page new;
1723 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001724
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001725 if (page->freelist) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001726 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001727 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001728 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001729
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001730 c->tid = next_tid(c->tid);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001731 c->page = NULL;
1732 freelist = c->freelist;
1733 c->freelist = NULL;
1734
1735 /*
1736 * Stage one: Free all available per cpu objects back
1737 * to the page freelist while it is still frozen. Leave the
1738 * last one.
1739 *
1740 * There is no need to take the list->lock because the page
1741 * is still frozen.
1742 */
1743 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1744 void *prior;
1745 unsigned long counters;
1746
1747 do {
1748 prior = page->freelist;
1749 counters = page->counters;
1750 set_freepointer(s, freelist, prior);
1751 new.counters = counters;
1752 new.inuse--;
1753 VM_BUG_ON(!new.frozen);
1754
Christoph Lameter1d071712011-07-14 12:49:12 -05001755 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001756 prior, counters,
1757 freelist, new.counters,
1758 "drain percpu freelist"));
1759
1760 freelist = nextfree;
1761 }
1762
1763 /*
1764 * Stage two: Ensure that the page is unfrozen while the
1765 * list presence reflects the actual number of objects
1766 * during unfreeze.
1767 *
1768 * We setup the list membership and then perform a cmpxchg
1769 * with the count. If there is a mismatch then the page
1770 * is not unfrozen but the page is on the wrong list.
1771 *
1772 * Then we restart the process which may have to remove
1773 * the page from the list that we just put it on again
1774 * because the number of objects in the slab may have
1775 * changed.
1776 */
1777redo:
1778
1779 old.freelist = page->freelist;
1780 old.counters = page->counters;
1781 VM_BUG_ON(!old.frozen);
1782
1783 /* Determine target state of the slab */
1784 new.counters = old.counters;
1785 if (freelist) {
1786 new.inuse--;
1787 set_freepointer(s, freelist, old.freelist);
1788 new.freelist = freelist;
1789 } else
1790 new.freelist = old.freelist;
1791
1792 new.frozen = 0;
1793
Christoph Lameter81107182011-08-09 13:01:32 -05001794 if (!new.inuse && n->nr_partial > s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001795 m = M_FREE;
1796 else if (new.freelist) {
1797 m = M_PARTIAL;
1798 if (!lock) {
1799 lock = 1;
1800 /*
1801 * Taking the spinlock removes the possiblity
1802 * that acquire_slab() will see a slab page that
1803 * is frozen
1804 */
1805 spin_lock(&n->list_lock);
1806 }
1807 } else {
1808 m = M_FULL;
1809 if (kmem_cache_debug(s) && !lock) {
1810 lock = 1;
1811 /*
1812 * This also ensures that the scanning of full
1813 * slabs from diagnostic functions will not see
1814 * any frozen slabs.
1815 */
1816 spin_lock(&n->list_lock);
1817 }
1818 }
1819
1820 if (l != m) {
1821
1822 if (l == M_PARTIAL)
1823
1824 remove_partial(n, page);
1825
1826 else if (l == M_FULL)
1827
1828 remove_full(s, page);
1829
1830 if (m == M_PARTIAL) {
1831
1832 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001833 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001834
1835 } else if (m == M_FULL) {
1836
1837 stat(s, DEACTIVATE_FULL);
1838 add_full(s, n, page);
1839
1840 }
1841 }
1842
1843 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001844 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001845 old.freelist, old.counters,
1846 new.freelist, new.counters,
1847 "unfreezing slab"))
1848 goto redo;
1849
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001850 if (lock)
1851 spin_unlock(&n->list_lock);
1852
1853 if (m == M_FREE) {
1854 stat(s, DEACTIVATE_EMPTY);
1855 discard_slab(s, page);
1856 stat(s, FREE_SLAB);
1857 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001858}
1859
Christoph Lameter49e22582011-08-09 16:12:27 -05001860/* Unfreeze all the cpu partial slabs */
1861static void unfreeze_partials(struct kmem_cache *s)
1862{
1863 struct kmem_cache_node *n = NULL;
1864 struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab);
1865 struct page *page;
1866
1867 while ((page = c->partial)) {
1868 enum slab_modes { M_PARTIAL, M_FREE };
1869 enum slab_modes l, m;
1870 struct page new;
1871 struct page old;
1872
1873 c->partial = page->next;
1874 l = M_FREE;
1875
1876 do {
1877
1878 old.freelist = page->freelist;
1879 old.counters = page->counters;
1880 VM_BUG_ON(!old.frozen);
1881
1882 new.counters = old.counters;
1883 new.freelist = old.freelist;
1884
1885 new.frozen = 0;
1886
Alex Shidcc3be62011-09-06 14:46:01 +08001887 if (!new.inuse && (!n || n->nr_partial > s->min_partial))
Christoph Lameter49e22582011-08-09 16:12:27 -05001888 m = M_FREE;
1889 else {
1890 struct kmem_cache_node *n2 = get_node(s,
1891 page_to_nid(page));
1892
1893 m = M_PARTIAL;
1894 if (n != n2) {
1895 if (n)
1896 spin_unlock(&n->list_lock);
1897
1898 n = n2;
1899 spin_lock(&n->list_lock);
1900 }
1901 }
1902
1903 if (l != m) {
1904 if (l == M_PARTIAL)
1905 remove_partial(n, page);
1906 else
1907 add_partial(n, page, 1);
1908
1909 l = m;
1910 }
1911
1912 } while (!cmpxchg_double_slab(s, page,
1913 old.freelist, old.counters,
1914 new.freelist, new.counters,
1915 "unfreezing slab"));
1916
1917 if (m == M_FREE) {
1918 stat(s, DEACTIVATE_EMPTY);
1919 discard_slab(s, page);
1920 stat(s, FREE_SLAB);
1921 }
1922 }
1923
1924 if (n)
1925 spin_unlock(&n->list_lock);
1926}
1927
1928/*
1929 * Put a page that was just frozen (in __slab_free) into a partial page
1930 * slot if available. This is done without interrupts disabled and without
1931 * preemption disabled. The cmpxchg is racy and may put the partial page
1932 * onto a random cpus partial slot.
1933 *
1934 * If we did not find a slot then simply move all the partials to the
1935 * per node partial list.
1936 */
1937int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
1938{
1939 struct page *oldpage;
1940 int pages;
1941 int pobjects;
1942
1943 do {
1944 pages = 0;
1945 pobjects = 0;
1946 oldpage = this_cpu_read(s->cpu_slab->partial);
1947
1948 if (oldpage) {
1949 pobjects = oldpage->pobjects;
1950 pages = oldpage->pages;
1951 if (drain && pobjects > s->cpu_partial) {
1952 unsigned long flags;
1953 /*
1954 * partial array is full. Move the existing
1955 * set to the per node partial list.
1956 */
1957 local_irq_save(flags);
1958 unfreeze_partials(s);
1959 local_irq_restore(flags);
1960 pobjects = 0;
1961 pages = 0;
1962 }
1963 }
1964
1965 pages++;
1966 pobjects += page->objects - page->inuse;
1967
1968 page->pages = pages;
1969 page->pobjects = pobjects;
1970 page->next = oldpage;
1971
1972 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage);
1973 stat(s, CPU_PARTIAL_FREE);
1974 return pobjects;
1975}
1976
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001977static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001978{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001979 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001980 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001981}
1982
1983/*
1984 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001985 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001986 * Called from IPI handler with interrupts disabled.
1987 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001988static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001989{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001990 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001991
Christoph Lameter49e22582011-08-09 16:12:27 -05001992 if (likely(c)) {
1993 if (c->page)
1994 flush_slab(s, c);
1995
1996 unfreeze_partials(s);
1997 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001998}
1999
2000static void flush_cpu_slab(void *d)
2001{
2002 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002003
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002004 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002005}
2006
2007static void flush_all(struct kmem_cache *s)
2008{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002009 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002010}
2011
2012/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002013 * Check if the objects in a per cpu structure fit numa
2014 * locality expectations.
2015 */
2016static inline int node_match(struct kmem_cache_cpu *c, int node)
2017{
2018#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05002019 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002020 return 0;
2021#endif
2022 return 1;
2023}
2024
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002025static int count_free(struct page *page)
2026{
2027 return page->objects - page->inuse;
2028}
2029
2030static unsigned long count_partial(struct kmem_cache_node *n,
2031 int (*get_count)(struct page *))
2032{
2033 unsigned long flags;
2034 unsigned long x = 0;
2035 struct page *page;
2036
2037 spin_lock_irqsave(&n->list_lock, flags);
2038 list_for_each_entry(page, &n->partial, lru)
2039 x += get_count(page);
2040 spin_unlock_irqrestore(&n->list_lock, flags);
2041 return x;
2042}
2043
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002044static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2045{
2046#ifdef CONFIG_SLUB_DEBUG
2047 return atomic_long_read(&n->total_objects);
2048#else
2049 return 0;
2050#endif
2051}
2052
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002053static noinline void
2054slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2055{
2056 int node;
2057
2058 printk(KERN_WARNING
2059 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2060 nid, gfpflags);
2061 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
2062 "default order: %d, min order: %d\n", s->name, s->objsize,
2063 s->size, oo_order(s->oo), oo_order(s->min));
2064
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002065 if (oo_order(s->min) > get_order(s->objsize))
2066 printk(KERN_WARNING " %s debugging increased min order, use "
2067 "slub_debug=O to disable.\n", s->name);
2068
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002069 for_each_online_node(node) {
2070 struct kmem_cache_node *n = get_node(s, node);
2071 unsigned long nr_slabs;
2072 unsigned long nr_objs;
2073 unsigned long nr_free;
2074
2075 if (!n)
2076 continue;
2077
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002078 nr_free = count_partial(n, count_free);
2079 nr_slabs = node_nr_slabs(n);
2080 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002081
2082 printk(KERN_WARNING
2083 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2084 node, nr_slabs, nr_objs, nr_free);
2085 }
2086}
2087
Christoph Lameter497b66f2011-08-09 16:12:26 -05002088static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2089 int node, struct kmem_cache_cpu **pc)
2090{
2091 void *object;
2092 struct kmem_cache_cpu *c;
2093 struct page *page = new_slab(s, flags, node);
2094
2095 if (page) {
2096 c = __this_cpu_ptr(s->cpu_slab);
2097 if (c->page)
2098 flush_slab(s, c);
2099
2100 /*
2101 * No other reference to the page yet so we can
2102 * muck around with it freely without cmpxchg
2103 */
2104 object = page->freelist;
2105 page->freelist = NULL;
2106
2107 stat(s, ALLOC_SLAB);
2108 c->node = page_to_nid(page);
2109 c->page = page;
2110 *pc = c;
2111 } else
2112 object = NULL;
2113
2114 return object;
2115}
2116
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002117/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002118 * Slow path. The lockless freelist is empty or we need to perform
2119 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002120 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002121 * Processing is still very fast if new objects have been freed to the
2122 * regular freelist. In that case we simply take over the regular freelist
2123 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002124 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002125 * If that is not working then we fall back to the partial lists. We take the
2126 * first element of the freelist as the object to allocate now and move the
2127 * rest of the freelist to the lockless freelist.
2128 *
2129 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002130 * we need to allocate a new slab. This is the slowest path since it involves
2131 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002132 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002133static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2134 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002135{
Christoph Lameter81819f02007-05-06 14:49:36 -07002136 void **object;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002137 unsigned long flags;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002138 struct page new;
2139 unsigned long counters;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002140
2141 local_irq_save(flags);
2142#ifdef CONFIG_PREEMPT
2143 /*
2144 * We may have been preempted and rescheduled on a different
2145 * cpu before disabling interrupts. Need to reload cpu area
2146 * pointer.
2147 */
2148 c = this_cpu_ptr(s->cpu_slab);
2149#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002150
Christoph Lameter497b66f2011-08-09 16:12:26 -05002151 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002152 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002153redo:
Christoph Lameterfc59c052011-06-01 12:25:56 -05002154 if (unlikely(!node_match(c, node))) {
Christoph Lametere36a2652011-06-01 12:25:57 -05002155 stat(s, ALLOC_NODE_MISMATCH);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002156 deactivate_slab(s, c);
2157 goto new_slab;
2158 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002159
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002160 stat(s, ALLOC_SLOWPATH);
2161
2162 do {
Christoph Lameter497b66f2011-08-09 16:12:26 -05002163 object = c->page->freelist;
2164 counters = c->page->counters;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002165 new.counters = counters;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002166 VM_BUG_ON(!new.frozen);
2167
Christoph Lameter03e404a2011-06-01 12:25:58 -05002168 /*
2169 * If there is no object left then we use this loop to
2170 * deactivate the slab which is simple since no objects
2171 * are left in the slab and therefore we do not need to
2172 * put the page back onto the partial list.
2173 *
2174 * If there are objects left then we retrieve them
2175 * and use them to refill the per cpu queue.
Christoph Lameter497b66f2011-08-09 16:12:26 -05002176 */
Christoph Lameter03e404a2011-06-01 12:25:58 -05002177
Christoph Lameter497b66f2011-08-09 16:12:26 -05002178 new.inuse = c->page->objects;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002179 new.frozen = object != NULL;
2180
Christoph Lameter497b66f2011-08-09 16:12:26 -05002181 } while (!__cmpxchg_double_slab(s, c->page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002182 object, counters,
2183 NULL, new.counters,
2184 "__slab_alloc"));
Christoph Lameter6446faa2008-02-15 23:45:26 -08002185
Christoph Lameter49e22582011-08-09 16:12:27 -05002186 if (!object) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002187 c->page = NULL;
2188 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002189 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002190 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002191
Christoph Lameter81819f02007-05-06 14:49:36 -07002192 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002193
Christoph Lameter894b8782007-05-10 03:15:16 -07002194load_freelist:
Christoph Lameterff120592009-12-18 16:26:22 -06002195 c->freelist = get_freepointer(s, object);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002196 c->tid = next_tid(c->tid);
2197 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002198 return object;
2199
Christoph Lameter81819f02007-05-06 14:49:36 -07002200new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002201
Christoph Lameter49e22582011-08-09 16:12:27 -05002202 if (c->partial) {
2203 c->page = c->partial;
2204 c->partial = c->page->next;
2205 c->node = page_to_nid(c->page);
2206 stat(s, CPU_PARTIAL_ALLOC);
2207 c->freelist = NULL;
2208 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002209 }
2210
Christoph Lameter49e22582011-08-09 16:12:27 -05002211 /* Then do expensive stuff like retrieving pages from the partial lists */
Christoph Lameter497b66f2011-08-09 16:12:26 -05002212 object = get_partial(s, gfpflags, node, c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002213
Christoph Lameter497b66f2011-08-09 16:12:26 -05002214 if (unlikely(!object)) {
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002215
Christoph Lameter497b66f2011-08-09 16:12:26 -05002216 object = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002217
Christoph Lameter497b66f2011-08-09 16:12:26 -05002218 if (unlikely(!object)) {
2219 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
2220 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter9e577e82011-07-22 09:35:14 -05002221
Christoph Lameter497b66f2011-08-09 16:12:26 -05002222 local_irq_restore(flags);
2223 return NULL;
2224 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002225 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002226
Christoph Lameter497b66f2011-08-09 16:12:26 -05002227 if (likely(!kmem_cache_debug(s)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002228 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002229
Christoph Lameter497b66f2011-08-09 16:12:26 -05002230 /* Only entered in the debug case */
2231 if (!alloc_debug_processing(s, c->page, object, addr))
2232 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002233
2234 c->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05002235 deactivate_slab(s, c);
Pekka Enberg15b7c512010-10-02 11:32:32 +03002236 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05002237 local_irq_restore(flags);
2238 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07002239}
2240
2241/*
2242 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2243 * have the fastpath folded into their functions. So no function call
2244 * overhead for requests that can be satisfied on the fastpath.
2245 *
2246 * The fastpath works by first checking if the lockless freelist can be used.
2247 * If not then __slab_alloc is called for slow processing.
2248 *
2249 * Otherwise we can simply pick the next object from the lockless free list.
2250 */
Pekka Enberg06428782008-01-07 23:20:27 -08002251static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002252 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002253{
Christoph Lameter894b8782007-05-10 03:15:16 -07002254 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002255 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002256 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002257
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002258 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09002259 return NULL;
2260
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002261redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002262
2263 /*
2264 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2265 * enabled. We may switch back and forth between cpus while
2266 * reading from one cpu area. That does not matter as long
2267 * as we end up on the original cpu again when doing the cmpxchg.
2268 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002269 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002270
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002271 /*
2272 * The transaction ids are globally unique per cpu and per operation on
2273 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2274 * occurs on the right processor and that there was no operation on the
2275 * linked list in between.
2276 */
2277 tid = c->tid;
2278 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002279
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002280 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002281 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07002282
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002283 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07002284
2285 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002286 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002287 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002288 * operation and if we are on the right processor.
2289 *
2290 * The cmpxchg does the following atomically (without lock semantics!)
2291 * 1. Relocate first pointer to the current per cpu area.
2292 * 2. Verify that tid and freelist have not been changed
2293 * 3. If they were not changed replace tid and freelist
2294 *
2295 * Since this is without lock semantics the protection is only against
2296 * code executing on this cpu *not* from access by other cpus.
2297 */
Thomas Gleixner30106b82011-05-04 15:38:19 +02002298 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002299 s->cpu_slab->freelist, s->cpu_slab->tid,
2300 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05002301 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002302
2303 note_cmpxchg_failure("slab_alloc", s, tid);
2304 goto redo;
2305 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06002306 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002307 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002308
Pekka Enberg74e21342009-11-25 20:14:48 +02002309 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06002310 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002311
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002312 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002313
Christoph Lameter894b8782007-05-10 03:15:16 -07002314 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002315}
2316
2317void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2318{
Christoph Lameter2154a332010-07-09 14:07:10 -05002319 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002320
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002321 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002322
2323 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002324}
2325EXPORT_SYMBOL(kmem_cache_alloc);
2326
Li Zefan0f24f122009-12-11 15:45:30 +08002327#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002328void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002329{
Richard Kennedy4a923792010-10-21 10:29:19 +01002330 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
2331 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2332 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002333}
Richard Kennedy4a923792010-10-21 10:29:19 +01002334EXPORT_SYMBOL(kmem_cache_alloc_trace);
2335
2336void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2337{
2338 void *ret = kmalloc_order(size, flags, order);
2339 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2340 return ret;
2341}
2342EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002343#endif
2344
Christoph Lameter81819f02007-05-06 14:49:36 -07002345#ifdef CONFIG_NUMA
2346void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2347{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002348 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2349
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002350 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2351 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002352
2353 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002354}
2355EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002356
Li Zefan0f24f122009-12-11 15:45:30 +08002357#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002358void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002359 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002360 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002361{
Richard Kennedy4a923792010-10-21 10:29:19 +01002362 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2363
2364 trace_kmalloc_node(_RET_IP_, ret,
2365 size, s->size, gfpflags, node);
2366 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002367}
Richard Kennedy4a923792010-10-21 10:29:19 +01002368EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002369#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002370#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002371
Christoph Lameter81819f02007-05-06 14:49:36 -07002372/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002373 * Slow patch handling. This may still be called frequently since objects
2374 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002375 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002376 * So we still attempt to reduce cache line usage. Just take the slab
2377 * lock and free the item. If there is no additional partial page
2378 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002379 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002380static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002381 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002382{
2383 void *prior;
2384 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002385 int was_frozen;
2386 int inuse;
2387 struct page new;
2388 unsigned long counters;
2389 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002390 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002391
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002392 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002393
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002394 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002395 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002396
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002397 do {
2398 prior = page->freelist;
2399 counters = page->counters;
2400 set_freepointer(s, object, prior);
2401 new.counters = counters;
2402 was_frozen = new.frozen;
2403 new.inuse--;
2404 if ((!new.inuse || !prior) && !was_frozen && !n) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002405
2406 if (!kmem_cache_debug(s) && !prior)
2407
2408 /*
2409 * Slab was on no list before and will be partially empty
2410 * We can defer the list move and instead freeze it.
2411 */
2412 new.frozen = 1;
2413
2414 else { /* Needs to be taken off a list */
2415
2416 n = get_node(s, page_to_nid(page));
2417 /*
2418 * Speculatively acquire the list_lock.
2419 * If the cmpxchg does not succeed then we may
2420 * drop the list_lock without any processing.
2421 *
2422 * Otherwise the list_lock will synchronize with
2423 * other processors updating the list of slabs.
2424 */
2425 spin_lock_irqsave(&n->list_lock, flags);
2426
2427 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002428 }
2429 inuse = new.inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002430
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002431 } while (!cmpxchg_double_slab(s, page,
2432 prior, counters,
2433 object, new.counters,
2434 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002435
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002436 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002437
2438 /*
2439 * If we just froze the page then put it onto the
2440 * per cpu partial list.
2441 */
2442 if (new.frozen && !was_frozen)
2443 put_cpu_partial(s, page, 1);
2444
2445 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002446 * The list lock was not taken therefore no list
2447 * activity can be necessary.
2448 */
2449 if (was_frozen)
2450 stat(s, FREE_FROZEN);
Christoph Lameter80f08c12011-06-01 12:25:55 -05002451 return;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002452 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002453
2454 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002455 * was_frozen may have been set after we acquired the list_lock in
2456 * an earlier loop. So we need to check it here again.
Christoph Lameter81819f02007-05-06 14:49:36 -07002457 */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002458 if (was_frozen)
2459 stat(s, FREE_FROZEN);
2460 else {
2461 if (unlikely(!inuse && n->nr_partial > s->min_partial))
2462 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002463
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002464 /*
2465 * Objects left in the slab. If it was not on the partial list before
2466 * then add it.
2467 */
2468 if (unlikely(!prior)) {
2469 remove_full(s, page);
Shaohua Li136333d2011-08-24 08:57:52 +08002470 add_partial(n, page, DEACTIVATE_TO_TAIL);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002471 stat(s, FREE_ADD_PARTIAL);
2472 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002473 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002474 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002475 return;
2476
2477slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002478 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002479 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002480 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002481 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002482 remove_partial(n, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002483 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002484 } else
2485 /* Slab must be on the full list */
2486 remove_full(s, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002487
Christoph Lameter80f08c12011-06-01 12:25:55 -05002488 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002489 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002490 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002491}
2492
Christoph Lameter894b8782007-05-10 03:15:16 -07002493/*
2494 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2495 * can perform fastpath freeing without additional function calls.
2496 *
2497 * The fastpath is only possible if we are freeing to the current cpu slab
2498 * of this processor. This typically the case if we have just allocated
2499 * the item before.
2500 *
2501 * If fastpath is not possible then fall back to __slab_free where we deal
2502 * with all sorts of special processing.
2503 */
Pekka Enberg06428782008-01-07 23:20:27 -08002504static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002505 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002506{
2507 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002508 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002509 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002510
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002511 slab_free_hook(s, x);
2512
Christoph Lametera24c5a02011-03-15 12:45:21 -05002513redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002514 /*
2515 * Determine the currently cpus per cpu slab.
2516 * The cpu may change afterward. However that does not matter since
2517 * data is retrieved via this pointer. If we are on the same cpu
2518 * during the cmpxchg then the free will succedd.
2519 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002520 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002521
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002522 tid = c->tid;
2523 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002524
Christoph Lameter442b06b2011-05-17 16:29:31 -05002525 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002526 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002527
Thomas Gleixner30106b82011-05-04 15:38:19 +02002528 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002529 s->cpu_slab->freelist, s->cpu_slab->tid,
2530 c->freelist, tid,
2531 object, next_tid(tid)))) {
2532
2533 note_cmpxchg_failure("slab_free", s, tid);
2534 goto redo;
2535 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06002536 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002537 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002538 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002539
Christoph Lameter894b8782007-05-10 03:15:16 -07002540}
2541
Christoph Lameter81819f02007-05-06 14:49:36 -07002542void kmem_cache_free(struct kmem_cache *s, void *x)
2543{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002544 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002545
Christoph Lameterb49af682007-05-06 14:49:41 -07002546 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002547
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002548 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002549
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002550 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002551}
2552EXPORT_SYMBOL(kmem_cache_free);
2553
Christoph Lameter81819f02007-05-06 14:49:36 -07002554/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002555 * Object placement in a slab is made very easy because we always start at
2556 * offset 0. If we tune the size of the object to the alignment then we can
2557 * get the required alignment by putting one properly sized object after
2558 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002559 *
2560 * Notice that the allocation order determines the sizes of the per cpu
2561 * caches. Each processor has always one slab available for allocations.
2562 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002563 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002564 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002565 */
2566
2567/*
2568 * Mininum / Maximum order of slab pages. This influences locking overhead
2569 * and slab fragmentation. A higher order reduces the number of partial slabs
2570 * and increases the number of allocations possible without having to
2571 * take the list_lock.
2572 */
2573static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002574static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002575static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002576
2577/*
2578 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002579 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002580 */
2581static int slub_nomerge;
2582
2583/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002584 * Calculate the order of allocation given an slab object size.
2585 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002586 * The order of allocation has significant impact on performance and other
2587 * system components. Generally order 0 allocations should be preferred since
2588 * order 0 does not cause fragmentation in the page allocator. Larger objects
2589 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002590 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002591 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002592 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002593 * In order to reach satisfactory performance we must ensure that a minimum
2594 * number of objects is in one slab. Otherwise we may generate too much
2595 * activity on the partial lists which requires taking the list_lock. This is
2596 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002597 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002598 * slub_max_order specifies the order where we begin to stop considering the
2599 * number of objects in a slab as critical. If we reach slub_max_order then
2600 * we try to keep the page order as low as possible. So we accept more waste
2601 * of space in favor of a small page order.
2602 *
2603 * Higher order allocations also allow the placement of more objects in a
2604 * slab and thereby reduce object handling overhead. If the user has
2605 * requested a higher mininum order then we start with that one instead of
2606 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002607 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002608static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002609 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002610{
2611 int order;
2612 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002613 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002614
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002615 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002616 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002617
Christoph Lameter6300ea72007-07-17 04:03:20 -07002618 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002619 fls(min_objects * size - 1) - PAGE_SHIFT);
2620 order <= max_order; order++) {
2621
Christoph Lameter81819f02007-05-06 14:49:36 -07002622 unsigned long slab_size = PAGE_SIZE << order;
2623
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002624 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002625 continue;
2626
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002627 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002628
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002629 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002630 break;
2631
2632 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002633
Christoph Lameter81819f02007-05-06 14:49:36 -07002634 return order;
2635}
2636
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002637static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002638{
2639 int order;
2640 int min_objects;
2641 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002642 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002643
2644 /*
2645 * Attempt to find best configuration for a slab. This
2646 * works by first attempting to generate a layout with
2647 * the best configuration and backing off gradually.
2648 *
2649 * First we reduce the acceptable waste in a slab. Then
2650 * we reduce the minimum objects required in a slab.
2651 */
2652 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002653 if (!min_objects)
2654 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002655 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002656 min_objects = min(min_objects, max_objects);
2657
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002658 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002659 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002660 while (fraction >= 4) {
2661 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002662 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002663 if (order <= slub_max_order)
2664 return order;
2665 fraction /= 2;
2666 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002667 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002668 }
2669
2670 /*
2671 * We were unable to place multiple objects in a slab. Now
2672 * lets see if we can place a single object there.
2673 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002674 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002675 if (order <= slub_max_order)
2676 return order;
2677
2678 /*
2679 * Doh this slab cannot be placed using slub_max_order.
2680 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002681 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002682 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002683 return order;
2684 return -ENOSYS;
2685}
2686
Christoph Lameter81819f02007-05-06 14:49:36 -07002687/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002688 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002689 */
2690static unsigned long calculate_alignment(unsigned long flags,
2691 unsigned long align, unsigned long size)
2692{
2693 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002694 * If the user wants hardware cache aligned objects then follow that
2695 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002696 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002697 * The hardware cache alignment cannot override the specified
2698 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002699 */
Nick Pigginb6210382008-03-05 14:05:56 -08002700 if (flags & SLAB_HWCACHE_ALIGN) {
2701 unsigned long ralign = cache_line_size();
2702 while (size <= ralign / 2)
2703 ralign /= 2;
2704 align = max(align, ralign);
2705 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002706
2707 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002708 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002709
2710 return ALIGN(align, sizeof(void *));
2711}
2712
Pekka Enberg5595cff2008-08-05 09:28:47 +03002713static void
2714init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002715{
2716 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002717 spin_lock_init(&n->list_lock);
2718 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002719#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002720 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002721 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002722 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002723#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002724}
2725
Christoph Lameter55136592010-08-20 12:37:13 -05002726static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002727{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002728 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2729 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002730
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002731 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002732 * Must align to double word boundary for the double cmpxchg
2733 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002734 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002735 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2736 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002737
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002738 if (!s->cpu_slab)
2739 return 0;
2740
2741 init_kmem_cache_cpus(s);
2742
2743 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002744}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002745
Christoph Lameter51df1142010-08-20 12:37:15 -05002746static struct kmem_cache *kmem_cache_node;
2747
Christoph Lameter81819f02007-05-06 14:49:36 -07002748/*
2749 * No kmalloc_node yet so do it by hand. We know that this is the first
2750 * slab on the node for this slabcache. There are no concurrent accesses
2751 * possible.
2752 *
2753 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002754 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2755 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002756 */
Christoph Lameter55136592010-08-20 12:37:13 -05002757static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002758{
2759 struct page *page;
2760 struct kmem_cache_node *n;
2761
Christoph Lameter51df1142010-08-20 12:37:15 -05002762 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002763
Christoph Lameter51df1142010-08-20 12:37:15 -05002764 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002765
2766 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002767 if (page_to_nid(page) != node) {
2768 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2769 "node %d\n", node);
2770 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2771 "in order to be able to continue\n");
2772 }
2773
Christoph Lameter81819f02007-05-06 14:49:36 -07002774 n = page->freelist;
2775 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002776 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05002777 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002778 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002779 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002780#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002781 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002782 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002783#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002784 init_kmem_cache_node(n, kmem_cache_node);
2785 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002786
Shaohua Li136333d2011-08-24 08:57:52 +08002787 add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002788}
2789
2790static void free_kmem_cache_nodes(struct kmem_cache *s)
2791{
2792 int node;
2793
Christoph Lameterf64dc582007-10-16 01:25:33 -07002794 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002795 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002796
Alexander Duyck73367bd2010-05-21 14:41:35 -07002797 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002798 kmem_cache_free(kmem_cache_node, n);
2799
Christoph Lameter81819f02007-05-06 14:49:36 -07002800 s->node[node] = NULL;
2801 }
2802}
2803
Christoph Lameter55136592010-08-20 12:37:13 -05002804static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002805{
2806 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002807
Christoph Lameterf64dc582007-10-16 01:25:33 -07002808 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002809 struct kmem_cache_node *n;
2810
Alexander Duyck73367bd2010-05-21 14:41:35 -07002811 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002812 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002813 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002814 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002815 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002816 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002817
2818 if (!n) {
2819 free_kmem_cache_nodes(s);
2820 return 0;
2821 }
2822
Christoph Lameter81819f02007-05-06 14:49:36 -07002823 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002824 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002825 }
2826 return 1;
2827}
Christoph Lameter81819f02007-05-06 14:49:36 -07002828
David Rientjesc0bdb232009-02-25 09:16:35 +02002829static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002830{
2831 if (min < MIN_PARTIAL)
2832 min = MIN_PARTIAL;
2833 else if (min > MAX_PARTIAL)
2834 min = MAX_PARTIAL;
2835 s->min_partial = min;
2836}
2837
Christoph Lameter81819f02007-05-06 14:49:36 -07002838/*
2839 * calculate_sizes() determines the order and the distribution of data within
2840 * a slab object.
2841 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002842static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002843{
2844 unsigned long flags = s->flags;
2845 unsigned long size = s->objsize;
2846 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002847 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002848
2849 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002850 * Round up object size to the next word boundary. We can only
2851 * place the free pointer at word boundaries and this determines
2852 * the possible location of the free pointer.
2853 */
2854 size = ALIGN(size, sizeof(void *));
2855
2856#ifdef CONFIG_SLUB_DEBUG
2857 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002858 * Determine if we can poison the object itself. If the user of
2859 * the slab may touch the object after free or before allocation
2860 * then we should never poison the object itself.
2861 */
2862 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002863 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002864 s->flags |= __OBJECT_POISON;
2865 else
2866 s->flags &= ~__OBJECT_POISON;
2867
Christoph Lameter81819f02007-05-06 14:49:36 -07002868
2869 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002870 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002871 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002872 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002873 */
2874 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2875 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002876#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002877
2878 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002879 * With that we have determined the number of bytes in actual use
2880 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002881 */
2882 s->inuse = size;
2883
2884 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002885 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002886 /*
2887 * Relocate free pointer after the object if it is not
2888 * permitted to overwrite the first word of the object on
2889 * kmem_cache_free.
2890 *
2891 * This is the case if we do RCU, have a constructor or
2892 * destructor or are poisoning the objects.
2893 */
2894 s->offset = size;
2895 size += sizeof(void *);
2896 }
2897
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002898#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002899 if (flags & SLAB_STORE_USER)
2900 /*
2901 * Need to store information about allocs and frees after
2902 * the object.
2903 */
2904 size += 2 * sizeof(struct track);
2905
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002906 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002907 /*
2908 * Add some empty padding so that we can catch
2909 * overwrites from earlier objects rather than let
2910 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002911 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002912 * of the object.
2913 */
2914 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002915#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002916
Christoph Lameter81819f02007-05-06 14:49:36 -07002917 /*
2918 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002919 * user specified and the dynamic determination of cache line size
2920 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002921 */
2922 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002923 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002924
2925 /*
2926 * SLUB stores one object immediately after another beginning from
2927 * offset 0. In order to align the objects we have to simply size
2928 * each object to conform to the alignment.
2929 */
2930 size = ALIGN(size, align);
2931 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002932 if (forced_order >= 0)
2933 order = forced_order;
2934 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002935 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002936
Christoph Lameter834f3d12008-04-14 19:11:31 +03002937 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002938 return 0;
2939
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002940 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002941 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002942 s->allocflags |= __GFP_COMP;
2943
2944 if (s->flags & SLAB_CACHE_DMA)
2945 s->allocflags |= SLUB_DMA;
2946
2947 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2948 s->allocflags |= __GFP_RECLAIMABLE;
2949
Christoph Lameter81819f02007-05-06 14:49:36 -07002950 /*
2951 * Determine the number of objects per slab
2952 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002953 s->oo = oo_make(order, size, s->reserved);
2954 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002955 if (oo_objects(s->oo) > oo_objects(s->max))
2956 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002957
Christoph Lameter834f3d12008-04-14 19:11:31 +03002958 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002959
2960}
2961
Christoph Lameter55136592010-08-20 12:37:13 -05002962static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002963 const char *name, size_t size,
2964 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002965 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002966{
2967 memset(s, 0, kmem_size);
2968 s->name = name;
2969 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002970 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002971 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002972 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002973 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002974
Lai Jiangshanda9a6382011-03-10 15:22:00 +08002975 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2976 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07002977
Christoph Lameter06b285d2008-04-14 19:11:41 +03002978 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002979 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002980 if (disable_higher_order_debug) {
2981 /*
2982 * Disable debugging flags that store metadata if the min slab
2983 * order increased.
2984 */
2985 if (get_order(s->size) > get_order(s->objsize)) {
2986 s->flags &= ~DEBUG_METADATA_FLAGS;
2987 s->offset = 0;
2988 if (!calculate_sizes(s, -1))
2989 goto error;
2990 }
2991 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002992
Christoph Lameterb789ef52011-06-01 12:25:49 -05002993#ifdef CONFIG_CMPXCHG_DOUBLE
2994 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
2995 /* Enable fast mode */
2996 s->flags |= __CMPXCHG_DOUBLE;
2997#endif
2998
David Rientjes3b89d7d2009-02-22 17:40:07 -08002999 /*
3000 * The larger the object size is, the more pages we want on the partial
3001 * list to avoid pounding the page allocator excessively.
3002 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003003 set_min_partial(s, ilog2(s->size) / 2);
3004
3005 /*
3006 * cpu_partial determined the maximum number of objects kept in the
3007 * per cpu partial lists of a processor.
3008 *
3009 * Per cpu partial lists mainly contain slabs that just have one
3010 * object freed. If they are used for allocation then they can be
3011 * filled up again with minimal effort. The slab will never hit the
3012 * per node partial lists and therefore no locking will be required.
3013 *
3014 * This setting also determines
3015 *
3016 * A) The number of objects from per cpu partial slabs dumped to the
3017 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003018 * B) The number of objects in cpu partial slabs to extract from the
Christoph Lameter49e22582011-08-09 16:12:27 -05003019 * per node list when we run out of per cpu objects. We only fetch 50%
3020 * to keep some capacity around for frees.
3021 */
3022 if (s->size >= PAGE_SIZE)
3023 s->cpu_partial = 2;
3024 else if (s->size >= 1024)
3025 s->cpu_partial = 6;
3026 else if (s->size >= 256)
3027 s->cpu_partial = 13;
3028 else
3029 s->cpu_partial = 30;
3030
Christoph Lameter81819f02007-05-06 14:49:36 -07003031 s->refcount = 1;
3032#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003033 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003034#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003035 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003036 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003037
Christoph Lameter55136592010-08-20 12:37:13 -05003038 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07003039 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06003040
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003041 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003042error:
3043 if (flags & SLAB_PANIC)
3044 panic("Cannot create slab %s size=%lu realsize=%u "
3045 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03003046 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07003047 s->offset, flags);
3048 return 0;
3049}
Christoph Lameter81819f02007-05-06 14:49:36 -07003050
3051/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003052 * Determine the size of a slab object
3053 */
3054unsigned int kmem_cache_size(struct kmem_cache *s)
3055{
3056 return s->objsize;
3057}
3058EXPORT_SYMBOL(kmem_cache_size);
3059
Christoph Lameter33b12c32008-04-25 12:22:43 -07003060static void list_slab_objects(struct kmem_cache *s, struct page *page,
3061 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003062{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003063#ifdef CONFIG_SLUB_DEBUG
3064 void *addr = page_address(page);
3065 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003066 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3067 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003068 if (!map)
3069 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07003070 slab_err(s, page, "%s", text);
3071 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003072
Christoph Lameter5f80b132011-04-15 14:48:13 -05003073 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003074 for_each_object(p, s, addr, page->objects) {
3075
3076 if (!test_bit(slab_index(p, s, addr), map)) {
3077 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
3078 p, p - addr);
3079 print_tracking(s, p);
3080 }
3081 }
3082 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003083 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003084#endif
3085}
3086
Christoph Lameter81819f02007-05-06 14:49:36 -07003087/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003088 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003089 * This is called from kmem_cache_close(). We must be the last thread
3090 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003091 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003092static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003093{
Christoph Lameter81819f02007-05-06 14:49:36 -07003094 struct page *page, *h;
3095
Christoph Lameter33b12c32008-04-25 12:22:43 -07003096 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003097 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003098 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003099 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003100 } else {
3101 list_slab_objects(s, page,
3102 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003103 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003104 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003105}
3106
3107/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003108 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003109 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003110static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003111{
3112 int node;
3113
3114 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003115 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003116 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07003117 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003118 struct kmem_cache_node *n = get_node(s, node);
3119
Christoph Lameter599870b2008-04-23 12:36:52 -07003120 free_partial(s, n);
3121 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003122 return 1;
3123 }
3124 free_kmem_cache_nodes(s);
3125 return 0;
3126}
3127
3128/*
3129 * Close a cache and release the kmem_cache structure
3130 * (must be used for caches created using kmem_cache_create)
3131 */
3132void kmem_cache_destroy(struct kmem_cache *s)
3133{
3134 down_write(&slub_lock);
3135 s->refcount--;
3136 if (!s->refcount) {
3137 list_del(&s->list);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003138 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03003139 if (kmem_cache_close(s)) {
3140 printk(KERN_ERR "SLUB %s: %s called for cache that "
3141 "still has objects.\n", s->name, __func__);
3142 dump_stack();
3143 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03003144 if (s->flags & SLAB_DESTROY_BY_RCU)
3145 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07003146 sysfs_slab_remove(s);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003147 } else
3148 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003149}
3150EXPORT_SYMBOL(kmem_cache_destroy);
3151
3152/********************************************************************
3153 * Kmalloc subsystem
3154 *******************************************************************/
3155
Christoph Lameter51df1142010-08-20 12:37:15 -05003156struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07003157EXPORT_SYMBOL(kmalloc_caches);
3158
Christoph Lameter51df1142010-08-20 12:37:15 -05003159static struct kmem_cache *kmem_cache;
3160
Christoph Lameter55136592010-08-20 12:37:13 -05003161#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003162static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05003163#endif
3164
Christoph Lameter81819f02007-05-06 14:49:36 -07003165static int __init setup_slub_min_order(char *str)
3166{
Pekka Enberg06428782008-01-07 23:20:27 -08003167 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003168
3169 return 1;
3170}
3171
3172__setup("slub_min_order=", setup_slub_min_order);
3173
3174static int __init setup_slub_max_order(char *str)
3175{
Pekka Enberg06428782008-01-07 23:20:27 -08003176 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003177 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003178
3179 return 1;
3180}
3181
3182__setup("slub_max_order=", setup_slub_max_order);
3183
3184static int __init setup_slub_min_objects(char *str)
3185{
Pekka Enberg06428782008-01-07 23:20:27 -08003186 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003187
3188 return 1;
3189}
3190
3191__setup("slub_min_objects=", setup_slub_min_objects);
3192
3193static int __init setup_slub_nomerge(char *str)
3194{
3195 slub_nomerge = 1;
3196 return 1;
3197}
3198
3199__setup("slub_nomerge", setup_slub_nomerge);
3200
Christoph Lameter51df1142010-08-20 12:37:15 -05003201static struct kmem_cache *__init create_kmalloc_cache(const char *name,
3202 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003203{
Christoph Lameter51df1142010-08-20 12:37:15 -05003204 struct kmem_cache *s;
3205
3206 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3207
Pekka Enberg83b519e2009-06-10 19:40:04 +03003208 /*
3209 * This function is called with IRQs disabled during early-boot on
3210 * single CPU so there's no need to take slub_lock here.
3211 */
Christoph Lameter55136592010-08-20 12:37:13 -05003212 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03003213 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07003214 goto panic;
3215
3216 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05003217 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003218
3219panic:
3220 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003221 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003222}
3223
Christoph Lameterf1b26332007-07-17 04:03:26 -07003224/*
3225 * Conversion table for small slabs sizes / 8 to the index in the
3226 * kmalloc array. This is necessary for slabs < 192 since we have non power
3227 * of two cache sizes there. The size of larger slabs can be determined using
3228 * fls.
3229 */
3230static s8 size_index[24] = {
3231 3, /* 8 */
3232 4, /* 16 */
3233 5, /* 24 */
3234 5, /* 32 */
3235 6, /* 40 */
3236 6, /* 48 */
3237 6, /* 56 */
3238 6, /* 64 */
3239 1, /* 72 */
3240 1, /* 80 */
3241 1, /* 88 */
3242 1, /* 96 */
3243 7, /* 104 */
3244 7, /* 112 */
3245 7, /* 120 */
3246 7, /* 128 */
3247 2, /* 136 */
3248 2, /* 144 */
3249 2, /* 152 */
3250 2, /* 160 */
3251 2, /* 168 */
3252 2, /* 176 */
3253 2, /* 184 */
3254 2 /* 192 */
3255};
3256
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003257static inline int size_index_elem(size_t bytes)
3258{
3259 return (bytes - 1) / 8;
3260}
3261
Christoph Lameter81819f02007-05-06 14:49:36 -07003262static struct kmem_cache *get_slab(size_t size, gfp_t flags)
3263{
Christoph Lameterf1b26332007-07-17 04:03:26 -07003264 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07003265
Christoph Lameterf1b26332007-07-17 04:03:26 -07003266 if (size <= 192) {
3267 if (!size)
3268 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07003269
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003270 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003271 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07003272 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003273
3274#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07003275 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05003276 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07003277
Christoph Lameter81819f02007-05-06 14:49:36 -07003278#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003279 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07003280}
3281
3282void *__kmalloc(size_t size, gfp_t flags)
3283{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003284 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003285 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003286
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003287 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003288 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003289
3290 s = get_slab(size, flags);
3291
3292 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003293 return s;
3294
Christoph Lameter2154a332010-07-09 14:07:10 -05003295 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003296
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003297 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003298
3299 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003300}
3301EXPORT_SYMBOL(__kmalloc);
3302
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003303#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003304static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3305{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003306 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003307 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003308
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003309 flags |= __GFP_COMP | __GFP_NOTRACK;
3310 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003311 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003312 ptr = page_address(page);
3313
3314 kmemleak_alloc(ptr, size, 1, flags);
3315 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003316}
3317
Christoph Lameter81819f02007-05-06 14:49:36 -07003318void *__kmalloc_node(size_t size, gfp_t flags, int node)
3319{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003320 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003321 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003322
Ingo Molnar057685c2009-02-20 12:15:30 +01003323 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003324 ret = kmalloc_large_node(size, flags, node);
3325
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003326 trace_kmalloc_node(_RET_IP_, ret,
3327 size, PAGE_SIZE << get_order(size),
3328 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003329
3330 return ret;
3331 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003332
3333 s = get_slab(size, flags);
3334
3335 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003336 return s;
3337
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003338 ret = slab_alloc(s, flags, node, _RET_IP_);
3339
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003340 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003341
3342 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003343}
3344EXPORT_SYMBOL(__kmalloc_node);
3345#endif
3346
3347size_t ksize(const void *object)
3348{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003349 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003350
Christoph Lameteref8b4522007-10-16 01:24:46 -07003351 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003352 return 0;
3353
Vegard Nossum294a80a2007-12-04 23:45:30 -08003354 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003355
Pekka Enberg76994412008-05-22 19:22:25 +03003356 if (unlikely(!PageSlab(page))) {
3357 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003358 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003359 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003360
Eric Dumazetb3d41882011-02-14 18:35:22 +01003361 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003362}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003363EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003364
Ben Greeard18a90d2011-07-07 11:36:37 -07003365#ifdef CONFIG_SLUB_DEBUG
3366bool verify_mem_not_deleted(const void *x)
3367{
3368 struct page *page;
3369 void *object = (void *)x;
3370 unsigned long flags;
3371 bool rv;
3372
3373 if (unlikely(ZERO_OR_NULL_PTR(x)))
3374 return false;
3375
3376 local_irq_save(flags);
3377
3378 page = virt_to_head_page(x);
3379 if (unlikely(!PageSlab(page))) {
3380 /* maybe it was from stack? */
3381 rv = true;
3382 goto out_unlock;
3383 }
3384
3385 slab_lock(page);
3386 if (on_freelist(page->slab, page, object)) {
3387 object_err(page->slab, page, object, "Object is on free-list");
3388 rv = false;
3389 } else {
3390 rv = true;
3391 }
3392 slab_unlock(page);
3393
3394out_unlock:
3395 local_irq_restore(flags);
3396 return rv;
3397}
3398EXPORT_SYMBOL(verify_mem_not_deleted);
3399#endif
3400
Christoph Lameter81819f02007-05-06 14:49:36 -07003401void kfree(const void *x)
3402{
Christoph Lameter81819f02007-05-06 14:49:36 -07003403 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003404 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003405
Pekka Enberg2121db72009-03-25 11:05:57 +02003406 trace_kfree(_RET_IP_, x);
3407
Satyam Sharma2408c552007-10-16 01:24:44 -07003408 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003409 return;
3410
Christoph Lameterb49af682007-05-06 14:49:41 -07003411 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003412 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003413 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003414 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003415 put_page(page);
3416 return;
3417 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003418 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003419}
3420EXPORT_SYMBOL(kfree);
3421
Christoph Lameter2086d262007-05-06 14:49:46 -07003422/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003423 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3424 * the remaining slabs by the number of items in use. The slabs with the
3425 * most items in use come first. New allocations will then fill those up
3426 * and thus they can be removed from the partial lists.
3427 *
3428 * The slabs with the least items are placed last. This results in them
3429 * being allocated from last increasing the chance that the last objects
3430 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003431 */
3432int kmem_cache_shrink(struct kmem_cache *s)
3433{
3434 int node;
3435 int i;
3436 struct kmem_cache_node *n;
3437 struct page *page;
3438 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003439 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003440 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003441 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003442 unsigned long flags;
3443
3444 if (!slabs_by_inuse)
3445 return -ENOMEM;
3446
3447 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003448 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003449 n = get_node(s, node);
3450
3451 if (!n->nr_partial)
3452 continue;
3453
Christoph Lameter834f3d12008-04-14 19:11:31 +03003454 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003455 INIT_LIST_HEAD(slabs_by_inuse + i);
3456
3457 spin_lock_irqsave(&n->list_lock, flags);
3458
3459 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003460 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003461 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003462 * Note that concurrent frees may occur while we hold the
3463 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003464 */
3465 list_for_each_entry_safe(page, t, &n->partial, lru) {
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003466 list_move(&page->lru, slabs_by_inuse + page->inuse);
3467 if (!page->inuse)
3468 n->nr_partial--;
Christoph Lameter2086d262007-05-06 14:49:46 -07003469 }
3470
Christoph Lameter2086d262007-05-06 14:49:46 -07003471 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003472 * Rebuild the partial list with the slabs filled up most
3473 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003474 */
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003475 for (i = objects - 1; i > 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003476 list_splice(slabs_by_inuse + i, n->partial.prev);
3477
Christoph Lameter2086d262007-05-06 14:49:46 -07003478 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003479
3480 /* Release empty slabs */
3481 list_for_each_entry_safe(page, t, slabs_by_inuse, lru)
3482 discard_slab(s, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003483 }
3484
3485 kfree(slabs_by_inuse);
3486 return 0;
3487}
3488EXPORT_SYMBOL(kmem_cache_shrink);
3489
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003490#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003491static int slab_mem_going_offline_callback(void *arg)
3492{
3493 struct kmem_cache *s;
3494
3495 down_read(&slub_lock);
3496 list_for_each_entry(s, &slab_caches, list)
3497 kmem_cache_shrink(s);
3498 up_read(&slub_lock);
3499
3500 return 0;
3501}
3502
3503static void slab_mem_offline_callback(void *arg)
3504{
3505 struct kmem_cache_node *n;
3506 struct kmem_cache *s;
3507 struct memory_notify *marg = arg;
3508 int offline_node;
3509
3510 offline_node = marg->status_change_nid;
3511
3512 /*
3513 * If the node still has available memory. we need kmem_cache_node
3514 * for it yet.
3515 */
3516 if (offline_node < 0)
3517 return;
3518
3519 down_read(&slub_lock);
3520 list_for_each_entry(s, &slab_caches, list) {
3521 n = get_node(s, offline_node);
3522 if (n) {
3523 /*
3524 * if n->nr_slabs > 0, slabs still exist on the node
3525 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003526 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003527 * callback. So, we must fail.
3528 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003529 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003530
3531 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003532 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003533 }
3534 }
3535 up_read(&slub_lock);
3536}
3537
3538static int slab_mem_going_online_callback(void *arg)
3539{
3540 struct kmem_cache_node *n;
3541 struct kmem_cache *s;
3542 struct memory_notify *marg = arg;
3543 int nid = marg->status_change_nid;
3544 int ret = 0;
3545
3546 /*
3547 * If the node's memory is already available, then kmem_cache_node is
3548 * already created. Nothing to do.
3549 */
3550 if (nid < 0)
3551 return 0;
3552
3553 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003554 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003555 * allocate a kmem_cache_node structure in order to bring the node
3556 * online.
3557 */
3558 down_read(&slub_lock);
3559 list_for_each_entry(s, &slab_caches, list) {
3560 /*
3561 * XXX: kmem_cache_alloc_node will fallback to other nodes
3562 * since memory is not yet available from the node that
3563 * is brought up.
3564 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003565 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003566 if (!n) {
3567 ret = -ENOMEM;
3568 goto out;
3569 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003570 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003571 s->node[nid] = n;
3572 }
3573out:
3574 up_read(&slub_lock);
3575 return ret;
3576}
3577
3578static int slab_memory_callback(struct notifier_block *self,
3579 unsigned long action, void *arg)
3580{
3581 int ret = 0;
3582
3583 switch (action) {
3584 case MEM_GOING_ONLINE:
3585 ret = slab_mem_going_online_callback(arg);
3586 break;
3587 case MEM_GOING_OFFLINE:
3588 ret = slab_mem_going_offline_callback(arg);
3589 break;
3590 case MEM_OFFLINE:
3591 case MEM_CANCEL_ONLINE:
3592 slab_mem_offline_callback(arg);
3593 break;
3594 case MEM_ONLINE:
3595 case MEM_CANCEL_OFFLINE:
3596 break;
3597 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003598 if (ret)
3599 ret = notifier_from_errno(ret);
3600 else
3601 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003602 return ret;
3603}
3604
3605#endif /* CONFIG_MEMORY_HOTPLUG */
3606
Christoph Lameter81819f02007-05-06 14:49:36 -07003607/********************************************************************
3608 * Basic setup of slabs
3609 *******************************************************************/
3610
Christoph Lameter51df1142010-08-20 12:37:15 -05003611/*
3612 * Used for early kmem_cache structures that were allocated using
3613 * the page allocator
3614 */
3615
3616static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3617{
3618 int node;
3619
3620 list_add(&s->list, &slab_caches);
3621 s->refcount = -1;
3622
3623 for_each_node_state(node, N_NORMAL_MEMORY) {
3624 struct kmem_cache_node *n = get_node(s, node);
3625 struct page *p;
3626
3627 if (n) {
3628 list_for_each_entry(p, &n->partial, lru)
3629 p->slab = s;
3630
Li Zefan607bf322011-04-12 15:22:26 +08003631#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003632 list_for_each_entry(p, &n->full, lru)
3633 p->slab = s;
3634#endif
3635 }
3636 }
3637}
3638
Christoph Lameter81819f02007-05-06 14:49:36 -07003639void __init kmem_cache_init(void)
3640{
3641 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003642 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003643 struct kmem_cache *temp_kmem_cache;
3644 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003645 struct kmem_cache *temp_kmem_cache_node;
3646 unsigned long kmalloc_size;
3647
3648 kmem_size = offsetof(struct kmem_cache, node) +
3649 nr_node_ids * sizeof(struct kmem_cache_node *);
3650
3651 /* Allocate two kmem_caches from the page allocator */
3652 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3653 order = get_order(2 * kmalloc_size);
3654 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3655
Christoph Lameter81819f02007-05-06 14:49:36 -07003656 /*
3657 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003658 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003659 * kmem_cache_open for slab_state == DOWN.
3660 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003661 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3662
3663 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3664 sizeof(struct kmem_cache_node),
3665 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003666
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003667 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003668
3669 /* Able to allocate the per node structures */
3670 slab_state = PARTIAL;
3671
Christoph Lameter51df1142010-08-20 12:37:15 -05003672 temp_kmem_cache = kmem_cache;
3673 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3674 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3675 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3676 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003677
Christoph Lameter51df1142010-08-20 12:37:15 -05003678 /*
3679 * Allocate kmem_cache_node properly from the kmem_cache slab.
3680 * kmem_cache_node is separately allocated so no need to
3681 * update any list pointers.
3682 */
3683 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003684
Christoph Lameter51df1142010-08-20 12:37:15 -05003685 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3686 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3687
3688 kmem_cache_bootstrap_fixup(kmem_cache_node);
3689
3690 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003691 kmem_cache_bootstrap_fixup(kmem_cache);
3692 caches++;
3693 /* Free temporary boot structure */
3694 free_pages((unsigned long)temp_kmem_cache, order);
3695
3696 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003697
3698 /*
3699 * Patch up the size_index table if we have strange large alignment
3700 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003701 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003702 *
3703 * Largest permitted alignment is 256 bytes due to the way we
3704 * handle the index determination for the smaller caches.
3705 *
3706 * Make sure that nothing crazy happens if someone starts tinkering
3707 * around with ARCH_KMALLOC_MINALIGN
3708 */
3709 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3710 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3711
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003712 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3713 int elem = size_index_elem(i);
3714 if (elem >= ARRAY_SIZE(size_index))
3715 break;
3716 size_index[elem] = KMALLOC_SHIFT_LOW;
3717 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003718
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003719 if (KMALLOC_MIN_SIZE == 64) {
3720 /*
3721 * The 96 byte size cache is not used if the alignment
3722 * is 64 byte.
3723 */
3724 for (i = 64 + 8; i <= 96; i += 8)
3725 size_index[size_index_elem(i)] = 7;
3726 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003727 /*
3728 * The 192 byte sized cache is not used if the alignment
3729 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3730 * instead.
3731 */
3732 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003733 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003734 }
3735
Christoph Lameter51df1142010-08-20 12:37:15 -05003736 /* Caches that are not of the two-to-the-power-of size */
3737 if (KMALLOC_MIN_SIZE <= 32) {
3738 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3739 caches++;
3740 }
3741
3742 if (KMALLOC_MIN_SIZE <= 64) {
3743 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3744 caches++;
3745 }
3746
3747 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3748 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3749 caches++;
3750 }
3751
Christoph Lameter81819f02007-05-06 14:49:36 -07003752 slab_state = UP;
3753
3754 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003755 if (KMALLOC_MIN_SIZE <= 32) {
3756 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3757 BUG_ON(!kmalloc_caches[1]->name);
3758 }
3759
3760 if (KMALLOC_MIN_SIZE <= 64) {
3761 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3762 BUG_ON(!kmalloc_caches[2]->name);
3763 }
3764
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003765 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3766 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3767
3768 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003769 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003770 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003771
3772#ifdef CONFIG_SMP
3773 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003774#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003775
Christoph Lameter55136592010-08-20 12:37:13 -05003776#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003777 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3778 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003779
Christoph Lameter51df1142010-08-20 12:37:15 -05003780 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003781 char *name = kasprintf(GFP_NOWAIT,
3782 "dma-kmalloc-%d", s->objsize);
3783
3784 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003785 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3786 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003787 }
3788 }
3789#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003790 printk(KERN_INFO
3791 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003792 " CPUs=%d, Nodes=%d\n",
3793 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003794 slub_min_order, slub_max_order, slub_min_objects,
3795 nr_cpu_ids, nr_node_ids);
3796}
3797
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003798void __init kmem_cache_init_late(void)
3799{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003800}
3801
Christoph Lameter81819f02007-05-06 14:49:36 -07003802/*
3803 * Find a mergeable slab cache
3804 */
3805static int slab_unmergeable(struct kmem_cache *s)
3806{
3807 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3808 return 1;
3809
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003810 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003811 return 1;
3812
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003813 /*
3814 * We may have set a slab to be unmergeable during bootstrap.
3815 */
3816 if (s->refcount < 0)
3817 return 1;
3818
Christoph Lameter81819f02007-05-06 14:49:36 -07003819 return 0;
3820}
3821
3822static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003823 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003824 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003825{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003826 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003827
3828 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3829 return NULL;
3830
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003831 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003832 return NULL;
3833
3834 size = ALIGN(size, sizeof(void *));
3835 align = calculate_alignment(flags, align, size);
3836 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003837 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003838
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003839 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003840 if (slab_unmergeable(s))
3841 continue;
3842
3843 if (size > s->size)
3844 continue;
3845
Christoph Lameterba0268a2007-09-11 15:24:11 -07003846 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003847 continue;
3848 /*
3849 * Check if alignment is compatible.
3850 * Courtesy of Adrian Drzewiecki
3851 */
Pekka Enberg06428782008-01-07 23:20:27 -08003852 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003853 continue;
3854
3855 if (s->size - size >= sizeof(void *))
3856 continue;
3857
3858 return s;
3859 }
3860 return NULL;
3861}
3862
3863struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003864 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003865{
3866 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003867 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003868
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003869 if (WARN_ON(!name))
3870 return NULL;
3871
Christoph Lameter81819f02007-05-06 14:49:36 -07003872 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003873 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003874 if (s) {
3875 s->refcount++;
3876 /*
3877 * Adjust the object sizes so that we clear
3878 * the complete object on kzalloc.
3879 */
3880 s->objsize = max(s->objsize, (int)size);
3881 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003882
David Rientjes7b8f3b62008-12-17 22:09:46 -08003883 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003884 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003885 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003886 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003887 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003888 return s;
3889 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003890
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003891 n = kstrdup(name, GFP_KERNEL);
3892 if (!n)
3893 goto err;
3894
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003895 s = kmalloc(kmem_size, GFP_KERNEL);
3896 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003897 if (kmem_cache_open(s, n,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003898 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003899 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003900 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003901 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003902 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003903 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003904 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003905 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003906 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003907 return s;
3908 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003909 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003910 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003911 }
Pavel Emelyanov68cee4f2010-10-28 13:50:37 +04003912err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003913 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003914
Christoph Lameter81819f02007-05-06 14:49:36 -07003915 if (flags & SLAB_PANIC)
3916 panic("Cannot create slabcache %s\n", name);
3917 else
3918 s = NULL;
3919 return s;
3920}
3921EXPORT_SYMBOL(kmem_cache_create);
3922
Christoph Lameter81819f02007-05-06 14:49:36 -07003923#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003924/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003925 * Use the cpu notifier to insure that the cpu slabs are flushed when
3926 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003927 */
3928static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3929 unsigned long action, void *hcpu)
3930{
3931 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003932 struct kmem_cache *s;
3933 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003934
3935 switch (action) {
3936 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003937 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003938 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003939 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003940 down_read(&slub_lock);
3941 list_for_each_entry(s, &slab_caches, list) {
3942 local_irq_save(flags);
3943 __flush_cpu_slab(s, cpu);
3944 local_irq_restore(flags);
3945 }
3946 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003947 break;
3948 default:
3949 break;
3950 }
3951 return NOTIFY_OK;
3952}
3953
Pekka Enberg06428782008-01-07 23:20:27 -08003954static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003955 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003956};
Christoph Lameter81819f02007-05-06 14:49:36 -07003957
3958#endif
3959
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003960void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003961{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003962 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003963 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003964
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003965 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003966 return kmalloc_large(size, gfpflags);
3967
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003968 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003969
Satyam Sharma2408c552007-10-16 01:24:44 -07003970 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003971 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003972
Christoph Lameter2154a332010-07-09 14:07:10 -05003973 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003974
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003975 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003976 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003977
3978 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003979}
3980
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003981#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003982void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003983 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003984{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003985 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003986 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003987
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003988 if (unlikely(size > SLUB_MAX_SIZE)) {
3989 ret = kmalloc_large_node(size, gfpflags, node);
3990
3991 trace_kmalloc_node(caller, ret,
3992 size, PAGE_SIZE << get_order(size),
3993 gfpflags, node);
3994
3995 return ret;
3996 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003997
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003998 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003999
Satyam Sharma2408c552007-10-16 01:24:44 -07004000 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004001 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004002
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004003 ret = slab_alloc(s, gfpflags, node, caller);
4004
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004005 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02004006 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004007
4008 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004009}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004010#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004011
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004012#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004013static int count_inuse(struct page *page)
4014{
4015 return page->inuse;
4016}
4017
4018static int count_total(struct page *page)
4019{
4020 return page->objects;
4021}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004022#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004023
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004024#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004025static int validate_slab(struct kmem_cache *s, struct page *page,
4026 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004027{
4028 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004029 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004030
4031 if (!check_slab(s, page) ||
4032 !on_freelist(s, page, NULL))
4033 return 0;
4034
4035 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004036 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004037
Christoph Lameter5f80b132011-04-15 14:48:13 -05004038 get_map(s, page, map);
4039 for_each_object(p, s, addr, page->objects) {
4040 if (test_bit(slab_index(p, s, addr), map))
4041 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4042 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004043 }
4044
Christoph Lameter224a88b2008-04-14 19:11:31 +03004045 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004046 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004047 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004048 return 0;
4049 return 1;
4050}
4051
Christoph Lameter434e2452007-07-17 04:03:30 -07004052static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4053 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004054{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004055 slab_lock(page);
4056 validate_slab(s, page, map);
4057 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004058}
4059
Christoph Lameter434e2452007-07-17 04:03:30 -07004060static int validate_slab_node(struct kmem_cache *s,
4061 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004062{
4063 unsigned long count = 0;
4064 struct page *page;
4065 unsigned long flags;
4066
4067 spin_lock_irqsave(&n->list_lock, flags);
4068
4069 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004070 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004071 count++;
4072 }
4073 if (count != n->nr_partial)
4074 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
4075 "counter=%ld\n", s->name, count, n->nr_partial);
4076
4077 if (!(s->flags & SLAB_STORE_USER))
4078 goto out;
4079
4080 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004081 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004082 count++;
4083 }
4084 if (count != atomic_long_read(&n->nr_slabs))
4085 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
4086 "counter=%ld\n", s->name, count,
4087 atomic_long_read(&n->nr_slabs));
4088
4089out:
4090 spin_unlock_irqrestore(&n->list_lock, flags);
4091 return count;
4092}
4093
Christoph Lameter434e2452007-07-17 04:03:30 -07004094static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004095{
4096 int node;
4097 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004098 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004099 sizeof(unsigned long), GFP_KERNEL);
4100
4101 if (!map)
4102 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004103
4104 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07004105 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07004106 struct kmem_cache_node *n = get_node(s, node);
4107
Christoph Lameter434e2452007-07-17 04:03:30 -07004108 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004109 }
Christoph Lameter434e2452007-07-17 04:03:30 -07004110 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004111 return count;
4112}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004113/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004114 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004115 * and freed.
4116 */
4117
4118struct location {
4119 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004120 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004121 long long sum_time;
4122 long min_time;
4123 long max_time;
4124 long min_pid;
4125 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304126 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004127 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004128};
4129
4130struct loc_track {
4131 unsigned long max;
4132 unsigned long count;
4133 struct location *loc;
4134};
4135
4136static void free_loc_track(struct loc_track *t)
4137{
4138 if (t->max)
4139 free_pages((unsigned long)t->loc,
4140 get_order(sizeof(struct location) * t->max));
4141}
4142
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004143static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004144{
4145 struct location *l;
4146 int order;
4147
Christoph Lameter88a420e2007-05-06 14:49:45 -07004148 order = get_order(sizeof(struct location) * max);
4149
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004150 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004151 if (!l)
4152 return 0;
4153
4154 if (t->count) {
4155 memcpy(l, t->loc, sizeof(struct location) * t->count);
4156 free_loc_track(t);
4157 }
4158 t->max = max;
4159 t->loc = l;
4160 return 1;
4161}
4162
4163static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004164 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004165{
4166 long start, end, pos;
4167 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004168 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004169 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004170
4171 start = -1;
4172 end = t->count;
4173
4174 for ( ; ; ) {
4175 pos = start + (end - start + 1) / 2;
4176
4177 /*
4178 * There is nothing at "end". If we end up there
4179 * we need to add something to before end.
4180 */
4181 if (pos == end)
4182 break;
4183
4184 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004185 if (track->addr == caddr) {
4186
4187 l = &t->loc[pos];
4188 l->count++;
4189 if (track->when) {
4190 l->sum_time += age;
4191 if (age < l->min_time)
4192 l->min_time = age;
4193 if (age > l->max_time)
4194 l->max_time = age;
4195
4196 if (track->pid < l->min_pid)
4197 l->min_pid = track->pid;
4198 if (track->pid > l->max_pid)
4199 l->max_pid = track->pid;
4200
Rusty Russell174596a2009-01-01 10:12:29 +10304201 cpumask_set_cpu(track->cpu,
4202 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004203 }
4204 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004205 return 1;
4206 }
4207
Christoph Lameter45edfa52007-05-09 02:32:45 -07004208 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004209 end = pos;
4210 else
4211 start = pos;
4212 }
4213
4214 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004215 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004216 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004217 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004218 return 0;
4219
4220 l = t->loc + pos;
4221 if (pos < t->count)
4222 memmove(l + 1, l,
4223 (t->count - pos) * sizeof(struct location));
4224 t->count++;
4225 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004226 l->addr = track->addr;
4227 l->sum_time = age;
4228 l->min_time = age;
4229 l->max_time = age;
4230 l->min_pid = track->pid;
4231 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304232 cpumask_clear(to_cpumask(l->cpus));
4233 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004234 nodes_clear(l->nodes);
4235 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004236 return 1;
4237}
4238
4239static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004240 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004241 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004242{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004243 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004244 void *p;
4245
Christoph Lameter39b26462008-04-14 19:11:30 +03004246 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004247 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004248
Christoph Lameter224a88b2008-04-14 19:11:31 +03004249 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004250 if (!test_bit(slab_index(p, s, addr), map))
4251 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004252}
4253
4254static int list_locations(struct kmem_cache *s, char *buf,
4255 enum track_item alloc)
4256{
Harvey Harrisone374d482008-01-31 15:20:50 -08004257 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004258 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004259 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004260 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004261 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4262 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004263
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004264 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4265 GFP_TEMPORARY)) {
4266 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004267 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004268 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004269 /* Push back cpu slabs */
4270 flush_all(s);
4271
Christoph Lameterf64dc582007-10-16 01:25:33 -07004272 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004273 struct kmem_cache_node *n = get_node(s, node);
4274 unsigned long flags;
4275 struct page *page;
4276
Christoph Lameter9e869432007-08-22 14:01:56 -07004277 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004278 continue;
4279
4280 spin_lock_irqsave(&n->list_lock, flags);
4281 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004282 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004283 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004284 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004285 spin_unlock_irqrestore(&n->list_lock, flags);
4286 }
4287
4288 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004289 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004290
Hugh Dickins9c246242008-12-09 13:14:27 -08004291 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004292 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004293 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004294
4295 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004296 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004297 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004298 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004299
4300 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004301 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004302 l->min_time,
4303 (long)div_u64(l->sum_time, l->count),
4304 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004305 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004306 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004307 l->min_time);
4308
4309 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004310 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004311 l->min_pid, l->max_pid);
4312 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004313 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004314 l->min_pid);
4315
Rusty Russell174596a2009-01-01 10:12:29 +10304316 if (num_online_cpus() > 1 &&
4317 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004318 len < PAGE_SIZE - 60) {
4319 len += sprintf(buf + len, " cpus=");
4320 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10304321 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004322 }
4323
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004324 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004325 len < PAGE_SIZE - 60) {
4326 len += sprintf(buf + len, " nodes=");
4327 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004328 l->nodes);
4329 }
4330
Harvey Harrisone374d482008-01-31 15:20:50 -08004331 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004332 }
4333
4334 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004335 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004336 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004337 len += sprintf(buf, "No data\n");
4338 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004339}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004340#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004341
Christoph Lametera5a84752010-10-05 13:57:27 -05004342#ifdef SLUB_RESILIENCY_TEST
4343static void resiliency_test(void)
4344{
4345 u8 *p;
4346
4347 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
4348
4349 printk(KERN_ERR "SLUB resiliency testing\n");
4350 printk(KERN_ERR "-----------------------\n");
4351 printk(KERN_ERR "A. Corruption after allocation\n");
4352
4353 p = kzalloc(16, GFP_KERNEL);
4354 p[16] = 0x12;
4355 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
4356 " 0x12->0x%p\n\n", p + 16);
4357
4358 validate_slab_cache(kmalloc_caches[4]);
4359
4360 /* Hmmm... The next two are dangerous */
4361 p = kzalloc(32, GFP_KERNEL);
4362 p[32 + sizeof(void *)] = 0x34;
4363 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
4364 " 0x34 -> -0x%p\n", p);
4365 printk(KERN_ERR
4366 "If allocated object is overwritten then not detectable\n\n");
4367
4368 validate_slab_cache(kmalloc_caches[5]);
4369 p = kzalloc(64, GFP_KERNEL);
4370 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4371 *p = 0x56;
4372 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4373 p);
4374 printk(KERN_ERR
4375 "If allocated object is overwritten then not detectable\n\n");
4376 validate_slab_cache(kmalloc_caches[6]);
4377
4378 printk(KERN_ERR "\nB. Corruption after free\n");
4379 p = kzalloc(128, GFP_KERNEL);
4380 kfree(p);
4381 *p = 0x78;
4382 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4383 validate_slab_cache(kmalloc_caches[7]);
4384
4385 p = kzalloc(256, GFP_KERNEL);
4386 kfree(p);
4387 p[50] = 0x9a;
4388 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4389 p);
4390 validate_slab_cache(kmalloc_caches[8]);
4391
4392 p = kzalloc(512, GFP_KERNEL);
4393 kfree(p);
4394 p[512] = 0xab;
4395 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4396 validate_slab_cache(kmalloc_caches[9]);
4397}
4398#else
4399#ifdef CONFIG_SYSFS
4400static void resiliency_test(void) {};
4401#endif
4402#endif
4403
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004404#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004405enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004406 SL_ALL, /* All slabs */
4407 SL_PARTIAL, /* Only partially allocated slabs */
4408 SL_CPU, /* Only slabs used for cpu caches */
4409 SL_OBJECTS, /* Determine allocated objects not slabs */
4410 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004411};
4412
Christoph Lameter205ab992008-04-14 19:11:40 +03004413#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004414#define SO_PARTIAL (1 << SL_PARTIAL)
4415#define SO_CPU (1 << SL_CPU)
4416#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004417#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004418
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004419static ssize_t show_slab_objects(struct kmem_cache *s,
4420 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004421{
4422 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004423 int node;
4424 int x;
4425 unsigned long *nodes;
4426 unsigned long *per_cpu;
4427
4428 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004429 if (!nodes)
4430 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004431 per_cpu = nodes + nr_node_ids;
4432
Christoph Lameter205ab992008-04-14 19:11:40 +03004433 if (flags & SO_CPU) {
4434 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004435
Christoph Lameter205ab992008-04-14 19:11:40 +03004436 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004437 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter49e22582011-08-09 16:12:27 -05004438 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004439
Christoph Lameter205ab992008-04-14 19:11:40 +03004440 if (!c || c->node < 0)
4441 continue;
4442
4443 if (c->page) {
4444 if (flags & SO_TOTAL)
4445 x = c->page->objects;
4446 else if (flags & SO_OBJECTS)
4447 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07004448 else
4449 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03004450
Christoph Lameter81819f02007-05-06 14:49:36 -07004451 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03004452 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004453 }
Christoph Lameter49e22582011-08-09 16:12:27 -05004454 page = c->partial;
4455
4456 if (page) {
4457 x = page->pobjects;
4458 total += x;
4459 nodes[c->node] += x;
4460 }
Christoph Lameter205ab992008-04-14 19:11:40 +03004461 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004462 }
4463 }
4464
Christoph Lameter04d94872011-01-10 10:15:15 -06004465 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004466#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004467 if (flags & SO_ALL) {
4468 for_each_node_state(node, N_NORMAL_MEMORY) {
4469 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004470
Christoph Lameter205ab992008-04-14 19:11:40 +03004471 if (flags & SO_TOTAL)
4472 x = atomic_long_read(&n->total_objects);
4473 else if (flags & SO_OBJECTS)
4474 x = atomic_long_read(&n->total_objects) -
4475 count_partial(n, count_free);
4476
4477 else
4478 x = atomic_long_read(&n->nr_slabs);
4479 total += x;
4480 nodes[node] += x;
4481 }
4482
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004483 } else
4484#endif
4485 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004486 for_each_node_state(node, N_NORMAL_MEMORY) {
4487 struct kmem_cache_node *n = get_node(s, node);
4488
4489 if (flags & SO_TOTAL)
4490 x = count_partial(n, count_total);
4491 else if (flags & SO_OBJECTS)
4492 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004493 else
4494 x = n->nr_partial;
4495 total += x;
4496 nodes[node] += x;
4497 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004498 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004499 x = sprintf(buf, "%lu", total);
4500#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004501 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004502 if (nodes[node])
4503 x += sprintf(buf + x, " N%d=%lu",
4504 node, nodes[node]);
4505#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004506 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004507 kfree(nodes);
4508 return x + sprintf(buf + x, "\n");
4509}
4510
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004511#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004512static int any_slab_objects(struct kmem_cache *s)
4513{
4514 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004515
4516 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004517 struct kmem_cache_node *n = get_node(s, node);
4518
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004519 if (!n)
4520 continue;
4521
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004522 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004523 return 1;
4524 }
4525 return 0;
4526}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004527#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004528
4529#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004530#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004531
4532struct slab_attribute {
4533 struct attribute attr;
4534 ssize_t (*show)(struct kmem_cache *s, char *buf);
4535 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4536};
4537
4538#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004539 static struct slab_attribute _name##_attr = \
4540 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004541
4542#define SLAB_ATTR(_name) \
4543 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004544 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004545
Christoph Lameter81819f02007-05-06 14:49:36 -07004546static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4547{
4548 return sprintf(buf, "%d\n", s->size);
4549}
4550SLAB_ATTR_RO(slab_size);
4551
4552static ssize_t align_show(struct kmem_cache *s, char *buf)
4553{
4554 return sprintf(buf, "%d\n", s->align);
4555}
4556SLAB_ATTR_RO(align);
4557
4558static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4559{
4560 return sprintf(buf, "%d\n", s->objsize);
4561}
4562SLAB_ATTR_RO(object_size);
4563
4564static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4565{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004566 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004567}
4568SLAB_ATTR_RO(objs_per_slab);
4569
Christoph Lameter06b285d2008-04-14 19:11:41 +03004570static ssize_t order_store(struct kmem_cache *s,
4571 const char *buf, size_t length)
4572{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004573 unsigned long order;
4574 int err;
4575
4576 err = strict_strtoul(buf, 10, &order);
4577 if (err)
4578 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004579
4580 if (order > slub_max_order || order < slub_min_order)
4581 return -EINVAL;
4582
4583 calculate_sizes(s, order);
4584 return length;
4585}
4586
Christoph Lameter81819f02007-05-06 14:49:36 -07004587static ssize_t order_show(struct kmem_cache *s, char *buf)
4588{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004589 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004590}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004591SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004592
David Rientjes73d342b2009-02-22 17:40:09 -08004593static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4594{
4595 return sprintf(buf, "%lu\n", s->min_partial);
4596}
4597
4598static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4599 size_t length)
4600{
4601 unsigned long min;
4602 int err;
4603
4604 err = strict_strtoul(buf, 10, &min);
4605 if (err)
4606 return err;
4607
David Rientjesc0bdb232009-02-25 09:16:35 +02004608 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004609 return length;
4610}
4611SLAB_ATTR(min_partial);
4612
Christoph Lameter49e22582011-08-09 16:12:27 -05004613static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4614{
4615 return sprintf(buf, "%u\n", s->cpu_partial);
4616}
4617
4618static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4619 size_t length)
4620{
4621 unsigned long objects;
4622 int err;
4623
4624 err = strict_strtoul(buf, 10, &objects);
4625 if (err)
4626 return err;
4627
4628 s->cpu_partial = objects;
4629 flush_all(s);
4630 return length;
4631}
4632SLAB_ATTR(cpu_partial);
4633
Christoph Lameter81819f02007-05-06 14:49:36 -07004634static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4635{
Joe Perches62c70bc2011-01-13 15:45:52 -08004636 if (!s->ctor)
4637 return 0;
4638 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004639}
4640SLAB_ATTR_RO(ctor);
4641
Christoph Lameter81819f02007-05-06 14:49:36 -07004642static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4643{
4644 return sprintf(buf, "%d\n", s->refcount - 1);
4645}
4646SLAB_ATTR_RO(aliases);
4647
Christoph Lameter81819f02007-05-06 14:49:36 -07004648static ssize_t partial_show(struct kmem_cache *s, char *buf)
4649{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004650 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004651}
4652SLAB_ATTR_RO(partial);
4653
4654static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4655{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004656 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004657}
4658SLAB_ATTR_RO(cpu_slabs);
4659
4660static ssize_t objects_show(struct kmem_cache *s, char *buf)
4661{
Christoph Lameter205ab992008-04-14 19:11:40 +03004662 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004663}
4664SLAB_ATTR_RO(objects);
4665
Christoph Lameter205ab992008-04-14 19:11:40 +03004666static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4667{
4668 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4669}
4670SLAB_ATTR_RO(objects_partial);
4671
Christoph Lameter49e22582011-08-09 16:12:27 -05004672static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4673{
4674 int objects = 0;
4675 int pages = 0;
4676 int cpu;
4677 int len;
4678
4679 for_each_online_cpu(cpu) {
4680 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4681
4682 if (page) {
4683 pages += page->pages;
4684 objects += page->pobjects;
4685 }
4686 }
4687
4688 len = sprintf(buf, "%d(%d)", objects, pages);
4689
4690#ifdef CONFIG_SMP
4691 for_each_online_cpu(cpu) {
4692 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4693
4694 if (page && len < PAGE_SIZE - 20)
4695 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4696 page->pobjects, page->pages);
4697 }
4698#endif
4699 return len + sprintf(buf + len, "\n");
4700}
4701SLAB_ATTR_RO(slabs_cpu_partial);
4702
Christoph Lameter81819f02007-05-06 14:49:36 -07004703static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4704{
4705 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4706}
4707
4708static ssize_t reclaim_account_store(struct kmem_cache *s,
4709 const char *buf, size_t length)
4710{
4711 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4712 if (buf[0] == '1')
4713 s->flags |= SLAB_RECLAIM_ACCOUNT;
4714 return length;
4715}
4716SLAB_ATTR(reclaim_account);
4717
4718static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4719{
Christoph Lameter5af60832007-05-06 14:49:56 -07004720 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004721}
4722SLAB_ATTR_RO(hwcache_align);
4723
4724#ifdef CONFIG_ZONE_DMA
4725static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4726{
4727 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4728}
4729SLAB_ATTR_RO(cache_dma);
4730#endif
4731
4732static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4733{
4734 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4735}
4736SLAB_ATTR_RO(destroy_by_rcu);
4737
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004738static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4739{
4740 return sprintf(buf, "%d\n", s->reserved);
4741}
4742SLAB_ATTR_RO(reserved);
4743
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004744#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004745static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4746{
4747 return show_slab_objects(s, buf, SO_ALL);
4748}
4749SLAB_ATTR_RO(slabs);
4750
4751static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4752{
4753 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4754}
4755SLAB_ATTR_RO(total_objects);
4756
4757static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4758{
4759 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4760}
4761
4762static ssize_t sanity_checks_store(struct kmem_cache *s,
4763 const char *buf, size_t length)
4764{
4765 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004766 if (buf[0] == '1') {
4767 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004768 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004769 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004770 return length;
4771}
4772SLAB_ATTR(sanity_checks);
4773
4774static ssize_t trace_show(struct kmem_cache *s, char *buf)
4775{
4776 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4777}
4778
4779static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4780 size_t length)
4781{
4782 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004783 if (buf[0] == '1') {
4784 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004785 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004786 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004787 return length;
4788}
4789SLAB_ATTR(trace);
4790
Christoph Lameter81819f02007-05-06 14:49:36 -07004791static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4792{
4793 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4794}
4795
4796static ssize_t red_zone_store(struct kmem_cache *s,
4797 const char *buf, size_t length)
4798{
4799 if (any_slab_objects(s))
4800 return -EBUSY;
4801
4802 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004803 if (buf[0] == '1') {
4804 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004805 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004806 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004807 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004808 return length;
4809}
4810SLAB_ATTR(red_zone);
4811
4812static ssize_t poison_show(struct kmem_cache *s, char *buf)
4813{
4814 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4815}
4816
4817static ssize_t poison_store(struct kmem_cache *s,
4818 const char *buf, size_t length)
4819{
4820 if (any_slab_objects(s))
4821 return -EBUSY;
4822
4823 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004824 if (buf[0] == '1') {
4825 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004826 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004827 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004828 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004829 return length;
4830}
4831SLAB_ATTR(poison);
4832
4833static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4834{
4835 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4836}
4837
4838static ssize_t store_user_store(struct kmem_cache *s,
4839 const char *buf, size_t length)
4840{
4841 if (any_slab_objects(s))
4842 return -EBUSY;
4843
4844 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004845 if (buf[0] == '1') {
4846 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004847 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004848 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004849 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004850 return length;
4851}
4852SLAB_ATTR(store_user);
4853
Christoph Lameter53e15af2007-05-06 14:49:43 -07004854static ssize_t validate_show(struct kmem_cache *s, char *buf)
4855{
4856 return 0;
4857}
4858
4859static ssize_t validate_store(struct kmem_cache *s,
4860 const char *buf, size_t length)
4861{
Christoph Lameter434e2452007-07-17 04:03:30 -07004862 int ret = -EINVAL;
4863
4864 if (buf[0] == '1') {
4865 ret = validate_slab_cache(s);
4866 if (ret >= 0)
4867 ret = length;
4868 }
4869 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004870}
4871SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004872
4873static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4874{
4875 if (!(s->flags & SLAB_STORE_USER))
4876 return -ENOSYS;
4877 return list_locations(s, buf, TRACK_ALLOC);
4878}
4879SLAB_ATTR_RO(alloc_calls);
4880
4881static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4882{
4883 if (!(s->flags & SLAB_STORE_USER))
4884 return -ENOSYS;
4885 return list_locations(s, buf, TRACK_FREE);
4886}
4887SLAB_ATTR_RO(free_calls);
4888#endif /* CONFIG_SLUB_DEBUG */
4889
4890#ifdef CONFIG_FAILSLAB
4891static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4892{
4893 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4894}
4895
4896static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4897 size_t length)
4898{
4899 s->flags &= ~SLAB_FAILSLAB;
4900 if (buf[0] == '1')
4901 s->flags |= SLAB_FAILSLAB;
4902 return length;
4903}
4904SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004905#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004906
Christoph Lameter2086d262007-05-06 14:49:46 -07004907static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4908{
4909 return 0;
4910}
4911
4912static ssize_t shrink_store(struct kmem_cache *s,
4913 const char *buf, size_t length)
4914{
4915 if (buf[0] == '1') {
4916 int rc = kmem_cache_shrink(s);
4917
4918 if (rc)
4919 return rc;
4920 } else
4921 return -EINVAL;
4922 return length;
4923}
4924SLAB_ATTR(shrink);
4925
Christoph Lameter81819f02007-05-06 14:49:36 -07004926#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004927static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004928{
Christoph Lameter98246012008-01-07 23:20:26 -08004929 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004930}
4931
Christoph Lameter98246012008-01-07 23:20:26 -08004932static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004933 const char *buf, size_t length)
4934{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004935 unsigned long ratio;
4936 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004937
Christoph Lameter0121c6192008-04-29 16:11:12 -07004938 err = strict_strtoul(buf, 10, &ratio);
4939 if (err)
4940 return err;
4941
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004942 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004943 s->remote_node_defrag_ratio = ratio * 10;
4944
Christoph Lameter81819f02007-05-06 14:49:36 -07004945 return length;
4946}
Christoph Lameter98246012008-01-07 23:20:26 -08004947SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004948#endif
4949
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004950#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004951static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4952{
4953 unsigned long sum = 0;
4954 int cpu;
4955 int len;
4956 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4957
4958 if (!data)
4959 return -ENOMEM;
4960
4961 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004962 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004963
4964 data[cpu] = x;
4965 sum += x;
4966 }
4967
4968 len = sprintf(buf, "%lu", sum);
4969
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004970#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004971 for_each_online_cpu(cpu) {
4972 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004973 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004974 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004975#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004976 kfree(data);
4977 return len + sprintf(buf + len, "\n");
4978}
4979
David Rientjes78eb00c2009-10-15 02:20:22 -07004980static void clear_stat(struct kmem_cache *s, enum stat_item si)
4981{
4982 int cpu;
4983
4984 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004985 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004986}
4987
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004988#define STAT_ATTR(si, text) \
4989static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4990{ \
4991 return show_stat(s, buf, si); \
4992} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004993static ssize_t text##_store(struct kmem_cache *s, \
4994 const char *buf, size_t length) \
4995{ \
4996 if (buf[0] != '0') \
4997 return -EINVAL; \
4998 clear_stat(s, si); \
4999 return length; \
5000} \
5001SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005002
5003STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5004STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5005STAT_ATTR(FREE_FASTPATH, free_fastpath);
5006STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5007STAT_ATTR(FREE_FROZEN, free_frozen);
5008STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5009STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5010STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5011STAT_ATTR(ALLOC_SLAB, alloc_slab);
5012STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005013STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005014STAT_ATTR(FREE_SLAB, free_slab);
5015STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5016STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5017STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5018STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5019STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5020STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005021STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005022STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005023STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5024STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005025STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5026STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005027#endif
5028
Pekka Enberg06428782008-01-07 23:20:27 -08005029static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005030 &slab_size_attr.attr,
5031 &object_size_attr.attr,
5032 &objs_per_slab_attr.attr,
5033 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005034 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005035 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005036 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005037 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005038 &partial_attr.attr,
5039 &cpu_slabs_attr.attr,
5040 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005041 &aliases_attr.attr,
5042 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005043 &hwcache_align_attr.attr,
5044 &reclaim_account_attr.attr,
5045 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005046 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005047 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005048 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005049#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005050 &total_objects_attr.attr,
5051 &slabs_attr.attr,
5052 &sanity_checks_attr.attr,
5053 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005054 &red_zone_attr.attr,
5055 &poison_attr.attr,
5056 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005057 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005058 &alloc_calls_attr.attr,
5059 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005060#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005061#ifdef CONFIG_ZONE_DMA
5062 &cache_dma_attr.attr,
5063#endif
5064#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005065 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005066#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005067#ifdef CONFIG_SLUB_STATS
5068 &alloc_fastpath_attr.attr,
5069 &alloc_slowpath_attr.attr,
5070 &free_fastpath_attr.attr,
5071 &free_slowpath_attr.attr,
5072 &free_frozen_attr.attr,
5073 &free_add_partial_attr.attr,
5074 &free_remove_partial_attr.attr,
5075 &alloc_from_partial_attr.attr,
5076 &alloc_slab_attr.attr,
5077 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005078 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005079 &free_slab_attr.attr,
5080 &cpuslab_flush_attr.attr,
5081 &deactivate_full_attr.attr,
5082 &deactivate_empty_attr.attr,
5083 &deactivate_to_head_attr.attr,
5084 &deactivate_to_tail_attr.attr,
5085 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005086 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005087 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005088 &cmpxchg_double_fail_attr.attr,
5089 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005090 &cpu_partial_alloc_attr.attr,
5091 &cpu_partial_free_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005092#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005093#ifdef CONFIG_FAILSLAB
5094 &failslab_attr.attr,
5095#endif
5096
Christoph Lameter81819f02007-05-06 14:49:36 -07005097 NULL
5098};
5099
5100static struct attribute_group slab_attr_group = {
5101 .attrs = slab_attrs,
5102};
5103
5104static ssize_t slab_attr_show(struct kobject *kobj,
5105 struct attribute *attr,
5106 char *buf)
5107{
5108 struct slab_attribute *attribute;
5109 struct kmem_cache *s;
5110 int err;
5111
5112 attribute = to_slab_attr(attr);
5113 s = to_slab(kobj);
5114
5115 if (!attribute->show)
5116 return -EIO;
5117
5118 err = attribute->show(s, buf);
5119
5120 return err;
5121}
5122
5123static ssize_t slab_attr_store(struct kobject *kobj,
5124 struct attribute *attr,
5125 const char *buf, size_t len)
5126{
5127 struct slab_attribute *attribute;
5128 struct kmem_cache *s;
5129 int err;
5130
5131 attribute = to_slab_attr(attr);
5132 s = to_slab(kobj);
5133
5134 if (!attribute->store)
5135 return -EIO;
5136
5137 err = attribute->store(s, buf, len);
5138
5139 return err;
5140}
5141
Christoph Lameter151c6022008-01-07 22:29:05 -08005142static void kmem_cache_release(struct kobject *kobj)
5143{
5144 struct kmem_cache *s = to_slab(kobj);
5145
Pekka Enberg84c1cf62010-09-14 23:21:12 +03005146 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08005147 kfree(s);
5148}
5149
Emese Revfy52cf25d2010-01-19 02:58:23 +01005150static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005151 .show = slab_attr_show,
5152 .store = slab_attr_store,
5153};
5154
5155static struct kobj_type slab_ktype = {
5156 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08005157 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07005158};
5159
5160static int uevent_filter(struct kset *kset, struct kobject *kobj)
5161{
5162 struct kobj_type *ktype = get_ktype(kobj);
5163
5164 if (ktype == &slab_ktype)
5165 return 1;
5166 return 0;
5167}
5168
Emese Revfy9cd43612009-12-31 14:52:51 +01005169static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005170 .filter = uevent_filter,
5171};
5172
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005173static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005174
5175#define ID_STR_LENGTH 64
5176
5177/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005178 *
5179 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005180 */
5181static char *create_unique_id(struct kmem_cache *s)
5182{
5183 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5184 char *p = name;
5185
5186 BUG_ON(!name);
5187
5188 *p++ = ':';
5189 /*
5190 * First flags affecting slabcache operations. We will only
5191 * get here for aliasable slabs so we do not need to support
5192 * too many flags. The flags here must cover all flags that
5193 * are matched during merging to guarantee that the id is
5194 * unique.
5195 */
5196 if (s->flags & SLAB_CACHE_DMA)
5197 *p++ = 'd';
5198 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5199 *p++ = 'a';
5200 if (s->flags & SLAB_DEBUG_FREE)
5201 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005202 if (!(s->flags & SLAB_NOTRACK))
5203 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005204 if (p != name + 1)
5205 *p++ = '-';
5206 p += sprintf(p, "%07d", s->size);
5207 BUG_ON(p > name + ID_STR_LENGTH - 1);
5208 return name;
5209}
5210
5211static int sysfs_slab_add(struct kmem_cache *s)
5212{
5213 int err;
5214 const char *name;
5215 int unmergeable;
5216
5217 if (slab_state < SYSFS)
5218 /* Defer until later */
5219 return 0;
5220
5221 unmergeable = slab_unmergeable(s);
5222 if (unmergeable) {
5223 /*
5224 * Slabcache can never be merged so we can use the name proper.
5225 * This is typically the case for debug situations. In that
5226 * case we can catch duplicate names easily.
5227 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005228 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005229 name = s->name;
5230 } else {
5231 /*
5232 * Create a unique name for the slab as a target
5233 * for the symlinks.
5234 */
5235 name = create_unique_id(s);
5236 }
5237
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005238 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005239 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
5240 if (err) {
5241 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005242 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005243 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005244
5245 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005246 if (err) {
5247 kobject_del(&s->kobj);
5248 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005249 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005250 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005251 kobject_uevent(&s->kobj, KOBJ_ADD);
5252 if (!unmergeable) {
5253 /* Setup first alias */
5254 sysfs_slab_alias(s, s->name);
5255 kfree(name);
5256 }
5257 return 0;
5258}
5259
5260static void sysfs_slab_remove(struct kmem_cache *s)
5261{
Christoph Lameter2bce6482010-07-19 11:39:11 -05005262 if (slab_state < SYSFS)
5263 /*
5264 * Sysfs has not been setup yet so no need to remove the
5265 * cache from sysfs.
5266 */
5267 return;
5268
Christoph Lameter81819f02007-05-06 14:49:36 -07005269 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5270 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005271 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005272}
5273
5274/*
5275 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005276 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005277 */
5278struct saved_alias {
5279 struct kmem_cache *s;
5280 const char *name;
5281 struct saved_alias *next;
5282};
5283
Adrian Bunk5af328a2007-07-17 04:03:27 -07005284static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005285
5286static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5287{
5288 struct saved_alias *al;
5289
5290 if (slab_state == SYSFS) {
5291 /*
5292 * If we have a leftover link then remove it.
5293 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005294 sysfs_remove_link(&slab_kset->kobj, name);
5295 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005296 }
5297
5298 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5299 if (!al)
5300 return -ENOMEM;
5301
5302 al->s = s;
5303 al->name = name;
5304 al->next = alias_list;
5305 alias_list = al;
5306 return 0;
5307}
5308
5309static int __init slab_sysfs_init(void)
5310{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07005311 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005312 int err;
5313
Christoph Lameter2bce6482010-07-19 11:39:11 -05005314 down_write(&slub_lock);
5315
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005316 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005317 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05005318 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005319 printk(KERN_ERR "Cannot register slab subsystem.\n");
5320 return -ENOSYS;
5321 }
5322
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005323 slab_state = SYSFS;
5324
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07005325 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005326 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005327 if (err)
5328 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
5329 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005330 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005331
5332 while (alias_list) {
5333 struct saved_alias *al = alias_list;
5334
5335 alias_list = alias_list->next;
5336 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005337 if (err)
5338 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
5339 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005340 kfree(al);
5341 }
5342
Christoph Lameter2bce6482010-07-19 11:39:11 -05005343 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005344 resiliency_test();
5345 return 0;
5346}
5347
5348__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005349#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005350
5351/*
5352 * The /proc/slabinfo ABI
5353 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005354#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005355static void print_slabinfo_header(struct seq_file *m)
5356{
5357 seq_puts(m, "slabinfo - version: 2.1\n");
5358 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
5359 "<objperslab> <pagesperslab>");
5360 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
5361 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5362 seq_putc(m, '\n');
5363}
5364
5365static void *s_start(struct seq_file *m, loff_t *pos)
5366{
5367 loff_t n = *pos;
5368
5369 down_read(&slub_lock);
5370 if (!n)
5371 print_slabinfo_header(m);
5372
5373 return seq_list_start(&slab_caches, *pos);
5374}
5375
5376static void *s_next(struct seq_file *m, void *p, loff_t *pos)
5377{
5378 return seq_list_next(p, &slab_caches, pos);
5379}
5380
5381static void s_stop(struct seq_file *m, void *p)
5382{
5383 up_read(&slub_lock);
5384}
5385
5386static int s_show(struct seq_file *m, void *p)
5387{
5388 unsigned long nr_partials = 0;
5389 unsigned long nr_slabs = 0;
5390 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005391 unsigned long nr_objs = 0;
5392 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005393 struct kmem_cache *s;
5394 int node;
5395
5396 s = list_entry(p, struct kmem_cache, list);
5397
5398 for_each_online_node(node) {
5399 struct kmem_cache_node *n = get_node(s, node);
5400
5401 if (!n)
5402 continue;
5403
5404 nr_partials += n->nr_partial;
5405 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03005406 nr_objs += atomic_long_read(&n->total_objects);
5407 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005408 }
5409
Christoph Lameter205ab992008-04-14 19:11:40 +03005410 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005411
5412 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03005413 nr_objs, s->size, oo_objects(s->oo),
5414 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005415 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
5416 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
5417 0UL);
5418 seq_putc(m, '\n');
5419 return 0;
5420}
5421
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005422static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005423 .start = s_start,
5424 .next = s_next,
5425 .stop = s_stop,
5426 .show = s_show,
5427};
5428
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005429static int slabinfo_open(struct inode *inode, struct file *file)
5430{
5431 return seq_open(file, &slabinfo_op);
5432}
5433
5434static const struct file_operations proc_slabinfo_operations = {
5435 .open = slabinfo_open,
5436 .read = seq_read,
5437 .llseek = seq_lseek,
5438 .release = seq_release,
5439};
5440
5441static int __init slab_proc_init(void)
5442{
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04005443 proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005444 return 0;
5445}
5446module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08005447#endif /* CONFIG_SLABINFO */