blob: 91a120f185d18bb28a8050ec8279efb016271a5e [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 Lameter84e554e62009-12-18 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{
470 int i, offset;
471 int newline = 1;
472 char ascii[17];
473
474 ascii[16] = 0;
475
476 for (i = 0; i < length; i++) {
477 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700478 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700479 newline = 0;
480 }
Pekka Enberg06428782008-01-07 23:20:27 -0800481 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700482 offset = i % 16;
483 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
484 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800485 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700486 newline = 1;
487 }
488 }
489 if (!newline) {
490 i %= 16;
491 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800492 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700493 ascii[i] = ' ';
494 i++;
495 }
Pekka Enberg06428782008-01-07 23:20:27 -0800496 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700497 }
498}
499
Christoph Lameter81819f02007-05-06 14:49:36 -0700500static struct track *get_track(struct kmem_cache *s, void *object,
501 enum track_item alloc)
502{
503 struct track *p;
504
505 if (s->offset)
506 p = object + s->offset + sizeof(void *);
507 else
508 p = object + s->inuse;
509
510 return p + alloc;
511}
512
513static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300514 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700515{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900516 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700517
Christoph Lameter81819f02007-05-06 14:49:36 -0700518 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700519#ifdef CONFIG_STACKTRACE
520 struct stack_trace trace;
521 int i;
522
523 trace.nr_entries = 0;
524 trace.max_entries = TRACK_ADDRS_COUNT;
525 trace.entries = p->addrs;
526 trace.skip = 3;
527 save_stack_trace(&trace);
528
529 /* See rant in lockdep.c */
530 if (trace.nr_entries != 0 &&
531 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
532 trace.nr_entries--;
533
534 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
535 p->addrs[i] = 0;
536#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700537 p->addr = addr;
538 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400539 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700540 p->when = jiffies;
541 } else
542 memset(p, 0, sizeof(struct track));
543}
544
Christoph Lameter81819f02007-05-06 14:49:36 -0700545static void init_tracking(struct kmem_cache *s, void *object)
546{
Christoph Lameter24922682007-07-17 04:03:18 -0700547 if (!(s->flags & SLAB_STORE_USER))
548 return;
549
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300550 set_track(s, object, TRACK_FREE, 0UL);
551 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700552}
553
554static void print_track(const char *s, struct track *t)
555{
556 if (!t->addr)
557 return;
558
Linus Torvalds7daf7052008-07-14 12:12:53 -0700559 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300560 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700561#ifdef CONFIG_STACKTRACE
562 {
563 int i;
564 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
565 if (t->addrs[i])
566 printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
567 else
568 break;
569 }
570#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700571}
572
Christoph Lameter24922682007-07-17 04:03:18 -0700573static void print_tracking(struct kmem_cache *s, void *object)
574{
575 if (!(s->flags & SLAB_STORE_USER))
576 return;
577
578 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
579 print_track("Freed", get_track(s, object, TRACK_FREE));
580}
581
582static void print_page_info(struct page *page)
583{
Christoph Lameter39b26462008-04-14 19:11:30 +0300584 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
585 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700586
587}
588
589static void slab_bug(struct kmem_cache *s, char *fmt, ...)
590{
591 va_list args;
592 char buf[100];
593
594 va_start(args, fmt);
595 vsnprintf(buf, sizeof(buf), fmt, args);
596 va_end(args);
597 printk(KERN_ERR "========================================"
598 "=====================================\n");
599 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
600 printk(KERN_ERR "----------------------------------------"
601 "-------------------------------------\n\n");
602}
603
604static void slab_fix(struct kmem_cache *s, char *fmt, ...)
605{
606 va_list args;
607 char buf[100];
608
609 va_start(args, fmt);
610 vsnprintf(buf, sizeof(buf), fmt, args);
611 va_end(args);
612 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
613}
614
615static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700616{
617 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800618 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700619
620 print_tracking(s, p);
621
622 print_page_info(page);
623
624 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
625 p, p - addr, get_freepointer(s, p));
626
627 if (p > addr + 16)
628 print_section("Bytes b4", p - 16, 16);
629
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300630 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700631
632 if (s->flags & SLAB_RED_ZONE)
633 print_section("Redzone", p + s->objsize,
634 s->inuse - s->objsize);
635
Christoph Lameter81819f02007-05-06 14:49:36 -0700636 if (s->offset)
637 off = s->offset + sizeof(void *);
638 else
639 off = s->inuse;
640
Christoph Lameter24922682007-07-17 04:03:18 -0700641 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700642 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700643
644 if (off != s->size)
645 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700646 print_section("Padding", p + off, s->size - off);
647
648 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700649}
650
651static void object_err(struct kmem_cache *s, struct page *page,
652 u8 *object, char *reason)
653{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700654 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700655 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700656}
657
Christoph Lameter24922682007-07-17 04:03:18 -0700658static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700659{
660 va_list args;
661 char buf[100];
662
Christoph Lameter24922682007-07-17 04:03:18 -0700663 va_start(args, fmt);
664 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700665 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700666 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700667 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700668 dump_stack();
669}
670
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500671static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700672{
673 u8 *p = object;
674
675 if (s->flags & __OBJECT_POISON) {
676 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800677 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700678 }
679
680 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500681 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700682}
683
Marcin Slusarzc4089f92011-06-26 21:39:18 +0200684static u8 *check_bytes8(u8 *start, u8 value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700685{
686 while (bytes) {
Marcin Slusarzc4089f92011-06-26 21:39:18 +0200687 if (*start != value)
Christoph Lameter24922682007-07-17 04:03:18 -0700688 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700689 start++;
690 bytes--;
691 }
Christoph Lameter24922682007-07-17 04:03:18 -0700692 return NULL;
693}
694
Marcin Slusarzc4089f92011-06-26 21:39:18 +0200695static u8 *check_bytes(u8 *start, u8 value, unsigned int bytes)
696{
697 u64 value64;
698 unsigned int words, prefix;
699
700 if (bytes <= 16)
701 return check_bytes8(start, value, bytes);
702
703 value64 = value | value << 8 | value << 16 | value << 24;
Akinobu Mitaef62fb32011-08-07 18:30:38 +0900704 value64 = (value64 & 0xffffffff) | value64 << 32;
Marcin Slusarzc4089f92011-06-26 21:39:18 +0200705 prefix = 8 - ((unsigned long)start) % 8;
706
707 if (prefix) {
708 u8 *r = check_bytes8(start, value, prefix);
709 if (r)
710 return r;
711 start += prefix;
712 bytes -= prefix;
713 }
714
715 words = bytes / 8;
716
717 while (words) {
718 if (*(u64 *)start != value64)
719 return check_bytes8(start, value, 8);
720 start += 8;
721 words--;
722 }
723
724 return check_bytes8(start, value, bytes % 8);
725}
726
Christoph Lameter24922682007-07-17 04:03:18 -0700727static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
728 void *from, void *to)
729{
730 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
731 memset(from, data, to - from);
732}
733
734static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
735 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800736 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700737{
738 u8 *fault;
739 u8 *end;
740
741 fault = check_bytes(start, value, bytes);
742 if (!fault)
743 return 1;
744
745 end = start + bytes;
746 while (end > fault && end[-1] == value)
747 end--;
748
749 slab_bug(s, "%s overwritten", what);
750 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
751 fault, end - 1, fault[0], value);
752 print_trailer(s, page, object);
753
754 restore_bytes(s, what, value, fault, end);
755 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700756}
757
Christoph Lameter81819f02007-05-06 14:49:36 -0700758/*
759 * Object layout:
760 *
761 * object address
762 * Bytes of the object to be managed.
763 * If the freepointer may overlay the object then the free
764 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700765 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700766 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
767 * 0xa5 (POISON_END)
768 *
769 * object + s->objsize
770 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700771 * Padding is extended by another word if Redzoning is enabled and
772 * objsize == inuse.
773 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700774 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
775 * 0xcc (RED_ACTIVE) for objects in use.
776 *
777 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700778 * Meta data starts here.
779 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700780 * A. Free pointer (if we cannot overwrite object on free)
781 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700782 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800783 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700784 * before the word boundary.
785 *
786 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700787 *
788 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700789 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700790 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700791 * If slabcaches are merged then the objsize and inuse boundaries are mostly
792 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700793 * may be used with merged slabcaches.
794 */
795
Christoph Lameter81819f02007-05-06 14:49:36 -0700796static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
797{
798 unsigned long off = s->inuse; /* The end of info */
799
800 if (s->offset)
801 /* Freepointer is placed after the object. */
802 off += sizeof(void *);
803
804 if (s->flags & SLAB_STORE_USER)
805 /* We also have user information there */
806 off += 2 * sizeof(struct track);
807
808 if (s->size == off)
809 return 1;
810
Christoph Lameter24922682007-07-17 04:03:18 -0700811 return check_bytes_and_report(s, page, p, "Object padding",
812 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700813}
814
Christoph Lameter39b26462008-04-14 19:11:30 +0300815/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700816static int slab_pad_check(struct kmem_cache *s, struct page *page)
817{
Christoph Lameter24922682007-07-17 04:03:18 -0700818 u8 *start;
819 u8 *fault;
820 u8 *end;
821 int length;
822 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700823
824 if (!(s->flags & SLAB_POISON))
825 return 1;
826
Christoph Lametera973e9d2008-03-01 13:40:44 -0800827 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800828 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300829 end = start + length;
830 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700831 if (!remainder)
832 return 1;
833
Christoph Lameter39b26462008-04-14 19:11:30 +0300834 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700835 if (!fault)
836 return 1;
837 while (end > fault && end[-1] == POISON_INUSE)
838 end--;
839
840 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300841 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700842
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200843 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700844 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700845}
846
847static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500848 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700849{
850 u8 *p = object;
851 u8 *endobject = object + s->objsize;
852
853 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700854 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500855 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700856 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700857 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800858 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
859 check_bytes_and_report(s, page, p, "Alignment padding",
860 endobject, POISON_INUSE, s->inuse - s->objsize);
861 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700862 }
863
864 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500865 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700866 (!check_bytes_and_report(s, page, p, "Poison", p,
867 POISON_FREE, s->objsize - 1) ||
868 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800869 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700870 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700871 /*
872 * check_pad_bytes cleans up on its own.
873 */
874 check_pad_bytes(s, page, p);
875 }
876
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500877 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700878 /*
879 * Object and freepointer overlap. Cannot check
880 * freepointer while object is allocated.
881 */
882 return 1;
883
884 /* Check free pointer validity */
885 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
886 object_err(s, page, p, "Freepointer corrupt");
887 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100888 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700889 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700890 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700891 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800892 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700893 return 0;
894 }
895 return 1;
896}
897
898static int check_slab(struct kmem_cache *s, struct page *page)
899{
Christoph Lameter39b26462008-04-14 19:11:30 +0300900 int maxobj;
901
Christoph Lameter81819f02007-05-06 14:49:36 -0700902 VM_BUG_ON(!irqs_disabled());
903
904 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700905 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700906 return 0;
907 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300908
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800909 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300910 if (page->objects > maxobj) {
911 slab_err(s, page, "objects %u > max %u",
912 s->name, page->objects, maxobj);
913 return 0;
914 }
915 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700916 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300917 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700918 return 0;
919 }
920 /* Slab_pad_check fixes things up after itself */
921 slab_pad_check(s, page);
922 return 1;
923}
924
925/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700926 * Determine if a certain object on a page is on the freelist. Must hold the
927 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700928 */
929static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
930{
931 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500932 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700933 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300934 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700935
Christoph Lameter881db7f2011-06-01 12:25:53 -0500936 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300937 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700938 if (fp == search)
939 return 1;
940 if (!check_valid_pointer(s, page, fp)) {
941 if (object) {
942 object_err(s, page, object,
943 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800944 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700945 break;
946 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700947 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800948 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300949 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700950 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700951 return 0;
952 }
953 break;
954 }
955 object = fp;
956 fp = get_freepointer(s, object);
957 nr++;
958 }
959
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800960 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400961 if (max_objects > MAX_OBJS_PER_PAGE)
962 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300963
964 if (page->objects != max_objects) {
965 slab_err(s, page, "Wrong number of objects. Found %d but "
966 "should be %d", page->objects, max_objects);
967 page->objects = max_objects;
968 slab_fix(s, "Number of objects adjusted.");
969 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300970 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700971 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300972 "counted were %d", page->inuse, page->objects - nr);
973 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700974 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700975 }
976 return search == NULL;
977}
978
Christoph Lameter0121c6192008-04-29 16:11:12 -0700979static void trace(struct kmem_cache *s, struct page *page, void *object,
980 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700981{
982 if (s->flags & SLAB_TRACE) {
983 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
984 s->name,
985 alloc ? "alloc" : "free",
986 object, page->inuse,
987 page->freelist);
988
989 if (!alloc)
990 print_section("Object", (void *)object, s->objsize);
991
992 dump_stack();
993 }
994}
995
Christoph Lameter643b1132007-05-06 14:49:42 -0700996/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500997 * Hooks for other subsystems that check memory allocations. In a typical
998 * production configuration these hooks all should produce no code at all.
999 */
1000static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1001{
Christoph Lameterc1d50832010-08-20 12:37:17 -05001002 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001003 lockdep_trace_alloc(flags);
1004 might_sleep_if(flags & __GFP_WAIT);
1005
1006 return should_failslab(s->objsize, flags, s->flags);
1007}
1008
1009static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
1010{
Christoph Lameterc1d50832010-08-20 12:37:17 -05001011 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +01001012 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001013 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
1014}
1015
1016static inline void slab_free_hook(struct kmem_cache *s, void *x)
1017{
1018 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001019
Christoph Lameterd3f661d2011-02-25 11:38:52 -06001020 /*
1021 * Trouble is that we may no longer disable interupts in the fast path
1022 * So in order to make the debug calls that expect irqs to be
1023 * disabled we need to disable interrupts temporarily.
1024 */
1025#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1026 {
1027 unsigned long flags;
1028
1029 local_irq_save(flags);
1030 kmemcheck_slab_free(s, x, s->objsize);
1031 debug_check_no_locks_freed(x, s->objsize);
Christoph Lameterd3f661d2011-02-25 11:38:52 -06001032 local_irq_restore(flags);
1033 }
1034#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +02001035 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1036 debug_check_no_obj_freed(x, s->objsize);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001037}
1038
1039/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001040 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001041 *
1042 * list_lock must be held.
Christoph Lameter643b1132007-05-06 14:49:42 -07001043 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001044static void add_full(struct kmem_cache *s,
1045 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -07001046{
Christoph Lameter643b1132007-05-06 14:49:42 -07001047 if (!(s->flags & SLAB_STORE_USER))
1048 return;
1049
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001050 list_add(&page->lru, &n->full);
1051}
Christoph Lameter643b1132007-05-06 14:49:42 -07001052
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001053/*
1054 * list_lock must be held.
1055 */
1056static void remove_full(struct kmem_cache *s, struct page *page)
1057{
1058 if (!(s->flags & SLAB_STORE_USER))
1059 return;
1060
Christoph Lameter643b1132007-05-06 14:49:42 -07001061 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001062}
1063
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001064/* Tracking of the number of slabs for debugging purposes */
1065static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1066{
1067 struct kmem_cache_node *n = get_node(s, node);
1068
1069 return atomic_long_read(&n->nr_slabs);
1070}
1071
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001072static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1073{
1074 return atomic_long_read(&n->nr_slabs);
1075}
1076
Christoph Lameter205ab992008-04-14 19:11:40 +03001077static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001078{
1079 struct kmem_cache_node *n = get_node(s, node);
1080
1081 /*
1082 * May be called early in order to allocate a slab for the
1083 * kmem_cache_node structure. Solve the chicken-egg
1084 * dilemma by deferring the increment of the count during
1085 * bootstrap (see early_kmem_cache_node_alloc).
1086 */
Christoph Lameter7340cc82010-09-28 08:10:26 -05001087 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001088 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001089 atomic_long_add(objects, &n->total_objects);
1090 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001091}
Christoph Lameter205ab992008-04-14 19:11:40 +03001092static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001093{
1094 struct kmem_cache_node *n = get_node(s, node);
1095
1096 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001097 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001098}
1099
1100/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001101static void setup_object_debug(struct kmem_cache *s, struct page *page,
1102 void *object)
1103{
1104 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1105 return;
1106
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001107 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001108 init_tracking(s, object);
1109}
1110
Christoph Lameter15370662010-08-20 12:37:12 -05001111static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001112 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001113{
1114 if (!check_slab(s, page))
1115 goto bad;
1116
Christoph Lameter81819f02007-05-06 14:49:36 -07001117 if (!check_valid_pointer(s, page, object)) {
1118 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001119 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001120 }
1121
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001122 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001123 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001124
Christoph Lameter3ec09742007-05-16 22:11:00 -07001125 /* Success perform special debug activities for allocs */
1126 if (s->flags & SLAB_STORE_USER)
1127 set_track(s, object, TRACK_ALLOC, addr);
1128 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001129 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001130 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001131
Christoph Lameter81819f02007-05-06 14:49:36 -07001132bad:
1133 if (PageSlab(page)) {
1134 /*
1135 * If this is a slab page then lets do the best we can
1136 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001137 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001138 */
Christoph Lameter24922682007-07-17 04:03:18 -07001139 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001140 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001141 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001142 }
1143 return 0;
1144}
1145
Christoph Lameter15370662010-08-20 12:37:12 -05001146static noinline int free_debug_processing(struct kmem_cache *s,
1147 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001148{
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001149 unsigned long flags;
1150 int rc = 0;
1151
1152 local_irq_save(flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001153 slab_lock(page);
1154
Christoph Lameter81819f02007-05-06 14:49:36 -07001155 if (!check_slab(s, page))
1156 goto fail;
1157
1158 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001159 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001160 goto fail;
1161 }
1162
1163 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001164 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001165 goto fail;
1166 }
1167
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001168 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001169 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001170
1171 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001172 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001173 slab_err(s, page, "Attempt to free object(0x%p) "
1174 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001175 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001176 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001177 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001178 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001179 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001180 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001181 object_err(s, page, object,
1182 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001183 goto fail;
1184 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001185
Christoph Lameter3ec09742007-05-16 22:11:00 -07001186 if (s->flags & SLAB_STORE_USER)
1187 set_track(s, object, TRACK_FREE, addr);
1188 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001189 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001190 rc = 1;
1191out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001192 slab_unlock(page);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001193 local_irq_restore(flags);
1194 return rc;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001195
Christoph Lameter81819f02007-05-06 14:49:36 -07001196fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001197 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001198 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001199}
1200
Christoph Lameter41ecc552007-05-09 02:32:44 -07001201static int __init setup_slub_debug(char *str)
1202{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001203 slub_debug = DEBUG_DEFAULT_FLAGS;
1204 if (*str++ != '=' || !*str)
1205 /*
1206 * No options specified. Switch on full debugging.
1207 */
1208 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001209
1210 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001211 /*
1212 * No options but restriction on slabs. This means full
1213 * debugging for slabs matching a pattern.
1214 */
1215 goto check_slabs;
1216
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001217 if (tolower(*str) == 'o') {
1218 /*
1219 * Avoid enabling debugging on caches if its minimum order
1220 * would increase as a result.
1221 */
1222 disable_higher_order_debug = 1;
1223 goto out;
1224 }
1225
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001226 slub_debug = 0;
1227 if (*str == '-')
1228 /*
1229 * Switch off all debugging measures.
1230 */
1231 goto out;
1232
1233 /*
1234 * Determine which debug features should be switched on
1235 */
Pekka Enberg06428782008-01-07 23:20:27 -08001236 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001237 switch (tolower(*str)) {
1238 case 'f':
1239 slub_debug |= SLAB_DEBUG_FREE;
1240 break;
1241 case 'z':
1242 slub_debug |= SLAB_RED_ZONE;
1243 break;
1244 case 'p':
1245 slub_debug |= SLAB_POISON;
1246 break;
1247 case 'u':
1248 slub_debug |= SLAB_STORE_USER;
1249 break;
1250 case 't':
1251 slub_debug |= SLAB_TRACE;
1252 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001253 case 'a':
1254 slub_debug |= SLAB_FAILSLAB;
1255 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001256 default:
1257 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001258 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001259 }
1260 }
1261
1262check_slabs:
1263 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001264 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001265out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001266 return 1;
1267}
1268
1269__setup("slub_debug", setup_slub_debug);
1270
Christoph Lameterba0268a2007-09-11 15:24:11 -07001271static unsigned long kmem_cache_flags(unsigned long objsize,
1272 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001273 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001274{
1275 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001276 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001277 */
Christoph Lametere1533622008-02-15 23:45:24 -08001278 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001279 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1280 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001281
1282 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001283}
1284#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001285static inline void setup_object_debug(struct kmem_cache *s,
1286 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001287
Christoph Lameter3ec09742007-05-16 22:11:00 -07001288static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001289 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001290
Christoph Lameter3ec09742007-05-16 22:11:00 -07001291static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001292 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001293
Christoph Lameter41ecc552007-05-09 02:32:44 -07001294static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1295 { return 1; }
1296static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001297 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001298static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1299 struct page *page) {}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001300static inline void remove_full(struct kmem_cache *s, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001301static inline unsigned long kmem_cache_flags(unsigned long objsize,
1302 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001303 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001304{
1305 return flags;
1306}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001307#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001308
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001309#define disable_higher_order_debug 0
1310
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001311static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1312 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001313static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1314 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001315static inline void inc_slabs_node(struct kmem_cache *s, int node,
1316 int objects) {}
1317static inline void dec_slabs_node(struct kmem_cache *s, int node,
1318 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001319
1320static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1321 { return 0; }
1322
1323static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1324 void *object) {}
1325
1326static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1327
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001328#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001329
Christoph Lameter81819f02007-05-06 14:49:36 -07001330/*
1331 * Slab allocation and freeing
1332 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001333static inline struct page *alloc_slab_page(gfp_t flags, int node,
1334 struct kmem_cache_order_objects oo)
1335{
1336 int order = oo_order(oo);
1337
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001338 flags |= __GFP_NOTRACK;
1339
Christoph Lameter2154a332010-07-09 14:07:10 -05001340 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001341 return alloc_pages(flags, order);
1342 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001343 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001344}
1345
Christoph Lameter81819f02007-05-06 14:49:36 -07001346static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1347{
Pekka Enberg06428782008-01-07 23:20:27 -08001348 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001349 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001350 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001351
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001352 flags &= gfp_allowed_mask;
1353
1354 if (flags & __GFP_WAIT)
1355 local_irq_enable();
1356
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001357 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001358
Pekka Enbergba522702009-06-24 21:59:51 +03001359 /*
1360 * Let the initial higher-order allocation fail under memory pressure
1361 * so we fall-back to the minimum order allocation.
1362 */
1363 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1364
1365 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001366 if (unlikely(!page)) {
1367 oo = s->min;
1368 /*
1369 * Allocation may have failed due to fragmentation.
1370 * Try a lower order alloc if possible
1371 */
1372 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001373
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001374 if (page)
1375 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001376 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001377
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001378 if (flags & __GFP_WAIT)
1379 local_irq_disable();
1380
1381 if (!page)
1382 return NULL;
1383
Vegard Nossum5a896d92008-04-04 00:54:48 +02001384 if (kmemcheck_enabled
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001385 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001386 int pages = 1 << oo_order(oo);
1387
1388 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1389
1390 /*
1391 * Objects from caches that have a constructor don't get
1392 * cleared when they're allocated, so we need to do it here.
1393 */
1394 if (s->ctor)
1395 kmemcheck_mark_uninitialized_pages(page, pages);
1396 else
1397 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001398 }
1399
Christoph Lameter834f3d12008-04-14 19:11:31 +03001400 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001401 mod_zone_page_state(page_zone(page),
1402 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1403 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001404 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001405
1406 return page;
1407}
1408
1409static void setup_object(struct kmem_cache *s, struct page *page,
1410 void *object)
1411{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001412 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001413 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001414 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001415}
1416
1417static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1418{
1419 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001420 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001421 void *last;
1422 void *p;
1423
Christoph Lameter6cb06222007-10-16 01:25:41 -07001424 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001425
Christoph Lameter6cb06222007-10-16 01:25:41 -07001426 page = allocate_slab(s,
1427 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001428 if (!page)
1429 goto out;
1430
Christoph Lameter205ab992008-04-14 19:11:40 +03001431 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001432 page->slab = s;
1433 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001434
1435 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001436
1437 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001438 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001439
1440 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001441 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001442 setup_object(s, page, last);
1443 set_freepointer(s, last, p);
1444 last = p;
1445 }
1446 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001447 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001448
1449 page->freelist = start;
1450 page->inuse = 0;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001451 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001452out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001453 return page;
1454}
1455
1456static void __free_slab(struct kmem_cache *s, struct page *page)
1457{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001458 int order = compound_order(page);
1459 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001460
Christoph Lameteraf537b02010-07-09 14:07:14 -05001461 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001462 void *p;
1463
1464 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001465 for_each_object(p, s, page_address(page),
1466 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001467 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001468 }
1469
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001470 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001471
Christoph Lameter81819f02007-05-06 14:49:36 -07001472 mod_zone_page_state(page_zone(page),
1473 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1474 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001475 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001476
Christoph Lameter49bd5222008-04-14 18:52:18 +03001477 __ClearPageSlab(page);
1478 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001479 if (current->reclaim_state)
1480 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001481 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001482}
1483
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001484#define need_reserve_slab_rcu \
1485 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1486
Christoph Lameter81819f02007-05-06 14:49:36 -07001487static void rcu_free_slab(struct rcu_head *h)
1488{
1489 struct page *page;
1490
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001491 if (need_reserve_slab_rcu)
1492 page = virt_to_head_page(h);
1493 else
1494 page = container_of((struct list_head *)h, struct page, lru);
1495
Christoph Lameter81819f02007-05-06 14:49:36 -07001496 __free_slab(page->slab, page);
1497}
1498
1499static void free_slab(struct kmem_cache *s, struct page *page)
1500{
1501 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001502 struct rcu_head *head;
1503
1504 if (need_reserve_slab_rcu) {
1505 int order = compound_order(page);
1506 int offset = (PAGE_SIZE << order) - s->reserved;
1507
1508 VM_BUG_ON(s->reserved != sizeof(*head));
1509 head = page_address(page) + offset;
1510 } else {
1511 /*
1512 * RCU free overloads the RCU head over the LRU
1513 */
1514 head = (void *)&page->lru;
1515 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001516
1517 call_rcu(head, rcu_free_slab);
1518 } else
1519 __free_slab(s, page);
1520}
1521
1522static void discard_slab(struct kmem_cache *s, struct page *page)
1523{
Christoph Lameter205ab992008-04-14 19:11:40 +03001524 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001525 free_slab(s, page);
1526}
1527
1528/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001529 * Management of partially allocated slabs.
1530 *
1531 * list_lock must be held.
Christoph Lameter81819f02007-05-06 14:49:36 -07001532 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001533static inline void add_partial(struct kmem_cache_node *n,
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001534 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001535{
Christoph Lametere95eed52007-05-06 14:49:44 -07001536 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001537 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001538 list_add_tail(&page->lru, &n->partial);
1539 else
1540 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001541}
1542
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001543/*
1544 * list_lock must be held.
1545 */
1546static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001547 struct page *page)
1548{
1549 list_del(&page->lru);
1550 n->nr_partial--;
1551}
1552
Christoph Lameter81819f02007-05-06 14:49:36 -07001553/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001554 * Lock slab, remove from the partial list and put the object into the
1555 * per cpu freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001556 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001557 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001558 */
Christoph Lameter881db7f2011-06-01 12:25:53 -05001559static inline int acquire_slab(struct kmem_cache *s,
Christoph Lameter61728d12011-06-01 12:25:51 -05001560 struct kmem_cache_node *n, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001561{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001562 void *freelist;
1563 unsigned long counters;
1564 struct page new;
1565
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001566 /*
1567 * Zap the freelist and set the frozen bit.
1568 * The old freelist is the list of objects for the
1569 * per cpu allocation list.
1570 */
1571 do {
1572 freelist = page->freelist;
1573 counters = page->counters;
1574 new.counters = counters;
1575 new.inuse = page->objects;
1576
1577 VM_BUG_ON(new.frozen);
1578 new.frozen = 1;
1579
Christoph Lameter1d071712011-07-14 12:49:12 -05001580 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001581 freelist, counters,
1582 NULL, new.counters,
1583 "lock and freeze"));
1584
1585 remove_partial(n, page);
1586
1587 if (freelist) {
1588 /* Populate the per cpu freelist */
1589 this_cpu_write(s->cpu_slab->freelist, freelist);
1590 this_cpu_write(s->cpu_slab->page, page);
1591 this_cpu_write(s->cpu_slab->node, page_to_nid(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001592 return 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001593 } else {
1594 /*
1595 * Slab page came from the wrong list. No object to allocate
1596 * from. Put it onto the correct list and continue partial
1597 * scan.
1598 */
1599 printk(KERN_ERR "SLUB: %s : Page without available objects on"
1600 " partial list\n", s->name);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001601 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001602 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001603}
1604
1605/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001606 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001607 */
Christoph Lameter61728d12011-06-01 12:25:51 -05001608static struct page *get_partial_node(struct kmem_cache *s,
1609 struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07001610{
1611 struct page *page;
1612
1613 /*
1614 * Racy check. If we mistakenly see no partial slabs then we
1615 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001616 * partial slab and there is none available then get_partials()
1617 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001618 */
1619 if (!n || !n->nr_partial)
1620 return NULL;
1621
1622 spin_lock(&n->list_lock);
1623 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter881db7f2011-06-01 12:25:53 -05001624 if (acquire_slab(s, n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001625 goto out;
1626 page = NULL;
1627out:
1628 spin_unlock(&n->list_lock);
1629 return page;
1630}
1631
1632/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001633 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001634 */
1635static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1636{
1637#ifdef CONFIG_NUMA
1638 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001639 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001640 struct zone *zone;
1641 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001642 struct page *page;
1643
1644 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001645 * The defrag ratio allows a configuration of the tradeoffs between
1646 * inter node defragmentation and node local allocations. A lower
1647 * defrag_ratio increases the tendency to do local allocations
1648 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001649 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001650 * If the defrag_ratio is set to 0 then kmalloc() always
1651 * returns node local objects. If the ratio is higher then kmalloc()
1652 * may return off node objects because partial slabs are obtained
1653 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001654 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001655 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001656 * defrag_ratio = 1000) then every (well almost) allocation will
1657 * first attempt to defrag slab caches on other nodes. This means
1658 * scanning over all nodes to look for partial slabs which may be
1659 * expensive if we do it every time we are trying to find a slab
1660 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001661 */
Christoph Lameter98246012008-01-07 23:20:26 -08001662 if (!s->remote_node_defrag_ratio ||
1663 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001664 return NULL;
1665
Miao Xiec0ff7452010-05-24 14:32:08 -07001666 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001667 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001668 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001669 struct kmem_cache_node *n;
1670
Mel Gorman54a6eb52008-04-28 02:12:16 -07001671 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001672
Mel Gorman54a6eb52008-04-28 02:12:16 -07001673 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001674 n->nr_partial > s->min_partial) {
Christoph Lameter61728d12011-06-01 12:25:51 -05001675 page = get_partial_node(s, n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001676 if (page) {
1677 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001678 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001679 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001680 }
1681 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001682 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001683#endif
1684 return NULL;
1685}
1686
1687/*
1688 * Get a partial page, lock it and return it.
1689 */
1690static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1691{
1692 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001693 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001694
Christoph Lameter61728d12011-06-01 12:25:51 -05001695 page = get_partial_node(s, get_node(s, searchnode));
Christoph Lameter33de04e2011-04-15 14:48:12 -05001696 if (page || node != NUMA_NO_NODE)
Christoph Lameter81819f02007-05-06 14:49:36 -07001697 return page;
1698
1699 return get_any_partial(s, flags);
1700}
1701
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001702#ifdef CONFIG_PREEMPT
1703/*
1704 * Calculate the next globally unique transaction for disambiguiation
1705 * during cmpxchg. The transactions start with the cpu number and are then
1706 * incremented by CONFIG_NR_CPUS.
1707 */
1708#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1709#else
1710/*
1711 * No preemption supported therefore also no need to check for
1712 * different cpus.
1713 */
1714#define TID_STEP 1
1715#endif
1716
1717static inline unsigned long next_tid(unsigned long tid)
1718{
1719 return tid + TID_STEP;
1720}
1721
1722static inline unsigned int tid_to_cpu(unsigned long tid)
1723{
1724 return tid % TID_STEP;
1725}
1726
1727static inline unsigned long tid_to_event(unsigned long tid)
1728{
1729 return tid / TID_STEP;
1730}
1731
1732static inline unsigned int init_tid(int cpu)
1733{
1734 return cpu;
1735}
1736
1737static inline void note_cmpxchg_failure(const char *n,
1738 const struct kmem_cache *s, unsigned long tid)
1739{
1740#ifdef SLUB_DEBUG_CMPXCHG
1741 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1742
1743 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1744
1745#ifdef CONFIG_PREEMPT
1746 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1747 printk("due to cpu change %d -> %d\n",
1748 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1749 else
1750#endif
1751 if (tid_to_event(tid) != tid_to_event(actual_tid))
1752 printk("due to cpu running other code. Event %ld->%ld\n",
1753 tid_to_event(tid), tid_to_event(actual_tid));
1754 else
1755 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1756 actual_tid, tid, next_tid(tid));
1757#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001758 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001759}
1760
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001761void init_kmem_cache_cpus(struct kmem_cache *s)
1762{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001763 int cpu;
1764
1765 for_each_possible_cpu(cpu)
1766 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001767}
Christoph Lameter81819f02007-05-06 14:49:36 -07001768/*
1769 * Remove the cpu slab
1770 */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001771
1772/*
1773 * Remove the cpu slab
1774 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001775static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001776{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001777 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001778 struct page *page = c->page;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001779 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1780 int lock = 0;
1781 enum slab_modes l = M_NONE, m = M_NONE;
1782 void *freelist;
1783 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001784 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001785 struct page new;
1786 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001787
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001788 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001789 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001790 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001791 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001792
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001793 c->tid = next_tid(c->tid);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001794 c->page = NULL;
1795 freelist = c->freelist;
1796 c->freelist = NULL;
1797
1798 /*
1799 * Stage one: Free all available per cpu objects back
1800 * to the page freelist while it is still frozen. Leave the
1801 * last one.
1802 *
1803 * There is no need to take the list->lock because the page
1804 * is still frozen.
1805 */
1806 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1807 void *prior;
1808 unsigned long counters;
1809
1810 do {
1811 prior = page->freelist;
1812 counters = page->counters;
1813 set_freepointer(s, freelist, prior);
1814 new.counters = counters;
1815 new.inuse--;
1816 VM_BUG_ON(!new.frozen);
1817
Christoph Lameter1d071712011-07-14 12:49:12 -05001818 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001819 prior, counters,
1820 freelist, new.counters,
1821 "drain percpu freelist"));
1822
1823 freelist = nextfree;
1824 }
1825
1826 /*
1827 * Stage two: Ensure that the page is unfrozen while the
1828 * list presence reflects the actual number of objects
1829 * during unfreeze.
1830 *
1831 * We setup the list membership and then perform a cmpxchg
1832 * with the count. If there is a mismatch then the page
1833 * is not unfrozen but the page is on the wrong list.
1834 *
1835 * Then we restart the process which may have to remove
1836 * the page from the list that we just put it on again
1837 * because the number of objects in the slab may have
1838 * changed.
1839 */
1840redo:
1841
1842 old.freelist = page->freelist;
1843 old.counters = page->counters;
1844 VM_BUG_ON(!old.frozen);
1845
1846 /* Determine target state of the slab */
1847 new.counters = old.counters;
1848 if (freelist) {
1849 new.inuse--;
1850 set_freepointer(s, freelist, old.freelist);
1851 new.freelist = freelist;
1852 } else
1853 new.freelist = old.freelist;
1854
1855 new.frozen = 0;
1856
Christoph Lameter81107182011-08-09 13:01:32 -05001857 if (!new.inuse && n->nr_partial > s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001858 m = M_FREE;
1859 else if (new.freelist) {
1860 m = M_PARTIAL;
1861 if (!lock) {
1862 lock = 1;
1863 /*
1864 * Taking the spinlock removes the possiblity
1865 * that acquire_slab() will see a slab page that
1866 * is frozen
1867 */
1868 spin_lock(&n->list_lock);
1869 }
1870 } else {
1871 m = M_FULL;
1872 if (kmem_cache_debug(s) && !lock) {
1873 lock = 1;
1874 /*
1875 * This also ensures that the scanning of full
1876 * slabs from diagnostic functions will not see
1877 * any frozen slabs.
1878 */
1879 spin_lock(&n->list_lock);
1880 }
1881 }
1882
1883 if (l != m) {
1884
1885 if (l == M_PARTIAL)
1886
1887 remove_partial(n, page);
1888
1889 else if (l == M_FULL)
1890
1891 remove_full(s, page);
1892
1893 if (m == M_PARTIAL) {
1894
1895 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001896 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001897
1898 } else if (m == M_FULL) {
1899
1900 stat(s, DEACTIVATE_FULL);
1901 add_full(s, n, page);
1902
1903 }
1904 }
1905
1906 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001907 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001908 old.freelist, old.counters,
1909 new.freelist, new.counters,
1910 "unfreezing slab"))
1911 goto redo;
1912
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001913 if (lock)
1914 spin_unlock(&n->list_lock);
1915
1916 if (m == M_FREE) {
1917 stat(s, DEACTIVATE_EMPTY);
1918 discard_slab(s, page);
1919 stat(s, FREE_SLAB);
1920 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001921}
1922
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001923static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001924{
Christoph Lameter84e554e62009-12-18 16:26:23 -06001925 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001926 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001927}
1928
1929/*
1930 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001931 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001932 * Called from IPI handler with interrupts disabled.
1933 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001934static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001935{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001936 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001937
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001938 if (likely(c && c->page))
1939 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001940}
1941
1942static void flush_cpu_slab(void *d)
1943{
1944 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001945
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001946 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001947}
1948
1949static void flush_all(struct kmem_cache *s)
1950{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001951 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001952}
1953
1954/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001955 * Check if the objects in a per cpu structure fit numa
1956 * locality expectations.
1957 */
1958static inline int node_match(struct kmem_cache_cpu *c, int node)
1959{
1960#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001961 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001962 return 0;
1963#endif
1964 return 1;
1965}
1966
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001967static int count_free(struct page *page)
1968{
1969 return page->objects - page->inuse;
1970}
1971
1972static unsigned long count_partial(struct kmem_cache_node *n,
1973 int (*get_count)(struct page *))
1974{
1975 unsigned long flags;
1976 unsigned long x = 0;
1977 struct page *page;
1978
1979 spin_lock_irqsave(&n->list_lock, flags);
1980 list_for_each_entry(page, &n->partial, lru)
1981 x += get_count(page);
1982 spin_unlock_irqrestore(&n->list_lock, flags);
1983 return x;
1984}
1985
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001986static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1987{
1988#ifdef CONFIG_SLUB_DEBUG
1989 return atomic_long_read(&n->total_objects);
1990#else
1991 return 0;
1992#endif
1993}
1994
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001995static noinline void
1996slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1997{
1998 int node;
1999
2000 printk(KERN_WARNING
2001 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2002 nid, gfpflags);
2003 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
2004 "default order: %d, min order: %d\n", s->name, s->objsize,
2005 s->size, oo_order(s->oo), oo_order(s->min));
2006
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002007 if (oo_order(s->min) > get_order(s->objsize))
2008 printk(KERN_WARNING " %s debugging increased min order, use "
2009 "slub_debug=O to disable.\n", s->name);
2010
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002011 for_each_online_node(node) {
2012 struct kmem_cache_node *n = get_node(s, node);
2013 unsigned long nr_slabs;
2014 unsigned long nr_objs;
2015 unsigned long nr_free;
2016
2017 if (!n)
2018 continue;
2019
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002020 nr_free = count_partial(n, count_free);
2021 nr_slabs = node_nr_slabs(n);
2022 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002023
2024 printk(KERN_WARNING
2025 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2026 node, nr_slabs, nr_objs, nr_free);
2027 }
2028}
2029
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002030/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002031 * Slow path. The lockless freelist is empty or we need to perform
2032 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002033 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002034 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07002035 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002036 * Processing is still very fast if new objects have been freed to the
2037 * regular freelist. In that case we simply take over the regular freelist
2038 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002039 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002040 * If that is not working then we fall back to the partial lists. We take the
2041 * first element of the freelist as the object to allocate now and move the
2042 * rest of the freelist to the lockless freelist.
2043 *
2044 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002045 * we need to allocate a new slab. This is the slowest path since it involves
2046 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002047 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002048static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2049 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002050{
Christoph Lameter81819f02007-05-06 14:49:36 -07002051 void **object;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002052 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002053 unsigned long flags;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002054 struct page new;
2055 unsigned long counters;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002056
2057 local_irq_save(flags);
2058#ifdef CONFIG_PREEMPT
2059 /*
2060 * We may have been preempted and rescheduled on a different
2061 * cpu before disabling interrupts. Need to reload cpu area
2062 * pointer.
2063 */
2064 c = this_cpu_ptr(s->cpu_slab);
2065#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002066
Linus Torvaldse72e9c22008-03-27 20:56:33 -07002067 /* We handle __GFP_ZERO in the caller */
2068 gfpflags &= ~__GFP_ZERO;
2069
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002070 page = c->page;
2071 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002072 goto new_slab;
2073
Christoph Lameterfc59c052011-06-01 12:25:56 -05002074 if (unlikely(!node_match(c, node))) {
Christoph Lametere36a2652011-06-01 12:25:57 -05002075 stat(s, ALLOC_NODE_MISMATCH);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002076 deactivate_slab(s, c);
2077 goto new_slab;
2078 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002079
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002080 stat(s, ALLOC_SLOWPATH);
2081
2082 do {
2083 object = page->freelist;
2084 counters = page->counters;
2085 new.counters = counters;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002086 VM_BUG_ON(!new.frozen);
2087
Christoph Lameter03e404a2011-06-01 12:25:58 -05002088 /*
2089 * If there is no object left then we use this loop to
2090 * deactivate the slab which is simple since no objects
2091 * are left in the slab and therefore we do not need to
2092 * put the page back onto the partial list.
2093 *
2094 * If there are objects left then we retrieve them
2095 * and use them to refill the per cpu queue.
2096 */
2097
2098 new.inuse = page->objects;
2099 new.frozen = object != NULL;
2100
Christoph Lameter1d071712011-07-14 12:49:12 -05002101 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002102 object, counters,
2103 NULL, new.counters,
2104 "__slab_alloc"));
Christoph Lameter6446faa2008-02-15 23:45:26 -08002105
Christoph Lameter03e404a2011-06-01 12:25:58 -05002106 if (unlikely(!object)) {
2107 c->page = NULL;
2108 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002109 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002110 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002111
Christoph Lameter81819f02007-05-06 14:49:36 -07002112 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002113
Christoph Lameter894b8782007-05-10 03:15:16 -07002114load_freelist:
Christoph Lameter4eade542011-06-01 12:25:59 -05002115 VM_BUG_ON(!page->frozen);
Christoph Lameterff120592009-12-18 16:26:22 -06002116 c->freelist = get_freepointer(s, object);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002117 c->tid = next_tid(c->tid);
2118 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002119 return object;
2120
Christoph Lameter81819f02007-05-06 14:49:36 -07002121new_slab:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002122 page = get_partial(s, gfpflags, node);
2123 if (page) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002124 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002125 object = c->freelist;
2126
2127 if (kmem_cache_debug(s))
2128 goto debug;
Christoph Lameter894b8782007-05-10 03:15:16 -07002129 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002130 }
2131
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002132 page = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07002133
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002134 if (page) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002135 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter05aa3452007-11-05 11:31:58 -08002136 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002137 flush_slab(s, c);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002138
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002139 /*
2140 * No other reference to the page yet so we can
2141 * muck around with it freely without cmpxchg
2142 */
2143 object = page->freelist;
2144 page->freelist = NULL;
2145 page->inuse = page->objects;
2146
2147 stat(s, ALLOC_SLAB);
David Rientjesbd07d872011-05-12 13:10:49 -07002148 c->node = page_to_nid(page);
2149 c->page = page;
Christoph Lameter9e577e82011-07-22 09:35:14 -05002150
2151 if (kmem_cache_debug(s))
2152 goto debug;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07002153 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002154 }
Pekka Enberg95f85982009-06-11 16:18:09 +03002155 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
2156 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter2fd66c52011-03-22 13:32:53 -05002157 local_irq_restore(flags);
Christoph Lameter71c7a062008-02-14 14:28:01 -08002158 return NULL;
Christoph Lameter894b8782007-05-10 03:15:16 -07002159
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002160debug:
2161 if (!object || !alloc_debug_processing(s, page, object, addr))
2162 goto new_slab;
2163
2164 c->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05002165 deactivate_slab(s, c);
2166 c->page = NULL;
Pekka Enberg15b7c512010-10-02 11:32:32 +03002167 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05002168 local_irq_restore(flags);
2169 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07002170}
2171
2172/*
2173 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2174 * have the fastpath folded into their functions. So no function call
2175 * overhead for requests that can be satisfied on the fastpath.
2176 *
2177 * The fastpath works by first checking if the lockless freelist can be used.
2178 * If not then __slab_alloc is called for slow processing.
2179 *
2180 * Otherwise we can simply pick the next object from the lockless free list.
2181 */
Pekka Enberg06428782008-01-07 23:20:27 -08002182static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002183 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002184{
Christoph Lameter894b8782007-05-10 03:15:16 -07002185 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002186 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002187 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002188
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002189 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09002190 return NULL;
2191
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002192redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002193
2194 /*
2195 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2196 * enabled. We may switch back and forth between cpus while
2197 * reading from one cpu area. That does not matter as long
2198 * as we end up on the original cpu again when doing the cmpxchg.
2199 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002200 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002201
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002202 /*
2203 * The transaction ids are globally unique per cpu and per operation on
2204 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2205 * occurs on the right processor and that there was no operation on the
2206 * linked list in between.
2207 */
2208 tid = c->tid;
2209 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002210
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002211 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002212 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07002213
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002214 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07002215
2216 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002217 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002218 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002219 * operation and if we are on the right processor.
2220 *
2221 * The cmpxchg does the following atomically (without lock semantics!)
2222 * 1. Relocate first pointer to the current per cpu area.
2223 * 2. Verify that tid and freelist have not been changed
2224 * 3. If they were not changed replace tid and freelist
2225 *
2226 * Since this is without lock semantics the protection is only against
2227 * code executing on this cpu *not* from access by other cpus.
2228 */
Thomas Gleixner30106b82011-05-04 15:38:19 +02002229 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002230 s->cpu_slab->freelist, s->cpu_slab->tid,
2231 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05002232 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002233
2234 note_cmpxchg_failure("slab_alloc", s, tid);
2235 goto redo;
2236 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002237 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002238 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002239
Pekka Enberg74e21342009-11-25 20:14:48 +02002240 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06002241 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002242
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002243 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002244
Christoph Lameter894b8782007-05-10 03:15:16 -07002245 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002246}
2247
2248void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2249{
Christoph Lameter2154a332010-07-09 14:07:10 -05002250 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002251
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002252 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002253
2254 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002255}
2256EXPORT_SYMBOL(kmem_cache_alloc);
2257
Li Zefan0f24f122009-12-11 15:45:30 +08002258#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002259void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002260{
Richard Kennedy4a923792010-10-21 10:29:19 +01002261 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
2262 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2263 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002264}
Richard Kennedy4a923792010-10-21 10:29:19 +01002265EXPORT_SYMBOL(kmem_cache_alloc_trace);
2266
2267void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2268{
2269 void *ret = kmalloc_order(size, flags, order);
2270 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2271 return ret;
2272}
2273EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002274#endif
2275
Christoph Lameter81819f02007-05-06 14:49:36 -07002276#ifdef CONFIG_NUMA
2277void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2278{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002279 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2280
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002281 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2282 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002283
2284 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002285}
2286EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002287
Li Zefan0f24f122009-12-11 15:45:30 +08002288#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002289void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002290 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002291 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002292{
Richard Kennedy4a923792010-10-21 10:29:19 +01002293 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2294
2295 trace_kmalloc_node(_RET_IP_, ret,
2296 size, s->size, gfpflags, node);
2297 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002298}
Richard Kennedy4a923792010-10-21 10:29:19 +01002299EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002300#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002301#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002302
Christoph Lameter81819f02007-05-06 14:49:36 -07002303/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002304 * Slow patch handling. This may still be called frequently since objects
2305 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002306 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002307 * So we still attempt to reduce cache line usage. Just take the slab
2308 * lock and free the item. If there is no additional partial page
2309 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002310 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002311static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002312 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002313{
2314 void *prior;
2315 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002316 int was_frozen;
2317 int inuse;
2318 struct page new;
2319 unsigned long counters;
2320 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002321 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002322
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002323 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002324
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002325 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002326 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002327
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002328 do {
2329 prior = page->freelist;
2330 counters = page->counters;
2331 set_freepointer(s, object, prior);
2332 new.counters = counters;
2333 was_frozen = new.frozen;
2334 new.inuse--;
2335 if ((!new.inuse || !prior) && !was_frozen && !n) {
2336 n = get_node(s, page_to_nid(page));
2337 /*
2338 * Speculatively acquire the list_lock.
2339 * If the cmpxchg does not succeed then we may
2340 * drop the list_lock without any processing.
2341 *
2342 * Otherwise the list_lock will synchronize with
2343 * other processors updating the list of slabs.
2344 */
Christoph Lameter80f08c12011-06-01 12:25:55 -05002345 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002346 }
2347 inuse = new.inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002348
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002349 } while (!cmpxchg_double_slab(s, page,
2350 prior, counters,
2351 object, new.counters,
2352 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002353
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002354 if (likely(!n)) {
2355 /*
2356 * The list lock was not taken therefore no list
2357 * activity can be necessary.
2358 */
2359 if (was_frozen)
2360 stat(s, FREE_FROZEN);
Christoph Lameter80f08c12011-06-01 12:25:55 -05002361 return;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002362 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002363
2364 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002365 * was_frozen may have been set after we acquired the list_lock in
2366 * an earlier loop. So we need to check it here again.
Christoph Lameter81819f02007-05-06 14:49:36 -07002367 */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002368 if (was_frozen)
2369 stat(s, FREE_FROZEN);
2370 else {
2371 if (unlikely(!inuse && n->nr_partial > s->min_partial))
2372 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002373
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002374 /*
2375 * Objects left in the slab. If it was not on the partial list before
2376 * then add it.
2377 */
2378 if (unlikely(!prior)) {
2379 remove_full(s, page);
Shaohua Li136333d2011-08-24 08:57:52 +08002380 add_partial(n, page, DEACTIVATE_TO_TAIL);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002381 stat(s, FREE_ADD_PARTIAL);
2382 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002383 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002384 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002385 return;
2386
2387slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002388 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002389 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002390 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002391 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002392 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002393 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002394 } else
2395 /* Slab must be on the full list */
2396 remove_full(s, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002397
Christoph Lameter80f08c12011-06-01 12:25:55 -05002398 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002399 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002400 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002401}
2402
Christoph Lameter894b8782007-05-10 03:15:16 -07002403/*
2404 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2405 * can perform fastpath freeing without additional function calls.
2406 *
2407 * The fastpath is only possible if we are freeing to the current cpu slab
2408 * of this processor. This typically the case if we have just allocated
2409 * the item before.
2410 *
2411 * If fastpath is not possible then fall back to __slab_free where we deal
2412 * with all sorts of special processing.
2413 */
Pekka Enberg06428782008-01-07 23:20:27 -08002414static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002415 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002416{
2417 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002418 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002419 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002420
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002421 slab_free_hook(s, x);
2422
Christoph Lametera24c5a02011-03-15 12:45:21 -05002423redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002424
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002425 /*
2426 * Determine the currently cpus per cpu slab.
2427 * The cpu may change afterward. However that does not matter since
2428 * data is retrieved via this pointer. If we are on the same cpu
2429 * during the cmpxchg then the free will succedd.
2430 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002431 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002432
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002433 tid = c->tid;
2434 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002435
Christoph Lameter442b06b2011-05-17 16:29:31 -05002436 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002437 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002438
Thomas Gleixner30106b82011-05-04 15:38:19 +02002439 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002440 s->cpu_slab->freelist, s->cpu_slab->tid,
2441 c->freelist, tid,
2442 object, next_tid(tid)))) {
2443
2444 note_cmpxchg_failure("slab_free", s, tid);
2445 goto redo;
2446 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002447 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002448 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002449 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002450
Christoph Lameter894b8782007-05-10 03:15:16 -07002451}
2452
Christoph Lameter81819f02007-05-06 14:49:36 -07002453void kmem_cache_free(struct kmem_cache *s, void *x)
2454{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002455 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002456
Christoph Lameterb49af682007-05-06 14:49:41 -07002457 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002458
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002459 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002460
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002461 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002462}
2463EXPORT_SYMBOL(kmem_cache_free);
2464
Christoph Lameter81819f02007-05-06 14:49:36 -07002465/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002466 * Object placement in a slab is made very easy because we always start at
2467 * offset 0. If we tune the size of the object to the alignment then we can
2468 * get the required alignment by putting one properly sized object after
2469 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002470 *
2471 * Notice that the allocation order determines the sizes of the per cpu
2472 * caches. Each processor has always one slab available for allocations.
2473 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002474 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002475 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002476 */
2477
2478/*
2479 * Mininum / Maximum order of slab pages. This influences locking overhead
2480 * and slab fragmentation. A higher order reduces the number of partial slabs
2481 * and increases the number of allocations possible without having to
2482 * take the list_lock.
2483 */
2484static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002485static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002486static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002487
2488/*
2489 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002490 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002491 */
2492static int slub_nomerge;
2493
2494/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002495 * Calculate the order of allocation given an slab object size.
2496 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002497 * The order of allocation has significant impact on performance and other
2498 * system components. Generally order 0 allocations should be preferred since
2499 * order 0 does not cause fragmentation in the page allocator. Larger objects
2500 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002501 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002502 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002503 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002504 * In order to reach satisfactory performance we must ensure that a minimum
2505 * number of objects is in one slab. Otherwise we may generate too much
2506 * activity on the partial lists which requires taking the list_lock. This is
2507 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002508 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002509 * slub_max_order specifies the order where we begin to stop considering the
2510 * number of objects in a slab as critical. If we reach slub_max_order then
2511 * we try to keep the page order as low as possible. So we accept more waste
2512 * of space in favor of a small page order.
2513 *
2514 * Higher order allocations also allow the placement of more objects in a
2515 * slab and thereby reduce object handling overhead. If the user has
2516 * requested a higher mininum order then we start with that one instead of
2517 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002518 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002519static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002520 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002521{
2522 int order;
2523 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002524 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002525
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002526 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002527 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002528
Christoph Lameter6300ea72007-07-17 04:03:20 -07002529 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002530 fls(min_objects * size - 1) - PAGE_SHIFT);
2531 order <= max_order; order++) {
2532
Christoph Lameter81819f02007-05-06 14:49:36 -07002533 unsigned long slab_size = PAGE_SIZE << order;
2534
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002535 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002536 continue;
2537
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002538 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002539
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002540 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002541 break;
2542
2543 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002544
Christoph Lameter81819f02007-05-06 14:49:36 -07002545 return order;
2546}
2547
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002548static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002549{
2550 int order;
2551 int min_objects;
2552 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002553 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002554
2555 /*
2556 * Attempt to find best configuration for a slab. This
2557 * works by first attempting to generate a layout with
2558 * the best configuration and backing off gradually.
2559 *
2560 * First we reduce the acceptable waste in a slab. Then
2561 * we reduce the minimum objects required in a slab.
2562 */
2563 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002564 if (!min_objects)
2565 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002566 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002567 min_objects = min(min_objects, max_objects);
2568
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002569 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002570 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002571 while (fraction >= 4) {
2572 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002573 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002574 if (order <= slub_max_order)
2575 return order;
2576 fraction /= 2;
2577 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002578 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002579 }
2580
2581 /*
2582 * We were unable to place multiple objects in a slab. Now
2583 * lets see if we can place a single object there.
2584 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002585 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002586 if (order <= slub_max_order)
2587 return order;
2588
2589 /*
2590 * Doh this slab cannot be placed using slub_max_order.
2591 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002592 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002593 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002594 return order;
2595 return -ENOSYS;
2596}
2597
Christoph Lameter81819f02007-05-06 14:49:36 -07002598/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002599 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002600 */
2601static unsigned long calculate_alignment(unsigned long flags,
2602 unsigned long align, unsigned long size)
2603{
2604 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002605 * If the user wants hardware cache aligned objects then follow that
2606 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002607 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002608 * The hardware cache alignment cannot override the specified
2609 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002610 */
Nick Pigginb6210382008-03-05 14:05:56 -08002611 if (flags & SLAB_HWCACHE_ALIGN) {
2612 unsigned long ralign = cache_line_size();
2613 while (size <= ralign / 2)
2614 ralign /= 2;
2615 align = max(align, ralign);
2616 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002617
2618 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002619 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002620
2621 return ALIGN(align, sizeof(void *));
2622}
2623
Pekka Enberg5595cff2008-08-05 09:28:47 +03002624static void
2625init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002626{
2627 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002628 spin_lock_init(&n->list_lock);
2629 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002630#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002631 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002632 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002633 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002634#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002635}
2636
Christoph Lameter55136592010-08-20 12:37:13 -05002637static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002638{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002639 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2640 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002641
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002642 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002643 * Must align to double word boundary for the double cmpxchg
2644 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002645 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002646 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2647 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002648
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002649 if (!s->cpu_slab)
2650 return 0;
2651
2652 init_kmem_cache_cpus(s);
2653
2654 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002655}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002656
Christoph Lameter51df1142010-08-20 12:37:15 -05002657static struct kmem_cache *kmem_cache_node;
2658
Christoph Lameter81819f02007-05-06 14:49:36 -07002659/*
2660 * No kmalloc_node yet so do it by hand. We know that this is the first
2661 * slab on the node for this slabcache. There are no concurrent accesses
2662 * possible.
2663 *
2664 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002665 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2666 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002667 */
Christoph Lameter55136592010-08-20 12:37:13 -05002668static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002669{
2670 struct page *page;
2671 struct kmem_cache_node *n;
2672
Christoph Lameter51df1142010-08-20 12:37:15 -05002673 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002674
Christoph Lameter51df1142010-08-20 12:37:15 -05002675 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002676
2677 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002678 if (page_to_nid(page) != node) {
2679 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2680 "node %d\n", node);
2681 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2682 "in order to be able to continue\n");
2683 }
2684
Christoph Lameter81819f02007-05-06 14:49:36 -07002685 n = page->freelist;
2686 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002687 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002688 page->inuse++;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002689 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002690 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002691#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002692 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002693 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002694#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002695 init_kmem_cache_node(n, kmem_cache_node);
2696 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002697
Shaohua Li136333d2011-08-24 08:57:52 +08002698 add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002699}
2700
2701static void free_kmem_cache_nodes(struct kmem_cache *s)
2702{
2703 int node;
2704
Christoph Lameterf64dc582007-10-16 01:25:33 -07002705 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002706 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002707
Alexander Duyck73367bd2010-05-21 14:41:35 -07002708 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002709 kmem_cache_free(kmem_cache_node, n);
2710
Christoph Lameter81819f02007-05-06 14:49:36 -07002711 s->node[node] = NULL;
2712 }
2713}
2714
Christoph Lameter55136592010-08-20 12:37:13 -05002715static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002716{
2717 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002718
Christoph Lameterf64dc582007-10-16 01:25:33 -07002719 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002720 struct kmem_cache_node *n;
2721
Alexander Duyck73367bd2010-05-21 14:41:35 -07002722 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002723 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002724 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002725 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002726 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002727 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002728
2729 if (!n) {
2730 free_kmem_cache_nodes(s);
2731 return 0;
2732 }
2733
Christoph Lameter81819f02007-05-06 14:49:36 -07002734 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002735 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002736 }
2737 return 1;
2738}
Christoph Lameter81819f02007-05-06 14:49:36 -07002739
David Rientjesc0bdb232009-02-25 09:16:35 +02002740static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002741{
2742 if (min < MIN_PARTIAL)
2743 min = MIN_PARTIAL;
2744 else if (min > MAX_PARTIAL)
2745 min = MAX_PARTIAL;
2746 s->min_partial = min;
2747}
2748
Christoph Lameter81819f02007-05-06 14:49:36 -07002749/*
2750 * calculate_sizes() determines the order and the distribution of data within
2751 * a slab object.
2752 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002753static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002754{
2755 unsigned long flags = s->flags;
2756 unsigned long size = s->objsize;
2757 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002758 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002759
2760 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002761 * Round up object size to the next word boundary. We can only
2762 * place the free pointer at word boundaries and this determines
2763 * the possible location of the free pointer.
2764 */
2765 size = ALIGN(size, sizeof(void *));
2766
2767#ifdef CONFIG_SLUB_DEBUG
2768 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002769 * Determine if we can poison the object itself. If the user of
2770 * the slab may touch the object after free or before allocation
2771 * then we should never poison the object itself.
2772 */
2773 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002774 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002775 s->flags |= __OBJECT_POISON;
2776 else
2777 s->flags &= ~__OBJECT_POISON;
2778
Christoph Lameter81819f02007-05-06 14:49:36 -07002779
2780 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002781 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002782 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002783 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002784 */
2785 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2786 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002787#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002788
2789 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002790 * With that we have determined the number of bytes in actual use
2791 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002792 */
2793 s->inuse = size;
2794
2795 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002796 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002797 /*
2798 * Relocate free pointer after the object if it is not
2799 * permitted to overwrite the first word of the object on
2800 * kmem_cache_free.
2801 *
2802 * This is the case if we do RCU, have a constructor or
2803 * destructor or are poisoning the objects.
2804 */
2805 s->offset = size;
2806 size += sizeof(void *);
2807 }
2808
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002809#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002810 if (flags & SLAB_STORE_USER)
2811 /*
2812 * Need to store information about allocs and frees after
2813 * the object.
2814 */
2815 size += 2 * sizeof(struct track);
2816
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002817 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002818 /*
2819 * Add some empty padding so that we can catch
2820 * overwrites from earlier objects rather than let
2821 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002822 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002823 * of the object.
2824 */
2825 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002826#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002827
Christoph Lameter81819f02007-05-06 14:49:36 -07002828 /*
2829 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002830 * user specified and the dynamic determination of cache line size
2831 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002832 */
2833 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002834 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002835
2836 /*
2837 * SLUB stores one object immediately after another beginning from
2838 * offset 0. In order to align the objects we have to simply size
2839 * each object to conform to the alignment.
2840 */
2841 size = ALIGN(size, align);
2842 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002843 if (forced_order >= 0)
2844 order = forced_order;
2845 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002846 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002847
Christoph Lameter834f3d12008-04-14 19:11:31 +03002848 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002849 return 0;
2850
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002851 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002852 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002853 s->allocflags |= __GFP_COMP;
2854
2855 if (s->flags & SLAB_CACHE_DMA)
2856 s->allocflags |= SLUB_DMA;
2857
2858 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2859 s->allocflags |= __GFP_RECLAIMABLE;
2860
Christoph Lameter81819f02007-05-06 14:49:36 -07002861 /*
2862 * Determine the number of objects per slab
2863 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002864 s->oo = oo_make(order, size, s->reserved);
2865 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002866 if (oo_objects(s->oo) > oo_objects(s->max))
2867 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002868
Christoph Lameter834f3d12008-04-14 19:11:31 +03002869 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002870
2871}
2872
Christoph Lameter55136592010-08-20 12:37:13 -05002873static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002874 const char *name, size_t size,
2875 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002876 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002877{
2878 memset(s, 0, kmem_size);
2879 s->name = name;
2880 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002881 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002882 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002883 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002884 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002885
Lai Jiangshanda9a6382011-03-10 15:22:00 +08002886 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2887 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07002888
Christoph Lameter06b285d2008-04-14 19:11:41 +03002889 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002890 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002891 if (disable_higher_order_debug) {
2892 /*
2893 * Disable debugging flags that store metadata if the min slab
2894 * order increased.
2895 */
2896 if (get_order(s->size) > get_order(s->objsize)) {
2897 s->flags &= ~DEBUG_METADATA_FLAGS;
2898 s->offset = 0;
2899 if (!calculate_sizes(s, -1))
2900 goto error;
2901 }
2902 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002903
Christoph Lameterb789ef52011-06-01 12:25:49 -05002904#ifdef CONFIG_CMPXCHG_DOUBLE
2905 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
2906 /* Enable fast mode */
2907 s->flags |= __CMPXCHG_DOUBLE;
2908#endif
2909
David Rientjes3b89d7d2009-02-22 17:40:07 -08002910 /*
2911 * The larger the object size is, the more pages we want on the partial
2912 * list to avoid pounding the page allocator excessively.
2913 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002914 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002915 s->refcount = 1;
2916#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002917 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002918#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002919 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002920 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002921
Christoph Lameter55136592010-08-20 12:37:13 -05002922 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002923 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002924
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002925 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002926error:
2927 if (flags & SLAB_PANIC)
2928 panic("Cannot create slab %s size=%lu realsize=%u "
2929 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002930 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002931 s->offset, flags);
2932 return 0;
2933}
Christoph Lameter81819f02007-05-06 14:49:36 -07002934
2935/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002936 * Determine the size of a slab object
2937 */
2938unsigned int kmem_cache_size(struct kmem_cache *s)
2939{
2940 return s->objsize;
2941}
2942EXPORT_SYMBOL(kmem_cache_size);
2943
Christoph Lameter33b12c32008-04-25 12:22:43 -07002944static void list_slab_objects(struct kmem_cache *s, struct page *page,
2945 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002946{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002947#ifdef CONFIG_SLUB_DEBUG
2948 void *addr = page_address(page);
2949 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002950 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2951 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002952 if (!map)
2953 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002954 slab_err(s, page, "%s", text);
2955 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002956
Christoph Lameter5f80b132011-04-15 14:48:13 -05002957 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002958 for_each_object(p, s, addr, page->objects) {
2959
2960 if (!test_bit(slab_index(p, s, addr), map)) {
2961 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2962 p, p - addr);
2963 print_tracking(s, p);
2964 }
2965 }
2966 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002967 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002968#endif
2969}
2970
Christoph Lameter81819f02007-05-06 14:49:36 -07002971/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002972 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002973 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002974static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002975{
Christoph Lameter81819f02007-05-06 14:49:36 -07002976 unsigned long flags;
2977 struct page *page, *h;
2978
2979 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002980 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002981 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002982 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002983 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002984 } else {
2985 list_slab_objects(s, page,
2986 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002987 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002988 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002989 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002990}
2991
2992/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002993 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002994 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002995static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002996{
2997 int node;
2998
2999 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003000 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003001 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07003002 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003003 struct kmem_cache_node *n = get_node(s, node);
3004
Christoph Lameter599870b2008-04-23 12:36:52 -07003005 free_partial(s, n);
3006 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003007 return 1;
3008 }
3009 free_kmem_cache_nodes(s);
3010 return 0;
3011}
3012
3013/*
3014 * Close a cache and release the kmem_cache structure
3015 * (must be used for caches created using kmem_cache_create)
3016 */
3017void kmem_cache_destroy(struct kmem_cache *s)
3018{
3019 down_write(&slub_lock);
3020 s->refcount--;
3021 if (!s->refcount) {
3022 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03003023 if (kmem_cache_close(s)) {
3024 printk(KERN_ERR "SLUB %s: %s called for cache that "
3025 "still has objects.\n", s->name, __func__);
3026 dump_stack();
3027 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03003028 if (s->flags & SLAB_DESTROY_BY_RCU)
3029 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07003030 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05003031 }
3032 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003033}
3034EXPORT_SYMBOL(kmem_cache_destroy);
3035
3036/********************************************************************
3037 * Kmalloc subsystem
3038 *******************************************************************/
3039
Christoph Lameter51df1142010-08-20 12:37:15 -05003040struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07003041EXPORT_SYMBOL(kmalloc_caches);
3042
Christoph Lameter51df1142010-08-20 12:37:15 -05003043static struct kmem_cache *kmem_cache;
3044
Christoph Lameter55136592010-08-20 12:37:13 -05003045#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003046static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05003047#endif
3048
Christoph Lameter81819f02007-05-06 14:49:36 -07003049static int __init setup_slub_min_order(char *str)
3050{
Pekka Enberg06428782008-01-07 23:20:27 -08003051 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003052
3053 return 1;
3054}
3055
3056__setup("slub_min_order=", setup_slub_min_order);
3057
3058static int __init setup_slub_max_order(char *str)
3059{
Pekka Enberg06428782008-01-07 23:20:27 -08003060 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003061 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003062
3063 return 1;
3064}
3065
3066__setup("slub_max_order=", setup_slub_max_order);
3067
3068static int __init setup_slub_min_objects(char *str)
3069{
Pekka Enberg06428782008-01-07 23:20:27 -08003070 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003071
3072 return 1;
3073}
3074
3075__setup("slub_min_objects=", setup_slub_min_objects);
3076
3077static int __init setup_slub_nomerge(char *str)
3078{
3079 slub_nomerge = 1;
3080 return 1;
3081}
3082
3083__setup("slub_nomerge", setup_slub_nomerge);
3084
Christoph Lameter51df1142010-08-20 12:37:15 -05003085static struct kmem_cache *__init create_kmalloc_cache(const char *name,
3086 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003087{
Christoph Lameter51df1142010-08-20 12:37:15 -05003088 struct kmem_cache *s;
3089
3090 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3091
Pekka Enberg83b519e2009-06-10 19:40:04 +03003092 /*
3093 * This function is called with IRQs disabled during early-boot on
3094 * single CPU so there's no need to take slub_lock here.
3095 */
Christoph Lameter55136592010-08-20 12:37:13 -05003096 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03003097 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07003098 goto panic;
3099
3100 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05003101 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003102
3103panic:
3104 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003105 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003106}
3107
Christoph Lameterf1b26332007-07-17 04:03:26 -07003108/*
3109 * Conversion table for small slabs sizes / 8 to the index in the
3110 * kmalloc array. This is necessary for slabs < 192 since we have non power
3111 * of two cache sizes there. The size of larger slabs can be determined using
3112 * fls.
3113 */
3114static s8 size_index[24] = {
3115 3, /* 8 */
3116 4, /* 16 */
3117 5, /* 24 */
3118 5, /* 32 */
3119 6, /* 40 */
3120 6, /* 48 */
3121 6, /* 56 */
3122 6, /* 64 */
3123 1, /* 72 */
3124 1, /* 80 */
3125 1, /* 88 */
3126 1, /* 96 */
3127 7, /* 104 */
3128 7, /* 112 */
3129 7, /* 120 */
3130 7, /* 128 */
3131 2, /* 136 */
3132 2, /* 144 */
3133 2, /* 152 */
3134 2, /* 160 */
3135 2, /* 168 */
3136 2, /* 176 */
3137 2, /* 184 */
3138 2 /* 192 */
3139};
3140
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003141static inline int size_index_elem(size_t bytes)
3142{
3143 return (bytes - 1) / 8;
3144}
3145
Christoph Lameter81819f02007-05-06 14:49:36 -07003146static struct kmem_cache *get_slab(size_t size, gfp_t flags)
3147{
Christoph Lameterf1b26332007-07-17 04:03:26 -07003148 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07003149
Christoph Lameterf1b26332007-07-17 04:03:26 -07003150 if (size <= 192) {
3151 if (!size)
3152 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07003153
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003154 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003155 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07003156 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003157
3158#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07003159 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05003160 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07003161
Christoph Lameter81819f02007-05-06 14:49:36 -07003162#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003163 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07003164}
3165
3166void *__kmalloc(size_t size, gfp_t flags)
3167{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003168 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003169 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003170
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003171 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003172 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003173
3174 s = get_slab(size, flags);
3175
3176 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003177 return s;
3178
Christoph Lameter2154a332010-07-09 14:07:10 -05003179 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003180
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003181 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003182
3183 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003184}
3185EXPORT_SYMBOL(__kmalloc);
3186
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003187#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003188static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3189{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003190 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003191 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003192
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003193 flags |= __GFP_COMP | __GFP_NOTRACK;
3194 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003195 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003196 ptr = page_address(page);
3197
3198 kmemleak_alloc(ptr, size, 1, flags);
3199 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003200}
3201
Christoph Lameter81819f02007-05-06 14:49:36 -07003202void *__kmalloc_node(size_t size, gfp_t flags, int node)
3203{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003204 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003205 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003206
Ingo Molnar057685c2009-02-20 12:15:30 +01003207 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003208 ret = kmalloc_large_node(size, flags, node);
3209
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003210 trace_kmalloc_node(_RET_IP_, ret,
3211 size, PAGE_SIZE << get_order(size),
3212 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003213
3214 return ret;
3215 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003216
3217 s = get_slab(size, flags);
3218
3219 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003220 return s;
3221
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003222 ret = slab_alloc(s, flags, node, _RET_IP_);
3223
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003224 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003225
3226 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003227}
3228EXPORT_SYMBOL(__kmalloc_node);
3229#endif
3230
3231size_t ksize(const void *object)
3232{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003233 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003234
Christoph Lameteref8b4522007-10-16 01:24:46 -07003235 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003236 return 0;
3237
Vegard Nossum294a80a2007-12-04 23:45:30 -08003238 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003239
Pekka Enberg76994412008-05-22 19:22:25 +03003240 if (unlikely(!PageSlab(page))) {
3241 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003242 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003243 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003244
Eric Dumazetb3d41882011-02-14 18:35:22 +01003245 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003246}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003247EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003248
Ben Greeard18a90d2011-07-07 11:36:37 -07003249#ifdef CONFIG_SLUB_DEBUG
3250bool verify_mem_not_deleted(const void *x)
3251{
3252 struct page *page;
3253 void *object = (void *)x;
3254 unsigned long flags;
3255 bool rv;
3256
3257 if (unlikely(ZERO_OR_NULL_PTR(x)))
3258 return false;
3259
3260 local_irq_save(flags);
3261
3262 page = virt_to_head_page(x);
3263 if (unlikely(!PageSlab(page))) {
3264 /* maybe it was from stack? */
3265 rv = true;
3266 goto out_unlock;
3267 }
3268
3269 slab_lock(page);
3270 if (on_freelist(page->slab, page, object)) {
3271 object_err(page->slab, page, object, "Object is on free-list");
3272 rv = false;
3273 } else {
3274 rv = true;
3275 }
3276 slab_unlock(page);
3277
3278out_unlock:
3279 local_irq_restore(flags);
3280 return rv;
3281}
3282EXPORT_SYMBOL(verify_mem_not_deleted);
3283#endif
3284
Christoph Lameter81819f02007-05-06 14:49:36 -07003285void kfree(const void *x)
3286{
Christoph Lameter81819f02007-05-06 14:49:36 -07003287 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003288 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003289
Pekka Enberg2121db72009-03-25 11:05:57 +02003290 trace_kfree(_RET_IP_, x);
3291
Satyam Sharma2408c552007-10-16 01:24:44 -07003292 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003293 return;
3294
Christoph Lameterb49af682007-05-06 14:49:41 -07003295 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003296 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003297 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003298 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003299 put_page(page);
3300 return;
3301 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003302 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003303}
3304EXPORT_SYMBOL(kfree);
3305
Christoph Lameter2086d262007-05-06 14:49:46 -07003306/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003307 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3308 * the remaining slabs by the number of items in use. The slabs with the
3309 * most items in use come first. New allocations will then fill those up
3310 * and thus they can be removed from the partial lists.
3311 *
3312 * The slabs with the least items are placed last. This results in them
3313 * being allocated from last increasing the chance that the last objects
3314 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003315 */
3316int kmem_cache_shrink(struct kmem_cache *s)
3317{
3318 int node;
3319 int i;
3320 struct kmem_cache_node *n;
3321 struct page *page;
3322 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003323 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003324 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003325 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003326 unsigned long flags;
3327
3328 if (!slabs_by_inuse)
3329 return -ENOMEM;
3330
3331 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003332 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003333 n = get_node(s, node);
3334
3335 if (!n->nr_partial)
3336 continue;
3337
Christoph Lameter834f3d12008-04-14 19:11:31 +03003338 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003339 INIT_LIST_HEAD(slabs_by_inuse + i);
3340
3341 spin_lock_irqsave(&n->list_lock, flags);
3342
3343 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003344 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003345 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003346 * Note that concurrent frees may occur while we hold the
3347 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003348 */
3349 list_for_each_entry_safe(page, t, &n->partial, lru) {
Christoph Lameter881db7f2011-06-01 12:25:53 -05003350 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003351 remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003352 discard_slab(s, page);
3353 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07003354 list_move(&page->lru,
3355 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07003356 }
3357 }
3358
Christoph Lameter2086d262007-05-06 14:49:46 -07003359 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003360 * Rebuild the partial list with the slabs filled up most
3361 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003362 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003363 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003364 list_splice(slabs_by_inuse + i, n->partial.prev);
3365
Christoph Lameter2086d262007-05-06 14:49:46 -07003366 spin_unlock_irqrestore(&n->list_lock, flags);
3367 }
3368
3369 kfree(slabs_by_inuse);
3370 return 0;
3371}
3372EXPORT_SYMBOL(kmem_cache_shrink);
3373
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003374#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003375static int slab_mem_going_offline_callback(void *arg)
3376{
3377 struct kmem_cache *s;
3378
3379 down_read(&slub_lock);
3380 list_for_each_entry(s, &slab_caches, list)
3381 kmem_cache_shrink(s);
3382 up_read(&slub_lock);
3383
3384 return 0;
3385}
3386
3387static void slab_mem_offline_callback(void *arg)
3388{
3389 struct kmem_cache_node *n;
3390 struct kmem_cache *s;
3391 struct memory_notify *marg = arg;
3392 int offline_node;
3393
3394 offline_node = marg->status_change_nid;
3395
3396 /*
3397 * If the node still has available memory. we need kmem_cache_node
3398 * for it yet.
3399 */
3400 if (offline_node < 0)
3401 return;
3402
3403 down_read(&slub_lock);
3404 list_for_each_entry(s, &slab_caches, list) {
3405 n = get_node(s, offline_node);
3406 if (n) {
3407 /*
3408 * if n->nr_slabs > 0, slabs still exist on the node
3409 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003410 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003411 * callback. So, we must fail.
3412 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003413 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003414
3415 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003416 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003417 }
3418 }
3419 up_read(&slub_lock);
3420}
3421
3422static int slab_mem_going_online_callback(void *arg)
3423{
3424 struct kmem_cache_node *n;
3425 struct kmem_cache *s;
3426 struct memory_notify *marg = arg;
3427 int nid = marg->status_change_nid;
3428 int ret = 0;
3429
3430 /*
3431 * If the node's memory is already available, then kmem_cache_node is
3432 * already created. Nothing to do.
3433 */
3434 if (nid < 0)
3435 return 0;
3436
3437 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003438 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003439 * allocate a kmem_cache_node structure in order to bring the node
3440 * online.
3441 */
3442 down_read(&slub_lock);
3443 list_for_each_entry(s, &slab_caches, list) {
3444 /*
3445 * XXX: kmem_cache_alloc_node will fallback to other nodes
3446 * since memory is not yet available from the node that
3447 * is brought up.
3448 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003449 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003450 if (!n) {
3451 ret = -ENOMEM;
3452 goto out;
3453 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003454 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003455 s->node[nid] = n;
3456 }
3457out:
3458 up_read(&slub_lock);
3459 return ret;
3460}
3461
3462static int slab_memory_callback(struct notifier_block *self,
3463 unsigned long action, void *arg)
3464{
3465 int ret = 0;
3466
3467 switch (action) {
3468 case MEM_GOING_ONLINE:
3469 ret = slab_mem_going_online_callback(arg);
3470 break;
3471 case MEM_GOING_OFFLINE:
3472 ret = slab_mem_going_offline_callback(arg);
3473 break;
3474 case MEM_OFFLINE:
3475 case MEM_CANCEL_ONLINE:
3476 slab_mem_offline_callback(arg);
3477 break;
3478 case MEM_ONLINE:
3479 case MEM_CANCEL_OFFLINE:
3480 break;
3481 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003482 if (ret)
3483 ret = notifier_from_errno(ret);
3484 else
3485 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003486 return ret;
3487}
3488
3489#endif /* CONFIG_MEMORY_HOTPLUG */
3490
Christoph Lameter81819f02007-05-06 14:49:36 -07003491/********************************************************************
3492 * Basic setup of slabs
3493 *******************************************************************/
3494
Christoph Lameter51df1142010-08-20 12:37:15 -05003495/*
3496 * Used for early kmem_cache structures that were allocated using
3497 * the page allocator
3498 */
3499
3500static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3501{
3502 int node;
3503
3504 list_add(&s->list, &slab_caches);
3505 s->refcount = -1;
3506
3507 for_each_node_state(node, N_NORMAL_MEMORY) {
3508 struct kmem_cache_node *n = get_node(s, node);
3509 struct page *p;
3510
3511 if (n) {
3512 list_for_each_entry(p, &n->partial, lru)
3513 p->slab = s;
3514
Li Zefan607bf322011-04-12 15:22:26 +08003515#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003516 list_for_each_entry(p, &n->full, lru)
3517 p->slab = s;
3518#endif
3519 }
3520 }
3521}
3522
Christoph Lameter81819f02007-05-06 14:49:36 -07003523void __init kmem_cache_init(void)
3524{
3525 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003526 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003527 struct kmem_cache *temp_kmem_cache;
3528 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003529 struct kmem_cache *temp_kmem_cache_node;
3530 unsigned long kmalloc_size;
3531
3532 kmem_size = offsetof(struct kmem_cache, node) +
3533 nr_node_ids * sizeof(struct kmem_cache_node *);
3534
3535 /* Allocate two kmem_caches from the page allocator */
3536 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3537 order = get_order(2 * kmalloc_size);
3538 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3539
Christoph Lameter81819f02007-05-06 14:49:36 -07003540 /*
3541 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003542 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003543 * kmem_cache_open for slab_state == DOWN.
3544 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003545 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3546
3547 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3548 sizeof(struct kmem_cache_node),
3549 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003550
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003551 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003552
3553 /* Able to allocate the per node structures */
3554 slab_state = PARTIAL;
3555
Christoph Lameter51df1142010-08-20 12:37:15 -05003556 temp_kmem_cache = kmem_cache;
3557 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3558 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3559 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3560 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003561
Christoph Lameter51df1142010-08-20 12:37:15 -05003562 /*
3563 * Allocate kmem_cache_node properly from the kmem_cache slab.
3564 * kmem_cache_node is separately allocated so no need to
3565 * update any list pointers.
3566 */
3567 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003568
Christoph Lameter51df1142010-08-20 12:37:15 -05003569 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3570 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3571
3572 kmem_cache_bootstrap_fixup(kmem_cache_node);
3573
3574 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003575 kmem_cache_bootstrap_fixup(kmem_cache);
3576 caches++;
3577 /* Free temporary boot structure */
3578 free_pages((unsigned long)temp_kmem_cache, order);
3579
3580 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003581
3582 /*
3583 * Patch up the size_index table if we have strange large alignment
3584 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003585 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003586 *
3587 * Largest permitted alignment is 256 bytes due to the way we
3588 * handle the index determination for the smaller caches.
3589 *
3590 * Make sure that nothing crazy happens if someone starts tinkering
3591 * around with ARCH_KMALLOC_MINALIGN
3592 */
3593 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3594 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3595
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003596 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3597 int elem = size_index_elem(i);
3598 if (elem >= ARRAY_SIZE(size_index))
3599 break;
3600 size_index[elem] = KMALLOC_SHIFT_LOW;
3601 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003602
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003603 if (KMALLOC_MIN_SIZE == 64) {
3604 /*
3605 * The 96 byte size cache is not used if the alignment
3606 * is 64 byte.
3607 */
3608 for (i = 64 + 8; i <= 96; i += 8)
3609 size_index[size_index_elem(i)] = 7;
3610 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003611 /*
3612 * The 192 byte sized cache is not used if the alignment
3613 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3614 * instead.
3615 */
3616 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003617 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003618 }
3619
Christoph Lameter51df1142010-08-20 12:37:15 -05003620 /* Caches that are not of the two-to-the-power-of size */
3621 if (KMALLOC_MIN_SIZE <= 32) {
3622 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3623 caches++;
3624 }
3625
3626 if (KMALLOC_MIN_SIZE <= 64) {
3627 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3628 caches++;
3629 }
3630
3631 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3632 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3633 caches++;
3634 }
3635
Christoph Lameter81819f02007-05-06 14:49:36 -07003636 slab_state = UP;
3637
3638 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003639 if (KMALLOC_MIN_SIZE <= 32) {
3640 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3641 BUG_ON(!kmalloc_caches[1]->name);
3642 }
3643
3644 if (KMALLOC_MIN_SIZE <= 64) {
3645 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3646 BUG_ON(!kmalloc_caches[2]->name);
3647 }
3648
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003649 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3650 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3651
3652 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003653 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003654 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003655
3656#ifdef CONFIG_SMP
3657 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003658#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003659
Christoph Lameter55136592010-08-20 12:37:13 -05003660#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003661 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3662 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003663
Christoph Lameter51df1142010-08-20 12:37:15 -05003664 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003665 char *name = kasprintf(GFP_NOWAIT,
3666 "dma-kmalloc-%d", s->objsize);
3667
3668 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003669 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3670 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003671 }
3672 }
3673#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003674 printk(KERN_INFO
3675 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003676 " CPUs=%d, Nodes=%d\n",
3677 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003678 slub_min_order, slub_max_order, slub_min_objects,
3679 nr_cpu_ids, nr_node_ids);
3680}
3681
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003682void __init kmem_cache_init_late(void)
3683{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003684}
3685
Christoph Lameter81819f02007-05-06 14:49:36 -07003686/*
3687 * Find a mergeable slab cache
3688 */
3689static int slab_unmergeable(struct kmem_cache *s)
3690{
3691 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3692 return 1;
3693
Christoph Lameterc59def92007-05-16 22:10:50 -07003694 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003695 return 1;
3696
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003697 /*
3698 * We may have set a slab to be unmergeable during bootstrap.
3699 */
3700 if (s->refcount < 0)
3701 return 1;
3702
Christoph Lameter81819f02007-05-06 14:49:36 -07003703 return 0;
3704}
3705
3706static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003707 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003708 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003709{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003710 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003711
3712 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3713 return NULL;
3714
Christoph Lameterc59def92007-05-16 22:10:50 -07003715 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003716 return NULL;
3717
3718 size = ALIGN(size, sizeof(void *));
3719 align = calculate_alignment(flags, align, size);
3720 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003721 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003722
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003723 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003724 if (slab_unmergeable(s))
3725 continue;
3726
3727 if (size > s->size)
3728 continue;
3729
Christoph Lameterba0268a2007-09-11 15:24:11 -07003730 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003731 continue;
3732 /*
3733 * Check if alignment is compatible.
3734 * Courtesy of Adrian Drzewiecki
3735 */
Pekka Enberg06428782008-01-07 23:20:27 -08003736 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003737 continue;
3738
3739 if (s->size - size >= sizeof(void *))
3740 continue;
3741
3742 return s;
3743 }
3744 return NULL;
3745}
3746
3747struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003748 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003749{
3750 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003751 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003752
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003753 if (WARN_ON(!name))
3754 return NULL;
3755
Christoph Lameter81819f02007-05-06 14:49:36 -07003756 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003757 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003758 if (s) {
3759 s->refcount++;
3760 /*
3761 * Adjust the object sizes so that we clear
3762 * the complete object on kzalloc.
3763 */
3764 s->objsize = max(s->objsize, (int)size);
3765 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003766
David Rientjes7b8f3b62008-12-17 22:09:46 -08003767 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003768 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003769 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003770 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003771 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003772 return s;
3773 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003774
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003775 n = kstrdup(name, GFP_KERNEL);
3776 if (!n)
3777 goto err;
3778
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003779 s = kmalloc(kmem_size, GFP_KERNEL);
3780 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003781 if (kmem_cache_open(s, n,
Christoph Lameterc59def92007-05-16 22:10:50 -07003782 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003783 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003784 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003785 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003786 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003787 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003788 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003789 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003790 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003791 return s;
3792 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003793 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003794 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003795 }
Pavel Emelyanov68cee4f12010-10-28 13:50:37 +04003796err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003797 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003798
Christoph Lameter81819f02007-05-06 14:49:36 -07003799 if (flags & SLAB_PANIC)
3800 panic("Cannot create slabcache %s\n", name);
3801 else
3802 s = NULL;
3803 return s;
3804}
3805EXPORT_SYMBOL(kmem_cache_create);
3806
Christoph Lameter81819f02007-05-06 14:49:36 -07003807#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003808/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003809 * Use the cpu notifier to insure that the cpu slabs are flushed when
3810 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003811 */
3812static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3813 unsigned long action, void *hcpu)
3814{
3815 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003816 struct kmem_cache *s;
3817 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003818
3819 switch (action) {
3820 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003821 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003822 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003823 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003824 down_read(&slub_lock);
3825 list_for_each_entry(s, &slab_caches, list) {
3826 local_irq_save(flags);
3827 __flush_cpu_slab(s, cpu);
3828 local_irq_restore(flags);
3829 }
3830 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003831 break;
3832 default:
3833 break;
3834 }
3835 return NOTIFY_OK;
3836}
3837
Pekka Enberg06428782008-01-07 23:20:27 -08003838static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003839 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003840};
Christoph Lameter81819f02007-05-06 14:49:36 -07003841
3842#endif
3843
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003844void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003845{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003846 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003847 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003848
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003849 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003850 return kmalloc_large(size, gfpflags);
3851
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003852 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003853
Satyam Sharma2408c552007-10-16 01:24:44 -07003854 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003855 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003856
Christoph Lameter2154a332010-07-09 14:07:10 -05003857 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003858
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003859 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003860 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003861
3862 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003863}
3864
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003865#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003866void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003867 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003868{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003869 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003870 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003871
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003872 if (unlikely(size > SLUB_MAX_SIZE)) {
3873 ret = kmalloc_large_node(size, gfpflags, node);
3874
3875 trace_kmalloc_node(caller, ret,
3876 size, PAGE_SIZE << get_order(size),
3877 gfpflags, node);
3878
3879 return ret;
3880 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003881
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003882 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003883
Satyam Sharma2408c552007-10-16 01:24:44 -07003884 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003885 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003886
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003887 ret = slab_alloc(s, gfpflags, node, caller);
3888
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003889 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003890 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003891
3892 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003893}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003894#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003895
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003896#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003897static int count_inuse(struct page *page)
3898{
3899 return page->inuse;
3900}
3901
3902static int count_total(struct page *page)
3903{
3904 return page->objects;
3905}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003906#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003907
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003908#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003909static int validate_slab(struct kmem_cache *s, struct page *page,
3910 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003911{
3912 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003913 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003914
3915 if (!check_slab(s, page) ||
3916 !on_freelist(s, page, NULL))
3917 return 0;
3918
3919 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003920 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003921
Christoph Lameter5f80b132011-04-15 14:48:13 -05003922 get_map(s, page, map);
3923 for_each_object(p, s, addr, page->objects) {
3924 if (test_bit(slab_index(p, s, addr), map))
3925 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3926 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003927 }
3928
Christoph Lameter224a88b2008-04-14 19:11:31 +03003929 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003930 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003931 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003932 return 0;
3933 return 1;
3934}
3935
Christoph Lameter434e2452007-07-17 04:03:30 -07003936static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3937 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003938{
Christoph Lameter881db7f2011-06-01 12:25:53 -05003939 slab_lock(page);
3940 validate_slab(s, page, map);
3941 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003942}
3943
Christoph Lameter434e2452007-07-17 04:03:30 -07003944static int validate_slab_node(struct kmem_cache *s,
3945 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003946{
3947 unsigned long count = 0;
3948 struct page *page;
3949 unsigned long flags;
3950
3951 spin_lock_irqsave(&n->list_lock, flags);
3952
3953 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003954 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003955 count++;
3956 }
3957 if (count != n->nr_partial)
3958 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3959 "counter=%ld\n", s->name, count, n->nr_partial);
3960
3961 if (!(s->flags & SLAB_STORE_USER))
3962 goto out;
3963
3964 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003965 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003966 count++;
3967 }
3968 if (count != atomic_long_read(&n->nr_slabs))
3969 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3970 "counter=%ld\n", s->name, count,
3971 atomic_long_read(&n->nr_slabs));
3972
3973out:
3974 spin_unlock_irqrestore(&n->list_lock, flags);
3975 return count;
3976}
3977
Christoph Lameter434e2452007-07-17 04:03:30 -07003978static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003979{
3980 int node;
3981 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003982 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003983 sizeof(unsigned long), GFP_KERNEL);
3984
3985 if (!map)
3986 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003987
3988 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003989 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003990 struct kmem_cache_node *n = get_node(s, node);
3991
Christoph Lameter434e2452007-07-17 04:03:30 -07003992 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003993 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003994 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003995 return count;
3996}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003997/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003998 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003999 * and freed.
4000 */
4001
4002struct location {
4003 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004004 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004005 long long sum_time;
4006 long min_time;
4007 long max_time;
4008 long min_pid;
4009 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304010 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004011 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004012};
4013
4014struct loc_track {
4015 unsigned long max;
4016 unsigned long count;
4017 struct location *loc;
4018};
4019
4020static void free_loc_track(struct loc_track *t)
4021{
4022 if (t->max)
4023 free_pages((unsigned long)t->loc,
4024 get_order(sizeof(struct location) * t->max));
4025}
4026
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004027static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004028{
4029 struct location *l;
4030 int order;
4031
Christoph Lameter88a420e2007-05-06 14:49:45 -07004032 order = get_order(sizeof(struct location) * max);
4033
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004034 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004035 if (!l)
4036 return 0;
4037
4038 if (t->count) {
4039 memcpy(l, t->loc, sizeof(struct location) * t->count);
4040 free_loc_track(t);
4041 }
4042 t->max = max;
4043 t->loc = l;
4044 return 1;
4045}
4046
4047static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004048 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004049{
4050 long start, end, pos;
4051 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004052 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004053 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004054
4055 start = -1;
4056 end = t->count;
4057
4058 for ( ; ; ) {
4059 pos = start + (end - start + 1) / 2;
4060
4061 /*
4062 * There is nothing at "end". If we end up there
4063 * we need to add something to before end.
4064 */
4065 if (pos == end)
4066 break;
4067
4068 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004069 if (track->addr == caddr) {
4070
4071 l = &t->loc[pos];
4072 l->count++;
4073 if (track->when) {
4074 l->sum_time += age;
4075 if (age < l->min_time)
4076 l->min_time = age;
4077 if (age > l->max_time)
4078 l->max_time = age;
4079
4080 if (track->pid < l->min_pid)
4081 l->min_pid = track->pid;
4082 if (track->pid > l->max_pid)
4083 l->max_pid = track->pid;
4084
Rusty Russell174596a2009-01-01 10:12:29 +10304085 cpumask_set_cpu(track->cpu,
4086 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004087 }
4088 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004089 return 1;
4090 }
4091
Christoph Lameter45edfa52007-05-09 02:32:45 -07004092 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004093 end = pos;
4094 else
4095 start = pos;
4096 }
4097
4098 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004099 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004100 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004101 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004102 return 0;
4103
4104 l = t->loc + pos;
4105 if (pos < t->count)
4106 memmove(l + 1, l,
4107 (t->count - pos) * sizeof(struct location));
4108 t->count++;
4109 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004110 l->addr = track->addr;
4111 l->sum_time = age;
4112 l->min_time = age;
4113 l->max_time = age;
4114 l->min_pid = track->pid;
4115 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304116 cpumask_clear(to_cpumask(l->cpus));
4117 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004118 nodes_clear(l->nodes);
4119 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004120 return 1;
4121}
4122
4123static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004124 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004125 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004126{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004127 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004128 void *p;
4129
Christoph Lameter39b26462008-04-14 19:11:30 +03004130 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004131 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004132
Christoph Lameter224a88b2008-04-14 19:11:31 +03004133 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004134 if (!test_bit(slab_index(p, s, addr), map))
4135 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004136}
4137
4138static int list_locations(struct kmem_cache *s, char *buf,
4139 enum track_item alloc)
4140{
Harvey Harrisone374d482008-01-31 15:20:50 -08004141 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004142 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004143 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004144 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004145 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4146 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004147
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004148 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4149 GFP_TEMPORARY)) {
4150 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004151 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004152 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004153 /* Push back cpu slabs */
4154 flush_all(s);
4155
Christoph Lameterf64dc582007-10-16 01:25:33 -07004156 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004157 struct kmem_cache_node *n = get_node(s, node);
4158 unsigned long flags;
4159 struct page *page;
4160
Christoph Lameter9e869432007-08-22 14:01:56 -07004161 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004162 continue;
4163
4164 spin_lock_irqsave(&n->list_lock, flags);
4165 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004166 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004167 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004168 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004169 spin_unlock_irqrestore(&n->list_lock, flags);
4170 }
4171
4172 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004173 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004174
Hugh Dickins9c246242008-12-09 13:14:27 -08004175 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004176 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004177 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004178
4179 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004180 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004181 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004182 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004183
4184 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004185 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004186 l->min_time,
4187 (long)div_u64(l->sum_time, l->count),
4188 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004189 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004190 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004191 l->min_time);
4192
4193 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004194 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004195 l->min_pid, l->max_pid);
4196 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004197 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004198 l->min_pid);
4199
Rusty Russell174596a2009-01-01 10:12:29 +10304200 if (num_online_cpus() > 1 &&
4201 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004202 len < PAGE_SIZE - 60) {
4203 len += sprintf(buf + len, " cpus=");
4204 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10304205 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004206 }
4207
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004208 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004209 len < PAGE_SIZE - 60) {
4210 len += sprintf(buf + len, " nodes=");
4211 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004212 l->nodes);
4213 }
4214
Harvey Harrisone374d482008-01-31 15:20:50 -08004215 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004216 }
4217
4218 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004219 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004220 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004221 len += sprintf(buf, "No data\n");
4222 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004223}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004224#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004225
Christoph Lametera5a84752010-10-05 13:57:27 -05004226#ifdef SLUB_RESILIENCY_TEST
4227static void resiliency_test(void)
4228{
4229 u8 *p;
4230
4231 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
4232
4233 printk(KERN_ERR "SLUB resiliency testing\n");
4234 printk(KERN_ERR "-----------------------\n");
4235 printk(KERN_ERR "A. Corruption after allocation\n");
4236
4237 p = kzalloc(16, GFP_KERNEL);
4238 p[16] = 0x12;
4239 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
4240 " 0x12->0x%p\n\n", p + 16);
4241
4242 validate_slab_cache(kmalloc_caches[4]);
4243
4244 /* Hmmm... The next two are dangerous */
4245 p = kzalloc(32, GFP_KERNEL);
4246 p[32 + sizeof(void *)] = 0x34;
4247 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
4248 " 0x34 -> -0x%p\n", p);
4249 printk(KERN_ERR
4250 "If allocated object is overwritten then not detectable\n\n");
4251
4252 validate_slab_cache(kmalloc_caches[5]);
4253 p = kzalloc(64, GFP_KERNEL);
4254 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4255 *p = 0x56;
4256 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4257 p);
4258 printk(KERN_ERR
4259 "If allocated object is overwritten then not detectable\n\n");
4260 validate_slab_cache(kmalloc_caches[6]);
4261
4262 printk(KERN_ERR "\nB. Corruption after free\n");
4263 p = kzalloc(128, GFP_KERNEL);
4264 kfree(p);
4265 *p = 0x78;
4266 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4267 validate_slab_cache(kmalloc_caches[7]);
4268
4269 p = kzalloc(256, GFP_KERNEL);
4270 kfree(p);
4271 p[50] = 0x9a;
4272 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4273 p);
4274 validate_slab_cache(kmalloc_caches[8]);
4275
4276 p = kzalloc(512, GFP_KERNEL);
4277 kfree(p);
4278 p[512] = 0xab;
4279 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4280 validate_slab_cache(kmalloc_caches[9]);
4281}
4282#else
4283#ifdef CONFIG_SYSFS
4284static void resiliency_test(void) {};
4285#endif
4286#endif
4287
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004288#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004289enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004290 SL_ALL, /* All slabs */
4291 SL_PARTIAL, /* Only partially allocated slabs */
4292 SL_CPU, /* Only slabs used for cpu caches */
4293 SL_OBJECTS, /* Determine allocated objects not slabs */
4294 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004295};
4296
Christoph Lameter205ab992008-04-14 19:11:40 +03004297#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004298#define SO_PARTIAL (1 << SL_PARTIAL)
4299#define SO_CPU (1 << SL_CPU)
4300#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004301#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004302
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004303static ssize_t show_slab_objects(struct kmem_cache *s,
4304 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004305{
4306 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004307 int node;
4308 int x;
4309 unsigned long *nodes;
4310 unsigned long *per_cpu;
4311
4312 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004313 if (!nodes)
4314 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004315 per_cpu = nodes + nr_node_ids;
4316
Christoph Lameter205ab992008-04-14 19:11:40 +03004317 if (flags & SO_CPU) {
4318 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004319
Christoph Lameter205ab992008-04-14 19:11:40 +03004320 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004321 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004322
Christoph Lameter205ab992008-04-14 19:11:40 +03004323 if (!c || c->node < 0)
4324 continue;
4325
4326 if (c->page) {
4327 if (flags & SO_TOTAL)
4328 x = c->page->objects;
4329 else if (flags & SO_OBJECTS)
4330 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07004331 else
4332 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03004333
Christoph Lameter81819f02007-05-06 14:49:36 -07004334 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03004335 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004336 }
Christoph Lameter205ab992008-04-14 19:11:40 +03004337 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004338 }
4339 }
4340
Christoph Lameter04d94872011-01-10 10:15:15 -06004341 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004342#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004343 if (flags & SO_ALL) {
4344 for_each_node_state(node, N_NORMAL_MEMORY) {
4345 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004346
Christoph Lameter205ab992008-04-14 19:11:40 +03004347 if (flags & SO_TOTAL)
4348 x = atomic_long_read(&n->total_objects);
4349 else if (flags & SO_OBJECTS)
4350 x = atomic_long_read(&n->total_objects) -
4351 count_partial(n, count_free);
4352
4353 else
4354 x = atomic_long_read(&n->nr_slabs);
4355 total += x;
4356 nodes[node] += x;
4357 }
4358
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004359 } else
4360#endif
4361 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004362 for_each_node_state(node, N_NORMAL_MEMORY) {
4363 struct kmem_cache_node *n = get_node(s, node);
4364
4365 if (flags & SO_TOTAL)
4366 x = count_partial(n, count_total);
4367 else if (flags & SO_OBJECTS)
4368 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004369 else
4370 x = n->nr_partial;
4371 total += x;
4372 nodes[node] += x;
4373 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004374 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004375 x = sprintf(buf, "%lu", total);
4376#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004377 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004378 if (nodes[node])
4379 x += sprintf(buf + x, " N%d=%lu",
4380 node, nodes[node]);
4381#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004382 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004383 kfree(nodes);
4384 return x + sprintf(buf + x, "\n");
4385}
4386
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004387#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004388static int any_slab_objects(struct kmem_cache *s)
4389{
4390 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004391
4392 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004393 struct kmem_cache_node *n = get_node(s, node);
4394
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004395 if (!n)
4396 continue;
4397
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004398 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004399 return 1;
4400 }
4401 return 0;
4402}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004403#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004404
4405#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004406#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004407
4408struct slab_attribute {
4409 struct attribute attr;
4410 ssize_t (*show)(struct kmem_cache *s, char *buf);
4411 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4412};
4413
4414#define SLAB_ATTR_RO(_name) \
4415 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4416
4417#define SLAB_ATTR(_name) \
4418 static struct slab_attribute _name##_attr = \
4419 __ATTR(_name, 0644, _name##_show, _name##_store)
4420
Christoph Lameter81819f02007-05-06 14:49:36 -07004421static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4422{
4423 return sprintf(buf, "%d\n", s->size);
4424}
4425SLAB_ATTR_RO(slab_size);
4426
4427static ssize_t align_show(struct kmem_cache *s, char *buf)
4428{
4429 return sprintf(buf, "%d\n", s->align);
4430}
4431SLAB_ATTR_RO(align);
4432
4433static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4434{
4435 return sprintf(buf, "%d\n", s->objsize);
4436}
4437SLAB_ATTR_RO(object_size);
4438
4439static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4440{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004441 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004442}
4443SLAB_ATTR_RO(objs_per_slab);
4444
Christoph Lameter06b285d2008-04-14 19:11:41 +03004445static ssize_t order_store(struct kmem_cache *s,
4446 const char *buf, size_t length)
4447{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004448 unsigned long order;
4449 int err;
4450
4451 err = strict_strtoul(buf, 10, &order);
4452 if (err)
4453 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004454
4455 if (order > slub_max_order || order < slub_min_order)
4456 return -EINVAL;
4457
4458 calculate_sizes(s, order);
4459 return length;
4460}
4461
Christoph Lameter81819f02007-05-06 14:49:36 -07004462static ssize_t order_show(struct kmem_cache *s, char *buf)
4463{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004464 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004465}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004466SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004467
David Rientjes73d342b2009-02-22 17:40:09 -08004468static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4469{
4470 return sprintf(buf, "%lu\n", s->min_partial);
4471}
4472
4473static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4474 size_t length)
4475{
4476 unsigned long min;
4477 int err;
4478
4479 err = strict_strtoul(buf, 10, &min);
4480 if (err)
4481 return err;
4482
David Rientjesc0bdb232009-02-25 09:16:35 +02004483 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004484 return length;
4485}
4486SLAB_ATTR(min_partial);
4487
Christoph Lameter81819f02007-05-06 14:49:36 -07004488static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4489{
Joe Perches62c70bc2011-01-13 15:45:52 -08004490 if (!s->ctor)
4491 return 0;
4492 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004493}
4494SLAB_ATTR_RO(ctor);
4495
Christoph Lameter81819f02007-05-06 14:49:36 -07004496static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4497{
4498 return sprintf(buf, "%d\n", s->refcount - 1);
4499}
4500SLAB_ATTR_RO(aliases);
4501
Christoph Lameter81819f02007-05-06 14:49:36 -07004502static ssize_t partial_show(struct kmem_cache *s, char *buf)
4503{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004504 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004505}
4506SLAB_ATTR_RO(partial);
4507
4508static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4509{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004510 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004511}
4512SLAB_ATTR_RO(cpu_slabs);
4513
4514static ssize_t objects_show(struct kmem_cache *s, char *buf)
4515{
Christoph Lameter205ab992008-04-14 19:11:40 +03004516 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004517}
4518SLAB_ATTR_RO(objects);
4519
Christoph Lameter205ab992008-04-14 19:11:40 +03004520static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4521{
4522 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4523}
4524SLAB_ATTR_RO(objects_partial);
4525
Christoph Lameter81819f02007-05-06 14:49:36 -07004526static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4527{
4528 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4529}
4530
4531static ssize_t reclaim_account_store(struct kmem_cache *s,
4532 const char *buf, size_t length)
4533{
4534 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4535 if (buf[0] == '1')
4536 s->flags |= SLAB_RECLAIM_ACCOUNT;
4537 return length;
4538}
4539SLAB_ATTR(reclaim_account);
4540
4541static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4542{
Christoph Lameter5af60832007-05-06 14:49:56 -07004543 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004544}
4545SLAB_ATTR_RO(hwcache_align);
4546
4547#ifdef CONFIG_ZONE_DMA
4548static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4549{
4550 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4551}
4552SLAB_ATTR_RO(cache_dma);
4553#endif
4554
4555static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4556{
4557 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4558}
4559SLAB_ATTR_RO(destroy_by_rcu);
4560
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004561static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4562{
4563 return sprintf(buf, "%d\n", s->reserved);
4564}
4565SLAB_ATTR_RO(reserved);
4566
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004567#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004568static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4569{
4570 return show_slab_objects(s, buf, SO_ALL);
4571}
4572SLAB_ATTR_RO(slabs);
4573
4574static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4575{
4576 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4577}
4578SLAB_ATTR_RO(total_objects);
4579
4580static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4581{
4582 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4583}
4584
4585static ssize_t sanity_checks_store(struct kmem_cache *s,
4586 const char *buf, size_t length)
4587{
4588 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004589 if (buf[0] == '1') {
4590 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004591 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004592 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004593 return length;
4594}
4595SLAB_ATTR(sanity_checks);
4596
4597static ssize_t trace_show(struct kmem_cache *s, char *buf)
4598{
4599 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4600}
4601
4602static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4603 size_t length)
4604{
4605 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004606 if (buf[0] == '1') {
4607 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004608 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004609 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004610 return length;
4611}
4612SLAB_ATTR(trace);
4613
Christoph Lameter81819f02007-05-06 14:49:36 -07004614static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4615{
4616 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4617}
4618
4619static ssize_t red_zone_store(struct kmem_cache *s,
4620 const char *buf, size_t length)
4621{
4622 if (any_slab_objects(s))
4623 return -EBUSY;
4624
4625 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004626 if (buf[0] == '1') {
4627 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004628 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004629 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004630 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004631 return length;
4632}
4633SLAB_ATTR(red_zone);
4634
4635static ssize_t poison_show(struct kmem_cache *s, char *buf)
4636{
4637 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4638}
4639
4640static ssize_t poison_store(struct kmem_cache *s,
4641 const char *buf, size_t length)
4642{
4643 if (any_slab_objects(s))
4644 return -EBUSY;
4645
4646 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004647 if (buf[0] == '1') {
4648 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004649 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004650 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004651 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004652 return length;
4653}
4654SLAB_ATTR(poison);
4655
4656static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4657{
4658 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4659}
4660
4661static ssize_t store_user_store(struct kmem_cache *s,
4662 const char *buf, size_t length)
4663{
4664 if (any_slab_objects(s))
4665 return -EBUSY;
4666
4667 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004668 if (buf[0] == '1') {
4669 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004670 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004671 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004672 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004673 return length;
4674}
4675SLAB_ATTR(store_user);
4676
Christoph Lameter53e15af2007-05-06 14:49:43 -07004677static ssize_t validate_show(struct kmem_cache *s, char *buf)
4678{
4679 return 0;
4680}
4681
4682static ssize_t validate_store(struct kmem_cache *s,
4683 const char *buf, size_t length)
4684{
Christoph Lameter434e2452007-07-17 04:03:30 -07004685 int ret = -EINVAL;
4686
4687 if (buf[0] == '1') {
4688 ret = validate_slab_cache(s);
4689 if (ret >= 0)
4690 ret = length;
4691 }
4692 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004693}
4694SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004695
4696static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4697{
4698 if (!(s->flags & SLAB_STORE_USER))
4699 return -ENOSYS;
4700 return list_locations(s, buf, TRACK_ALLOC);
4701}
4702SLAB_ATTR_RO(alloc_calls);
4703
4704static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4705{
4706 if (!(s->flags & SLAB_STORE_USER))
4707 return -ENOSYS;
4708 return list_locations(s, buf, TRACK_FREE);
4709}
4710SLAB_ATTR_RO(free_calls);
4711#endif /* CONFIG_SLUB_DEBUG */
4712
4713#ifdef CONFIG_FAILSLAB
4714static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4715{
4716 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4717}
4718
4719static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4720 size_t length)
4721{
4722 s->flags &= ~SLAB_FAILSLAB;
4723 if (buf[0] == '1')
4724 s->flags |= SLAB_FAILSLAB;
4725 return length;
4726}
4727SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004728#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004729
Christoph Lameter2086d262007-05-06 14:49:46 -07004730static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4731{
4732 return 0;
4733}
4734
4735static ssize_t shrink_store(struct kmem_cache *s,
4736 const char *buf, size_t length)
4737{
4738 if (buf[0] == '1') {
4739 int rc = kmem_cache_shrink(s);
4740
4741 if (rc)
4742 return rc;
4743 } else
4744 return -EINVAL;
4745 return length;
4746}
4747SLAB_ATTR(shrink);
4748
Christoph Lameter81819f02007-05-06 14:49:36 -07004749#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004750static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004751{
Christoph Lameter98246012008-01-07 23:20:26 -08004752 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004753}
4754
Christoph Lameter98246012008-01-07 23:20:26 -08004755static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004756 const char *buf, size_t length)
4757{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004758 unsigned long ratio;
4759 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004760
Christoph Lameter0121c6192008-04-29 16:11:12 -07004761 err = strict_strtoul(buf, 10, &ratio);
4762 if (err)
4763 return err;
4764
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004765 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004766 s->remote_node_defrag_ratio = ratio * 10;
4767
Christoph Lameter81819f02007-05-06 14:49:36 -07004768 return length;
4769}
Christoph Lameter98246012008-01-07 23:20:26 -08004770SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004771#endif
4772
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004773#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004774static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4775{
4776 unsigned long sum = 0;
4777 int cpu;
4778 int len;
4779 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4780
4781 if (!data)
4782 return -ENOMEM;
4783
4784 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004785 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004786
4787 data[cpu] = x;
4788 sum += x;
4789 }
4790
4791 len = sprintf(buf, "%lu", sum);
4792
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004793#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004794 for_each_online_cpu(cpu) {
4795 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004796 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004797 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004798#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004799 kfree(data);
4800 return len + sprintf(buf + len, "\n");
4801}
4802
David Rientjes78eb00c2009-10-15 02:20:22 -07004803static void clear_stat(struct kmem_cache *s, enum stat_item si)
4804{
4805 int cpu;
4806
4807 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004808 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004809}
4810
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004811#define STAT_ATTR(si, text) \
4812static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4813{ \
4814 return show_stat(s, buf, si); \
4815} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004816static ssize_t text##_store(struct kmem_cache *s, \
4817 const char *buf, size_t length) \
4818{ \
4819 if (buf[0] != '0') \
4820 return -EINVAL; \
4821 clear_stat(s, si); \
4822 return length; \
4823} \
4824SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004825
4826STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4827STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4828STAT_ATTR(FREE_FASTPATH, free_fastpath);
4829STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4830STAT_ATTR(FREE_FROZEN, free_frozen);
4831STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4832STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4833STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4834STAT_ATTR(ALLOC_SLAB, alloc_slab);
4835STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05004836STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004837STAT_ATTR(FREE_SLAB, free_slab);
4838STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4839STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4840STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4841STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4842STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4843STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05004844STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03004845STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004846STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4847STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004848#endif
4849
Pekka Enberg06428782008-01-07 23:20:27 -08004850static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004851 &slab_size_attr.attr,
4852 &object_size_attr.attr,
4853 &objs_per_slab_attr.attr,
4854 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004855 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004856 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004857 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004858 &partial_attr.attr,
4859 &cpu_slabs_attr.attr,
4860 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004861 &aliases_attr.attr,
4862 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004863 &hwcache_align_attr.attr,
4864 &reclaim_account_attr.attr,
4865 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004866 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004867 &reserved_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004868#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004869 &total_objects_attr.attr,
4870 &slabs_attr.attr,
4871 &sanity_checks_attr.attr,
4872 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004873 &red_zone_attr.attr,
4874 &poison_attr.attr,
4875 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004876 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004877 &alloc_calls_attr.attr,
4878 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004879#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004880#ifdef CONFIG_ZONE_DMA
4881 &cache_dma_attr.attr,
4882#endif
4883#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004884 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004885#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004886#ifdef CONFIG_SLUB_STATS
4887 &alloc_fastpath_attr.attr,
4888 &alloc_slowpath_attr.attr,
4889 &free_fastpath_attr.attr,
4890 &free_slowpath_attr.attr,
4891 &free_frozen_attr.attr,
4892 &free_add_partial_attr.attr,
4893 &free_remove_partial_attr.attr,
4894 &alloc_from_partial_attr.attr,
4895 &alloc_slab_attr.attr,
4896 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05004897 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004898 &free_slab_attr.attr,
4899 &cpuslab_flush_attr.attr,
4900 &deactivate_full_attr.attr,
4901 &deactivate_empty_attr.attr,
4902 &deactivate_to_head_attr.attr,
4903 &deactivate_to_tail_attr.attr,
4904 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05004905 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004906 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05004907 &cmpxchg_double_fail_attr.attr,
4908 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004909#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004910#ifdef CONFIG_FAILSLAB
4911 &failslab_attr.attr,
4912#endif
4913
Christoph Lameter81819f02007-05-06 14:49:36 -07004914 NULL
4915};
4916
4917static struct attribute_group slab_attr_group = {
4918 .attrs = slab_attrs,
4919};
4920
4921static ssize_t slab_attr_show(struct kobject *kobj,
4922 struct attribute *attr,
4923 char *buf)
4924{
4925 struct slab_attribute *attribute;
4926 struct kmem_cache *s;
4927 int err;
4928
4929 attribute = to_slab_attr(attr);
4930 s = to_slab(kobj);
4931
4932 if (!attribute->show)
4933 return -EIO;
4934
4935 err = attribute->show(s, buf);
4936
4937 return err;
4938}
4939
4940static ssize_t slab_attr_store(struct kobject *kobj,
4941 struct attribute *attr,
4942 const char *buf, size_t len)
4943{
4944 struct slab_attribute *attribute;
4945 struct kmem_cache *s;
4946 int err;
4947
4948 attribute = to_slab_attr(attr);
4949 s = to_slab(kobj);
4950
4951 if (!attribute->store)
4952 return -EIO;
4953
4954 err = attribute->store(s, buf, len);
4955
4956 return err;
4957}
4958
Christoph Lameter151c6022008-01-07 22:29:05 -08004959static void kmem_cache_release(struct kobject *kobj)
4960{
4961 struct kmem_cache *s = to_slab(kobj);
4962
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004963 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004964 kfree(s);
4965}
4966
Emese Revfy52cf25d2010-01-19 02:58:23 +01004967static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004968 .show = slab_attr_show,
4969 .store = slab_attr_store,
4970};
4971
4972static struct kobj_type slab_ktype = {
4973 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004974 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004975};
4976
4977static int uevent_filter(struct kset *kset, struct kobject *kobj)
4978{
4979 struct kobj_type *ktype = get_ktype(kobj);
4980
4981 if (ktype == &slab_ktype)
4982 return 1;
4983 return 0;
4984}
4985
Emese Revfy9cd43612009-12-31 14:52:51 +01004986static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004987 .filter = uevent_filter,
4988};
4989
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004990static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004991
4992#define ID_STR_LENGTH 64
4993
4994/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004995 *
4996 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004997 */
4998static char *create_unique_id(struct kmem_cache *s)
4999{
5000 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5001 char *p = name;
5002
5003 BUG_ON(!name);
5004
5005 *p++ = ':';
5006 /*
5007 * First flags affecting slabcache operations. We will only
5008 * get here for aliasable slabs so we do not need to support
5009 * too many flags. The flags here must cover all flags that
5010 * are matched during merging to guarantee that the id is
5011 * unique.
5012 */
5013 if (s->flags & SLAB_CACHE_DMA)
5014 *p++ = 'd';
5015 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5016 *p++ = 'a';
5017 if (s->flags & SLAB_DEBUG_FREE)
5018 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005019 if (!(s->flags & SLAB_NOTRACK))
5020 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005021 if (p != name + 1)
5022 *p++ = '-';
5023 p += sprintf(p, "%07d", s->size);
5024 BUG_ON(p > name + ID_STR_LENGTH - 1);
5025 return name;
5026}
5027
5028static int sysfs_slab_add(struct kmem_cache *s)
5029{
5030 int err;
5031 const char *name;
5032 int unmergeable;
5033
5034 if (slab_state < SYSFS)
5035 /* Defer until later */
5036 return 0;
5037
5038 unmergeable = slab_unmergeable(s);
5039 if (unmergeable) {
5040 /*
5041 * Slabcache can never be merged so we can use the name proper.
5042 * This is typically the case for debug situations. In that
5043 * case we can catch duplicate names easily.
5044 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005045 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005046 name = s->name;
5047 } else {
5048 /*
5049 * Create a unique name for the slab as a target
5050 * for the symlinks.
5051 */
5052 name = create_unique_id(s);
5053 }
5054
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005055 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005056 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
5057 if (err) {
5058 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005059 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005060 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005061
5062 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005063 if (err) {
5064 kobject_del(&s->kobj);
5065 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005066 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005067 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005068 kobject_uevent(&s->kobj, KOBJ_ADD);
5069 if (!unmergeable) {
5070 /* Setup first alias */
5071 sysfs_slab_alias(s, s->name);
5072 kfree(name);
5073 }
5074 return 0;
5075}
5076
5077static void sysfs_slab_remove(struct kmem_cache *s)
5078{
Christoph Lameter2bce6482010-07-19 11:39:11 -05005079 if (slab_state < SYSFS)
5080 /*
5081 * Sysfs has not been setup yet so no need to remove the
5082 * cache from sysfs.
5083 */
5084 return;
5085
Christoph Lameter81819f02007-05-06 14:49:36 -07005086 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5087 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005088 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005089}
5090
5091/*
5092 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005093 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005094 */
5095struct saved_alias {
5096 struct kmem_cache *s;
5097 const char *name;
5098 struct saved_alias *next;
5099};
5100
Adrian Bunk5af328a2007-07-17 04:03:27 -07005101static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005102
5103static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5104{
5105 struct saved_alias *al;
5106
5107 if (slab_state == SYSFS) {
5108 /*
5109 * If we have a leftover link then remove it.
5110 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005111 sysfs_remove_link(&slab_kset->kobj, name);
5112 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005113 }
5114
5115 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5116 if (!al)
5117 return -ENOMEM;
5118
5119 al->s = s;
5120 al->name = name;
5121 al->next = alias_list;
5122 alias_list = al;
5123 return 0;
5124}
5125
5126static int __init slab_sysfs_init(void)
5127{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005128 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005129 int err;
5130
Christoph Lameter2bce6482010-07-19 11:39:11 -05005131 down_write(&slub_lock);
5132
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005133 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005134 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05005135 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005136 printk(KERN_ERR "Cannot register slab subsystem.\n");
5137 return -ENOSYS;
5138 }
5139
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005140 slab_state = SYSFS;
5141
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005142 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005143 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005144 if (err)
5145 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
5146 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005147 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005148
5149 while (alias_list) {
5150 struct saved_alias *al = alias_list;
5151
5152 alias_list = alias_list->next;
5153 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005154 if (err)
5155 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
5156 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005157 kfree(al);
5158 }
5159
Christoph Lameter2bce6482010-07-19 11:39:11 -05005160 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005161 resiliency_test();
5162 return 0;
5163}
5164
5165__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005166#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005167
5168/*
5169 * The /proc/slabinfo ABI
5170 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005171#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005172static void print_slabinfo_header(struct seq_file *m)
5173{
5174 seq_puts(m, "slabinfo - version: 2.1\n");
5175 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
5176 "<objperslab> <pagesperslab>");
5177 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
5178 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5179 seq_putc(m, '\n');
5180}
5181
5182static void *s_start(struct seq_file *m, loff_t *pos)
5183{
5184 loff_t n = *pos;
5185
5186 down_read(&slub_lock);
5187 if (!n)
5188 print_slabinfo_header(m);
5189
5190 return seq_list_start(&slab_caches, *pos);
5191}
5192
5193static void *s_next(struct seq_file *m, void *p, loff_t *pos)
5194{
5195 return seq_list_next(p, &slab_caches, pos);
5196}
5197
5198static void s_stop(struct seq_file *m, void *p)
5199{
5200 up_read(&slub_lock);
5201}
5202
5203static int s_show(struct seq_file *m, void *p)
5204{
5205 unsigned long nr_partials = 0;
5206 unsigned long nr_slabs = 0;
5207 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005208 unsigned long nr_objs = 0;
5209 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005210 struct kmem_cache *s;
5211 int node;
5212
5213 s = list_entry(p, struct kmem_cache, list);
5214
5215 for_each_online_node(node) {
5216 struct kmem_cache_node *n = get_node(s, node);
5217
5218 if (!n)
5219 continue;
5220
5221 nr_partials += n->nr_partial;
5222 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03005223 nr_objs += atomic_long_read(&n->total_objects);
5224 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005225 }
5226
Christoph Lameter205ab992008-04-14 19:11:40 +03005227 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005228
5229 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03005230 nr_objs, s->size, oo_objects(s->oo),
5231 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005232 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
5233 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
5234 0UL);
5235 seq_putc(m, '\n');
5236 return 0;
5237}
5238
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005239static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005240 .start = s_start,
5241 .next = s_next,
5242 .stop = s_stop,
5243 .show = s_show,
5244};
5245
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005246static int slabinfo_open(struct inode *inode, struct file *file)
5247{
5248 return seq_open(file, &slabinfo_op);
5249}
5250
5251static const struct file_operations proc_slabinfo_operations = {
5252 .open = slabinfo_open,
5253 .read = seq_read,
5254 .llseek = seq_lseek,
5255 .release = seq_release,
5256};
5257
5258static int __init slab_proc_init(void)
5259{
WANG Congcf5d1132009-08-18 19:11:40 +03005260 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005261 return 0;
5262}
5263module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08005264#endif /* CONFIG_SLABINFO */