blob: 4907563ef7ff7e199577f5d2ffedda6fa358271e [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
Christoph Lameter881db7f2011-06-01 12:25:53 -05005 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07007 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter881db7f2011-06-01 12:25:53 -05009 * (C) 2011 Linux Foundation, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -070010 */
11
12#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100013#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070014#include <linux/module.h>
15#include <linux/bit_spinlock.h>
16#include <linux/interrupt.h>
17#include <linux/bitops.h>
18#include <linux/slab.h>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040019#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070020#include <linux/seq_file.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020021#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070022#include <linux/cpu.h>
23#include <linux/cpuset.h>
24#include <linux/mempolicy.h>
25#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070026#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070027#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070028#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070029#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090030#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030031#include <linux/stacktrace.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070032
Richard Kennedy4a923792010-10-21 10:29:19 +010033#include <trace/events/kmem.h>
34
Christoph Lameter81819f02007-05-06 14:49:36 -070035/*
36 * Lock order:
Christoph Lameter881db7f2011-06-01 12:25:53 -050037 * 1. slub_lock (Global Semaphore)
38 * 2. node->list_lock
39 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070040 *
Christoph Lameter881db7f2011-06-01 12:25:53 -050041 * slub_lock
42 *
43 * The role of the slub_lock is to protect the list of all the slabs
44 * and to synchronize major metadata changes to slab cache structures.
45 *
46 * The slab_lock is only used for debugging and on arches that do not
47 * have the ability to do a cmpxchg_double. It only protects the second
48 * double word in the page struct. Meaning
49 * A. page->freelist -> List of object free in a page
50 * B. page->counters -> Counters of objects
51 * C. page->frozen -> frozen state
52 *
53 * If a slab is frozen then it is exempt from list management. It is not
54 * on any list. The processor that froze the slab is the one who can
55 * perform list operations on the page. Other processors may put objects
56 * onto the freelist but the processor that froze the slab is the only
57 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070058 *
59 * The list_lock protects the partial and full list on each node and
60 * the partial slab counter. If taken then no new slabs may be added or
61 * removed from the lists nor make the number of partial slabs be modified.
62 * (Note that the total number of slabs is an atomic value that may be
63 * modified without taking the list lock).
64 *
65 * The list_lock is a centralized lock and thus we avoid taking it as
66 * much as possible. As long as SLUB does not have to handle partial
67 * slabs, operations can continue without any centralized lock. F.e.
68 * allocating a long series of objects that fill up slabs does not require
69 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070070 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
74 *
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
77 *
Christoph Lameter672bba32007-05-09 02:32:39 -070078 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070080 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070081 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070083 *
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
87 *
88 * Overloading of page flags that are otherwise used for LRU management.
89 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070090 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
98 *
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700102 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700105 *
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700108 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700109 */
110
Christoph Lameteraf537b02010-07-09 14:07:14 -0500111#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
113
114static inline int kmem_cache_debug(struct kmem_cache *s)
115{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700116#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500117 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700118#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500119 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500121}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700122
Christoph Lameter81819f02007-05-06 14:49:36 -0700123/*
124 * Issues still to be resolved:
125 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
127 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700128 * - Variable sizing of the per node arrays
129 */
130
131/* Enable to test recovery from slab corruption on boot */
132#undef SLUB_RESILIENCY_TEST
133
Christoph Lameterb789ef52011-06-01 12:25:49 -0500134/* Enable to log cmpxchg failures */
135#undef SLUB_DEBUG_CMPXCHG
136
Christoph Lameter81819f02007-05-06 14:49:36 -0700137/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700138 * Mininum number of partial slabs. These will be left on the partial
139 * lists even if they are empty. kmem_cache_shrink may reclaim them.
140 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800141#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700142
Christoph Lameter2086d262007-05-06 14:49:46 -0700143/*
144 * Maximum number of desirable partial slabs.
145 * The existence of more partial slabs makes kmem_cache_shrink
146 * sort the partial list by the number of objects in the.
147 */
148#define MAX_PARTIAL 10
149
Christoph Lameter81819f02007-05-06 14:49:36 -0700150#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
151 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700152
Christoph Lameter81819f02007-05-06 14:49:36 -0700153/*
David Rientjes3de47212009-07-27 18:30:35 -0700154 * Debugging flags that require metadata to be stored in the slab. These get
155 * disabled when slub_debug=O is used and a cache's min order increases with
156 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700157 */
David Rientjes3de47212009-07-27 18:30:35 -0700158#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700159
160/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700161 * Set of flags that will prevent slab merging
162 */
163#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300164 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
165 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166
167#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200168 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700169
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400170#define OO_SHIFT 16
171#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500172#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400173
Christoph Lameter81819f02007-05-06 14:49:36 -0700174/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500175#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500176#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700177
178static int kmem_size = sizeof(struct kmem_cache);
179
180#ifdef CONFIG_SMP
181static struct notifier_block slab_notifier;
182#endif
183
184static enum {
185 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500186 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700187 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700188 SYSFS /* Sysfs up */
189} slab_state = DOWN;
190
191/* A list of all slab caches on the system */
192static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700193static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700194
Christoph Lameter02cbc872007-05-09 02:32:43 -0700195/*
196 * Tracking user of a slab.
197 */
Ben Greeard6543e32011-07-07 11:36:36 -0700198#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700199struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300200 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700201#ifdef CONFIG_STACKTRACE
202 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
203#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700204 int cpu; /* Was running on cpu */
205 int pid; /* Pid context */
206 unsigned long when; /* When did the operation occur */
207};
208
209enum track_item { TRACK_ALLOC, TRACK_FREE };
210
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500211#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700212static int sysfs_slab_add(struct kmem_cache *);
213static int sysfs_slab_alias(struct kmem_cache *, const char *);
214static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800215
Christoph Lameter81819f02007-05-06 14:49:36 -0700216#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700217static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
218static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
219 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800220static inline void sysfs_slab_remove(struct kmem_cache *s)
221{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300222 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800223 kfree(s);
224}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800225
Christoph Lameter81819f02007-05-06 14:49:36 -0700226#endif
227
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500228static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800229{
230#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600231 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800232#endif
233}
234
Christoph Lameter81819f02007-05-06 14:49:36 -0700235/********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
238
239int slab_is_available(void)
240{
241 return slab_state >= UP;
242}
243
244static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
245{
Christoph Lameter81819f02007-05-06 14:49:36 -0700246 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700247}
248
Christoph Lameter6446faa2008-02-15 23:45:26 -0800249/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700250static inline int check_valid_pointer(struct kmem_cache *s,
251 struct page *page, const void *object)
252{
253 void *base;
254
Christoph Lametera973e9d2008-03-01 13:40:44 -0800255 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700256 return 1;
257
Christoph Lametera973e9d2008-03-01 13:40:44 -0800258 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300259 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700260 (object - base) % s->size) {
261 return 0;
262 }
263
264 return 1;
265}
266
Christoph Lameter7656c722007-05-09 02:32:40 -0700267static inline void *get_freepointer(struct kmem_cache *s, void *object)
268{
269 return *(void **)(object + s->offset);
270}
271
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500272static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
273{
274 void *p;
275
276#ifdef CONFIG_DEBUG_PAGEALLOC
277 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
278#else
279 p = get_freepointer(s, object);
280#endif
281 return p;
282}
283
Christoph Lameter7656c722007-05-09 02:32:40 -0700284static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
285{
286 *(void **)(object + s->offset) = fp;
287}
288
289/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300290#define for_each_object(__p, __s, __addr, __objects) \
291 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700292 __p += (__s)->size)
293
Christoph Lameter7656c722007-05-09 02:32:40 -0700294/* Determine object index from a given position */
295static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
296{
297 return (p - addr) / s->size;
298}
299
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100300static inline size_t slab_ksize(const struct kmem_cache *s)
301{
302#ifdef CONFIG_SLUB_DEBUG
303 /*
304 * Debugging requires use of the padding between object
305 * and whatever may come after it.
306 */
307 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
308 return s->objsize;
309
310#endif
311 /*
312 * If we have the need to store the freelist pointer
313 * back there or track user information then we can
314 * only use the space before that information.
315 */
316 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
317 return s->inuse;
318 /*
319 * Else we can use all the padding etc for the allocation
320 */
321 return s->size;
322}
323
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800324static inline int order_objects(int order, unsigned long size, int reserved)
325{
326 return ((PAGE_SIZE << order) - reserved) / size;
327}
328
Christoph Lameter834f3d12008-04-14 19:11:31 +0300329static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800330 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300331{
332 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800333 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300334 };
335
336 return x;
337}
338
339static inline int oo_order(struct kmem_cache_order_objects x)
340{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400341 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300342}
343
344static inline int oo_objects(struct kmem_cache_order_objects x)
345{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400346 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300347}
348
Christoph Lameter881db7f2011-06-01 12:25:53 -0500349/*
350 * Per slab locking using the pagelock
351 */
352static __always_inline void slab_lock(struct page *page)
353{
354 bit_spin_lock(PG_locked, &page->flags);
355}
356
357static __always_inline void slab_unlock(struct page *page)
358{
359 __bit_spin_unlock(PG_locked, &page->flags);
360}
361
Christoph Lameter1d071712011-07-14 12:49:12 -0500362/* Interrupts must be disabled (for the fallback code to work right) */
363static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500364 void *freelist_old, unsigned long counters_old,
365 void *freelist_new, unsigned long counters_new,
366 const char *n)
367{
Christoph Lameter1d071712011-07-14 12:49:12 -0500368 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800369#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
370 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500371 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000372 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500373 freelist_old, counters_old,
374 freelist_new, counters_new))
375 return 1;
376 } else
377#endif
378 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500379 slab_lock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500380 if (page->freelist == freelist_old && page->counters == counters_old) {
381 page->freelist = freelist_new;
382 page->counters = counters_new;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500383 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500384 return 1;
385 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500386 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500387 }
388
389 cpu_relax();
390 stat(s, CMPXCHG_DOUBLE_FAIL);
391
392#ifdef SLUB_DEBUG_CMPXCHG
393 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
394#endif
395
396 return 0;
397}
398
Christoph Lameter1d071712011-07-14 12:49:12 -0500399static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
400 void *freelist_old, unsigned long counters_old,
401 void *freelist_new, unsigned long counters_new,
402 const char *n)
403{
Heiko Carstens25654092012-01-12 17:17:33 -0800404#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
405 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500406 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000407 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameter1d071712011-07-14 12:49:12 -0500408 freelist_old, counters_old,
409 freelist_new, counters_new))
410 return 1;
411 } else
412#endif
413 {
414 unsigned long flags;
415
416 local_irq_save(flags);
417 slab_lock(page);
418 if (page->freelist == freelist_old && page->counters == counters_old) {
419 page->freelist = freelist_new;
420 page->counters = counters_new;
421 slab_unlock(page);
422 local_irq_restore(flags);
423 return 1;
424 }
425 slab_unlock(page);
426 local_irq_restore(flags);
427 }
428
429 cpu_relax();
430 stat(s, CMPXCHG_DOUBLE_FAIL);
431
432#ifdef SLUB_DEBUG_CMPXCHG
433 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
434#endif
435
436 return 0;
437}
438
Christoph Lameter41ecc552007-05-09 02:32:44 -0700439#ifdef CONFIG_SLUB_DEBUG
440/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500441 * Determine a map of object in use on a page.
442 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500443 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500444 * not vanish from under us.
445 */
446static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
447{
448 void *p;
449 void *addr = page_address(page);
450
451 for (p = page->freelist; p; p = get_freepointer(s, p))
452 set_bit(slab_index(p, s, addr), map);
453}
454
Christoph Lameter41ecc552007-05-09 02:32:44 -0700455/*
456 * Debug settings:
457 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700458#ifdef CONFIG_SLUB_DEBUG_ON
459static int slub_debug = DEBUG_DEFAULT_FLAGS;
460#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700461static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700462#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700463
464static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700465static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700466
Christoph Lameter7656c722007-05-09 02:32:40 -0700467/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700468 * Object debugging
469 */
470static void print_section(char *text, u8 *addr, unsigned int length)
471{
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200472 print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
473 length, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700474}
475
Christoph Lameter81819f02007-05-06 14:49:36 -0700476static struct track *get_track(struct kmem_cache *s, void *object,
477 enum track_item alloc)
478{
479 struct track *p;
480
481 if (s->offset)
482 p = object + s->offset + sizeof(void *);
483 else
484 p = object + s->inuse;
485
486 return p + alloc;
487}
488
489static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300490 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700491{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900492 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700493
Christoph Lameter81819f02007-05-06 14:49:36 -0700494 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700495#ifdef CONFIG_STACKTRACE
496 struct stack_trace trace;
497 int i;
498
499 trace.nr_entries = 0;
500 trace.max_entries = TRACK_ADDRS_COUNT;
501 trace.entries = p->addrs;
502 trace.skip = 3;
503 save_stack_trace(&trace);
504
505 /* See rant in lockdep.c */
506 if (trace.nr_entries != 0 &&
507 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
508 trace.nr_entries--;
509
510 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
511 p->addrs[i] = 0;
512#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700513 p->addr = addr;
514 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400515 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700516 p->when = jiffies;
517 } else
518 memset(p, 0, sizeof(struct track));
519}
520
Christoph Lameter81819f02007-05-06 14:49:36 -0700521static void init_tracking(struct kmem_cache *s, void *object)
522{
Christoph Lameter24922682007-07-17 04:03:18 -0700523 if (!(s->flags & SLAB_STORE_USER))
524 return;
525
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300526 set_track(s, object, TRACK_FREE, 0UL);
527 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700528}
529
530static void print_track(const char *s, struct track *t)
531{
532 if (!t->addr)
533 return;
534
Linus Torvalds7daf7052008-07-14 12:12:53 -0700535 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300536 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700537#ifdef CONFIG_STACKTRACE
538 {
539 int i;
540 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
541 if (t->addrs[i])
542 printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
543 else
544 break;
545 }
546#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700547}
548
Christoph Lameter24922682007-07-17 04:03:18 -0700549static void print_tracking(struct kmem_cache *s, void *object)
550{
551 if (!(s->flags & SLAB_STORE_USER))
552 return;
553
554 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
555 print_track("Freed", get_track(s, object, TRACK_FREE));
556}
557
558static void print_page_info(struct page *page)
559{
Christoph Lameter39b26462008-04-14 19:11:30 +0300560 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
561 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700562
563}
564
565static void slab_bug(struct kmem_cache *s, char *fmt, ...)
566{
567 va_list args;
568 char buf[100];
569
570 va_start(args, fmt);
571 vsnprintf(buf, sizeof(buf), fmt, args);
572 va_end(args);
573 printk(KERN_ERR "========================================"
574 "=====================================\n");
Dave Jones265d47e2011-11-15 15:04:00 -0800575 printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700576 printk(KERN_ERR "----------------------------------------"
577 "-------------------------------------\n\n");
578}
579
580static void slab_fix(struct kmem_cache *s, char *fmt, ...)
581{
582 va_list args;
583 char buf[100];
584
585 va_start(args, fmt);
586 vsnprintf(buf, sizeof(buf), fmt, args);
587 va_end(args);
588 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
589}
590
591static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700592{
593 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800594 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700595
596 print_tracking(s, p);
597
598 print_page_info(page);
599
600 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
601 p, p - addr, get_freepointer(s, p));
602
603 if (p > addr + 16)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200604 print_section("Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700605
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200606 print_section("Object ", p, min_t(unsigned long, s->objsize,
607 PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700608 if (s->flags & SLAB_RED_ZONE)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200609 print_section("Redzone ", p + s->objsize,
Christoph Lameter81819f02007-05-06 14:49:36 -0700610 s->inuse - s->objsize);
611
Christoph Lameter81819f02007-05-06 14:49:36 -0700612 if (s->offset)
613 off = s->offset + sizeof(void *);
614 else
615 off = s->inuse;
616
Christoph Lameter24922682007-07-17 04:03:18 -0700617 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700618 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700619
620 if (off != s->size)
621 /* Beginning of the filler is the free pointer */
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200622 print_section("Padding ", p + off, s->size - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700623
624 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700625}
626
627static void object_err(struct kmem_cache *s, struct page *page,
628 u8 *object, char *reason)
629{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700630 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700631 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700632}
633
Christoph Lameter24922682007-07-17 04:03:18 -0700634static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700635{
636 va_list args;
637 char buf[100];
638
Christoph Lameter24922682007-07-17 04:03:18 -0700639 va_start(args, fmt);
640 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700641 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700642 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700643 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700644 dump_stack();
645}
646
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500647static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700648{
649 u8 *p = object;
650
651 if (s->flags & __OBJECT_POISON) {
652 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800653 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700654 }
655
656 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500657 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700658}
659
Christoph Lameter24922682007-07-17 04:03:18 -0700660static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
661 void *from, void *to)
662{
663 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
664 memset(from, data, to - from);
665}
666
667static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
668 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800669 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700670{
671 u8 *fault;
672 u8 *end;
673
Akinobu Mita79824822011-10-31 17:08:07 -0700674 fault = memchr_inv(start, value, bytes);
Christoph Lameter24922682007-07-17 04:03:18 -0700675 if (!fault)
676 return 1;
677
678 end = start + bytes;
679 while (end > fault && end[-1] == value)
680 end--;
681
682 slab_bug(s, "%s overwritten", what);
683 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
684 fault, end - 1, fault[0], value);
685 print_trailer(s, page, object);
686
687 restore_bytes(s, what, value, fault, end);
688 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700689}
690
Christoph Lameter81819f02007-05-06 14:49:36 -0700691/*
692 * Object layout:
693 *
694 * object address
695 * Bytes of the object to be managed.
696 * If the freepointer may overlay the object then the free
697 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700698 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700699 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
700 * 0xa5 (POISON_END)
701 *
702 * object + s->objsize
703 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700704 * Padding is extended by another word if Redzoning is enabled and
705 * objsize == inuse.
706 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700707 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
708 * 0xcc (RED_ACTIVE) for objects in use.
709 *
710 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700711 * Meta data starts here.
712 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700713 * A. Free pointer (if we cannot overwrite object on free)
714 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700715 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800716 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700717 * before the word boundary.
718 *
719 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700720 *
721 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700722 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700723 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700724 * If slabcaches are merged then the objsize and inuse boundaries are mostly
725 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700726 * may be used with merged slabcaches.
727 */
728
Christoph Lameter81819f02007-05-06 14:49:36 -0700729static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
730{
731 unsigned long off = s->inuse; /* The end of info */
732
733 if (s->offset)
734 /* Freepointer is placed after the object. */
735 off += sizeof(void *);
736
737 if (s->flags & SLAB_STORE_USER)
738 /* We also have user information there */
739 off += 2 * sizeof(struct track);
740
741 if (s->size == off)
742 return 1;
743
Christoph Lameter24922682007-07-17 04:03:18 -0700744 return check_bytes_and_report(s, page, p, "Object padding",
745 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700746}
747
Christoph Lameter39b26462008-04-14 19:11:30 +0300748/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700749static int slab_pad_check(struct kmem_cache *s, struct page *page)
750{
Christoph Lameter24922682007-07-17 04:03:18 -0700751 u8 *start;
752 u8 *fault;
753 u8 *end;
754 int length;
755 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700756
757 if (!(s->flags & SLAB_POISON))
758 return 1;
759
Christoph Lametera973e9d2008-03-01 13:40:44 -0800760 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800761 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300762 end = start + length;
763 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700764 if (!remainder)
765 return 1;
766
Akinobu Mita79824822011-10-31 17:08:07 -0700767 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700768 if (!fault)
769 return 1;
770 while (end > fault && end[-1] == POISON_INUSE)
771 end--;
772
773 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200774 print_section("Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700775
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200776 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700777 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700778}
779
780static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500781 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700782{
783 u8 *p = object;
784 u8 *endobject = object + s->objsize;
785
786 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700787 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500788 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700789 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700790 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800791 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
792 check_bytes_and_report(s, page, p, "Alignment padding",
793 endobject, POISON_INUSE, s->inuse - s->objsize);
794 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700795 }
796
797 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500798 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700799 (!check_bytes_and_report(s, page, p, "Poison", p,
800 POISON_FREE, s->objsize - 1) ||
801 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800802 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700803 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700804 /*
805 * check_pad_bytes cleans up on its own.
806 */
807 check_pad_bytes(s, page, p);
808 }
809
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500810 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700811 /*
812 * Object and freepointer overlap. Cannot check
813 * freepointer while object is allocated.
814 */
815 return 1;
816
817 /* Check free pointer validity */
818 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
819 object_err(s, page, p, "Freepointer corrupt");
820 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100821 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700822 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700823 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700824 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800825 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700826 return 0;
827 }
828 return 1;
829}
830
831static int check_slab(struct kmem_cache *s, struct page *page)
832{
Christoph Lameter39b26462008-04-14 19:11:30 +0300833 int maxobj;
834
Christoph Lameter81819f02007-05-06 14:49:36 -0700835 VM_BUG_ON(!irqs_disabled());
836
837 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700838 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700839 return 0;
840 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300841
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800842 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300843 if (page->objects > maxobj) {
844 slab_err(s, page, "objects %u > max %u",
845 s->name, page->objects, maxobj);
846 return 0;
847 }
848 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700849 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300850 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700851 return 0;
852 }
853 /* Slab_pad_check fixes things up after itself */
854 slab_pad_check(s, page);
855 return 1;
856}
857
858/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700859 * Determine if a certain object on a page is on the freelist. Must hold the
860 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700861 */
862static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
863{
864 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500865 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700866 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300867 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700868
Christoph Lameter881db7f2011-06-01 12:25:53 -0500869 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300870 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700871 if (fp == search)
872 return 1;
873 if (!check_valid_pointer(s, page, fp)) {
874 if (object) {
875 object_err(s, page, object,
876 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800877 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700878 break;
879 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700880 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800881 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300882 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700883 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700884 return 0;
885 }
886 break;
887 }
888 object = fp;
889 fp = get_freepointer(s, object);
890 nr++;
891 }
892
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800893 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400894 if (max_objects > MAX_OBJS_PER_PAGE)
895 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300896
897 if (page->objects != max_objects) {
898 slab_err(s, page, "Wrong number of objects. Found %d but "
899 "should be %d", page->objects, max_objects);
900 page->objects = max_objects;
901 slab_fix(s, "Number of objects adjusted.");
902 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300903 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700904 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300905 "counted were %d", page->inuse, page->objects - nr);
906 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700907 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700908 }
909 return search == NULL;
910}
911
Christoph Lameter0121c6192008-04-29 16:11:12 -0700912static void trace(struct kmem_cache *s, struct page *page, void *object,
913 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700914{
915 if (s->flags & SLAB_TRACE) {
916 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
917 s->name,
918 alloc ? "alloc" : "free",
919 object, page->inuse,
920 page->freelist);
921
922 if (!alloc)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200923 print_section("Object ", (void *)object, s->objsize);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700924
925 dump_stack();
926 }
927}
928
Christoph Lameter643b1132007-05-06 14:49:42 -0700929/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500930 * Hooks for other subsystems that check memory allocations. In a typical
931 * production configuration these hooks all should produce no code at all.
932 */
933static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
934{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500935 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500936 lockdep_trace_alloc(flags);
937 might_sleep_if(flags & __GFP_WAIT);
938
939 return should_failslab(s->objsize, flags, s->flags);
940}
941
942static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
943{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500944 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100945 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500946 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
947}
948
949static inline void slab_free_hook(struct kmem_cache *s, void *x)
950{
951 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500952
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600953 /*
954 * Trouble is that we may no longer disable interupts in the fast path
955 * So in order to make the debug calls that expect irqs to be
956 * disabled we need to disable interrupts temporarily.
957 */
958#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
959 {
960 unsigned long flags;
961
962 local_irq_save(flags);
963 kmemcheck_slab_free(s, x, s->objsize);
964 debug_check_no_locks_freed(x, s->objsize);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600965 local_irq_restore(flags);
966 }
967#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200968 if (!(s->flags & SLAB_DEBUG_OBJECTS))
969 debug_check_no_obj_freed(x, s->objsize);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500970}
971
972/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700973 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500974 *
975 * list_lock must be held.
Christoph Lameter643b1132007-05-06 14:49:42 -0700976 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500977static void add_full(struct kmem_cache *s,
978 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700979{
Christoph Lameter643b1132007-05-06 14:49:42 -0700980 if (!(s->flags & SLAB_STORE_USER))
981 return;
982
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500983 list_add(&page->lru, &n->full);
984}
Christoph Lameter643b1132007-05-06 14:49:42 -0700985
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500986/*
987 * list_lock must be held.
988 */
989static void remove_full(struct kmem_cache *s, struct page *page)
990{
991 if (!(s->flags & SLAB_STORE_USER))
992 return;
993
Christoph Lameter643b1132007-05-06 14:49:42 -0700994 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -0700995}
996
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300997/* Tracking of the number of slabs for debugging purposes */
998static inline unsigned long slabs_node(struct kmem_cache *s, int node)
999{
1000 struct kmem_cache_node *n = get_node(s, node);
1001
1002 return atomic_long_read(&n->nr_slabs);
1003}
1004
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001005static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1006{
1007 return atomic_long_read(&n->nr_slabs);
1008}
1009
Christoph Lameter205ab992008-04-14 19:11:40 +03001010static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001011{
1012 struct kmem_cache_node *n = get_node(s, node);
1013
1014 /*
1015 * May be called early in order to allocate a slab for the
1016 * kmem_cache_node structure. Solve the chicken-egg
1017 * dilemma by deferring the increment of the count during
1018 * bootstrap (see early_kmem_cache_node_alloc).
1019 */
Christoph Lameter7340cc82010-09-28 08:10:26 -05001020 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001021 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001022 atomic_long_add(objects, &n->total_objects);
1023 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001024}
Christoph Lameter205ab992008-04-14 19:11:40 +03001025static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001026{
1027 struct kmem_cache_node *n = get_node(s, node);
1028
1029 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001030 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001031}
1032
1033/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001034static void setup_object_debug(struct kmem_cache *s, struct page *page,
1035 void *object)
1036{
1037 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1038 return;
1039
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001040 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001041 init_tracking(s, object);
1042}
1043
Christoph Lameter15370662010-08-20 12:37:12 -05001044static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001045 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001046{
1047 if (!check_slab(s, page))
1048 goto bad;
1049
Christoph Lameter81819f02007-05-06 14:49:36 -07001050 if (!check_valid_pointer(s, page, object)) {
1051 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001052 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001053 }
1054
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001055 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001056 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001057
Christoph Lameter3ec09742007-05-16 22:11:00 -07001058 /* Success perform special debug activities for allocs */
1059 if (s->flags & SLAB_STORE_USER)
1060 set_track(s, object, TRACK_ALLOC, addr);
1061 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001062 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001063 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001064
Christoph Lameter81819f02007-05-06 14:49:36 -07001065bad:
1066 if (PageSlab(page)) {
1067 /*
1068 * If this is a slab page then lets do the best we can
1069 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001070 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001071 */
Christoph Lameter24922682007-07-17 04:03:18 -07001072 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001073 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001074 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001075 }
1076 return 0;
1077}
1078
Christoph Lameter15370662010-08-20 12:37:12 -05001079static noinline int free_debug_processing(struct kmem_cache *s,
1080 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001081{
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001082 unsigned long flags;
1083 int rc = 0;
1084
1085 local_irq_save(flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001086 slab_lock(page);
1087
Christoph Lameter81819f02007-05-06 14:49:36 -07001088 if (!check_slab(s, page))
1089 goto fail;
1090
1091 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001092 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001093 goto fail;
1094 }
1095
1096 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001097 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001098 goto fail;
1099 }
1100
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001101 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001102 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001103
1104 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001105 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001106 slab_err(s, page, "Attempt to free object(0x%p) "
1107 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001108 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001109 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001110 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001111 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001112 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001113 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001114 object_err(s, page, object,
1115 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001116 goto fail;
1117 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001118
Christoph Lameter3ec09742007-05-16 22:11:00 -07001119 if (s->flags & SLAB_STORE_USER)
1120 set_track(s, object, TRACK_FREE, addr);
1121 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001122 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001123 rc = 1;
1124out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001125 slab_unlock(page);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001126 local_irq_restore(flags);
1127 return rc;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001128
Christoph Lameter81819f02007-05-06 14:49:36 -07001129fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001130 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001131 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001132}
1133
Christoph Lameter41ecc552007-05-09 02:32:44 -07001134static int __init setup_slub_debug(char *str)
1135{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001136 slub_debug = DEBUG_DEFAULT_FLAGS;
1137 if (*str++ != '=' || !*str)
1138 /*
1139 * No options specified. Switch on full debugging.
1140 */
1141 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001142
1143 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001144 /*
1145 * No options but restriction on slabs. This means full
1146 * debugging for slabs matching a pattern.
1147 */
1148 goto check_slabs;
1149
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001150 if (tolower(*str) == 'o') {
1151 /*
1152 * Avoid enabling debugging on caches if its minimum order
1153 * would increase as a result.
1154 */
1155 disable_higher_order_debug = 1;
1156 goto out;
1157 }
1158
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001159 slub_debug = 0;
1160 if (*str == '-')
1161 /*
1162 * Switch off all debugging measures.
1163 */
1164 goto out;
1165
1166 /*
1167 * Determine which debug features should be switched on
1168 */
Pekka Enberg06428782008-01-07 23:20:27 -08001169 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001170 switch (tolower(*str)) {
1171 case 'f':
1172 slub_debug |= SLAB_DEBUG_FREE;
1173 break;
1174 case 'z':
1175 slub_debug |= SLAB_RED_ZONE;
1176 break;
1177 case 'p':
1178 slub_debug |= SLAB_POISON;
1179 break;
1180 case 'u':
1181 slub_debug |= SLAB_STORE_USER;
1182 break;
1183 case 't':
1184 slub_debug |= SLAB_TRACE;
1185 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001186 case 'a':
1187 slub_debug |= SLAB_FAILSLAB;
1188 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001189 default:
1190 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001191 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001192 }
1193 }
1194
1195check_slabs:
1196 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001197 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001198out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001199 return 1;
1200}
1201
1202__setup("slub_debug", setup_slub_debug);
1203
Christoph Lameterba0268a2007-09-11 15:24:11 -07001204static unsigned long kmem_cache_flags(unsigned long objsize,
1205 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001206 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001207{
1208 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001209 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001210 */
Christoph Lametere1533622008-02-15 23:45:24 -08001211 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001212 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1213 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001214
1215 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001216}
1217#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001218static inline void setup_object_debug(struct kmem_cache *s,
1219 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001220
Christoph Lameter3ec09742007-05-16 22:11:00 -07001221static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001222 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001223
Christoph Lameter3ec09742007-05-16 22:11:00 -07001224static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001225 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001226
Christoph Lameter41ecc552007-05-09 02:32:44 -07001227static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1228 { return 1; }
1229static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001230 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001231static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1232 struct page *page) {}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001233static inline void remove_full(struct kmem_cache *s, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001234static inline unsigned long kmem_cache_flags(unsigned long objsize,
1235 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001236 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001237{
1238 return flags;
1239}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001240#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001241
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001242#define disable_higher_order_debug 0
1243
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001244static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1245 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001246static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1247 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001248static inline void inc_slabs_node(struct kmem_cache *s, int node,
1249 int objects) {}
1250static inline void dec_slabs_node(struct kmem_cache *s, int node,
1251 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001252
1253static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1254 { return 0; }
1255
1256static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1257 void *object) {}
1258
1259static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1260
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001261#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001262
Christoph Lameter81819f02007-05-06 14:49:36 -07001263/*
1264 * Slab allocation and freeing
1265 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001266static inline struct page *alloc_slab_page(gfp_t flags, int node,
1267 struct kmem_cache_order_objects oo)
1268{
1269 int order = oo_order(oo);
1270
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001271 flags |= __GFP_NOTRACK;
1272
Christoph Lameter2154a332010-07-09 14:07:10 -05001273 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001274 return alloc_pages(flags, order);
1275 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001276 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001277}
1278
Christoph Lameter81819f02007-05-06 14:49:36 -07001279static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1280{
Pekka Enberg06428782008-01-07 23:20:27 -08001281 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001282 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001283 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001284
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001285 flags &= gfp_allowed_mask;
1286
1287 if (flags & __GFP_WAIT)
1288 local_irq_enable();
1289
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001290 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001291
Pekka Enbergba522702009-06-24 21:59:51 +03001292 /*
1293 * Let the initial higher-order allocation fail under memory pressure
1294 * so we fall-back to the minimum order allocation.
1295 */
1296 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1297
1298 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001299 if (unlikely(!page)) {
1300 oo = s->min;
1301 /*
1302 * Allocation may have failed due to fragmentation.
1303 * Try a lower order alloc if possible
1304 */
1305 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001306
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001307 if (page)
1308 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001309 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001310
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001311 if (flags & __GFP_WAIT)
1312 local_irq_disable();
1313
1314 if (!page)
1315 return NULL;
1316
Vegard Nossum5a896d92008-04-04 00:54:48 +02001317 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001318 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001319 int pages = 1 << oo_order(oo);
1320
1321 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1322
1323 /*
1324 * Objects from caches that have a constructor don't get
1325 * cleared when they're allocated, so we need to do it here.
1326 */
1327 if (s->ctor)
1328 kmemcheck_mark_uninitialized_pages(page, pages);
1329 else
1330 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001331 }
1332
Christoph Lameter834f3d12008-04-14 19:11:31 +03001333 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001334 mod_zone_page_state(page_zone(page),
1335 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1336 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001337 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001338
1339 return page;
1340}
1341
1342static void setup_object(struct kmem_cache *s, struct page *page,
1343 void *object)
1344{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001345 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001346 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001347 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001348}
1349
1350static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1351{
1352 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001353 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001354 void *last;
1355 void *p;
1356
Christoph Lameter6cb06222007-10-16 01:25:41 -07001357 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001358
Christoph Lameter6cb06222007-10-16 01:25:41 -07001359 page = allocate_slab(s,
1360 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001361 if (!page)
1362 goto out;
1363
Christoph Lameter205ab992008-04-14 19:11:40 +03001364 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001365 page->slab = s;
1366 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001367
1368 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001369
1370 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001371 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001372
1373 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001374 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001375 setup_object(s, page, last);
1376 set_freepointer(s, last, p);
1377 last = p;
1378 }
1379 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001380 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001381
1382 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001383 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001384 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001385out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001386 return page;
1387}
1388
1389static void __free_slab(struct kmem_cache *s, struct page *page)
1390{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001391 int order = compound_order(page);
1392 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001393
Christoph Lameteraf537b02010-07-09 14:07:14 -05001394 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001395 void *p;
1396
1397 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001398 for_each_object(p, s, page_address(page),
1399 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001400 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001401 }
1402
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001403 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001404
Christoph Lameter81819f02007-05-06 14:49:36 -07001405 mod_zone_page_state(page_zone(page),
1406 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1407 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001408 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001409
Christoph Lameter49bd5222008-04-14 18:52:18 +03001410 __ClearPageSlab(page);
1411 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001412 if (current->reclaim_state)
1413 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001414 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001415}
1416
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001417#define need_reserve_slab_rcu \
1418 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1419
Christoph Lameter81819f02007-05-06 14:49:36 -07001420static void rcu_free_slab(struct rcu_head *h)
1421{
1422 struct page *page;
1423
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001424 if (need_reserve_slab_rcu)
1425 page = virt_to_head_page(h);
1426 else
1427 page = container_of((struct list_head *)h, struct page, lru);
1428
Christoph Lameter81819f02007-05-06 14:49:36 -07001429 __free_slab(page->slab, page);
1430}
1431
1432static void free_slab(struct kmem_cache *s, struct page *page)
1433{
1434 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001435 struct rcu_head *head;
1436
1437 if (need_reserve_slab_rcu) {
1438 int order = compound_order(page);
1439 int offset = (PAGE_SIZE << order) - s->reserved;
1440
1441 VM_BUG_ON(s->reserved != sizeof(*head));
1442 head = page_address(page) + offset;
1443 } else {
1444 /*
1445 * RCU free overloads the RCU head over the LRU
1446 */
1447 head = (void *)&page->lru;
1448 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001449
1450 call_rcu(head, rcu_free_slab);
1451 } else
1452 __free_slab(s, page);
1453}
1454
1455static void discard_slab(struct kmem_cache *s, struct page *page)
1456{
Christoph Lameter205ab992008-04-14 19:11:40 +03001457 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001458 free_slab(s, page);
1459}
1460
1461/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001462 * Management of partially allocated slabs.
1463 *
1464 * list_lock must be held.
Christoph Lameter81819f02007-05-06 14:49:36 -07001465 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001466static inline void add_partial(struct kmem_cache_node *n,
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001467 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001468{
Christoph Lametere95eed52007-05-06 14:49:44 -07001469 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001470 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001471 list_add_tail(&page->lru, &n->partial);
1472 else
1473 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001474}
1475
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001476/*
1477 * list_lock must be held.
1478 */
1479static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001480 struct page *page)
1481{
1482 list_del(&page->lru);
1483 n->nr_partial--;
1484}
1485
Christoph Lameter81819f02007-05-06 14:49:36 -07001486/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001487 * Lock slab, remove from the partial list and put the object into the
1488 * per cpu freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001489 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001490 * Returns a list of objects or NULL if it fails.
1491 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001492 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001493 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001494static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001495 struct kmem_cache_node *n, struct page *page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001496 int mode)
Christoph Lameter81819f02007-05-06 14:49:36 -07001497{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001498 void *freelist;
1499 unsigned long counters;
1500 struct page new;
1501
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001502 /*
1503 * Zap the freelist and set the frozen bit.
1504 * The old freelist is the list of objects for the
1505 * per cpu allocation list.
1506 */
1507 do {
1508 freelist = page->freelist;
1509 counters = page->counters;
1510 new.counters = counters;
Christoph Lameter49e22582011-08-09 16:12:27 -05001511 if (mode)
1512 new.inuse = page->objects;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001513
1514 VM_BUG_ON(new.frozen);
1515 new.frozen = 1;
1516
Christoph Lameter1d071712011-07-14 12:49:12 -05001517 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001518 freelist, counters,
1519 NULL, new.counters,
1520 "lock and freeze"));
1521
1522 remove_partial(n, page);
Christoph Lameter49e22582011-08-09 16:12:27 -05001523 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001524}
1525
Christoph Lameter49e22582011-08-09 16:12:27 -05001526static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1527
Christoph Lameter81819f02007-05-06 14:49:36 -07001528/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001529 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001530 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001531static void *get_partial_node(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001532 struct kmem_cache_node *n, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001533{
Christoph Lameter49e22582011-08-09 16:12:27 -05001534 struct page *page, *page2;
1535 void *object = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001536
1537 /*
1538 * Racy check. If we mistakenly see no partial slabs then we
1539 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001540 * partial slab and there is none available then get_partials()
1541 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001542 */
1543 if (!n || !n->nr_partial)
1544 return NULL;
1545
1546 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001547 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Alex,Shi12d79632011-09-07 10:26:36 +08001548 void *t = acquire_slab(s, n, page, object == NULL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001549 int available;
1550
1551 if (!t)
1552 break;
1553
Alex,Shi12d79632011-09-07 10:26:36 +08001554 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001555 c->page = page;
1556 c->node = page_to_nid(page);
1557 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001558 object = t;
1559 available = page->objects - page->inuse;
1560 } else {
1561 page->freelist = t;
1562 available = put_cpu_partial(s, page, 0);
1563 }
1564 if (kmem_cache_debug(s) || available > s->cpu_partial / 2)
1565 break;
1566
Christoph Lameter497b66f2011-08-09 16:12:26 -05001567 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001568 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001569 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001570}
1571
1572/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001573 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001574 */
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001575static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags,
1576 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001577{
1578#ifdef CONFIG_NUMA
1579 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001580 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001581 struct zone *zone;
1582 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001583 void *object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001584
1585 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001586 * The defrag ratio allows a configuration of the tradeoffs between
1587 * inter node defragmentation and node local allocations. A lower
1588 * defrag_ratio increases the tendency to do local allocations
1589 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001590 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001591 * If the defrag_ratio is set to 0 then kmalloc() always
1592 * returns node local objects. If the ratio is higher then kmalloc()
1593 * may return off node objects because partial slabs are obtained
1594 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001595 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001596 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001597 * defrag_ratio = 1000) then every (well almost) allocation will
1598 * first attempt to defrag slab caches on other nodes. This means
1599 * scanning over all nodes to look for partial slabs which may be
1600 * expensive if we do it every time we are trying to find a slab
1601 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001602 */
Christoph Lameter98246012008-01-07 23:20:26 -08001603 if (!s->remote_node_defrag_ratio ||
1604 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001605 return NULL;
1606
Miao Xiec0ff7452010-05-24 14:32:08 -07001607 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001608 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001609 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001610 struct kmem_cache_node *n;
1611
Mel Gorman54a6eb52008-04-28 02:12:16 -07001612 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001613
Mel Gorman54a6eb52008-04-28 02:12:16 -07001614 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001615 n->nr_partial > s->min_partial) {
Christoph Lameter497b66f2011-08-09 16:12:26 -05001616 object = get_partial_node(s, n, c);
1617 if (object) {
Miao Xiec0ff7452010-05-24 14:32:08 -07001618 put_mems_allowed();
Christoph Lameter497b66f2011-08-09 16:12:26 -05001619 return object;
Miao Xiec0ff7452010-05-24 14:32:08 -07001620 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001621 }
1622 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001623 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001624#endif
1625 return NULL;
1626}
1627
1628/*
1629 * Get a partial page, lock it and return it.
1630 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001631static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001632 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001633{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001634 void *object;
Christoph Lameter2154a332010-07-09 14:07:10 -05001635 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001636
Christoph Lameter497b66f2011-08-09 16:12:26 -05001637 object = get_partial_node(s, get_node(s, searchnode), c);
1638 if (object || node != NUMA_NO_NODE)
1639 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001640
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001641 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001642}
1643
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001644#ifdef CONFIG_PREEMPT
1645/*
1646 * Calculate the next globally unique transaction for disambiguiation
1647 * during cmpxchg. The transactions start with the cpu number and are then
1648 * incremented by CONFIG_NR_CPUS.
1649 */
1650#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1651#else
1652/*
1653 * No preemption supported therefore also no need to check for
1654 * different cpus.
1655 */
1656#define TID_STEP 1
1657#endif
1658
1659static inline unsigned long next_tid(unsigned long tid)
1660{
1661 return tid + TID_STEP;
1662}
1663
1664static inline unsigned int tid_to_cpu(unsigned long tid)
1665{
1666 return tid % TID_STEP;
1667}
1668
1669static inline unsigned long tid_to_event(unsigned long tid)
1670{
1671 return tid / TID_STEP;
1672}
1673
1674static inline unsigned int init_tid(int cpu)
1675{
1676 return cpu;
1677}
1678
1679static inline void note_cmpxchg_failure(const char *n,
1680 const struct kmem_cache *s, unsigned long tid)
1681{
1682#ifdef SLUB_DEBUG_CMPXCHG
1683 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1684
1685 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1686
1687#ifdef CONFIG_PREEMPT
1688 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1689 printk("due to cpu change %d -> %d\n",
1690 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1691 else
1692#endif
1693 if (tid_to_event(tid) != tid_to_event(actual_tid))
1694 printk("due to cpu running other code. Event %ld->%ld\n",
1695 tid_to_event(tid), tid_to_event(actual_tid));
1696 else
1697 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1698 actual_tid, tid, next_tid(tid));
1699#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001700 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001701}
1702
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001703void init_kmem_cache_cpus(struct kmem_cache *s)
1704{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001705 int cpu;
1706
1707 for_each_possible_cpu(cpu)
1708 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001709}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001710
1711/*
1712 * Remove the cpu slab
1713 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001714static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001715{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001716 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001717 struct page *page = c->page;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001718 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1719 int lock = 0;
1720 enum slab_modes l = M_NONE, m = M_NONE;
1721 void *freelist;
1722 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001723 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001724 struct page new;
1725 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001726
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001727 if (page->freelist) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001728 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001729 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001730 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001731
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001732 c->tid = next_tid(c->tid);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001733 c->page = NULL;
1734 freelist = c->freelist;
1735 c->freelist = NULL;
1736
1737 /*
1738 * Stage one: Free all available per cpu objects back
1739 * to the page freelist while it is still frozen. Leave the
1740 * last one.
1741 *
1742 * There is no need to take the list->lock because the page
1743 * is still frozen.
1744 */
1745 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1746 void *prior;
1747 unsigned long counters;
1748
1749 do {
1750 prior = page->freelist;
1751 counters = page->counters;
1752 set_freepointer(s, freelist, prior);
1753 new.counters = counters;
1754 new.inuse--;
1755 VM_BUG_ON(!new.frozen);
1756
Christoph Lameter1d071712011-07-14 12:49:12 -05001757 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001758 prior, counters,
1759 freelist, new.counters,
1760 "drain percpu freelist"));
1761
1762 freelist = nextfree;
1763 }
1764
1765 /*
1766 * Stage two: Ensure that the page is unfrozen while the
1767 * list presence reflects the actual number of objects
1768 * during unfreeze.
1769 *
1770 * We setup the list membership and then perform a cmpxchg
1771 * with the count. If there is a mismatch then the page
1772 * is not unfrozen but the page is on the wrong list.
1773 *
1774 * Then we restart the process which may have to remove
1775 * the page from the list that we just put it on again
1776 * because the number of objects in the slab may have
1777 * changed.
1778 */
1779redo:
1780
1781 old.freelist = page->freelist;
1782 old.counters = page->counters;
1783 VM_BUG_ON(!old.frozen);
1784
1785 /* Determine target state of the slab */
1786 new.counters = old.counters;
1787 if (freelist) {
1788 new.inuse--;
1789 set_freepointer(s, freelist, old.freelist);
1790 new.freelist = freelist;
1791 } else
1792 new.freelist = old.freelist;
1793
1794 new.frozen = 0;
1795
Christoph Lameter81107182011-08-09 13:01:32 -05001796 if (!new.inuse && n->nr_partial > s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001797 m = M_FREE;
1798 else if (new.freelist) {
1799 m = M_PARTIAL;
1800 if (!lock) {
1801 lock = 1;
1802 /*
1803 * Taking the spinlock removes the possiblity
1804 * that acquire_slab() will see a slab page that
1805 * is frozen
1806 */
1807 spin_lock(&n->list_lock);
1808 }
1809 } else {
1810 m = M_FULL;
1811 if (kmem_cache_debug(s) && !lock) {
1812 lock = 1;
1813 /*
1814 * This also ensures that the scanning of full
1815 * slabs from diagnostic functions will not see
1816 * any frozen slabs.
1817 */
1818 spin_lock(&n->list_lock);
1819 }
1820 }
1821
1822 if (l != m) {
1823
1824 if (l == M_PARTIAL)
1825
1826 remove_partial(n, page);
1827
1828 else if (l == M_FULL)
1829
1830 remove_full(s, page);
1831
1832 if (m == M_PARTIAL) {
1833
1834 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001835 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001836
1837 } else if (m == M_FULL) {
1838
1839 stat(s, DEACTIVATE_FULL);
1840 add_full(s, n, page);
1841
1842 }
1843 }
1844
1845 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001846 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001847 old.freelist, old.counters,
1848 new.freelist, new.counters,
1849 "unfreezing slab"))
1850 goto redo;
1851
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001852 if (lock)
1853 spin_unlock(&n->list_lock);
1854
1855 if (m == M_FREE) {
1856 stat(s, DEACTIVATE_EMPTY);
1857 discard_slab(s, page);
1858 stat(s, FREE_SLAB);
1859 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001860}
1861
Christoph Lameter49e22582011-08-09 16:12:27 -05001862/* Unfreeze all the cpu partial slabs */
1863static void unfreeze_partials(struct kmem_cache *s)
1864{
1865 struct kmem_cache_node *n = NULL;
1866 struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab);
Shaohua Li9ada1932011-11-14 13:34:13 +08001867 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001868
1869 while ((page = c->partial)) {
1870 enum slab_modes { M_PARTIAL, M_FREE };
1871 enum slab_modes l, m;
1872 struct page new;
1873 struct page old;
1874
1875 c->partial = page->next;
1876 l = M_FREE;
1877
1878 do {
1879
1880 old.freelist = page->freelist;
1881 old.counters = page->counters;
1882 VM_BUG_ON(!old.frozen);
1883
1884 new.counters = old.counters;
1885 new.freelist = old.freelist;
1886
1887 new.frozen = 0;
1888
Alex Shidcc3be62011-09-06 14:46:01 +08001889 if (!new.inuse && (!n || n->nr_partial > s->min_partial))
Christoph Lameter49e22582011-08-09 16:12:27 -05001890 m = M_FREE;
1891 else {
1892 struct kmem_cache_node *n2 = get_node(s,
1893 page_to_nid(page));
1894
1895 m = M_PARTIAL;
1896 if (n != n2) {
1897 if (n)
1898 spin_unlock(&n->list_lock);
1899
1900 n = n2;
1901 spin_lock(&n->list_lock);
1902 }
1903 }
1904
1905 if (l != m) {
Shaohua Li4c493a52011-11-11 14:54:14 +08001906 if (l == M_PARTIAL) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001907 remove_partial(n, page);
Shaohua Li4c493a52011-11-11 14:54:14 +08001908 stat(s, FREE_REMOVE_PARTIAL);
1909 } else {
Shaohua Lif64ae042011-11-11 08:33:48 +08001910 add_partial(n, page,
1911 DEACTIVATE_TO_TAIL);
Shaohua Li4c493a52011-11-11 14:54:14 +08001912 stat(s, FREE_ADD_PARTIAL);
1913 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001914
1915 l = m;
1916 }
1917
1918 } while (!cmpxchg_double_slab(s, page,
1919 old.freelist, old.counters,
1920 new.freelist, new.counters,
1921 "unfreezing slab"));
1922
1923 if (m == M_FREE) {
Shaohua Li9ada1932011-11-14 13:34:13 +08001924 page->next = discard_page;
1925 discard_page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001926 }
1927 }
1928
1929 if (n)
1930 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08001931
1932 while (discard_page) {
1933 page = discard_page;
1934 discard_page = discard_page->next;
1935
1936 stat(s, DEACTIVATE_EMPTY);
1937 discard_slab(s, page);
1938 stat(s, FREE_SLAB);
1939 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001940}
1941
1942/*
1943 * Put a page that was just frozen (in __slab_free) into a partial page
1944 * slot if available. This is done without interrupts disabled and without
1945 * preemption disabled. The cmpxchg is racy and may put the partial page
1946 * onto a random cpus partial slot.
1947 *
1948 * If we did not find a slot then simply move all the partials to the
1949 * per node partial list.
1950 */
1951int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
1952{
1953 struct page *oldpage;
1954 int pages;
1955 int pobjects;
1956
1957 do {
1958 pages = 0;
1959 pobjects = 0;
1960 oldpage = this_cpu_read(s->cpu_slab->partial);
1961
1962 if (oldpage) {
1963 pobjects = oldpage->pobjects;
1964 pages = oldpage->pages;
1965 if (drain && pobjects > s->cpu_partial) {
1966 unsigned long flags;
1967 /*
1968 * partial array is full. Move the existing
1969 * set to the per node partial list.
1970 */
1971 local_irq_save(flags);
1972 unfreeze_partials(s);
1973 local_irq_restore(flags);
1974 pobjects = 0;
1975 pages = 0;
1976 }
1977 }
1978
1979 pages++;
1980 pobjects += page->objects - page->inuse;
1981
1982 page->pages = pages;
1983 page->pobjects = pobjects;
1984 page->next = oldpage;
1985
Christoph Lameter933393f2011-12-22 11:58:51 -06001986 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage);
Christoph Lameter49e22582011-08-09 16:12:27 -05001987 stat(s, CPU_PARTIAL_FREE);
1988 return pobjects;
1989}
1990
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001991static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001992{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001993 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001994 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001995}
1996
1997/*
1998 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001999 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002000 * Called from IPI handler with interrupts disabled.
2001 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002002static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002003{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002004 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002005
Christoph Lameter49e22582011-08-09 16:12:27 -05002006 if (likely(c)) {
2007 if (c->page)
2008 flush_slab(s, c);
2009
2010 unfreeze_partials(s);
2011 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002012}
2013
2014static void flush_cpu_slab(void *d)
2015{
2016 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002017
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002018 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002019}
2020
2021static void flush_all(struct kmem_cache *s)
2022{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002023 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002024}
2025
2026/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002027 * Check if the objects in a per cpu structure fit numa
2028 * locality expectations.
2029 */
2030static inline int node_match(struct kmem_cache_cpu *c, int node)
2031{
2032#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05002033 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002034 return 0;
2035#endif
2036 return 1;
2037}
2038
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002039static int count_free(struct page *page)
2040{
2041 return page->objects - page->inuse;
2042}
2043
2044static unsigned long count_partial(struct kmem_cache_node *n,
2045 int (*get_count)(struct page *))
2046{
2047 unsigned long flags;
2048 unsigned long x = 0;
2049 struct page *page;
2050
2051 spin_lock_irqsave(&n->list_lock, flags);
2052 list_for_each_entry(page, &n->partial, lru)
2053 x += get_count(page);
2054 spin_unlock_irqrestore(&n->list_lock, flags);
2055 return x;
2056}
2057
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002058static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2059{
2060#ifdef CONFIG_SLUB_DEBUG
2061 return atomic_long_read(&n->total_objects);
2062#else
2063 return 0;
2064#endif
2065}
2066
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002067static noinline void
2068slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2069{
2070 int node;
2071
2072 printk(KERN_WARNING
2073 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2074 nid, gfpflags);
2075 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
2076 "default order: %d, min order: %d\n", s->name, s->objsize,
2077 s->size, oo_order(s->oo), oo_order(s->min));
2078
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002079 if (oo_order(s->min) > get_order(s->objsize))
2080 printk(KERN_WARNING " %s debugging increased min order, use "
2081 "slub_debug=O to disable.\n", s->name);
2082
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002083 for_each_online_node(node) {
2084 struct kmem_cache_node *n = get_node(s, node);
2085 unsigned long nr_slabs;
2086 unsigned long nr_objs;
2087 unsigned long nr_free;
2088
2089 if (!n)
2090 continue;
2091
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002092 nr_free = count_partial(n, count_free);
2093 nr_slabs = node_nr_slabs(n);
2094 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002095
2096 printk(KERN_WARNING
2097 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2098 node, nr_slabs, nr_objs, nr_free);
2099 }
2100}
2101
Christoph Lameter497b66f2011-08-09 16:12:26 -05002102static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2103 int node, struct kmem_cache_cpu **pc)
2104{
2105 void *object;
2106 struct kmem_cache_cpu *c;
2107 struct page *page = new_slab(s, flags, node);
2108
2109 if (page) {
2110 c = __this_cpu_ptr(s->cpu_slab);
2111 if (c->page)
2112 flush_slab(s, c);
2113
2114 /*
2115 * No other reference to the page yet so we can
2116 * muck around with it freely without cmpxchg
2117 */
2118 object = page->freelist;
2119 page->freelist = NULL;
2120
2121 stat(s, ALLOC_SLAB);
2122 c->node = page_to_nid(page);
2123 c->page = page;
2124 *pc = c;
2125 } else
2126 object = NULL;
2127
2128 return object;
2129}
2130
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002131/*
Christoph Lameter213eeb92011-11-11 14:07:14 -06002132 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2133 * or deactivate the page.
2134 *
2135 * The page is still frozen if the return value is not NULL.
2136 *
2137 * If this function returns NULL then the page has been unfrozen.
2138 */
2139static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2140{
2141 struct page new;
2142 unsigned long counters;
2143 void *freelist;
2144
2145 do {
2146 freelist = page->freelist;
2147 counters = page->counters;
2148 new.counters = counters;
2149 VM_BUG_ON(!new.frozen);
2150
2151 new.inuse = page->objects;
2152 new.frozen = freelist != NULL;
2153
2154 } while (!cmpxchg_double_slab(s, page,
2155 freelist, counters,
2156 NULL, new.counters,
2157 "get_freelist"));
2158
2159 return freelist;
2160}
2161
2162/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002163 * Slow path. The lockless freelist is empty or we need to perform
2164 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002165 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002166 * Processing is still very fast if new objects have been freed to the
2167 * regular freelist. In that case we simply take over the regular freelist
2168 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002169 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002170 * If that is not working then we fall back to the partial lists. We take the
2171 * first element of the freelist as the object to allocate now and move the
2172 * rest of the freelist to the lockless freelist.
2173 *
2174 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002175 * we need to allocate a new slab. This is the slowest path since it involves
2176 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002177 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002178static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2179 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002180{
Christoph Lameter81819f02007-05-06 14:49:36 -07002181 void **object;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002182 unsigned long flags;
2183
2184 local_irq_save(flags);
2185#ifdef CONFIG_PREEMPT
2186 /*
2187 * We may have been preempted and rescheduled on a different
2188 * cpu before disabling interrupts. Need to reload cpu area
2189 * pointer.
2190 */
2191 c = this_cpu_ptr(s->cpu_slab);
2192#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002193
Christoph Lameter497b66f2011-08-09 16:12:26 -05002194 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002195 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002196redo:
Christoph Lameterfc59c052011-06-01 12:25:56 -05002197 if (unlikely(!node_match(c, node))) {
Christoph Lametere36a2652011-06-01 12:25:57 -05002198 stat(s, ALLOC_NODE_MISMATCH);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002199 deactivate_slab(s, c);
2200 goto new_slab;
2201 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002202
Eric Dumazet73736e02011-12-13 04:57:06 +01002203 /* must check again c->freelist in case of cpu migration or IRQ */
2204 object = c->freelist;
2205 if (object)
2206 goto load_freelist;
2207
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002208 stat(s, ALLOC_SLOWPATH);
2209
Christoph Lameter213eeb92011-11-11 14:07:14 -06002210 object = get_freelist(s, c->page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002211
Christoph Lameter49e22582011-08-09 16:12:27 -05002212 if (!object) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002213 c->page = NULL;
2214 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002215 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002216 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002217
Christoph Lameter81819f02007-05-06 14:49:36 -07002218 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002219
Christoph Lameter894b8782007-05-10 03:15:16 -07002220load_freelist:
Christoph Lameterff120592009-12-18 16:26:22 -06002221 c->freelist = get_freepointer(s, object);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002222 c->tid = next_tid(c->tid);
2223 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002224 return object;
2225
Christoph Lameter81819f02007-05-06 14:49:36 -07002226new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002227
Christoph Lameter49e22582011-08-09 16:12:27 -05002228 if (c->partial) {
2229 c->page = c->partial;
2230 c->partial = c->page->next;
2231 c->node = page_to_nid(c->page);
2232 stat(s, CPU_PARTIAL_ALLOC);
2233 c->freelist = NULL;
2234 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002235 }
2236
Christoph Lameter49e22582011-08-09 16:12:27 -05002237 /* Then do expensive stuff like retrieving pages from the partial lists */
Christoph Lameter497b66f2011-08-09 16:12:26 -05002238 object = get_partial(s, gfpflags, node, c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002239
Christoph Lameter497b66f2011-08-09 16:12:26 -05002240 if (unlikely(!object)) {
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002241
Christoph Lameter497b66f2011-08-09 16:12:26 -05002242 object = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002243
Christoph Lameter497b66f2011-08-09 16:12:26 -05002244 if (unlikely(!object)) {
2245 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
2246 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter9e577e82011-07-22 09:35:14 -05002247
Christoph Lameter497b66f2011-08-09 16:12:26 -05002248 local_irq_restore(flags);
2249 return NULL;
2250 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002251 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002252
Christoph Lameter497b66f2011-08-09 16:12:26 -05002253 if (likely(!kmem_cache_debug(s)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002254 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002255
Christoph Lameter497b66f2011-08-09 16:12:26 -05002256 /* Only entered in the debug case */
2257 if (!alloc_debug_processing(s, c->page, object, addr))
2258 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002259
2260 c->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05002261 deactivate_slab(s, c);
Pekka Enberg15b7c512010-10-02 11:32:32 +03002262 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05002263 local_irq_restore(flags);
2264 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07002265}
2266
2267/*
2268 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2269 * have the fastpath folded into their functions. So no function call
2270 * overhead for requests that can be satisfied on the fastpath.
2271 *
2272 * The fastpath works by first checking if the lockless freelist can be used.
2273 * If not then __slab_alloc is called for slow processing.
2274 *
2275 * Otherwise we can simply pick the next object from the lockless free list.
2276 */
Pekka Enberg06428782008-01-07 23:20:27 -08002277static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002278 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002279{
Christoph Lameter894b8782007-05-10 03:15:16 -07002280 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002281 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002282 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002283
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002284 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09002285 return NULL;
2286
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002287redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002288
2289 /*
2290 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2291 * enabled. We may switch back and forth between cpus while
2292 * reading from one cpu area. That does not matter as long
2293 * as we end up on the original cpu again when doing the cmpxchg.
2294 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002295 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002296
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002297 /*
2298 * The transaction ids are globally unique per cpu and per operation on
2299 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2300 * occurs on the right processor and that there was no operation on the
2301 * linked list in between.
2302 */
2303 tid = c->tid;
2304 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002305
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002306 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002307 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07002308
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002309 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07002310
2311 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002312 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002313 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002314 * operation and if we are on the right processor.
2315 *
2316 * The cmpxchg does the following atomically (without lock semantics!)
2317 * 1. Relocate first pointer to the current per cpu area.
2318 * 2. Verify that tid and freelist have not been changed
2319 * 3. If they were not changed replace tid and freelist
2320 *
2321 * Since this is without lock semantics the protection is only against
2322 * code executing on this cpu *not* from access by other cpus.
2323 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002324 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002325 s->cpu_slab->freelist, s->cpu_slab->tid,
2326 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05002327 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002328
2329 note_cmpxchg_failure("slab_alloc", s, tid);
2330 goto redo;
2331 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06002332 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002333 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002334
Pekka Enberg74e21342009-11-25 20:14:48 +02002335 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06002336 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002337
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002338 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002339
Christoph Lameter894b8782007-05-10 03:15:16 -07002340 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002341}
2342
2343void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2344{
Christoph Lameter2154a332010-07-09 14:07:10 -05002345 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002346
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002347 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002348
2349 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002350}
2351EXPORT_SYMBOL(kmem_cache_alloc);
2352
Li Zefan0f24f122009-12-11 15:45:30 +08002353#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002354void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002355{
Richard Kennedy4a923792010-10-21 10:29:19 +01002356 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
2357 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2358 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002359}
Richard Kennedy4a923792010-10-21 10:29:19 +01002360EXPORT_SYMBOL(kmem_cache_alloc_trace);
2361
2362void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2363{
2364 void *ret = kmalloc_order(size, flags, order);
2365 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2366 return ret;
2367}
2368EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002369#endif
2370
Christoph Lameter81819f02007-05-06 14:49:36 -07002371#ifdef CONFIG_NUMA
2372void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2373{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002374 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2375
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002376 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2377 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002378
2379 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002380}
2381EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002382
Li Zefan0f24f122009-12-11 15:45:30 +08002383#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002384void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002385 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002386 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002387{
Richard Kennedy4a923792010-10-21 10:29:19 +01002388 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2389
2390 trace_kmalloc_node(_RET_IP_, ret,
2391 size, s->size, gfpflags, node);
2392 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002393}
Richard Kennedy4a923792010-10-21 10:29:19 +01002394EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002395#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002396#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002397
Christoph Lameter81819f02007-05-06 14:49:36 -07002398/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002399 * Slow patch handling. This may still be called frequently since objects
2400 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002401 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002402 * So we still attempt to reduce cache line usage. Just take the slab
2403 * lock and free the item. If there is no additional partial page
2404 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002405 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002406static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002407 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002408{
2409 void *prior;
2410 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002411 int was_frozen;
2412 int inuse;
2413 struct page new;
2414 unsigned long counters;
2415 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002416 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002417
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002418 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002419
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002420 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002421 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002422
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002423 do {
2424 prior = page->freelist;
2425 counters = page->counters;
2426 set_freepointer(s, object, prior);
2427 new.counters = counters;
2428 was_frozen = new.frozen;
2429 new.inuse--;
2430 if ((!new.inuse || !prior) && !was_frozen && !n) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002431
2432 if (!kmem_cache_debug(s) && !prior)
2433
2434 /*
2435 * Slab was on no list before and will be partially empty
2436 * We can defer the list move and instead freeze it.
2437 */
2438 new.frozen = 1;
2439
2440 else { /* Needs to be taken off a list */
2441
2442 n = get_node(s, page_to_nid(page));
2443 /*
2444 * Speculatively acquire the list_lock.
2445 * If the cmpxchg does not succeed then we may
2446 * drop the list_lock without any processing.
2447 *
2448 * Otherwise the list_lock will synchronize with
2449 * other processors updating the list of slabs.
2450 */
2451 spin_lock_irqsave(&n->list_lock, flags);
2452
2453 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002454 }
2455 inuse = new.inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002456
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002457 } while (!cmpxchg_double_slab(s, page,
2458 prior, counters,
2459 object, new.counters,
2460 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002461
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002462 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002463
2464 /*
2465 * If we just froze the page then put it onto the
2466 * per cpu partial list.
2467 */
2468 if (new.frozen && !was_frozen)
2469 put_cpu_partial(s, page, 1);
2470
2471 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002472 * The list lock was not taken therefore no list
2473 * activity can be necessary.
2474 */
2475 if (was_frozen)
2476 stat(s, FREE_FROZEN);
Christoph Lameter80f08c12011-06-01 12:25:55 -05002477 return;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002478 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002479
2480 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002481 * was_frozen may have been set after we acquired the list_lock in
2482 * an earlier loop. So we need to check it here again.
Christoph Lameter81819f02007-05-06 14:49:36 -07002483 */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002484 if (was_frozen)
2485 stat(s, FREE_FROZEN);
2486 else {
2487 if (unlikely(!inuse && n->nr_partial > s->min_partial))
2488 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002489
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002490 /*
2491 * Objects left in the slab. If it was not on the partial list before
2492 * then add it.
2493 */
2494 if (unlikely(!prior)) {
2495 remove_full(s, page);
Shaohua Li136333d2011-08-24 08:57:52 +08002496 add_partial(n, page, DEACTIVATE_TO_TAIL);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002497 stat(s, FREE_ADD_PARTIAL);
2498 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002499 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002500 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002501 return;
2502
2503slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002504 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002505 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002506 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002507 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002508 remove_partial(n, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002509 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002510 } else
2511 /* Slab must be on the full list */
2512 remove_full(s, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002513
Christoph Lameter80f08c12011-06-01 12:25:55 -05002514 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002515 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002516 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002517}
2518
Christoph Lameter894b8782007-05-10 03:15:16 -07002519/*
2520 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2521 * can perform fastpath freeing without additional function calls.
2522 *
2523 * The fastpath is only possible if we are freeing to the current cpu slab
2524 * of this processor. This typically the case if we have just allocated
2525 * the item before.
2526 *
2527 * If fastpath is not possible then fall back to __slab_free where we deal
2528 * with all sorts of special processing.
2529 */
Pekka Enberg06428782008-01-07 23:20:27 -08002530static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002531 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002532{
2533 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002534 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002535 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002536
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002537 slab_free_hook(s, x);
2538
Christoph Lametera24c5a02011-03-15 12:45:21 -05002539redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002540 /*
2541 * Determine the currently cpus per cpu slab.
2542 * The cpu may change afterward. However that does not matter since
2543 * data is retrieved via this pointer. If we are on the same cpu
2544 * during the cmpxchg then the free will succedd.
2545 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002546 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002547
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002548 tid = c->tid;
2549 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002550
Christoph Lameter442b06b2011-05-17 16:29:31 -05002551 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002552 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002553
Christoph Lameter933393f2011-12-22 11:58:51 -06002554 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002555 s->cpu_slab->freelist, s->cpu_slab->tid,
2556 c->freelist, tid,
2557 object, next_tid(tid)))) {
2558
2559 note_cmpxchg_failure("slab_free", s, tid);
2560 goto redo;
2561 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06002562 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002563 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002564 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002565
Christoph Lameter894b8782007-05-10 03:15:16 -07002566}
2567
Christoph Lameter81819f02007-05-06 14:49:36 -07002568void kmem_cache_free(struct kmem_cache *s, void *x)
2569{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002570 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002571
Christoph Lameterb49af682007-05-06 14:49:41 -07002572 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002573
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002574 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002575
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002576 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002577}
2578EXPORT_SYMBOL(kmem_cache_free);
2579
Christoph Lameter81819f02007-05-06 14:49:36 -07002580/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002581 * Object placement in a slab is made very easy because we always start at
2582 * offset 0. If we tune the size of the object to the alignment then we can
2583 * get the required alignment by putting one properly sized object after
2584 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002585 *
2586 * Notice that the allocation order determines the sizes of the per cpu
2587 * caches. Each processor has always one slab available for allocations.
2588 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002589 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002590 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002591 */
2592
2593/*
2594 * Mininum / Maximum order of slab pages. This influences locking overhead
2595 * and slab fragmentation. A higher order reduces the number of partial slabs
2596 * and increases the number of allocations possible without having to
2597 * take the list_lock.
2598 */
2599static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002600static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002601static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002602
2603/*
2604 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002605 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002606 */
2607static int slub_nomerge;
2608
2609/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002610 * Calculate the order of allocation given an slab object size.
2611 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002612 * The order of allocation has significant impact on performance and other
2613 * system components. Generally order 0 allocations should be preferred since
2614 * order 0 does not cause fragmentation in the page allocator. Larger objects
2615 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002616 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002617 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002618 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002619 * In order to reach satisfactory performance we must ensure that a minimum
2620 * number of objects is in one slab. Otherwise we may generate too much
2621 * activity on the partial lists which requires taking the list_lock. This is
2622 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002623 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002624 * slub_max_order specifies the order where we begin to stop considering the
2625 * number of objects in a slab as critical. If we reach slub_max_order then
2626 * we try to keep the page order as low as possible. So we accept more waste
2627 * of space in favor of a small page order.
2628 *
2629 * Higher order allocations also allow the placement of more objects in a
2630 * slab and thereby reduce object handling overhead. If the user has
2631 * requested a higher mininum order then we start with that one instead of
2632 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002633 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002634static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002635 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002636{
2637 int order;
2638 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002639 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002640
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002641 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002642 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002643
Christoph Lameter6300ea72007-07-17 04:03:20 -07002644 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002645 fls(min_objects * size - 1) - PAGE_SHIFT);
2646 order <= max_order; order++) {
2647
Christoph Lameter81819f02007-05-06 14:49:36 -07002648 unsigned long slab_size = PAGE_SIZE << order;
2649
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002650 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002651 continue;
2652
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002653 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002654
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002655 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002656 break;
2657
2658 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002659
Christoph Lameter81819f02007-05-06 14:49:36 -07002660 return order;
2661}
2662
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002663static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002664{
2665 int order;
2666 int min_objects;
2667 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002668 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002669
2670 /*
2671 * Attempt to find best configuration for a slab. This
2672 * works by first attempting to generate a layout with
2673 * the best configuration and backing off gradually.
2674 *
2675 * First we reduce the acceptable waste in a slab. Then
2676 * we reduce the minimum objects required in a slab.
2677 */
2678 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002679 if (!min_objects)
2680 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002681 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002682 min_objects = min(min_objects, max_objects);
2683
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002684 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002685 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002686 while (fraction >= 4) {
2687 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002688 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002689 if (order <= slub_max_order)
2690 return order;
2691 fraction /= 2;
2692 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002693 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002694 }
2695
2696 /*
2697 * We were unable to place multiple objects in a slab. Now
2698 * lets see if we can place a single object there.
2699 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002700 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002701 if (order <= slub_max_order)
2702 return order;
2703
2704 /*
2705 * Doh this slab cannot be placed using slub_max_order.
2706 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002707 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002708 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002709 return order;
2710 return -ENOSYS;
2711}
2712
Christoph Lameter81819f02007-05-06 14:49:36 -07002713/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002714 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002715 */
2716static unsigned long calculate_alignment(unsigned long flags,
2717 unsigned long align, unsigned long size)
2718{
2719 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002720 * If the user wants hardware cache aligned objects then follow that
2721 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002722 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002723 * The hardware cache alignment cannot override the specified
2724 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002725 */
Nick Pigginb6210382008-03-05 14:05:56 -08002726 if (flags & SLAB_HWCACHE_ALIGN) {
2727 unsigned long ralign = cache_line_size();
2728 while (size <= ralign / 2)
2729 ralign /= 2;
2730 align = max(align, ralign);
2731 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002732
2733 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002734 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002735
2736 return ALIGN(align, sizeof(void *));
2737}
2738
Pekka Enberg5595cff2008-08-05 09:28:47 +03002739static void
2740init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002741{
2742 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002743 spin_lock_init(&n->list_lock);
2744 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002745#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002746 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002747 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002748 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002749#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002750}
2751
Christoph Lameter55136592010-08-20 12:37:13 -05002752static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002753{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002754 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2755 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002756
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002757 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002758 * Must align to double word boundary for the double cmpxchg
2759 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002760 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002761 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2762 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002763
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002764 if (!s->cpu_slab)
2765 return 0;
2766
2767 init_kmem_cache_cpus(s);
2768
2769 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002770}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002771
Christoph Lameter51df1142010-08-20 12:37:15 -05002772static struct kmem_cache *kmem_cache_node;
2773
Christoph Lameter81819f02007-05-06 14:49:36 -07002774/*
2775 * No kmalloc_node yet so do it by hand. We know that this is the first
2776 * slab on the node for this slabcache. There are no concurrent accesses
2777 * possible.
2778 *
2779 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002780 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2781 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002782 */
Christoph Lameter55136592010-08-20 12:37:13 -05002783static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002784{
2785 struct page *page;
2786 struct kmem_cache_node *n;
2787
Christoph Lameter51df1142010-08-20 12:37:15 -05002788 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002789
Christoph Lameter51df1142010-08-20 12:37:15 -05002790 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002791
2792 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002793 if (page_to_nid(page) != node) {
2794 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2795 "node %d\n", node);
2796 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2797 "in order to be able to continue\n");
2798 }
2799
Christoph Lameter81819f02007-05-06 14:49:36 -07002800 n = page->freelist;
2801 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002802 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05002803 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002804 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002805 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002806#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002807 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002808 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002809#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002810 init_kmem_cache_node(n, kmem_cache_node);
2811 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002812
Shaohua Li136333d2011-08-24 08:57:52 +08002813 add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002814}
2815
2816static void free_kmem_cache_nodes(struct kmem_cache *s)
2817{
2818 int node;
2819
Christoph Lameterf64dc582007-10-16 01:25:33 -07002820 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002821 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002822
Alexander Duyck73367bd2010-05-21 14:41:35 -07002823 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002824 kmem_cache_free(kmem_cache_node, n);
2825
Christoph Lameter81819f02007-05-06 14:49:36 -07002826 s->node[node] = NULL;
2827 }
2828}
2829
Christoph Lameter55136592010-08-20 12:37:13 -05002830static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002831{
2832 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002833
Christoph Lameterf64dc582007-10-16 01:25:33 -07002834 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002835 struct kmem_cache_node *n;
2836
Alexander Duyck73367bd2010-05-21 14:41:35 -07002837 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002838 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002839 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002840 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002841 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002842 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002843
2844 if (!n) {
2845 free_kmem_cache_nodes(s);
2846 return 0;
2847 }
2848
Christoph Lameter81819f02007-05-06 14:49:36 -07002849 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002850 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002851 }
2852 return 1;
2853}
Christoph Lameter81819f02007-05-06 14:49:36 -07002854
David Rientjesc0bdb232009-02-25 09:16:35 +02002855static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002856{
2857 if (min < MIN_PARTIAL)
2858 min = MIN_PARTIAL;
2859 else if (min > MAX_PARTIAL)
2860 min = MAX_PARTIAL;
2861 s->min_partial = min;
2862}
2863
Christoph Lameter81819f02007-05-06 14:49:36 -07002864/*
2865 * calculate_sizes() determines the order and the distribution of data within
2866 * a slab object.
2867 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002868static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002869{
2870 unsigned long flags = s->flags;
2871 unsigned long size = s->objsize;
2872 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002873 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002874
2875 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002876 * Round up object size to the next word boundary. We can only
2877 * place the free pointer at word boundaries and this determines
2878 * the possible location of the free pointer.
2879 */
2880 size = ALIGN(size, sizeof(void *));
2881
2882#ifdef CONFIG_SLUB_DEBUG
2883 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002884 * Determine if we can poison the object itself. If the user of
2885 * the slab may touch the object after free or before allocation
2886 * then we should never poison the object itself.
2887 */
2888 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002889 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002890 s->flags |= __OBJECT_POISON;
2891 else
2892 s->flags &= ~__OBJECT_POISON;
2893
Christoph Lameter81819f02007-05-06 14:49:36 -07002894
2895 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002896 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002897 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002898 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002899 */
2900 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2901 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002902#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002903
2904 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002905 * With that we have determined the number of bytes in actual use
2906 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002907 */
2908 s->inuse = size;
2909
2910 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002911 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002912 /*
2913 * Relocate free pointer after the object if it is not
2914 * permitted to overwrite the first word of the object on
2915 * kmem_cache_free.
2916 *
2917 * This is the case if we do RCU, have a constructor or
2918 * destructor or are poisoning the objects.
2919 */
2920 s->offset = size;
2921 size += sizeof(void *);
2922 }
2923
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002924#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002925 if (flags & SLAB_STORE_USER)
2926 /*
2927 * Need to store information about allocs and frees after
2928 * the object.
2929 */
2930 size += 2 * sizeof(struct track);
2931
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002932 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002933 /*
2934 * Add some empty padding so that we can catch
2935 * overwrites from earlier objects rather than let
2936 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002937 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002938 * of the object.
2939 */
2940 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002941#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002942
Christoph Lameter81819f02007-05-06 14:49:36 -07002943 /*
2944 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002945 * user specified and the dynamic determination of cache line size
2946 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002947 */
2948 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002949 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002950
2951 /*
2952 * SLUB stores one object immediately after another beginning from
2953 * offset 0. In order to align the objects we have to simply size
2954 * each object to conform to the alignment.
2955 */
2956 size = ALIGN(size, align);
2957 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002958 if (forced_order >= 0)
2959 order = forced_order;
2960 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002961 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002962
Christoph Lameter834f3d12008-04-14 19:11:31 +03002963 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002964 return 0;
2965
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002966 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002967 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002968 s->allocflags |= __GFP_COMP;
2969
2970 if (s->flags & SLAB_CACHE_DMA)
2971 s->allocflags |= SLUB_DMA;
2972
2973 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2974 s->allocflags |= __GFP_RECLAIMABLE;
2975
Christoph Lameter81819f02007-05-06 14:49:36 -07002976 /*
2977 * Determine the number of objects per slab
2978 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002979 s->oo = oo_make(order, size, s->reserved);
2980 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002981 if (oo_objects(s->oo) > oo_objects(s->max))
2982 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002983
Christoph Lameter834f3d12008-04-14 19:11:31 +03002984 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002985
2986}
2987
Christoph Lameter55136592010-08-20 12:37:13 -05002988static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002989 const char *name, size_t size,
2990 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002991 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002992{
2993 memset(s, 0, kmem_size);
2994 s->name = name;
2995 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002996 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002997 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002998 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002999 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003000
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003001 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3002 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003003
Christoph Lameter06b285d2008-04-14 19:11:41 +03003004 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003005 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003006 if (disable_higher_order_debug) {
3007 /*
3008 * Disable debugging flags that store metadata if the min slab
3009 * order increased.
3010 */
3011 if (get_order(s->size) > get_order(s->objsize)) {
3012 s->flags &= ~DEBUG_METADATA_FLAGS;
3013 s->offset = 0;
3014 if (!calculate_sizes(s, -1))
3015 goto error;
3016 }
3017 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003018
Heiko Carstens25654092012-01-12 17:17:33 -08003019#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3020 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003021 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
3022 /* Enable fast mode */
3023 s->flags |= __CMPXCHG_DOUBLE;
3024#endif
3025
David Rientjes3b89d7d2009-02-22 17:40:07 -08003026 /*
3027 * The larger the object size is, the more pages we want on the partial
3028 * list to avoid pounding the page allocator excessively.
3029 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003030 set_min_partial(s, ilog2(s->size) / 2);
3031
3032 /*
3033 * cpu_partial determined the maximum number of objects kept in the
3034 * per cpu partial lists of a processor.
3035 *
3036 * Per cpu partial lists mainly contain slabs that just have one
3037 * object freed. If they are used for allocation then they can be
3038 * filled up again with minimal effort. The slab will never hit the
3039 * per node partial lists and therefore no locking will be required.
3040 *
3041 * This setting also determines
3042 *
3043 * A) The number of objects from per cpu partial slabs dumped to the
3044 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003045 * B) The number of objects in cpu partial slabs to extract from the
Christoph Lameter49e22582011-08-09 16:12:27 -05003046 * per node list when we run out of per cpu objects. We only fetch 50%
3047 * to keep some capacity around for frees.
3048 */
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003049 if (kmem_cache_debug(s))
3050 s->cpu_partial = 0;
3051 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003052 s->cpu_partial = 2;
3053 else if (s->size >= 1024)
3054 s->cpu_partial = 6;
3055 else if (s->size >= 256)
3056 s->cpu_partial = 13;
3057 else
3058 s->cpu_partial = 30;
3059
Christoph Lameter81819f02007-05-06 14:49:36 -07003060 s->refcount = 1;
3061#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003062 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003063#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003064 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003065 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003066
Christoph Lameter55136592010-08-20 12:37:13 -05003067 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07003068 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06003069
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003070 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003071error:
3072 if (flags & SLAB_PANIC)
3073 panic("Cannot create slab %s size=%lu realsize=%u "
3074 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03003075 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07003076 s->offset, flags);
3077 return 0;
3078}
Christoph Lameter81819f02007-05-06 14:49:36 -07003079
3080/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003081 * Determine the size of a slab object
3082 */
3083unsigned int kmem_cache_size(struct kmem_cache *s)
3084{
3085 return s->objsize;
3086}
3087EXPORT_SYMBOL(kmem_cache_size);
3088
Christoph Lameter33b12c32008-04-25 12:22:43 -07003089static void list_slab_objects(struct kmem_cache *s, struct page *page,
3090 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003091{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003092#ifdef CONFIG_SLUB_DEBUG
3093 void *addr = page_address(page);
3094 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003095 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3096 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003097 if (!map)
3098 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07003099 slab_err(s, page, "%s", text);
3100 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003101
Christoph Lameter5f80b132011-04-15 14:48:13 -05003102 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003103 for_each_object(p, s, addr, page->objects) {
3104
3105 if (!test_bit(slab_index(p, s, addr), map)) {
3106 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
3107 p, p - addr);
3108 print_tracking(s, p);
3109 }
3110 }
3111 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003112 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003113#endif
3114}
3115
Christoph Lameter81819f02007-05-06 14:49:36 -07003116/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003117 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003118 * This is called from kmem_cache_close(). We must be the last thread
3119 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003120 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003121static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003122{
Christoph Lameter81819f02007-05-06 14:49:36 -07003123 struct page *page, *h;
3124
Christoph Lameter33b12c32008-04-25 12:22:43 -07003125 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003126 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003127 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003128 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003129 } else {
3130 list_slab_objects(s, page,
3131 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003132 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003133 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003134}
3135
3136/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003137 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003138 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003139static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003140{
3141 int node;
3142
3143 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003144 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003145 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07003146 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003147 struct kmem_cache_node *n = get_node(s, node);
3148
Christoph Lameter599870b2008-04-23 12:36:52 -07003149 free_partial(s, n);
3150 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003151 return 1;
3152 }
3153 free_kmem_cache_nodes(s);
3154 return 0;
3155}
3156
3157/*
3158 * Close a cache and release the kmem_cache structure
3159 * (must be used for caches created using kmem_cache_create)
3160 */
3161void kmem_cache_destroy(struct kmem_cache *s)
3162{
3163 down_write(&slub_lock);
3164 s->refcount--;
3165 if (!s->refcount) {
3166 list_del(&s->list);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003167 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03003168 if (kmem_cache_close(s)) {
3169 printk(KERN_ERR "SLUB %s: %s called for cache that "
3170 "still has objects.\n", s->name, __func__);
3171 dump_stack();
3172 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03003173 if (s->flags & SLAB_DESTROY_BY_RCU)
3174 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07003175 sysfs_slab_remove(s);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003176 } else
3177 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003178}
3179EXPORT_SYMBOL(kmem_cache_destroy);
3180
3181/********************************************************************
3182 * Kmalloc subsystem
3183 *******************************************************************/
3184
Christoph Lameter51df1142010-08-20 12:37:15 -05003185struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07003186EXPORT_SYMBOL(kmalloc_caches);
3187
Christoph Lameter51df1142010-08-20 12:37:15 -05003188static struct kmem_cache *kmem_cache;
3189
Christoph Lameter55136592010-08-20 12:37:13 -05003190#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003191static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05003192#endif
3193
Christoph Lameter81819f02007-05-06 14:49:36 -07003194static int __init setup_slub_min_order(char *str)
3195{
Pekka Enberg06428782008-01-07 23:20:27 -08003196 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003197
3198 return 1;
3199}
3200
3201__setup("slub_min_order=", setup_slub_min_order);
3202
3203static int __init setup_slub_max_order(char *str)
3204{
Pekka Enberg06428782008-01-07 23:20:27 -08003205 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003206 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003207
3208 return 1;
3209}
3210
3211__setup("slub_max_order=", setup_slub_max_order);
3212
3213static int __init setup_slub_min_objects(char *str)
3214{
Pekka Enberg06428782008-01-07 23:20:27 -08003215 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003216
3217 return 1;
3218}
3219
3220__setup("slub_min_objects=", setup_slub_min_objects);
3221
3222static int __init setup_slub_nomerge(char *str)
3223{
3224 slub_nomerge = 1;
3225 return 1;
3226}
3227
3228__setup("slub_nomerge", setup_slub_nomerge);
3229
Christoph Lameter51df1142010-08-20 12:37:15 -05003230static struct kmem_cache *__init create_kmalloc_cache(const char *name,
3231 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003232{
Christoph Lameter51df1142010-08-20 12:37:15 -05003233 struct kmem_cache *s;
3234
3235 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3236
Pekka Enberg83b519e2009-06-10 19:40:04 +03003237 /*
3238 * This function is called with IRQs disabled during early-boot on
3239 * single CPU so there's no need to take slub_lock here.
3240 */
Christoph Lameter55136592010-08-20 12:37:13 -05003241 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03003242 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07003243 goto panic;
3244
3245 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05003246 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003247
3248panic:
3249 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003250 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003251}
3252
Christoph Lameterf1b26332007-07-17 04:03:26 -07003253/*
3254 * Conversion table for small slabs sizes / 8 to the index in the
3255 * kmalloc array. This is necessary for slabs < 192 since we have non power
3256 * of two cache sizes there. The size of larger slabs can be determined using
3257 * fls.
3258 */
3259static s8 size_index[24] = {
3260 3, /* 8 */
3261 4, /* 16 */
3262 5, /* 24 */
3263 5, /* 32 */
3264 6, /* 40 */
3265 6, /* 48 */
3266 6, /* 56 */
3267 6, /* 64 */
3268 1, /* 72 */
3269 1, /* 80 */
3270 1, /* 88 */
3271 1, /* 96 */
3272 7, /* 104 */
3273 7, /* 112 */
3274 7, /* 120 */
3275 7, /* 128 */
3276 2, /* 136 */
3277 2, /* 144 */
3278 2, /* 152 */
3279 2, /* 160 */
3280 2, /* 168 */
3281 2, /* 176 */
3282 2, /* 184 */
3283 2 /* 192 */
3284};
3285
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003286static inline int size_index_elem(size_t bytes)
3287{
3288 return (bytes - 1) / 8;
3289}
3290
Christoph Lameter81819f02007-05-06 14:49:36 -07003291static struct kmem_cache *get_slab(size_t size, gfp_t flags)
3292{
Christoph Lameterf1b26332007-07-17 04:03:26 -07003293 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07003294
Christoph Lameterf1b26332007-07-17 04:03:26 -07003295 if (size <= 192) {
3296 if (!size)
3297 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07003298
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003299 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003300 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07003301 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003302
3303#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07003304 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05003305 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07003306
Christoph Lameter81819f02007-05-06 14:49:36 -07003307#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003308 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07003309}
3310
3311void *__kmalloc(size_t size, gfp_t flags)
3312{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003313 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003314 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003315
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003316 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003317 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003318
3319 s = get_slab(size, flags);
3320
3321 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003322 return s;
3323
Christoph Lameter2154a332010-07-09 14:07:10 -05003324 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003325
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003326 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003327
3328 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003329}
3330EXPORT_SYMBOL(__kmalloc);
3331
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003332#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003333static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3334{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003335 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003336 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003337
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003338 flags |= __GFP_COMP | __GFP_NOTRACK;
3339 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003340 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003341 ptr = page_address(page);
3342
3343 kmemleak_alloc(ptr, size, 1, flags);
3344 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003345}
3346
Christoph Lameter81819f02007-05-06 14:49:36 -07003347void *__kmalloc_node(size_t size, gfp_t flags, int node)
3348{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003349 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003350 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003351
Ingo Molnar057685c2009-02-20 12:15:30 +01003352 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003353 ret = kmalloc_large_node(size, flags, node);
3354
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003355 trace_kmalloc_node(_RET_IP_, ret,
3356 size, PAGE_SIZE << get_order(size),
3357 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003358
3359 return ret;
3360 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003361
3362 s = get_slab(size, flags);
3363
3364 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003365 return s;
3366
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003367 ret = slab_alloc(s, flags, node, _RET_IP_);
3368
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003369 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003370
3371 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003372}
3373EXPORT_SYMBOL(__kmalloc_node);
3374#endif
3375
3376size_t ksize(const void *object)
3377{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003378 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003379
Christoph Lameteref8b4522007-10-16 01:24:46 -07003380 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003381 return 0;
3382
Vegard Nossum294a80a2007-12-04 23:45:30 -08003383 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003384
Pekka Enberg76994412008-05-22 19:22:25 +03003385 if (unlikely(!PageSlab(page))) {
3386 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003387 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003388 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003389
Eric Dumazetb3d41882011-02-14 18:35:22 +01003390 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003391}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003392EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003393
Ben Greeard18a90d2011-07-07 11:36:37 -07003394#ifdef CONFIG_SLUB_DEBUG
3395bool verify_mem_not_deleted(const void *x)
3396{
3397 struct page *page;
3398 void *object = (void *)x;
3399 unsigned long flags;
3400 bool rv;
3401
3402 if (unlikely(ZERO_OR_NULL_PTR(x)))
3403 return false;
3404
3405 local_irq_save(flags);
3406
3407 page = virt_to_head_page(x);
3408 if (unlikely(!PageSlab(page))) {
3409 /* maybe it was from stack? */
3410 rv = true;
3411 goto out_unlock;
3412 }
3413
3414 slab_lock(page);
3415 if (on_freelist(page->slab, page, object)) {
3416 object_err(page->slab, page, object, "Object is on free-list");
3417 rv = false;
3418 } else {
3419 rv = true;
3420 }
3421 slab_unlock(page);
3422
3423out_unlock:
3424 local_irq_restore(flags);
3425 return rv;
3426}
3427EXPORT_SYMBOL(verify_mem_not_deleted);
3428#endif
3429
Christoph Lameter81819f02007-05-06 14:49:36 -07003430void kfree(const void *x)
3431{
Christoph Lameter81819f02007-05-06 14:49:36 -07003432 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003433 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003434
Pekka Enberg2121db72009-03-25 11:05:57 +02003435 trace_kfree(_RET_IP_, x);
3436
Satyam Sharma2408c552007-10-16 01:24:44 -07003437 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003438 return;
3439
Christoph Lameterb49af682007-05-06 14:49:41 -07003440 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003441 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003442 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003443 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003444 put_page(page);
3445 return;
3446 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003447 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003448}
3449EXPORT_SYMBOL(kfree);
3450
Christoph Lameter2086d262007-05-06 14:49:46 -07003451/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003452 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3453 * the remaining slabs by the number of items in use. The slabs with the
3454 * most items in use come first. New allocations will then fill those up
3455 * and thus they can be removed from the partial lists.
3456 *
3457 * The slabs with the least items are placed last. This results in them
3458 * being allocated from last increasing the chance that the last objects
3459 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003460 */
3461int kmem_cache_shrink(struct kmem_cache *s)
3462{
3463 int node;
3464 int i;
3465 struct kmem_cache_node *n;
3466 struct page *page;
3467 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003468 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003469 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003470 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003471 unsigned long flags;
3472
3473 if (!slabs_by_inuse)
3474 return -ENOMEM;
3475
3476 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003477 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003478 n = get_node(s, node);
3479
3480 if (!n->nr_partial)
3481 continue;
3482
Christoph Lameter834f3d12008-04-14 19:11:31 +03003483 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003484 INIT_LIST_HEAD(slabs_by_inuse + i);
3485
3486 spin_lock_irqsave(&n->list_lock, flags);
3487
3488 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003489 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003490 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003491 * Note that concurrent frees may occur while we hold the
3492 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003493 */
3494 list_for_each_entry_safe(page, t, &n->partial, lru) {
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003495 list_move(&page->lru, slabs_by_inuse + page->inuse);
3496 if (!page->inuse)
3497 n->nr_partial--;
Christoph Lameter2086d262007-05-06 14:49:46 -07003498 }
3499
Christoph Lameter2086d262007-05-06 14:49:46 -07003500 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003501 * Rebuild the partial list with the slabs filled up most
3502 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003503 */
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003504 for (i = objects - 1; i > 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003505 list_splice(slabs_by_inuse + i, n->partial.prev);
3506
Christoph Lameter2086d262007-05-06 14:49:46 -07003507 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003508
3509 /* Release empty slabs */
3510 list_for_each_entry_safe(page, t, slabs_by_inuse, lru)
3511 discard_slab(s, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003512 }
3513
3514 kfree(slabs_by_inuse);
3515 return 0;
3516}
3517EXPORT_SYMBOL(kmem_cache_shrink);
3518
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003519#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003520static int slab_mem_going_offline_callback(void *arg)
3521{
3522 struct kmem_cache *s;
3523
3524 down_read(&slub_lock);
3525 list_for_each_entry(s, &slab_caches, list)
3526 kmem_cache_shrink(s);
3527 up_read(&slub_lock);
3528
3529 return 0;
3530}
3531
3532static void slab_mem_offline_callback(void *arg)
3533{
3534 struct kmem_cache_node *n;
3535 struct kmem_cache *s;
3536 struct memory_notify *marg = arg;
3537 int offline_node;
3538
3539 offline_node = marg->status_change_nid;
3540
3541 /*
3542 * If the node still has available memory. we need kmem_cache_node
3543 * for it yet.
3544 */
3545 if (offline_node < 0)
3546 return;
3547
3548 down_read(&slub_lock);
3549 list_for_each_entry(s, &slab_caches, list) {
3550 n = get_node(s, offline_node);
3551 if (n) {
3552 /*
3553 * if n->nr_slabs > 0, slabs still exist on the node
3554 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003555 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003556 * callback. So, we must fail.
3557 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003558 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003559
3560 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003561 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003562 }
3563 }
3564 up_read(&slub_lock);
3565}
3566
3567static int slab_mem_going_online_callback(void *arg)
3568{
3569 struct kmem_cache_node *n;
3570 struct kmem_cache *s;
3571 struct memory_notify *marg = arg;
3572 int nid = marg->status_change_nid;
3573 int ret = 0;
3574
3575 /*
3576 * If the node's memory is already available, then kmem_cache_node is
3577 * already created. Nothing to do.
3578 */
3579 if (nid < 0)
3580 return 0;
3581
3582 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003583 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003584 * allocate a kmem_cache_node structure in order to bring the node
3585 * online.
3586 */
3587 down_read(&slub_lock);
3588 list_for_each_entry(s, &slab_caches, list) {
3589 /*
3590 * XXX: kmem_cache_alloc_node will fallback to other nodes
3591 * since memory is not yet available from the node that
3592 * is brought up.
3593 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003594 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003595 if (!n) {
3596 ret = -ENOMEM;
3597 goto out;
3598 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003599 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003600 s->node[nid] = n;
3601 }
3602out:
3603 up_read(&slub_lock);
3604 return ret;
3605}
3606
3607static int slab_memory_callback(struct notifier_block *self,
3608 unsigned long action, void *arg)
3609{
3610 int ret = 0;
3611
3612 switch (action) {
3613 case MEM_GOING_ONLINE:
3614 ret = slab_mem_going_online_callback(arg);
3615 break;
3616 case MEM_GOING_OFFLINE:
3617 ret = slab_mem_going_offline_callback(arg);
3618 break;
3619 case MEM_OFFLINE:
3620 case MEM_CANCEL_ONLINE:
3621 slab_mem_offline_callback(arg);
3622 break;
3623 case MEM_ONLINE:
3624 case MEM_CANCEL_OFFLINE:
3625 break;
3626 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003627 if (ret)
3628 ret = notifier_from_errno(ret);
3629 else
3630 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003631 return ret;
3632}
3633
3634#endif /* CONFIG_MEMORY_HOTPLUG */
3635
Christoph Lameter81819f02007-05-06 14:49:36 -07003636/********************************************************************
3637 * Basic setup of slabs
3638 *******************************************************************/
3639
Christoph Lameter51df1142010-08-20 12:37:15 -05003640/*
3641 * Used for early kmem_cache structures that were allocated using
3642 * the page allocator
3643 */
3644
3645static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3646{
3647 int node;
3648
3649 list_add(&s->list, &slab_caches);
3650 s->refcount = -1;
3651
3652 for_each_node_state(node, N_NORMAL_MEMORY) {
3653 struct kmem_cache_node *n = get_node(s, node);
3654 struct page *p;
3655
3656 if (n) {
3657 list_for_each_entry(p, &n->partial, lru)
3658 p->slab = s;
3659
Li Zefan607bf322011-04-12 15:22:26 +08003660#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003661 list_for_each_entry(p, &n->full, lru)
3662 p->slab = s;
3663#endif
3664 }
3665 }
3666}
3667
Christoph Lameter81819f02007-05-06 14:49:36 -07003668void __init kmem_cache_init(void)
3669{
3670 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003671 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003672 struct kmem_cache *temp_kmem_cache;
3673 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003674 struct kmem_cache *temp_kmem_cache_node;
3675 unsigned long kmalloc_size;
3676
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08003677 if (debug_guardpage_minorder())
3678 slub_max_order = 0;
3679
Christoph Lameter51df1142010-08-20 12:37:15 -05003680 kmem_size = offsetof(struct kmem_cache, node) +
3681 nr_node_ids * sizeof(struct kmem_cache_node *);
3682
3683 /* Allocate two kmem_caches from the page allocator */
3684 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3685 order = get_order(2 * kmalloc_size);
3686 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3687
Christoph Lameter81819f02007-05-06 14:49:36 -07003688 /*
3689 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003690 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003691 * kmem_cache_open for slab_state == DOWN.
3692 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003693 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3694
3695 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3696 sizeof(struct kmem_cache_node),
3697 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003698
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003699 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003700
3701 /* Able to allocate the per node structures */
3702 slab_state = PARTIAL;
3703
Christoph Lameter51df1142010-08-20 12:37:15 -05003704 temp_kmem_cache = kmem_cache;
3705 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3706 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3707 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3708 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003709
Christoph Lameter51df1142010-08-20 12:37:15 -05003710 /*
3711 * Allocate kmem_cache_node properly from the kmem_cache slab.
3712 * kmem_cache_node is separately allocated so no need to
3713 * update any list pointers.
3714 */
3715 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003716
Christoph Lameter51df1142010-08-20 12:37:15 -05003717 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3718 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3719
3720 kmem_cache_bootstrap_fixup(kmem_cache_node);
3721
3722 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003723 kmem_cache_bootstrap_fixup(kmem_cache);
3724 caches++;
3725 /* Free temporary boot structure */
3726 free_pages((unsigned long)temp_kmem_cache, order);
3727
3728 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003729
3730 /*
3731 * Patch up the size_index table if we have strange large alignment
3732 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003733 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003734 *
3735 * Largest permitted alignment is 256 bytes due to the way we
3736 * handle the index determination for the smaller caches.
3737 *
3738 * Make sure that nothing crazy happens if someone starts tinkering
3739 * around with ARCH_KMALLOC_MINALIGN
3740 */
3741 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3742 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3743
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003744 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3745 int elem = size_index_elem(i);
3746 if (elem >= ARRAY_SIZE(size_index))
3747 break;
3748 size_index[elem] = KMALLOC_SHIFT_LOW;
3749 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003750
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003751 if (KMALLOC_MIN_SIZE == 64) {
3752 /*
3753 * The 96 byte size cache is not used if the alignment
3754 * is 64 byte.
3755 */
3756 for (i = 64 + 8; i <= 96; i += 8)
3757 size_index[size_index_elem(i)] = 7;
3758 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003759 /*
3760 * The 192 byte sized cache is not used if the alignment
3761 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3762 * instead.
3763 */
3764 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003765 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003766 }
3767
Christoph Lameter51df1142010-08-20 12:37:15 -05003768 /* Caches that are not of the two-to-the-power-of size */
3769 if (KMALLOC_MIN_SIZE <= 32) {
3770 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3771 caches++;
3772 }
3773
3774 if (KMALLOC_MIN_SIZE <= 64) {
3775 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3776 caches++;
3777 }
3778
3779 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3780 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3781 caches++;
3782 }
3783
Christoph Lameter81819f02007-05-06 14:49:36 -07003784 slab_state = UP;
3785
3786 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003787 if (KMALLOC_MIN_SIZE <= 32) {
3788 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3789 BUG_ON(!kmalloc_caches[1]->name);
3790 }
3791
3792 if (KMALLOC_MIN_SIZE <= 64) {
3793 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3794 BUG_ON(!kmalloc_caches[2]->name);
3795 }
3796
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003797 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3798 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3799
3800 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003801 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003802 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003803
3804#ifdef CONFIG_SMP
3805 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003806#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003807
Christoph Lameter55136592010-08-20 12:37:13 -05003808#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003809 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3810 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003811
Christoph Lameter51df1142010-08-20 12:37:15 -05003812 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003813 char *name = kasprintf(GFP_NOWAIT,
3814 "dma-kmalloc-%d", s->objsize);
3815
3816 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003817 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3818 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003819 }
3820 }
3821#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003822 printk(KERN_INFO
3823 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003824 " CPUs=%d, Nodes=%d\n",
3825 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003826 slub_min_order, slub_max_order, slub_min_objects,
3827 nr_cpu_ids, nr_node_ids);
3828}
3829
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003830void __init kmem_cache_init_late(void)
3831{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003832}
3833
Christoph Lameter81819f02007-05-06 14:49:36 -07003834/*
3835 * Find a mergeable slab cache
3836 */
3837static int slab_unmergeable(struct kmem_cache *s)
3838{
3839 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3840 return 1;
3841
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003842 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003843 return 1;
3844
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003845 /*
3846 * We may have set a slab to be unmergeable during bootstrap.
3847 */
3848 if (s->refcount < 0)
3849 return 1;
3850
Christoph Lameter81819f02007-05-06 14:49:36 -07003851 return 0;
3852}
3853
3854static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003855 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003856 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003857{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003858 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003859
3860 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3861 return NULL;
3862
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003863 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003864 return NULL;
3865
3866 size = ALIGN(size, sizeof(void *));
3867 align = calculate_alignment(flags, align, size);
3868 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003869 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003870
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003871 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003872 if (slab_unmergeable(s))
3873 continue;
3874
3875 if (size > s->size)
3876 continue;
3877
Christoph Lameterba0268a2007-09-11 15:24:11 -07003878 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003879 continue;
3880 /*
3881 * Check if alignment is compatible.
3882 * Courtesy of Adrian Drzewiecki
3883 */
Pekka Enberg06428782008-01-07 23:20:27 -08003884 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003885 continue;
3886
3887 if (s->size - size >= sizeof(void *))
3888 continue;
3889
3890 return s;
3891 }
3892 return NULL;
3893}
3894
3895struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003896 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003897{
3898 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003899 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003900
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003901 if (WARN_ON(!name))
3902 return NULL;
3903
Christoph Lameter81819f02007-05-06 14:49:36 -07003904 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003905 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003906 if (s) {
3907 s->refcount++;
3908 /*
3909 * Adjust the object sizes so that we clear
3910 * the complete object on kzalloc.
3911 */
3912 s->objsize = max(s->objsize, (int)size);
3913 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003914
David Rientjes7b8f3b62008-12-17 22:09:46 -08003915 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003916 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003917 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003918 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003919 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003920 return s;
3921 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003922
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003923 n = kstrdup(name, GFP_KERNEL);
3924 if (!n)
3925 goto err;
3926
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003927 s = kmalloc(kmem_size, GFP_KERNEL);
3928 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003929 if (kmem_cache_open(s, n,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003930 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003931 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003932 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003933 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003934 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003935 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003936 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003937 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003938 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003939 return s;
3940 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003941 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003942 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003943 }
Pavel Emelyanov68cee4f2010-10-28 13:50:37 +04003944err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003945 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003946
Christoph Lameter81819f02007-05-06 14:49:36 -07003947 if (flags & SLAB_PANIC)
3948 panic("Cannot create slabcache %s\n", name);
3949 else
3950 s = NULL;
3951 return s;
3952}
3953EXPORT_SYMBOL(kmem_cache_create);
3954
Christoph Lameter81819f02007-05-06 14:49:36 -07003955#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003956/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003957 * Use the cpu notifier to insure that the cpu slabs are flushed when
3958 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003959 */
3960static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3961 unsigned long action, void *hcpu)
3962{
3963 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003964 struct kmem_cache *s;
3965 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003966
3967 switch (action) {
3968 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003969 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003970 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003971 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003972 down_read(&slub_lock);
3973 list_for_each_entry(s, &slab_caches, list) {
3974 local_irq_save(flags);
3975 __flush_cpu_slab(s, cpu);
3976 local_irq_restore(flags);
3977 }
3978 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003979 break;
3980 default:
3981 break;
3982 }
3983 return NOTIFY_OK;
3984}
3985
Pekka Enberg06428782008-01-07 23:20:27 -08003986static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003987 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003988};
Christoph Lameter81819f02007-05-06 14:49:36 -07003989
3990#endif
3991
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003992void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003993{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003994 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003995 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003996
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003997 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003998 return kmalloc_large(size, gfpflags);
3999
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004000 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004001
Satyam Sharma2408c552007-10-16 01:24:44 -07004002 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004003 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004004
Christoph Lameter2154a332010-07-09 14:07:10 -05004005 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004006
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004007 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02004008 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004009
4010 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004011}
4012
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004013#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004014void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004015 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004016{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004017 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004018 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004019
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004020 if (unlikely(size > SLUB_MAX_SIZE)) {
4021 ret = kmalloc_large_node(size, gfpflags, node);
4022
4023 trace_kmalloc_node(caller, ret,
4024 size, PAGE_SIZE << get_order(size),
4025 gfpflags, node);
4026
4027 return ret;
4028 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004029
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004030 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004031
Satyam Sharma2408c552007-10-16 01:24:44 -07004032 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004033 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004034
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004035 ret = slab_alloc(s, gfpflags, node, caller);
4036
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004037 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02004038 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004039
4040 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004041}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004042#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004043
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004044#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004045static int count_inuse(struct page *page)
4046{
4047 return page->inuse;
4048}
4049
4050static int count_total(struct page *page)
4051{
4052 return page->objects;
4053}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004054#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004055
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004056#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004057static int validate_slab(struct kmem_cache *s, struct page *page,
4058 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004059{
4060 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004061 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004062
4063 if (!check_slab(s, page) ||
4064 !on_freelist(s, page, NULL))
4065 return 0;
4066
4067 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004068 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004069
Christoph Lameter5f80b132011-04-15 14:48:13 -05004070 get_map(s, page, map);
4071 for_each_object(p, s, addr, page->objects) {
4072 if (test_bit(slab_index(p, s, addr), map))
4073 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4074 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004075 }
4076
Christoph Lameter224a88b2008-04-14 19:11:31 +03004077 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004078 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004079 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004080 return 0;
4081 return 1;
4082}
4083
Christoph Lameter434e2452007-07-17 04:03:30 -07004084static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4085 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004086{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004087 slab_lock(page);
4088 validate_slab(s, page, map);
4089 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004090}
4091
Christoph Lameter434e2452007-07-17 04:03:30 -07004092static int validate_slab_node(struct kmem_cache *s,
4093 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004094{
4095 unsigned long count = 0;
4096 struct page *page;
4097 unsigned long flags;
4098
4099 spin_lock_irqsave(&n->list_lock, flags);
4100
4101 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004102 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004103 count++;
4104 }
4105 if (count != n->nr_partial)
4106 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
4107 "counter=%ld\n", s->name, count, n->nr_partial);
4108
4109 if (!(s->flags & SLAB_STORE_USER))
4110 goto out;
4111
4112 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004113 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004114 count++;
4115 }
4116 if (count != atomic_long_read(&n->nr_slabs))
4117 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
4118 "counter=%ld\n", s->name, count,
4119 atomic_long_read(&n->nr_slabs));
4120
4121out:
4122 spin_unlock_irqrestore(&n->list_lock, flags);
4123 return count;
4124}
4125
Christoph Lameter434e2452007-07-17 04:03:30 -07004126static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004127{
4128 int node;
4129 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004130 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004131 sizeof(unsigned long), GFP_KERNEL);
4132
4133 if (!map)
4134 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004135
4136 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07004137 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07004138 struct kmem_cache_node *n = get_node(s, node);
4139
Christoph Lameter434e2452007-07-17 04:03:30 -07004140 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004141 }
Christoph Lameter434e2452007-07-17 04:03:30 -07004142 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004143 return count;
4144}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004145/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004146 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004147 * and freed.
4148 */
4149
4150struct location {
4151 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004152 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004153 long long sum_time;
4154 long min_time;
4155 long max_time;
4156 long min_pid;
4157 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304158 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004159 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004160};
4161
4162struct loc_track {
4163 unsigned long max;
4164 unsigned long count;
4165 struct location *loc;
4166};
4167
4168static void free_loc_track(struct loc_track *t)
4169{
4170 if (t->max)
4171 free_pages((unsigned long)t->loc,
4172 get_order(sizeof(struct location) * t->max));
4173}
4174
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004175static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004176{
4177 struct location *l;
4178 int order;
4179
Christoph Lameter88a420e2007-05-06 14:49:45 -07004180 order = get_order(sizeof(struct location) * max);
4181
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004182 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004183 if (!l)
4184 return 0;
4185
4186 if (t->count) {
4187 memcpy(l, t->loc, sizeof(struct location) * t->count);
4188 free_loc_track(t);
4189 }
4190 t->max = max;
4191 t->loc = l;
4192 return 1;
4193}
4194
4195static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004196 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004197{
4198 long start, end, pos;
4199 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004200 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004201 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004202
4203 start = -1;
4204 end = t->count;
4205
4206 for ( ; ; ) {
4207 pos = start + (end - start + 1) / 2;
4208
4209 /*
4210 * There is nothing at "end". If we end up there
4211 * we need to add something to before end.
4212 */
4213 if (pos == end)
4214 break;
4215
4216 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004217 if (track->addr == caddr) {
4218
4219 l = &t->loc[pos];
4220 l->count++;
4221 if (track->when) {
4222 l->sum_time += age;
4223 if (age < l->min_time)
4224 l->min_time = age;
4225 if (age > l->max_time)
4226 l->max_time = age;
4227
4228 if (track->pid < l->min_pid)
4229 l->min_pid = track->pid;
4230 if (track->pid > l->max_pid)
4231 l->max_pid = track->pid;
4232
Rusty Russell174596a2009-01-01 10:12:29 +10304233 cpumask_set_cpu(track->cpu,
4234 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004235 }
4236 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004237 return 1;
4238 }
4239
Christoph Lameter45edfa52007-05-09 02:32:45 -07004240 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004241 end = pos;
4242 else
4243 start = pos;
4244 }
4245
4246 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004247 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004248 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004249 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004250 return 0;
4251
4252 l = t->loc + pos;
4253 if (pos < t->count)
4254 memmove(l + 1, l,
4255 (t->count - pos) * sizeof(struct location));
4256 t->count++;
4257 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004258 l->addr = track->addr;
4259 l->sum_time = age;
4260 l->min_time = age;
4261 l->max_time = age;
4262 l->min_pid = track->pid;
4263 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304264 cpumask_clear(to_cpumask(l->cpus));
4265 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004266 nodes_clear(l->nodes);
4267 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004268 return 1;
4269}
4270
4271static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004272 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004273 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004274{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004275 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004276 void *p;
4277
Christoph Lameter39b26462008-04-14 19:11:30 +03004278 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004279 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004280
Christoph Lameter224a88b2008-04-14 19:11:31 +03004281 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004282 if (!test_bit(slab_index(p, s, addr), map))
4283 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004284}
4285
4286static int list_locations(struct kmem_cache *s, char *buf,
4287 enum track_item alloc)
4288{
Harvey Harrisone374d482008-01-31 15:20:50 -08004289 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004290 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004291 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004292 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004293 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4294 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004295
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004296 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4297 GFP_TEMPORARY)) {
4298 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004299 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004300 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004301 /* Push back cpu slabs */
4302 flush_all(s);
4303
Christoph Lameterf64dc582007-10-16 01:25:33 -07004304 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004305 struct kmem_cache_node *n = get_node(s, node);
4306 unsigned long flags;
4307 struct page *page;
4308
Christoph Lameter9e869432007-08-22 14:01:56 -07004309 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004310 continue;
4311
4312 spin_lock_irqsave(&n->list_lock, flags);
4313 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004314 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004315 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004316 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004317 spin_unlock_irqrestore(&n->list_lock, flags);
4318 }
4319
4320 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004321 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004322
Hugh Dickins9c246242008-12-09 13:14:27 -08004323 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004324 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004325 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004326
4327 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004328 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004329 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004330 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004331
4332 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004333 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004334 l->min_time,
4335 (long)div_u64(l->sum_time, l->count),
4336 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004337 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004338 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004339 l->min_time);
4340
4341 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004342 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004343 l->min_pid, l->max_pid);
4344 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004345 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004346 l->min_pid);
4347
Rusty Russell174596a2009-01-01 10:12:29 +10304348 if (num_online_cpus() > 1 &&
4349 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004350 len < PAGE_SIZE - 60) {
4351 len += sprintf(buf + len, " cpus=");
4352 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10304353 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004354 }
4355
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004356 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004357 len < PAGE_SIZE - 60) {
4358 len += sprintf(buf + len, " nodes=");
4359 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004360 l->nodes);
4361 }
4362
Harvey Harrisone374d482008-01-31 15:20:50 -08004363 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004364 }
4365
4366 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004367 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004368 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004369 len += sprintf(buf, "No data\n");
4370 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004371}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004372#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004373
Christoph Lametera5a84752010-10-05 13:57:27 -05004374#ifdef SLUB_RESILIENCY_TEST
4375static void resiliency_test(void)
4376{
4377 u8 *p;
4378
4379 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
4380
4381 printk(KERN_ERR "SLUB resiliency testing\n");
4382 printk(KERN_ERR "-----------------------\n");
4383 printk(KERN_ERR "A. Corruption after allocation\n");
4384
4385 p = kzalloc(16, GFP_KERNEL);
4386 p[16] = 0x12;
4387 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
4388 " 0x12->0x%p\n\n", p + 16);
4389
4390 validate_slab_cache(kmalloc_caches[4]);
4391
4392 /* Hmmm... The next two are dangerous */
4393 p = kzalloc(32, GFP_KERNEL);
4394 p[32 + sizeof(void *)] = 0x34;
4395 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
4396 " 0x34 -> -0x%p\n", p);
4397 printk(KERN_ERR
4398 "If allocated object is overwritten then not detectable\n\n");
4399
4400 validate_slab_cache(kmalloc_caches[5]);
4401 p = kzalloc(64, GFP_KERNEL);
4402 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4403 *p = 0x56;
4404 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4405 p);
4406 printk(KERN_ERR
4407 "If allocated object is overwritten then not detectable\n\n");
4408 validate_slab_cache(kmalloc_caches[6]);
4409
4410 printk(KERN_ERR "\nB. Corruption after free\n");
4411 p = kzalloc(128, GFP_KERNEL);
4412 kfree(p);
4413 *p = 0x78;
4414 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4415 validate_slab_cache(kmalloc_caches[7]);
4416
4417 p = kzalloc(256, GFP_KERNEL);
4418 kfree(p);
4419 p[50] = 0x9a;
4420 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4421 p);
4422 validate_slab_cache(kmalloc_caches[8]);
4423
4424 p = kzalloc(512, GFP_KERNEL);
4425 kfree(p);
4426 p[512] = 0xab;
4427 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4428 validate_slab_cache(kmalloc_caches[9]);
4429}
4430#else
4431#ifdef CONFIG_SYSFS
4432static void resiliency_test(void) {};
4433#endif
4434#endif
4435
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004436#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004437enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004438 SL_ALL, /* All slabs */
4439 SL_PARTIAL, /* Only partially allocated slabs */
4440 SL_CPU, /* Only slabs used for cpu caches */
4441 SL_OBJECTS, /* Determine allocated objects not slabs */
4442 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004443};
4444
Christoph Lameter205ab992008-04-14 19:11:40 +03004445#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004446#define SO_PARTIAL (1 << SL_PARTIAL)
4447#define SO_CPU (1 << SL_CPU)
4448#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004449#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004450
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004451static ssize_t show_slab_objects(struct kmem_cache *s,
4452 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004453{
4454 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004455 int node;
4456 int x;
4457 unsigned long *nodes;
4458 unsigned long *per_cpu;
4459
4460 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004461 if (!nodes)
4462 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004463 per_cpu = nodes + nr_node_ids;
4464
Christoph Lameter205ab992008-04-14 19:11:40 +03004465 if (flags & SO_CPU) {
4466 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004467
Christoph Lameter205ab992008-04-14 19:11:40 +03004468 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004469 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004470 int node = ACCESS_ONCE(c->node);
Christoph Lameter49e22582011-08-09 16:12:27 -05004471 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004472
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004473 if (node < 0)
Christoph Lameter205ab992008-04-14 19:11:40 +03004474 continue;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004475 page = ACCESS_ONCE(c->page);
4476 if (page) {
4477 if (flags & SO_TOTAL)
4478 x = page->objects;
Christoph Lameter205ab992008-04-14 19:11:40 +03004479 else if (flags & SO_OBJECTS)
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004480 x = page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07004481 else
4482 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03004483
Christoph Lameter81819f02007-05-06 14:49:36 -07004484 total += x;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004485 nodes[node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004486 }
Christoph Lameter49e22582011-08-09 16:12:27 -05004487 page = c->partial;
4488
4489 if (page) {
4490 x = page->pobjects;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004491 total += x;
4492 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004493 }
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004494 per_cpu[node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004495 }
4496 }
4497
Christoph Lameter04d94872011-01-10 10:15:15 -06004498 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004499#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004500 if (flags & SO_ALL) {
4501 for_each_node_state(node, N_NORMAL_MEMORY) {
4502 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004503
Christoph Lameter205ab992008-04-14 19:11:40 +03004504 if (flags & SO_TOTAL)
4505 x = atomic_long_read(&n->total_objects);
4506 else if (flags & SO_OBJECTS)
4507 x = atomic_long_read(&n->total_objects) -
4508 count_partial(n, count_free);
4509
4510 else
4511 x = atomic_long_read(&n->nr_slabs);
4512 total += x;
4513 nodes[node] += x;
4514 }
4515
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004516 } else
4517#endif
4518 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004519 for_each_node_state(node, N_NORMAL_MEMORY) {
4520 struct kmem_cache_node *n = get_node(s, node);
4521
4522 if (flags & SO_TOTAL)
4523 x = count_partial(n, count_total);
4524 else if (flags & SO_OBJECTS)
4525 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004526 else
4527 x = n->nr_partial;
4528 total += x;
4529 nodes[node] += x;
4530 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004531 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004532 x = sprintf(buf, "%lu", total);
4533#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004534 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004535 if (nodes[node])
4536 x += sprintf(buf + x, " N%d=%lu",
4537 node, nodes[node]);
4538#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004539 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004540 kfree(nodes);
4541 return x + sprintf(buf + x, "\n");
4542}
4543
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004544#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004545static int any_slab_objects(struct kmem_cache *s)
4546{
4547 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004548
4549 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004550 struct kmem_cache_node *n = get_node(s, node);
4551
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004552 if (!n)
4553 continue;
4554
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004555 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004556 return 1;
4557 }
4558 return 0;
4559}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004560#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004561
4562#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004563#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004564
4565struct slab_attribute {
4566 struct attribute attr;
4567 ssize_t (*show)(struct kmem_cache *s, char *buf);
4568 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4569};
4570
4571#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004572 static struct slab_attribute _name##_attr = \
4573 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004574
4575#define SLAB_ATTR(_name) \
4576 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004577 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004578
Christoph Lameter81819f02007-05-06 14:49:36 -07004579static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4580{
4581 return sprintf(buf, "%d\n", s->size);
4582}
4583SLAB_ATTR_RO(slab_size);
4584
4585static ssize_t align_show(struct kmem_cache *s, char *buf)
4586{
4587 return sprintf(buf, "%d\n", s->align);
4588}
4589SLAB_ATTR_RO(align);
4590
4591static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4592{
4593 return sprintf(buf, "%d\n", s->objsize);
4594}
4595SLAB_ATTR_RO(object_size);
4596
4597static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4598{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004599 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004600}
4601SLAB_ATTR_RO(objs_per_slab);
4602
Christoph Lameter06b285d2008-04-14 19:11:41 +03004603static ssize_t order_store(struct kmem_cache *s,
4604 const char *buf, size_t length)
4605{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004606 unsigned long order;
4607 int err;
4608
4609 err = strict_strtoul(buf, 10, &order);
4610 if (err)
4611 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004612
4613 if (order > slub_max_order || order < slub_min_order)
4614 return -EINVAL;
4615
4616 calculate_sizes(s, order);
4617 return length;
4618}
4619
Christoph Lameter81819f02007-05-06 14:49:36 -07004620static ssize_t order_show(struct kmem_cache *s, char *buf)
4621{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004622 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004623}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004624SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004625
David Rientjes73d342b2009-02-22 17:40:09 -08004626static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4627{
4628 return sprintf(buf, "%lu\n", s->min_partial);
4629}
4630
4631static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4632 size_t length)
4633{
4634 unsigned long min;
4635 int err;
4636
4637 err = strict_strtoul(buf, 10, &min);
4638 if (err)
4639 return err;
4640
David Rientjesc0bdb232009-02-25 09:16:35 +02004641 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004642 return length;
4643}
4644SLAB_ATTR(min_partial);
4645
Christoph Lameter49e22582011-08-09 16:12:27 -05004646static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4647{
4648 return sprintf(buf, "%u\n", s->cpu_partial);
4649}
4650
4651static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4652 size_t length)
4653{
4654 unsigned long objects;
4655 int err;
4656
4657 err = strict_strtoul(buf, 10, &objects);
4658 if (err)
4659 return err;
David Rientjes74ee4ef2012-01-09 13:19:45 -08004660 if (objects && kmem_cache_debug(s))
4661 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004662
4663 s->cpu_partial = objects;
4664 flush_all(s);
4665 return length;
4666}
4667SLAB_ATTR(cpu_partial);
4668
Christoph Lameter81819f02007-05-06 14:49:36 -07004669static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4670{
Joe Perches62c70bc2011-01-13 15:45:52 -08004671 if (!s->ctor)
4672 return 0;
4673 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004674}
4675SLAB_ATTR_RO(ctor);
4676
Christoph Lameter81819f02007-05-06 14:49:36 -07004677static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4678{
4679 return sprintf(buf, "%d\n", s->refcount - 1);
4680}
4681SLAB_ATTR_RO(aliases);
4682
Christoph Lameter81819f02007-05-06 14:49:36 -07004683static ssize_t partial_show(struct kmem_cache *s, char *buf)
4684{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004685 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004686}
4687SLAB_ATTR_RO(partial);
4688
4689static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4690{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004691 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004692}
4693SLAB_ATTR_RO(cpu_slabs);
4694
4695static ssize_t objects_show(struct kmem_cache *s, char *buf)
4696{
Christoph Lameter205ab992008-04-14 19:11:40 +03004697 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004698}
4699SLAB_ATTR_RO(objects);
4700
Christoph Lameter205ab992008-04-14 19:11:40 +03004701static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4702{
4703 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4704}
4705SLAB_ATTR_RO(objects_partial);
4706
Christoph Lameter49e22582011-08-09 16:12:27 -05004707static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4708{
4709 int objects = 0;
4710 int pages = 0;
4711 int cpu;
4712 int len;
4713
4714 for_each_online_cpu(cpu) {
4715 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4716
4717 if (page) {
4718 pages += page->pages;
4719 objects += page->pobjects;
4720 }
4721 }
4722
4723 len = sprintf(buf, "%d(%d)", objects, pages);
4724
4725#ifdef CONFIG_SMP
4726 for_each_online_cpu(cpu) {
4727 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4728
4729 if (page && len < PAGE_SIZE - 20)
4730 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4731 page->pobjects, page->pages);
4732 }
4733#endif
4734 return len + sprintf(buf + len, "\n");
4735}
4736SLAB_ATTR_RO(slabs_cpu_partial);
4737
Christoph Lameter81819f02007-05-06 14:49:36 -07004738static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4739{
4740 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4741}
4742
4743static ssize_t reclaim_account_store(struct kmem_cache *s,
4744 const char *buf, size_t length)
4745{
4746 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4747 if (buf[0] == '1')
4748 s->flags |= SLAB_RECLAIM_ACCOUNT;
4749 return length;
4750}
4751SLAB_ATTR(reclaim_account);
4752
4753static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4754{
Christoph Lameter5af60832007-05-06 14:49:56 -07004755 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004756}
4757SLAB_ATTR_RO(hwcache_align);
4758
4759#ifdef CONFIG_ZONE_DMA
4760static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4761{
4762 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4763}
4764SLAB_ATTR_RO(cache_dma);
4765#endif
4766
4767static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4768{
4769 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4770}
4771SLAB_ATTR_RO(destroy_by_rcu);
4772
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004773static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4774{
4775 return sprintf(buf, "%d\n", s->reserved);
4776}
4777SLAB_ATTR_RO(reserved);
4778
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004779#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004780static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4781{
4782 return show_slab_objects(s, buf, SO_ALL);
4783}
4784SLAB_ATTR_RO(slabs);
4785
4786static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4787{
4788 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4789}
4790SLAB_ATTR_RO(total_objects);
4791
4792static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4793{
4794 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4795}
4796
4797static ssize_t sanity_checks_store(struct kmem_cache *s,
4798 const char *buf, size_t length)
4799{
4800 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004801 if (buf[0] == '1') {
4802 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004803 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004804 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004805 return length;
4806}
4807SLAB_ATTR(sanity_checks);
4808
4809static ssize_t trace_show(struct kmem_cache *s, char *buf)
4810{
4811 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4812}
4813
4814static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4815 size_t length)
4816{
4817 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004818 if (buf[0] == '1') {
4819 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004820 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004821 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004822 return length;
4823}
4824SLAB_ATTR(trace);
4825
Christoph Lameter81819f02007-05-06 14:49:36 -07004826static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4827{
4828 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4829}
4830
4831static ssize_t red_zone_store(struct kmem_cache *s,
4832 const char *buf, size_t length)
4833{
4834 if (any_slab_objects(s))
4835 return -EBUSY;
4836
4837 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004838 if (buf[0] == '1') {
4839 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004840 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004841 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004842 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004843 return length;
4844}
4845SLAB_ATTR(red_zone);
4846
4847static ssize_t poison_show(struct kmem_cache *s, char *buf)
4848{
4849 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4850}
4851
4852static ssize_t poison_store(struct kmem_cache *s,
4853 const char *buf, size_t length)
4854{
4855 if (any_slab_objects(s))
4856 return -EBUSY;
4857
4858 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004859 if (buf[0] == '1') {
4860 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004861 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004862 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004863 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004864 return length;
4865}
4866SLAB_ATTR(poison);
4867
4868static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4869{
4870 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4871}
4872
4873static ssize_t store_user_store(struct kmem_cache *s,
4874 const char *buf, size_t length)
4875{
4876 if (any_slab_objects(s))
4877 return -EBUSY;
4878
4879 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004880 if (buf[0] == '1') {
4881 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004882 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004883 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004884 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004885 return length;
4886}
4887SLAB_ATTR(store_user);
4888
Christoph Lameter53e15af2007-05-06 14:49:43 -07004889static ssize_t validate_show(struct kmem_cache *s, char *buf)
4890{
4891 return 0;
4892}
4893
4894static ssize_t validate_store(struct kmem_cache *s,
4895 const char *buf, size_t length)
4896{
Christoph Lameter434e2452007-07-17 04:03:30 -07004897 int ret = -EINVAL;
4898
4899 if (buf[0] == '1') {
4900 ret = validate_slab_cache(s);
4901 if (ret >= 0)
4902 ret = length;
4903 }
4904 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004905}
4906SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004907
4908static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4909{
4910 if (!(s->flags & SLAB_STORE_USER))
4911 return -ENOSYS;
4912 return list_locations(s, buf, TRACK_ALLOC);
4913}
4914SLAB_ATTR_RO(alloc_calls);
4915
4916static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4917{
4918 if (!(s->flags & SLAB_STORE_USER))
4919 return -ENOSYS;
4920 return list_locations(s, buf, TRACK_FREE);
4921}
4922SLAB_ATTR_RO(free_calls);
4923#endif /* CONFIG_SLUB_DEBUG */
4924
4925#ifdef CONFIG_FAILSLAB
4926static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4927{
4928 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4929}
4930
4931static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4932 size_t length)
4933{
4934 s->flags &= ~SLAB_FAILSLAB;
4935 if (buf[0] == '1')
4936 s->flags |= SLAB_FAILSLAB;
4937 return length;
4938}
4939SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004940#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004941
Christoph Lameter2086d262007-05-06 14:49:46 -07004942static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4943{
4944 return 0;
4945}
4946
4947static ssize_t shrink_store(struct kmem_cache *s,
4948 const char *buf, size_t length)
4949{
4950 if (buf[0] == '1') {
4951 int rc = kmem_cache_shrink(s);
4952
4953 if (rc)
4954 return rc;
4955 } else
4956 return -EINVAL;
4957 return length;
4958}
4959SLAB_ATTR(shrink);
4960
Christoph Lameter81819f02007-05-06 14:49:36 -07004961#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004962static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004963{
Christoph Lameter98246012008-01-07 23:20:26 -08004964 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004965}
4966
Christoph Lameter98246012008-01-07 23:20:26 -08004967static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004968 const char *buf, size_t length)
4969{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004970 unsigned long ratio;
4971 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004972
Christoph Lameter0121c6192008-04-29 16:11:12 -07004973 err = strict_strtoul(buf, 10, &ratio);
4974 if (err)
4975 return err;
4976
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004977 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004978 s->remote_node_defrag_ratio = ratio * 10;
4979
Christoph Lameter81819f02007-05-06 14:49:36 -07004980 return length;
4981}
Christoph Lameter98246012008-01-07 23:20:26 -08004982SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004983#endif
4984
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004985#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004986static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4987{
4988 unsigned long sum = 0;
4989 int cpu;
4990 int len;
4991 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4992
4993 if (!data)
4994 return -ENOMEM;
4995
4996 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004997 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004998
4999 data[cpu] = x;
5000 sum += x;
5001 }
5002
5003 len = sprintf(buf, "%lu", sum);
5004
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005005#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005006 for_each_online_cpu(cpu) {
5007 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005008 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005009 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005010#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005011 kfree(data);
5012 return len + sprintf(buf + len, "\n");
5013}
5014
David Rientjes78eb00c2009-10-15 02:20:22 -07005015static void clear_stat(struct kmem_cache *s, enum stat_item si)
5016{
5017 int cpu;
5018
5019 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005020 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005021}
5022
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005023#define STAT_ATTR(si, text) \
5024static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5025{ \
5026 return show_stat(s, buf, si); \
5027} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005028static ssize_t text##_store(struct kmem_cache *s, \
5029 const char *buf, size_t length) \
5030{ \
5031 if (buf[0] != '0') \
5032 return -EINVAL; \
5033 clear_stat(s, si); \
5034 return length; \
5035} \
5036SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005037
5038STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5039STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5040STAT_ATTR(FREE_FASTPATH, free_fastpath);
5041STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5042STAT_ATTR(FREE_FROZEN, free_frozen);
5043STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5044STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5045STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5046STAT_ATTR(ALLOC_SLAB, alloc_slab);
5047STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005048STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005049STAT_ATTR(FREE_SLAB, free_slab);
5050STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5051STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5052STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5053STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5054STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5055STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005056STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005057STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005058STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5059STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005060STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5061STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005062#endif
5063
Pekka Enberg06428782008-01-07 23:20:27 -08005064static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005065 &slab_size_attr.attr,
5066 &object_size_attr.attr,
5067 &objs_per_slab_attr.attr,
5068 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005069 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005070 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005071 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005072 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005073 &partial_attr.attr,
5074 &cpu_slabs_attr.attr,
5075 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005076 &aliases_attr.attr,
5077 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005078 &hwcache_align_attr.attr,
5079 &reclaim_account_attr.attr,
5080 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005081 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005082 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005083 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005084#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005085 &total_objects_attr.attr,
5086 &slabs_attr.attr,
5087 &sanity_checks_attr.attr,
5088 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005089 &red_zone_attr.attr,
5090 &poison_attr.attr,
5091 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005092 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005093 &alloc_calls_attr.attr,
5094 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005095#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005096#ifdef CONFIG_ZONE_DMA
5097 &cache_dma_attr.attr,
5098#endif
5099#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005100 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005101#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005102#ifdef CONFIG_SLUB_STATS
5103 &alloc_fastpath_attr.attr,
5104 &alloc_slowpath_attr.attr,
5105 &free_fastpath_attr.attr,
5106 &free_slowpath_attr.attr,
5107 &free_frozen_attr.attr,
5108 &free_add_partial_attr.attr,
5109 &free_remove_partial_attr.attr,
5110 &alloc_from_partial_attr.attr,
5111 &alloc_slab_attr.attr,
5112 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005113 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005114 &free_slab_attr.attr,
5115 &cpuslab_flush_attr.attr,
5116 &deactivate_full_attr.attr,
5117 &deactivate_empty_attr.attr,
5118 &deactivate_to_head_attr.attr,
5119 &deactivate_to_tail_attr.attr,
5120 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005121 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005122 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005123 &cmpxchg_double_fail_attr.attr,
5124 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005125 &cpu_partial_alloc_attr.attr,
5126 &cpu_partial_free_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005127#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005128#ifdef CONFIG_FAILSLAB
5129 &failslab_attr.attr,
5130#endif
5131
Christoph Lameter81819f02007-05-06 14:49:36 -07005132 NULL
5133};
5134
5135static struct attribute_group slab_attr_group = {
5136 .attrs = slab_attrs,
5137};
5138
5139static ssize_t slab_attr_show(struct kobject *kobj,
5140 struct attribute *attr,
5141 char *buf)
5142{
5143 struct slab_attribute *attribute;
5144 struct kmem_cache *s;
5145 int err;
5146
5147 attribute = to_slab_attr(attr);
5148 s = to_slab(kobj);
5149
5150 if (!attribute->show)
5151 return -EIO;
5152
5153 err = attribute->show(s, buf);
5154
5155 return err;
5156}
5157
5158static ssize_t slab_attr_store(struct kobject *kobj,
5159 struct attribute *attr,
5160 const char *buf, size_t len)
5161{
5162 struct slab_attribute *attribute;
5163 struct kmem_cache *s;
5164 int err;
5165
5166 attribute = to_slab_attr(attr);
5167 s = to_slab(kobj);
5168
5169 if (!attribute->store)
5170 return -EIO;
5171
5172 err = attribute->store(s, buf, len);
5173
5174 return err;
5175}
5176
Christoph Lameter151c6022008-01-07 22:29:05 -08005177static void kmem_cache_release(struct kobject *kobj)
5178{
5179 struct kmem_cache *s = to_slab(kobj);
5180
Pekka Enberg84c1cf62010-09-14 23:21:12 +03005181 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08005182 kfree(s);
5183}
5184
Emese Revfy52cf25d2010-01-19 02:58:23 +01005185static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005186 .show = slab_attr_show,
5187 .store = slab_attr_store,
5188};
5189
5190static struct kobj_type slab_ktype = {
5191 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08005192 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07005193};
5194
5195static int uevent_filter(struct kset *kset, struct kobject *kobj)
5196{
5197 struct kobj_type *ktype = get_ktype(kobj);
5198
5199 if (ktype == &slab_ktype)
5200 return 1;
5201 return 0;
5202}
5203
Emese Revfy9cd43612009-12-31 14:52:51 +01005204static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005205 .filter = uevent_filter,
5206};
5207
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005208static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005209
5210#define ID_STR_LENGTH 64
5211
5212/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005213 *
5214 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005215 */
5216static char *create_unique_id(struct kmem_cache *s)
5217{
5218 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5219 char *p = name;
5220
5221 BUG_ON(!name);
5222
5223 *p++ = ':';
5224 /*
5225 * First flags affecting slabcache operations. We will only
5226 * get here for aliasable slabs so we do not need to support
5227 * too many flags. The flags here must cover all flags that
5228 * are matched during merging to guarantee that the id is
5229 * unique.
5230 */
5231 if (s->flags & SLAB_CACHE_DMA)
5232 *p++ = 'd';
5233 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5234 *p++ = 'a';
5235 if (s->flags & SLAB_DEBUG_FREE)
5236 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005237 if (!(s->flags & SLAB_NOTRACK))
5238 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005239 if (p != name + 1)
5240 *p++ = '-';
5241 p += sprintf(p, "%07d", s->size);
5242 BUG_ON(p > name + ID_STR_LENGTH - 1);
5243 return name;
5244}
5245
5246static int sysfs_slab_add(struct kmem_cache *s)
5247{
5248 int err;
5249 const char *name;
5250 int unmergeable;
5251
5252 if (slab_state < SYSFS)
5253 /* Defer until later */
5254 return 0;
5255
5256 unmergeable = slab_unmergeable(s);
5257 if (unmergeable) {
5258 /*
5259 * Slabcache can never be merged so we can use the name proper.
5260 * This is typically the case for debug situations. In that
5261 * case we can catch duplicate names easily.
5262 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005263 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005264 name = s->name;
5265 } else {
5266 /*
5267 * Create a unique name for the slab as a target
5268 * for the symlinks.
5269 */
5270 name = create_unique_id(s);
5271 }
5272
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005273 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005274 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
5275 if (err) {
5276 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005277 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005278 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005279
5280 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005281 if (err) {
5282 kobject_del(&s->kobj);
5283 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005284 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005285 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005286 kobject_uevent(&s->kobj, KOBJ_ADD);
5287 if (!unmergeable) {
5288 /* Setup first alias */
5289 sysfs_slab_alias(s, s->name);
5290 kfree(name);
5291 }
5292 return 0;
5293}
5294
5295static void sysfs_slab_remove(struct kmem_cache *s)
5296{
Christoph Lameter2bce6482010-07-19 11:39:11 -05005297 if (slab_state < SYSFS)
5298 /*
5299 * Sysfs has not been setup yet so no need to remove the
5300 * cache from sysfs.
5301 */
5302 return;
5303
Christoph Lameter81819f02007-05-06 14:49:36 -07005304 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5305 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005306 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005307}
5308
5309/*
5310 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005311 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005312 */
5313struct saved_alias {
5314 struct kmem_cache *s;
5315 const char *name;
5316 struct saved_alias *next;
5317};
5318
Adrian Bunk5af328a2007-07-17 04:03:27 -07005319static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005320
5321static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5322{
5323 struct saved_alias *al;
5324
5325 if (slab_state == SYSFS) {
5326 /*
5327 * If we have a leftover link then remove it.
5328 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005329 sysfs_remove_link(&slab_kset->kobj, name);
5330 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005331 }
5332
5333 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5334 if (!al)
5335 return -ENOMEM;
5336
5337 al->s = s;
5338 al->name = name;
5339 al->next = alias_list;
5340 alias_list = al;
5341 return 0;
5342}
5343
5344static int __init slab_sysfs_init(void)
5345{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07005346 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005347 int err;
5348
Christoph Lameter2bce6482010-07-19 11:39:11 -05005349 down_write(&slub_lock);
5350
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005351 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005352 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05005353 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005354 printk(KERN_ERR "Cannot register slab subsystem.\n");
5355 return -ENOSYS;
5356 }
5357
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005358 slab_state = SYSFS;
5359
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07005360 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005361 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005362 if (err)
5363 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
5364 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005365 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005366
5367 while (alias_list) {
5368 struct saved_alias *al = alias_list;
5369
5370 alias_list = alias_list->next;
5371 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005372 if (err)
5373 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
5374 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005375 kfree(al);
5376 }
5377
Christoph Lameter2bce6482010-07-19 11:39:11 -05005378 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005379 resiliency_test();
5380 return 0;
5381}
5382
5383__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005384#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005385
5386/*
5387 * The /proc/slabinfo ABI
5388 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005389#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005390static void print_slabinfo_header(struct seq_file *m)
5391{
5392 seq_puts(m, "slabinfo - version: 2.1\n");
5393 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
5394 "<objperslab> <pagesperslab>");
5395 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
5396 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5397 seq_putc(m, '\n');
5398}
5399
5400static void *s_start(struct seq_file *m, loff_t *pos)
5401{
5402 loff_t n = *pos;
5403
5404 down_read(&slub_lock);
5405 if (!n)
5406 print_slabinfo_header(m);
5407
5408 return seq_list_start(&slab_caches, *pos);
5409}
5410
5411static void *s_next(struct seq_file *m, void *p, loff_t *pos)
5412{
5413 return seq_list_next(p, &slab_caches, pos);
5414}
5415
5416static void s_stop(struct seq_file *m, void *p)
5417{
5418 up_read(&slub_lock);
5419}
5420
5421static int s_show(struct seq_file *m, void *p)
5422{
5423 unsigned long nr_partials = 0;
5424 unsigned long nr_slabs = 0;
5425 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005426 unsigned long nr_objs = 0;
5427 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005428 struct kmem_cache *s;
5429 int node;
5430
5431 s = list_entry(p, struct kmem_cache, list);
5432
5433 for_each_online_node(node) {
5434 struct kmem_cache_node *n = get_node(s, node);
5435
5436 if (!n)
5437 continue;
5438
5439 nr_partials += n->nr_partial;
5440 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03005441 nr_objs += atomic_long_read(&n->total_objects);
5442 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005443 }
5444
Christoph Lameter205ab992008-04-14 19:11:40 +03005445 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005446
5447 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03005448 nr_objs, s->size, oo_objects(s->oo),
5449 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005450 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
5451 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
5452 0UL);
5453 seq_putc(m, '\n');
5454 return 0;
5455}
5456
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005457static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005458 .start = s_start,
5459 .next = s_next,
5460 .stop = s_stop,
5461 .show = s_show,
5462};
5463
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005464static int slabinfo_open(struct inode *inode, struct file *file)
5465{
5466 return seq_open(file, &slabinfo_op);
5467}
5468
5469static const struct file_operations proc_slabinfo_operations = {
5470 .open = slabinfo_open,
5471 .read = seq_read,
5472 .llseek = seq_lseek,
5473 .release = seq_release,
5474};
5475
5476static int __init slab_proc_init(void)
5477{
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04005478 proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005479 return 0;
5480}
5481module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08005482#endif /* CONFIG_SLABINFO */