blob: 97bb5b8d935fb0f6a9c029032a73d56076ee7208 [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 *
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
7 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -07009 */
10
11#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100012#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070013#include <linux/module.h>
14#include <linux/bit_spinlock.h>
15#include <linux/interrupt.h>
16#include <linux/bitops.h>
17#include <linux/slab.h>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040018#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070019#include <linux/seq_file.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020020#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070021#include <linux/cpu.h>
22#include <linux/cpuset.h>
23#include <linux/mempolicy.h>
24#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070025#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070026#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070027#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070028#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090029#include <linux/fault-inject.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070030
Richard Kennedy4a923792010-10-21 10:29:19 +010031#include <trace/events/kmem.h>
32
Christoph Lameter81819f02007-05-06 14:49:36 -070033/*
34 * Lock order:
35 * 1. slab_lock(page)
36 * 2. slab->list_lock
37 *
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
44 *
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
50 *
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
55 * the list lock.
56 *
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
69 *
70 * 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 Lameter81819f02007-05-06 14:49:36 -0700134/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700135 * Mininum number of partial slabs. These will be left on the partial
136 * lists even if they are empty. kmem_cache_shrink may reclaim them.
137 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800138#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700139
Christoph Lameter2086d262007-05-06 14:49:46 -0700140/*
141 * Maximum number of desirable partial slabs.
142 * The existence of more partial slabs makes kmem_cache_shrink
143 * sort the partial list by the number of objects in the.
144 */
145#define MAX_PARTIAL 10
146
Christoph Lameter81819f02007-05-06 14:49:36 -0700147#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
148 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700149
Christoph Lameter81819f02007-05-06 14:49:36 -0700150/*
David Rientjes3de47212009-07-27 18:30:35 -0700151 * Debugging flags that require metadata to be stored in the slab. These get
152 * disabled when slub_debug=O is used and a cache's min order increases with
153 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700154 */
David Rientjes3de47212009-07-27 18:30:35 -0700155#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700156
157/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700158 * Set of flags that will prevent slab merging
159 */
160#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300161 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
162 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700163
164#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200165 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400167#define OO_SHIFT 16
168#define OO_MASK ((1 << OO_SHIFT) - 1)
169#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
170
Christoph Lameter81819f02007-05-06 14:49:36 -0700171/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500172#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameter81819f02007-05-06 14:49:36 -0700173
174static int kmem_size = sizeof(struct kmem_cache);
175
176#ifdef CONFIG_SMP
177static struct notifier_block slab_notifier;
178#endif
179
180static enum {
181 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500182 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700183 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700184 SYSFS /* Sysfs up */
185} slab_state = DOWN;
186
187/* A list of all slab caches on the system */
188static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700189static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700190
Christoph Lameter02cbc872007-05-09 02:32:43 -0700191/*
192 * Tracking user of a slab.
193 */
194struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300195 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700196 int cpu; /* Was running on cpu */
197 int pid; /* Pid context */
198 unsigned long when; /* When did the operation occur */
199};
200
201enum track_item { TRACK_ALLOC, TRACK_FREE };
202
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500203#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700204static int sysfs_slab_add(struct kmem_cache *);
205static int sysfs_slab_alias(struct kmem_cache *, const char *);
206static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800207
Christoph Lameter81819f02007-05-06 14:49:36 -0700208#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700209static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
210static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
211 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800212static inline void sysfs_slab_remove(struct kmem_cache *s)
213{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300214 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800215 kfree(s);
216}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800217
Christoph Lameter81819f02007-05-06 14:49:36 -0700218#endif
219
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500220static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800221{
222#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600223 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800224#endif
225}
226
Christoph Lameter81819f02007-05-06 14:49:36 -0700227/********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
230
231int slab_is_available(void)
232{
233 return slab_state >= UP;
234}
235
236static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
237{
Christoph Lameter81819f02007-05-06 14:49:36 -0700238 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700239}
240
Christoph Lameter6446faa2008-02-15 23:45:26 -0800241/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700242static inline int check_valid_pointer(struct kmem_cache *s,
243 struct page *page, const void *object)
244{
245 void *base;
246
Christoph Lametera973e9d2008-03-01 13:40:44 -0800247 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700248 return 1;
249
Christoph Lametera973e9d2008-03-01 13:40:44 -0800250 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300251 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700252 (object - base) % s->size) {
253 return 0;
254 }
255
256 return 1;
257}
258
Christoph Lameter7656c722007-05-09 02:32:40 -0700259static inline void *get_freepointer(struct kmem_cache *s, void *object)
260{
261 return *(void **)(object + s->offset);
262}
263
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500264static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
265{
266 void *p;
267
268#ifdef CONFIG_DEBUG_PAGEALLOC
269 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
270#else
271 p = get_freepointer(s, object);
272#endif
273 return p;
274}
275
Christoph Lameter7656c722007-05-09 02:32:40 -0700276static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
277{
278 *(void **)(object + s->offset) = fp;
279}
280
281/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300282#define for_each_object(__p, __s, __addr, __objects) \
283 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700284 __p += (__s)->size)
285
Christoph Lameter7656c722007-05-09 02:32:40 -0700286/* Determine object index from a given position */
287static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
288{
289 return (p - addr) / s->size;
290}
291
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100292static inline size_t slab_ksize(const struct kmem_cache *s)
293{
294#ifdef CONFIG_SLUB_DEBUG
295 /*
296 * Debugging requires use of the padding between object
297 * and whatever may come after it.
298 */
299 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
300 return s->objsize;
301
302#endif
303 /*
304 * If we have the need to store the freelist pointer
305 * back there or track user information then we can
306 * only use the space before that information.
307 */
308 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
309 return s->inuse;
310 /*
311 * Else we can use all the padding etc for the allocation
312 */
313 return s->size;
314}
315
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800316static inline int order_objects(int order, unsigned long size, int reserved)
317{
318 return ((PAGE_SIZE << order) - reserved) / size;
319}
320
Christoph Lameter834f3d12008-04-14 19:11:31 +0300321static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800322 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300323{
324 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800325 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300326 };
327
328 return x;
329}
330
331static inline int oo_order(struct kmem_cache_order_objects x)
332{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400333 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300334}
335
336static inline int oo_objects(struct kmem_cache_order_objects x)
337{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400338 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300339}
340
Christoph Lameter6332aa92011-05-16 15:26:06 -0500341#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter5f80b132011-04-15 14:48:13 -0500342/*
343 * Determine a map of object in use on a page.
344 *
345 * Slab lock or node listlock must be held to guarantee that the page does
346 * not vanish from under us.
347 */
348static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
349{
350 void *p;
351 void *addr = page_address(page);
352
353 for (p = page->freelist; p; p = get_freepointer(s, p))
354 set_bit(slab_index(p, s, addr), map);
355}
356
Christoph Lameter41ecc552007-05-09 02:32:44 -0700357/*
358 * Debug settings:
359 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700360#ifdef CONFIG_SLUB_DEBUG_ON
361static int slub_debug = DEBUG_DEFAULT_FLAGS;
362#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700363static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700364#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700365
366static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700367static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700368
Christoph Lameter7656c722007-05-09 02:32:40 -0700369/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700370 * Object debugging
371 */
372static void print_section(char *text, u8 *addr, unsigned int length)
373{
374 int i, offset;
375 int newline = 1;
376 char ascii[17];
377
378 ascii[16] = 0;
379
380 for (i = 0; i < length; i++) {
381 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700382 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700383 newline = 0;
384 }
Pekka Enberg06428782008-01-07 23:20:27 -0800385 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700386 offset = i % 16;
387 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
388 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800389 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700390 newline = 1;
391 }
392 }
393 if (!newline) {
394 i %= 16;
395 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800396 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700397 ascii[i] = ' ';
398 i++;
399 }
Pekka Enberg06428782008-01-07 23:20:27 -0800400 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700401 }
402}
403
Christoph Lameter81819f02007-05-06 14:49:36 -0700404static struct track *get_track(struct kmem_cache *s, void *object,
405 enum track_item alloc)
406{
407 struct track *p;
408
409 if (s->offset)
410 p = object + s->offset + sizeof(void *);
411 else
412 p = object + s->inuse;
413
414 return p + alloc;
415}
416
417static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300418 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700419{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900420 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700421
Christoph Lameter81819f02007-05-06 14:49:36 -0700422 if (addr) {
423 p->addr = addr;
424 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400425 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700426 p->when = jiffies;
427 } else
428 memset(p, 0, sizeof(struct track));
429}
430
Christoph Lameter81819f02007-05-06 14:49:36 -0700431static void init_tracking(struct kmem_cache *s, void *object)
432{
Christoph Lameter24922682007-07-17 04:03:18 -0700433 if (!(s->flags & SLAB_STORE_USER))
434 return;
435
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300436 set_track(s, object, TRACK_FREE, 0UL);
437 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700438}
439
440static void print_track(const char *s, struct track *t)
441{
442 if (!t->addr)
443 return;
444
Linus Torvalds7daf7052008-07-14 12:12:53 -0700445 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300446 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700447}
448
Christoph Lameter24922682007-07-17 04:03:18 -0700449static void print_tracking(struct kmem_cache *s, void *object)
450{
451 if (!(s->flags & SLAB_STORE_USER))
452 return;
453
454 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
455 print_track("Freed", get_track(s, object, TRACK_FREE));
456}
457
458static void print_page_info(struct page *page)
459{
Christoph Lameter39b26462008-04-14 19:11:30 +0300460 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
461 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700462
463}
464
465static void slab_bug(struct kmem_cache *s, char *fmt, ...)
466{
467 va_list args;
468 char buf[100];
469
470 va_start(args, fmt);
471 vsnprintf(buf, sizeof(buf), fmt, args);
472 va_end(args);
473 printk(KERN_ERR "========================================"
474 "=====================================\n");
475 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
476 printk(KERN_ERR "----------------------------------------"
477 "-------------------------------------\n\n");
478}
479
480static void slab_fix(struct kmem_cache *s, char *fmt, ...)
481{
482 va_list args;
483 char buf[100];
484
485 va_start(args, fmt);
486 vsnprintf(buf, sizeof(buf), fmt, args);
487 va_end(args);
488 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
489}
490
491static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700492{
493 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800494 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700495
496 print_tracking(s, p);
497
498 print_page_info(page);
499
500 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
501 p, p - addr, get_freepointer(s, p));
502
503 if (p > addr + 16)
504 print_section("Bytes b4", p - 16, 16);
505
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300506 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700507
508 if (s->flags & SLAB_RED_ZONE)
509 print_section("Redzone", p + s->objsize,
510 s->inuse - s->objsize);
511
Christoph Lameter81819f02007-05-06 14:49:36 -0700512 if (s->offset)
513 off = s->offset + sizeof(void *);
514 else
515 off = s->inuse;
516
Christoph Lameter24922682007-07-17 04:03:18 -0700517 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700518 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700519
520 if (off != s->size)
521 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700522 print_section("Padding", p + off, s->size - off);
523
524 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700525}
526
527static void object_err(struct kmem_cache *s, struct page *page,
528 u8 *object, char *reason)
529{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700530 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700531 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700532}
533
Christoph Lameter24922682007-07-17 04:03:18 -0700534static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700535{
536 va_list args;
537 char buf[100];
538
Christoph Lameter24922682007-07-17 04:03:18 -0700539 va_start(args, fmt);
540 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700541 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700542 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700543 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700544 dump_stack();
545}
546
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500547static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700548{
549 u8 *p = object;
550
551 if (s->flags & __OBJECT_POISON) {
552 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800553 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700554 }
555
556 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500557 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700558}
559
Christoph Lameter24922682007-07-17 04:03:18 -0700560static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700561{
562 while (bytes) {
563 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700564 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700565 start++;
566 bytes--;
567 }
Christoph Lameter24922682007-07-17 04:03:18 -0700568 return NULL;
569}
570
571static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
572 void *from, void *to)
573{
574 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
575 memset(from, data, to - from);
576}
577
578static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
579 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800580 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700581{
582 u8 *fault;
583 u8 *end;
584
585 fault = check_bytes(start, value, bytes);
586 if (!fault)
587 return 1;
588
589 end = start + bytes;
590 while (end > fault && end[-1] == value)
591 end--;
592
593 slab_bug(s, "%s overwritten", what);
594 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
595 fault, end - 1, fault[0], value);
596 print_trailer(s, page, object);
597
598 restore_bytes(s, what, value, fault, end);
599 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700600}
601
Christoph Lameter81819f02007-05-06 14:49:36 -0700602/*
603 * Object layout:
604 *
605 * object address
606 * Bytes of the object to be managed.
607 * If the freepointer may overlay the object then the free
608 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700609 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700610 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
611 * 0xa5 (POISON_END)
612 *
613 * object + s->objsize
614 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700615 * Padding is extended by another word if Redzoning is enabled and
616 * objsize == inuse.
617 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700618 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
619 * 0xcc (RED_ACTIVE) for objects in use.
620 *
621 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700622 * Meta data starts here.
623 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700624 * A. Free pointer (if we cannot overwrite object on free)
625 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700626 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800627 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700628 * before the word boundary.
629 *
630 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700631 *
632 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700633 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700634 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700635 * If slabcaches are merged then the objsize and inuse boundaries are mostly
636 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700637 * may be used with merged slabcaches.
638 */
639
Christoph Lameter81819f02007-05-06 14:49:36 -0700640static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
641{
642 unsigned long off = s->inuse; /* The end of info */
643
644 if (s->offset)
645 /* Freepointer is placed after the object. */
646 off += sizeof(void *);
647
648 if (s->flags & SLAB_STORE_USER)
649 /* We also have user information there */
650 off += 2 * sizeof(struct track);
651
652 if (s->size == off)
653 return 1;
654
Christoph Lameter24922682007-07-17 04:03:18 -0700655 return check_bytes_and_report(s, page, p, "Object padding",
656 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700657}
658
Christoph Lameter39b26462008-04-14 19:11:30 +0300659/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700660static int slab_pad_check(struct kmem_cache *s, struct page *page)
661{
Christoph Lameter24922682007-07-17 04:03:18 -0700662 u8 *start;
663 u8 *fault;
664 u8 *end;
665 int length;
666 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700667
668 if (!(s->flags & SLAB_POISON))
669 return 1;
670
Christoph Lametera973e9d2008-03-01 13:40:44 -0800671 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800672 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300673 end = start + length;
674 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700675 if (!remainder)
676 return 1;
677
Christoph Lameter39b26462008-04-14 19:11:30 +0300678 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700679 if (!fault)
680 return 1;
681 while (end > fault && end[-1] == POISON_INUSE)
682 end--;
683
684 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300685 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700686
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200687 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700688 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700689}
690
691static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500692 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700693{
694 u8 *p = object;
695 u8 *endobject = object + s->objsize;
696
697 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700698 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500699 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700700 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700701 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800702 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
703 check_bytes_and_report(s, page, p, "Alignment padding",
704 endobject, POISON_INUSE, s->inuse - s->objsize);
705 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700706 }
707
708 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500709 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700710 (!check_bytes_and_report(s, page, p, "Poison", p,
711 POISON_FREE, s->objsize - 1) ||
712 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800713 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700714 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700715 /*
716 * check_pad_bytes cleans up on its own.
717 */
718 check_pad_bytes(s, page, p);
719 }
720
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500721 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700722 /*
723 * Object and freepointer overlap. Cannot check
724 * freepointer while object is allocated.
725 */
726 return 1;
727
728 /* Check free pointer validity */
729 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
730 object_err(s, page, p, "Freepointer corrupt");
731 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100732 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700733 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700734 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700735 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800736 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700737 return 0;
738 }
739 return 1;
740}
741
742static int check_slab(struct kmem_cache *s, struct page *page)
743{
Christoph Lameter39b26462008-04-14 19:11:30 +0300744 int maxobj;
745
Christoph Lameter81819f02007-05-06 14:49:36 -0700746 VM_BUG_ON(!irqs_disabled());
747
748 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700749 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700750 return 0;
751 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300752
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800753 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300754 if (page->objects > maxobj) {
755 slab_err(s, page, "objects %u > max %u",
756 s->name, page->objects, maxobj);
757 return 0;
758 }
759 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700760 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300761 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700762 return 0;
763 }
764 /* Slab_pad_check fixes things up after itself */
765 slab_pad_check(s, page);
766 return 1;
767}
768
769/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700770 * Determine if a certain object on a page is on the freelist. Must hold the
771 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700772 */
773static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
774{
775 int nr = 0;
776 void *fp = page->freelist;
777 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300778 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700779
Christoph Lameter39b26462008-04-14 19:11:30 +0300780 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700781 if (fp == search)
782 return 1;
783 if (!check_valid_pointer(s, page, fp)) {
784 if (object) {
785 object_err(s, page, object,
786 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800787 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700788 break;
789 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700790 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800791 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300792 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700793 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700794 return 0;
795 }
796 break;
797 }
798 object = fp;
799 fp = get_freepointer(s, object);
800 nr++;
801 }
802
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800803 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400804 if (max_objects > MAX_OBJS_PER_PAGE)
805 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300806
807 if (page->objects != max_objects) {
808 slab_err(s, page, "Wrong number of objects. Found %d but "
809 "should be %d", page->objects, max_objects);
810 page->objects = max_objects;
811 slab_fix(s, "Number of objects adjusted.");
812 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300813 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700814 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300815 "counted were %d", page->inuse, page->objects - nr);
816 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700817 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700818 }
819 return search == NULL;
820}
821
Christoph Lameter0121c6192008-04-29 16:11:12 -0700822static void trace(struct kmem_cache *s, struct page *page, void *object,
823 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700824{
825 if (s->flags & SLAB_TRACE) {
826 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
827 s->name,
828 alloc ? "alloc" : "free",
829 object, page->inuse,
830 page->freelist);
831
832 if (!alloc)
833 print_section("Object", (void *)object, s->objsize);
834
835 dump_stack();
836 }
837}
838
Christoph Lameter643b1132007-05-06 14:49:42 -0700839/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500840 * Hooks for other subsystems that check memory allocations. In a typical
841 * production configuration these hooks all should produce no code at all.
842 */
843static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
844{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500845 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500846 lockdep_trace_alloc(flags);
847 might_sleep_if(flags & __GFP_WAIT);
848
849 return should_failslab(s->objsize, flags, s->flags);
850}
851
852static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
853{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500854 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100855 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500856 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
857}
858
859static inline void slab_free_hook(struct kmem_cache *s, void *x)
860{
861 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500862
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600863 /*
864 * Trouble is that we may no longer disable interupts in the fast path
865 * So in order to make the debug calls that expect irqs to be
866 * disabled we need to disable interrupts temporarily.
867 */
868#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
869 {
870 unsigned long flags;
871
872 local_irq_save(flags);
873 kmemcheck_slab_free(s, x, s->objsize);
874 debug_check_no_locks_freed(x, s->objsize);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600875 local_irq_restore(flags);
876 }
877#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200878 if (!(s->flags & SLAB_DEBUG_OBJECTS))
879 debug_check_no_obj_freed(x, s->objsize);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500880}
881
882/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700883 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700884 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700885static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700886{
Christoph Lameter643b1132007-05-06 14:49:42 -0700887 spin_lock(&n->list_lock);
888 list_add(&page->lru, &n->full);
889 spin_unlock(&n->list_lock);
890}
891
892static void remove_full(struct kmem_cache *s, struct page *page)
893{
894 struct kmem_cache_node *n;
895
896 if (!(s->flags & SLAB_STORE_USER))
897 return;
898
899 n = get_node(s, page_to_nid(page));
900
901 spin_lock(&n->list_lock);
902 list_del(&page->lru);
903 spin_unlock(&n->list_lock);
904}
905
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300906/* Tracking of the number of slabs for debugging purposes */
907static inline unsigned long slabs_node(struct kmem_cache *s, int node)
908{
909 struct kmem_cache_node *n = get_node(s, node);
910
911 return atomic_long_read(&n->nr_slabs);
912}
913
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400914static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
915{
916 return atomic_long_read(&n->nr_slabs);
917}
918
Christoph Lameter205ab992008-04-14 19:11:40 +0300919static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300920{
921 struct kmem_cache_node *n = get_node(s, node);
922
923 /*
924 * May be called early in order to allocate a slab for the
925 * kmem_cache_node structure. Solve the chicken-egg
926 * dilemma by deferring the increment of the count during
927 * bootstrap (see early_kmem_cache_node_alloc).
928 */
Christoph Lameter7340cc82010-09-28 08:10:26 -0500929 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300930 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300931 atomic_long_add(objects, &n->total_objects);
932 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300933}
Christoph Lameter205ab992008-04-14 19:11:40 +0300934static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300935{
936 struct kmem_cache_node *n = get_node(s, node);
937
938 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300939 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300940}
941
942/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700943static void setup_object_debug(struct kmem_cache *s, struct page *page,
944 void *object)
945{
946 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
947 return;
948
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500949 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700950 init_tracking(s, object);
951}
952
Christoph Lameter15370662010-08-20 12:37:12 -0500953static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300954 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700955{
956 if (!check_slab(s, page))
957 goto bad;
958
Christoph Lameterd692ef62008-02-15 23:45:24 -0800959 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700960 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700961 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700962 }
963
964 if (!check_valid_pointer(s, page, object)) {
965 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700966 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700967 }
968
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500969 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -0700970 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700971
Christoph Lameter3ec09742007-05-16 22:11:00 -0700972 /* Success perform special debug activities for allocs */
973 if (s->flags & SLAB_STORE_USER)
974 set_track(s, object, TRACK_ALLOC, addr);
975 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500976 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700977 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700978
Christoph Lameter81819f02007-05-06 14:49:36 -0700979bad:
980 if (PageSlab(page)) {
981 /*
982 * If this is a slab page then lets do the best we can
983 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700984 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700985 */
Christoph Lameter24922682007-07-17 04:03:18 -0700986 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300987 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800988 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700989 }
990 return 0;
991}
992
Christoph Lameter15370662010-08-20 12:37:12 -0500993static noinline int free_debug_processing(struct kmem_cache *s,
994 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700995{
996 if (!check_slab(s, page))
997 goto fail;
998
999 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001000 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001001 goto fail;
1002 }
1003
1004 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001005 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001006 goto fail;
1007 }
1008
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001009 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001010 return 0;
1011
1012 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001013 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001014 slab_err(s, page, "Attempt to free object(0x%p) "
1015 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001016 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001017 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001018 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001019 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001020 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001021 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001022 object_err(s, page, object,
1023 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001024 goto fail;
1025 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001026
1027 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -07001028 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -07001029 remove_full(s, page);
1030 if (s->flags & SLAB_STORE_USER)
1031 set_track(s, object, TRACK_FREE, addr);
1032 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001033 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001034 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001035
Christoph Lameter81819f02007-05-06 14:49:36 -07001036fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001037 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001038 return 0;
1039}
1040
Christoph Lameter41ecc552007-05-09 02:32:44 -07001041static int __init setup_slub_debug(char *str)
1042{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001043 slub_debug = DEBUG_DEFAULT_FLAGS;
1044 if (*str++ != '=' || !*str)
1045 /*
1046 * No options specified. Switch on full debugging.
1047 */
1048 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001049
1050 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001051 /*
1052 * No options but restriction on slabs. This means full
1053 * debugging for slabs matching a pattern.
1054 */
1055 goto check_slabs;
1056
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001057 if (tolower(*str) == 'o') {
1058 /*
1059 * Avoid enabling debugging on caches if its minimum order
1060 * would increase as a result.
1061 */
1062 disable_higher_order_debug = 1;
1063 goto out;
1064 }
1065
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001066 slub_debug = 0;
1067 if (*str == '-')
1068 /*
1069 * Switch off all debugging measures.
1070 */
1071 goto out;
1072
1073 /*
1074 * Determine which debug features should be switched on
1075 */
Pekka Enberg06428782008-01-07 23:20:27 -08001076 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001077 switch (tolower(*str)) {
1078 case 'f':
1079 slub_debug |= SLAB_DEBUG_FREE;
1080 break;
1081 case 'z':
1082 slub_debug |= SLAB_RED_ZONE;
1083 break;
1084 case 'p':
1085 slub_debug |= SLAB_POISON;
1086 break;
1087 case 'u':
1088 slub_debug |= SLAB_STORE_USER;
1089 break;
1090 case 't':
1091 slub_debug |= SLAB_TRACE;
1092 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001093 case 'a':
1094 slub_debug |= SLAB_FAILSLAB;
1095 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001096 default:
1097 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001098 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001099 }
1100 }
1101
1102check_slabs:
1103 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001104 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001105out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001106 return 1;
1107}
1108
1109__setup("slub_debug", setup_slub_debug);
1110
Christoph Lameterba0268a2007-09-11 15:24:11 -07001111static unsigned long kmem_cache_flags(unsigned long objsize,
1112 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001113 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001114{
1115 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001116 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001117 */
Christoph Lametere1533622008-02-15 23:45:24 -08001118 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001119 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1120 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001121
1122 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001123}
1124#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001125static inline void setup_object_debug(struct kmem_cache *s,
1126 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001127
Christoph Lameter3ec09742007-05-16 22:11:00 -07001128static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001129 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001130
Christoph Lameter3ec09742007-05-16 22:11:00 -07001131static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001132 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001133
Christoph Lameter41ecc552007-05-09 02:32:44 -07001134static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1135 { return 1; }
1136static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001137 void *object, u8 val) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001138static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001139static inline unsigned long kmem_cache_flags(unsigned long objsize,
1140 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001141 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001142{
1143 return flags;
1144}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001145#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001146
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001147#define disable_higher_order_debug 0
1148
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001149static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1150 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001151static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1152 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001153static inline void inc_slabs_node(struct kmem_cache *s, int node,
1154 int objects) {}
1155static inline void dec_slabs_node(struct kmem_cache *s, int node,
1156 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001157
1158static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1159 { return 0; }
1160
1161static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1162 void *object) {}
1163
1164static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1165
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001166#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001167
Christoph Lameter81819f02007-05-06 14:49:36 -07001168/*
1169 * Slab allocation and freeing
1170 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001171static inline struct page *alloc_slab_page(gfp_t flags, int node,
1172 struct kmem_cache_order_objects oo)
1173{
1174 int order = oo_order(oo);
1175
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001176 flags |= __GFP_NOTRACK;
1177
Christoph Lameter2154a332010-07-09 14:07:10 -05001178 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001179 return alloc_pages(flags, order);
1180 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001181 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001182}
1183
Christoph Lameter81819f02007-05-06 14:49:36 -07001184static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1185{
Pekka Enberg06428782008-01-07 23:20:27 -08001186 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001187 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001188 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001189
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001190 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001191
Pekka Enbergba522702009-06-24 21:59:51 +03001192 /*
1193 * Let the initial higher-order allocation fail under memory pressure
1194 * so we fall-back to the minimum order allocation.
1195 */
1196 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1197
1198 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001199 if (unlikely(!page)) {
1200 oo = s->min;
1201 /*
1202 * Allocation may have failed due to fragmentation.
1203 * Try a lower order alloc if possible
1204 */
1205 page = alloc_slab_page(flags, node, oo);
1206 if (!page)
1207 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001208
Christoph Lameter84e554e2009-12-18 16:26:23 -06001209 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001210 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001211
1212 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001213 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001214 int pages = 1 << oo_order(oo);
1215
1216 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1217
1218 /*
1219 * Objects from caches that have a constructor don't get
1220 * cleared when they're allocated, so we need to do it here.
1221 */
1222 if (s->ctor)
1223 kmemcheck_mark_uninitialized_pages(page, pages);
1224 else
1225 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001226 }
1227
Christoph Lameter834f3d12008-04-14 19:11:31 +03001228 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001229 mod_zone_page_state(page_zone(page),
1230 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1231 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001232 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001233
1234 return page;
1235}
1236
1237static void setup_object(struct kmem_cache *s, struct page *page,
1238 void *object)
1239{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001240 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001241 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001242 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001243}
1244
1245static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1246{
1247 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001248 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001249 void *last;
1250 void *p;
1251
Christoph Lameter6cb06222007-10-16 01:25:41 -07001252 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001253
Christoph Lameter6cb06222007-10-16 01:25:41 -07001254 page = allocate_slab(s,
1255 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001256 if (!page)
1257 goto out;
1258
Christoph Lameter205ab992008-04-14 19:11:40 +03001259 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001260 page->slab = s;
1261 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001262
1263 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001264
1265 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001266 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001267
1268 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001269 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001270 setup_object(s, page, last);
1271 set_freepointer(s, last, p);
1272 last = p;
1273 }
1274 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001275 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001276
1277 page->freelist = start;
1278 page->inuse = 0;
1279out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001280 return page;
1281}
1282
1283static void __free_slab(struct kmem_cache *s, struct page *page)
1284{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001285 int order = compound_order(page);
1286 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001287
Christoph Lameteraf537b02010-07-09 14:07:14 -05001288 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001289 void *p;
1290
1291 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001292 for_each_object(p, s, page_address(page),
1293 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001294 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001295 }
1296
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001297 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001298
Christoph Lameter81819f02007-05-06 14:49:36 -07001299 mod_zone_page_state(page_zone(page),
1300 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1301 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001302 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001303
Christoph Lameter49bd5222008-04-14 18:52:18 +03001304 __ClearPageSlab(page);
1305 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001306 if (current->reclaim_state)
1307 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001308 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001309}
1310
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001311#define need_reserve_slab_rcu \
1312 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1313
Christoph Lameter81819f02007-05-06 14:49:36 -07001314static void rcu_free_slab(struct rcu_head *h)
1315{
1316 struct page *page;
1317
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001318 if (need_reserve_slab_rcu)
1319 page = virt_to_head_page(h);
1320 else
1321 page = container_of((struct list_head *)h, struct page, lru);
1322
Christoph Lameter81819f02007-05-06 14:49:36 -07001323 __free_slab(page->slab, page);
1324}
1325
1326static void free_slab(struct kmem_cache *s, struct page *page)
1327{
1328 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001329 struct rcu_head *head;
1330
1331 if (need_reserve_slab_rcu) {
1332 int order = compound_order(page);
1333 int offset = (PAGE_SIZE << order) - s->reserved;
1334
1335 VM_BUG_ON(s->reserved != sizeof(*head));
1336 head = page_address(page) + offset;
1337 } else {
1338 /*
1339 * RCU free overloads the RCU head over the LRU
1340 */
1341 head = (void *)&page->lru;
1342 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001343
1344 call_rcu(head, rcu_free_slab);
1345 } else
1346 __free_slab(s, page);
1347}
1348
1349static void discard_slab(struct kmem_cache *s, struct page *page)
1350{
Christoph Lameter205ab992008-04-14 19:11:40 +03001351 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001352 free_slab(s, page);
1353}
1354
1355/*
1356 * Per slab locking using the pagelock
1357 */
1358static __always_inline void slab_lock(struct page *page)
1359{
1360 bit_spin_lock(PG_locked, &page->flags);
1361}
1362
1363static __always_inline void slab_unlock(struct page *page)
1364{
Nick Piggina76d3542008-01-07 23:20:27 -08001365 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001366}
1367
1368static __always_inline int slab_trylock(struct page *page)
1369{
1370 int rc = 1;
1371
1372 rc = bit_spin_trylock(PG_locked, &page->flags);
1373 return rc;
1374}
1375
1376/*
1377 * Management of partially allocated slabs
1378 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001379static void add_partial(struct kmem_cache_node *n,
1380 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001381{
Christoph Lametere95eed52007-05-06 14:49:44 -07001382 spin_lock(&n->list_lock);
1383 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001384 if (tail)
1385 list_add_tail(&page->lru, &n->partial);
1386 else
1387 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001388 spin_unlock(&n->list_lock);
1389}
1390
Christoph Lameter62e346a2010-09-28 08:10:28 -05001391static inline void __remove_partial(struct kmem_cache_node *n,
1392 struct page *page)
1393{
1394 list_del(&page->lru);
1395 n->nr_partial--;
1396}
1397
Christoph Lameter0121c6192008-04-29 16:11:12 -07001398static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001399{
1400 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1401
1402 spin_lock(&n->list_lock);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001403 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001404 spin_unlock(&n->list_lock);
1405}
1406
1407/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001408 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001409 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001410 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001411 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001412static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1413 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001414{
1415 if (slab_trylock(page)) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05001416 __remove_partial(n, page);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001417 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001418 return 1;
1419 }
1420 return 0;
1421}
1422
1423/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001424 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001425 */
1426static struct page *get_partial_node(struct kmem_cache_node *n)
1427{
1428 struct page *page;
1429
1430 /*
1431 * Racy check. If we mistakenly see no partial slabs then we
1432 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001433 * partial slab and there is none available then get_partials()
1434 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001435 */
1436 if (!n || !n->nr_partial)
1437 return NULL;
1438
1439 spin_lock(&n->list_lock);
1440 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001441 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001442 goto out;
1443 page = NULL;
1444out:
1445 spin_unlock(&n->list_lock);
1446 return page;
1447}
1448
1449/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001450 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001451 */
1452static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1453{
1454#ifdef CONFIG_NUMA
1455 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001456 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001457 struct zone *zone;
1458 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001459 struct page *page;
1460
1461 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001462 * The defrag ratio allows a configuration of the tradeoffs between
1463 * inter node defragmentation and node local allocations. A lower
1464 * defrag_ratio increases the tendency to do local allocations
1465 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001466 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001467 * If the defrag_ratio is set to 0 then kmalloc() always
1468 * returns node local objects. If the ratio is higher then kmalloc()
1469 * may return off node objects because partial slabs are obtained
1470 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001471 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001472 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001473 * defrag_ratio = 1000) then every (well almost) allocation will
1474 * first attempt to defrag slab caches on other nodes. This means
1475 * scanning over all nodes to look for partial slabs which may be
1476 * expensive if we do it every time we are trying to find a slab
1477 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001478 */
Christoph Lameter98246012008-01-07 23:20:26 -08001479 if (!s->remote_node_defrag_ratio ||
1480 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001481 return NULL;
1482
Miao Xiec0ff7452010-05-24 14:32:08 -07001483 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001484 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001485 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001486 struct kmem_cache_node *n;
1487
Mel Gorman54a6eb52008-04-28 02:12:16 -07001488 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001489
Mel Gorman54a6eb52008-04-28 02:12:16 -07001490 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001491 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001492 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001493 if (page) {
1494 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001495 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001496 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001497 }
1498 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001499 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001500#endif
1501 return NULL;
1502}
1503
1504/*
1505 * Get a partial page, lock it and return it.
1506 */
1507static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1508{
1509 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001510 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001511
1512 page = get_partial_node(get_node(s, searchnode));
Christoph Lameter33de04e2011-04-15 14:48:12 -05001513 if (page || node != NUMA_NO_NODE)
Christoph Lameter81819f02007-05-06 14:49:36 -07001514 return page;
1515
1516 return get_any_partial(s, flags);
1517}
1518
1519/*
1520 * Move a page back to the lists.
1521 *
1522 * Must be called with the slab lock held.
1523 *
1524 * On exit the slab lock will have been dropped.
1525 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001526static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Namhyung Kim34789732010-09-29 21:02:14 +09001527 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001528{
Christoph Lametere95eed52007-05-06 14:49:44 -07001529 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1530
Andy Whitcroft8a380822008-07-23 21:27:18 -07001531 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001532 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001533
Christoph Lametera973e9d2008-03-01 13:40:44 -08001534 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001535 add_partial(n, page, tail);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001536 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001537 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001538 stat(s, DEACTIVATE_FULL);
Christoph Lameteraf537b02010-07-09 14:07:14 -05001539 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001540 add_full(n, page);
1541 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001542 slab_unlock(page);
1543 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001544 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001545 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001546 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001547 * Adding an empty slab to the partial slabs in order
1548 * to avoid page allocator overhead. This slab needs
1549 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001550 * so that the others get filled first. That way the
1551 * size of the partial list stays small.
1552 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001553 * kmem_cache_shrink can reclaim any empty slabs from
1554 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001555 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001556 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001557 slab_unlock(page);
1558 } else {
1559 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001560 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001561 discard_slab(s, page);
1562 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001563 }
1564}
1565
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001566#ifdef CONFIG_PREEMPT
1567/*
1568 * Calculate the next globally unique transaction for disambiguiation
1569 * during cmpxchg. The transactions start with the cpu number and are then
1570 * incremented by CONFIG_NR_CPUS.
1571 */
1572#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1573#else
1574/*
1575 * No preemption supported therefore also no need to check for
1576 * different cpus.
1577 */
1578#define TID_STEP 1
1579#endif
1580
1581static inline unsigned long next_tid(unsigned long tid)
1582{
1583 return tid + TID_STEP;
1584}
1585
1586static inline unsigned int tid_to_cpu(unsigned long tid)
1587{
1588 return tid % TID_STEP;
1589}
1590
1591static inline unsigned long tid_to_event(unsigned long tid)
1592{
1593 return tid / TID_STEP;
1594}
1595
1596static inline unsigned int init_tid(int cpu)
1597{
1598 return cpu;
1599}
1600
1601static inline void note_cmpxchg_failure(const char *n,
1602 const struct kmem_cache *s, unsigned long tid)
1603{
1604#ifdef SLUB_DEBUG_CMPXCHG
1605 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1606
1607 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1608
1609#ifdef CONFIG_PREEMPT
1610 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1611 printk("due to cpu change %d -> %d\n",
1612 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1613 else
1614#endif
1615 if (tid_to_event(tid) != tid_to_event(actual_tid))
1616 printk("due to cpu running other code. Event %ld->%ld\n",
1617 tid_to_event(tid), tid_to_event(actual_tid));
1618 else
1619 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1620 actual_tid, tid, next_tid(tid));
1621#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001622 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001623}
1624
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001625void init_kmem_cache_cpus(struct kmem_cache *s)
1626{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001627 int cpu;
1628
1629 for_each_possible_cpu(cpu)
1630 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001631}
Christoph Lameter81819f02007-05-06 14:49:36 -07001632/*
1633 * Remove the cpu slab
1634 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001635static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Namhyung Kim34789732010-09-29 21:02:14 +09001636 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001637{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001638 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001639 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001640
Christoph Lameterb773ad72008-03-04 11:10:17 -08001641 if (page->freelist)
Christoph Lameter84e554e2009-12-18 16:26:23 -06001642 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001643 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001644 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001645 * because both freelists are empty. So this is unlikely
1646 * to occur.
1647 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001648 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001649 void **object;
1650
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001651 tail = 0; /* Hot objects. Put the slab first */
1652
Christoph Lameter894b8782007-05-10 03:15:16 -07001653 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001654 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001655 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001656
1657 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001658 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001659 page->freelist = object;
1660 page->inuse--;
1661 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001662 c->page = NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001663 c->tid = next_tid(c->tid);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001664 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001665}
1666
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001667static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001668{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001669 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001670 slab_lock(c->page);
1671 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001672}
1673
1674/*
1675 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001676 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001677 * Called from IPI handler with interrupts disabled.
1678 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001679static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001680{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001681 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001682
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001683 if (likely(c && c->page))
1684 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001685}
1686
1687static void flush_cpu_slab(void *d)
1688{
1689 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001690
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001691 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001692}
1693
1694static void flush_all(struct kmem_cache *s)
1695{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001696 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001697}
1698
1699/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001700 * Check if the objects in a per cpu structure fit numa
1701 * locality expectations.
1702 */
1703static inline int node_match(struct kmem_cache_cpu *c, int node)
1704{
1705#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001706 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001707 return 0;
1708#endif
1709 return 1;
1710}
1711
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001712static int count_free(struct page *page)
1713{
1714 return page->objects - page->inuse;
1715}
1716
1717static unsigned long count_partial(struct kmem_cache_node *n,
1718 int (*get_count)(struct page *))
1719{
1720 unsigned long flags;
1721 unsigned long x = 0;
1722 struct page *page;
1723
1724 spin_lock_irqsave(&n->list_lock, flags);
1725 list_for_each_entry(page, &n->partial, lru)
1726 x += get_count(page);
1727 spin_unlock_irqrestore(&n->list_lock, flags);
1728 return x;
1729}
1730
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001731static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1732{
1733#ifdef CONFIG_SLUB_DEBUG
1734 return atomic_long_read(&n->total_objects);
1735#else
1736 return 0;
1737#endif
1738}
1739
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001740static noinline void
1741slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1742{
1743 int node;
1744
1745 printk(KERN_WARNING
1746 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1747 nid, gfpflags);
1748 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1749 "default order: %d, min order: %d\n", s->name, s->objsize,
1750 s->size, oo_order(s->oo), oo_order(s->min));
1751
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001752 if (oo_order(s->min) > get_order(s->objsize))
1753 printk(KERN_WARNING " %s debugging increased min order, use "
1754 "slub_debug=O to disable.\n", s->name);
1755
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001756 for_each_online_node(node) {
1757 struct kmem_cache_node *n = get_node(s, node);
1758 unsigned long nr_slabs;
1759 unsigned long nr_objs;
1760 unsigned long nr_free;
1761
1762 if (!n)
1763 continue;
1764
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001765 nr_free = count_partial(n, count_free);
1766 nr_slabs = node_nr_slabs(n);
1767 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001768
1769 printk(KERN_WARNING
1770 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1771 node, nr_slabs, nr_objs, nr_free);
1772 }
1773}
1774
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001775/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001776 * Slow path. The lockless freelist is empty or we need to perform
1777 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001778 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001779 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001780 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001781 * Processing is still very fast if new objects have been freed to the
1782 * regular freelist. In that case we simply take over the regular freelist
1783 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001784 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001785 * If that is not working then we fall back to the partial lists. We take the
1786 * first element of the freelist as the object to allocate now and move the
1787 * rest of the freelist to the lockless freelist.
1788 *
1789 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001790 * we need to allocate a new slab. This is the slowest path since it involves
1791 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001792 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001793static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1794 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001795{
Christoph Lameter81819f02007-05-06 14:49:36 -07001796 void **object;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001797 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001798 unsigned long flags;
1799
1800 local_irq_save(flags);
1801#ifdef CONFIG_PREEMPT
1802 /*
1803 * We may have been preempted and rescheduled on a different
1804 * cpu before disabling interrupts. Need to reload cpu area
1805 * pointer.
1806 */
1807 c = this_cpu_ptr(s->cpu_slab);
1808#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07001809
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001810 /* We handle __GFP_ZERO in the caller */
1811 gfpflags &= ~__GFP_ZERO;
1812
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001813 page = c->page;
1814 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001815 goto new_slab;
1816
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001817 slab_lock(page);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001818 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001819 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001820
Christoph Lameter84e554e2009-12-18 16:26:23 -06001821 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001822
Christoph Lameter894b8782007-05-10 03:15:16 -07001823load_freelist:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001824 object = page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001825 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001826 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001827 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001828 goto debug;
1829
Christoph Lameterff120592009-12-18 16:26:22 -06001830 c->freelist = get_freepointer(s, object);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001831 page->inuse = page->objects;
1832 page->freelist = NULL;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001833
Christoph Lameter1f842602008-01-07 23:20:30 -08001834unlock_out:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001835 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001836 c->tid = next_tid(c->tid);
1837 local_irq_restore(flags);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001838 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001839 return object;
1840
1841another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001842 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001843
1844new_slab:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001845 page = get_partial(s, gfpflags, node);
1846 if (page) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001847 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameterdc1fb7f2011-04-15 14:48:15 -05001848load_from_page:
1849 c->node = page_to_nid(page);
1850 c->page = page;
Christoph Lameter894b8782007-05-10 03:15:16 -07001851 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001852 }
1853
Christoph Lameterc1d50832010-08-20 12:37:17 -05001854 gfpflags &= gfp_allowed_mask;
Christoph Lameterb811c202007-10-16 23:25:51 -07001855 if (gfpflags & __GFP_WAIT)
1856 local_irq_enable();
1857
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001858 page = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001859
1860 if (gfpflags & __GFP_WAIT)
1861 local_irq_disable();
1862
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001863 if (page) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001864 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001865 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001866 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001867 flush_slab(s, c);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001868
1869 slab_lock(page);
1870 __SetPageSlubFrozen(page);
Christoph Lameterdc1fb7f2011-04-15 14:48:15 -05001871
1872 goto load_from_page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001873 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001874 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1875 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter2fd66c52011-03-22 13:32:53 -05001876 local_irq_restore(flags);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001877 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001878debug:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001879 if (!alloc_debug_processing(s, page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001880 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001881
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001882 page->inuse++;
1883 page->freelist = get_freepointer(s, object);
Pekka Enberg15b7c512010-10-02 11:32:32 +03001884 c->node = NUMA_NO_NODE;
Christoph Lameter1f842602008-01-07 23:20:30 -08001885 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001886}
1887
1888/*
1889 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1890 * have the fastpath folded into their functions. So no function call
1891 * overhead for requests that can be satisfied on the fastpath.
1892 *
1893 * The fastpath works by first checking if the lockless freelist can be used.
1894 * If not then __slab_alloc is called for slow processing.
1895 *
1896 * Otherwise we can simply pick the next object from the lockless free list.
1897 */
Pekka Enberg06428782008-01-07 23:20:27 -08001898static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001899 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001900{
Christoph Lameter894b8782007-05-10 03:15:16 -07001901 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001902 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001903 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08001904
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001905 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001906 return NULL;
1907
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001908redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001909
1910 /*
1911 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1912 * enabled. We may switch back and forth between cpus while
1913 * reading from one cpu area. That does not matter as long
1914 * as we end up on the original cpu again when doing the cmpxchg.
1915 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001916 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001917
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001918 /*
1919 * The transaction ids are globally unique per cpu and per operation on
1920 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1921 * occurs on the right processor and that there was no operation on the
1922 * linked list in between.
1923 */
1924 tid = c->tid;
1925 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001926
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001927 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001928 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001929
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001930 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001931
1932 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001933 /*
1934 * The cmpxchg will only match if there was no additonal
1935 * operation and if we are on the right processor.
1936 *
1937 * The cmpxchg does the following atomically (without lock semantics!)
1938 * 1. Relocate first pointer to the current per cpu area.
1939 * 2. Verify that tid and freelist have not been changed
1940 * 3. If they were not changed replace tid and freelist
1941 *
1942 * Since this is without lock semantics the protection is only against
1943 * code executing on this cpu *not* from access by other cpus.
1944 */
1945 if (unlikely(!this_cpu_cmpxchg_double(
1946 s->cpu_slab->freelist, s->cpu_slab->tid,
1947 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05001948 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001949
1950 note_cmpxchg_failure("slab_alloc", s, tid);
1951 goto redo;
1952 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06001953 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001954 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001955
Pekka Enberg74e21342009-11-25 20:14:48 +02001956 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001957 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001958
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001959 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001960
Christoph Lameter894b8782007-05-10 03:15:16 -07001961 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001962}
1963
1964void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1965{
Christoph Lameter2154a332010-07-09 14:07:10 -05001966 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001967
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001968 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001969
1970 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001971}
1972EXPORT_SYMBOL(kmem_cache_alloc);
1973
Li Zefan0f24f122009-12-11 15:45:30 +08001974#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01001975void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001976{
Richard Kennedy4a923792010-10-21 10:29:19 +01001977 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1978 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1979 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001980}
Richard Kennedy4a923792010-10-21 10:29:19 +01001981EXPORT_SYMBOL(kmem_cache_alloc_trace);
1982
1983void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1984{
1985 void *ret = kmalloc_order(size, flags, order);
1986 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1987 return ret;
1988}
1989EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001990#endif
1991
Christoph Lameter81819f02007-05-06 14:49:36 -07001992#ifdef CONFIG_NUMA
1993void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1994{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001995 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1996
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001997 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1998 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001999
2000 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002001}
2002EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002003
Li Zefan0f24f122009-12-11 15:45:30 +08002004#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002005void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002006 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002007 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002008{
Richard Kennedy4a923792010-10-21 10:29:19 +01002009 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2010
2011 trace_kmalloc_node(_RET_IP_, ret,
2012 size, s->size, gfpflags, node);
2013 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002014}
Richard Kennedy4a923792010-10-21 10:29:19 +01002015EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002016#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002017#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002018
Christoph Lameter81819f02007-05-06 14:49:36 -07002019/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002020 * Slow patch handling. This may still be called frequently since objects
2021 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002022 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002023 * So we still attempt to reduce cache line usage. Just take the slab
2024 * lock and free the item. If there is no additional partial page
2025 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002026 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002027static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002028 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002029{
2030 void *prior;
2031 void **object = (void *)x;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002032 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002033
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002034 local_irq_save(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002035 slab_lock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002036 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002037
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002038 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2039 goto out_unlock;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002040
Christoph Lameterff120592009-12-18 16:26:22 -06002041 prior = page->freelist;
2042 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07002043 page->freelist = object;
2044 page->inuse--;
2045
Andy Whitcroft8a380822008-07-23 21:27:18 -07002046 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06002047 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07002048 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002049 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002050
2051 if (unlikely(!page->inuse))
2052 goto slab_empty;
2053
2054 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002055 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07002056 * then add it.
2057 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08002058 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002059 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002060 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002061 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002062
2063out_unlock:
2064 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002065 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002066 return;
2067
2068slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002069 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002070 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002071 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002072 */
2073 remove_partial(s, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002074 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002075 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002076 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002077 local_irq_restore(flags);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002078 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002079 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002080}
2081
Christoph Lameter894b8782007-05-10 03:15:16 -07002082/*
2083 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2084 * can perform fastpath freeing without additional function calls.
2085 *
2086 * The fastpath is only possible if we are freeing to the current cpu slab
2087 * of this processor. This typically the case if we have just allocated
2088 * the item before.
2089 *
2090 * If fastpath is not possible then fall back to __slab_free where we deal
2091 * with all sorts of special processing.
2092 */
Pekka Enberg06428782008-01-07 23:20:27 -08002093static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002094 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002095{
2096 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002097 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002098 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002099
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002100 slab_free_hook(s, x);
2101
Christoph Lametera24c5a02011-03-15 12:45:21 -05002102redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002103
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002104 /*
2105 * Determine the currently cpus per cpu slab.
2106 * The cpu may change afterward. However that does not matter since
2107 * data is retrieved via this pointer. If we are on the same cpu
2108 * during the cmpxchg then the free will succedd.
2109 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002110 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002111
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002112 tid = c->tid;
2113 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002114
Pekka Enberg15b7c512010-10-02 11:32:32 +03002115 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002116 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002117
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002118 if (unlikely(!this_cpu_cmpxchg_double(
2119 s->cpu_slab->freelist, s->cpu_slab->tid,
2120 c->freelist, tid,
2121 object, next_tid(tid)))) {
2122
2123 note_cmpxchg_failure("slab_free", s, tid);
2124 goto redo;
2125 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06002126 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002127 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002128 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002129
Christoph Lameter894b8782007-05-10 03:15:16 -07002130}
2131
Christoph Lameter81819f02007-05-06 14:49:36 -07002132void kmem_cache_free(struct kmem_cache *s, void *x)
2133{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002134 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002135
Christoph Lameterb49af682007-05-06 14:49:41 -07002136 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002137
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002138 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002139
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002140 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002141}
2142EXPORT_SYMBOL(kmem_cache_free);
2143
Christoph Lameter81819f02007-05-06 14:49:36 -07002144/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002145 * Object placement in a slab is made very easy because we always start at
2146 * offset 0. If we tune the size of the object to the alignment then we can
2147 * get the required alignment by putting one properly sized object after
2148 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002149 *
2150 * Notice that the allocation order determines the sizes of the per cpu
2151 * caches. Each processor has always one slab available for allocations.
2152 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002153 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002154 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002155 */
2156
2157/*
2158 * Mininum / Maximum order of slab pages. This influences locking overhead
2159 * and slab fragmentation. A higher order reduces the number of partial slabs
2160 * and increases the number of allocations possible without having to
2161 * take the list_lock.
2162 */
2163static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002164static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002165static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002166
2167/*
2168 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002169 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002170 */
2171static int slub_nomerge;
2172
2173/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002174 * Calculate the order of allocation given an slab object size.
2175 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002176 * The order of allocation has significant impact on performance and other
2177 * system components. Generally order 0 allocations should be preferred since
2178 * order 0 does not cause fragmentation in the page allocator. Larger objects
2179 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002180 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002181 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002182 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002183 * In order to reach satisfactory performance we must ensure that a minimum
2184 * number of objects is in one slab. Otherwise we may generate too much
2185 * activity on the partial lists which requires taking the list_lock. This is
2186 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002187 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002188 * slub_max_order specifies the order where we begin to stop considering the
2189 * number of objects in a slab as critical. If we reach slub_max_order then
2190 * we try to keep the page order as low as possible. So we accept more waste
2191 * of space in favor of a small page order.
2192 *
2193 * Higher order allocations also allow the placement of more objects in a
2194 * slab and thereby reduce object handling overhead. If the user has
2195 * requested a higher mininum order then we start with that one instead of
2196 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002197 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002198static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002199 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002200{
2201 int order;
2202 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002203 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002204
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002205 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002206 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002207
Christoph Lameter6300ea72007-07-17 04:03:20 -07002208 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002209 fls(min_objects * size - 1) - PAGE_SHIFT);
2210 order <= max_order; order++) {
2211
Christoph Lameter81819f02007-05-06 14:49:36 -07002212 unsigned long slab_size = PAGE_SIZE << order;
2213
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002214 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002215 continue;
2216
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002217 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002218
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002219 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002220 break;
2221
2222 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002223
Christoph Lameter81819f02007-05-06 14:49:36 -07002224 return order;
2225}
2226
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002227static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002228{
2229 int order;
2230 int min_objects;
2231 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002232 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002233
2234 /*
2235 * Attempt to find best configuration for a slab. This
2236 * works by first attempting to generate a layout with
2237 * the best configuration and backing off gradually.
2238 *
2239 * First we reduce the acceptable waste in a slab. Then
2240 * we reduce the minimum objects required in a slab.
2241 */
2242 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002243 if (!min_objects)
2244 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002245 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002246 min_objects = min(min_objects, max_objects);
2247
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002248 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002249 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002250 while (fraction >= 4) {
2251 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002252 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002253 if (order <= slub_max_order)
2254 return order;
2255 fraction /= 2;
2256 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002257 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002258 }
2259
2260 /*
2261 * We were unable to place multiple objects in a slab. Now
2262 * lets see if we can place a single object there.
2263 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002264 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002265 if (order <= slub_max_order)
2266 return order;
2267
2268 /*
2269 * Doh this slab cannot be placed using slub_max_order.
2270 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002271 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002272 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002273 return order;
2274 return -ENOSYS;
2275}
2276
Christoph Lameter81819f02007-05-06 14:49:36 -07002277/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002278 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002279 */
2280static unsigned long calculate_alignment(unsigned long flags,
2281 unsigned long align, unsigned long size)
2282{
2283 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002284 * If the user wants hardware cache aligned objects then follow that
2285 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002286 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002287 * The hardware cache alignment cannot override the specified
2288 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002289 */
Nick Pigginb6210382008-03-05 14:05:56 -08002290 if (flags & SLAB_HWCACHE_ALIGN) {
2291 unsigned long ralign = cache_line_size();
2292 while (size <= ralign / 2)
2293 ralign /= 2;
2294 align = max(align, ralign);
2295 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002296
2297 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002298 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002299
2300 return ALIGN(align, sizeof(void *));
2301}
2302
Pekka Enberg5595cff2008-08-05 09:28:47 +03002303static void
2304init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002305{
2306 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002307 spin_lock_init(&n->list_lock);
2308 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002309#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002310 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002311 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002312 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002313#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002314}
2315
Christoph Lameter55136592010-08-20 12:37:13 -05002316static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002317{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002318 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2319 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002320
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002321#ifdef CONFIG_CMPXCHG_LOCAL
2322 /*
2323 * Must align to double word boundary for the double cmpxchg instructions
2324 * to work.
2325 */
2326 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
2327#else
2328 /* Regular alignment is sufficient */
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002329 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002330#endif
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002331
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002332 if (!s->cpu_slab)
2333 return 0;
2334
2335 init_kmem_cache_cpus(s);
2336
2337 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002338}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002339
Christoph Lameter51df1142010-08-20 12:37:15 -05002340static struct kmem_cache *kmem_cache_node;
2341
Christoph Lameter81819f02007-05-06 14:49:36 -07002342/*
2343 * No kmalloc_node yet so do it by hand. We know that this is the first
2344 * slab on the node for this slabcache. There are no concurrent accesses
2345 * possible.
2346 *
2347 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002348 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2349 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002350 */
Christoph Lameter55136592010-08-20 12:37:13 -05002351static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002352{
2353 struct page *page;
2354 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002355 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002356
Christoph Lameter51df1142010-08-20 12:37:15 -05002357 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002358
Christoph Lameter51df1142010-08-20 12:37:15 -05002359 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002360
2361 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002362 if (page_to_nid(page) != node) {
2363 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2364 "node %d\n", node);
2365 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2366 "in order to be able to continue\n");
2367 }
2368
Christoph Lameter81819f02007-05-06 14:49:36 -07002369 n = page->freelist;
2370 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002371 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002372 page->inuse++;
Christoph Lameter51df1142010-08-20 12:37:15 -05002373 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002374#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002375 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002376 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002377#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002378 init_kmem_cache_node(n, kmem_cache_node);
2379 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002380
rootba84c732008-01-07 23:20:28 -08002381 /*
2382 * lockdep requires consistent irq usage for each lock
2383 * so even though there cannot be a race this early in
2384 * the boot sequence, we still disable irqs.
2385 */
2386 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002387 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002388 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002389}
2390
2391static void free_kmem_cache_nodes(struct kmem_cache *s)
2392{
2393 int node;
2394
Christoph Lameterf64dc582007-10-16 01:25:33 -07002395 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002396 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002397
Alexander Duyck73367bd2010-05-21 14:41:35 -07002398 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002399 kmem_cache_free(kmem_cache_node, n);
2400
Christoph Lameter81819f02007-05-06 14:49:36 -07002401 s->node[node] = NULL;
2402 }
2403}
2404
Christoph Lameter55136592010-08-20 12:37:13 -05002405static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002406{
2407 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002408
Christoph Lameterf64dc582007-10-16 01:25:33 -07002409 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002410 struct kmem_cache_node *n;
2411
Alexander Duyck73367bd2010-05-21 14:41:35 -07002412 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002413 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002414 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002415 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002416 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002417 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002418
2419 if (!n) {
2420 free_kmem_cache_nodes(s);
2421 return 0;
2422 }
2423
Christoph Lameter81819f02007-05-06 14:49:36 -07002424 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002425 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002426 }
2427 return 1;
2428}
Christoph Lameter81819f02007-05-06 14:49:36 -07002429
David Rientjesc0bdb232009-02-25 09:16:35 +02002430static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002431{
2432 if (min < MIN_PARTIAL)
2433 min = MIN_PARTIAL;
2434 else if (min > MAX_PARTIAL)
2435 min = MAX_PARTIAL;
2436 s->min_partial = min;
2437}
2438
Christoph Lameter81819f02007-05-06 14:49:36 -07002439/*
2440 * calculate_sizes() determines the order and the distribution of data within
2441 * a slab object.
2442 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002443static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002444{
2445 unsigned long flags = s->flags;
2446 unsigned long size = s->objsize;
2447 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002448 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002449
2450 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002451 * Round up object size to the next word boundary. We can only
2452 * place the free pointer at word boundaries and this determines
2453 * the possible location of the free pointer.
2454 */
2455 size = ALIGN(size, sizeof(void *));
2456
2457#ifdef CONFIG_SLUB_DEBUG
2458 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002459 * Determine if we can poison the object itself. If the user of
2460 * the slab may touch the object after free or before allocation
2461 * then we should never poison the object itself.
2462 */
2463 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002464 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002465 s->flags |= __OBJECT_POISON;
2466 else
2467 s->flags &= ~__OBJECT_POISON;
2468
Christoph Lameter81819f02007-05-06 14:49:36 -07002469
2470 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002471 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002472 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002473 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002474 */
2475 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2476 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002477#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002478
2479 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002480 * With that we have determined the number of bytes in actual use
2481 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002482 */
2483 s->inuse = size;
2484
2485 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002486 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002487 /*
2488 * Relocate free pointer after the object if it is not
2489 * permitted to overwrite the first word of the object on
2490 * kmem_cache_free.
2491 *
2492 * This is the case if we do RCU, have a constructor or
2493 * destructor or are poisoning the objects.
2494 */
2495 s->offset = size;
2496 size += sizeof(void *);
2497 }
2498
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002499#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002500 if (flags & SLAB_STORE_USER)
2501 /*
2502 * Need to store information about allocs and frees after
2503 * the object.
2504 */
2505 size += 2 * sizeof(struct track);
2506
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002507 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002508 /*
2509 * Add some empty padding so that we can catch
2510 * overwrites from earlier objects rather than let
2511 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002512 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002513 * of the object.
2514 */
2515 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002516#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002517
Christoph Lameter81819f02007-05-06 14:49:36 -07002518 /*
2519 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002520 * user specified and the dynamic determination of cache line size
2521 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002522 */
2523 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002524 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002525
2526 /*
2527 * SLUB stores one object immediately after another beginning from
2528 * offset 0. In order to align the objects we have to simply size
2529 * each object to conform to the alignment.
2530 */
2531 size = ALIGN(size, align);
2532 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002533 if (forced_order >= 0)
2534 order = forced_order;
2535 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002536 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002537
Christoph Lameter834f3d12008-04-14 19:11:31 +03002538 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002539 return 0;
2540
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002541 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002542 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002543 s->allocflags |= __GFP_COMP;
2544
2545 if (s->flags & SLAB_CACHE_DMA)
2546 s->allocflags |= SLUB_DMA;
2547
2548 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2549 s->allocflags |= __GFP_RECLAIMABLE;
2550
Christoph Lameter81819f02007-05-06 14:49:36 -07002551 /*
2552 * Determine the number of objects per slab
2553 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002554 s->oo = oo_make(order, size, s->reserved);
2555 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002556 if (oo_objects(s->oo) > oo_objects(s->max))
2557 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002558
Christoph Lameter834f3d12008-04-14 19:11:31 +03002559 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002560
2561}
2562
Christoph Lameter55136592010-08-20 12:37:13 -05002563static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002564 const char *name, size_t size,
2565 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002566 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002567{
2568 memset(s, 0, kmem_size);
2569 s->name = name;
2570 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002571 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002572 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002573 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002574 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002575
Lai Jiangshanda9a6382011-03-10 15:22:00 +08002576 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2577 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07002578
Christoph Lameter06b285d2008-04-14 19:11:41 +03002579 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002580 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002581 if (disable_higher_order_debug) {
2582 /*
2583 * Disable debugging flags that store metadata if the min slab
2584 * order increased.
2585 */
2586 if (get_order(s->size) > get_order(s->objsize)) {
2587 s->flags &= ~DEBUG_METADATA_FLAGS;
2588 s->offset = 0;
2589 if (!calculate_sizes(s, -1))
2590 goto error;
2591 }
2592 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002593
David Rientjes3b89d7d2009-02-22 17:40:07 -08002594 /*
2595 * The larger the object size is, the more pages we want on the partial
2596 * list to avoid pounding the page allocator excessively.
2597 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002598 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002599 s->refcount = 1;
2600#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002601 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002602#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002603 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002604 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002605
Christoph Lameter55136592010-08-20 12:37:13 -05002606 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002607 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002608
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002609 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002610error:
2611 if (flags & SLAB_PANIC)
2612 panic("Cannot create slab %s size=%lu realsize=%u "
2613 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002614 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002615 s->offset, flags);
2616 return 0;
2617}
Christoph Lameter81819f02007-05-06 14:49:36 -07002618
2619/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002620 * Determine the size of a slab object
2621 */
2622unsigned int kmem_cache_size(struct kmem_cache *s)
2623{
2624 return s->objsize;
2625}
2626EXPORT_SYMBOL(kmem_cache_size);
2627
Christoph Lameter33b12c32008-04-25 12:22:43 -07002628static void list_slab_objects(struct kmem_cache *s, struct page *page,
2629 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002630{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002631#ifdef CONFIG_SLUB_DEBUG
2632 void *addr = page_address(page);
2633 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002634 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2635 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002636 if (!map)
2637 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002638 slab_err(s, page, "%s", text);
2639 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002640
Christoph Lameter5f80b132011-04-15 14:48:13 -05002641 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002642 for_each_object(p, s, addr, page->objects) {
2643
2644 if (!test_bit(slab_index(p, s, addr), map)) {
2645 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2646 p, p - addr);
2647 print_tracking(s, p);
2648 }
2649 }
2650 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002651 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002652#endif
2653}
2654
Christoph Lameter81819f02007-05-06 14:49:36 -07002655/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002656 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002657 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002658static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002659{
Christoph Lameter81819f02007-05-06 14:49:36 -07002660 unsigned long flags;
2661 struct page *page, *h;
2662
2663 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002664 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002665 if (!page->inuse) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05002666 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002667 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002668 } else {
2669 list_slab_objects(s, page,
2670 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002671 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002672 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002673 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002674}
2675
2676/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002677 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002678 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002679static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002680{
2681 int node;
2682
2683 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002684 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002685 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002686 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002687 struct kmem_cache_node *n = get_node(s, node);
2688
Christoph Lameter599870b2008-04-23 12:36:52 -07002689 free_partial(s, n);
2690 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002691 return 1;
2692 }
2693 free_kmem_cache_nodes(s);
2694 return 0;
2695}
2696
2697/*
2698 * Close a cache and release the kmem_cache structure
2699 * (must be used for caches created using kmem_cache_create)
2700 */
2701void kmem_cache_destroy(struct kmem_cache *s)
2702{
2703 down_write(&slub_lock);
2704 s->refcount--;
2705 if (!s->refcount) {
2706 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002707 if (kmem_cache_close(s)) {
2708 printk(KERN_ERR "SLUB %s: %s called for cache that "
2709 "still has objects.\n", s->name, __func__);
2710 dump_stack();
2711 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002712 if (s->flags & SLAB_DESTROY_BY_RCU)
2713 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002714 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002715 }
2716 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002717}
2718EXPORT_SYMBOL(kmem_cache_destroy);
2719
2720/********************************************************************
2721 * Kmalloc subsystem
2722 *******************************************************************/
2723
Christoph Lameter51df1142010-08-20 12:37:15 -05002724struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002725EXPORT_SYMBOL(kmalloc_caches);
2726
Christoph Lameter51df1142010-08-20 12:37:15 -05002727static struct kmem_cache *kmem_cache;
2728
Christoph Lameter55136592010-08-20 12:37:13 -05002729#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002730static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002731#endif
2732
Christoph Lameter81819f02007-05-06 14:49:36 -07002733static int __init setup_slub_min_order(char *str)
2734{
Pekka Enberg06428782008-01-07 23:20:27 -08002735 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002736
2737 return 1;
2738}
2739
2740__setup("slub_min_order=", setup_slub_min_order);
2741
2742static int __init setup_slub_max_order(char *str)
2743{
Pekka Enberg06428782008-01-07 23:20:27 -08002744 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002745 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002746
2747 return 1;
2748}
2749
2750__setup("slub_max_order=", setup_slub_max_order);
2751
2752static int __init setup_slub_min_objects(char *str)
2753{
Pekka Enberg06428782008-01-07 23:20:27 -08002754 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002755
2756 return 1;
2757}
2758
2759__setup("slub_min_objects=", setup_slub_min_objects);
2760
2761static int __init setup_slub_nomerge(char *str)
2762{
2763 slub_nomerge = 1;
2764 return 1;
2765}
2766
2767__setup("slub_nomerge", setup_slub_nomerge);
2768
Christoph Lameter51df1142010-08-20 12:37:15 -05002769static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2770 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002771{
Christoph Lameter51df1142010-08-20 12:37:15 -05002772 struct kmem_cache *s;
2773
2774 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2775
Pekka Enberg83b519e2009-06-10 19:40:04 +03002776 /*
2777 * This function is called with IRQs disabled during early-boot on
2778 * single CPU so there's no need to take slub_lock here.
2779 */
Christoph Lameter55136592010-08-20 12:37:13 -05002780 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002781 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002782 goto panic;
2783
2784 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002785 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002786
2787panic:
2788 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002789 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002790}
2791
Christoph Lameterf1b26332007-07-17 04:03:26 -07002792/*
2793 * Conversion table for small slabs sizes / 8 to the index in the
2794 * kmalloc array. This is necessary for slabs < 192 since we have non power
2795 * of two cache sizes there. The size of larger slabs can be determined using
2796 * fls.
2797 */
2798static s8 size_index[24] = {
2799 3, /* 8 */
2800 4, /* 16 */
2801 5, /* 24 */
2802 5, /* 32 */
2803 6, /* 40 */
2804 6, /* 48 */
2805 6, /* 56 */
2806 6, /* 64 */
2807 1, /* 72 */
2808 1, /* 80 */
2809 1, /* 88 */
2810 1, /* 96 */
2811 7, /* 104 */
2812 7, /* 112 */
2813 7, /* 120 */
2814 7, /* 128 */
2815 2, /* 136 */
2816 2, /* 144 */
2817 2, /* 152 */
2818 2, /* 160 */
2819 2, /* 168 */
2820 2, /* 176 */
2821 2, /* 184 */
2822 2 /* 192 */
2823};
2824
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002825static inline int size_index_elem(size_t bytes)
2826{
2827 return (bytes - 1) / 8;
2828}
2829
Christoph Lameter81819f02007-05-06 14:49:36 -07002830static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2831{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002832 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002833
Christoph Lameterf1b26332007-07-17 04:03:26 -07002834 if (size <= 192) {
2835 if (!size)
2836 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002837
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002838 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002839 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002840 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002841
2842#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002843 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002844 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002845
Christoph Lameter81819f02007-05-06 14:49:36 -07002846#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002847 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002848}
2849
2850void *__kmalloc(size_t size, gfp_t flags)
2851{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002852 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002853 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002854
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002855 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002856 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002857
2858 s = get_slab(size, flags);
2859
2860 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002861 return s;
2862
Christoph Lameter2154a332010-07-09 14:07:10 -05002863 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002864
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002865 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002866
2867 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002868}
2869EXPORT_SYMBOL(__kmalloc);
2870
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002871#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002872static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2873{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002874 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002875 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002876
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002877 flags |= __GFP_COMP | __GFP_NOTRACK;
2878 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002879 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002880 ptr = page_address(page);
2881
2882 kmemleak_alloc(ptr, size, 1, flags);
2883 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002884}
2885
Christoph Lameter81819f02007-05-06 14:49:36 -07002886void *__kmalloc_node(size_t size, gfp_t flags, int node)
2887{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002888 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002889 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002890
Ingo Molnar057685c2009-02-20 12:15:30 +01002891 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002892 ret = kmalloc_large_node(size, flags, node);
2893
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002894 trace_kmalloc_node(_RET_IP_, ret,
2895 size, PAGE_SIZE << get_order(size),
2896 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002897
2898 return ret;
2899 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002900
2901 s = get_slab(size, flags);
2902
2903 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002904 return s;
2905
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002906 ret = slab_alloc(s, flags, node, _RET_IP_);
2907
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002908 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002909
2910 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002911}
2912EXPORT_SYMBOL(__kmalloc_node);
2913#endif
2914
2915size_t ksize(const void *object)
2916{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002917 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002918
Christoph Lameteref8b4522007-10-16 01:24:46 -07002919 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002920 return 0;
2921
Vegard Nossum294a80a2007-12-04 23:45:30 -08002922 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002923
Pekka Enberg76994412008-05-22 19:22:25 +03002924 if (unlikely(!PageSlab(page))) {
2925 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002926 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002927 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002928
Eric Dumazetb3d41882011-02-14 18:35:22 +01002929 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002930}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002931EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002932
2933void kfree(const void *x)
2934{
Christoph Lameter81819f02007-05-06 14:49:36 -07002935 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002936 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002937
Pekka Enberg2121db72009-03-25 11:05:57 +02002938 trace_kfree(_RET_IP_, x);
2939
Satyam Sharma2408c552007-10-16 01:24:44 -07002940 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002941 return;
2942
Christoph Lameterb49af682007-05-06 14:49:41 -07002943 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002944 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002945 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002946 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002947 put_page(page);
2948 return;
2949 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002950 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002951}
2952EXPORT_SYMBOL(kfree);
2953
Christoph Lameter2086d262007-05-06 14:49:46 -07002954/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002955 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2956 * the remaining slabs by the number of items in use. The slabs with the
2957 * most items in use come first. New allocations will then fill those up
2958 * and thus they can be removed from the partial lists.
2959 *
2960 * The slabs with the least items are placed last. This results in them
2961 * being allocated from last increasing the chance that the last objects
2962 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002963 */
2964int kmem_cache_shrink(struct kmem_cache *s)
2965{
2966 int node;
2967 int i;
2968 struct kmem_cache_node *n;
2969 struct page *page;
2970 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002971 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002972 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002973 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002974 unsigned long flags;
2975
2976 if (!slabs_by_inuse)
2977 return -ENOMEM;
2978
2979 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002980 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002981 n = get_node(s, node);
2982
2983 if (!n->nr_partial)
2984 continue;
2985
Christoph Lameter834f3d12008-04-14 19:11:31 +03002986 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002987 INIT_LIST_HEAD(slabs_by_inuse + i);
2988
2989 spin_lock_irqsave(&n->list_lock, flags);
2990
2991 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002992 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002993 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002994 * Note that concurrent frees may occur while we hold the
2995 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002996 */
2997 list_for_each_entry_safe(page, t, &n->partial, lru) {
2998 if (!page->inuse && slab_trylock(page)) {
2999 /*
3000 * Must hold slab lock here because slab_free
3001 * may have freed the last object and be
3002 * waiting to release the slab.
3003 */
Christoph Lameter62e346a2010-09-28 08:10:28 -05003004 __remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003005 slab_unlock(page);
3006 discard_slab(s, page);
3007 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07003008 list_move(&page->lru,
3009 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07003010 }
3011 }
3012
Christoph Lameter2086d262007-05-06 14:49:46 -07003013 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003014 * Rebuild the partial list with the slabs filled up most
3015 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003016 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003017 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003018 list_splice(slabs_by_inuse + i, n->partial.prev);
3019
Christoph Lameter2086d262007-05-06 14:49:46 -07003020 spin_unlock_irqrestore(&n->list_lock, flags);
3021 }
3022
3023 kfree(slabs_by_inuse);
3024 return 0;
3025}
3026EXPORT_SYMBOL(kmem_cache_shrink);
3027
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003028#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003029static int slab_mem_going_offline_callback(void *arg)
3030{
3031 struct kmem_cache *s;
3032
3033 down_read(&slub_lock);
3034 list_for_each_entry(s, &slab_caches, list)
3035 kmem_cache_shrink(s);
3036 up_read(&slub_lock);
3037
3038 return 0;
3039}
3040
3041static void slab_mem_offline_callback(void *arg)
3042{
3043 struct kmem_cache_node *n;
3044 struct kmem_cache *s;
3045 struct memory_notify *marg = arg;
3046 int offline_node;
3047
3048 offline_node = marg->status_change_nid;
3049
3050 /*
3051 * If the node still has available memory. we need kmem_cache_node
3052 * for it yet.
3053 */
3054 if (offline_node < 0)
3055 return;
3056
3057 down_read(&slub_lock);
3058 list_for_each_entry(s, &slab_caches, list) {
3059 n = get_node(s, offline_node);
3060 if (n) {
3061 /*
3062 * if n->nr_slabs > 0, slabs still exist on the node
3063 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003064 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003065 * callback. So, we must fail.
3066 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003067 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003068
3069 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003070 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003071 }
3072 }
3073 up_read(&slub_lock);
3074}
3075
3076static int slab_mem_going_online_callback(void *arg)
3077{
3078 struct kmem_cache_node *n;
3079 struct kmem_cache *s;
3080 struct memory_notify *marg = arg;
3081 int nid = marg->status_change_nid;
3082 int ret = 0;
3083
3084 /*
3085 * If the node's memory is already available, then kmem_cache_node is
3086 * already created. Nothing to do.
3087 */
3088 if (nid < 0)
3089 return 0;
3090
3091 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003092 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003093 * allocate a kmem_cache_node structure in order to bring the node
3094 * online.
3095 */
3096 down_read(&slub_lock);
3097 list_for_each_entry(s, &slab_caches, list) {
3098 /*
3099 * XXX: kmem_cache_alloc_node will fallback to other nodes
3100 * since memory is not yet available from the node that
3101 * is brought up.
3102 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003103 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003104 if (!n) {
3105 ret = -ENOMEM;
3106 goto out;
3107 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003108 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003109 s->node[nid] = n;
3110 }
3111out:
3112 up_read(&slub_lock);
3113 return ret;
3114}
3115
3116static int slab_memory_callback(struct notifier_block *self,
3117 unsigned long action, void *arg)
3118{
3119 int ret = 0;
3120
3121 switch (action) {
3122 case MEM_GOING_ONLINE:
3123 ret = slab_mem_going_online_callback(arg);
3124 break;
3125 case MEM_GOING_OFFLINE:
3126 ret = slab_mem_going_offline_callback(arg);
3127 break;
3128 case MEM_OFFLINE:
3129 case MEM_CANCEL_ONLINE:
3130 slab_mem_offline_callback(arg);
3131 break;
3132 case MEM_ONLINE:
3133 case MEM_CANCEL_OFFLINE:
3134 break;
3135 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003136 if (ret)
3137 ret = notifier_from_errno(ret);
3138 else
3139 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003140 return ret;
3141}
3142
3143#endif /* CONFIG_MEMORY_HOTPLUG */
3144
Christoph Lameter81819f02007-05-06 14:49:36 -07003145/********************************************************************
3146 * Basic setup of slabs
3147 *******************************************************************/
3148
Christoph Lameter51df1142010-08-20 12:37:15 -05003149/*
3150 * Used for early kmem_cache structures that were allocated using
3151 * the page allocator
3152 */
3153
3154static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3155{
3156 int node;
3157
3158 list_add(&s->list, &slab_caches);
3159 s->refcount = -1;
3160
3161 for_each_node_state(node, N_NORMAL_MEMORY) {
3162 struct kmem_cache_node *n = get_node(s, node);
3163 struct page *p;
3164
3165 if (n) {
3166 list_for_each_entry(p, &n->partial, lru)
3167 p->slab = s;
3168
Li Zefan607bf322011-04-12 15:22:26 +08003169#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003170 list_for_each_entry(p, &n->full, lru)
3171 p->slab = s;
3172#endif
3173 }
3174 }
3175}
3176
Christoph Lameter81819f02007-05-06 14:49:36 -07003177void __init kmem_cache_init(void)
3178{
3179 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003180 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003181 struct kmem_cache *temp_kmem_cache;
3182 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003183 struct kmem_cache *temp_kmem_cache_node;
3184 unsigned long kmalloc_size;
3185
3186 kmem_size = offsetof(struct kmem_cache, node) +
3187 nr_node_ids * sizeof(struct kmem_cache_node *);
3188
3189 /* Allocate two kmem_caches from the page allocator */
3190 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3191 order = get_order(2 * kmalloc_size);
3192 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3193
Christoph Lameter81819f02007-05-06 14:49:36 -07003194 /*
3195 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003196 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003197 * kmem_cache_open for slab_state == DOWN.
3198 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003199 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3200
3201 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3202 sizeof(struct kmem_cache_node),
3203 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003204
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003205 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003206
3207 /* Able to allocate the per node structures */
3208 slab_state = PARTIAL;
3209
Christoph Lameter51df1142010-08-20 12:37:15 -05003210 temp_kmem_cache = kmem_cache;
3211 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3212 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3213 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3214 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003215
Christoph Lameter51df1142010-08-20 12:37:15 -05003216 /*
3217 * Allocate kmem_cache_node properly from the kmem_cache slab.
3218 * kmem_cache_node is separately allocated so no need to
3219 * update any list pointers.
3220 */
3221 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003222
Christoph Lameter51df1142010-08-20 12:37:15 -05003223 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3224 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3225
3226 kmem_cache_bootstrap_fixup(kmem_cache_node);
3227
3228 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003229 kmem_cache_bootstrap_fixup(kmem_cache);
3230 caches++;
3231 /* Free temporary boot structure */
3232 free_pages((unsigned long)temp_kmem_cache, order);
3233
3234 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003235
3236 /*
3237 * Patch up the size_index table if we have strange large alignment
3238 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003239 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003240 *
3241 * Largest permitted alignment is 256 bytes due to the way we
3242 * handle the index determination for the smaller caches.
3243 *
3244 * Make sure that nothing crazy happens if someone starts tinkering
3245 * around with ARCH_KMALLOC_MINALIGN
3246 */
3247 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3248 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3249
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003250 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3251 int elem = size_index_elem(i);
3252 if (elem >= ARRAY_SIZE(size_index))
3253 break;
3254 size_index[elem] = KMALLOC_SHIFT_LOW;
3255 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003256
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003257 if (KMALLOC_MIN_SIZE == 64) {
3258 /*
3259 * The 96 byte size cache is not used if the alignment
3260 * is 64 byte.
3261 */
3262 for (i = 64 + 8; i <= 96; i += 8)
3263 size_index[size_index_elem(i)] = 7;
3264 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003265 /*
3266 * The 192 byte sized cache is not used if the alignment
3267 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3268 * instead.
3269 */
3270 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003271 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003272 }
3273
Christoph Lameter51df1142010-08-20 12:37:15 -05003274 /* Caches that are not of the two-to-the-power-of size */
3275 if (KMALLOC_MIN_SIZE <= 32) {
3276 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3277 caches++;
3278 }
3279
3280 if (KMALLOC_MIN_SIZE <= 64) {
3281 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3282 caches++;
3283 }
3284
3285 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3286 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3287 caches++;
3288 }
3289
Christoph Lameter81819f02007-05-06 14:49:36 -07003290 slab_state = UP;
3291
3292 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003293 if (KMALLOC_MIN_SIZE <= 32) {
3294 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3295 BUG_ON(!kmalloc_caches[1]->name);
3296 }
3297
3298 if (KMALLOC_MIN_SIZE <= 64) {
3299 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3300 BUG_ON(!kmalloc_caches[2]->name);
3301 }
3302
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003303 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3304 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3305
3306 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003307 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003308 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003309
3310#ifdef CONFIG_SMP
3311 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003312#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003313
Christoph Lameter55136592010-08-20 12:37:13 -05003314#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003315 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3316 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003317
Christoph Lameter51df1142010-08-20 12:37:15 -05003318 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003319 char *name = kasprintf(GFP_NOWAIT,
3320 "dma-kmalloc-%d", s->objsize);
3321
3322 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003323 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3324 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003325 }
3326 }
3327#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003328 printk(KERN_INFO
3329 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003330 " CPUs=%d, Nodes=%d\n",
3331 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003332 slub_min_order, slub_max_order, slub_min_objects,
3333 nr_cpu_ids, nr_node_ids);
3334}
3335
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003336void __init kmem_cache_init_late(void)
3337{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003338}
3339
Christoph Lameter81819f02007-05-06 14:49:36 -07003340/*
3341 * Find a mergeable slab cache
3342 */
3343static int slab_unmergeable(struct kmem_cache *s)
3344{
3345 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3346 return 1;
3347
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003348 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003349 return 1;
3350
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003351 /*
3352 * We may have set a slab to be unmergeable during bootstrap.
3353 */
3354 if (s->refcount < 0)
3355 return 1;
3356
Christoph Lameter81819f02007-05-06 14:49:36 -07003357 return 0;
3358}
3359
3360static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003361 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003362 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003363{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003364 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003365
3366 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3367 return NULL;
3368
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003369 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003370 return NULL;
3371
3372 size = ALIGN(size, sizeof(void *));
3373 align = calculate_alignment(flags, align, size);
3374 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003375 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003376
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003377 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003378 if (slab_unmergeable(s))
3379 continue;
3380
3381 if (size > s->size)
3382 continue;
3383
Christoph Lameterba0268a2007-09-11 15:24:11 -07003384 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003385 continue;
3386 /*
3387 * Check if alignment is compatible.
3388 * Courtesy of Adrian Drzewiecki
3389 */
Pekka Enberg06428782008-01-07 23:20:27 -08003390 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003391 continue;
3392
3393 if (s->size - size >= sizeof(void *))
3394 continue;
3395
3396 return s;
3397 }
3398 return NULL;
3399}
3400
3401struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003402 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003403{
3404 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003405 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003406
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003407 if (WARN_ON(!name))
3408 return NULL;
3409
Christoph Lameter81819f02007-05-06 14:49:36 -07003410 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003411 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003412 if (s) {
3413 s->refcount++;
3414 /*
3415 * Adjust the object sizes so that we clear
3416 * the complete object on kzalloc.
3417 */
3418 s->objsize = max(s->objsize, (int)size);
3419 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003420
David Rientjes7b8f3b62008-12-17 22:09:46 -08003421 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003422 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003423 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003424 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003425 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003426 return s;
3427 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003428
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003429 n = kstrdup(name, GFP_KERNEL);
3430 if (!n)
3431 goto err;
3432
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003433 s = kmalloc(kmem_size, GFP_KERNEL);
3434 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003435 if (kmem_cache_open(s, n,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003436 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003437 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003438 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003439 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003440 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003441 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003442 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003443 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003444 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003445 return s;
3446 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003447 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003448 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003449 }
Pavel Emelyanov68cee4f2010-10-28 13:50:37 +04003450err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003451 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003452
Christoph Lameter81819f02007-05-06 14:49:36 -07003453 if (flags & SLAB_PANIC)
3454 panic("Cannot create slabcache %s\n", name);
3455 else
3456 s = NULL;
3457 return s;
3458}
3459EXPORT_SYMBOL(kmem_cache_create);
3460
Christoph Lameter81819f02007-05-06 14:49:36 -07003461#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003462/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003463 * Use the cpu notifier to insure that the cpu slabs are flushed when
3464 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003465 */
3466static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3467 unsigned long action, void *hcpu)
3468{
3469 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003470 struct kmem_cache *s;
3471 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003472
3473 switch (action) {
3474 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003475 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003476 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003477 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003478 down_read(&slub_lock);
3479 list_for_each_entry(s, &slab_caches, list) {
3480 local_irq_save(flags);
3481 __flush_cpu_slab(s, cpu);
3482 local_irq_restore(flags);
3483 }
3484 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003485 break;
3486 default:
3487 break;
3488 }
3489 return NOTIFY_OK;
3490}
3491
Pekka Enberg06428782008-01-07 23:20:27 -08003492static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003493 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003494};
Christoph Lameter81819f02007-05-06 14:49:36 -07003495
3496#endif
3497
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003498void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003499{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003500 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003501 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003502
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003503 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003504 return kmalloc_large(size, gfpflags);
3505
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003506 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003507
Satyam Sharma2408c552007-10-16 01:24:44 -07003508 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003509 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003510
Christoph Lameter2154a332010-07-09 14:07:10 -05003511 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003512
3513 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003514 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003515
3516 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003517}
3518
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003519#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003520void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003521 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003522{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003523 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003524 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003525
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003526 if (unlikely(size > SLUB_MAX_SIZE)) {
3527 ret = kmalloc_large_node(size, gfpflags, node);
3528
3529 trace_kmalloc_node(caller, ret,
3530 size, PAGE_SIZE << get_order(size),
3531 gfpflags, node);
3532
3533 return ret;
3534 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003535
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003536 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003537
Satyam Sharma2408c552007-10-16 01:24:44 -07003538 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003539 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003540
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003541 ret = slab_alloc(s, gfpflags, node, caller);
3542
3543 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003544 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003545
3546 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003547}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003548#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003549
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003550#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003551static int count_inuse(struct page *page)
3552{
3553 return page->inuse;
3554}
3555
3556static int count_total(struct page *page)
3557{
3558 return page->objects;
3559}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003560#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003561
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003562#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003563static int validate_slab(struct kmem_cache *s, struct page *page,
3564 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003565{
3566 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003567 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003568
3569 if (!check_slab(s, page) ||
3570 !on_freelist(s, page, NULL))
3571 return 0;
3572
3573 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003574 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003575
Christoph Lameter5f80b132011-04-15 14:48:13 -05003576 get_map(s, page, map);
3577 for_each_object(p, s, addr, page->objects) {
3578 if (test_bit(slab_index(p, s, addr), map))
3579 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3580 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003581 }
3582
Christoph Lameter224a88b2008-04-14 19:11:31 +03003583 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003584 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003585 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003586 return 0;
3587 return 1;
3588}
3589
Christoph Lameter434e2452007-07-17 04:03:30 -07003590static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3591 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003592{
3593 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003594 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003595 slab_unlock(page);
3596 } else
3597 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3598 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003599}
3600
Christoph Lameter434e2452007-07-17 04:03:30 -07003601static int validate_slab_node(struct kmem_cache *s,
3602 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003603{
3604 unsigned long count = 0;
3605 struct page *page;
3606 unsigned long flags;
3607
3608 spin_lock_irqsave(&n->list_lock, flags);
3609
3610 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003611 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003612 count++;
3613 }
3614 if (count != n->nr_partial)
3615 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3616 "counter=%ld\n", s->name, count, n->nr_partial);
3617
3618 if (!(s->flags & SLAB_STORE_USER))
3619 goto out;
3620
3621 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003622 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003623 count++;
3624 }
3625 if (count != atomic_long_read(&n->nr_slabs))
3626 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3627 "counter=%ld\n", s->name, count,
3628 atomic_long_read(&n->nr_slabs));
3629
3630out:
3631 spin_unlock_irqrestore(&n->list_lock, flags);
3632 return count;
3633}
3634
Christoph Lameter434e2452007-07-17 04:03:30 -07003635static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003636{
3637 int node;
3638 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003639 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003640 sizeof(unsigned long), GFP_KERNEL);
3641
3642 if (!map)
3643 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003644
3645 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003646 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003647 struct kmem_cache_node *n = get_node(s, node);
3648
Christoph Lameter434e2452007-07-17 04:03:30 -07003649 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003650 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003651 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003652 return count;
3653}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003654/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003655 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003656 * and freed.
3657 */
3658
3659struct location {
3660 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003661 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003662 long long sum_time;
3663 long min_time;
3664 long max_time;
3665 long min_pid;
3666 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303667 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003668 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003669};
3670
3671struct loc_track {
3672 unsigned long max;
3673 unsigned long count;
3674 struct location *loc;
3675};
3676
3677static void free_loc_track(struct loc_track *t)
3678{
3679 if (t->max)
3680 free_pages((unsigned long)t->loc,
3681 get_order(sizeof(struct location) * t->max));
3682}
3683
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003684static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003685{
3686 struct location *l;
3687 int order;
3688
Christoph Lameter88a420e2007-05-06 14:49:45 -07003689 order = get_order(sizeof(struct location) * max);
3690
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003691 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003692 if (!l)
3693 return 0;
3694
3695 if (t->count) {
3696 memcpy(l, t->loc, sizeof(struct location) * t->count);
3697 free_loc_track(t);
3698 }
3699 t->max = max;
3700 t->loc = l;
3701 return 1;
3702}
3703
3704static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003705 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003706{
3707 long start, end, pos;
3708 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003709 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003710 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003711
3712 start = -1;
3713 end = t->count;
3714
3715 for ( ; ; ) {
3716 pos = start + (end - start + 1) / 2;
3717
3718 /*
3719 * There is nothing at "end". If we end up there
3720 * we need to add something to before end.
3721 */
3722 if (pos == end)
3723 break;
3724
3725 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003726 if (track->addr == caddr) {
3727
3728 l = &t->loc[pos];
3729 l->count++;
3730 if (track->when) {
3731 l->sum_time += age;
3732 if (age < l->min_time)
3733 l->min_time = age;
3734 if (age > l->max_time)
3735 l->max_time = age;
3736
3737 if (track->pid < l->min_pid)
3738 l->min_pid = track->pid;
3739 if (track->pid > l->max_pid)
3740 l->max_pid = track->pid;
3741
Rusty Russell174596a2009-01-01 10:12:29 +10303742 cpumask_set_cpu(track->cpu,
3743 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003744 }
3745 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003746 return 1;
3747 }
3748
Christoph Lameter45edfa52007-05-09 02:32:45 -07003749 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003750 end = pos;
3751 else
3752 start = pos;
3753 }
3754
3755 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003756 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003757 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003758 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003759 return 0;
3760
3761 l = t->loc + pos;
3762 if (pos < t->count)
3763 memmove(l + 1, l,
3764 (t->count - pos) * sizeof(struct location));
3765 t->count++;
3766 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003767 l->addr = track->addr;
3768 l->sum_time = age;
3769 l->min_time = age;
3770 l->max_time = age;
3771 l->min_pid = track->pid;
3772 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303773 cpumask_clear(to_cpumask(l->cpus));
3774 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003775 nodes_clear(l->nodes);
3776 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003777 return 1;
3778}
3779
3780static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003781 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003782 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003783{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003784 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003785 void *p;
3786
Christoph Lameter39b26462008-04-14 19:11:30 +03003787 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05003788 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003789
Christoph Lameter224a88b2008-04-14 19:11:31 +03003790 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003791 if (!test_bit(slab_index(p, s, addr), map))
3792 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003793}
3794
3795static int list_locations(struct kmem_cache *s, char *buf,
3796 enum track_item alloc)
3797{
Harvey Harrisone374d482008-01-31 15:20:50 -08003798 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003799 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003800 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003801 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003802 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3803 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003804
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003805 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3806 GFP_TEMPORARY)) {
3807 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003808 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003809 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003810 /* Push back cpu slabs */
3811 flush_all(s);
3812
Christoph Lameterf64dc582007-10-16 01:25:33 -07003813 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003814 struct kmem_cache_node *n = get_node(s, node);
3815 unsigned long flags;
3816 struct page *page;
3817
Christoph Lameter9e869432007-08-22 14:01:56 -07003818 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003819 continue;
3820
3821 spin_lock_irqsave(&n->list_lock, flags);
3822 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003823 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003824 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003825 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003826 spin_unlock_irqrestore(&n->list_lock, flags);
3827 }
3828
3829 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003830 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003831
Hugh Dickins9c246242008-12-09 13:14:27 -08003832 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003833 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003834 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003835
3836 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08003837 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003838 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003839 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003840
3841 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003842 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003843 l->min_time,
3844 (long)div_u64(l->sum_time, l->count),
3845 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003846 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003847 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003848 l->min_time);
3849
3850 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003851 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003852 l->min_pid, l->max_pid);
3853 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003854 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003855 l->min_pid);
3856
Rusty Russell174596a2009-01-01 10:12:29 +10303857 if (num_online_cpus() > 1 &&
3858 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003859 len < PAGE_SIZE - 60) {
3860 len += sprintf(buf + len, " cpus=");
3861 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303862 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003863 }
3864
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003865 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003866 len < PAGE_SIZE - 60) {
3867 len += sprintf(buf + len, " nodes=");
3868 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003869 l->nodes);
3870 }
3871
Harvey Harrisone374d482008-01-31 15:20:50 -08003872 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003873 }
3874
3875 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003876 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003877 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003878 len += sprintf(buf, "No data\n");
3879 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003880}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003881#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07003882
Christoph Lametera5a84752010-10-05 13:57:27 -05003883#ifdef SLUB_RESILIENCY_TEST
3884static void resiliency_test(void)
3885{
3886 u8 *p;
3887
3888 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3889
3890 printk(KERN_ERR "SLUB resiliency testing\n");
3891 printk(KERN_ERR "-----------------------\n");
3892 printk(KERN_ERR "A. Corruption after allocation\n");
3893
3894 p = kzalloc(16, GFP_KERNEL);
3895 p[16] = 0x12;
3896 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3897 " 0x12->0x%p\n\n", p + 16);
3898
3899 validate_slab_cache(kmalloc_caches[4]);
3900
3901 /* Hmmm... The next two are dangerous */
3902 p = kzalloc(32, GFP_KERNEL);
3903 p[32 + sizeof(void *)] = 0x34;
3904 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3905 " 0x34 -> -0x%p\n", p);
3906 printk(KERN_ERR
3907 "If allocated object is overwritten then not detectable\n\n");
3908
3909 validate_slab_cache(kmalloc_caches[5]);
3910 p = kzalloc(64, GFP_KERNEL);
3911 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3912 *p = 0x56;
3913 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3914 p);
3915 printk(KERN_ERR
3916 "If allocated object is overwritten then not detectable\n\n");
3917 validate_slab_cache(kmalloc_caches[6]);
3918
3919 printk(KERN_ERR "\nB. Corruption after free\n");
3920 p = kzalloc(128, GFP_KERNEL);
3921 kfree(p);
3922 *p = 0x78;
3923 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3924 validate_slab_cache(kmalloc_caches[7]);
3925
3926 p = kzalloc(256, GFP_KERNEL);
3927 kfree(p);
3928 p[50] = 0x9a;
3929 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3930 p);
3931 validate_slab_cache(kmalloc_caches[8]);
3932
3933 p = kzalloc(512, GFP_KERNEL);
3934 kfree(p);
3935 p[512] = 0xab;
3936 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3937 validate_slab_cache(kmalloc_caches[9]);
3938}
3939#else
3940#ifdef CONFIG_SYSFS
3941static void resiliency_test(void) {};
3942#endif
3943#endif
3944
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003945#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07003946enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003947 SL_ALL, /* All slabs */
3948 SL_PARTIAL, /* Only partially allocated slabs */
3949 SL_CPU, /* Only slabs used for cpu caches */
3950 SL_OBJECTS, /* Determine allocated objects not slabs */
3951 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003952};
3953
Christoph Lameter205ab992008-04-14 19:11:40 +03003954#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003955#define SO_PARTIAL (1 << SL_PARTIAL)
3956#define SO_CPU (1 << SL_CPU)
3957#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003958#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003959
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003960static ssize_t show_slab_objects(struct kmem_cache *s,
3961 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003962{
3963 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003964 int node;
3965 int x;
3966 unsigned long *nodes;
3967 unsigned long *per_cpu;
3968
3969 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003970 if (!nodes)
3971 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003972 per_cpu = nodes + nr_node_ids;
3973
Christoph Lameter205ab992008-04-14 19:11:40 +03003974 if (flags & SO_CPU) {
3975 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003976
Christoph Lameter205ab992008-04-14 19:11:40 +03003977 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003978 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003979
Christoph Lameter205ab992008-04-14 19:11:40 +03003980 if (!c || c->node < 0)
3981 continue;
3982
3983 if (c->page) {
3984 if (flags & SO_TOTAL)
3985 x = c->page->objects;
3986 else if (flags & SO_OBJECTS)
3987 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003988 else
3989 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003990
Christoph Lameter81819f02007-05-06 14:49:36 -07003991 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003992 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003993 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003994 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003995 }
3996 }
3997
Christoph Lameter04d94872011-01-10 10:15:15 -06003998 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003999#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004000 if (flags & SO_ALL) {
4001 for_each_node_state(node, N_NORMAL_MEMORY) {
4002 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004003
Christoph Lameter205ab992008-04-14 19:11:40 +03004004 if (flags & SO_TOTAL)
4005 x = atomic_long_read(&n->total_objects);
4006 else if (flags & SO_OBJECTS)
4007 x = atomic_long_read(&n->total_objects) -
4008 count_partial(n, count_free);
4009
4010 else
4011 x = atomic_long_read(&n->nr_slabs);
4012 total += x;
4013 nodes[node] += x;
4014 }
4015
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004016 } else
4017#endif
4018 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004019 for_each_node_state(node, N_NORMAL_MEMORY) {
4020 struct kmem_cache_node *n = get_node(s, node);
4021
4022 if (flags & SO_TOTAL)
4023 x = count_partial(n, count_total);
4024 else if (flags & SO_OBJECTS)
4025 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004026 else
4027 x = n->nr_partial;
4028 total += x;
4029 nodes[node] += x;
4030 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004031 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004032 x = sprintf(buf, "%lu", total);
4033#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004034 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004035 if (nodes[node])
4036 x += sprintf(buf + x, " N%d=%lu",
4037 node, nodes[node]);
4038#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004039 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004040 kfree(nodes);
4041 return x + sprintf(buf + x, "\n");
4042}
4043
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004044#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004045static int any_slab_objects(struct kmem_cache *s)
4046{
4047 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004048
4049 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004050 struct kmem_cache_node *n = get_node(s, node);
4051
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004052 if (!n)
4053 continue;
4054
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004055 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004056 return 1;
4057 }
4058 return 0;
4059}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004060#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004061
4062#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4063#define to_slab(n) container_of(n, struct kmem_cache, kobj);
4064
4065struct slab_attribute {
4066 struct attribute attr;
4067 ssize_t (*show)(struct kmem_cache *s, char *buf);
4068 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4069};
4070
4071#define SLAB_ATTR_RO(_name) \
4072 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4073
4074#define SLAB_ATTR(_name) \
4075 static struct slab_attribute _name##_attr = \
4076 __ATTR(_name, 0644, _name##_show, _name##_store)
4077
Christoph Lameter81819f02007-05-06 14:49:36 -07004078static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4079{
4080 return sprintf(buf, "%d\n", s->size);
4081}
4082SLAB_ATTR_RO(slab_size);
4083
4084static ssize_t align_show(struct kmem_cache *s, char *buf)
4085{
4086 return sprintf(buf, "%d\n", s->align);
4087}
4088SLAB_ATTR_RO(align);
4089
4090static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4091{
4092 return sprintf(buf, "%d\n", s->objsize);
4093}
4094SLAB_ATTR_RO(object_size);
4095
4096static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4097{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004098 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004099}
4100SLAB_ATTR_RO(objs_per_slab);
4101
Christoph Lameter06b285d2008-04-14 19:11:41 +03004102static ssize_t order_store(struct kmem_cache *s,
4103 const char *buf, size_t length)
4104{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004105 unsigned long order;
4106 int err;
4107
4108 err = strict_strtoul(buf, 10, &order);
4109 if (err)
4110 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004111
4112 if (order > slub_max_order || order < slub_min_order)
4113 return -EINVAL;
4114
4115 calculate_sizes(s, order);
4116 return length;
4117}
4118
Christoph Lameter81819f02007-05-06 14:49:36 -07004119static ssize_t order_show(struct kmem_cache *s, char *buf)
4120{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004121 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004122}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004123SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004124
David Rientjes73d342b2009-02-22 17:40:09 -08004125static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4126{
4127 return sprintf(buf, "%lu\n", s->min_partial);
4128}
4129
4130static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4131 size_t length)
4132{
4133 unsigned long min;
4134 int err;
4135
4136 err = strict_strtoul(buf, 10, &min);
4137 if (err)
4138 return err;
4139
David Rientjesc0bdb232009-02-25 09:16:35 +02004140 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004141 return length;
4142}
4143SLAB_ATTR(min_partial);
4144
Christoph Lameter81819f02007-05-06 14:49:36 -07004145static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4146{
Joe Perches62c70bc2011-01-13 15:45:52 -08004147 if (!s->ctor)
4148 return 0;
4149 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004150}
4151SLAB_ATTR_RO(ctor);
4152
Christoph Lameter81819f02007-05-06 14:49:36 -07004153static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4154{
4155 return sprintf(buf, "%d\n", s->refcount - 1);
4156}
4157SLAB_ATTR_RO(aliases);
4158
Christoph Lameter81819f02007-05-06 14:49:36 -07004159static ssize_t partial_show(struct kmem_cache *s, char *buf)
4160{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004161 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004162}
4163SLAB_ATTR_RO(partial);
4164
4165static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4166{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004167 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004168}
4169SLAB_ATTR_RO(cpu_slabs);
4170
4171static ssize_t objects_show(struct kmem_cache *s, char *buf)
4172{
Christoph Lameter205ab992008-04-14 19:11:40 +03004173 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004174}
4175SLAB_ATTR_RO(objects);
4176
Christoph Lameter205ab992008-04-14 19:11:40 +03004177static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4178{
4179 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4180}
4181SLAB_ATTR_RO(objects_partial);
4182
Christoph Lameter81819f02007-05-06 14:49:36 -07004183static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4184{
4185 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4186}
4187
4188static ssize_t reclaim_account_store(struct kmem_cache *s,
4189 const char *buf, size_t length)
4190{
4191 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4192 if (buf[0] == '1')
4193 s->flags |= SLAB_RECLAIM_ACCOUNT;
4194 return length;
4195}
4196SLAB_ATTR(reclaim_account);
4197
4198static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4199{
Christoph Lameter5af60832007-05-06 14:49:56 -07004200 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004201}
4202SLAB_ATTR_RO(hwcache_align);
4203
4204#ifdef CONFIG_ZONE_DMA
4205static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4206{
4207 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4208}
4209SLAB_ATTR_RO(cache_dma);
4210#endif
4211
4212static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4213{
4214 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4215}
4216SLAB_ATTR_RO(destroy_by_rcu);
4217
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004218static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4219{
4220 return sprintf(buf, "%d\n", s->reserved);
4221}
4222SLAB_ATTR_RO(reserved);
4223
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004224#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004225static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4226{
4227 return show_slab_objects(s, buf, SO_ALL);
4228}
4229SLAB_ATTR_RO(slabs);
4230
4231static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4232{
4233 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4234}
4235SLAB_ATTR_RO(total_objects);
4236
4237static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4238{
4239 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4240}
4241
4242static ssize_t sanity_checks_store(struct kmem_cache *s,
4243 const char *buf, size_t length)
4244{
4245 s->flags &= ~SLAB_DEBUG_FREE;
4246 if (buf[0] == '1')
4247 s->flags |= SLAB_DEBUG_FREE;
4248 return length;
4249}
4250SLAB_ATTR(sanity_checks);
4251
4252static ssize_t trace_show(struct kmem_cache *s, char *buf)
4253{
4254 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4255}
4256
4257static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4258 size_t length)
4259{
4260 s->flags &= ~SLAB_TRACE;
4261 if (buf[0] == '1')
4262 s->flags |= SLAB_TRACE;
4263 return length;
4264}
4265SLAB_ATTR(trace);
4266
Christoph Lameter81819f02007-05-06 14:49:36 -07004267static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4268{
4269 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4270}
4271
4272static ssize_t red_zone_store(struct kmem_cache *s,
4273 const char *buf, size_t length)
4274{
4275 if (any_slab_objects(s))
4276 return -EBUSY;
4277
4278 s->flags &= ~SLAB_RED_ZONE;
4279 if (buf[0] == '1')
4280 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004281 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004282 return length;
4283}
4284SLAB_ATTR(red_zone);
4285
4286static ssize_t poison_show(struct kmem_cache *s, char *buf)
4287{
4288 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4289}
4290
4291static ssize_t poison_store(struct kmem_cache *s,
4292 const char *buf, size_t length)
4293{
4294 if (any_slab_objects(s))
4295 return -EBUSY;
4296
4297 s->flags &= ~SLAB_POISON;
4298 if (buf[0] == '1')
4299 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004300 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004301 return length;
4302}
4303SLAB_ATTR(poison);
4304
4305static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4306{
4307 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4308}
4309
4310static ssize_t store_user_store(struct kmem_cache *s,
4311 const char *buf, size_t length)
4312{
4313 if (any_slab_objects(s))
4314 return -EBUSY;
4315
4316 s->flags &= ~SLAB_STORE_USER;
4317 if (buf[0] == '1')
4318 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004319 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004320 return length;
4321}
4322SLAB_ATTR(store_user);
4323
Christoph Lameter53e15af2007-05-06 14:49:43 -07004324static ssize_t validate_show(struct kmem_cache *s, char *buf)
4325{
4326 return 0;
4327}
4328
4329static ssize_t validate_store(struct kmem_cache *s,
4330 const char *buf, size_t length)
4331{
Christoph Lameter434e2452007-07-17 04:03:30 -07004332 int ret = -EINVAL;
4333
4334 if (buf[0] == '1') {
4335 ret = validate_slab_cache(s);
4336 if (ret >= 0)
4337 ret = length;
4338 }
4339 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004340}
4341SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004342
4343static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4344{
4345 if (!(s->flags & SLAB_STORE_USER))
4346 return -ENOSYS;
4347 return list_locations(s, buf, TRACK_ALLOC);
4348}
4349SLAB_ATTR_RO(alloc_calls);
4350
4351static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4352{
4353 if (!(s->flags & SLAB_STORE_USER))
4354 return -ENOSYS;
4355 return list_locations(s, buf, TRACK_FREE);
4356}
4357SLAB_ATTR_RO(free_calls);
4358#endif /* CONFIG_SLUB_DEBUG */
4359
4360#ifdef CONFIG_FAILSLAB
4361static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4362{
4363 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4364}
4365
4366static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4367 size_t length)
4368{
4369 s->flags &= ~SLAB_FAILSLAB;
4370 if (buf[0] == '1')
4371 s->flags |= SLAB_FAILSLAB;
4372 return length;
4373}
4374SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004375#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004376
Christoph Lameter2086d262007-05-06 14:49:46 -07004377static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4378{
4379 return 0;
4380}
4381
4382static ssize_t shrink_store(struct kmem_cache *s,
4383 const char *buf, size_t length)
4384{
4385 if (buf[0] == '1') {
4386 int rc = kmem_cache_shrink(s);
4387
4388 if (rc)
4389 return rc;
4390 } else
4391 return -EINVAL;
4392 return length;
4393}
4394SLAB_ATTR(shrink);
4395
Christoph Lameter81819f02007-05-06 14:49:36 -07004396#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004397static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004398{
Christoph Lameter98246012008-01-07 23:20:26 -08004399 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004400}
4401
Christoph Lameter98246012008-01-07 23:20:26 -08004402static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004403 const char *buf, size_t length)
4404{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004405 unsigned long ratio;
4406 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004407
Christoph Lameter0121c6192008-04-29 16:11:12 -07004408 err = strict_strtoul(buf, 10, &ratio);
4409 if (err)
4410 return err;
4411
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004412 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004413 s->remote_node_defrag_ratio = ratio * 10;
4414
Christoph Lameter81819f02007-05-06 14:49:36 -07004415 return length;
4416}
Christoph Lameter98246012008-01-07 23:20:26 -08004417SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004418#endif
4419
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004420#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004421static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4422{
4423 unsigned long sum = 0;
4424 int cpu;
4425 int len;
4426 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4427
4428 if (!data)
4429 return -ENOMEM;
4430
4431 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004432 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004433
4434 data[cpu] = x;
4435 sum += x;
4436 }
4437
4438 len = sprintf(buf, "%lu", sum);
4439
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004440#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004441 for_each_online_cpu(cpu) {
4442 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004443 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004444 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004445#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004446 kfree(data);
4447 return len + sprintf(buf + len, "\n");
4448}
4449
David Rientjes78eb00c2009-10-15 02:20:22 -07004450static void clear_stat(struct kmem_cache *s, enum stat_item si)
4451{
4452 int cpu;
4453
4454 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004455 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004456}
4457
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004458#define STAT_ATTR(si, text) \
4459static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4460{ \
4461 return show_stat(s, buf, si); \
4462} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004463static ssize_t text##_store(struct kmem_cache *s, \
4464 const char *buf, size_t length) \
4465{ \
4466 if (buf[0] != '0') \
4467 return -EINVAL; \
4468 clear_stat(s, si); \
4469 return length; \
4470} \
4471SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004472
4473STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4474STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4475STAT_ATTR(FREE_FASTPATH, free_fastpath);
4476STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4477STAT_ATTR(FREE_FROZEN, free_frozen);
4478STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4479STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4480STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4481STAT_ATTR(ALLOC_SLAB, alloc_slab);
4482STAT_ATTR(ALLOC_REFILL, alloc_refill);
4483STAT_ATTR(FREE_SLAB, free_slab);
4484STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4485STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4486STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4487STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4488STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4489STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004490STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004491#endif
4492
Pekka Enberg06428782008-01-07 23:20:27 -08004493static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004494 &slab_size_attr.attr,
4495 &object_size_attr.attr,
4496 &objs_per_slab_attr.attr,
4497 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004498 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004499 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004500 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004501 &partial_attr.attr,
4502 &cpu_slabs_attr.attr,
4503 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004504 &aliases_attr.attr,
4505 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004506 &hwcache_align_attr.attr,
4507 &reclaim_account_attr.attr,
4508 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004509 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004510 &reserved_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004511#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004512 &total_objects_attr.attr,
4513 &slabs_attr.attr,
4514 &sanity_checks_attr.attr,
4515 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004516 &red_zone_attr.attr,
4517 &poison_attr.attr,
4518 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004519 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004520 &alloc_calls_attr.attr,
4521 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004522#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004523#ifdef CONFIG_ZONE_DMA
4524 &cache_dma_attr.attr,
4525#endif
4526#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004527 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004528#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004529#ifdef CONFIG_SLUB_STATS
4530 &alloc_fastpath_attr.attr,
4531 &alloc_slowpath_attr.attr,
4532 &free_fastpath_attr.attr,
4533 &free_slowpath_attr.attr,
4534 &free_frozen_attr.attr,
4535 &free_add_partial_attr.attr,
4536 &free_remove_partial_attr.attr,
4537 &alloc_from_partial_attr.attr,
4538 &alloc_slab_attr.attr,
4539 &alloc_refill_attr.attr,
4540 &free_slab_attr.attr,
4541 &cpuslab_flush_attr.attr,
4542 &deactivate_full_attr.attr,
4543 &deactivate_empty_attr.attr,
4544 &deactivate_to_head_attr.attr,
4545 &deactivate_to_tail_attr.attr,
4546 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004547 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004548#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004549#ifdef CONFIG_FAILSLAB
4550 &failslab_attr.attr,
4551#endif
4552
Christoph Lameter81819f02007-05-06 14:49:36 -07004553 NULL
4554};
4555
4556static struct attribute_group slab_attr_group = {
4557 .attrs = slab_attrs,
4558};
4559
4560static ssize_t slab_attr_show(struct kobject *kobj,
4561 struct attribute *attr,
4562 char *buf)
4563{
4564 struct slab_attribute *attribute;
4565 struct kmem_cache *s;
4566 int err;
4567
4568 attribute = to_slab_attr(attr);
4569 s = to_slab(kobj);
4570
4571 if (!attribute->show)
4572 return -EIO;
4573
4574 err = attribute->show(s, buf);
4575
4576 return err;
4577}
4578
4579static ssize_t slab_attr_store(struct kobject *kobj,
4580 struct attribute *attr,
4581 const char *buf, size_t len)
4582{
4583 struct slab_attribute *attribute;
4584 struct kmem_cache *s;
4585 int err;
4586
4587 attribute = to_slab_attr(attr);
4588 s = to_slab(kobj);
4589
4590 if (!attribute->store)
4591 return -EIO;
4592
4593 err = attribute->store(s, buf, len);
4594
4595 return err;
4596}
4597
Christoph Lameter151c6022008-01-07 22:29:05 -08004598static void kmem_cache_release(struct kobject *kobj)
4599{
4600 struct kmem_cache *s = to_slab(kobj);
4601
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004602 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004603 kfree(s);
4604}
4605
Emese Revfy52cf25d2010-01-19 02:58:23 +01004606static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004607 .show = slab_attr_show,
4608 .store = slab_attr_store,
4609};
4610
4611static struct kobj_type slab_ktype = {
4612 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004613 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004614};
4615
4616static int uevent_filter(struct kset *kset, struct kobject *kobj)
4617{
4618 struct kobj_type *ktype = get_ktype(kobj);
4619
4620 if (ktype == &slab_ktype)
4621 return 1;
4622 return 0;
4623}
4624
Emese Revfy9cd43612009-12-31 14:52:51 +01004625static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004626 .filter = uevent_filter,
4627};
4628
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004629static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004630
4631#define ID_STR_LENGTH 64
4632
4633/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004634 *
4635 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004636 */
4637static char *create_unique_id(struct kmem_cache *s)
4638{
4639 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4640 char *p = name;
4641
4642 BUG_ON(!name);
4643
4644 *p++ = ':';
4645 /*
4646 * First flags affecting slabcache operations. We will only
4647 * get here for aliasable slabs so we do not need to support
4648 * too many flags. The flags here must cover all flags that
4649 * are matched during merging to guarantee that the id is
4650 * unique.
4651 */
4652 if (s->flags & SLAB_CACHE_DMA)
4653 *p++ = 'd';
4654 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4655 *p++ = 'a';
4656 if (s->flags & SLAB_DEBUG_FREE)
4657 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004658 if (!(s->flags & SLAB_NOTRACK))
4659 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004660 if (p != name + 1)
4661 *p++ = '-';
4662 p += sprintf(p, "%07d", s->size);
4663 BUG_ON(p > name + ID_STR_LENGTH - 1);
4664 return name;
4665}
4666
4667static int sysfs_slab_add(struct kmem_cache *s)
4668{
4669 int err;
4670 const char *name;
4671 int unmergeable;
4672
4673 if (slab_state < SYSFS)
4674 /* Defer until later */
4675 return 0;
4676
4677 unmergeable = slab_unmergeable(s);
4678 if (unmergeable) {
4679 /*
4680 * Slabcache can never be merged so we can use the name proper.
4681 * This is typically the case for debug situations. In that
4682 * case we can catch duplicate names easily.
4683 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004684 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004685 name = s->name;
4686 } else {
4687 /*
4688 * Create a unique name for the slab as a target
4689 * for the symlinks.
4690 */
4691 name = create_unique_id(s);
4692 }
4693
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004694 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004695 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4696 if (err) {
4697 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004698 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004699 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004700
4701 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004702 if (err) {
4703 kobject_del(&s->kobj);
4704 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004705 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004706 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004707 kobject_uevent(&s->kobj, KOBJ_ADD);
4708 if (!unmergeable) {
4709 /* Setup first alias */
4710 sysfs_slab_alias(s, s->name);
4711 kfree(name);
4712 }
4713 return 0;
4714}
4715
4716static void sysfs_slab_remove(struct kmem_cache *s)
4717{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004718 if (slab_state < SYSFS)
4719 /*
4720 * Sysfs has not been setup yet so no need to remove the
4721 * cache from sysfs.
4722 */
4723 return;
4724
Christoph Lameter81819f02007-05-06 14:49:36 -07004725 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4726 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004727 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004728}
4729
4730/*
4731 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004732 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004733 */
4734struct saved_alias {
4735 struct kmem_cache *s;
4736 const char *name;
4737 struct saved_alias *next;
4738};
4739
Adrian Bunk5af328a2007-07-17 04:03:27 -07004740static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004741
4742static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4743{
4744 struct saved_alias *al;
4745
4746 if (slab_state == SYSFS) {
4747 /*
4748 * If we have a leftover link then remove it.
4749 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004750 sysfs_remove_link(&slab_kset->kobj, name);
4751 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004752 }
4753
4754 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4755 if (!al)
4756 return -ENOMEM;
4757
4758 al->s = s;
4759 al->name = name;
4760 al->next = alias_list;
4761 alias_list = al;
4762 return 0;
4763}
4764
4765static int __init slab_sysfs_init(void)
4766{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004767 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004768 int err;
4769
Christoph Lameter2bce6482010-07-19 11:39:11 -05004770 down_write(&slub_lock);
4771
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004772 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004773 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004774 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004775 printk(KERN_ERR "Cannot register slab subsystem.\n");
4776 return -ENOSYS;
4777 }
4778
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004779 slab_state = SYSFS;
4780
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004781 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004782 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004783 if (err)
4784 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4785 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004786 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004787
4788 while (alias_list) {
4789 struct saved_alias *al = alias_list;
4790
4791 alias_list = alias_list->next;
4792 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004793 if (err)
4794 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4795 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004796 kfree(al);
4797 }
4798
Christoph Lameter2bce6482010-07-19 11:39:11 -05004799 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004800 resiliency_test();
4801 return 0;
4802}
4803
4804__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004805#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004806
4807/*
4808 * The /proc/slabinfo ABI
4809 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004810#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004811static void print_slabinfo_header(struct seq_file *m)
4812{
4813 seq_puts(m, "slabinfo - version: 2.1\n");
4814 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4815 "<objperslab> <pagesperslab>");
4816 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4817 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4818 seq_putc(m, '\n');
4819}
4820
4821static void *s_start(struct seq_file *m, loff_t *pos)
4822{
4823 loff_t n = *pos;
4824
4825 down_read(&slub_lock);
4826 if (!n)
4827 print_slabinfo_header(m);
4828
4829 return seq_list_start(&slab_caches, *pos);
4830}
4831
4832static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4833{
4834 return seq_list_next(p, &slab_caches, pos);
4835}
4836
4837static void s_stop(struct seq_file *m, void *p)
4838{
4839 up_read(&slub_lock);
4840}
4841
4842static int s_show(struct seq_file *m, void *p)
4843{
4844 unsigned long nr_partials = 0;
4845 unsigned long nr_slabs = 0;
4846 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004847 unsigned long nr_objs = 0;
4848 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004849 struct kmem_cache *s;
4850 int node;
4851
4852 s = list_entry(p, struct kmem_cache, list);
4853
4854 for_each_online_node(node) {
4855 struct kmem_cache_node *n = get_node(s, node);
4856
4857 if (!n)
4858 continue;
4859
4860 nr_partials += n->nr_partial;
4861 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004862 nr_objs += atomic_long_read(&n->total_objects);
4863 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004864 }
4865
Christoph Lameter205ab992008-04-14 19:11:40 +03004866 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004867
4868 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004869 nr_objs, s->size, oo_objects(s->oo),
4870 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004871 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4872 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4873 0UL);
4874 seq_putc(m, '\n');
4875 return 0;
4876}
4877
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004878static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004879 .start = s_start,
4880 .next = s_next,
4881 .stop = s_stop,
4882 .show = s_show,
4883};
4884
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004885static int slabinfo_open(struct inode *inode, struct file *file)
4886{
4887 return seq_open(file, &slabinfo_op);
4888}
4889
4890static const struct file_operations proc_slabinfo_operations = {
4891 .open = slabinfo_open,
4892 .read = seq_read,
4893 .llseek = seq_lseek,
4894 .release = seq_release,
4895};
4896
4897static int __init slab_proc_init(void)
4898{
WANG Congcf5d1132009-08-18 19:11:40 +03004899 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004900 return 0;
4901}
4902module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004903#endif /* CONFIG_SLABINFO */