blob: 7be0223531b090dbd9f716104714bc4d9ca9c7b5 [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
Lucas De Marchi25985ed2011-03-30 22:57:33 -030067 * a partial slab. A new slab has no one operating on it and thus there is
Christoph Lameter81819f02007-05-06 14:49:36 -070068 * 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 Lameter41ecc552007-05-09 02:32:44 -0700341#ifdef CONFIG_SLUB_DEBUG
342/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500343 * 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 Lameter01ad8a72011-04-15 14:48:14 -05001834 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001835 c->tid = next_tid(c->tid);
1836 local_irq_restore(flags);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001837 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001838 return object;
1839
1840another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001841 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001842
1843new_slab:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001844 page = get_partial(s, gfpflags, node);
1845 if (page) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001846 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameterdc1fb7f2011-04-15 14:48:15 -05001847 c->node = page_to_nid(page);
1848 c->page = page;
Christoph Lameter894b8782007-05-10 03:15:16 -07001849 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001850 }
1851
Christoph Lameterc1d50832010-08-20 12:37:17 -05001852 gfpflags &= gfp_allowed_mask;
Christoph Lameterb811c202007-10-16 23:25:51 -07001853 if (gfpflags & __GFP_WAIT)
1854 local_irq_enable();
1855
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001856 page = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001857
1858 if (gfpflags & __GFP_WAIT)
1859 local_irq_disable();
1860
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001861 if (page) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001862 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001863 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001864 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001865 flush_slab(s, c);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001866
1867 slab_lock(page);
1868 __SetPageSlubFrozen(page);
David Rientjesbd07d872011-05-12 13:10:49 -07001869 c->node = page_to_nid(page);
1870 c->page = page;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001871 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001872 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001873 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1874 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter2fd66c52011-03-22 13:32:53 -05001875 local_irq_restore(flags);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001876 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001877debug:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001878 if (!alloc_debug_processing(s, page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001879 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001880
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001881 page->inuse++;
1882 page->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05001883 deactivate_slab(s, c);
1884 c->page = NULL;
Pekka Enberg15b7c512010-10-02 11:32:32 +03001885 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05001886 local_irq_restore(flags);
1887 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07001888}
1889
1890/*
1891 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1892 * have the fastpath folded into their functions. So no function call
1893 * overhead for requests that can be satisfied on the fastpath.
1894 *
1895 * The fastpath works by first checking if the lockless freelist can be used.
1896 * If not then __slab_alloc is called for slow processing.
1897 *
1898 * Otherwise we can simply pick the next object from the lockless free list.
1899 */
Pekka Enberg06428782008-01-07 23:20:27 -08001900static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001901 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001902{
Christoph Lameter894b8782007-05-10 03:15:16 -07001903 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001904 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001905 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08001906
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001907 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001908 return NULL;
1909
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001910redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001911
1912 /*
1913 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1914 * enabled. We may switch back and forth between cpus while
1915 * reading from one cpu area. That does not matter as long
1916 * as we end up on the original cpu again when doing the cmpxchg.
1917 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001918 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001919
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001920 /*
1921 * The transaction ids are globally unique per cpu and per operation on
1922 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1923 * occurs on the right processor and that there was no operation on the
1924 * linked list in between.
1925 */
1926 tid = c->tid;
1927 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001928
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001929 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001930 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001931
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001932 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001933
1934 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001935 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001936 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001937 * operation and if we are on the right processor.
1938 *
1939 * The cmpxchg does the following atomically (without lock semantics!)
1940 * 1. Relocate first pointer to the current per cpu area.
1941 * 2. Verify that tid and freelist have not been changed
1942 * 3. If they were not changed replace tid and freelist
1943 *
1944 * Since this is without lock semantics the protection is only against
1945 * code executing on this cpu *not* from access by other cpus.
1946 */
Thomas Gleixner30106b82011-05-04 15:38:19 +02001947 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001948 s->cpu_slab->freelist, s->cpu_slab->tid,
1949 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05001950 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001951
1952 note_cmpxchg_failure("slab_alloc", s, tid);
1953 goto redo;
1954 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06001955 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001956 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001957
Pekka Enberg74e21342009-11-25 20:14:48 +02001958 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001959 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001960
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001961 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001962
Christoph Lameter894b8782007-05-10 03:15:16 -07001963 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001964}
1965
1966void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1967{
Christoph Lameter2154a332010-07-09 14:07:10 -05001968 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001969
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001970 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001971
1972 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001973}
1974EXPORT_SYMBOL(kmem_cache_alloc);
1975
Li Zefan0f24f122009-12-11 15:45:30 +08001976#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01001977void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001978{
Richard Kennedy4a923792010-10-21 10:29:19 +01001979 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1980 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1981 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001982}
Richard Kennedy4a923792010-10-21 10:29:19 +01001983EXPORT_SYMBOL(kmem_cache_alloc_trace);
1984
1985void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1986{
1987 void *ret = kmalloc_order(size, flags, order);
1988 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1989 return ret;
1990}
1991EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001992#endif
1993
Christoph Lameter81819f02007-05-06 14:49:36 -07001994#ifdef CONFIG_NUMA
1995void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1996{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001997 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1998
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001999 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2000 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002001
2002 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002003}
2004EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002005
Li Zefan0f24f122009-12-11 15:45:30 +08002006#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002007void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002008 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002009 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002010{
Richard Kennedy4a923792010-10-21 10:29:19 +01002011 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2012
2013 trace_kmalloc_node(_RET_IP_, ret,
2014 size, s->size, gfpflags, node);
2015 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002016}
Richard Kennedy4a923792010-10-21 10:29:19 +01002017EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002018#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002019#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002020
Christoph Lameter81819f02007-05-06 14:49:36 -07002021/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002022 * Slow patch handling. This may still be called frequently since objects
2023 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002024 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002025 * So we still attempt to reduce cache line usage. Just take the slab
2026 * lock and free the item. If there is no additional partial page
2027 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002028 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002029static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002030 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002031{
2032 void *prior;
2033 void **object = (void *)x;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002034 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002035
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002036 local_irq_save(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002037 slab_lock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002038 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002039
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002040 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2041 goto out_unlock;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002042
Christoph Lameterff120592009-12-18 16:26:22 -06002043 prior = page->freelist;
2044 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07002045 page->freelist = object;
2046 page->inuse--;
2047
Andy Whitcroft8a380822008-07-23 21:27:18 -07002048 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06002049 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07002050 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002051 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002052
2053 if (unlikely(!page->inuse))
2054 goto slab_empty;
2055
2056 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002057 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07002058 * then add it.
2059 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08002060 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002061 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002062 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002063 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002064
2065out_unlock:
2066 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002067 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002068 return;
2069
2070slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002071 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002072 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002073 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002074 */
2075 remove_partial(s, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002076 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002077 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002078 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002079 local_irq_restore(flags);
Christoph Lameter84e554e2009-12-18 16:26:23 -06002080 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002081 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002082}
2083
Christoph Lameter894b8782007-05-10 03:15:16 -07002084/*
2085 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2086 * can perform fastpath freeing without additional function calls.
2087 *
2088 * The fastpath is only possible if we are freeing to the current cpu slab
2089 * of this processor. This typically the case if we have just allocated
2090 * the item before.
2091 *
2092 * If fastpath is not possible then fall back to __slab_free where we deal
2093 * with all sorts of special processing.
2094 */
Pekka Enberg06428782008-01-07 23:20:27 -08002095static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002096 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002097{
2098 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002099 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002100 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002101
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002102 slab_free_hook(s, x);
2103
Christoph Lametera24c5a02011-03-15 12:45:21 -05002104redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002105
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002106 /*
2107 * Determine the currently cpus per cpu slab.
2108 * The cpu may change afterward. However that does not matter since
2109 * data is retrieved via this pointer. If we are on the same cpu
2110 * during the cmpxchg then the free will succedd.
2111 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002112 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002113
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002114 tid = c->tid;
2115 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002116
Christoph Lameter442b06b2011-05-17 16:29:31 -05002117 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002118 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002119
Thomas Gleixner30106b82011-05-04 15:38:19 +02002120 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002121 s->cpu_slab->freelist, s->cpu_slab->tid,
2122 c->freelist, tid,
2123 object, next_tid(tid)))) {
2124
2125 note_cmpxchg_failure("slab_free", s, tid);
2126 goto redo;
2127 }
Christoph Lameter84e554e2009-12-18 16:26:23 -06002128 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002129 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002130 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002131
Christoph Lameter894b8782007-05-10 03:15:16 -07002132}
2133
Christoph Lameter81819f02007-05-06 14:49:36 -07002134void kmem_cache_free(struct kmem_cache *s, void *x)
2135{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002136 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002137
Christoph Lameterb49af682007-05-06 14:49:41 -07002138 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002139
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002140 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002141
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002142 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002143}
2144EXPORT_SYMBOL(kmem_cache_free);
2145
Christoph Lameter81819f02007-05-06 14:49:36 -07002146/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002147 * Object placement in a slab is made very easy because we always start at
2148 * offset 0. If we tune the size of the object to the alignment then we can
2149 * get the required alignment by putting one properly sized object after
2150 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002151 *
2152 * Notice that the allocation order determines the sizes of the per cpu
2153 * caches. Each processor has always one slab available for allocations.
2154 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002155 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002156 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002157 */
2158
2159/*
2160 * Mininum / Maximum order of slab pages. This influences locking overhead
2161 * and slab fragmentation. A higher order reduces the number of partial slabs
2162 * and increases the number of allocations possible without having to
2163 * take the list_lock.
2164 */
2165static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002166static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002167static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002168
2169/*
2170 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002171 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002172 */
2173static int slub_nomerge;
2174
2175/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002176 * Calculate the order of allocation given an slab object size.
2177 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002178 * The order of allocation has significant impact on performance and other
2179 * system components. Generally order 0 allocations should be preferred since
2180 * order 0 does not cause fragmentation in the page allocator. Larger objects
2181 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002182 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002183 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002184 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002185 * In order to reach satisfactory performance we must ensure that a minimum
2186 * number of objects is in one slab. Otherwise we may generate too much
2187 * activity on the partial lists which requires taking the list_lock. This is
2188 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002189 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002190 * slub_max_order specifies the order where we begin to stop considering the
2191 * number of objects in a slab as critical. If we reach slub_max_order then
2192 * we try to keep the page order as low as possible. So we accept more waste
2193 * of space in favor of a small page order.
2194 *
2195 * Higher order allocations also allow the placement of more objects in a
2196 * slab and thereby reduce object handling overhead. If the user has
2197 * requested a higher mininum order then we start with that one instead of
2198 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002199 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002200static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002201 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002202{
2203 int order;
2204 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002205 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002206
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002207 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002208 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002209
Christoph Lameter6300ea72007-07-17 04:03:20 -07002210 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002211 fls(min_objects * size - 1) - PAGE_SHIFT);
2212 order <= max_order; order++) {
2213
Christoph Lameter81819f02007-05-06 14:49:36 -07002214 unsigned long slab_size = PAGE_SIZE << order;
2215
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002216 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002217 continue;
2218
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002219 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002220
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002221 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002222 break;
2223
2224 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002225
Christoph Lameter81819f02007-05-06 14:49:36 -07002226 return order;
2227}
2228
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002229static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002230{
2231 int order;
2232 int min_objects;
2233 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002234 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002235
2236 /*
2237 * Attempt to find best configuration for a slab. This
2238 * works by first attempting to generate a layout with
2239 * the best configuration and backing off gradually.
2240 *
2241 * First we reduce the acceptable waste in a slab. Then
2242 * we reduce the minimum objects required in a slab.
2243 */
2244 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002245 if (!min_objects)
2246 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002247 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002248 min_objects = min(min_objects, max_objects);
2249
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002250 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002251 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002252 while (fraction >= 4) {
2253 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002254 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002255 if (order <= slub_max_order)
2256 return order;
2257 fraction /= 2;
2258 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002259 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002260 }
2261
2262 /*
2263 * We were unable to place multiple objects in a slab. Now
2264 * lets see if we can place a single object there.
2265 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002266 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002267 if (order <= slub_max_order)
2268 return order;
2269
2270 /*
2271 * Doh this slab cannot be placed using slub_max_order.
2272 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002273 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002274 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002275 return order;
2276 return -ENOSYS;
2277}
2278
Christoph Lameter81819f02007-05-06 14:49:36 -07002279/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002280 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002281 */
2282static unsigned long calculate_alignment(unsigned long flags,
2283 unsigned long align, unsigned long size)
2284{
2285 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002286 * If the user wants hardware cache aligned objects then follow that
2287 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002288 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002289 * The hardware cache alignment cannot override the specified
2290 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002291 */
Nick Pigginb6210382008-03-05 14:05:56 -08002292 if (flags & SLAB_HWCACHE_ALIGN) {
2293 unsigned long ralign = cache_line_size();
2294 while (size <= ralign / 2)
2295 ralign /= 2;
2296 align = max(align, ralign);
2297 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002298
2299 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002300 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002301
2302 return ALIGN(align, sizeof(void *));
2303}
2304
Pekka Enberg5595cff2008-08-05 09:28:47 +03002305static void
2306init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002307{
2308 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002309 spin_lock_init(&n->list_lock);
2310 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002311#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002312 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002313 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002314 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002315#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002316}
2317
Christoph Lameter55136592010-08-20 12:37:13 -05002318static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002319{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002320 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2321 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002322
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002323#ifdef CONFIG_CMPXCHG_LOCAL
2324 /*
2325 * Must align to double word boundary for the double cmpxchg instructions
2326 * to work.
2327 */
2328 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
2329#else
2330 /* Regular alignment is sufficient */
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002331 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002332#endif
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002333
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002334 if (!s->cpu_slab)
2335 return 0;
2336
2337 init_kmem_cache_cpus(s);
2338
2339 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002340}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002341
Christoph Lameter51df1142010-08-20 12:37:15 -05002342static struct kmem_cache *kmem_cache_node;
2343
Christoph Lameter81819f02007-05-06 14:49:36 -07002344/*
2345 * No kmalloc_node yet so do it by hand. We know that this is the first
2346 * slab on the node for this slabcache. There are no concurrent accesses
2347 * possible.
2348 *
2349 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002350 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2351 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002352 */
Christoph Lameter55136592010-08-20 12:37:13 -05002353static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002354{
2355 struct page *page;
2356 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002357 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002358
Christoph Lameter51df1142010-08-20 12:37:15 -05002359 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002360
Christoph Lameter51df1142010-08-20 12:37:15 -05002361 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002362
2363 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002364 if (page_to_nid(page) != node) {
2365 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2366 "node %d\n", node);
2367 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2368 "in order to be able to continue\n");
2369 }
2370
Christoph Lameter81819f02007-05-06 14:49:36 -07002371 n = page->freelist;
2372 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002373 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002374 page->inuse++;
Christoph Lameter51df1142010-08-20 12:37:15 -05002375 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002376#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002377 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002378 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002379#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002380 init_kmem_cache_node(n, kmem_cache_node);
2381 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002382
rootba84c732008-01-07 23:20:28 -08002383 /*
2384 * lockdep requires consistent irq usage for each lock
2385 * so even though there cannot be a race this early in
2386 * the boot sequence, we still disable irqs.
2387 */
2388 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002389 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002390 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002391}
2392
2393static void free_kmem_cache_nodes(struct kmem_cache *s)
2394{
2395 int node;
2396
Christoph Lameterf64dc582007-10-16 01:25:33 -07002397 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002398 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002399
Alexander Duyck73367bd2010-05-21 14:41:35 -07002400 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002401 kmem_cache_free(kmem_cache_node, n);
2402
Christoph Lameter81819f02007-05-06 14:49:36 -07002403 s->node[node] = NULL;
2404 }
2405}
2406
Christoph Lameter55136592010-08-20 12:37:13 -05002407static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002408{
2409 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002410
Christoph Lameterf64dc582007-10-16 01:25:33 -07002411 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002412 struct kmem_cache_node *n;
2413
Alexander Duyck73367bd2010-05-21 14:41:35 -07002414 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002415 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002416 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002417 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002418 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002419 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002420
2421 if (!n) {
2422 free_kmem_cache_nodes(s);
2423 return 0;
2424 }
2425
Christoph Lameter81819f02007-05-06 14:49:36 -07002426 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002427 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002428 }
2429 return 1;
2430}
Christoph Lameter81819f02007-05-06 14:49:36 -07002431
David Rientjesc0bdb232009-02-25 09:16:35 +02002432static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002433{
2434 if (min < MIN_PARTIAL)
2435 min = MIN_PARTIAL;
2436 else if (min > MAX_PARTIAL)
2437 min = MAX_PARTIAL;
2438 s->min_partial = min;
2439}
2440
Christoph Lameter81819f02007-05-06 14:49:36 -07002441/*
2442 * calculate_sizes() determines the order and the distribution of data within
2443 * a slab object.
2444 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002445static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002446{
2447 unsigned long flags = s->flags;
2448 unsigned long size = s->objsize;
2449 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002450 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002451
2452 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002453 * Round up object size to the next word boundary. We can only
2454 * place the free pointer at word boundaries and this determines
2455 * the possible location of the free pointer.
2456 */
2457 size = ALIGN(size, sizeof(void *));
2458
2459#ifdef CONFIG_SLUB_DEBUG
2460 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002461 * Determine if we can poison the object itself. If the user of
2462 * the slab may touch the object after free or before allocation
2463 * then we should never poison the object itself.
2464 */
2465 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002466 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002467 s->flags |= __OBJECT_POISON;
2468 else
2469 s->flags &= ~__OBJECT_POISON;
2470
Christoph Lameter81819f02007-05-06 14:49:36 -07002471
2472 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002473 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002474 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002475 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002476 */
2477 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2478 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002479#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002480
2481 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002482 * With that we have determined the number of bytes in actual use
2483 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002484 */
2485 s->inuse = size;
2486
2487 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002488 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002489 /*
2490 * Relocate free pointer after the object if it is not
2491 * permitted to overwrite the first word of the object on
2492 * kmem_cache_free.
2493 *
2494 * This is the case if we do RCU, have a constructor or
2495 * destructor or are poisoning the objects.
2496 */
2497 s->offset = size;
2498 size += sizeof(void *);
2499 }
2500
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002501#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002502 if (flags & SLAB_STORE_USER)
2503 /*
2504 * Need to store information about allocs and frees after
2505 * the object.
2506 */
2507 size += 2 * sizeof(struct track);
2508
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002509 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002510 /*
2511 * Add some empty padding so that we can catch
2512 * overwrites from earlier objects rather than let
2513 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002514 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002515 * of the object.
2516 */
2517 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002518#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002519
Christoph Lameter81819f02007-05-06 14:49:36 -07002520 /*
2521 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002522 * user specified and the dynamic determination of cache line size
2523 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002524 */
2525 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002526 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002527
2528 /*
2529 * SLUB stores one object immediately after another beginning from
2530 * offset 0. In order to align the objects we have to simply size
2531 * each object to conform to the alignment.
2532 */
2533 size = ALIGN(size, align);
2534 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002535 if (forced_order >= 0)
2536 order = forced_order;
2537 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002538 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002539
Christoph Lameter834f3d12008-04-14 19:11:31 +03002540 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002541 return 0;
2542
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002543 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002544 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002545 s->allocflags |= __GFP_COMP;
2546
2547 if (s->flags & SLAB_CACHE_DMA)
2548 s->allocflags |= SLUB_DMA;
2549
2550 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2551 s->allocflags |= __GFP_RECLAIMABLE;
2552
Christoph Lameter81819f02007-05-06 14:49:36 -07002553 /*
2554 * Determine the number of objects per slab
2555 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002556 s->oo = oo_make(order, size, s->reserved);
2557 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002558 if (oo_objects(s->oo) > oo_objects(s->max))
2559 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002560
Christoph Lameter834f3d12008-04-14 19:11:31 +03002561 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002562
2563}
2564
Christoph Lameter55136592010-08-20 12:37:13 -05002565static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002566 const char *name, size_t size,
2567 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002568 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002569{
2570 memset(s, 0, kmem_size);
2571 s->name = name;
2572 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002573 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002574 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002575 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002576 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002577
Lai Jiangshanda9a6382011-03-10 15:22:00 +08002578 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2579 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07002580
Christoph Lameter06b285d2008-04-14 19:11:41 +03002581 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002582 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002583 if (disable_higher_order_debug) {
2584 /*
2585 * Disable debugging flags that store metadata if the min slab
2586 * order increased.
2587 */
2588 if (get_order(s->size) > get_order(s->objsize)) {
2589 s->flags &= ~DEBUG_METADATA_FLAGS;
2590 s->offset = 0;
2591 if (!calculate_sizes(s, -1))
2592 goto error;
2593 }
2594 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002595
David Rientjes3b89d7d2009-02-22 17:40:07 -08002596 /*
2597 * The larger the object size is, the more pages we want on the partial
2598 * list to avoid pounding the page allocator excessively.
2599 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002600 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002601 s->refcount = 1;
2602#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002603 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002604#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002605 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002606 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002607
Christoph Lameter55136592010-08-20 12:37:13 -05002608 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002609 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002610
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002611 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002612error:
2613 if (flags & SLAB_PANIC)
2614 panic("Cannot create slab %s size=%lu realsize=%u "
2615 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002616 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002617 s->offset, flags);
2618 return 0;
2619}
Christoph Lameter81819f02007-05-06 14:49:36 -07002620
2621/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002622 * Determine the size of a slab object
2623 */
2624unsigned int kmem_cache_size(struct kmem_cache *s)
2625{
2626 return s->objsize;
2627}
2628EXPORT_SYMBOL(kmem_cache_size);
2629
Christoph Lameter33b12c32008-04-25 12:22:43 -07002630static void list_slab_objects(struct kmem_cache *s, struct page *page,
2631 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002632{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002633#ifdef CONFIG_SLUB_DEBUG
2634 void *addr = page_address(page);
2635 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002636 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2637 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002638 if (!map)
2639 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002640 slab_err(s, page, "%s", text);
2641 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002642
Christoph Lameter5f80b132011-04-15 14:48:13 -05002643 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002644 for_each_object(p, s, addr, page->objects) {
2645
2646 if (!test_bit(slab_index(p, s, addr), map)) {
2647 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2648 p, p - addr);
2649 print_tracking(s, p);
2650 }
2651 }
2652 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002653 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002654#endif
2655}
2656
Christoph Lameter81819f02007-05-06 14:49:36 -07002657/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002658 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002659 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002660static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002661{
Christoph Lameter81819f02007-05-06 14:49:36 -07002662 unsigned long flags;
2663 struct page *page, *h;
2664
2665 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002666 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002667 if (!page->inuse) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05002668 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002669 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002670 } else {
2671 list_slab_objects(s, page,
2672 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002673 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002674 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002675 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002676}
2677
2678/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002679 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002680 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002681static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002682{
2683 int node;
2684
2685 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002686 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002687 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002688 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002689 struct kmem_cache_node *n = get_node(s, node);
2690
Christoph Lameter599870b2008-04-23 12:36:52 -07002691 free_partial(s, n);
2692 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002693 return 1;
2694 }
2695 free_kmem_cache_nodes(s);
2696 return 0;
2697}
2698
2699/*
2700 * Close a cache and release the kmem_cache structure
2701 * (must be used for caches created using kmem_cache_create)
2702 */
2703void kmem_cache_destroy(struct kmem_cache *s)
2704{
2705 down_write(&slub_lock);
2706 s->refcount--;
2707 if (!s->refcount) {
2708 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002709 if (kmem_cache_close(s)) {
2710 printk(KERN_ERR "SLUB %s: %s called for cache that "
2711 "still has objects.\n", s->name, __func__);
2712 dump_stack();
2713 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002714 if (s->flags & SLAB_DESTROY_BY_RCU)
2715 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002716 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002717 }
2718 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002719}
2720EXPORT_SYMBOL(kmem_cache_destroy);
2721
2722/********************************************************************
2723 * Kmalloc subsystem
2724 *******************************************************************/
2725
Christoph Lameter51df1142010-08-20 12:37:15 -05002726struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002727EXPORT_SYMBOL(kmalloc_caches);
2728
Christoph Lameter51df1142010-08-20 12:37:15 -05002729static struct kmem_cache *kmem_cache;
2730
Christoph Lameter55136592010-08-20 12:37:13 -05002731#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002732static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002733#endif
2734
Christoph Lameter81819f02007-05-06 14:49:36 -07002735static int __init setup_slub_min_order(char *str)
2736{
Pekka Enberg06428782008-01-07 23:20:27 -08002737 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002738
2739 return 1;
2740}
2741
2742__setup("slub_min_order=", setup_slub_min_order);
2743
2744static int __init setup_slub_max_order(char *str)
2745{
Pekka Enberg06428782008-01-07 23:20:27 -08002746 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002747 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002748
2749 return 1;
2750}
2751
2752__setup("slub_max_order=", setup_slub_max_order);
2753
2754static int __init setup_slub_min_objects(char *str)
2755{
Pekka Enberg06428782008-01-07 23:20:27 -08002756 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002757
2758 return 1;
2759}
2760
2761__setup("slub_min_objects=", setup_slub_min_objects);
2762
2763static int __init setup_slub_nomerge(char *str)
2764{
2765 slub_nomerge = 1;
2766 return 1;
2767}
2768
2769__setup("slub_nomerge", setup_slub_nomerge);
2770
Christoph Lameter51df1142010-08-20 12:37:15 -05002771static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2772 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002773{
Christoph Lameter51df1142010-08-20 12:37:15 -05002774 struct kmem_cache *s;
2775
2776 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2777
Pekka Enberg83b519e2009-06-10 19:40:04 +03002778 /*
2779 * This function is called with IRQs disabled during early-boot on
2780 * single CPU so there's no need to take slub_lock here.
2781 */
Christoph Lameter55136592010-08-20 12:37:13 -05002782 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002783 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002784 goto panic;
2785
2786 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002787 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002788
2789panic:
2790 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002791 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002792}
2793
Christoph Lameterf1b26332007-07-17 04:03:26 -07002794/*
2795 * Conversion table for small slabs sizes / 8 to the index in the
2796 * kmalloc array. This is necessary for slabs < 192 since we have non power
2797 * of two cache sizes there. The size of larger slabs can be determined using
2798 * fls.
2799 */
2800static s8 size_index[24] = {
2801 3, /* 8 */
2802 4, /* 16 */
2803 5, /* 24 */
2804 5, /* 32 */
2805 6, /* 40 */
2806 6, /* 48 */
2807 6, /* 56 */
2808 6, /* 64 */
2809 1, /* 72 */
2810 1, /* 80 */
2811 1, /* 88 */
2812 1, /* 96 */
2813 7, /* 104 */
2814 7, /* 112 */
2815 7, /* 120 */
2816 7, /* 128 */
2817 2, /* 136 */
2818 2, /* 144 */
2819 2, /* 152 */
2820 2, /* 160 */
2821 2, /* 168 */
2822 2, /* 176 */
2823 2, /* 184 */
2824 2 /* 192 */
2825};
2826
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002827static inline int size_index_elem(size_t bytes)
2828{
2829 return (bytes - 1) / 8;
2830}
2831
Christoph Lameter81819f02007-05-06 14:49:36 -07002832static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2833{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002834 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002835
Christoph Lameterf1b26332007-07-17 04:03:26 -07002836 if (size <= 192) {
2837 if (!size)
2838 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002839
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002840 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002841 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002842 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002843
2844#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002845 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002846 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002847
Christoph Lameter81819f02007-05-06 14:49:36 -07002848#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002849 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002850}
2851
2852void *__kmalloc(size_t size, gfp_t flags)
2853{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002854 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002855 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002856
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002857 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002858 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002859
2860 s = get_slab(size, flags);
2861
2862 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002863 return s;
2864
Christoph Lameter2154a332010-07-09 14:07:10 -05002865 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002866
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002867 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002868
2869 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002870}
2871EXPORT_SYMBOL(__kmalloc);
2872
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002873#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002874static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2875{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002876 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002877 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002878
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002879 flags |= __GFP_COMP | __GFP_NOTRACK;
2880 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002881 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002882 ptr = page_address(page);
2883
2884 kmemleak_alloc(ptr, size, 1, flags);
2885 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002886}
2887
Christoph Lameter81819f02007-05-06 14:49:36 -07002888void *__kmalloc_node(size_t size, gfp_t flags, int node)
2889{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002890 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002891 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002892
Ingo Molnar057685c2009-02-20 12:15:30 +01002893 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002894 ret = kmalloc_large_node(size, flags, node);
2895
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002896 trace_kmalloc_node(_RET_IP_, ret,
2897 size, PAGE_SIZE << get_order(size),
2898 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002899
2900 return ret;
2901 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002902
2903 s = get_slab(size, flags);
2904
2905 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002906 return s;
2907
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002908 ret = slab_alloc(s, flags, node, _RET_IP_);
2909
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002910 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002911
2912 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002913}
2914EXPORT_SYMBOL(__kmalloc_node);
2915#endif
2916
2917size_t ksize(const void *object)
2918{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002919 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002920
Christoph Lameteref8b4522007-10-16 01:24:46 -07002921 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002922 return 0;
2923
Vegard Nossum294a80a2007-12-04 23:45:30 -08002924 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002925
Pekka Enberg76994412008-05-22 19:22:25 +03002926 if (unlikely(!PageSlab(page))) {
2927 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002928 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002929 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002930
Eric Dumazetb3d41882011-02-14 18:35:22 +01002931 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002932}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002933EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002934
2935void kfree(const void *x)
2936{
Christoph Lameter81819f02007-05-06 14:49:36 -07002937 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002938 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002939
Pekka Enberg2121db72009-03-25 11:05:57 +02002940 trace_kfree(_RET_IP_, x);
2941
Satyam Sharma2408c552007-10-16 01:24:44 -07002942 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002943 return;
2944
Christoph Lameterb49af682007-05-06 14:49:41 -07002945 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002946 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002947 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002948 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002949 put_page(page);
2950 return;
2951 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002952 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002953}
2954EXPORT_SYMBOL(kfree);
2955
Christoph Lameter2086d262007-05-06 14:49:46 -07002956/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002957 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2958 * the remaining slabs by the number of items in use. The slabs with the
2959 * most items in use come first. New allocations will then fill those up
2960 * and thus they can be removed from the partial lists.
2961 *
2962 * The slabs with the least items are placed last. This results in them
2963 * being allocated from last increasing the chance that the last objects
2964 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002965 */
2966int kmem_cache_shrink(struct kmem_cache *s)
2967{
2968 int node;
2969 int i;
2970 struct kmem_cache_node *n;
2971 struct page *page;
2972 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002973 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002974 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002975 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002976 unsigned long flags;
2977
2978 if (!slabs_by_inuse)
2979 return -ENOMEM;
2980
2981 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002982 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002983 n = get_node(s, node);
2984
2985 if (!n->nr_partial)
2986 continue;
2987
Christoph Lameter834f3d12008-04-14 19:11:31 +03002988 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002989 INIT_LIST_HEAD(slabs_by_inuse + i);
2990
2991 spin_lock_irqsave(&n->list_lock, flags);
2992
2993 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002994 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002995 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002996 * Note that concurrent frees may occur while we hold the
2997 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002998 */
2999 list_for_each_entry_safe(page, t, &n->partial, lru) {
3000 if (!page->inuse && slab_trylock(page)) {
3001 /*
3002 * Must hold slab lock here because slab_free
3003 * may have freed the last object and be
3004 * waiting to release the slab.
3005 */
Christoph Lameter62e346a2010-09-28 08:10:28 -05003006 __remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003007 slab_unlock(page);
3008 discard_slab(s, page);
3009 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07003010 list_move(&page->lru,
3011 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07003012 }
3013 }
3014
Christoph Lameter2086d262007-05-06 14:49:46 -07003015 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003016 * Rebuild the partial list with the slabs filled up most
3017 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003018 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003019 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003020 list_splice(slabs_by_inuse + i, n->partial.prev);
3021
Christoph Lameter2086d262007-05-06 14:49:46 -07003022 spin_unlock_irqrestore(&n->list_lock, flags);
3023 }
3024
3025 kfree(slabs_by_inuse);
3026 return 0;
3027}
3028EXPORT_SYMBOL(kmem_cache_shrink);
3029
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003030#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003031static int slab_mem_going_offline_callback(void *arg)
3032{
3033 struct kmem_cache *s;
3034
3035 down_read(&slub_lock);
3036 list_for_each_entry(s, &slab_caches, list)
3037 kmem_cache_shrink(s);
3038 up_read(&slub_lock);
3039
3040 return 0;
3041}
3042
3043static void slab_mem_offline_callback(void *arg)
3044{
3045 struct kmem_cache_node *n;
3046 struct kmem_cache *s;
3047 struct memory_notify *marg = arg;
3048 int offline_node;
3049
3050 offline_node = marg->status_change_nid;
3051
3052 /*
3053 * If the node still has available memory. we need kmem_cache_node
3054 * for it yet.
3055 */
3056 if (offline_node < 0)
3057 return;
3058
3059 down_read(&slub_lock);
3060 list_for_each_entry(s, &slab_caches, list) {
3061 n = get_node(s, offline_node);
3062 if (n) {
3063 /*
3064 * if n->nr_slabs > 0, slabs still exist on the node
3065 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003066 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003067 * callback. So, we must fail.
3068 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003069 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003070
3071 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003072 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003073 }
3074 }
3075 up_read(&slub_lock);
3076}
3077
3078static int slab_mem_going_online_callback(void *arg)
3079{
3080 struct kmem_cache_node *n;
3081 struct kmem_cache *s;
3082 struct memory_notify *marg = arg;
3083 int nid = marg->status_change_nid;
3084 int ret = 0;
3085
3086 /*
3087 * If the node's memory is already available, then kmem_cache_node is
3088 * already created. Nothing to do.
3089 */
3090 if (nid < 0)
3091 return 0;
3092
3093 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003094 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003095 * allocate a kmem_cache_node structure in order to bring the node
3096 * online.
3097 */
3098 down_read(&slub_lock);
3099 list_for_each_entry(s, &slab_caches, list) {
3100 /*
3101 * XXX: kmem_cache_alloc_node will fallback to other nodes
3102 * since memory is not yet available from the node that
3103 * is brought up.
3104 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003105 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003106 if (!n) {
3107 ret = -ENOMEM;
3108 goto out;
3109 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003110 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003111 s->node[nid] = n;
3112 }
3113out:
3114 up_read(&slub_lock);
3115 return ret;
3116}
3117
3118static int slab_memory_callback(struct notifier_block *self,
3119 unsigned long action, void *arg)
3120{
3121 int ret = 0;
3122
3123 switch (action) {
3124 case MEM_GOING_ONLINE:
3125 ret = slab_mem_going_online_callback(arg);
3126 break;
3127 case MEM_GOING_OFFLINE:
3128 ret = slab_mem_going_offline_callback(arg);
3129 break;
3130 case MEM_OFFLINE:
3131 case MEM_CANCEL_ONLINE:
3132 slab_mem_offline_callback(arg);
3133 break;
3134 case MEM_ONLINE:
3135 case MEM_CANCEL_OFFLINE:
3136 break;
3137 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003138 if (ret)
3139 ret = notifier_from_errno(ret);
3140 else
3141 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003142 return ret;
3143}
3144
3145#endif /* CONFIG_MEMORY_HOTPLUG */
3146
Christoph Lameter81819f02007-05-06 14:49:36 -07003147/********************************************************************
3148 * Basic setup of slabs
3149 *******************************************************************/
3150
Christoph Lameter51df1142010-08-20 12:37:15 -05003151/*
3152 * Used for early kmem_cache structures that were allocated using
3153 * the page allocator
3154 */
3155
3156static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3157{
3158 int node;
3159
3160 list_add(&s->list, &slab_caches);
3161 s->refcount = -1;
3162
3163 for_each_node_state(node, N_NORMAL_MEMORY) {
3164 struct kmem_cache_node *n = get_node(s, node);
3165 struct page *p;
3166
3167 if (n) {
3168 list_for_each_entry(p, &n->partial, lru)
3169 p->slab = s;
3170
Li Zefan607bf322011-04-12 15:22:26 +08003171#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003172 list_for_each_entry(p, &n->full, lru)
3173 p->slab = s;
3174#endif
3175 }
3176 }
3177}
3178
Christoph Lameter81819f02007-05-06 14:49:36 -07003179void __init kmem_cache_init(void)
3180{
3181 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003182 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003183 struct kmem_cache *temp_kmem_cache;
3184 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003185 struct kmem_cache *temp_kmem_cache_node;
3186 unsigned long kmalloc_size;
3187
3188 kmem_size = offsetof(struct kmem_cache, node) +
3189 nr_node_ids * sizeof(struct kmem_cache_node *);
3190
3191 /* Allocate two kmem_caches from the page allocator */
3192 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3193 order = get_order(2 * kmalloc_size);
3194 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3195
Christoph Lameter81819f02007-05-06 14:49:36 -07003196 /*
3197 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003198 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003199 * kmem_cache_open for slab_state == DOWN.
3200 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003201 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3202
3203 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3204 sizeof(struct kmem_cache_node),
3205 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003206
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003207 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003208
3209 /* Able to allocate the per node structures */
3210 slab_state = PARTIAL;
3211
Christoph Lameter51df1142010-08-20 12:37:15 -05003212 temp_kmem_cache = kmem_cache;
3213 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3214 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3215 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3216 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003217
Christoph Lameter51df1142010-08-20 12:37:15 -05003218 /*
3219 * Allocate kmem_cache_node properly from the kmem_cache slab.
3220 * kmem_cache_node is separately allocated so no need to
3221 * update any list pointers.
3222 */
3223 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003224
Christoph Lameter51df1142010-08-20 12:37:15 -05003225 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3226 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3227
3228 kmem_cache_bootstrap_fixup(kmem_cache_node);
3229
3230 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003231 kmem_cache_bootstrap_fixup(kmem_cache);
3232 caches++;
3233 /* Free temporary boot structure */
3234 free_pages((unsigned long)temp_kmem_cache, order);
3235
3236 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003237
3238 /*
3239 * Patch up the size_index table if we have strange large alignment
3240 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003241 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003242 *
3243 * Largest permitted alignment is 256 bytes due to the way we
3244 * handle the index determination for the smaller caches.
3245 *
3246 * Make sure that nothing crazy happens if someone starts tinkering
3247 * around with ARCH_KMALLOC_MINALIGN
3248 */
3249 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3250 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3251
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003252 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3253 int elem = size_index_elem(i);
3254 if (elem >= ARRAY_SIZE(size_index))
3255 break;
3256 size_index[elem] = KMALLOC_SHIFT_LOW;
3257 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003258
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003259 if (KMALLOC_MIN_SIZE == 64) {
3260 /*
3261 * The 96 byte size cache is not used if the alignment
3262 * is 64 byte.
3263 */
3264 for (i = 64 + 8; i <= 96; i += 8)
3265 size_index[size_index_elem(i)] = 7;
3266 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003267 /*
3268 * The 192 byte sized cache is not used if the alignment
3269 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3270 * instead.
3271 */
3272 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003273 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003274 }
3275
Christoph Lameter51df1142010-08-20 12:37:15 -05003276 /* Caches that are not of the two-to-the-power-of size */
3277 if (KMALLOC_MIN_SIZE <= 32) {
3278 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3279 caches++;
3280 }
3281
3282 if (KMALLOC_MIN_SIZE <= 64) {
3283 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3284 caches++;
3285 }
3286
3287 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3288 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3289 caches++;
3290 }
3291
Christoph Lameter81819f02007-05-06 14:49:36 -07003292 slab_state = UP;
3293
3294 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003295 if (KMALLOC_MIN_SIZE <= 32) {
3296 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3297 BUG_ON(!kmalloc_caches[1]->name);
3298 }
3299
3300 if (KMALLOC_MIN_SIZE <= 64) {
3301 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3302 BUG_ON(!kmalloc_caches[2]->name);
3303 }
3304
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003305 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3306 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3307
3308 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003309 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003310 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003311
3312#ifdef CONFIG_SMP
3313 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003314#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003315
Christoph Lameter55136592010-08-20 12:37:13 -05003316#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003317 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3318 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003319
Christoph Lameter51df1142010-08-20 12:37:15 -05003320 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003321 char *name = kasprintf(GFP_NOWAIT,
3322 "dma-kmalloc-%d", s->objsize);
3323
3324 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003325 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3326 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003327 }
3328 }
3329#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003330 printk(KERN_INFO
3331 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003332 " CPUs=%d, Nodes=%d\n",
3333 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003334 slub_min_order, slub_max_order, slub_min_objects,
3335 nr_cpu_ids, nr_node_ids);
3336}
3337
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003338void __init kmem_cache_init_late(void)
3339{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003340}
3341
Christoph Lameter81819f02007-05-06 14:49:36 -07003342/*
3343 * Find a mergeable slab cache
3344 */
3345static int slab_unmergeable(struct kmem_cache *s)
3346{
3347 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3348 return 1;
3349
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003350 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003351 return 1;
3352
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003353 /*
3354 * We may have set a slab to be unmergeable during bootstrap.
3355 */
3356 if (s->refcount < 0)
3357 return 1;
3358
Christoph Lameter81819f02007-05-06 14:49:36 -07003359 return 0;
3360}
3361
3362static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003363 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003364 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003365{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003366 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003367
3368 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3369 return NULL;
3370
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003371 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003372 return NULL;
3373
3374 size = ALIGN(size, sizeof(void *));
3375 align = calculate_alignment(flags, align, size);
3376 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003377 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003378
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003379 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003380 if (slab_unmergeable(s))
3381 continue;
3382
3383 if (size > s->size)
3384 continue;
3385
Christoph Lameterba0268a2007-09-11 15:24:11 -07003386 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003387 continue;
3388 /*
3389 * Check if alignment is compatible.
3390 * Courtesy of Adrian Drzewiecki
3391 */
Pekka Enberg06428782008-01-07 23:20:27 -08003392 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003393 continue;
3394
3395 if (s->size - size >= sizeof(void *))
3396 continue;
3397
3398 return s;
3399 }
3400 return NULL;
3401}
3402
3403struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003404 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003405{
3406 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003407 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003408
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003409 if (WARN_ON(!name))
3410 return NULL;
3411
Christoph Lameter81819f02007-05-06 14:49:36 -07003412 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003413 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003414 if (s) {
3415 s->refcount++;
3416 /*
3417 * Adjust the object sizes so that we clear
3418 * the complete object on kzalloc.
3419 */
3420 s->objsize = max(s->objsize, (int)size);
3421 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003422
David Rientjes7b8f3b62008-12-17 22:09:46 -08003423 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003424 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003425 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003426 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003427 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003428 return s;
3429 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003430
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003431 n = kstrdup(name, GFP_KERNEL);
3432 if (!n)
3433 goto err;
3434
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003435 s = kmalloc(kmem_size, GFP_KERNEL);
3436 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003437 if (kmem_cache_open(s, n,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003438 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003439 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003440 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003441 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003442 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003443 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003444 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003445 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003446 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003447 return s;
3448 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003449 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003450 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003451 }
Pavel Emelyanov68cee4f2010-10-28 13:50:37 +04003452err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003453 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003454
Christoph Lameter81819f02007-05-06 14:49:36 -07003455 if (flags & SLAB_PANIC)
3456 panic("Cannot create slabcache %s\n", name);
3457 else
3458 s = NULL;
3459 return s;
3460}
3461EXPORT_SYMBOL(kmem_cache_create);
3462
Christoph Lameter81819f02007-05-06 14:49:36 -07003463#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003464/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003465 * Use the cpu notifier to insure that the cpu slabs are flushed when
3466 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003467 */
3468static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3469 unsigned long action, void *hcpu)
3470{
3471 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003472 struct kmem_cache *s;
3473 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003474
3475 switch (action) {
3476 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003477 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003478 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003479 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003480 down_read(&slub_lock);
3481 list_for_each_entry(s, &slab_caches, list) {
3482 local_irq_save(flags);
3483 __flush_cpu_slab(s, cpu);
3484 local_irq_restore(flags);
3485 }
3486 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003487 break;
3488 default:
3489 break;
3490 }
3491 return NOTIFY_OK;
3492}
3493
Pekka Enberg06428782008-01-07 23:20:27 -08003494static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003495 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003496};
Christoph Lameter81819f02007-05-06 14:49:36 -07003497
3498#endif
3499
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003500void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003501{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003502 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003503 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003504
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003505 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003506 return kmalloc_large(size, gfpflags);
3507
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003508 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003509
Satyam Sharma2408c552007-10-16 01:24:44 -07003510 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003511 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003512
Christoph Lameter2154a332010-07-09 14:07:10 -05003513 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003514
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003515 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003516 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003517
3518 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003519}
3520
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003521#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003522void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003523 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003524{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003525 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003526 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003527
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003528 if (unlikely(size > SLUB_MAX_SIZE)) {
3529 ret = kmalloc_large_node(size, gfpflags, node);
3530
3531 trace_kmalloc_node(caller, ret,
3532 size, PAGE_SIZE << get_order(size),
3533 gfpflags, node);
3534
3535 return ret;
3536 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003537
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003538 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003539
Satyam Sharma2408c552007-10-16 01:24:44 -07003540 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003541 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003542
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003543 ret = slab_alloc(s, gfpflags, node, caller);
3544
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003545 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003546 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003547
3548 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003549}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003550#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003551
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003552#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003553static int count_inuse(struct page *page)
3554{
3555 return page->inuse;
3556}
3557
3558static int count_total(struct page *page)
3559{
3560 return page->objects;
3561}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003562#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003563
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003564#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003565static int validate_slab(struct kmem_cache *s, struct page *page,
3566 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003567{
3568 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003569 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003570
3571 if (!check_slab(s, page) ||
3572 !on_freelist(s, page, NULL))
3573 return 0;
3574
3575 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003576 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003577
Christoph Lameter5f80b132011-04-15 14:48:13 -05003578 get_map(s, page, map);
3579 for_each_object(p, s, addr, page->objects) {
3580 if (test_bit(slab_index(p, s, addr), map))
3581 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3582 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003583 }
3584
Christoph Lameter224a88b2008-04-14 19:11:31 +03003585 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003586 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003587 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003588 return 0;
3589 return 1;
3590}
3591
Christoph Lameter434e2452007-07-17 04:03:30 -07003592static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3593 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003594{
3595 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003596 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003597 slab_unlock(page);
3598 } else
3599 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3600 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003601}
3602
Christoph Lameter434e2452007-07-17 04:03:30 -07003603static int validate_slab_node(struct kmem_cache *s,
3604 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003605{
3606 unsigned long count = 0;
3607 struct page *page;
3608 unsigned long flags;
3609
3610 spin_lock_irqsave(&n->list_lock, flags);
3611
3612 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003613 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003614 count++;
3615 }
3616 if (count != n->nr_partial)
3617 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3618 "counter=%ld\n", s->name, count, n->nr_partial);
3619
3620 if (!(s->flags & SLAB_STORE_USER))
3621 goto out;
3622
3623 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003624 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003625 count++;
3626 }
3627 if (count != atomic_long_read(&n->nr_slabs))
3628 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3629 "counter=%ld\n", s->name, count,
3630 atomic_long_read(&n->nr_slabs));
3631
3632out:
3633 spin_unlock_irqrestore(&n->list_lock, flags);
3634 return count;
3635}
3636
Christoph Lameter434e2452007-07-17 04:03:30 -07003637static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003638{
3639 int node;
3640 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003641 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003642 sizeof(unsigned long), GFP_KERNEL);
3643
3644 if (!map)
3645 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003646
3647 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003648 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003649 struct kmem_cache_node *n = get_node(s, node);
3650
Christoph Lameter434e2452007-07-17 04:03:30 -07003651 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003652 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003653 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003654 return count;
3655}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003656/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003657 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003658 * and freed.
3659 */
3660
3661struct location {
3662 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003663 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003664 long long sum_time;
3665 long min_time;
3666 long max_time;
3667 long min_pid;
3668 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303669 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003670 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003671};
3672
3673struct loc_track {
3674 unsigned long max;
3675 unsigned long count;
3676 struct location *loc;
3677};
3678
3679static void free_loc_track(struct loc_track *t)
3680{
3681 if (t->max)
3682 free_pages((unsigned long)t->loc,
3683 get_order(sizeof(struct location) * t->max));
3684}
3685
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003686static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003687{
3688 struct location *l;
3689 int order;
3690
Christoph Lameter88a420e2007-05-06 14:49:45 -07003691 order = get_order(sizeof(struct location) * max);
3692
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003693 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003694 if (!l)
3695 return 0;
3696
3697 if (t->count) {
3698 memcpy(l, t->loc, sizeof(struct location) * t->count);
3699 free_loc_track(t);
3700 }
3701 t->max = max;
3702 t->loc = l;
3703 return 1;
3704}
3705
3706static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003707 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003708{
3709 long start, end, pos;
3710 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003711 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003712 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003713
3714 start = -1;
3715 end = t->count;
3716
3717 for ( ; ; ) {
3718 pos = start + (end - start + 1) / 2;
3719
3720 /*
3721 * There is nothing at "end". If we end up there
3722 * we need to add something to before end.
3723 */
3724 if (pos == end)
3725 break;
3726
3727 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003728 if (track->addr == caddr) {
3729
3730 l = &t->loc[pos];
3731 l->count++;
3732 if (track->when) {
3733 l->sum_time += age;
3734 if (age < l->min_time)
3735 l->min_time = age;
3736 if (age > l->max_time)
3737 l->max_time = age;
3738
3739 if (track->pid < l->min_pid)
3740 l->min_pid = track->pid;
3741 if (track->pid > l->max_pid)
3742 l->max_pid = track->pid;
3743
Rusty Russell174596a2009-01-01 10:12:29 +10303744 cpumask_set_cpu(track->cpu,
3745 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003746 }
3747 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003748 return 1;
3749 }
3750
Christoph Lameter45edfa52007-05-09 02:32:45 -07003751 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003752 end = pos;
3753 else
3754 start = pos;
3755 }
3756
3757 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003758 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003759 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003760 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003761 return 0;
3762
3763 l = t->loc + pos;
3764 if (pos < t->count)
3765 memmove(l + 1, l,
3766 (t->count - pos) * sizeof(struct location));
3767 t->count++;
3768 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003769 l->addr = track->addr;
3770 l->sum_time = age;
3771 l->min_time = age;
3772 l->max_time = age;
3773 l->min_pid = track->pid;
3774 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303775 cpumask_clear(to_cpumask(l->cpus));
3776 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003777 nodes_clear(l->nodes);
3778 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003779 return 1;
3780}
3781
3782static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003783 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003784 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003785{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003786 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003787 void *p;
3788
Christoph Lameter39b26462008-04-14 19:11:30 +03003789 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05003790 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003791
Christoph Lameter224a88b2008-04-14 19:11:31 +03003792 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003793 if (!test_bit(slab_index(p, s, addr), map))
3794 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003795}
3796
3797static int list_locations(struct kmem_cache *s, char *buf,
3798 enum track_item alloc)
3799{
Harvey Harrisone374d482008-01-31 15:20:50 -08003800 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003801 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003802 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003803 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003804 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3805 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003806
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003807 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3808 GFP_TEMPORARY)) {
3809 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003810 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003811 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003812 /* Push back cpu slabs */
3813 flush_all(s);
3814
Christoph Lameterf64dc582007-10-16 01:25:33 -07003815 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003816 struct kmem_cache_node *n = get_node(s, node);
3817 unsigned long flags;
3818 struct page *page;
3819
Christoph Lameter9e869432007-08-22 14:01:56 -07003820 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003821 continue;
3822
3823 spin_lock_irqsave(&n->list_lock, flags);
3824 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003825 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003826 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003827 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003828 spin_unlock_irqrestore(&n->list_lock, flags);
3829 }
3830
3831 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003832 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003833
Hugh Dickins9c246242008-12-09 13:14:27 -08003834 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003835 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003836 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003837
3838 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08003839 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003840 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003841 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003842
3843 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003844 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003845 l->min_time,
3846 (long)div_u64(l->sum_time, l->count),
3847 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003848 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003849 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003850 l->min_time);
3851
3852 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003853 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003854 l->min_pid, l->max_pid);
3855 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003856 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003857 l->min_pid);
3858
Rusty Russell174596a2009-01-01 10:12:29 +10303859 if (num_online_cpus() > 1 &&
3860 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003861 len < PAGE_SIZE - 60) {
3862 len += sprintf(buf + len, " cpus=");
3863 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303864 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003865 }
3866
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003867 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003868 len < PAGE_SIZE - 60) {
3869 len += sprintf(buf + len, " nodes=");
3870 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003871 l->nodes);
3872 }
3873
Harvey Harrisone374d482008-01-31 15:20:50 -08003874 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003875 }
3876
3877 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003878 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003879 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003880 len += sprintf(buf, "No data\n");
3881 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003882}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003883#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07003884
Christoph Lametera5a84752010-10-05 13:57:27 -05003885#ifdef SLUB_RESILIENCY_TEST
3886static void resiliency_test(void)
3887{
3888 u8 *p;
3889
3890 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3891
3892 printk(KERN_ERR "SLUB resiliency testing\n");
3893 printk(KERN_ERR "-----------------------\n");
3894 printk(KERN_ERR "A. Corruption after allocation\n");
3895
3896 p = kzalloc(16, GFP_KERNEL);
3897 p[16] = 0x12;
3898 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3899 " 0x12->0x%p\n\n", p + 16);
3900
3901 validate_slab_cache(kmalloc_caches[4]);
3902
3903 /* Hmmm... The next two are dangerous */
3904 p = kzalloc(32, GFP_KERNEL);
3905 p[32 + sizeof(void *)] = 0x34;
3906 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3907 " 0x34 -> -0x%p\n", p);
3908 printk(KERN_ERR
3909 "If allocated object is overwritten then not detectable\n\n");
3910
3911 validate_slab_cache(kmalloc_caches[5]);
3912 p = kzalloc(64, GFP_KERNEL);
3913 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3914 *p = 0x56;
3915 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3916 p);
3917 printk(KERN_ERR
3918 "If allocated object is overwritten then not detectable\n\n");
3919 validate_slab_cache(kmalloc_caches[6]);
3920
3921 printk(KERN_ERR "\nB. Corruption after free\n");
3922 p = kzalloc(128, GFP_KERNEL);
3923 kfree(p);
3924 *p = 0x78;
3925 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3926 validate_slab_cache(kmalloc_caches[7]);
3927
3928 p = kzalloc(256, GFP_KERNEL);
3929 kfree(p);
3930 p[50] = 0x9a;
3931 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3932 p);
3933 validate_slab_cache(kmalloc_caches[8]);
3934
3935 p = kzalloc(512, GFP_KERNEL);
3936 kfree(p);
3937 p[512] = 0xab;
3938 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3939 validate_slab_cache(kmalloc_caches[9]);
3940}
3941#else
3942#ifdef CONFIG_SYSFS
3943static void resiliency_test(void) {};
3944#endif
3945#endif
3946
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003947#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07003948enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003949 SL_ALL, /* All slabs */
3950 SL_PARTIAL, /* Only partially allocated slabs */
3951 SL_CPU, /* Only slabs used for cpu caches */
3952 SL_OBJECTS, /* Determine allocated objects not slabs */
3953 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003954};
3955
Christoph Lameter205ab992008-04-14 19:11:40 +03003956#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003957#define SO_PARTIAL (1 << SL_PARTIAL)
3958#define SO_CPU (1 << SL_CPU)
3959#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003960#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003961
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003962static ssize_t show_slab_objects(struct kmem_cache *s,
3963 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003964{
3965 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003966 int node;
3967 int x;
3968 unsigned long *nodes;
3969 unsigned long *per_cpu;
3970
3971 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003972 if (!nodes)
3973 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003974 per_cpu = nodes + nr_node_ids;
3975
Christoph Lameter205ab992008-04-14 19:11:40 +03003976 if (flags & SO_CPU) {
3977 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003978
Christoph Lameter205ab992008-04-14 19:11:40 +03003979 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003980 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003981
Christoph Lameter205ab992008-04-14 19:11:40 +03003982 if (!c || c->node < 0)
3983 continue;
3984
3985 if (c->page) {
3986 if (flags & SO_TOTAL)
3987 x = c->page->objects;
3988 else if (flags & SO_OBJECTS)
3989 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003990 else
3991 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003992
Christoph Lameter81819f02007-05-06 14:49:36 -07003993 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003994 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003995 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003996 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003997 }
3998 }
3999
Christoph Lameter04d94872011-01-10 10:15:15 -06004000 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004001#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004002 if (flags & SO_ALL) {
4003 for_each_node_state(node, N_NORMAL_MEMORY) {
4004 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004005
Christoph Lameter205ab992008-04-14 19:11:40 +03004006 if (flags & SO_TOTAL)
4007 x = atomic_long_read(&n->total_objects);
4008 else if (flags & SO_OBJECTS)
4009 x = atomic_long_read(&n->total_objects) -
4010 count_partial(n, count_free);
4011
4012 else
4013 x = atomic_long_read(&n->nr_slabs);
4014 total += x;
4015 nodes[node] += x;
4016 }
4017
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004018 } else
4019#endif
4020 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004021 for_each_node_state(node, N_NORMAL_MEMORY) {
4022 struct kmem_cache_node *n = get_node(s, node);
4023
4024 if (flags & SO_TOTAL)
4025 x = count_partial(n, count_total);
4026 else if (flags & SO_OBJECTS)
4027 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004028 else
4029 x = n->nr_partial;
4030 total += x;
4031 nodes[node] += x;
4032 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004033 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004034 x = sprintf(buf, "%lu", total);
4035#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004036 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004037 if (nodes[node])
4038 x += sprintf(buf + x, " N%d=%lu",
4039 node, nodes[node]);
4040#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004041 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004042 kfree(nodes);
4043 return x + sprintf(buf + x, "\n");
4044}
4045
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004046#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004047static int any_slab_objects(struct kmem_cache *s)
4048{
4049 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004050
4051 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004052 struct kmem_cache_node *n = get_node(s, node);
4053
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004054 if (!n)
4055 continue;
4056
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004057 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004058 return 1;
4059 }
4060 return 0;
4061}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004062#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004063
4064#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4065#define to_slab(n) container_of(n, struct kmem_cache, kobj);
4066
4067struct slab_attribute {
4068 struct attribute attr;
4069 ssize_t (*show)(struct kmem_cache *s, char *buf);
4070 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4071};
4072
4073#define SLAB_ATTR_RO(_name) \
4074 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4075
4076#define SLAB_ATTR(_name) \
4077 static struct slab_attribute _name##_attr = \
4078 __ATTR(_name, 0644, _name##_show, _name##_store)
4079
Christoph Lameter81819f02007-05-06 14:49:36 -07004080static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4081{
4082 return sprintf(buf, "%d\n", s->size);
4083}
4084SLAB_ATTR_RO(slab_size);
4085
4086static ssize_t align_show(struct kmem_cache *s, char *buf)
4087{
4088 return sprintf(buf, "%d\n", s->align);
4089}
4090SLAB_ATTR_RO(align);
4091
4092static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4093{
4094 return sprintf(buf, "%d\n", s->objsize);
4095}
4096SLAB_ATTR_RO(object_size);
4097
4098static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4099{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004100 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004101}
4102SLAB_ATTR_RO(objs_per_slab);
4103
Christoph Lameter06b285d2008-04-14 19:11:41 +03004104static ssize_t order_store(struct kmem_cache *s,
4105 const char *buf, size_t length)
4106{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004107 unsigned long order;
4108 int err;
4109
4110 err = strict_strtoul(buf, 10, &order);
4111 if (err)
4112 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004113
4114 if (order > slub_max_order || order < slub_min_order)
4115 return -EINVAL;
4116
4117 calculate_sizes(s, order);
4118 return length;
4119}
4120
Christoph Lameter81819f02007-05-06 14:49:36 -07004121static ssize_t order_show(struct kmem_cache *s, char *buf)
4122{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004123 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004124}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004125SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004126
David Rientjes73d342b2009-02-22 17:40:09 -08004127static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4128{
4129 return sprintf(buf, "%lu\n", s->min_partial);
4130}
4131
4132static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4133 size_t length)
4134{
4135 unsigned long min;
4136 int err;
4137
4138 err = strict_strtoul(buf, 10, &min);
4139 if (err)
4140 return err;
4141
David Rientjesc0bdb232009-02-25 09:16:35 +02004142 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004143 return length;
4144}
4145SLAB_ATTR(min_partial);
4146
Christoph Lameter81819f02007-05-06 14:49:36 -07004147static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4148{
Joe Perches62c70bc2011-01-13 15:45:52 -08004149 if (!s->ctor)
4150 return 0;
4151 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004152}
4153SLAB_ATTR_RO(ctor);
4154
Christoph Lameter81819f02007-05-06 14:49:36 -07004155static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4156{
4157 return sprintf(buf, "%d\n", s->refcount - 1);
4158}
4159SLAB_ATTR_RO(aliases);
4160
Christoph Lameter81819f02007-05-06 14:49:36 -07004161static ssize_t partial_show(struct kmem_cache *s, char *buf)
4162{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004163 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004164}
4165SLAB_ATTR_RO(partial);
4166
4167static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4168{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004169 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004170}
4171SLAB_ATTR_RO(cpu_slabs);
4172
4173static ssize_t objects_show(struct kmem_cache *s, char *buf)
4174{
Christoph Lameter205ab992008-04-14 19:11:40 +03004175 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004176}
4177SLAB_ATTR_RO(objects);
4178
Christoph Lameter205ab992008-04-14 19:11:40 +03004179static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4180{
4181 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4182}
4183SLAB_ATTR_RO(objects_partial);
4184
Christoph Lameter81819f02007-05-06 14:49:36 -07004185static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4186{
4187 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4188}
4189
4190static ssize_t reclaim_account_store(struct kmem_cache *s,
4191 const char *buf, size_t length)
4192{
4193 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4194 if (buf[0] == '1')
4195 s->flags |= SLAB_RECLAIM_ACCOUNT;
4196 return length;
4197}
4198SLAB_ATTR(reclaim_account);
4199
4200static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4201{
Christoph Lameter5af60832007-05-06 14:49:56 -07004202 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004203}
4204SLAB_ATTR_RO(hwcache_align);
4205
4206#ifdef CONFIG_ZONE_DMA
4207static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4208{
4209 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4210}
4211SLAB_ATTR_RO(cache_dma);
4212#endif
4213
4214static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4215{
4216 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4217}
4218SLAB_ATTR_RO(destroy_by_rcu);
4219
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004220static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4221{
4222 return sprintf(buf, "%d\n", s->reserved);
4223}
4224SLAB_ATTR_RO(reserved);
4225
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004226#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004227static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4228{
4229 return show_slab_objects(s, buf, SO_ALL);
4230}
4231SLAB_ATTR_RO(slabs);
4232
4233static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4234{
4235 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4236}
4237SLAB_ATTR_RO(total_objects);
4238
4239static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4240{
4241 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4242}
4243
4244static ssize_t sanity_checks_store(struct kmem_cache *s,
4245 const char *buf, size_t length)
4246{
4247 s->flags &= ~SLAB_DEBUG_FREE;
4248 if (buf[0] == '1')
4249 s->flags |= SLAB_DEBUG_FREE;
4250 return length;
4251}
4252SLAB_ATTR(sanity_checks);
4253
4254static ssize_t trace_show(struct kmem_cache *s, char *buf)
4255{
4256 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4257}
4258
4259static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4260 size_t length)
4261{
4262 s->flags &= ~SLAB_TRACE;
4263 if (buf[0] == '1')
4264 s->flags |= SLAB_TRACE;
4265 return length;
4266}
4267SLAB_ATTR(trace);
4268
Christoph Lameter81819f02007-05-06 14:49:36 -07004269static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4270{
4271 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4272}
4273
4274static ssize_t red_zone_store(struct kmem_cache *s,
4275 const char *buf, size_t length)
4276{
4277 if (any_slab_objects(s))
4278 return -EBUSY;
4279
4280 s->flags &= ~SLAB_RED_ZONE;
4281 if (buf[0] == '1')
4282 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004283 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004284 return length;
4285}
4286SLAB_ATTR(red_zone);
4287
4288static ssize_t poison_show(struct kmem_cache *s, char *buf)
4289{
4290 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4291}
4292
4293static ssize_t poison_store(struct kmem_cache *s,
4294 const char *buf, size_t length)
4295{
4296 if (any_slab_objects(s))
4297 return -EBUSY;
4298
4299 s->flags &= ~SLAB_POISON;
4300 if (buf[0] == '1')
4301 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004302 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004303 return length;
4304}
4305SLAB_ATTR(poison);
4306
4307static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4308{
4309 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4310}
4311
4312static ssize_t store_user_store(struct kmem_cache *s,
4313 const char *buf, size_t length)
4314{
4315 if (any_slab_objects(s))
4316 return -EBUSY;
4317
4318 s->flags &= ~SLAB_STORE_USER;
4319 if (buf[0] == '1')
4320 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004321 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004322 return length;
4323}
4324SLAB_ATTR(store_user);
4325
Christoph Lameter53e15af2007-05-06 14:49:43 -07004326static ssize_t validate_show(struct kmem_cache *s, char *buf)
4327{
4328 return 0;
4329}
4330
4331static ssize_t validate_store(struct kmem_cache *s,
4332 const char *buf, size_t length)
4333{
Christoph Lameter434e2452007-07-17 04:03:30 -07004334 int ret = -EINVAL;
4335
4336 if (buf[0] == '1') {
4337 ret = validate_slab_cache(s);
4338 if (ret >= 0)
4339 ret = length;
4340 }
4341 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004342}
4343SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004344
4345static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4346{
4347 if (!(s->flags & SLAB_STORE_USER))
4348 return -ENOSYS;
4349 return list_locations(s, buf, TRACK_ALLOC);
4350}
4351SLAB_ATTR_RO(alloc_calls);
4352
4353static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4354{
4355 if (!(s->flags & SLAB_STORE_USER))
4356 return -ENOSYS;
4357 return list_locations(s, buf, TRACK_FREE);
4358}
4359SLAB_ATTR_RO(free_calls);
4360#endif /* CONFIG_SLUB_DEBUG */
4361
4362#ifdef CONFIG_FAILSLAB
4363static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4364{
4365 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4366}
4367
4368static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4369 size_t length)
4370{
4371 s->flags &= ~SLAB_FAILSLAB;
4372 if (buf[0] == '1')
4373 s->flags |= SLAB_FAILSLAB;
4374 return length;
4375}
4376SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004377#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004378
Christoph Lameter2086d262007-05-06 14:49:46 -07004379static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4380{
4381 return 0;
4382}
4383
4384static ssize_t shrink_store(struct kmem_cache *s,
4385 const char *buf, size_t length)
4386{
4387 if (buf[0] == '1') {
4388 int rc = kmem_cache_shrink(s);
4389
4390 if (rc)
4391 return rc;
4392 } else
4393 return -EINVAL;
4394 return length;
4395}
4396SLAB_ATTR(shrink);
4397
Christoph Lameter81819f02007-05-06 14:49:36 -07004398#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004399static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004400{
Christoph Lameter98246012008-01-07 23:20:26 -08004401 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004402}
4403
Christoph Lameter98246012008-01-07 23:20:26 -08004404static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004405 const char *buf, size_t length)
4406{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004407 unsigned long ratio;
4408 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004409
Christoph Lameter0121c6192008-04-29 16:11:12 -07004410 err = strict_strtoul(buf, 10, &ratio);
4411 if (err)
4412 return err;
4413
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004414 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004415 s->remote_node_defrag_ratio = ratio * 10;
4416
Christoph Lameter81819f02007-05-06 14:49:36 -07004417 return length;
4418}
Christoph Lameter98246012008-01-07 23:20:26 -08004419SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004420#endif
4421
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004422#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004423static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4424{
4425 unsigned long sum = 0;
4426 int cpu;
4427 int len;
4428 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4429
4430 if (!data)
4431 return -ENOMEM;
4432
4433 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004434 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004435
4436 data[cpu] = x;
4437 sum += x;
4438 }
4439
4440 len = sprintf(buf, "%lu", sum);
4441
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004442#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004443 for_each_online_cpu(cpu) {
4444 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004445 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004446 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004447#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004448 kfree(data);
4449 return len + sprintf(buf + len, "\n");
4450}
4451
David Rientjes78eb00c2009-10-15 02:20:22 -07004452static void clear_stat(struct kmem_cache *s, enum stat_item si)
4453{
4454 int cpu;
4455
4456 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004457 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004458}
4459
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004460#define STAT_ATTR(si, text) \
4461static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4462{ \
4463 return show_stat(s, buf, si); \
4464} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004465static ssize_t text##_store(struct kmem_cache *s, \
4466 const char *buf, size_t length) \
4467{ \
4468 if (buf[0] != '0') \
4469 return -EINVAL; \
4470 clear_stat(s, si); \
4471 return length; \
4472} \
4473SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004474
4475STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4476STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4477STAT_ATTR(FREE_FASTPATH, free_fastpath);
4478STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4479STAT_ATTR(FREE_FROZEN, free_frozen);
4480STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4481STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4482STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4483STAT_ATTR(ALLOC_SLAB, alloc_slab);
4484STAT_ATTR(ALLOC_REFILL, alloc_refill);
4485STAT_ATTR(FREE_SLAB, free_slab);
4486STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4487STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4488STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4489STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4490STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4491STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004492STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004493#endif
4494
Pekka Enberg06428782008-01-07 23:20:27 -08004495static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004496 &slab_size_attr.attr,
4497 &object_size_attr.attr,
4498 &objs_per_slab_attr.attr,
4499 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004500 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004501 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004502 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004503 &partial_attr.attr,
4504 &cpu_slabs_attr.attr,
4505 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004506 &aliases_attr.attr,
4507 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004508 &hwcache_align_attr.attr,
4509 &reclaim_account_attr.attr,
4510 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004511 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004512 &reserved_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004513#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004514 &total_objects_attr.attr,
4515 &slabs_attr.attr,
4516 &sanity_checks_attr.attr,
4517 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004518 &red_zone_attr.attr,
4519 &poison_attr.attr,
4520 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004521 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004522 &alloc_calls_attr.attr,
4523 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004524#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004525#ifdef CONFIG_ZONE_DMA
4526 &cache_dma_attr.attr,
4527#endif
4528#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004529 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004530#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004531#ifdef CONFIG_SLUB_STATS
4532 &alloc_fastpath_attr.attr,
4533 &alloc_slowpath_attr.attr,
4534 &free_fastpath_attr.attr,
4535 &free_slowpath_attr.attr,
4536 &free_frozen_attr.attr,
4537 &free_add_partial_attr.attr,
4538 &free_remove_partial_attr.attr,
4539 &alloc_from_partial_attr.attr,
4540 &alloc_slab_attr.attr,
4541 &alloc_refill_attr.attr,
4542 &free_slab_attr.attr,
4543 &cpuslab_flush_attr.attr,
4544 &deactivate_full_attr.attr,
4545 &deactivate_empty_attr.attr,
4546 &deactivate_to_head_attr.attr,
4547 &deactivate_to_tail_attr.attr,
4548 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004549 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004550#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004551#ifdef CONFIG_FAILSLAB
4552 &failslab_attr.attr,
4553#endif
4554
Christoph Lameter81819f02007-05-06 14:49:36 -07004555 NULL
4556};
4557
4558static struct attribute_group slab_attr_group = {
4559 .attrs = slab_attrs,
4560};
4561
4562static ssize_t slab_attr_show(struct kobject *kobj,
4563 struct attribute *attr,
4564 char *buf)
4565{
4566 struct slab_attribute *attribute;
4567 struct kmem_cache *s;
4568 int err;
4569
4570 attribute = to_slab_attr(attr);
4571 s = to_slab(kobj);
4572
4573 if (!attribute->show)
4574 return -EIO;
4575
4576 err = attribute->show(s, buf);
4577
4578 return err;
4579}
4580
4581static ssize_t slab_attr_store(struct kobject *kobj,
4582 struct attribute *attr,
4583 const char *buf, size_t len)
4584{
4585 struct slab_attribute *attribute;
4586 struct kmem_cache *s;
4587 int err;
4588
4589 attribute = to_slab_attr(attr);
4590 s = to_slab(kobj);
4591
4592 if (!attribute->store)
4593 return -EIO;
4594
4595 err = attribute->store(s, buf, len);
4596
4597 return err;
4598}
4599
Christoph Lameter151c6022008-01-07 22:29:05 -08004600static void kmem_cache_release(struct kobject *kobj)
4601{
4602 struct kmem_cache *s = to_slab(kobj);
4603
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004604 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004605 kfree(s);
4606}
4607
Emese Revfy52cf25d2010-01-19 02:58:23 +01004608static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004609 .show = slab_attr_show,
4610 .store = slab_attr_store,
4611};
4612
4613static struct kobj_type slab_ktype = {
4614 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004615 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004616};
4617
4618static int uevent_filter(struct kset *kset, struct kobject *kobj)
4619{
4620 struct kobj_type *ktype = get_ktype(kobj);
4621
4622 if (ktype == &slab_ktype)
4623 return 1;
4624 return 0;
4625}
4626
Emese Revfy9cd43612009-12-31 14:52:51 +01004627static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004628 .filter = uevent_filter,
4629};
4630
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004631static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004632
4633#define ID_STR_LENGTH 64
4634
4635/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004636 *
4637 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004638 */
4639static char *create_unique_id(struct kmem_cache *s)
4640{
4641 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4642 char *p = name;
4643
4644 BUG_ON(!name);
4645
4646 *p++ = ':';
4647 /*
4648 * First flags affecting slabcache operations. We will only
4649 * get here for aliasable slabs so we do not need to support
4650 * too many flags. The flags here must cover all flags that
4651 * are matched during merging to guarantee that the id is
4652 * unique.
4653 */
4654 if (s->flags & SLAB_CACHE_DMA)
4655 *p++ = 'd';
4656 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4657 *p++ = 'a';
4658 if (s->flags & SLAB_DEBUG_FREE)
4659 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004660 if (!(s->flags & SLAB_NOTRACK))
4661 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004662 if (p != name + 1)
4663 *p++ = '-';
4664 p += sprintf(p, "%07d", s->size);
4665 BUG_ON(p > name + ID_STR_LENGTH - 1);
4666 return name;
4667}
4668
4669static int sysfs_slab_add(struct kmem_cache *s)
4670{
4671 int err;
4672 const char *name;
4673 int unmergeable;
4674
4675 if (slab_state < SYSFS)
4676 /* Defer until later */
4677 return 0;
4678
4679 unmergeable = slab_unmergeable(s);
4680 if (unmergeable) {
4681 /*
4682 * Slabcache can never be merged so we can use the name proper.
4683 * This is typically the case for debug situations. In that
4684 * case we can catch duplicate names easily.
4685 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004686 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004687 name = s->name;
4688 } else {
4689 /*
4690 * Create a unique name for the slab as a target
4691 * for the symlinks.
4692 */
4693 name = create_unique_id(s);
4694 }
4695
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004696 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004697 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4698 if (err) {
4699 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004700 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004701 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004702
4703 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004704 if (err) {
4705 kobject_del(&s->kobj);
4706 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004707 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004708 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004709 kobject_uevent(&s->kobj, KOBJ_ADD);
4710 if (!unmergeable) {
4711 /* Setup first alias */
4712 sysfs_slab_alias(s, s->name);
4713 kfree(name);
4714 }
4715 return 0;
4716}
4717
4718static void sysfs_slab_remove(struct kmem_cache *s)
4719{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004720 if (slab_state < SYSFS)
4721 /*
4722 * Sysfs has not been setup yet so no need to remove the
4723 * cache from sysfs.
4724 */
4725 return;
4726
Christoph Lameter81819f02007-05-06 14:49:36 -07004727 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4728 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004729 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004730}
4731
4732/*
4733 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004734 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004735 */
4736struct saved_alias {
4737 struct kmem_cache *s;
4738 const char *name;
4739 struct saved_alias *next;
4740};
4741
Adrian Bunk5af328a2007-07-17 04:03:27 -07004742static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004743
4744static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4745{
4746 struct saved_alias *al;
4747
4748 if (slab_state == SYSFS) {
4749 /*
4750 * If we have a leftover link then remove it.
4751 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004752 sysfs_remove_link(&slab_kset->kobj, name);
4753 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004754 }
4755
4756 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4757 if (!al)
4758 return -ENOMEM;
4759
4760 al->s = s;
4761 al->name = name;
4762 al->next = alias_list;
4763 alias_list = al;
4764 return 0;
4765}
4766
4767static int __init slab_sysfs_init(void)
4768{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004769 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004770 int err;
4771
Christoph Lameter2bce6482010-07-19 11:39:11 -05004772 down_write(&slub_lock);
4773
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004774 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004775 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004776 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004777 printk(KERN_ERR "Cannot register slab subsystem.\n");
4778 return -ENOSYS;
4779 }
4780
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004781 slab_state = SYSFS;
4782
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004783 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004784 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004785 if (err)
4786 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4787 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004788 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004789
4790 while (alias_list) {
4791 struct saved_alias *al = alias_list;
4792
4793 alias_list = alias_list->next;
4794 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004795 if (err)
4796 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4797 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004798 kfree(al);
4799 }
4800
Christoph Lameter2bce6482010-07-19 11:39:11 -05004801 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004802 resiliency_test();
4803 return 0;
4804}
4805
4806__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004807#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004808
4809/*
4810 * The /proc/slabinfo ABI
4811 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004812#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004813static void print_slabinfo_header(struct seq_file *m)
4814{
4815 seq_puts(m, "slabinfo - version: 2.1\n");
4816 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4817 "<objperslab> <pagesperslab>");
4818 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4819 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4820 seq_putc(m, '\n');
4821}
4822
4823static void *s_start(struct seq_file *m, loff_t *pos)
4824{
4825 loff_t n = *pos;
4826
4827 down_read(&slub_lock);
4828 if (!n)
4829 print_slabinfo_header(m);
4830
4831 return seq_list_start(&slab_caches, *pos);
4832}
4833
4834static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4835{
4836 return seq_list_next(p, &slab_caches, pos);
4837}
4838
4839static void s_stop(struct seq_file *m, void *p)
4840{
4841 up_read(&slub_lock);
4842}
4843
4844static int s_show(struct seq_file *m, void *p)
4845{
4846 unsigned long nr_partials = 0;
4847 unsigned long nr_slabs = 0;
4848 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004849 unsigned long nr_objs = 0;
4850 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004851 struct kmem_cache *s;
4852 int node;
4853
4854 s = list_entry(p, struct kmem_cache, list);
4855
4856 for_each_online_node(node) {
4857 struct kmem_cache_node *n = get_node(s, node);
4858
4859 if (!n)
4860 continue;
4861
4862 nr_partials += n->nr_partial;
4863 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004864 nr_objs += atomic_long_read(&n->total_objects);
4865 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004866 }
4867
Christoph Lameter205ab992008-04-14 19:11:40 +03004868 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004869
4870 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004871 nr_objs, s->size, oo_objects(s->oo),
4872 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004873 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4874 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4875 0UL);
4876 seq_putc(m, '\n');
4877 return 0;
4878}
4879
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004880static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004881 .start = s_start,
4882 .next = s_next,
4883 .stop = s_stop,
4884 .show = s_show,
4885};
4886
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004887static int slabinfo_open(struct inode *inode, struct file *file)
4888{
4889 return seq_open(file, &slabinfo_op);
4890}
4891
4892static const struct file_operations proc_slabinfo_operations = {
4893 .open = slabinfo_open,
4894 .read = seq_read,
4895 .llseek = seq_lseek,
4896 .release = seq_release,
4897};
4898
4899static int __init slab_proc_init(void)
4900{
WANG Congcf5d1132009-08-18 19:11:40 +03004901 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004902 return 0;
4903}
4904module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004905#endif /* CONFIG_SLABINFO */