blob: be6715dd0ee8e1e34d16d9d75ab4f7e7f9e71fa0 [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 Lameterb789ef52011-06-01 12:25:49 -0500134/* Enable to log cmpxchg failures */
135#undef SLUB_DEBUG_CMPXCHG
136
Christoph Lameter81819f02007-05-06 14:49:36 -0700137/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700138 * Mininum number of partial slabs. These will be left on the partial
139 * lists even if they are empty. kmem_cache_shrink may reclaim them.
140 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800141#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700142
Christoph Lameter2086d262007-05-06 14:49:46 -0700143/*
144 * Maximum number of desirable partial slabs.
145 * The existence of more partial slabs makes kmem_cache_shrink
146 * sort the partial list by the number of objects in the.
147 */
148#define MAX_PARTIAL 10
149
Christoph Lameter81819f02007-05-06 14:49:36 -0700150#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
151 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700152
Christoph Lameter81819f02007-05-06 14:49:36 -0700153/*
David Rientjes3de47212009-07-27 18:30:35 -0700154 * Debugging flags that require metadata to be stored in the slab. These get
155 * disabled when slub_debug=O is used and a cache's min order increases with
156 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700157 */
David Rientjes3de47212009-07-27 18:30:35 -0700158#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700159
160/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700161 * Set of flags that will prevent slab merging
162 */
163#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300164 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
165 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166
167#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200168 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700169
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400170#define OO_SHIFT 16
171#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500172#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400173
Christoph Lameter81819f02007-05-06 14:49:36 -0700174/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500175#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500176#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700177
178static int kmem_size = sizeof(struct kmem_cache);
179
180#ifdef CONFIG_SMP
181static struct notifier_block slab_notifier;
182#endif
183
184static enum {
185 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500186 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700187 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700188 SYSFS /* Sysfs up */
189} slab_state = DOWN;
190
191/* A list of all slab caches on the system */
192static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700193static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700194
Christoph Lameter02cbc872007-05-09 02:32:43 -0700195/*
196 * Tracking user of a slab.
197 */
198struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300199 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700200 int cpu; /* Was running on cpu */
201 int pid; /* Pid context */
202 unsigned long when; /* When did the operation occur */
203};
204
205enum track_item { TRACK_ALLOC, TRACK_FREE };
206
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500207#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700208static int sysfs_slab_add(struct kmem_cache *);
209static int sysfs_slab_alias(struct kmem_cache *, const char *);
210static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800211
Christoph Lameter81819f02007-05-06 14:49:36 -0700212#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700213static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
214static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
215 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800216static inline void sysfs_slab_remove(struct kmem_cache *s)
217{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300218 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800219 kfree(s);
220}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800221
Christoph Lameter81819f02007-05-06 14:49:36 -0700222#endif
223
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500224static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800225{
226#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e62009-12-18 16:26:23 -0600227 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800228#endif
229}
230
Christoph Lameter81819f02007-05-06 14:49:36 -0700231/********************************************************************
232 * Core slab cache functions
233 *******************************************************************/
234
235int slab_is_available(void)
236{
237 return slab_state >= UP;
238}
239
240static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
241{
Christoph Lameter81819f02007-05-06 14:49:36 -0700242 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700243}
244
Christoph Lameter6446faa2008-02-15 23:45:26 -0800245/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700246static inline int check_valid_pointer(struct kmem_cache *s,
247 struct page *page, const void *object)
248{
249 void *base;
250
Christoph Lametera973e9d2008-03-01 13:40:44 -0800251 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700252 return 1;
253
Christoph Lametera973e9d2008-03-01 13:40:44 -0800254 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300255 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700256 (object - base) % s->size) {
257 return 0;
258 }
259
260 return 1;
261}
262
Christoph Lameter7656c722007-05-09 02:32:40 -0700263static inline void *get_freepointer(struct kmem_cache *s, void *object)
264{
265 return *(void **)(object + s->offset);
266}
267
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500268static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
269{
270 void *p;
271
272#ifdef CONFIG_DEBUG_PAGEALLOC
273 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
274#else
275 p = get_freepointer(s, object);
276#endif
277 return p;
278}
279
Christoph Lameter7656c722007-05-09 02:32:40 -0700280static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
281{
282 *(void **)(object + s->offset) = fp;
283}
284
285/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300286#define for_each_object(__p, __s, __addr, __objects) \
287 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700288 __p += (__s)->size)
289
Christoph Lameter7656c722007-05-09 02:32:40 -0700290/* Determine object index from a given position */
291static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
292{
293 return (p - addr) / s->size;
294}
295
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100296static inline size_t slab_ksize(const struct kmem_cache *s)
297{
298#ifdef CONFIG_SLUB_DEBUG
299 /*
300 * Debugging requires use of the padding between object
301 * and whatever may come after it.
302 */
303 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
304 return s->objsize;
305
306#endif
307 /*
308 * If we have the need to store the freelist pointer
309 * back there or track user information then we can
310 * only use the space before that information.
311 */
312 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
313 return s->inuse;
314 /*
315 * Else we can use all the padding etc for the allocation
316 */
317 return s->size;
318}
319
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800320static inline int order_objects(int order, unsigned long size, int reserved)
321{
322 return ((PAGE_SIZE << order) - reserved) / size;
323}
324
Christoph Lameter834f3d12008-04-14 19:11:31 +0300325static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800326 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300327{
328 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800329 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300330 };
331
332 return x;
333}
334
335static inline int oo_order(struct kmem_cache_order_objects x)
336{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400337 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300338}
339
340static inline int oo_objects(struct kmem_cache_order_objects x)
341{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400342 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300343}
344
Christoph Lameterb789ef52011-06-01 12:25:49 -0500345static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
346 void *freelist_old, unsigned long counters_old,
347 void *freelist_new, unsigned long counters_new,
348 const char *n)
349{
350#ifdef CONFIG_CMPXCHG_DOUBLE
351 if (s->flags & __CMPXCHG_DOUBLE) {
352 if (cmpxchg_double(&page->freelist,
353 freelist_old, counters_old,
354 freelist_new, counters_new))
355 return 1;
356 } else
357#endif
358 {
359 if (page->freelist == freelist_old && page->counters == counters_old) {
360 page->freelist = freelist_new;
361 page->counters = counters_new;
362 return 1;
363 }
364 }
365
366 cpu_relax();
367 stat(s, CMPXCHG_DOUBLE_FAIL);
368
369#ifdef SLUB_DEBUG_CMPXCHG
370 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
371#endif
372
373 return 0;
374}
375
Christoph Lameter41ecc552007-05-09 02:32:44 -0700376#ifdef CONFIG_SLUB_DEBUG
377/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500378 * Determine a map of object in use on a page.
379 *
380 * Slab lock or node listlock must be held to guarantee that the page does
381 * not vanish from under us.
382 */
383static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
384{
385 void *p;
386 void *addr = page_address(page);
387
388 for (p = page->freelist; p; p = get_freepointer(s, p))
389 set_bit(slab_index(p, s, addr), map);
390}
391
Christoph Lameter41ecc552007-05-09 02:32:44 -0700392/*
393 * Debug settings:
394 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700395#ifdef CONFIG_SLUB_DEBUG_ON
396static int slub_debug = DEBUG_DEFAULT_FLAGS;
397#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700398static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700399#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700400
401static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700402static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700403
Christoph Lameter7656c722007-05-09 02:32:40 -0700404/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700405 * Object debugging
406 */
407static void print_section(char *text, u8 *addr, unsigned int length)
408{
409 int i, offset;
410 int newline = 1;
411 char ascii[17];
412
413 ascii[16] = 0;
414
415 for (i = 0; i < length; i++) {
416 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700417 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700418 newline = 0;
419 }
Pekka Enberg06428782008-01-07 23:20:27 -0800420 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700421 offset = i % 16;
422 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
423 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800424 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700425 newline = 1;
426 }
427 }
428 if (!newline) {
429 i %= 16;
430 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800431 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700432 ascii[i] = ' ';
433 i++;
434 }
Pekka Enberg06428782008-01-07 23:20:27 -0800435 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700436 }
437}
438
Christoph Lameter81819f02007-05-06 14:49:36 -0700439static struct track *get_track(struct kmem_cache *s, void *object,
440 enum track_item alloc)
441{
442 struct track *p;
443
444 if (s->offset)
445 p = object + s->offset + sizeof(void *);
446 else
447 p = object + s->inuse;
448
449 return p + alloc;
450}
451
452static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300453 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700454{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900455 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700456
Christoph Lameter81819f02007-05-06 14:49:36 -0700457 if (addr) {
458 p->addr = addr;
459 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400460 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700461 p->when = jiffies;
462 } else
463 memset(p, 0, sizeof(struct track));
464}
465
Christoph Lameter81819f02007-05-06 14:49:36 -0700466static void init_tracking(struct kmem_cache *s, void *object)
467{
Christoph Lameter24922682007-07-17 04:03:18 -0700468 if (!(s->flags & SLAB_STORE_USER))
469 return;
470
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300471 set_track(s, object, TRACK_FREE, 0UL);
472 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700473}
474
475static void print_track(const char *s, struct track *t)
476{
477 if (!t->addr)
478 return;
479
Linus Torvalds7daf7052008-07-14 12:12:53 -0700480 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300481 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700482}
483
Christoph Lameter24922682007-07-17 04:03:18 -0700484static void print_tracking(struct kmem_cache *s, void *object)
485{
486 if (!(s->flags & SLAB_STORE_USER))
487 return;
488
489 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
490 print_track("Freed", get_track(s, object, TRACK_FREE));
491}
492
493static void print_page_info(struct page *page)
494{
Christoph Lameter39b26462008-04-14 19:11:30 +0300495 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
496 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700497
498}
499
500static void slab_bug(struct kmem_cache *s, char *fmt, ...)
501{
502 va_list args;
503 char buf[100];
504
505 va_start(args, fmt);
506 vsnprintf(buf, sizeof(buf), fmt, args);
507 va_end(args);
508 printk(KERN_ERR "========================================"
509 "=====================================\n");
510 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
511 printk(KERN_ERR "----------------------------------------"
512 "-------------------------------------\n\n");
513}
514
515static void slab_fix(struct kmem_cache *s, char *fmt, ...)
516{
517 va_list args;
518 char buf[100];
519
520 va_start(args, fmt);
521 vsnprintf(buf, sizeof(buf), fmt, args);
522 va_end(args);
523 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
524}
525
526static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700527{
528 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800529 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700530
531 print_tracking(s, p);
532
533 print_page_info(page);
534
535 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
536 p, p - addr, get_freepointer(s, p));
537
538 if (p > addr + 16)
539 print_section("Bytes b4", p - 16, 16);
540
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300541 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700542
543 if (s->flags & SLAB_RED_ZONE)
544 print_section("Redzone", p + s->objsize,
545 s->inuse - s->objsize);
546
Christoph Lameter81819f02007-05-06 14:49:36 -0700547 if (s->offset)
548 off = s->offset + sizeof(void *);
549 else
550 off = s->inuse;
551
Christoph Lameter24922682007-07-17 04:03:18 -0700552 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700553 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700554
555 if (off != s->size)
556 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700557 print_section("Padding", p + off, s->size - off);
558
559 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700560}
561
562static void object_err(struct kmem_cache *s, struct page *page,
563 u8 *object, char *reason)
564{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700565 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700566 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700567}
568
Christoph Lameter24922682007-07-17 04:03:18 -0700569static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700570{
571 va_list args;
572 char buf[100];
573
Christoph Lameter24922682007-07-17 04:03:18 -0700574 va_start(args, fmt);
575 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700576 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700577 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700578 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700579 dump_stack();
580}
581
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500582static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700583{
584 u8 *p = object;
585
586 if (s->flags & __OBJECT_POISON) {
587 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800588 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700589 }
590
591 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500592 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700593}
594
Christoph Lameter24922682007-07-17 04:03:18 -0700595static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700596{
597 while (bytes) {
598 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700599 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700600 start++;
601 bytes--;
602 }
Christoph Lameter24922682007-07-17 04:03:18 -0700603 return NULL;
604}
605
606static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
607 void *from, void *to)
608{
609 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
610 memset(from, data, to - from);
611}
612
613static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
614 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800615 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700616{
617 u8 *fault;
618 u8 *end;
619
620 fault = check_bytes(start, value, bytes);
621 if (!fault)
622 return 1;
623
624 end = start + bytes;
625 while (end > fault && end[-1] == value)
626 end--;
627
628 slab_bug(s, "%s overwritten", what);
629 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
630 fault, end - 1, fault[0], value);
631 print_trailer(s, page, object);
632
633 restore_bytes(s, what, value, fault, end);
634 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700635}
636
Christoph Lameter81819f02007-05-06 14:49:36 -0700637/*
638 * Object layout:
639 *
640 * object address
641 * Bytes of the object to be managed.
642 * If the freepointer may overlay the object then the free
643 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700644 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700645 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
646 * 0xa5 (POISON_END)
647 *
648 * object + s->objsize
649 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700650 * Padding is extended by another word if Redzoning is enabled and
651 * objsize == inuse.
652 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700653 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
654 * 0xcc (RED_ACTIVE) for objects in use.
655 *
656 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700657 * Meta data starts here.
658 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700659 * A. Free pointer (if we cannot overwrite object on free)
660 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700661 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800662 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700663 * before the word boundary.
664 *
665 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700666 *
667 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700668 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700669 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700670 * If slabcaches are merged then the objsize and inuse boundaries are mostly
671 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700672 * may be used with merged slabcaches.
673 */
674
Christoph Lameter81819f02007-05-06 14:49:36 -0700675static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
676{
677 unsigned long off = s->inuse; /* The end of info */
678
679 if (s->offset)
680 /* Freepointer is placed after the object. */
681 off += sizeof(void *);
682
683 if (s->flags & SLAB_STORE_USER)
684 /* We also have user information there */
685 off += 2 * sizeof(struct track);
686
687 if (s->size == off)
688 return 1;
689
Christoph Lameter24922682007-07-17 04:03:18 -0700690 return check_bytes_and_report(s, page, p, "Object padding",
691 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700692}
693
Christoph Lameter39b26462008-04-14 19:11:30 +0300694/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700695static int slab_pad_check(struct kmem_cache *s, struct page *page)
696{
Christoph Lameter24922682007-07-17 04:03:18 -0700697 u8 *start;
698 u8 *fault;
699 u8 *end;
700 int length;
701 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700702
703 if (!(s->flags & SLAB_POISON))
704 return 1;
705
Christoph Lametera973e9d2008-03-01 13:40:44 -0800706 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800707 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300708 end = start + length;
709 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700710 if (!remainder)
711 return 1;
712
Christoph Lameter39b26462008-04-14 19:11:30 +0300713 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700714 if (!fault)
715 return 1;
716 while (end > fault && end[-1] == POISON_INUSE)
717 end--;
718
719 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300720 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700721
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200722 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700723 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700724}
725
726static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500727 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700728{
729 u8 *p = object;
730 u8 *endobject = object + s->objsize;
731
732 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700733 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500734 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700735 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700736 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800737 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
738 check_bytes_and_report(s, page, p, "Alignment padding",
739 endobject, POISON_INUSE, s->inuse - s->objsize);
740 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700741 }
742
743 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500744 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700745 (!check_bytes_and_report(s, page, p, "Poison", p,
746 POISON_FREE, s->objsize - 1) ||
747 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800748 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700749 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700750 /*
751 * check_pad_bytes cleans up on its own.
752 */
753 check_pad_bytes(s, page, p);
754 }
755
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500756 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700757 /*
758 * Object and freepointer overlap. Cannot check
759 * freepointer while object is allocated.
760 */
761 return 1;
762
763 /* Check free pointer validity */
764 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
765 object_err(s, page, p, "Freepointer corrupt");
766 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100767 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700768 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700769 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700770 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800771 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700772 return 0;
773 }
774 return 1;
775}
776
777static int check_slab(struct kmem_cache *s, struct page *page)
778{
Christoph Lameter39b26462008-04-14 19:11:30 +0300779 int maxobj;
780
Christoph Lameter81819f02007-05-06 14:49:36 -0700781 VM_BUG_ON(!irqs_disabled());
782
783 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700784 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700785 return 0;
786 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300787
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800788 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300789 if (page->objects > maxobj) {
790 slab_err(s, page, "objects %u > max %u",
791 s->name, page->objects, maxobj);
792 return 0;
793 }
794 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700795 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300796 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700797 return 0;
798 }
799 /* Slab_pad_check fixes things up after itself */
800 slab_pad_check(s, page);
801 return 1;
802}
803
804/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700805 * Determine if a certain object on a page is on the freelist. Must hold the
806 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700807 */
808static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
809{
810 int nr = 0;
811 void *fp = page->freelist;
812 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300813 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700814
Christoph Lameter39b26462008-04-14 19:11:30 +0300815 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700816 if (fp == search)
817 return 1;
818 if (!check_valid_pointer(s, page, fp)) {
819 if (object) {
820 object_err(s, page, object,
821 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800822 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700823 break;
824 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700825 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800826 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300827 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700828 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700829 return 0;
830 }
831 break;
832 }
833 object = fp;
834 fp = get_freepointer(s, object);
835 nr++;
836 }
837
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800838 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400839 if (max_objects > MAX_OBJS_PER_PAGE)
840 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300841
842 if (page->objects != max_objects) {
843 slab_err(s, page, "Wrong number of objects. Found %d but "
844 "should be %d", page->objects, max_objects);
845 page->objects = max_objects;
846 slab_fix(s, "Number of objects adjusted.");
847 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300848 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700849 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300850 "counted were %d", page->inuse, page->objects - nr);
851 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700852 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700853 }
854 return search == NULL;
855}
856
Christoph Lameter0121c6192008-04-29 16:11:12 -0700857static void trace(struct kmem_cache *s, struct page *page, void *object,
858 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700859{
860 if (s->flags & SLAB_TRACE) {
861 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
862 s->name,
863 alloc ? "alloc" : "free",
864 object, page->inuse,
865 page->freelist);
866
867 if (!alloc)
868 print_section("Object", (void *)object, s->objsize);
869
870 dump_stack();
871 }
872}
873
Christoph Lameter643b1132007-05-06 14:49:42 -0700874/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500875 * Hooks for other subsystems that check memory allocations. In a typical
876 * production configuration these hooks all should produce no code at all.
877 */
878static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
879{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500880 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500881 lockdep_trace_alloc(flags);
882 might_sleep_if(flags & __GFP_WAIT);
883
884 return should_failslab(s->objsize, flags, s->flags);
885}
886
887static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
888{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500889 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100890 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500891 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
892}
893
894static inline void slab_free_hook(struct kmem_cache *s, void *x)
895{
896 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500897
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600898 /*
899 * Trouble is that we may no longer disable interupts in the fast path
900 * So in order to make the debug calls that expect irqs to be
901 * disabled we need to disable interrupts temporarily.
902 */
903#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
904 {
905 unsigned long flags;
906
907 local_irq_save(flags);
908 kmemcheck_slab_free(s, x, s->objsize);
909 debug_check_no_locks_freed(x, s->objsize);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600910 local_irq_restore(flags);
911 }
912#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200913 if (!(s->flags & SLAB_DEBUG_OBJECTS))
914 debug_check_no_obj_freed(x, s->objsize);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500915}
916
917/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700918 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700919 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700920static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700921{
Christoph Lameter643b1132007-05-06 14:49:42 -0700922 spin_lock(&n->list_lock);
923 list_add(&page->lru, &n->full);
924 spin_unlock(&n->list_lock);
925}
926
927static void remove_full(struct kmem_cache *s, struct page *page)
928{
929 struct kmem_cache_node *n;
930
931 if (!(s->flags & SLAB_STORE_USER))
932 return;
933
934 n = get_node(s, page_to_nid(page));
935
936 spin_lock(&n->list_lock);
937 list_del(&page->lru);
938 spin_unlock(&n->list_lock);
939}
940
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300941/* Tracking of the number of slabs for debugging purposes */
942static inline unsigned long slabs_node(struct kmem_cache *s, int node)
943{
944 struct kmem_cache_node *n = get_node(s, node);
945
946 return atomic_long_read(&n->nr_slabs);
947}
948
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400949static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
950{
951 return atomic_long_read(&n->nr_slabs);
952}
953
Christoph Lameter205ab992008-04-14 19:11:40 +0300954static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300955{
956 struct kmem_cache_node *n = get_node(s, node);
957
958 /*
959 * May be called early in order to allocate a slab for the
960 * kmem_cache_node structure. Solve the chicken-egg
961 * dilemma by deferring the increment of the count during
962 * bootstrap (see early_kmem_cache_node_alloc).
963 */
Christoph Lameter7340cc82010-09-28 08:10:26 -0500964 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300965 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300966 atomic_long_add(objects, &n->total_objects);
967 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300968}
Christoph Lameter205ab992008-04-14 19:11:40 +0300969static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300970{
971 struct kmem_cache_node *n = get_node(s, node);
972
973 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300974 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300975}
976
977/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700978static void setup_object_debug(struct kmem_cache *s, struct page *page,
979 void *object)
980{
981 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
982 return;
983
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500984 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700985 init_tracking(s, object);
986}
987
Christoph Lameter15370662010-08-20 12:37:12 -0500988static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300989 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700990{
991 if (!check_slab(s, page))
992 goto bad;
993
Christoph Lameterd692ef62008-02-15 23:45:24 -0800994 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700995 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700996 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700997 }
998
999 if (!check_valid_pointer(s, page, object)) {
1000 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001001 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001002 }
1003
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001004 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001005 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001006
Christoph Lameter3ec09742007-05-16 22:11:00 -07001007 /* Success perform special debug activities for allocs */
1008 if (s->flags & SLAB_STORE_USER)
1009 set_track(s, object, TRACK_ALLOC, addr);
1010 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001011 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001012 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001013
Christoph Lameter81819f02007-05-06 14:49:36 -07001014bad:
1015 if (PageSlab(page)) {
1016 /*
1017 * If this is a slab page then lets do the best we can
1018 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001019 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001020 */
Christoph Lameter24922682007-07-17 04:03:18 -07001021 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001022 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001023 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001024 }
1025 return 0;
1026}
1027
Christoph Lameter15370662010-08-20 12:37:12 -05001028static noinline int free_debug_processing(struct kmem_cache *s,
1029 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001030{
1031 if (!check_slab(s, page))
1032 goto fail;
1033
1034 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001035 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001036 goto fail;
1037 }
1038
1039 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001040 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001041 goto fail;
1042 }
1043
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001044 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001045 return 0;
1046
1047 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001048 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001049 slab_err(s, page, "Attempt to free object(0x%p) "
1050 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001051 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001052 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001053 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001054 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001055 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001056 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001057 object_err(s, page, object,
1058 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001059 goto fail;
1060 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001061
1062 /* Special debug activities for freeing objects */
Christoph Lameter50d5c412011-06-01 12:25:45 -05001063 if (!page->frozen && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -07001064 remove_full(s, page);
1065 if (s->flags & SLAB_STORE_USER)
1066 set_track(s, object, TRACK_FREE, addr);
1067 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001068 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001069 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001070
Christoph Lameter81819f02007-05-06 14:49:36 -07001071fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001072 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001073 return 0;
1074}
1075
Christoph Lameter41ecc552007-05-09 02:32:44 -07001076static int __init setup_slub_debug(char *str)
1077{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001078 slub_debug = DEBUG_DEFAULT_FLAGS;
1079 if (*str++ != '=' || !*str)
1080 /*
1081 * No options specified. Switch on full debugging.
1082 */
1083 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001084
1085 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001086 /*
1087 * No options but restriction on slabs. This means full
1088 * debugging for slabs matching a pattern.
1089 */
1090 goto check_slabs;
1091
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001092 if (tolower(*str) == 'o') {
1093 /*
1094 * Avoid enabling debugging on caches if its minimum order
1095 * would increase as a result.
1096 */
1097 disable_higher_order_debug = 1;
1098 goto out;
1099 }
1100
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001101 slub_debug = 0;
1102 if (*str == '-')
1103 /*
1104 * Switch off all debugging measures.
1105 */
1106 goto out;
1107
1108 /*
1109 * Determine which debug features should be switched on
1110 */
Pekka Enberg06428782008-01-07 23:20:27 -08001111 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001112 switch (tolower(*str)) {
1113 case 'f':
1114 slub_debug |= SLAB_DEBUG_FREE;
1115 break;
1116 case 'z':
1117 slub_debug |= SLAB_RED_ZONE;
1118 break;
1119 case 'p':
1120 slub_debug |= SLAB_POISON;
1121 break;
1122 case 'u':
1123 slub_debug |= SLAB_STORE_USER;
1124 break;
1125 case 't':
1126 slub_debug |= SLAB_TRACE;
1127 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001128 case 'a':
1129 slub_debug |= SLAB_FAILSLAB;
1130 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001131 default:
1132 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001133 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001134 }
1135 }
1136
1137check_slabs:
1138 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001139 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001140out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001141 return 1;
1142}
1143
1144__setup("slub_debug", setup_slub_debug);
1145
Christoph Lameterba0268a2007-09-11 15:24:11 -07001146static unsigned long kmem_cache_flags(unsigned long objsize,
1147 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001148 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001149{
1150 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001151 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001152 */
Christoph Lametere1533622008-02-15 23:45:24 -08001153 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001154 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1155 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001156
1157 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001158}
1159#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001160static inline void setup_object_debug(struct kmem_cache *s,
1161 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001162
Christoph Lameter3ec09742007-05-16 22:11:00 -07001163static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001164 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001165
Christoph Lameter3ec09742007-05-16 22:11:00 -07001166static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001167 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001168
Christoph Lameter41ecc552007-05-09 02:32:44 -07001169static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1170 { return 1; }
1171static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001172 void *object, u8 val) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001173static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001174static inline unsigned long kmem_cache_flags(unsigned long objsize,
1175 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001176 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001177{
1178 return flags;
1179}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001180#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001181
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001182#define disable_higher_order_debug 0
1183
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001184static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1185 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001186static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1187 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001188static inline void inc_slabs_node(struct kmem_cache *s, int node,
1189 int objects) {}
1190static inline void dec_slabs_node(struct kmem_cache *s, int node,
1191 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001192
1193static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1194 { return 0; }
1195
1196static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1197 void *object) {}
1198
1199static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1200
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001201#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001202
Christoph Lameter81819f02007-05-06 14:49:36 -07001203/*
1204 * Slab allocation and freeing
1205 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001206static inline struct page *alloc_slab_page(gfp_t flags, int node,
1207 struct kmem_cache_order_objects oo)
1208{
1209 int order = oo_order(oo);
1210
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001211 flags |= __GFP_NOTRACK;
1212
Christoph Lameter2154a332010-07-09 14:07:10 -05001213 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001214 return alloc_pages(flags, order);
1215 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001216 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001217}
1218
Christoph Lameter81819f02007-05-06 14:49:36 -07001219static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1220{
Pekka Enberg06428782008-01-07 23:20:27 -08001221 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001222 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001223 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001224
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001225 flags &= gfp_allowed_mask;
1226
1227 if (flags & __GFP_WAIT)
1228 local_irq_enable();
1229
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001230 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001231
Pekka Enbergba522702009-06-24 21:59:51 +03001232 /*
1233 * Let the initial higher-order allocation fail under memory pressure
1234 * so we fall-back to the minimum order allocation.
1235 */
1236 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1237
1238 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001239 if (unlikely(!page)) {
1240 oo = s->min;
1241 /*
1242 * Allocation may have failed due to fragmentation.
1243 * Try a lower order alloc if possible
1244 */
1245 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001246
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001247 if (page)
1248 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001249 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001250
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001251 if (flags & __GFP_WAIT)
1252 local_irq_disable();
1253
1254 if (!page)
1255 return NULL;
1256
Vegard Nossum5a896d92008-04-04 00:54:48 +02001257 if (kmemcheck_enabled
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001258 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001259 int pages = 1 << oo_order(oo);
1260
1261 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1262
1263 /*
1264 * Objects from caches that have a constructor don't get
1265 * cleared when they're allocated, so we need to do it here.
1266 */
1267 if (s->ctor)
1268 kmemcheck_mark_uninitialized_pages(page, pages);
1269 else
1270 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001271 }
1272
Christoph Lameter834f3d12008-04-14 19:11:31 +03001273 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001274 mod_zone_page_state(page_zone(page),
1275 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1276 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001277 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001278
1279 return page;
1280}
1281
1282static void setup_object(struct kmem_cache *s, struct page *page,
1283 void *object)
1284{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001285 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001286 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001287 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001288}
1289
1290static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1291{
1292 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001293 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001294 void *last;
1295 void *p;
1296
Christoph Lameter6cb06222007-10-16 01:25:41 -07001297 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001298
Christoph Lameter6cb06222007-10-16 01:25:41 -07001299 page = allocate_slab(s,
1300 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001301 if (!page)
1302 goto out;
1303
Christoph Lameter205ab992008-04-14 19:11:40 +03001304 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001305 page->slab = s;
1306 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001307
1308 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001309
1310 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001311 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001312
1313 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001314 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001315 setup_object(s, page, last);
1316 set_freepointer(s, last, p);
1317 last = p;
1318 }
1319 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001320 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001321
1322 page->freelist = start;
1323 page->inuse = 0;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001324 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001325out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001326 return page;
1327}
1328
1329static void __free_slab(struct kmem_cache *s, struct page *page)
1330{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001331 int order = compound_order(page);
1332 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001333
Christoph Lameteraf537b02010-07-09 14:07:14 -05001334 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001335 void *p;
1336
1337 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001338 for_each_object(p, s, page_address(page),
1339 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001340 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001341 }
1342
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001343 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001344
Christoph Lameter81819f02007-05-06 14:49:36 -07001345 mod_zone_page_state(page_zone(page),
1346 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1347 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001348 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001349
Christoph Lameter49bd5222008-04-14 18:52:18 +03001350 __ClearPageSlab(page);
1351 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001352 if (current->reclaim_state)
1353 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001354 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001355}
1356
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001357#define need_reserve_slab_rcu \
1358 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1359
Christoph Lameter81819f02007-05-06 14:49:36 -07001360static void rcu_free_slab(struct rcu_head *h)
1361{
1362 struct page *page;
1363
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001364 if (need_reserve_slab_rcu)
1365 page = virt_to_head_page(h);
1366 else
1367 page = container_of((struct list_head *)h, struct page, lru);
1368
Christoph Lameter81819f02007-05-06 14:49:36 -07001369 __free_slab(page->slab, page);
1370}
1371
1372static void free_slab(struct kmem_cache *s, struct page *page)
1373{
1374 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001375 struct rcu_head *head;
1376
1377 if (need_reserve_slab_rcu) {
1378 int order = compound_order(page);
1379 int offset = (PAGE_SIZE << order) - s->reserved;
1380
1381 VM_BUG_ON(s->reserved != sizeof(*head));
1382 head = page_address(page) + offset;
1383 } else {
1384 /*
1385 * RCU free overloads the RCU head over the LRU
1386 */
1387 head = (void *)&page->lru;
1388 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001389
1390 call_rcu(head, rcu_free_slab);
1391 } else
1392 __free_slab(s, page);
1393}
1394
1395static void discard_slab(struct kmem_cache *s, struct page *page)
1396{
Christoph Lameter205ab992008-04-14 19:11:40 +03001397 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001398 free_slab(s, page);
1399}
1400
1401/*
1402 * Per slab locking using the pagelock
1403 */
1404static __always_inline void slab_lock(struct page *page)
1405{
1406 bit_spin_lock(PG_locked, &page->flags);
1407}
1408
1409static __always_inline void slab_unlock(struct page *page)
1410{
Nick Piggina76d3542008-01-07 23:20:27 -08001411 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001412}
1413
1414static __always_inline int slab_trylock(struct page *page)
1415{
1416 int rc = 1;
1417
1418 rc = bit_spin_trylock(PG_locked, &page->flags);
1419 return rc;
1420}
1421
1422/*
1423 * Management of partially allocated slabs
1424 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001425static void add_partial(struct kmem_cache_node *n,
1426 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001427{
Christoph Lametere95eed52007-05-06 14:49:44 -07001428 spin_lock(&n->list_lock);
1429 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001430 if (tail)
1431 list_add_tail(&page->lru, &n->partial);
1432 else
1433 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001434 spin_unlock(&n->list_lock);
1435}
1436
Christoph Lameter62e346a2010-09-28 08:10:28 -05001437static inline void __remove_partial(struct kmem_cache_node *n,
1438 struct page *page)
1439{
1440 list_del(&page->lru);
1441 n->nr_partial--;
1442}
1443
Christoph Lameter0121c6192008-04-29 16:11:12 -07001444static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001445{
1446 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1447
1448 spin_lock(&n->list_lock);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001449 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001450 spin_unlock(&n->list_lock);
1451}
1452
1453/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001454 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001455 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001456 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001457 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001458static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1459 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001460{
1461 if (slab_trylock(page)) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05001462 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001463 return 1;
1464 }
1465 return 0;
1466}
1467
1468/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001469 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001470 */
1471static struct page *get_partial_node(struct kmem_cache_node *n)
1472{
1473 struct page *page;
1474
1475 /*
1476 * Racy check. If we mistakenly see no partial slabs then we
1477 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001478 * partial slab and there is none available then get_partials()
1479 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001480 */
1481 if (!n || !n->nr_partial)
1482 return NULL;
1483
1484 spin_lock(&n->list_lock);
1485 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001486 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001487 goto out;
1488 page = NULL;
1489out:
1490 spin_unlock(&n->list_lock);
1491 return page;
1492}
1493
1494/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001495 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001496 */
1497static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1498{
1499#ifdef CONFIG_NUMA
1500 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001501 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001502 struct zone *zone;
1503 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001504 struct page *page;
1505
1506 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001507 * The defrag ratio allows a configuration of the tradeoffs between
1508 * inter node defragmentation and node local allocations. A lower
1509 * defrag_ratio increases the tendency to do local allocations
1510 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001511 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001512 * If the defrag_ratio is set to 0 then kmalloc() always
1513 * returns node local objects. If the ratio is higher then kmalloc()
1514 * may return off node objects because partial slabs are obtained
1515 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001516 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001517 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001518 * defrag_ratio = 1000) then every (well almost) allocation will
1519 * first attempt to defrag slab caches on other nodes. This means
1520 * scanning over all nodes to look for partial slabs which may be
1521 * expensive if we do it every time we are trying to find a slab
1522 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001523 */
Christoph Lameter98246012008-01-07 23:20:26 -08001524 if (!s->remote_node_defrag_ratio ||
1525 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001526 return NULL;
1527
Miao Xiec0ff7452010-05-24 14:32:08 -07001528 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001529 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001530 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001531 struct kmem_cache_node *n;
1532
Mel Gorman54a6eb52008-04-28 02:12:16 -07001533 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001534
Mel Gorman54a6eb52008-04-28 02:12:16 -07001535 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001536 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001537 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001538 if (page) {
1539 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001540 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001541 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001542 }
1543 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001544 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001545#endif
1546 return NULL;
1547}
1548
1549/*
1550 * Get a partial page, lock it and return it.
1551 */
1552static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1553{
1554 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001555 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001556
1557 page = get_partial_node(get_node(s, searchnode));
Christoph Lameter33de04e2011-04-15 14:48:12 -05001558 if (page || node != NUMA_NO_NODE)
Christoph Lameter81819f02007-05-06 14:49:36 -07001559 return page;
1560
1561 return get_any_partial(s, flags);
1562}
1563
1564/*
1565 * Move a page back to the lists.
1566 *
1567 * Must be called with the slab lock held.
1568 *
1569 * On exit the slab lock will have been dropped.
1570 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001571static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Namhyung Kim34789732010-09-29 21:02:14 +09001572 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001573{
Christoph Lametere95eed52007-05-06 14:49:44 -07001574 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1575
Christoph Lameter81819f02007-05-06 14:49:36 -07001576 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001577
Christoph Lametera973e9d2008-03-01 13:40:44 -08001578 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001579 add_partial(n, page, tail);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001580 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001581 } else {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001582 stat(s, DEACTIVATE_FULL);
Christoph Lameteraf537b02010-07-09 14:07:14 -05001583 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001584 add_full(n, page);
1585 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001586 slab_unlock(page);
1587 } else {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001588 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001589 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001590 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001591 * Adding an empty slab to the partial slabs in order
1592 * to avoid page allocator overhead. This slab needs
1593 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001594 * so that the others get filled first. That way the
1595 * size of the partial list stays small.
1596 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001597 * kmem_cache_shrink can reclaim any empty slabs from
1598 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001599 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001600 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001601 slab_unlock(page);
1602 } else {
1603 slab_unlock(page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001604 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001605 discard_slab(s, page);
1606 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001607 }
1608}
1609
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001610#ifdef CONFIG_PREEMPT
1611/*
1612 * Calculate the next globally unique transaction for disambiguiation
1613 * during cmpxchg. The transactions start with the cpu number and are then
1614 * incremented by CONFIG_NR_CPUS.
1615 */
1616#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1617#else
1618/*
1619 * No preemption supported therefore also no need to check for
1620 * different cpus.
1621 */
1622#define TID_STEP 1
1623#endif
1624
1625static inline unsigned long next_tid(unsigned long tid)
1626{
1627 return tid + TID_STEP;
1628}
1629
1630static inline unsigned int tid_to_cpu(unsigned long tid)
1631{
1632 return tid % TID_STEP;
1633}
1634
1635static inline unsigned long tid_to_event(unsigned long tid)
1636{
1637 return tid / TID_STEP;
1638}
1639
1640static inline unsigned int init_tid(int cpu)
1641{
1642 return cpu;
1643}
1644
1645static inline void note_cmpxchg_failure(const char *n,
1646 const struct kmem_cache *s, unsigned long tid)
1647{
1648#ifdef SLUB_DEBUG_CMPXCHG
1649 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1650
1651 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1652
1653#ifdef CONFIG_PREEMPT
1654 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1655 printk("due to cpu change %d -> %d\n",
1656 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1657 else
1658#endif
1659 if (tid_to_event(tid) != tid_to_event(actual_tid))
1660 printk("due to cpu running other code. Event %ld->%ld\n",
1661 tid_to_event(tid), tid_to_event(actual_tid));
1662 else
1663 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1664 actual_tid, tid, next_tid(tid));
1665#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001666 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001667}
1668
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001669void init_kmem_cache_cpus(struct kmem_cache *s)
1670{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001671 int cpu;
1672
1673 for_each_possible_cpu(cpu)
1674 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001675}
Christoph Lameter81819f02007-05-06 14:49:36 -07001676/*
1677 * Remove the cpu slab
1678 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001679static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Namhyung Kim34789732010-09-29 21:02:14 +09001680 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001681{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001682 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001683 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001684
Christoph Lameterb773ad72008-03-04 11:10:17 -08001685 if (page->freelist)
Christoph Lameter84e554e62009-12-18 16:26:23 -06001686 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001687 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001688 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001689 * because both freelists are empty. So this is unlikely
1690 * to occur.
1691 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001692 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001693 void **object;
1694
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001695 tail = 0; /* Hot objects. Put the slab first */
1696
Christoph Lameter894b8782007-05-10 03:15:16 -07001697 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001698 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001699 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001700
1701 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001702 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001703 page->freelist = object;
1704 page->inuse--;
1705 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001706 c->page = NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001707 c->tid = next_tid(c->tid);
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001708 page->frozen = 0;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001709 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001710}
1711
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001712static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001713{
Christoph Lameter84e554e62009-12-18 16:26:23 -06001714 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001715 slab_lock(c->page);
1716 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001717}
1718
1719/*
1720 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001721 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001722 * Called from IPI handler with interrupts disabled.
1723 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001724static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001725{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001726 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001727
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001728 if (likely(c && c->page))
1729 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001730}
1731
1732static void flush_cpu_slab(void *d)
1733{
1734 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001735
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001736 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001737}
1738
1739static void flush_all(struct kmem_cache *s)
1740{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001741 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001742}
1743
1744/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001745 * Check if the objects in a per cpu structure fit numa
1746 * locality expectations.
1747 */
1748static inline int node_match(struct kmem_cache_cpu *c, int node)
1749{
1750#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001751 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001752 return 0;
1753#endif
1754 return 1;
1755}
1756
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001757static int count_free(struct page *page)
1758{
1759 return page->objects - page->inuse;
1760}
1761
1762static unsigned long count_partial(struct kmem_cache_node *n,
1763 int (*get_count)(struct page *))
1764{
1765 unsigned long flags;
1766 unsigned long x = 0;
1767 struct page *page;
1768
1769 spin_lock_irqsave(&n->list_lock, flags);
1770 list_for_each_entry(page, &n->partial, lru)
1771 x += get_count(page);
1772 spin_unlock_irqrestore(&n->list_lock, flags);
1773 return x;
1774}
1775
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001776static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1777{
1778#ifdef CONFIG_SLUB_DEBUG
1779 return atomic_long_read(&n->total_objects);
1780#else
1781 return 0;
1782#endif
1783}
1784
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001785static noinline void
1786slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1787{
1788 int node;
1789
1790 printk(KERN_WARNING
1791 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1792 nid, gfpflags);
1793 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1794 "default order: %d, min order: %d\n", s->name, s->objsize,
1795 s->size, oo_order(s->oo), oo_order(s->min));
1796
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001797 if (oo_order(s->min) > get_order(s->objsize))
1798 printk(KERN_WARNING " %s debugging increased min order, use "
1799 "slub_debug=O to disable.\n", s->name);
1800
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001801 for_each_online_node(node) {
1802 struct kmem_cache_node *n = get_node(s, node);
1803 unsigned long nr_slabs;
1804 unsigned long nr_objs;
1805 unsigned long nr_free;
1806
1807 if (!n)
1808 continue;
1809
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001810 nr_free = count_partial(n, count_free);
1811 nr_slabs = node_nr_slabs(n);
1812 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001813
1814 printk(KERN_WARNING
1815 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1816 node, nr_slabs, nr_objs, nr_free);
1817 }
1818}
1819
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001820/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001821 * Slow path. The lockless freelist is empty or we need to perform
1822 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001823 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001824 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001825 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001826 * Processing is still very fast if new objects have been freed to the
1827 * regular freelist. In that case we simply take over the regular freelist
1828 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001829 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001830 * If that is not working then we fall back to the partial lists. We take the
1831 * first element of the freelist as the object to allocate now and move the
1832 * rest of the freelist to the lockless freelist.
1833 *
1834 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001835 * we need to allocate a new slab. This is the slowest path since it involves
1836 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001837 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001838static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1839 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001840{
Christoph Lameter81819f02007-05-06 14:49:36 -07001841 void **object;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001842 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001843 unsigned long flags;
1844
1845 local_irq_save(flags);
1846#ifdef CONFIG_PREEMPT
1847 /*
1848 * We may have been preempted and rescheduled on a different
1849 * cpu before disabling interrupts. Need to reload cpu area
1850 * pointer.
1851 */
1852 c = this_cpu_ptr(s->cpu_slab);
1853#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07001854
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001855 /* We handle __GFP_ZERO in the caller */
1856 gfpflags &= ~__GFP_ZERO;
1857
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001858 page = c->page;
1859 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001860 goto new_slab;
1861
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001862 slab_lock(page);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001863 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001864 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001865
Christoph Lameter84e554e62009-12-18 16:26:23 -06001866 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001867
Christoph Lameter894b8782007-05-10 03:15:16 -07001868load_freelist:
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001869 VM_BUG_ON(!page->frozen);
1870
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001871 object = page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001872 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001873 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001874 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001875 goto debug;
1876
Christoph Lameterff120592009-12-18 16:26:22 -06001877 c->freelist = get_freepointer(s, object);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001878 page->inuse = page->objects;
1879 page->freelist = NULL;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001880
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001881 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001882 c->tid = next_tid(c->tid);
1883 local_irq_restore(flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001884 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001885 return object;
1886
1887another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001888 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001889
1890new_slab:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001891 page = get_partial(s, gfpflags, node);
1892 if (page) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001893 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001894 page->frozen = 1;
Christoph Lameterdc1fb7f2011-04-15 14:48:15 -05001895 c->node = page_to_nid(page);
1896 c->page = page;
Christoph Lameter894b8782007-05-10 03:15:16 -07001897 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001898 }
1899
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001900 page = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001901
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001902 if (page) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001903 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001904 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001905 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001906 flush_slab(s, c);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001907
1908 slab_lock(page);
Christoph Lameter50d5c412011-06-01 12:25:45 -05001909 page->frozen = 1;
David Rientjesbd07d872011-05-12 13:10:49 -07001910 c->node = page_to_nid(page);
1911 c->page = page;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001912 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001913 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001914 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1915 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter2fd66c52011-03-22 13:32:53 -05001916 local_irq_restore(flags);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001917 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001918debug:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001919 if (!alloc_debug_processing(s, page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001920 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001921
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001922 page->inuse++;
1923 page->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05001924 deactivate_slab(s, c);
1925 c->page = NULL;
Pekka Enberg15b7c512010-10-02 11:32:32 +03001926 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05001927 local_irq_restore(flags);
1928 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07001929}
1930
1931/*
1932 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1933 * have the fastpath folded into their functions. So no function call
1934 * overhead for requests that can be satisfied on the fastpath.
1935 *
1936 * The fastpath works by first checking if the lockless freelist can be used.
1937 * If not then __slab_alloc is called for slow processing.
1938 *
1939 * Otherwise we can simply pick the next object from the lockless free list.
1940 */
Pekka Enberg06428782008-01-07 23:20:27 -08001941static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001942 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001943{
Christoph Lameter894b8782007-05-10 03:15:16 -07001944 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001945 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001946 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08001947
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001948 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001949 return NULL;
1950
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001951redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001952
1953 /*
1954 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1955 * enabled. We may switch back and forth between cpus while
1956 * reading from one cpu area. That does not matter as long
1957 * as we end up on the original cpu again when doing the cmpxchg.
1958 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001959 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001960
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001961 /*
1962 * The transaction ids are globally unique per cpu and per operation on
1963 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1964 * occurs on the right processor and that there was no operation on the
1965 * linked list in between.
1966 */
1967 tid = c->tid;
1968 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001969
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001970 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001971 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001972
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001973 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001974
1975 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001976 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001977 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001978 * operation and if we are on the right processor.
1979 *
1980 * The cmpxchg does the following atomically (without lock semantics!)
1981 * 1. Relocate first pointer to the current per cpu area.
1982 * 2. Verify that tid and freelist have not been changed
1983 * 3. If they were not changed replace tid and freelist
1984 *
1985 * Since this is without lock semantics the protection is only against
1986 * code executing on this cpu *not* from access by other cpus.
1987 */
Thomas Gleixner30106b82011-05-04 15:38:19 +02001988 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001989 s->cpu_slab->freelist, s->cpu_slab->tid,
1990 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05001991 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001992
1993 note_cmpxchg_failure("slab_alloc", s, tid);
1994 goto redo;
1995 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06001996 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001997 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001998
Pekka Enberg74e21342009-11-25 20:14:48 +02001999 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06002000 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002001
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002002 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002003
Christoph Lameter894b8782007-05-10 03:15:16 -07002004 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002005}
2006
2007void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2008{
Christoph Lameter2154a332010-07-09 14:07:10 -05002009 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002010
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002011 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002012
2013 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002014}
2015EXPORT_SYMBOL(kmem_cache_alloc);
2016
Li Zefan0f24f122009-12-11 15:45:30 +08002017#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002018void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002019{
Richard Kennedy4a923792010-10-21 10:29:19 +01002020 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
2021 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2022 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002023}
Richard Kennedy4a923792010-10-21 10:29:19 +01002024EXPORT_SYMBOL(kmem_cache_alloc_trace);
2025
2026void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2027{
2028 void *ret = kmalloc_order(size, flags, order);
2029 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2030 return ret;
2031}
2032EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002033#endif
2034
Christoph Lameter81819f02007-05-06 14:49:36 -07002035#ifdef CONFIG_NUMA
2036void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2037{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002038 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2039
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002040 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2041 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002042
2043 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002044}
2045EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002046
Li Zefan0f24f122009-12-11 15:45:30 +08002047#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002048void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002049 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002050 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002051{
Richard Kennedy4a923792010-10-21 10:29:19 +01002052 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2053
2054 trace_kmalloc_node(_RET_IP_, ret,
2055 size, s->size, gfpflags, node);
2056 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002057}
Richard Kennedy4a923792010-10-21 10:29:19 +01002058EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002059#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002060#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002061
Christoph Lameter81819f02007-05-06 14:49:36 -07002062/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002063 * Slow patch handling. This may still be called frequently since objects
2064 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002065 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002066 * So we still attempt to reduce cache line usage. Just take the slab
2067 * lock and free the item. If there is no additional partial page
2068 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002069 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002070static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002071 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002072{
2073 void *prior;
2074 void **object = (void *)x;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002075 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002076
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002077 local_irq_save(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002078 slab_lock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002079 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002080
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002081 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2082 goto out_unlock;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002083
Christoph Lameterff120592009-12-18 16:26:22 -06002084 prior = page->freelist;
2085 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07002086 page->freelist = object;
2087 page->inuse--;
2088
Christoph Lameter50d5c412011-06-01 12:25:45 -05002089 if (unlikely(page->frozen)) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002090 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07002091 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002092 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002093
2094 if (unlikely(!page->inuse))
2095 goto slab_empty;
2096
2097 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002098 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07002099 * then add it.
2100 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08002101 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002102 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002103 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002104 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002105
2106out_unlock:
2107 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002108 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002109 return;
2110
2111slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002112 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002113 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002114 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002115 */
2116 remove_partial(s, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002117 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002118 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002119 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002120 local_irq_restore(flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002121 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002122 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002123}
2124
Christoph Lameter894b8782007-05-10 03:15:16 -07002125/*
2126 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2127 * can perform fastpath freeing without additional function calls.
2128 *
2129 * The fastpath is only possible if we are freeing to the current cpu slab
2130 * of this processor. This typically the case if we have just allocated
2131 * the item before.
2132 *
2133 * If fastpath is not possible then fall back to __slab_free where we deal
2134 * with all sorts of special processing.
2135 */
Pekka Enberg06428782008-01-07 23:20:27 -08002136static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002137 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002138{
2139 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002140 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002141 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002142
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002143 slab_free_hook(s, x);
2144
Christoph Lametera24c5a02011-03-15 12:45:21 -05002145redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002146
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002147 /*
2148 * Determine the currently cpus per cpu slab.
2149 * The cpu may change afterward. However that does not matter since
2150 * data is retrieved via this pointer. If we are on the same cpu
2151 * during the cmpxchg then the free will succedd.
2152 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002153 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002154
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002155 tid = c->tid;
2156 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002157
Christoph Lameter442b06b2011-05-17 16:29:31 -05002158 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002159 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002160
Thomas Gleixner30106b82011-05-04 15:38:19 +02002161 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002162 s->cpu_slab->freelist, s->cpu_slab->tid,
2163 c->freelist, tid,
2164 object, next_tid(tid)))) {
2165
2166 note_cmpxchg_failure("slab_free", s, tid);
2167 goto redo;
2168 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002169 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002170 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002171 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002172
Christoph Lameter894b8782007-05-10 03:15:16 -07002173}
2174
Christoph Lameter81819f02007-05-06 14:49:36 -07002175void kmem_cache_free(struct kmem_cache *s, void *x)
2176{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002177 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002178
Christoph Lameterb49af682007-05-06 14:49:41 -07002179 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002180
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002181 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002182
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002183 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002184}
2185EXPORT_SYMBOL(kmem_cache_free);
2186
Christoph Lameter81819f02007-05-06 14:49:36 -07002187/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002188 * Object placement in a slab is made very easy because we always start at
2189 * offset 0. If we tune the size of the object to the alignment then we can
2190 * get the required alignment by putting one properly sized object after
2191 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002192 *
2193 * Notice that the allocation order determines the sizes of the per cpu
2194 * caches. Each processor has always one slab available for allocations.
2195 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002196 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002197 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002198 */
2199
2200/*
2201 * Mininum / Maximum order of slab pages. This influences locking overhead
2202 * and slab fragmentation. A higher order reduces the number of partial slabs
2203 * and increases the number of allocations possible without having to
2204 * take the list_lock.
2205 */
2206static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002207static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002208static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002209
2210/*
2211 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002212 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002213 */
2214static int slub_nomerge;
2215
2216/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002217 * Calculate the order of allocation given an slab object size.
2218 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002219 * The order of allocation has significant impact on performance and other
2220 * system components. Generally order 0 allocations should be preferred since
2221 * order 0 does not cause fragmentation in the page allocator. Larger objects
2222 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002223 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002224 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002225 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002226 * In order to reach satisfactory performance we must ensure that a minimum
2227 * number of objects is in one slab. Otherwise we may generate too much
2228 * activity on the partial lists which requires taking the list_lock. This is
2229 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002230 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002231 * slub_max_order specifies the order where we begin to stop considering the
2232 * number of objects in a slab as critical. If we reach slub_max_order then
2233 * we try to keep the page order as low as possible. So we accept more waste
2234 * of space in favor of a small page order.
2235 *
2236 * Higher order allocations also allow the placement of more objects in a
2237 * slab and thereby reduce object handling overhead. If the user has
2238 * requested a higher mininum order then we start with that one instead of
2239 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002240 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002241static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002242 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002243{
2244 int order;
2245 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002246 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002247
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002248 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002249 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002250
Christoph Lameter6300ea72007-07-17 04:03:20 -07002251 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002252 fls(min_objects * size - 1) - PAGE_SHIFT);
2253 order <= max_order; order++) {
2254
Christoph Lameter81819f02007-05-06 14:49:36 -07002255 unsigned long slab_size = PAGE_SIZE << order;
2256
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002257 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002258 continue;
2259
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002260 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002261
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002262 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002263 break;
2264
2265 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002266
Christoph Lameter81819f02007-05-06 14:49:36 -07002267 return order;
2268}
2269
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002270static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002271{
2272 int order;
2273 int min_objects;
2274 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002275 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002276
2277 /*
2278 * Attempt to find best configuration for a slab. This
2279 * works by first attempting to generate a layout with
2280 * the best configuration and backing off gradually.
2281 *
2282 * First we reduce the acceptable waste in a slab. Then
2283 * we reduce the minimum objects required in a slab.
2284 */
2285 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002286 if (!min_objects)
2287 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002288 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002289 min_objects = min(min_objects, max_objects);
2290
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002291 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002292 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002293 while (fraction >= 4) {
2294 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002295 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002296 if (order <= slub_max_order)
2297 return order;
2298 fraction /= 2;
2299 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002300 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002301 }
2302
2303 /*
2304 * We were unable to place multiple objects in a slab. Now
2305 * lets see if we can place a single object there.
2306 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002307 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002308 if (order <= slub_max_order)
2309 return order;
2310
2311 /*
2312 * Doh this slab cannot be placed using slub_max_order.
2313 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002314 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002315 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002316 return order;
2317 return -ENOSYS;
2318}
2319
Christoph Lameter81819f02007-05-06 14:49:36 -07002320/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002321 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002322 */
2323static unsigned long calculate_alignment(unsigned long flags,
2324 unsigned long align, unsigned long size)
2325{
2326 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002327 * If the user wants hardware cache aligned objects then follow that
2328 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002329 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002330 * The hardware cache alignment cannot override the specified
2331 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002332 */
Nick Pigginb6210382008-03-05 14:05:56 -08002333 if (flags & SLAB_HWCACHE_ALIGN) {
2334 unsigned long ralign = cache_line_size();
2335 while (size <= ralign / 2)
2336 ralign /= 2;
2337 align = max(align, ralign);
2338 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002339
2340 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002341 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002342
2343 return ALIGN(align, sizeof(void *));
2344}
2345
Pekka Enberg5595cff2008-08-05 09:28:47 +03002346static void
2347init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002348{
2349 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002350 spin_lock_init(&n->list_lock);
2351 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002352#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002353 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002354 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002355 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002356#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002357}
2358
Christoph Lameter55136592010-08-20 12:37:13 -05002359static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002360{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002361 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2362 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002363
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002364 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002365 * Must align to double word boundary for the double cmpxchg
2366 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002367 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002368 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2369 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002370
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002371 if (!s->cpu_slab)
2372 return 0;
2373
2374 init_kmem_cache_cpus(s);
2375
2376 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002377}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002378
Christoph Lameter51df1142010-08-20 12:37:15 -05002379static struct kmem_cache *kmem_cache_node;
2380
Christoph Lameter81819f02007-05-06 14:49:36 -07002381/*
2382 * No kmalloc_node yet so do it by hand. We know that this is the first
2383 * slab on the node for this slabcache. There are no concurrent accesses
2384 * possible.
2385 *
2386 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002387 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2388 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002389 */
Christoph Lameter55136592010-08-20 12:37:13 -05002390static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002391{
2392 struct page *page;
2393 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002394 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002395
Christoph Lameter51df1142010-08-20 12:37:15 -05002396 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002397
Christoph Lameter51df1142010-08-20 12:37:15 -05002398 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002399
2400 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002401 if (page_to_nid(page) != node) {
2402 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2403 "node %d\n", node);
2404 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2405 "in order to be able to continue\n");
2406 }
2407
Christoph Lameter81819f02007-05-06 14:49:36 -07002408 n = page->freelist;
2409 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002410 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002411 page->inuse++;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002412 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002413 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002414#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002415 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002416 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002417#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002418 init_kmem_cache_node(n, kmem_cache_node);
2419 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002420
rootba84c732008-01-07 23:20:28 -08002421 /*
2422 * lockdep requires consistent irq usage for each lock
2423 * so even though there cannot be a race this early in
2424 * the boot sequence, we still disable irqs.
2425 */
2426 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002427 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002428 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002429}
2430
2431static void free_kmem_cache_nodes(struct kmem_cache *s)
2432{
2433 int node;
2434
Christoph Lameterf64dc582007-10-16 01:25:33 -07002435 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002436 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002437
Alexander Duyck73367bd2010-05-21 14:41:35 -07002438 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002439 kmem_cache_free(kmem_cache_node, n);
2440
Christoph Lameter81819f02007-05-06 14:49:36 -07002441 s->node[node] = NULL;
2442 }
2443}
2444
Christoph Lameter55136592010-08-20 12:37:13 -05002445static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002446{
2447 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002448
Christoph Lameterf64dc582007-10-16 01:25:33 -07002449 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002450 struct kmem_cache_node *n;
2451
Alexander Duyck73367bd2010-05-21 14:41:35 -07002452 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002453 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002454 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002455 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002456 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002457 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002458
2459 if (!n) {
2460 free_kmem_cache_nodes(s);
2461 return 0;
2462 }
2463
Christoph Lameter81819f02007-05-06 14:49:36 -07002464 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002465 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002466 }
2467 return 1;
2468}
Christoph Lameter81819f02007-05-06 14:49:36 -07002469
David Rientjesc0bdb232009-02-25 09:16:35 +02002470static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002471{
2472 if (min < MIN_PARTIAL)
2473 min = MIN_PARTIAL;
2474 else if (min > MAX_PARTIAL)
2475 min = MAX_PARTIAL;
2476 s->min_partial = min;
2477}
2478
Christoph Lameter81819f02007-05-06 14:49:36 -07002479/*
2480 * calculate_sizes() determines the order and the distribution of data within
2481 * a slab object.
2482 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002483static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002484{
2485 unsigned long flags = s->flags;
2486 unsigned long size = s->objsize;
2487 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002488 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002489
2490 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002491 * Round up object size to the next word boundary. We can only
2492 * place the free pointer at word boundaries and this determines
2493 * the possible location of the free pointer.
2494 */
2495 size = ALIGN(size, sizeof(void *));
2496
2497#ifdef CONFIG_SLUB_DEBUG
2498 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002499 * Determine if we can poison the object itself. If the user of
2500 * the slab may touch the object after free or before allocation
2501 * then we should never poison the object itself.
2502 */
2503 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002504 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002505 s->flags |= __OBJECT_POISON;
2506 else
2507 s->flags &= ~__OBJECT_POISON;
2508
Christoph Lameter81819f02007-05-06 14:49:36 -07002509
2510 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002511 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002512 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002513 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002514 */
2515 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2516 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002517#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002518
2519 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002520 * With that we have determined the number of bytes in actual use
2521 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002522 */
2523 s->inuse = size;
2524
2525 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002526 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002527 /*
2528 * Relocate free pointer after the object if it is not
2529 * permitted to overwrite the first word of the object on
2530 * kmem_cache_free.
2531 *
2532 * This is the case if we do RCU, have a constructor or
2533 * destructor or are poisoning the objects.
2534 */
2535 s->offset = size;
2536 size += sizeof(void *);
2537 }
2538
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002539#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002540 if (flags & SLAB_STORE_USER)
2541 /*
2542 * Need to store information about allocs and frees after
2543 * the object.
2544 */
2545 size += 2 * sizeof(struct track);
2546
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002547 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002548 /*
2549 * Add some empty padding so that we can catch
2550 * overwrites from earlier objects rather than let
2551 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002552 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002553 * of the object.
2554 */
2555 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002556#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002557
Christoph Lameter81819f02007-05-06 14:49:36 -07002558 /*
2559 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002560 * user specified and the dynamic determination of cache line size
2561 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002562 */
2563 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002564 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002565
2566 /*
2567 * SLUB stores one object immediately after another beginning from
2568 * offset 0. In order to align the objects we have to simply size
2569 * each object to conform to the alignment.
2570 */
2571 size = ALIGN(size, align);
2572 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002573 if (forced_order >= 0)
2574 order = forced_order;
2575 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002576 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002577
Christoph Lameter834f3d12008-04-14 19:11:31 +03002578 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002579 return 0;
2580
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002581 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002582 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002583 s->allocflags |= __GFP_COMP;
2584
2585 if (s->flags & SLAB_CACHE_DMA)
2586 s->allocflags |= SLUB_DMA;
2587
2588 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2589 s->allocflags |= __GFP_RECLAIMABLE;
2590
Christoph Lameter81819f02007-05-06 14:49:36 -07002591 /*
2592 * Determine the number of objects per slab
2593 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002594 s->oo = oo_make(order, size, s->reserved);
2595 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002596 if (oo_objects(s->oo) > oo_objects(s->max))
2597 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002598
Christoph Lameter834f3d12008-04-14 19:11:31 +03002599 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002600
2601}
2602
Christoph Lameter55136592010-08-20 12:37:13 -05002603static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002604 const char *name, size_t size,
2605 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002606 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002607{
2608 memset(s, 0, kmem_size);
2609 s->name = name;
2610 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002611 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002612 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002613 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002614 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002615
Lai Jiangshanda9a6382011-03-10 15:22:00 +08002616 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2617 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07002618
Christoph Lameter06b285d2008-04-14 19:11:41 +03002619 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002620 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002621 if (disable_higher_order_debug) {
2622 /*
2623 * Disable debugging flags that store metadata if the min slab
2624 * order increased.
2625 */
2626 if (get_order(s->size) > get_order(s->objsize)) {
2627 s->flags &= ~DEBUG_METADATA_FLAGS;
2628 s->offset = 0;
2629 if (!calculate_sizes(s, -1))
2630 goto error;
2631 }
2632 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002633
Christoph Lameterb789ef52011-06-01 12:25:49 -05002634#ifdef CONFIG_CMPXCHG_DOUBLE
2635 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
2636 /* Enable fast mode */
2637 s->flags |= __CMPXCHG_DOUBLE;
2638#endif
2639
David Rientjes3b89d7d2009-02-22 17:40:07 -08002640 /*
2641 * The larger the object size is, the more pages we want on the partial
2642 * list to avoid pounding the page allocator excessively.
2643 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002644 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002645 s->refcount = 1;
2646#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002647 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002648#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002649 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002650 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002651
Christoph Lameter55136592010-08-20 12:37:13 -05002652 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002653 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002654
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002655 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002656error:
2657 if (flags & SLAB_PANIC)
2658 panic("Cannot create slab %s size=%lu realsize=%u "
2659 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002660 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002661 s->offset, flags);
2662 return 0;
2663}
Christoph Lameter81819f02007-05-06 14:49:36 -07002664
2665/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002666 * Determine the size of a slab object
2667 */
2668unsigned int kmem_cache_size(struct kmem_cache *s)
2669{
2670 return s->objsize;
2671}
2672EXPORT_SYMBOL(kmem_cache_size);
2673
Christoph Lameter33b12c32008-04-25 12:22:43 -07002674static void list_slab_objects(struct kmem_cache *s, struct page *page,
2675 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002676{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002677#ifdef CONFIG_SLUB_DEBUG
2678 void *addr = page_address(page);
2679 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002680 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2681 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002682 if (!map)
2683 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002684 slab_err(s, page, "%s", text);
2685 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002686
Christoph Lameter5f80b132011-04-15 14:48:13 -05002687 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002688 for_each_object(p, s, addr, page->objects) {
2689
2690 if (!test_bit(slab_index(p, s, addr), map)) {
2691 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2692 p, p - addr);
2693 print_tracking(s, p);
2694 }
2695 }
2696 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002697 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002698#endif
2699}
2700
Christoph Lameter81819f02007-05-06 14:49:36 -07002701/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002702 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002703 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002704static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002705{
Christoph Lameter81819f02007-05-06 14:49:36 -07002706 unsigned long flags;
2707 struct page *page, *h;
2708
2709 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002710 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002711 if (!page->inuse) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05002712 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002713 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002714 } else {
2715 list_slab_objects(s, page,
2716 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002717 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002718 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002719 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002720}
2721
2722/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002723 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002724 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002725static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002726{
2727 int node;
2728
2729 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002730 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002731 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002732 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002733 struct kmem_cache_node *n = get_node(s, node);
2734
Christoph Lameter599870b2008-04-23 12:36:52 -07002735 free_partial(s, n);
2736 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002737 return 1;
2738 }
2739 free_kmem_cache_nodes(s);
2740 return 0;
2741}
2742
2743/*
2744 * Close a cache and release the kmem_cache structure
2745 * (must be used for caches created using kmem_cache_create)
2746 */
2747void kmem_cache_destroy(struct kmem_cache *s)
2748{
2749 down_write(&slub_lock);
2750 s->refcount--;
2751 if (!s->refcount) {
2752 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002753 if (kmem_cache_close(s)) {
2754 printk(KERN_ERR "SLUB %s: %s called for cache that "
2755 "still has objects.\n", s->name, __func__);
2756 dump_stack();
2757 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002758 if (s->flags & SLAB_DESTROY_BY_RCU)
2759 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002760 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002761 }
2762 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002763}
2764EXPORT_SYMBOL(kmem_cache_destroy);
2765
2766/********************************************************************
2767 * Kmalloc subsystem
2768 *******************************************************************/
2769
Christoph Lameter51df1142010-08-20 12:37:15 -05002770struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002771EXPORT_SYMBOL(kmalloc_caches);
2772
Christoph Lameter51df1142010-08-20 12:37:15 -05002773static struct kmem_cache *kmem_cache;
2774
Christoph Lameter55136592010-08-20 12:37:13 -05002775#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002776static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002777#endif
2778
Christoph Lameter81819f02007-05-06 14:49:36 -07002779static int __init setup_slub_min_order(char *str)
2780{
Pekka Enberg06428782008-01-07 23:20:27 -08002781 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002782
2783 return 1;
2784}
2785
2786__setup("slub_min_order=", setup_slub_min_order);
2787
2788static int __init setup_slub_max_order(char *str)
2789{
Pekka Enberg06428782008-01-07 23:20:27 -08002790 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002791 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002792
2793 return 1;
2794}
2795
2796__setup("slub_max_order=", setup_slub_max_order);
2797
2798static int __init setup_slub_min_objects(char *str)
2799{
Pekka Enberg06428782008-01-07 23:20:27 -08002800 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002801
2802 return 1;
2803}
2804
2805__setup("slub_min_objects=", setup_slub_min_objects);
2806
2807static int __init setup_slub_nomerge(char *str)
2808{
2809 slub_nomerge = 1;
2810 return 1;
2811}
2812
2813__setup("slub_nomerge", setup_slub_nomerge);
2814
Christoph Lameter51df1142010-08-20 12:37:15 -05002815static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2816 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002817{
Christoph Lameter51df1142010-08-20 12:37:15 -05002818 struct kmem_cache *s;
2819
2820 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2821
Pekka Enberg83b519e2009-06-10 19:40:04 +03002822 /*
2823 * This function is called with IRQs disabled during early-boot on
2824 * single CPU so there's no need to take slub_lock here.
2825 */
Christoph Lameter55136592010-08-20 12:37:13 -05002826 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002827 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002828 goto panic;
2829
2830 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002831 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002832
2833panic:
2834 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002835 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002836}
2837
Christoph Lameterf1b26332007-07-17 04:03:26 -07002838/*
2839 * Conversion table for small slabs sizes / 8 to the index in the
2840 * kmalloc array. This is necessary for slabs < 192 since we have non power
2841 * of two cache sizes there. The size of larger slabs can be determined using
2842 * fls.
2843 */
2844static s8 size_index[24] = {
2845 3, /* 8 */
2846 4, /* 16 */
2847 5, /* 24 */
2848 5, /* 32 */
2849 6, /* 40 */
2850 6, /* 48 */
2851 6, /* 56 */
2852 6, /* 64 */
2853 1, /* 72 */
2854 1, /* 80 */
2855 1, /* 88 */
2856 1, /* 96 */
2857 7, /* 104 */
2858 7, /* 112 */
2859 7, /* 120 */
2860 7, /* 128 */
2861 2, /* 136 */
2862 2, /* 144 */
2863 2, /* 152 */
2864 2, /* 160 */
2865 2, /* 168 */
2866 2, /* 176 */
2867 2, /* 184 */
2868 2 /* 192 */
2869};
2870
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002871static inline int size_index_elem(size_t bytes)
2872{
2873 return (bytes - 1) / 8;
2874}
2875
Christoph Lameter81819f02007-05-06 14:49:36 -07002876static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2877{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002878 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002879
Christoph Lameterf1b26332007-07-17 04:03:26 -07002880 if (size <= 192) {
2881 if (!size)
2882 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002883
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002884 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002885 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002886 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002887
2888#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002889 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002890 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002891
Christoph Lameter81819f02007-05-06 14:49:36 -07002892#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002893 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002894}
2895
2896void *__kmalloc(size_t size, gfp_t flags)
2897{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002898 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002899 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002900
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002901 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002902 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002903
2904 s = get_slab(size, flags);
2905
2906 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002907 return s;
2908
Christoph Lameter2154a332010-07-09 14:07:10 -05002909 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002910
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002911 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002912
2913 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002914}
2915EXPORT_SYMBOL(__kmalloc);
2916
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002917#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002918static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2919{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002920 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002921 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002922
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002923 flags |= __GFP_COMP | __GFP_NOTRACK;
2924 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002925 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002926 ptr = page_address(page);
2927
2928 kmemleak_alloc(ptr, size, 1, flags);
2929 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002930}
2931
Christoph Lameter81819f02007-05-06 14:49:36 -07002932void *__kmalloc_node(size_t size, gfp_t flags, int node)
2933{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002934 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002935 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002936
Ingo Molnar057685c2009-02-20 12:15:30 +01002937 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002938 ret = kmalloc_large_node(size, flags, node);
2939
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002940 trace_kmalloc_node(_RET_IP_, ret,
2941 size, PAGE_SIZE << get_order(size),
2942 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002943
2944 return ret;
2945 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002946
2947 s = get_slab(size, flags);
2948
2949 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002950 return s;
2951
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002952 ret = slab_alloc(s, flags, node, _RET_IP_);
2953
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002954 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002955
2956 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002957}
2958EXPORT_SYMBOL(__kmalloc_node);
2959#endif
2960
2961size_t ksize(const void *object)
2962{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002963 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002964
Christoph Lameteref8b4522007-10-16 01:24:46 -07002965 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002966 return 0;
2967
Vegard Nossum294a80a2007-12-04 23:45:30 -08002968 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002969
Pekka Enberg76994412008-05-22 19:22:25 +03002970 if (unlikely(!PageSlab(page))) {
2971 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002972 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002973 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002974
Eric Dumazetb3d41882011-02-14 18:35:22 +01002975 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002976}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002977EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002978
2979void kfree(const void *x)
2980{
Christoph Lameter81819f02007-05-06 14:49:36 -07002981 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002982 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002983
Pekka Enberg2121db72009-03-25 11:05:57 +02002984 trace_kfree(_RET_IP_, x);
2985
Satyam Sharma2408c552007-10-16 01:24:44 -07002986 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002987 return;
2988
Christoph Lameterb49af682007-05-06 14:49:41 -07002989 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002990 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002991 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002992 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002993 put_page(page);
2994 return;
2995 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002996 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002997}
2998EXPORT_SYMBOL(kfree);
2999
Christoph Lameter2086d262007-05-06 14:49:46 -07003000/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003001 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3002 * the remaining slabs by the number of items in use. The slabs with the
3003 * most items in use come first. New allocations will then fill those up
3004 * and thus they can be removed from the partial lists.
3005 *
3006 * The slabs with the least items are placed last. This results in them
3007 * being allocated from last increasing the chance that the last objects
3008 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003009 */
3010int kmem_cache_shrink(struct kmem_cache *s)
3011{
3012 int node;
3013 int i;
3014 struct kmem_cache_node *n;
3015 struct page *page;
3016 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003017 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003018 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003019 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003020 unsigned long flags;
3021
3022 if (!slabs_by_inuse)
3023 return -ENOMEM;
3024
3025 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003026 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003027 n = get_node(s, node);
3028
3029 if (!n->nr_partial)
3030 continue;
3031
Christoph Lameter834f3d12008-04-14 19:11:31 +03003032 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003033 INIT_LIST_HEAD(slabs_by_inuse + i);
3034
3035 spin_lock_irqsave(&n->list_lock, flags);
3036
3037 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003038 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003039 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003040 * Note that concurrent frees may occur while we hold the
3041 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003042 */
3043 list_for_each_entry_safe(page, t, &n->partial, lru) {
3044 if (!page->inuse && slab_trylock(page)) {
3045 /*
3046 * Must hold slab lock here because slab_free
3047 * may have freed the last object and be
3048 * waiting to release the slab.
3049 */
Christoph Lameter62e346a2010-09-28 08:10:28 -05003050 __remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003051 slab_unlock(page);
3052 discard_slab(s, page);
3053 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07003054 list_move(&page->lru,
3055 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07003056 }
3057 }
3058
Christoph Lameter2086d262007-05-06 14:49:46 -07003059 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003060 * Rebuild the partial list with the slabs filled up most
3061 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003062 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003063 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003064 list_splice(slabs_by_inuse + i, n->partial.prev);
3065
Christoph Lameter2086d262007-05-06 14:49:46 -07003066 spin_unlock_irqrestore(&n->list_lock, flags);
3067 }
3068
3069 kfree(slabs_by_inuse);
3070 return 0;
3071}
3072EXPORT_SYMBOL(kmem_cache_shrink);
3073
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003074#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003075static int slab_mem_going_offline_callback(void *arg)
3076{
3077 struct kmem_cache *s;
3078
3079 down_read(&slub_lock);
3080 list_for_each_entry(s, &slab_caches, list)
3081 kmem_cache_shrink(s);
3082 up_read(&slub_lock);
3083
3084 return 0;
3085}
3086
3087static void slab_mem_offline_callback(void *arg)
3088{
3089 struct kmem_cache_node *n;
3090 struct kmem_cache *s;
3091 struct memory_notify *marg = arg;
3092 int offline_node;
3093
3094 offline_node = marg->status_change_nid;
3095
3096 /*
3097 * If the node still has available memory. we need kmem_cache_node
3098 * for it yet.
3099 */
3100 if (offline_node < 0)
3101 return;
3102
3103 down_read(&slub_lock);
3104 list_for_each_entry(s, &slab_caches, list) {
3105 n = get_node(s, offline_node);
3106 if (n) {
3107 /*
3108 * if n->nr_slabs > 0, slabs still exist on the node
3109 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003110 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003111 * callback. So, we must fail.
3112 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003113 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003114
3115 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003116 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003117 }
3118 }
3119 up_read(&slub_lock);
3120}
3121
3122static int slab_mem_going_online_callback(void *arg)
3123{
3124 struct kmem_cache_node *n;
3125 struct kmem_cache *s;
3126 struct memory_notify *marg = arg;
3127 int nid = marg->status_change_nid;
3128 int ret = 0;
3129
3130 /*
3131 * If the node's memory is already available, then kmem_cache_node is
3132 * already created. Nothing to do.
3133 */
3134 if (nid < 0)
3135 return 0;
3136
3137 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003138 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003139 * allocate a kmem_cache_node structure in order to bring the node
3140 * online.
3141 */
3142 down_read(&slub_lock);
3143 list_for_each_entry(s, &slab_caches, list) {
3144 /*
3145 * XXX: kmem_cache_alloc_node will fallback to other nodes
3146 * since memory is not yet available from the node that
3147 * is brought up.
3148 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003149 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003150 if (!n) {
3151 ret = -ENOMEM;
3152 goto out;
3153 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003154 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003155 s->node[nid] = n;
3156 }
3157out:
3158 up_read(&slub_lock);
3159 return ret;
3160}
3161
3162static int slab_memory_callback(struct notifier_block *self,
3163 unsigned long action, void *arg)
3164{
3165 int ret = 0;
3166
3167 switch (action) {
3168 case MEM_GOING_ONLINE:
3169 ret = slab_mem_going_online_callback(arg);
3170 break;
3171 case MEM_GOING_OFFLINE:
3172 ret = slab_mem_going_offline_callback(arg);
3173 break;
3174 case MEM_OFFLINE:
3175 case MEM_CANCEL_ONLINE:
3176 slab_mem_offline_callback(arg);
3177 break;
3178 case MEM_ONLINE:
3179 case MEM_CANCEL_OFFLINE:
3180 break;
3181 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003182 if (ret)
3183 ret = notifier_from_errno(ret);
3184 else
3185 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003186 return ret;
3187}
3188
3189#endif /* CONFIG_MEMORY_HOTPLUG */
3190
Christoph Lameter81819f02007-05-06 14:49:36 -07003191/********************************************************************
3192 * Basic setup of slabs
3193 *******************************************************************/
3194
Christoph Lameter51df1142010-08-20 12:37:15 -05003195/*
3196 * Used for early kmem_cache structures that were allocated using
3197 * the page allocator
3198 */
3199
3200static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3201{
3202 int node;
3203
3204 list_add(&s->list, &slab_caches);
3205 s->refcount = -1;
3206
3207 for_each_node_state(node, N_NORMAL_MEMORY) {
3208 struct kmem_cache_node *n = get_node(s, node);
3209 struct page *p;
3210
3211 if (n) {
3212 list_for_each_entry(p, &n->partial, lru)
3213 p->slab = s;
3214
Li Zefan607bf322011-04-12 15:22:26 +08003215#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003216 list_for_each_entry(p, &n->full, lru)
3217 p->slab = s;
3218#endif
3219 }
3220 }
3221}
3222
Christoph Lameter81819f02007-05-06 14:49:36 -07003223void __init kmem_cache_init(void)
3224{
3225 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003226 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003227 struct kmem_cache *temp_kmem_cache;
3228 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003229 struct kmem_cache *temp_kmem_cache_node;
3230 unsigned long kmalloc_size;
3231
3232 kmem_size = offsetof(struct kmem_cache, node) +
3233 nr_node_ids * sizeof(struct kmem_cache_node *);
3234
3235 /* Allocate two kmem_caches from the page allocator */
3236 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3237 order = get_order(2 * kmalloc_size);
3238 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3239
Christoph Lameter81819f02007-05-06 14:49:36 -07003240 /*
3241 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003242 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003243 * kmem_cache_open for slab_state == DOWN.
3244 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003245 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3246
3247 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3248 sizeof(struct kmem_cache_node),
3249 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003250
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003251 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003252
3253 /* Able to allocate the per node structures */
3254 slab_state = PARTIAL;
3255
Christoph Lameter51df1142010-08-20 12:37:15 -05003256 temp_kmem_cache = kmem_cache;
3257 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3258 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3259 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3260 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003261
Christoph Lameter51df1142010-08-20 12:37:15 -05003262 /*
3263 * Allocate kmem_cache_node properly from the kmem_cache slab.
3264 * kmem_cache_node is separately allocated so no need to
3265 * update any list pointers.
3266 */
3267 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003268
Christoph Lameter51df1142010-08-20 12:37:15 -05003269 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3270 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3271
3272 kmem_cache_bootstrap_fixup(kmem_cache_node);
3273
3274 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003275 kmem_cache_bootstrap_fixup(kmem_cache);
3276 caches++;
3277 /* Free temporary boot structure */
3278 free_pages((unsigned long)temp_kmem_cache, order);
3279
3280 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003281
3282 /*
3283 * Patch up the size_index table if we have strange large alignment
3284 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003285 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003286 *
3287 * Largest permitted alignment is 256 bytes due to the way we
3288 * handle the index determination for the smaller caches.
3289 *
3290 * Make sure that nothing crazy happens if someone starts tinkering
3291 * around with ARCH_KMALLOC_MINALIGN
3292 */
3293 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3294 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3295
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003296 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3297 int elem = size_index_elem(i);
3298 if (elem >= ARRAY_SIZE(size_index))
3299 break;
3300 size_index[elem] = KMALLOC_SHIFT_LOW;
3301 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003302
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003303 if (KMALLOC_MIN_SIZE == 64) {
3304 /*
3305 * The 96 byte size cache is not used if the alignment
3306 * is 64 byte.
3307 */
3308 for (i = 64 + 8; i <= 96; i += 8)
3309 size_index[size_index_elem(i)] = 7;
3310 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003311 /*
3312 * The 192 byte sized cache is not used if the alignment
3313 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3314 * instead.
3315 */
3316 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003317 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003318 }
3319
Christoph Lameter51df1142010-08-20 12:37:15 -05003320 /* Caches that are not of the two-to-the-power-of size */
3321 if (KMALLOC_MIN_SIZE <= 32) {
3322 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3323 caches++;
3324 }
3325
3326 if (KMALLOC_MIN_SIZE <= 64) {
3327 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3328 caches++;
3329 }
3330
3331 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3332 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3333 caches++;
3334 }
3335
Christoph Lameter81819f02007-05-06 14:49:36 -07003336 slab_state = UP;
3337
3338 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003339 if (KMALLOC_MIN_SIZE <= 32) {
3340 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3341 BUG_ON(!kmalloc_caches[1]->name);
3342 }
3343
3344 if (KMALLOC_MIN_SIZE <= 64) {
3345 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3346 BUG_ON(!kmalloc_caches[2]->name);
3347 }
3348
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003349 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3350 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3351
3352 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003353 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003354 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003355
3356#ifdef CONFIG_SMP
3357 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003358#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003359
Christoph Lameter55136592010-08-20 12:37:13 -05003360#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003361 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3362 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003363
Christoph Lameter51df1142010-08-20 12:37:15 -05003364 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003365 char *name = kasprintf(GFP_NOWAIT,
3366 "dma-kmalloc-%d", s->objsize);
3367
3368 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003369 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3370 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003371 }
3372 }
3373#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003374 printk(KERN_INFO
3375 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003376 " CPUs=%d, Nodes=%d\n",
3377 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003378 slub_min_order, slub_max_order, slub_min_objects,
3379 nr_cpu_ids, nr_node_ids);
3380}
3381
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003382void __init kmem_cache_init_late(void)
3383{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003384}
3385
Christoph Lameter81819f02007-05-06 14:49:36 -07003386/*
3387 * Find a mergeable slab cache
3388 */
3389static int slab_unmergeable(struct kmem_cache *s)
3390{
3391 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3392 return 1;
3393
Christoph Lameterc59def92007-05-16 22:10:50 -07003394 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003395 return 1;
3396
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003397 /*
3398 * We may have set a slab to be unmergeable during bootstrap.
3399 */
3400 if (s->refcount < 0)
3401 return 1;
3402
Christoph Lameter81819f02007-05-06 14:49:36 -07003403 return 0;
3404}
3405
3406static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003407 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003408 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003409{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003410 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003411
3412 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3413 return NULL;
3414
Christoph Lameterc59def92007-05-16 22:10:50 -07003415 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003416 return NULL;
3417
3418 size = ALIGN(size, sizeof(void *));
3419 align = calculate_alignment(flags, align, size);
3420 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003421 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003422
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003423 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003424 if (slab_unmergeable(s))
3425 continue;
3426
3427 if (size > s->size)
3428 continue;
3429
Christoph Lameterba0268a2007-09-11 15:24:11 -07003430 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003431 continue;
3432 /*
3433 * Check if alignment is compatible.
3434 * Courtesy of Adrian Drzewiecki
3435 */
Pekka Enberg06428782008-01-07 23:20:27 -08003436 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003437 continue;
3438
3439 if (s->size - size >= sizeof(void *))
3440 continue;
3441
3442 return s;
3443 }
3444 return NULL;
3445}
3446
3447struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003448 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003449{
3450 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003451 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003452
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003453 if (WARN_ON(!name))
3454 return NULL;
3455
Christoph Lameter81819f02007-05-06 14:49:36 -07003456 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003457 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003458 if (s) {
3459 s->refcount++;
3460 /*
3461 * Adjust the object sizes so that we clear
3462 * the complete object on kzalloc.
3463 */
3464 s->objsize = max(s->objsize, (int)size);
3465 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003466
David Rientjes7b8f3b62008-12-17 22:09:46 -08003467 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003468 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003469 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003470 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003471 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003472 return s;
3473 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003474
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003475 n = kstrdup(name, GFP_KERNEL);
3476 if (!n)
3477 goto err;
3478
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003479 s = kmalloc(kmem_size, GFP_KERNEL);
3480 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003481 if (kmem_cache_open(s, n,
Christoph Lameterc59def92007-05-16 22:10:50 -07003482 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003483 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003484 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003485 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003486 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003487 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003488 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003489 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003490 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003491 return s;
3492 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003493 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003494 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003495 }
Pavel Emelyanov68cee4f12010-10-28 13:50:37 +04003496err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003497 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003498
Christoph Lameter81819f02007-05-06 14:49:36 -07003499 if (flags & SLAB_PANIC)
3500 panic("Cannot create slabcache %s\n", name);
3501 else
3502 s = NULL;
3503 return s;
3504}
3505EXPORT_SYMBOL(kmem_cache_create);
3506
Christoph Lameter81819f02007-05-06 14:49:36 -07003507#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003508/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003509 * Use the cpu notifier to insure that the cpu slabs are flushed when
3510 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003511 */
3512static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3513 unsigned long action, void *hcpu)
3514{
3515 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003516 struct kmem_cache *s;
3517 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003518
3519 switch (action) {
3520 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003521 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003522 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003523 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003524 down_read(&slub_lock);
3525 list_for_each_entry(s, &slab_caches, list) {
3526 local_irq_save(flags);
3527 __flush_cpu_slab(s, cpu);
3528 local_irq_restore(flags);
3529 }
3530 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003531 break;
3532 default:
3533 break;
3534 }
3535 return NOTIFY_OK;
3536}
3537
Pekka Enberg06428782008-01-07 23:20:27 -08003538static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003539 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003540};
Christoph Lameter81819f02007-05-06 14:49:36 -07003541
3542#endif
3543
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003544void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003545{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003546 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003547 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003548
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003549 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003550 return kmalloc_large(size, gfpflags);
3551
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003552 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003553
Satyam Sharma2408c552007-10-16 01:24:44 -07003554 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003555 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003556
Christoph Lameter2154a332010-07-09 14:07:10 -05003557 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003558
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003559 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003560 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003561
3562 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003563}
3564
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003565#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003566void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003567 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003568{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003569 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003570 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003571
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003572 if (unlikely(size > SLUB_MAX_SIZE)) {
3573 ret = kmalloc_large_node(size, gfpflags, node);
3574
3575 trace_kmalloc_node(caller, ret,
3576 size, PAGE_SIZE << get_order(size),
3577 gfpflags, node);
3578
3579 return ret;
3580 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003581
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003582 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003583
Satyam Sharma2408c552007-10-16 01:24:44 -07003584 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003585 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003586
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003587 ret = slab_alloc(s, gfpflags, node, caller);
3588
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003589 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003590 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003591
3592 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003593}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003594#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003595
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003596#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003597static int count_inuse(struct page *page)
3598{
3599 return page->inuse;
3600}
3601
3602static int count_total(struct page *page)
3603{
3604 return page->objects;
3605}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003606#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003607
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003608#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003609static int validate_slab(struct kmem_cache *s, struct page *page,
3610 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003611{
3612 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003613 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003614
3615 if (!check_slab(s, page) ||
3616 !on_freelist(s, page, NULL))
3617 return 0;
3618
3619 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003620 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003621
Christoph Lameter5f80b132011-04-15 14:48:13 -05003622 get_map(s, page, map);
3623 for_each_object(p, s, addr, page->objects) {
3624 if (test_bit(slab_index(p, s, addr), map))
3625 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3626 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003627 }
3628
Christoph Lameter224a88b2008-04-14 19:11:31 +03003629 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003630 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003631 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003632 return 0;
3633 return 1;
3634}
3635
Christoph Lameter434e2452007-07-17 04:03:30 -07003636static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3637 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003638{
3639 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003640 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003641 slab_unlock(page);
3642 } else
3643 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3644 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003645}
3646
Christoph Lameter434e2452007-07-17 04:03:30 -07003647static int validate_slab_node(struct kmem_cache *s,
3648 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003649{
3650 unsigned long count = 0;
3651 struct page *page;
3652 unsigned long flags;
3653
3654 spin_lock_irqsave(&n->list_lock, flags);
3655
3656 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003657 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003658 count++;
3659 }
3660 if (count != n->nr_partial)
3661 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3662 "counter=%ld\n", s->name, count, n->nr_partial);
3663
3664 if (!(s->flags & SLAB_STORE_USER))
3665 goto out;
3666
3667 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003668 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003669 count++;
3670 }
3671 if (count != atomic_long_read(&n->nr_slabs))
3672 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3673 "counter=%ld\n", s->name, count,
3674 atomic_long_read(&n->nr_slabs));
3675
3676out:
3677 spin_unlock_irqrestore(&n->list_lock, flags);
3678 return count;
3679}
3680
Christoph Lameter434e2452007-07-17 04:03:30 -07003681static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003682{
3683 int node;
3684 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003685 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003686 sizeof(unsigned long), GFP_KERNEL);
3687
3688 if (!map)
3689 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003690
3691 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003692 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003693 struct kmem_cache_node *n = get_node(s, node);
3694
Christoph Lameter434e2452007-07-17 04:03:30 -07003695 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003696 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003697 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003698 return count;
3699}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003700/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003701 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003702 * and freed.
3703 */
3704
3705struct location {
3706 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003707 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003708 long long sum_time;
3709 long min_time;
3710 long max_time;
3711 long min_pid;
3712 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303713 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003714 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003715};
3716
3717struct loc_track {
3718 unsigned long max;
3719 unsigned long count;
3720 struct location *loc;
3721};
3722
3723static void free_loc_track(struct loc_track *t)
3724{
3725 if (t->max)
3726 free_pages((unsigned long)t->loc,
3727 get_order(sizeof(struct location) * t->max));
3728}
3729
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003730static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003731{
3732 struct location *l;
3733 int order;
3734
Christoph Lameter88a420e2007-05-06 14:49:45 -07003735 order = get_order(sizeof(struct location) * max);
3736
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003737 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003738 if (!l)
3739 return 0;
3740
3741 if (t->count) {
3742 memcpy(l, t->loc, sizeof(struct location) * t->count);
3743 free_loc_track(t);
3744 }
3745 t->max = max;
3746 t->loc = l;
3747 return 1;
3748}
3749
3750static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003751 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003752{
3753 long start, end, pos;
3754 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003755 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003756 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003757
3758 start = -1;
3759 end = t->count;
3760
3761 for ( ; ; ) {
3762 pos = start + (end - start + 1) / 2;
3763
3764 /*
3765 * There is nothing at "end". If we end up there
3766 * we need to add something to before end.
3767 */
3768 if (pos == end)
3769 break;
3770
3771 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003772 if (track->addr == caddr) {
3773
3774 l = &t->loc[pos];
3775 l->count++;
3776 if (track->when) {
3777 l->sum_time += age;
3778 if (age < l->min_time)
3779 l->min_time = age;
3780 if (age > l->max_time)
3781 l->max_time = age;
3782
3783 if (track->pid < l->min_pid)
3784 l->min_pid = track->pid;
3785 if (track->pid > l->max_pid)
3786 l->max_pid = track->pid;
3787
Rusty Russell174596a2009-01-01 10:12:29 +10303788 cpumask_set_cpu(track->cpu,
3789 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003790 }
3791 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003792 return 1;
3793 }
3794
Christoph Lameter45edfa52007-05-09 02:32:45 -07003795 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003796 end = pos;
3797 else
3798 start = pos;
3799 }
3800
3801 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003802 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003803 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003804 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003805 return 0;
3806
3807 l = t->loc + pos;
3808 if (pos < t->count)
3809 memmove(l + 1, l,
3810 (t->count - pos) * sizeof(struct location));
3811 t->count++;
3812 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003813 l->addr = track->addr;
3814 l->sum_time = age;
3815 l->min_time = age;
3816 l->max_time = age;
3817 l->min_pid = track->pid;
3818 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303819 cpumask_clear(to_cpumask(l->cpus));
3820 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003821 nodes_clear(l->nodes);
3822 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003823 return 1;
3824}
3825
3826static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003827 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003828 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003829{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003830 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003831 void *p;
3832
Christoph Lameter39b26462008-04-14 19:11:30 +03003833 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05003834 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003835
Christoph Lameter224a88b2008-04-14 19:11:31 +03003836 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003837 if (!test_bit(slab_index(p, s, addr), map))
3838 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003839}
3840
3841static int list_locations(struct kmem_cache *s, char *buf,
3842 enum track_item alloc)
3843{
Harvey Harrisone374d482008-01-31 15:20:50 -08003844 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003845 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003846 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003847 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003848 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3849 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003850
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003851 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3852 GFP_TEMPORARY)) {
3853 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003854 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003855 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003856 /* Push back cpu slabs */
3857 flush_all(s);
3858
Christoph Lameterf64dc582007-10-16 01:25:33 -07003859 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003860 struct kmem_cache_node *n = get_node(s, node);
3861 unsigned long flags;
3862 struct page *page;
3863
Christoph Lameter9e869432007-08-22 14:01:56 -07003864 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003865 continue;
3866
3867 spin_lock_irqsave(&n->list_lock, flags);
3868 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003869 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003870 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003871 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003872 spin_unlock_irqrestore(&n->list_lock, flags);
3873 }
3874
3875 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003876 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003877
Hugh Dickins9c246242008-12-09 13:14:27 -08003878 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003879 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003880 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003881
3882 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08003883 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003884 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003885 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003886
3887 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003888 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003889 l->min_time,
3890 (long)div_u64(l->sum_time, l->count),
3891 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003892 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003893 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003894 l->min_time);
3895
3896 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003897 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003898 l->min_pid, l->max_pid);
3899 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003900 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003901 l->min_pid);
3902
Rusty Russell174596a2009-01-01 10:12:29 +10303903 if (num_online_cpus() > 1 &&
3904 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003905 len < PAGE_SIZE - 60) {
3906 len += sprintf(buf + len, " cpus=");
3907 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303908 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003909 }
3910
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003911 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003912 len < PAGE_SIZE - 60) {
3913 len += sprintf(buf + len, " nodes=");
3914 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003915 l->nodes);
3916 }
3917
Harvey Harrisone374d482008-01-31 15:20:50 -08003918 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003919 }
3920
3921 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003922 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003923 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003924 len += sprintf(buf, "No data\n");
3925 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003926}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003927#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07003928
Christoph Lametera5a84752010-10-05 13:57:27 -05003929#ifdef SLUB_RESILIENCY_TEST
3930static void resiliency_test(void)
3931{
3932 u8 *p;
3933
3934 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3935
3936 printk(KERN_ERR "SLUB resiliency testing\n");
3937 printk(KERN_ERR "-----------------------\n");
3938 printk(KERN_ERR "A. Corruption after allocation\n");
3939
3940 p = kzalloc(16, GFP_KERNEL);
3941 p[16] = 0x12;
3942 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3943 " 0x12->0x%p\n\n", p + 16);
3944
3945 validate_slab_cache(kmalloc_caches[4]);
3946
3947 /* Hmmm... The next two are dangerous */
3948 p = kzalloc(32, GFP_KERNEL);
3949 p[32 + sizeof(void *)] = 0x34;
3950 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3951 " 0x34 -> -0x%p\n", p);
3952 printk(KERN_ERR
3953 "If allocated object is overwritten then not detectable\n\n");
3954
3955 validate_slab_cache(kmalloc_caches[5]);
3956 p = kzalloc(64, GFP_KERNEL);
3957 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3958 *p = 0x56;
3959 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3960 p);
3961 printk(KERN_ERR
3962 "If allocated object is overwritten then not detectable\n\n");
3963 validate_slab_cache(kmalloc_caches[6]);
3964
3965 printk(KERN_ERR "\nB. Corruption after free\n");
3966 p = kzalloc(128, GFP_KERNEL);
3967 kfree(p);
3968 *p = 0x78;
3969 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3970 validate_slab_cache(kmalloc_caches[7]);
3971
3972 p = kzalloc(256, GFP_KERNEL);
3973 kfree(p);
3974 p[50] = 0x9a;
3975 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3976 p);
3977 validate_slab_cache(kmalloc_caches[8]);
3978
3979 p = kzalloc(512, GFP_KERNEL);
3980 kfree(p);
3981 p[512] = 0xab;
3982 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3983 validate_slab_cache(kmalloc_caches[9]);
3984}
3985#else
3986#ifdef CONFIG_SYSFS
3987static void resiliency_test(void) {};
3988#endif
3989#endif
3990
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003991#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07003992enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003993 SL_ALL, /* All slabs */
3994 SL_PARTIAL, /* Only partially allocated slabs */
3995 SL_CPU, /* Only slabs used for cpu caches */
3996 SL_OBJECTS, /* Determine allocated objects not slabs */
3997 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003998};
3999
Christoph Lameter205ab992008-04-14 19:11:40 +03004000#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004001#define SO_PARTIAL (1 << SL_PARTIAL)
4002#define SO_CPU (1 << SL_CPU)
4003#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004004#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004005
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004006static ssize_t show_slab_objects(struct kmem_cache *s,
4007 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004008{
4009 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004010 int node;
4011 int x;
4012 unsigned long *nodes;
4013 unsigned long *per_cpu;
4014
4015 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004016 if (!nodes)
4017 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004018 per_cpu = nodes + nr_node_ids;
4019
Christoph Lameter205ab992008-04-14 19:11:40 +03004020 if (flags & SO_CPU) {
4021 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004022
Christoph Lameter205ab992008-04-14 19:11:40 +03004023 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004024 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004025
Christoph Lameter205ab992008-04-14 19:11:40 +03004026 if (!c || c->node < 0)
4027 continue;
4028
4029 if (c->page) {
4030 if (flags & SO_TOTAL)
4031 x = c->page->objects;
4032 else if (flags & SO_OBJECTS)
4033 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07004034 else
4035 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03004036
Christoph Lameter81819f02007-05-06 14:49:36 -07004037 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03004038 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004039 }
Christoph Lameter205ab992008-04-14 19:11:40 +03004040 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004041 }
4042 }
4043
Christoph Lameter04d94872011-01-10 10:15:15 -06004044 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004045#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004046 if (flags & SO_ALL) {
4047 for_each_node_state(node, N_NORMAL_MEMORY) {
4048 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004049
Christoph Lameter205ab992008-04-14 19:11:40 +03004050 if (flags & SO_TOTAL)
4051 x = atomic_long_read(&n->total_objects);
4052 else if (flags & SO_OBJECTS)
4053 x = atomic_long_read(&n->total_objects) -
4054 count_partial(n, count_free);
4055
4056 else
4057 x = atomic_long_read(&n->nr_slabs);
4058 total += x;
4059 nodes[node] += x;
4060 }
4061
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004062 } else
4063#endif
4064 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004065 for_each_node_state(node, N_NORMAL_MEMORY) {
4066 struct kmem_cache_node *n = get_node(s, node);
4067
4068 if (flags & SO_TOTAL)
4069 x = count_partial(n, count_total);
4070 else if (flags & SO_OBJECTS)
4071 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004072 else
4073 x = n->nr_partial;
4074 total += x;
4075 nodes[node] += x;
4076 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004077 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004078 x = sprintf(buf, "%lu", total);
4079#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004080 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004081 if (nodes[node])
4082 x += sprintf(buf + x, " N%d=%lu",
4083 node, nodes[node]);
4084#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004085 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004086 kfree(nodes);
4087 return x + sprintf(buf + x, "\n");
4088}
4089
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004090#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004091static int any_slab_objects(struct kmem_cache *s)
4092{
4093 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004094
4095 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004096 struct kmem_cache_node *n = get_node(s, node);
4097
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004098 if (!n)
4099 continue;
4100
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004101 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004102 return 1;
4103 }
4104 return 0;
4105}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004106#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004107
4108#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4109#define to_slab(n) container_of(n, struct kmem_cache, kobj);
4110
4111struct slab_attribute {
4112 struct attribute attr;
4113 ssize_t (*show)(struct kmem_cache *s, char *buf);
4114 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4115};
4116
4117#define SLAB_ATTR_RO(_name) \
4118 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4119
4120#define SLAB_ATTR(_name) \
4121 static struct slab_attribute _name##_attr = \
4122 __ATTR(_name, 0644, _name##_show, _name##_store)
4123
Christoph Lameter81819f02007-05-06 14:49:36 -07004124static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4125{
4126 return sprintf(buf, "%d\n", s->size);
4127}
4128SLAB_ATTR_RO(slab_size);
4129
4130static ssize_t align_show(struct kmem_cache *s, char *buf)
4131{
4132 return sprintf(buf, "%d\n", s->align);
4133}
4134SLAB_ATTR_RO(align);
4135
4136static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4137{
4138 return sprintf(buf, "%d\n", s->objsize);
4139}
4140SLAB_ATTR_RO(object_size);
4141
4142static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4143{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004144 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004145}
4146SLAB_ATTR_RO(objs_per_slab);
4147
Christoph Lameter06b285d2008-04-14 19:11:41 +03004148static ssize_t order_store(struct kmem_cache *s,
4149 const char *buf, size_t length)
4150{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004151 unsigned long order;
4152 int err;
4153
4154 err = strict_strtoul(buf, 10, &order);
4155 if (err)
4156 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004157
4158 if (order > slub_max_order || order < slub_min_order)
4159 return -EINVAL;
4160
4161 calculate_sizes(s, order);
4162 return length;
4163}
4164
Christoph Lameter81819f02007-05-06 14:49:36 -07004165static ssize_t order_show(struct kmem_cache *s, char *buf)
4166{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004167 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004168}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004169SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004170
David Rientjes73d342b2009-02-22 17:40:09 -08004171static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4172{
4173 return sprintf(buf, "%lu\n", s->min_partial);
4174}
4175
4176static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4177 size_t length)
4178{
4179 unsigned long min;
4180 int err;
4181
4182 err = strict_strtoul(buf, 10, &min);
4183 if (err)
4184 return err;
4185
David Rientjesc0bdb232009-02-25 09:16:35 +02004186 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004187 return length;
4188}
4189SLAB_ATTR(min_partial);
4190
Christoph Lameter81819f02007-05-06 14:49:36 -07004191static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4192{
Joe Perches62c70bc2011-01-13 15:45:52 -08004193 if (!s->ctor)
4194 return 0;
4195 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004196}
4197SLAB_ATTR_RO(ctor);
4198
Christoph Lameter81819f02007-05-06 14:49:36 -07004199static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4200{
4201 return sprintf(buf, "%d\n", s->refcount - 1);
4202}
4203SLAB_ATTR_RO(aliases);
4204
Christoph Lameter81819f02007-05-06 14:49:36 -07004205static ssize_t partial_show(struct kmem_cache *s, char *buf)
4206{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004207 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004208}
4209SLAB_ATTR_RO(partial);
4210
4211static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4212{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004213 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004214}
4215SLAB_ATTR_RO(cpu_slabs);
4216
4217static ssize_t objects_show(struct kmem_cache *s, char *buf)
4218{
Christoph Lameter205ab992008-04-14 19:11:40 +03004219 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004220}
4221SLAB_ATTR_RO(objects);
4222
Christoph Lameter205ab992008-04-14 19:11:40 +03004223static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4224{
4225 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4226}
4227SLAB_ATTR_RO(objects_partial);
4228
Christoph Lameter81819f02007-05-06 14:49:36 -07004229static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4230{
4231 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4232}
4233
4234static ssize_t reclaim_account_store(struct kmem_cache *s,
4235 const char *buf, size_t length)
4236{
4237 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4238 if (buf[0] == '1')
4239 s->flags |= SLAB_RECLAIM_ACCOUNT;
4240 return length;
4241}
4242SLAB_ATTR(reclaim_account);
4243
4244static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4245{
Christoph Lameter5af60832007-05-06 14:49:56 -07004246 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004247}
4248SLAB_ATTR_RO(hwcache_align);
4249
4250#ifdef CONFIG_ZONE_DMA
4251static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4252{
4253 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4254}
4255SLAB_ATTR_RO(cache_dma);
4256#endif
4257
4258static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4259{
4260 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4261}
4262SLAB_ATTR_RO(destroy_by_rcu);
4263
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004264static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4265{
4266 return sprintf(buf, "%d\n", s->reserved);
4267}
4268SLAB_ATTR_RO(reserved);
4269
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004270#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004271static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4272{
4273 return show_slab_objects(s, buf, SO_ALL);
4274}
4275SLAB_ATTR_RO(slabs);
4276
4277static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4278{
4279 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4280}
4281SLAB_ATTR_RO(total_objects);
4282
4283static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4284{
4285 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4286}
4287
4288static ssize_t sanity_checks_store(struct kmem_cache *s,
4289 const char *buf, size_t length)
4290{
4291 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004292 if (buf[0] == '1') {
4293 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004294 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004295 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004296 return length;
4297}
4298SLAB_ATTR(sanity_checks);
4299
4300static ssize_t trace_show(struct kmem_cache *s, char *buf)
4301{
4302 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4303}
4304
4305static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4306 size_t length)
4307{
4308 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004309 if (buf[0] == '1') {
4310 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004311 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004312 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004313 return length;
4314}
4315SLAB_ATTR(trace);
4316
Christoph Lameter81819f02007-05-06 14:49:36 -07004317static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4318{
4319 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4320}
4321
4322static ssize_t red_zone_store(struct kmem_cache *s,
4323 const char *buf, size_t length)
4324{
4325 if (any_slab_objects(s))
4326 return -EBUSY;
4327
4328 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004329 if (buf[0] == '1') {
4330 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004331 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004332 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004333 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004334 return length;
4335}
4336SLAB_ATTR(red_zone);
4337
4338static ssize_t poison_show(struct kmem_cache *s, char *buf)
4339{
4340 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4341}
4342
4343static ssize_t poison_store(struct kmem_cache *s,
4344 const char *buf, size_t length)
4345{
4346 if (any_slab_objects(s))
4347 return -EBUSY;
4348
4349 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004350 if (buf[0] == '1') {
4351 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004352 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004353 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004354 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004355 return length;
4356}
4357SLAB_ATTR(poison);
4358
4359static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4360{
4361 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4362}
4363
4364static ssize_t store_user_store(struct kmem_cache *s,
4365 const char *buf, size_t length)
4366{
4367 if (any_slab_objects(s))
4368 return -EBUSY;
4369
4370 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004371 if (buf[0] == '1') {
4372 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004373 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004374 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004375 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004376 return length;
4377}
4378SLAB_ATTR(store_user);
4379
Christoph Lameter53e15af2007-05-06 14:49:43 -07004380static ssize_t validate_show(struct kmem_cache *s, char *buf)
4381{
4382 return 0;
4383}
4384
4385static ssize_t validate_store(struct kmem_cache *s,
4386 const char *buf, size_t length)
4387{
Christoph Lameter434e2452007-07-17 04:03:30 -07004388 int ret = -EINVAL;
4389
4390 if (buf[0] == '1') {
4391 ret = validate_slab_cache(s);
4392 if (ret >= 0)
4393 ret = length;
4394 }
4395 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004396}
4397SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004398
4399static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4400{
4401 if (!(s->flags & SLAB_STORE_USER))
4402 return -ENOSYS;
4403 return list_locations(s, buf, TRACK_ALLOC);
4404}
4405SLAB_ATTR_RO(alloc_calls);
4406
4407static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4408{
4409 if (!(s->flags & SLAB_STORE_USER))
4410 return -ENOSYS;
4411 return list_locations(s, buf, TRACK_FREE);
4412}
4413SLAB_ATTR_RO(free_calls);
4414#endif /* CONFIG_SLUB_DEBUG */
4415
4416#ifdef CONFIG_FAILSLAB
4417static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4418{
4419 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4420}
4421
4422static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4423 size_t length)
4424{
4425 s->flags &= ~SLAB_FAILSLAB;
4426 if (buf[0] == '1')
4427 s->flags |= SLAB_FAILSLAB;
4428 return length;
4429}
4430SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004431#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004432
Christoph Lameter2086d262007-05-06 14:49:46 -07004433static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4434{
4435 return 0;
4436}
4437
4438static ssize_t shrink_store(struct kmem_cache *s,
4439 const char *buf, size_t length)
4440{
4441 if (buf[0] == '1') {
4442 int rc = kmem_cache_shrink(s);
4443
4444 if (rc)
4445 return rc;
4446 } else
4447 return -EINVAL;
4448 return length;
4449}
4450SLAB_ATTR(shrink);
4451
Christoph Lameter81819f02007-05-06 14:49:36 -07004452#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004453static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004454{
Christoph Lameter98246012008-01-07 23:20:26 -08004455 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004456}
4457
Christoph Lameter98246012008-01-07 23:20:26 -08004458static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004459 const char *buf, size_t length)
4460{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004461 unsigned long ratio;
4462 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004463
Christoph Lameter0121c6192008-04-29 16:11:12 -07004464 err = strict_strtoul(buf, 10, &ratio);
4465 if (err)
4466 return err;
4467
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004468 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004469 s->remote_node_defrag_ratio = ratio * 10;
4470
Christoph Lameter81819f02007-05-06 14:49:36 -07004471 return length;
4472}
Christoph Lameter98246012008-01-07 23:20:26 -08004473SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004474#endif
4475
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004476#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004477static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4478{
4479 unsigned long sum = 0;
4480 int cpu;
4481 int len;
4482 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4483
4484 if (!data)
4485 return -ENOMEM;
4486
4487 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004488 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004489
4490 data[cpu] = x;
4491 sum += x;
4492 }
4493
4494 len = sprintf(buf, "%lu", sum);
4495
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004496#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004497 for_each_online_cpu(cpu) {
4498 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004499 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004500 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004501#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004502 kfree(data);
4503 return len + sprintf(buf + len, "\n");
4504}
4505
David Rientjes78eb00c2009-10-15 02:20:22 -07004506static void clear_stat(struct kmem_cache *s, enum stat_item si)
4507{
4508 int cpu;
4509
4510 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004511 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004512}
4513
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004514#define STAT_ATTR(si, text) \
4515static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4516{ \
4517 return show_stat(s, buf, si); \
4518} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004519static ssize_t text##_store(struct kmem_cache *s, \
4520 const char *buf, size_t length) \
4521{ \
4522 if (buf[0] != '0') \
4523 return -EINVAL; \
4524 clear_stat(s, si); \
4525 return length; \
4526} \
4527SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004528
4529STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4530STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4531STAT_ATTR(FREE_FASTPATH, free_fastpath);
4532STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4533STAT_ATTR(FREE_FROZEN, free_frozen);
4534STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4535STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4536STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4537STAT_ATTR(ALLOC_SLAB, alloc_slab);
4538STAT_ATTR(ALLOC_REFILL, alloc_refill);
4539STAT_ATTR(FREE_SLAB, free_slab);
4540STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4541STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4542STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4543STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4544STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4545STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004546STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004547STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4548STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004549#endif
4550
Pekka Enberg06428782008-01-07 23:20:27 -08004551static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004552 &slab_size_attr.attr,
4553 &object_size_attr.attr,
4554 &objs_per_slab_attr.attr,
4555 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004556 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004557 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004558 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004559 &partial_attr.attr,
4560 &cpu_slabs_attr.attr,
4561 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004562 &aliases_attr.attr,
4563 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004564 &hwcache_align_attr.attr,
4565 &reclaim_account_attr.attr,
4566 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004567 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004568 &reserved_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004569#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004570 &total_objects_attr.attr,
4571 &slabs_attr.attr,
4572 &sanity_checks_attr.attr,
4573 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004574 &red_zone_attr.attr,
4575 &poison_attr.attr,
4576 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004577 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004578 &alloc_calls_attr.attr,
4579 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004580#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004581#ifdef CONFIG_ZONE_DMA
4582 &cache_dma_attr.attr,
4583#endif
4584#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004585 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004586#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004587#ifdef CONFIG_SLUB_STATS
4588 &alloc_fastpath_attr.attr,
4589 &alloc_slowpath_attr.attr,
4590 &free_fastpath_attr.attr,
4591 &free_slowpath_attr.attr,
4592 &free_frozen_attr.attr,
4593 &free_add_partial_attr.attr,
4594 &free_remove_partial_attr.attr,
4595 &alloc_from_partial_attr.attr,
4596 &alloc_slab_attr.attr,
4597 &alloc_refill_attr.attr,
4598 &free_slab_attr.attr,
4599 &cpuslab_flush_attr.attr,
4600 &deactivate_full_attr.attr,
4601 &deactivate_empty_attr.attr,
4602 &deactivate_to_head_attr.attr,
4603 &deactivate_to_tail_attr.attr,
4604 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004605 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05004606 &cmpxchg_double_fail_attr.attr,
4607 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004608#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004609#ifdef CONFIG_FAILSLAB
4610 &failslab_attr.attr,
4611#endif
4612
Christoph Lameter81819f02007-05-06 14:49:36 -07004613 NULL
4614};
4615
4616static struct attribute_group slab_attr_group = {
4617 .attrs = slab_attrs,
4618};
4619
4620static ssize_t slab_attr_show(struct kobject *kobj,
4621 struct attribute *attr,
4622 char *buf)
4623{
4624 struct slab_attribute *attribute;
4625 struct kmem_cache *s;
4626 int err;
4627
4628 attribute = to_slab_attr(attr);
4629 s = to_slab(kobj);
4630
4631 if (!attribute->show)
4632 return -EIO;
4633
4634 err = attribute->show(s, buf);
4635
4636 return err;
4637}
4638
4639static ssize_t slab_attr_store(struct kobject *kobj,
4640 struct attribute *attr,
4641 const char *buf, size_t len)
4642{
4643 struct slab_attribute *attribute;
4644 struct kmem_cache *s;
4645 int err;
4646
4647 attribute = to_slab_attr(attr);
4648 s = to_slab(kobj);
4649
4650 if (!attribute->store)
4651 return -EIO;
4652
4653 err = attribute->store(s, buf, len);
4654
4655 return err;
4656}
4657
Christoph Lameter151c6022008-01-07 22:29:05 -08004658static void kmem_cache_release(struct kobject *kobj)
4659{
4660 struct kmem_cache *s = to_slab(kobj);
4661
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004662 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004663 kfree(s);
4664}
4665
Emese Revfy52cf25d2010-01-19 02:58:23 +01004666static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004667 .show = slab_attr_show,
4668 .store = slab_attr_store,
4669};
4670
4671static struct kobj_type slab_ktype = {
4672 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004673 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004674};
4675
4676static int uevent_filter(struct kset *kset, struct kobject *kobj)
4677{
4678 struct kobj_type *ktype = get_ktype(kobj);
4679
4680 if (ktype == &slab_ktype)
4681 return 1;
4682 return 0;
4683}
4684
Emese Revfy9cd43612009-12-31 14:52:51 +01004685static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004686 .filter = uevent_filter,
4687};
4688
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004689static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004690
4691#define ID_STR_LENGTH 64
4692
4693/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004694 *
4695 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004696 */
4697static char *create_unique_id(struct kmem_cache *s)
4698{
4699 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4700 char *p = name;
4701
4702 BUG_ON(!name);
4703
4704 *p++ = ':';
4705 /*
4706 * First flags affecting slabcache operations. We will only
4707 * get here for aliasable slabs so we do not need to support
4708 * too many flags. The flags here must cover all flags that
4709 * are matched during merging to guarantee that the id is
4710 * unique.
4711 */
4712 if (s->flags & SLAB_CACHE_DMA)
4713 *p++ = 'd';
4714 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4715 *p++ = 'a';
4716 if (s->flags & SLAB_DEBUG_FREE)
4717 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004718 if (!(s->flags & SLAB_NOTRACK))
4719 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004720 if (p != name + 1)
4721 *p++ = '-';
4722 p += sprintf(p, "%07d", s->size);
4723 BUG_ON(p > name + ID_STR_LENGTH - 1);
4724 return name;
4725}
4726
4727static int sysfs_slab_add(struct kmem_cache *s)
4728{
4729 int err;
4730 const char *name;
4731 int unmergeable;
4732
4733 if (slab_state < SYSFS)
4734 /* Defer until later */
4735 return 0;
4736
4737 unmergeable = slab_unmergeable(s);
4738 if (unmergeable) {
4739 /*
4740 * Slabcache can never be merged so we can use the name proper.
4741 * This is typically the case for debug situations. In that
4742 * case we can catch duplicate names easily.
4743 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004744 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004745 name = s->name;
4746 } else {
4747 /*
4748 * Create a unique name for the slab as a target
4749 * for the symlinks.
4750 */
4751 name = create_unique_id(s);
4752 }
4753
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004754 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004755 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4756 if (err) {
4757 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004758 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004759 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004760
4761 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004762 if (err) {
4763 kobject_del(&s->kobj);
4764 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004765 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004766 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004767 kobject_uevent(&s->kobj, KOBJ_ADD);
4768 if (!unmergeable) {
4769 /* Setup first alias */
4770 sysfs_slab_alias(s, s->name);
4771 kfree(name);
4772 }
4773 return 0;
4774}
4775
4776static void sysfs_slab_remove(struct kmem_cache *s)
4777{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004778 if (slab_state < SYSFS)
4779 /*
4780 * Sysfs has not been setup yet so no need to remove the
4781 * cache from sysfs.
4782 */
4783 return;
4784
Christoph Lameter81819f02007-05-06 14:49:36 -07004785 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4786 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004787 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004788}
4789
4790/*
4791 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004792 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004793 */
4794struct saved_alias {
4795 struct kmem_cache *s;
4796 const char *name;
4797 struct saved_alias *next;
4798};
4799
Adrian Bunk5af328a2007-07-17 04:03:27 -07004800static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004801
4802static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4803{
4804 struct saved_alias *al;
4805
4806 if (slab_state == SYSFS) {
4807 /*
4808 * If we have a leftover link then remove it.
4809 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004810 sysfs_remove_link(&slab_kset->kobj, name);
4811 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004812 }
4813
4814 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4815 if (!al)
4816 return -ENOMEM;
4817
4818 al->s = s;
4819 al->name = name;
4820 al->next = alias_list;
4821 alias_list = al;
4822 return 0;
4823}
4824
4825static int __init slab_sysfs_init(void)
4826{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004827 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004828 int err;
4829
Christoph Lameter2bce6482010-07-19 11:39:11 -05004830 down_write(&slub_lock);
4831
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004832 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004833 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004834 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004835 printk(KERN_ERR "Cannot register slab subsystem.\n");
4836 return -ENOSYS;
4837 }
4838
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004839 slab_state = SYSFS;
4840
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004841 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004842 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004843 if (err)
4844 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4845 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004846 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004847
4848 while (alias_list) {
4849 struct saved_alias *al = alias_list;
4850
4851 alias_list = alias_list->next;
4852 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004853 if (err)
4854 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4855 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004856 kfree(al);
4857 }
4858
Christoph Lameter2bce6482010-07-19 11:39:11 -05004859 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004860 resiliency_test();
4861 return 0;
4862}
4863
4864__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004865#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004866
4867/*
4868 * The /proc/slabinfo ABI
4869 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004870#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004871static void print_slabinfo_header(struct seq_file *m)
4872{
4873 seq_puts(m, "slabinfo - version: 2.1\n");
4874 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4875 "<objperslab> <pagesperslab>");
4876 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4877 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4878 seq_putc(m, '\n');
4879}
4880
4881static void *s_start(struct seq_file *m, loff_t *pos)
4882{
4883 loff_t n = *pos;
4884
4885 down_read(&slub_lock);
4886 if (!n)
4887 print_slabinfo_header(m);
4888
4889 return seq_list_start(&slab_caches, *pos);
4890}
4891
4892static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4893{
4894 return seq_list_next(p, &slab_caches, pos);
4895}
4896
4897static void s_stop(struct seq_file *m, void *p)
4898{
4899 up_read(&slub_lock);
4900}
4901
4902static int s_show(struct seq_file *m, void *p)
4903{
4904 unsigned long nr_partials = 0;
4905 unsigned long nr_slabs = 0;
4906 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004907 unsigned long nr_objs = 0;
4908 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004909 struct kmem_cache *s;
4910 int node;
4911
4912 s = list_entry(p, struct kmem_cache, list);
4913
4914 for_each_online_node(node) {
4915 struct kmem_cache_node *n = get_node(s, node);
4916
4917 if (!n)
4918 continue;
4919
4920 nr_partials += n->nr_partial;
4921 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004922 nr_objs += atomic_long_read(&n->total_objects);
4923 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004924 }
4925
Christoph Lameter205ab992008-04-14 19:11:40 +03004926 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004927
4928 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004929 nr_objs, s->size, oo_objects(s->oo),
4930 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004931 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4932 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4933 0UL);
4934 seq_putc(m, '\n');
4935 return 0;
4936}
4937
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004938static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004939 .start = s_start,
4940 .next = s_next,
4941 .stop = s_stop,
4942 .show = s_show,
4943};
4944
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004945static int slabinfo_open(struct inode *inode, struct file *file)
4946{
4947 return seq_open(file, &slabinfo_op);
4948}
4949
4950static const struct file_operations proc_slabinfo_operations = {
4951 .open = slabinfo_open,
4952 .read = seq_read,
4953 .llseek = seq_lseek,
4954 .release = seq_release,
4955};
4956
4957static int __init slab_proc_init(void)
4958{
WANG Congcf5d1132009-08-18 19:11:40 +03004959 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004960 return 0;
4961}
4962module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004963#endif /* CONFIG_SLABINFO */