blob: e39be0928a22a1df33ca06493f59bbbb4ca185bf [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 Lameter5cc6eee2011-06-01 12:25:50 -0500919 *
920 * list_lock must be held.
Christoph Lameter643b1132007-05-06 14:49:42 -0700921 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500922static void add_full(struct kmem_cache *s,
923 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700924{
Christoph Lameter643b1132007-05-06 14:49:42 -0700925 if (!(s->flags & SLAB_STORE_USER))
926 return;
927
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500928 list_add(&page->lru, &n->full);
929}
Christoph Lameter643b1132007-05-06 14:49:42 -0700930
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500931/*
932 * list_lock must be held.
933 */
934static void remove_full(struct kmem_cache *s, struct page *page)
935{
936 if (!(s->flags & SLAB_STORE_USER))
937 return;
938
Christoph Lameter643b1132007-05-06 14:49:42 -0700939 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -0700940}
941
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300942/* Tracking of the number of slabs for debugging purposes */
943static inline unsigned long slabs_node(struct kmem_cache *s, int node)
944{
945 struct kmem_cache_node *n = get_node(s, node);
946
947 return atomic_long_read(&n->nr_slabs);
948}
949
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400950static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
951{
952 return atomic_long_read(&n->nr_slabs);
953}
954
Christoph Lameter205ab992008-04-14 19:11:40 +0300955static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300956{
957 struct kmem_cache_node *n = get_node(s, node);
958
959 /*
960 * May be called early in order to allocate a slab for the
961 * kmem_cache_node structure. Solve the chicken-egg
962 * dilemma by deferring the increment of the count during
963 * bootstrap (see early_kmem_cache_node_alloc).
964 */
Christoph Lameter7340cc82010-09-28 08:10:26 -0500965 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300966 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300967 atomic_long_add(objects, &n->total_objects);
968 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300969}
Christoph Lameter205ab992008-04-14 19:11:40 +0300970static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300971{
972 struct kmem_cache_node *n = get_node(s, node);
973
974 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300975 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300976}
977
978/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700979static void setup_object_debug(struct kmem_cache *s, struct page *page,
980 void *object)
981{
982 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
983 return;
984
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500985 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700986 init_tracking(s, object);
987}
988
Christoph Lameter15370662010-08-20 12:37:12 -0500989static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300990 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700991{
992 if (!check_slab(s, page))
993 goto bad;
994
Christoph Lameterd692ef62008-02-15 23:45:24 -0800995 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700996 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700997 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700998 }
999
1000 if (!check_valid_pointer(s, page, object)) {
1001 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001002 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001003 }
1004
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001005 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001006 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001007
Christoph Lameter3ec09742007-05-16 22:11:00 -07001008 /* Success perform special debug activities for allocs */
1009 if (s->flags & SLAB_STORE_USER)
1010 set_track(s, object, TRACK_ALLOC, addr);
1011 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001012 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001013 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001014
Christoph Lameter81819f02007-05-06 14:49:36 -07001015bad:
1016 if (PageSlab(page)) {
1017 /*
1018 * If this is a slab page then lets do the best we can
1019 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001020 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001021 */
Christoph Lameter24922682007-07-17 04:03:18 -07001022 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001023 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001024 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001025 }
1026 return 0;
1027}
1028
Christoph Lameter15370662010-08-20 12:37:12 -05001029static noinline int free_debug_processing(struct kmem_cache *s,
1030 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001031{
1032 if (!check_slab(s, page))
1033 goto fail;
1034
1035 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001036 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001037 goto fail;
1038 }
1039
1040 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001041 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001042 goto fail;
1043 }
1044
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001045 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001046 return 0;
1047
1048 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001049 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001050 slab_err(s, page, "Attempt to free object(0x%p) "
1051 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001052 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001053 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001054 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001055 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001056 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001057 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001058 object_err(s, page, object,
1059 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001060 goto fail;
1061 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001062
1063 /* Special debug activities for freeing objects */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001064 if (!page->frozen && !page->freelist) {
1065 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1066
1067 spin_lock(&n->list_lock);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001068 remove_full(s, page);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001069 spin_unlock(&n->list_lock);
1070 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001071 if (s->flags & SLAB_STORE_USER)
1072 set_track(s, object, TRACK_FREE, addr);
1073 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001074 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001075 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001076
Christoph Lameter81819f02007-05-06 14:49:36 -07001077fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001078 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001079 return 0;
1080}
1081
Christoph Lameter41ecc552007-05-09 02:32:44 -07001082static int __init setup_slub_debug(char *str)
1083{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001084 slub_debug = DEBUG_DEFAULT_FLAGS;
1085 if (*str++ != '=' || !*str)
1086 /*
1087 * No options specified. Switch on full debugging.
1088 */
1089 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001090
1091 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001092 /*
1093 * No options but restriction on slabs. This means full
1094 * debugging for slabs matching a pattern.
1095 */
1096 goto check_slabs;
1097
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001098 if (tolower(*str) == 'o') {
1099 /*
1100 * Avoid enabling debugging on caches if its minimum order
1101 * would increase as a result.
1102 */
1103 disable_higher_order_debug = 1;
1104 goto out;
1105 }
1106
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001107 slub_debug = 0;
1108 if (*str == '-')
1109 /*
1110 * Switch off all debugging measures.
1111 */
1112 goto out;
1113
1114 /*
1115 * Determine which debug features should be switched on
1116 */
Pekka Enberg06428782008-01-07 23:20:27 -08001117 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001118 switch (tolower(*str)) {
1119 case 'f':
1120 slub_debug |= SLAB_DEBUG_FREE;
1121 break;
1122 case 'z':
1123 slub_debug |= SLAB_RED_ZONE;
1124 break;
1125 case 'p':
1126 slub_debug |= SLAB_POISON;
1127 break;
1128 case 'u':
1129 slub_debug |= SLAB_STORE_USER;
1130 break;
1131 case 't':
1132 slub_debug |= SLAB_TRACE;
1133 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001134 case 'a':
1135 slub_debug |= SLAB_FAILSLAB;
1136 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001137 default:
1138 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001139 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001140 }
1141 }
1142
1143check_slabs:
1144 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001145 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001146out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001147 return 1;
1148}
1149
1150__setup("slub_debug", setup_slub_debug);
1151
Christoph Lameterba0268a2007-09-11 15:24:11 -07001152static unsigned long kmem_cache_flags(unsigned long objsize,
1153 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001154 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001155{
1156 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001157 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001158 */
Christoph Lametere1533622008-02-15 23:45:24 -08001159 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001160 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1161 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001162
1163 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001164}
1165#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001166static inline void setup_object_debug(struct kmem_cache *s,
1167 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001168
Christoph Lameter3ec09742007-05-16 22:11:00 -07001169static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001170 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001171
Christoph Lameter3ec09742007-05-16 22:11:00 -07001172static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001173 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001174
Christoph Lameter41ecc552007-05-09 02:32:44 -07001175static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1176 { return 1; }
1177static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001178 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001179static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1180 struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001181static inline unsigned long kmem_cache_flags(unsigned long objsize,
1182 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001183 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001184{
1185 return flags;
1186}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001187#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001188
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001189#define disable_higher_order_debug 0
1190
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001191static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1192 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001193static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1194 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001195static inline void inc_slabs_node(struct kmem_cache *s, int node,
1196 int objects) {}
1197static inline void dec_slabs_node(struct kmem_cache *s, int node,
1198 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001199
1200static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1201 { return 0; }
1202
1203static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1204 void *object) {}
1205
1206static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1207
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001208#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001209
Christoph Lameter81819f02007-05-06 14:49:36 -07001210/*
1211 * Slab allocation and freeing
1212 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001213static inline struct page *alloc_slab_page(gfp_t flags, int node,
1214 struct kmem_cache_order_objects oo)
1215{
1216 int order = oo_order(oo);
1217
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001218 flags |= __GFP_NOTRACK;
1219
Christoph Lameter2154a332010-07-09 14:07:10 -05001220 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001221 return alloc_pages(flags, order);
1222 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001223 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001224}
1225
Christoph Lameter81819f02007-05-06 14:49:36 -07001226static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1227{
Pekka Enberg06428782008-01-07 23:20:27 -08001228 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001229 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001230 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001231
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001232 flags &= gfp_allowed_mask;
1233
1234 if (flags & __GFP_WAIT)
1235 local_irq_enable();
1236
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001237 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001238
Pekka Enbergba522702009-06-24 21:59:51 +03001239 /*
1240 * Let the initial higher-order allocation fail under memory pressure
1241 * so we fall-back to the minimum order allocation.
1242 */
1243 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1244
1245 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001246 if (unlikely(!page)) {
1247 oo = s->min;
1248 /*
1249 * Allocation may have failed due to fragmentation.
1250 * Try a lower order alloc if possible
1251 */
1252 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001253
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001254 if (page)
1255 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001256 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001257
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001258 if (flags & __GFP_WAIT)
1259 local_irq_disable();
1260
1261 if (!page)
1262 return NULL;
1263
Vegard Nossum5a896d92008-04-04 00:54:48 +02001264 if (kmemcheck_enabled
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001265 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001266 int pages = 1 << oo_order(oo);
1267
1268 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1269
1270 /*
1271 * Objects from caches that have a constructor don't get
1272 * cleared when they're allocated, so we need to do it here.
1273 */
1274 if (s->ctor)
1275 kmemcheck_mark_uninitialized_pages(page, pages);
1276 else
1277 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001278 }
1279
Christoph Lameter834f3d12008-04-14 19:11:31 +03001280 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001281 mod_zone_page_state(page_zone(page),
1282 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1283 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001284 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001285
1286 return page;
1287}
1288
1289static void setup_object(struct kmem_cache *s, struct page *page,
1290 void *object)
1291{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001292 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001293 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001294 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001295}
1296
1297static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1298{
1299 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001300 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001301 void *last;
1302 void *p;
1303
Christoph Lameter6cb06222007-10-16 01:25:41 -07001304 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001305
Christoph Lameter6cb06222007-10-16 01:25:41 -07001306 page = allocate_slab(s,
1307 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001308 if (!page)
1309 goto out;
1310
Christoph Lameter205ab992008-04-14 19:11:40 +03001311 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001312 page->slab = s;
1313 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001314
1315 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001316
1317 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001318 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001319
1320 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001321 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001322 setup_object(s, page, last);
1323 set_freepointer(s, last, p);
1324 last = p;
1325 }
1326 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001327 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001328
1329 page->freelist = start;
1330 page->inuse = 0;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001331 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001332out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001333 return page;
1334}
1335
1336static void __free_slab(struct kmem_cache *s, struct page *page)
1337{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001338 int order = compound_order(page);
1339 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001340
Christoph Lameteraf537b02010-07-09 14:07:14 -05001341 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001342 void *p;
1343
1344 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001345 for_each_object(p, s, page_address(page),
1346 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001347 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001348 }
1349
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001350 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001351
Christoph Lameter81819f02007-05-06 14:49:36 -07001352 mod_zone_page_state(page_zone(page),
1353 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1354 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001355 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001356
Christoph Lameter49bd5222008-04-14 18:52:18 +03001357 __ClearPageSlab(page);
1358 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001359 if (current->reclaim_state)
1360 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001361 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001362}
1363
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001364#define need_reserve_slab_rcu \
1365 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1366
Christoph Lameter81819f02007-05-06 14:49:36 -07001367static void rcu_free_slab(struct rcu_head *h)
1368{
1369 struct page *page;
1370
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001371 if (need_reserve_slab_rcu)
1372 page = virt_to_head_page(h);
1373 else
1374 page = container_of((struct list_head *)h, struct page, lru);
1375
Christoph Lameter81819f02007-05-06 14:49:36 -07001376 __free_slab(page->slab, page);
1377}
1378
1379static void free_slab(struct kmem_cache *s, struct page *page)
1380{
1381 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001382 struct rcu_head *head;
1383
1384 if (need_reserve_slab_rcu) {
1385 int order = compound_order(page);
1386 int offset = (PAGE_SIZE << order) - s->reserved;
1387
1388 VM_BUG_ON(s->reserved != sizeof(*head));
1389 head = page_address(page) + offset;
1390 } else {
1391 /*
1392 * RCU free overloads the RCU head over the LRU
1393 */
1394 head = (void *)&page->lru;
1395 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001396
1397 call_rcu(head, rcu_free_slab);
1398 } else
1399 __free_slab(s, page);
1400}
1401
1402static void discard_slab(struct kmem_cache *s, struct page *page)
1403{
Christoph Lameter205ab992008-04-14 19:11:40 +03001404 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001405 free_slab(s, page);
1406}
1407
1408/*
1409 * Per slab locking using the pagelock
1410 */
1411static __always_inline void slab_lock(struct page *page)
1412{
1413 bit_spin_lock(PG_locked, &page->flags);
1414}
1415
1416static __always_inline void slab_unlock(struct page *page)
1417{
Nick Piggina76d3542008-01-07 23:20:27 -08001418 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001419}
1420
1421static __always_inline int slab_trylock(struct page *page)
1422{
1423 int rc = 1;
1424
1425 rc = bit_spin_trylock(PG_locked, &page->flags);
1426 return rc;
1427}
1428
1429/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001430 * Management of partially allocated slabs.
1431 *
1432 * list_lock must be held.
Christoph Lameter81819f02007-05-06 14:49:36 -07001433 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001434static inline void add_partial(struct kmem_cache_node *n,
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001435 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001436{
Christoph Lametere95eed52007-05-06 14:49:44 -07001437 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001438 if (tail)
1439 list_add_tail(&page->lru, &n->partial);
1440 else
1441 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001442}
1443
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001444/*
1445 * list_lock must be held.
1446 */
1447static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001448 struct page *page)
1449{
1450 list_del(&page->lru);
1451 n->nr_partial--;
1452}
1453
Christoph Lameter81819f02007-05-06 14:49:36 -07001454/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001455 * Lock slab, remove from the partial list and put the object into the
1456 * per cpu freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001457 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001458 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001459 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001460static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1461 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001462{
1463 if (slab_trylock(page)) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001464 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001465 return 1;
1466 }
1467 return 0;
1468}
1469
1470/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001471 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001472 */
1473static struct page *get_partial_node(struct kmem_cache_node *n)
1474{
1475 struct page *page;
1476
1477 /*
1478 * Racy check. If we mistakenly see no partial slabs then we
1479 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001480 * partial slab and there is none available then get_partials()
1481 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001482 */
1483 if (!n || !n->nr_partial)
1484 return NULL;
1485
1486 spin_lock(&n->list_lock);
1487 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001488 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001489 goto out;
1490 page = NULL;
1491out:
1492 spin_unlock(&n->list_lock);
1493 return page;
1494}
1495
1496/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001497 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001498 */
1499static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1500{
1501#ifdef CONFIG_NUMA
1502 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001503 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001504 struct zone *zone;
1505 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001506 struct page *page;
1507
1508 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001509 * The defrag ratio allows a configuration of the tradeoffs between
1510 * inter node defragmentation and node local allocations. A lower
1511 * defrag_ratio increases the tendency to do local allocations
1512 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001513 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001514 * If the defrag_ratio is set to 0 then kmalloc() always
1515 * returns node local objects. If the ratio is higher then kmalloc()
1516 * may return off node objects because partial slabs are obtained
1517 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001518 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001519 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001520 * defrag_ratio = 1000) then every (well almost) allocation will
1521 * first attempt to defrag slab caches on other nodes. This means
1522 * scanning over all nodes to look for partial slabs which may be
1523 * expensive if we do it every time we are trying to find a slab
1524 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001525 */
Christoph Lameter98246012008-01-07 23:20:26 -08001526 if (!s->remote_node_defrag_ratio ||
1527 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001528 return NULL;
1529
Miao Xiec0ff7452010-05-24 14:32:08 -07001530 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001531 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001532 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001533 struct kmem_cache_node *n;
1534
Mel Gorman54a6eb52008-04-28 02:12:16 -07001535 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001536
Mel Gorman54a6eb52008-04-28 02:12:16 -07001537 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001538 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001539 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001540 if (page) {
1541 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001542 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001543 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001544 }
1545 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001546 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001547#endif
1548 return NULL;
1549}
1550
1551/*
1552 * Get a partial page, lock it and return it.
1553 */
1554static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1555{
1556 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001557 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001558
1559 page = get_partial_node(get_node(s, searchnode));
Christoph Lameter33de04e2011-04-15 14:48:12 -05001560 if (page || node != NUMA_NO_NODE)
Christoph Lameter81819f02007-05-06 14:49:36 -07001561 return page;
1562
1563 return get_any_partial(s, flags);
1564}
1565
1566/*
1567 * Move a page back to the lists.
1568 *
1569 * Must be called with the slab lock held.
1570 *
1571 * On exit the slab lock will have been dropped.
1572 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001573static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Namhyung Kim34789732010-09-29 21:02:14 +09001574 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001575{
Christoph Lametere95eed52007-05-06 14:49:44 -07001576 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1577
Christoph Lameter81819f02007-05-06 14:49:36 -07001578 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001579
Christoph Lametera973e9d2008-03-01 13:40:44 -08001580 if (page->freelist) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001581 spin_lock(&n->list_lock);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001582 add_partial(n, page, tail);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001583 spin_unlock(&n->list_lock);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001584 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001585 } else {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001586 stat(s, DEACTIVATE_FULL);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001587 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER)) {
1588 spin_lock(&n->list_lock);
1589 add_full(s, n, page);
1590 spin_unlock(&n->list_lock);
1591 }
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001592 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001593 slab_unlock(page);
1594 } else {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001595 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001596 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001597 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001598 * Adding an empty slab to the partial slabs in order
1599 * to avoid page allocator overhead. This slab needs
1600 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001601 * so that the others get filled first. That way the
1602 * size of the partial list stays small.
1603 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001604 * kmem_cache_shrink can reclaim any empty slabs from
1605 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001606 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001607 spin_lock(&n->list_lock);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001608 add_partial(n, page, 1);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001609 spin_unlock(&n->list_lock);
Christoph Lametere95eed52007-05-06 14:49:44 -07001610 slab_unlock(page);
1611 } else {
1612 slab_unlock(page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001613 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001614 discard_slab(s, page);
1615 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001616 }
1617}
1618
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001619#ifdef CONFIG_PREEMPT
1620/*
1621 * Calculate the next globally unique transaction for disambiguiation
1622 * during cmpxchg. The transactions start with the cpu number and are then
1623 * incremented by CONFIG_NR_CPUS.
1624 */
1625#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1626#else
1627/*
1628 * No preemption supported therefore also no need to check for
1629 * different cpus.
1630 */
1631#define TID_STEP 1
1632#endif
1633
1634static inline unsigned long next_tid(unsigned long tid)
1635{
1636 return tid + TID_STEP;
1637}
1638
1639static inline unsigned int tid_to_cpu(unsigned long tid)
1640{
1641 return tid % TID_STEP;
1642}
1643
1644static inline unsigned long tid_to_event(unsigned long tid)
1645{
1646 return tid / TID_STEP;
1647}
1648
1649static inline unsigned int init_tid(int cpu)
1650{
1651 return cpu;
1652}
1653
1654static inline void note_cmpxchg_failure(const char *n,
1655 const struct kmem_cache *s, unsigned long tid)
1656{
1657#ifdef SLUB_DEBUG_CMPXCHG
1658 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1659
1660 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1661
1662#ifdef CONFIG_PREEMPT
1663 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1664 printk("due to cpu change %d -> %d\n",
1665 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1666 else
1667#endif
1668 if (tid_to_event(tid) != tid_to_event(actual_tid))
1669 printk("due to cpu running other code. Event %ld->%ld\n",
1670 tid_to_event(tid), tid_to_event(actual_tid));
1671 else
1672 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1673 actual_tid, tid, next_tid(tid));
1674#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001675 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001676}
1677
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001678void init_kmem_cache_cpus(struct kmem_cache *s)
1679{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001680 int cpu;
1681
1682 for_each_possible_cpu(cpu)
1683 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001684}
Christoph Lameter81819f02007-05-06 14:49:36 -07001685/*
1686 * Remove the cpu slab
1687 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001688static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Namhyung Kim34789732010-09-29 21:02:14 +09001689 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001690{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001691 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001692 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001693
Christoph Lameterb773ad72008-03-04 11:10:17 -08001694 if (page->freelist)
Christoph Lameter84e554e62009-12-18 16:26:23 -06001695 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001696 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001697 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001698 * because both freelists are empty. So this is unlikely
1699 * to occur.
1700 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001701 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001702 void **object;
1703
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001704 tail = 0; /* Hot objects. Put the slab first */
1705
Christoph Lameter894b8782007-05-10 03:15:16 -07001706 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001707 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001708 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001709
1710 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001711 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001712 page->freelist = object;
1713 page->inuse--;
1714 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001715 c->page = NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001716 c->tid = next_tid(c->tid);
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001717 page->frozen = 0;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001718 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001719}
1720
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001721static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001722{
Christoph Lameter84e554e62009-12-18 16:26:23 -06001723 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001724 slab_lock(c->page);
1725 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001726}
1727
1728/*
1729 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001730 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001731 * Called from IPI handler with interrupts disabled.
1732 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001733static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001734{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001735 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001736
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001737 if (likely(c && c->page))
1738 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001739}
1740
1741static void flush_cpu_slab(void *d)
1742{
1743 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001744
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001745 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001746}
1747
1748static void flush_all(struct kmem_cache *s)
1749{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001750 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001751}
1752
1753/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001754 * Check if the objects in a per cpu structure fit numa
1755 * locality expectations.
1756 */
1757static inline int node_match(struct kmem_cache_cpu *c, int node)
1758{
1759#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001760 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001761 return 0;
1762#endif
1763 return 1;
1764}
1765
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001766static int count_free(struct page *page)
1767{
1768 return page->objects - page->inuse;
1769}
1770
1771static unsigned long count_partial(struct kmem_cache_node *n,
1772 int (*get_count)(struct page *))
1773{
1774 unsigned long flags;
1775 unsigned long x = 0;
1776 struct page *page;
1777
1778 spin_lock_irqsave(&n->list_lock, flags);
1779 list_for_each_entry(page, &n->partial, lru)
1780 x += get_count(page);
1781 spin_unlock_irqrestore(&n->list_lock, flags);
1782 return x;
1783}
1784
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001785static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1786{
1787#ifdef CONFIG_SLUB_DEBUG
1788 return atomic_long_read(&n->total_objects);
1789#else
1790 return 0;
1791#endif
1792}
1793
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001794static noinline void
1795slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1796{
1797 int node;
1798
1799 printk(KERN_WARNING
1800 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1801 nid, gfpflags);
1802 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1803 "default order: %d, min order: %d\n", s->name, s->objsize,
1804 s->size, oo_order(s->oo), oo_order(s->min));
1805
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001806 if (oo_order(s->min) > get_order(s->objsize))
1807 printk(KERN_WARNING " %s debugging increased min order, use "
1808 "slub_debug=O to disable.\n", s->name);
1809
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001810 for_each_online_node(node) {
1811 struct kmem_cache_node *n = get_node(s, node);
1812 unsigned long nr_slabs;
1813 unsigned long nr_objs;
1814 unsigned long nr_free;
1815
1816 if (!n)
1817 continue;
1818
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001819 nr_free = count_partial(n, count_free);
1820 nr_slabs = node_nr_slabs(n);
1821 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001822
1823 printk(KERN_WARNING
1824 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1825 node, nr_slabs, nr_objs, nr_free);
1826 }
1827}
1828
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001829/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001830 * Slow path. The lockless freelist is empty or we need to perform
1831 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001832 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001833 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001834 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001835 * Processing is still very fast if new objects have been freed to the
1836 * regular freelist. In that case we simply take over the regular freelist
1837 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001838 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001839 * If that is not working then we fall back to the partial lists. We take the
1840 * first element of the freelist as the object to allocate now and move the
1841 * rest of the freelist to the lockless freelist.
1842 *
1843 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001844 * we need to allocate a new slab. This is the slowest path since it involves
1845 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001846 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001847static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1848 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001849{
Christoph Lameter81819f02007-05-06 14:49:36 -07001850 void **object;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001851 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001852 unsigned long flags;
1853
1854 local_irq_save(flags);
1855#ifdef CONFIG_PREEMPT
1856 /*
1857 * We may have been preempted and rescheduled on a different
1858 * cpu before disabling interrupts. Need to reload cpu area
1859 * pointer.
1860 */
1861 c = this_cpu_ptr(s->cpu_slab);
1862#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07001863
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001864 /* We handle __GFP_ZERO in the caller */
1865 gfpflags &= ~__GFP_ZERO;
1866
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001867 page = c->page;
1868 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001869 goto new_slab;
1870
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001871 slab_lock(page);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001872 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001873 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001874
Christoph Lameter84e554e62009-12-18 16:26:23 -06001875 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001876
Christoph Lameter894b8782007-05-10 03:15:16 -07001877load_freelist:
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001878 VM_BUG_ON(!page->frozen);
1879
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001880 object = page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001881 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001882 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001883 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001884 goto debug;
1885
Christoph Lameterff120592009-12-18 16:26:22 -06001886 c->freelist = get_freepointer(s, object);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001887 page->inuse = page->objects;
1888 page->freelist = NULL;
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001889
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001890 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001891 c->tid = next_tid(c->tid);
1892 local_irq_restore(flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001893 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001894 return object;
1895
1896another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001897 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001898
1899new_slab:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001900 page = get_partial(s, gfpflags, node);
1901 if (page) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001902 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001903 page->frozen = 1;
Christoph Lameterdc1fb7f2011-04-15 14:48:15 -05001904 c->node = page_to_nid(page);
1905 c->page = page;
Christoph Lameter894b8782007-05-10 03:15:16 -07001906 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001907 }
1908
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001909 page = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001910
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001911 if (page) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001912 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001913 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001914 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001915 flush_slab(s, c);
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001916
1917 slab_lock(page);
Christoph Lameter50d5c412011-06-01 12:25:45 -05001918 page->frozen = 1;
David Rientjesbd07d872011-05-12 13:10:49 -07001919 c->node = page_to_nid(page);
1920 c->page = page;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001921 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001922 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001923 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1924 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter2fd66c52011-03-22 13:32:53 -05001925 local_irq_restore(flags);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001926 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001927debug:
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001928 if (!alloc_debug_processing(s, page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001929 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001930
Christoph Lameter01ad8a72011-04-15 14:48:14 -05001931 page->inuse++;
1932 page->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05001933 deactivate_slab(s, c);
1934 c->page = NULL;
Pekka Enberg15b7c512010-10-02 11:32:32 +03001935 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05001936 local_irq_restore(flags);
1937 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07001938}
1939
1940/*
1941 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1942 * have the fastpath folded into their functions. So no function call
1943 * overhead for requests that can be satisfied on the fastpath.
1944 *
1945 * The fastpath works by first checking if the lockless freelist can be used.
1946 * If not then __slab_alloc is called for slow processing.
1947 *
1948 * Otherwise we can simply pick the next object from the lockless free list.
1949 */
Pekka Enberg06428782008-01-07 23:20:27 -08001950static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001951 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001952{
Christoph Lameter894b8782007-05-10 03:15:16 -07001953 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001954 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001955 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08001956
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001957 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001958 return NULL;
1959
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001960redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001961
1962 /*
1963 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1964 * enabled. We may switch back and forth between cpus while
1965 * reading from one cpu area. That does not matter as long
1966 * as we end up on the original cpu again when doing the cmpxchg.
1967 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001968 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001969
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001970 /*
1971 * The transaction ids are globally unique per cpu and per operation on
1972 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1973 * occurs on the right processor and that there was no operation on the
1974 * linked list in between.
1975 */
1976 tid = c->tid;
1977 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001978
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001979 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001980 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001981
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001982 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001983
1984 else {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001985 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001986 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001987 * operation and if we are on the right processor.
1988 *
1989 * The cmpxchg does the following atomically (without lock semantics!)
1990 * 1. Relocate first pointer to the current per cpu area.
1991 * 2. Verify that tid and freelist have not been changed
1992 * 3. If they were not changed replace tid and freelist
1993 *
1994 * Since this is without lock semantics the protection is only against
1995 * code executing on this cpu *not* from access by other cpus.
1996 */
Thomas Gleixner30106b82011-05-04 15:38:19 +02001997 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001998 s->cpu_slab->freelist, s->cpu_slab->tid,
1999 object, tid,
Christoph Lameter1393d9a2011-05-16 15:26:08 -05002000 get_freepointer_safe(s, object), next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002001
2002 note_cmpxchg_failure("slab_alloc", s, tid);
2003 goto redo;
2004 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002005 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002006 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002007
Pekka Enberg74e21342009-11-25 20:14:48 +02002008 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06002009 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002010
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002011 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002012
Christoph Lameter894b8782007-05-10 03:15:16 -07002013 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002014}
2015
2016void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2017{
Christoph Lameter2154a332010-07-09 14:07:10 -05002018 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002019
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002020 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002021
2022 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002023}
2024EXPORT_SYMBOL(kmem_cache_alloc);
2025
Li Zefan0f24f122009-12-11 15:45:30 +08002026#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002027void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002028{
Richard Kennedy4a923792010-10-21 10:29:19 +01002029 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
2030 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2031 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002032}
Richard Kennedy4a923792010-10-21 10:29:19 +01002033EXPORT_SYMBOL(kmem_cache_alloc_trace);
2034
2035void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2036{
2037 void *ret = kmalloc_order(size, flags, order);
2038 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2039 return ret;
2040}
2041EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002042#endif
2043
Christoph Lameter81819f02007-05-06 14:49:36 -07002044#ifdef CONFIG_NUMA
2045void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2046{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002047 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2048
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002049 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2050 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002051
2052 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002053}
2054EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002055
Li Zefan0f24f122009-12-11 15:45:30 +08002056#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002057void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002058 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002059 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002060{
Richard Kennedy4a923792010-10-21 10:29:19 +01002061 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2062
2063 trace_kmalloc_node(_RET_IP_, ret,
2064 size, s->size, gfpflags, node);
2065 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002066}
Richard Kennedy4a923792010-10-21 10:29:19 +01002067EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002068#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002069#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002070
Christoph Lameter81819f02007-05-06 14:49:36 -07002071/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002072 * Slow patch handling. This may still be called frequently since objects
2073 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002074 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002075 * So we still attempt to reduce cache line usage. Just take the slab
2076 * lock and free the item. If there is no additional partial page
2077 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002078 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002079static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002080 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002081{
2082 void *prior;
2083 void **object = (void *)x;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002084 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002085
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002086 local_irq_save(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002087 slab_lock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002088 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002089
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002090 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2091 goto out_unlock;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002092
Christoph Lameterff120592009-12-18 16:26:22 -06002093 prior = page->freelist;
2094 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07002095 page->freelist = object;
2096 page->inuse--;
2097
Christoph Lameter50d5c412011-06-01 12:25:45 -05002098 if (unlikely(page->frozen)) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002099 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07002100 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002101 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002102
2103 if (unlikely(!page->inuse))
2104 goto slab_empty;
2105
2106 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002107 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07002108 * then add it.
2109 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08002110 if (unlikely(!prior)) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002111 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2112
2113 spin_lock(&n->list_lock);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002114 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002115 spin_unlock(&n->list_lock);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002116 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002117 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002118
2119out_unlock:
2120 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002121 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002122 return;
2123
2124slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002125 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002126 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002127 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002128 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002129 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2130
2131 spin_lock(&n->list_lock);
2132 remove_partial(n, page);
2133 spin_unlock(&n->list_lock);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002134 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002135 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002136 slab_unlock(page);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002137 local_irq_restore(flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002138 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002139 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002140}
2141
Christoph Lameter894b8782007-05-10 03:15:16 -07002142/*
2143 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2144 * can perform fastpath freeing without additional function calls.
2145 *
2146 * The fastpath is only possible if we are freeing to the current cpu slab
2147 * of this processor. This typically the case if we have just allocated
2148 * the item before.
2149 *
2150 * If fastpath is not possible then fall back to __slab_free where we deal
2151 * with all sorts of special processing.
2152 */
Pekka Enberg06428782008-01-07 23:20:27 -08002153static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002154 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002155{
2156 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002157 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002158 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002159
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002160 slab_free_hook(s, x);
2161
Christoph Lametera24c5a02011-03-15 12:45:21 -05002162redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002163
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002164 /*
2165 * Determine the currently cpus per cpu slab.
2166 * The cpu may change afterward. However that does not matter since
2167 * data is retrieved via this pointer. If we are on the same cpu
2168 * during the cmpxchg then the free will succedd.
2169 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002170 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002171
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002172 tid = c->tid;
2173 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002174
Christoph Lameter442b06b2011-05-17 16:29:31 -05002175 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002176 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002177
Thomas Gleixner30106b82011-05-04 15:38:19 +02002178 if (unlikely(!irqsafe_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002179 s->cpu_slab->freelist, s->cpu_slab->tid,
2180 c->freelist, tid,
2181 object, next_tid(tid)))) {
2182
2183 note_cmpxchg_failure("slab_free", s, tid);
2184 goto redo;
2185 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002186 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002187 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002188 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002189
Christoph Lameter894b8782007-05-10 03:15:16 -07002190}
2191
Christoph Lameter81819f02007-05-06 14:49:36 -07002192void kmem_cache_free(struct kmem_cache *s, void *x)
2193{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002194 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002195
Christoph Lameterb49af682007-05-06 14:49:41 -07002196 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002197
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002198 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002199
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002200 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002201}
2202EXPORT_SYMBOL(kmem_cache_free);
2203
Christoph Lameter81819f02007-05-06 14:49:36 -07002204/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002205 * Object placement in a slab is made very easy because we always start at
2206 * offset 0. If we tune the size of the object to the alignment then we can
2207 * get the required alignment by putting one properly sized object after
2208 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002209 *
2210 * Notice that the allocation order determines the sizes of the per cpu
2211 * caches. Each processor has always one slab available for allocations.
2212 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002213 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002214 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002215 */
2216
2217/*
2218 * Mininum / Maximum order of slab pages. This influences locking overhead
2219 * and slab fragmentation. A higher order reduces the number of partial slabs
2220 * and increases the number of allocations possible without having to
2221 * take the list_lock.
2222 */
2223static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002224static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002225static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002226
2227/*
2228 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002229 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002230 */
2231static int slub_nomerge;
2232
2233/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002234 * Calculate the order of allocation given an slab object size.
2235 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002236 * The order of allocation has significant impact on performance and other
2237 * system components. Generally order 0 allocations should be preferred since
2238 * order 0 does not cause fragmentation in the page allocator. Larger objects
2239 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002240 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002241 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002242 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002243 * In order to reach satisfactory performance we must ensure that a minimum
2244 * number of objects is in one slab. Otherwise we may generate too much
2245 * activity on the partial lists which requires taking the list_lock. This is
2246 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002247 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002248 * slub_max_order specifies the order where we begin to stop considering the
2249 * number of objects in a slab as critical. If we reach slub_max_order then
2250 * we try to keep the page order as low as possible. So we accept more waste
2251 * of space in favor of a small page order.
2252 *
2253 * Higher order allocations also allow the placement of more objects in a
2254 * slab and thereby reduce object handling overhead. If the user has
2255 * requested a higher mininum order then we start with that one instead of
2256 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002257 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002258static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002259 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002260{
2261 int order;
2262 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002263 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002264
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002265 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002266 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002267
Christoph Lameter6300ea72007-07-17 04:03:20 -07002268 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002269 fls(min_objects * size - 1) - PAGE_SHIFT);
2270 order <= max_order; order++) {
2271
Christoph Lameter81819f02007-05-06 14:49:36 -07002272 unsigned long slab_size = PAGE_SIZE << order;
2273
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002274 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002275 continue;
2276
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002277 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002278
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002279 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002280 break;
2281
2282 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002283
Christoph Lameter81819f02007-05-06 14:49:36 -07002284 return order;
2285}
2286
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002287static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002288{
2289 int order;
2290 int min_objects;
2291 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002292 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002293
2294 /*
2295 * Attempt to find best configuration for a slab. This
2296 * works by first attempting to generate a layout with
2297 * the best configuration and backing off gradually.
2298 *
2299 * First we reduce the acceptable waste in a slab. Then
2300 * we reduce the minimum objects required in a slab.
2301 */
2302 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002303 if (!min_objects)
2304 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002305 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002306 min_objects = min(min_objects, max_objects);
2307
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002308 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002309 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002310 while (fraction >= 4) {
2311 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002312 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002313 if (order <= slub_max_order)
2314 return order;
2315 fraction /= 2;
2316 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002317 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002318 }
2319
2320 /*
2321 * We were unable to place multiple objects in a slab. Now
2322 * lets see if we can place a single object there.
2323 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002324 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002325 if (order <= slub_max_order)
2326 return order;
2327
2328 /*
2329 * Doh this slab cannot be placed using slub_max_order.
2330 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002331 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002332 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002333 return order;
2334 return -ENOSYS;
2335}
2336
Christoph Lameter81819f02007-05-06 14:49:36 -07002337/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002338 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002339 */
2340static unsigned long calculate_alignment(unsigned long flags,
2341 unsigned long align, unsigned long size)
2342{
2343 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002344 * If the user wants hardware cache aligned objects then follow that
2345 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002346 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002347 * The hardware cache alignment cannot override the specified
2348 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002349 */
Nick Pigginb6210382008-03-05 14:05:56 -08002350 if (flags & SLAB_HWCACHE_ALIGN) {
2351 unsigned long ralign = cache_line_size();
2352 while (size <= ralign / 2)
2353 ralign /= 2;
2354 align = max(align, ralign);
2355 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002356
2357 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002358 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002359
2360 return ALIGN(align, sizeof(void *));
2361}
2362
Pekka Enberg5595cff2008-08-05 09:28:47 +03002363static void
2364init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002365{
2366 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002367 spin_lock_init(&n->list_lock);
2368 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002369#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002370 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002371 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002372 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002373#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002374}
2375
Christoph Lameter55136592010-08-20 12:37:13 -05002376static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002377{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002378 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2379 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002380
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002381 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002382 * Must align to double word boundary for the double cmpxchg
2383 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002384 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002385 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2386 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002387
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002388 if (!s->cpu_slab)
2389 return 0;
2390
2391 init_kmem_cache_cpus(s);
2392
2393 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002394}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002395
Christoph Lameter51df1142010-08-20 12:37:15 -05002396static struct kmem_cache *kmem_cache_node;
2397
Christoph Lameter81819f02007-05-06 14:49:36 -07002398/*
2399 * No kmalloc_node yet so do it by hand. We know that this is the first
2400 * slab on the node for this slabcache. There are no concurrent accesses
2401 * possible.
2402 *
2403 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002404 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2405 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002406 */
Christoph Lameter55136592010-08-20 12:37:13 -05002407static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002408{
2409 struct page *page;
2410 struct kmem_cache_node *n;
2411
Christoph Lameter51df1142010-08-20 12:37:15 -05002412 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002413
Christoph Lameter51df1142010-08-20 12:37:15 -05002414 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002415
2416 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002417 if (page_to_nid(page) != node) {
2418 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2419 "node %d\n", node);
2420 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2421 "in order to be able to continue\n");
2422 }
2423
Christoph Lameter81819f02007-05-06 14:49:36 -07002424 n = page->freelist;
2425 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002426 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002427 page->inuse++;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002428 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002429 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002430#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002431 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002432 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002433#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002434 init_kmem_cache_node(n, kmem_cache_node);
2435 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002436
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002437 add_partial(n, page, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -07002438}
2439
2440static void free_kmem_cache_nodes(struct kmem_cache *s)
2441{
2442 int node;
2443
Christoph Lameterf64dc582007-10-16 01:25:33 -07002444 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002445 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002446
Alexander Duyck73367bd2010-05-21 14:41:35 -07002447 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002448 kmem_cache_free(kmem_cache_node, n);
2449
Christoph Lameter81819f02007-05-06 14:49:36 -07002450 s->node[node] = NULL;
2451 }
2452}
2453
Christoph Lameter55136592010-08-20 12:37:13 -05002454static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002455{
2456 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002457
Christoph Lameterf64dc582007-10-16 01:25:33 -07002458 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002459 struct kmem_cache_node *n;
2460
Alexander Duyck73367bd2010-05-21 14:41:35 -07002461 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002462 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002463 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002464 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002465 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002466 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002467
2468 if (!n) {
2469 free_kmem_cache_nodes(s);
2470 return 0;
2471 }
2472
Christoph Lameter81819f02007-05-06 14:49:36 -07002473 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002474 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002475 }
2476 return 1;
2477}
Christoph Lameter81819f02007-05-06 14:49:36 -07002478
David Rientjesc0bdb232009-02-25 09:16:35 +02002479static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002480{
2481 if (min < MIN_PARTIAL)
2482 min = MIN_PARTIAL;
2483 else if (min > MAX_PARTIAL)
2484 min = MAX_PARTIAL;
2485 s->min_partial = min;
2486}
2487
Christoph Lameter81819f02007-05-06 14:49:36 -07002488/*
2489 * calculate_sizes() determines the order and the distribution of data within
2490 * a slab object.
2491 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002492static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002493{
2494 unsigned long flags = s->flags;
2495 unsigned long size = s->objsize;
2496 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002497 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002498
2499 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002500 * Round up object size to the next word boundary. We can only
2501 * place the free pointer at word boundaries and this determines
2502 * the possible location of the free pointer.
2503 */
2504 size = ALIGN(size, sizeof(void *));
2505
2506#ifdef CONFIG_SLUB_DEBUG
2507 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002508 * Determine if we can poison the object itself. If the user of
2509 * the slab may touch the object after free or before allocation
2510 * then we should never poison the object itself.
2511 */
2512 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002513 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002514 s->flags |= __OBJECT_POISON;
2515 else
2516 s->flags &= ~__OBJECT_POISON;
2517
Christoph Lameter81819f02007-05-06 14:49:36 -07002518
2519 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002520 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002521 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002522 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002523 */
2524 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2525 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002526#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002527
2528 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002529 * With that we have determined the number of bytes in actual use
2530 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002531 */
2532 s->inuse = size;
2533
2534 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002535 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002536 /*
2537 * Relocate free pointer after the object if it is not
2538 * permitted to overwrite the first word of the object on
2539 * kmem_cache_free.
2540 *
2541 * This is the case if we do RCU, have a constructor or
2542 * destructor or are poisoning the objects.
2543 */
2544 s->offset = size;
2545 size += sizeof(void *);
2546 }
2547
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002548#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002549 if (flags & SLAB_STORE_USER)
2550 /*
2551 * Need to store information about allocs and frees after
2552 * the object.
2553 */
2554 size += 2 * sizeof(struct track);
2555
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002556 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002557 /*
2558 * Add some empty padding so that we can catch
2559 * overwrites from earlier objects rather than let
2560 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002561 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002562 * of the object.
2563 */
2564 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002565#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002566
Christoph Lameter81819f02007-05-06 14:49:36 -07002567 /*
2568 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002569 * user specified and the dynamic determination of cache line size
2570 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002571 */
2572 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002573 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002574
2575 /*
2576 * SLUB stores one object immediately after another beginning from
2577 * offset 0. In order to align the objects we have to simply size
2578 * each object to conform to the alignment.
2579 */
2580 size = ALIGN(size, align);
2581 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002582 if (forced_order >= 0)
2583 order = forced_order;
2584 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002585 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002586
Christoph Lameter834f3d12008-04-14 19:11:31 +03002587 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002588 return 0;
2589
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002590 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002591 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002592 s->allocflags |= __GFP_COMP;
2593
2594 if (s->flags & SLAB_CACHE_DMA)
2595 s->allocflags |= SLUB_DMA;
2596
2597 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2598 s->allocflags |= __GFP_RECLAIMABLE;
2599
Christoph Lameter81819f02007-05-06 14:49:36 -07002600 /*
2601 * Determine the number of objects per slab
2602 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002603 s->oo = oo_make(order, size, s->reserved);
2604 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03002605 if (oo_objects(s->oo) > oo_objects(s->max))
2606 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002607
Christoph Lameter834f3d12008-04-14 19:11:31 +03002608 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002609
2610}
2611
Christoph Lameter55136592010-08-20 12:37:13 -05002612static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002613 const char *name, size_t size,
2614 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002615 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002616{
2617 memset(s, 0, kmem_size);
2618 s->name = name;
2619 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002620 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002621 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002622 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002623 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002624
Lai Jiangshanda9a6382011-03-10 15:22:00 +08002625 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2626 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07002627
Christoph Lameter06b285d2008-04-14 19:11:41 +03002628 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002629 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002630 if (disable_higher_order_debug) {
2631 /*
2632 * Disable debugging flags that store metadata if the min slab
2633 * order increased.
2634 */
2635 if (get_order(s->size) > get_order(s->objsize)) {
2636 s->flags &= ~DEBUG_METADATA_FLAGS;
2637 s->offset = 0;
2638 if (!calculate_sizes(s, -1))
2639 goto error;
2640 }
2641 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002642
Christoph Lameterb789ef52011-06-01 12:25:49 -05002643#ifdef CONFIG_CMPXCHG_DOUBLE
2644 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
2645 /* Enable fast mode */
2646 s->flags |= __CMPXCHG_DOUBLE;
2647#endif
2648
David Rientjes3b89d7d2009-02-22 17:40:07 -08002649 /*
2650 * The larger the object size is, the more pages we want on the partial
2651 * list to avoid pounding the page allocator excessively.
2652 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002653 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002654 s->refcount = 1;
2655#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002656 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002657#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002658 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002659 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002660
Christoph Lameter55136592010-08-20 12:37:13 -05002661 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002662 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002663
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002664 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002665error:
2666 if (flags & SLAB_PANIC)
2667 panic("Cannot create slab %s size=%lu realsize=%u "
2668 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002669 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002670 s->offset, flags);
2671 return 0;
2672}
Christoph Lameter81819f02007-05-06 14:49:36 -07002673
2674/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002675 * Determine the size of a slab object
2676 */
2677unsigned int kmem_cache_size(struct kmem_cache *s)
2678{
2679 return s->objsize;
2680}
2681EXPORT_SYMBOL(kmem_cache_size);
2682
Christoph Lameter33b12c32008-04-25 12:22:43 -07002683static void list_slab_objects(struct kmem_cache *s, struct page *page,
2684 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002685{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002686#ifdef CONFIG_SLUB_DEBUG
2687 void *addr = page_address(page);
2688 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002689 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2690 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002691 if (!map)
2692 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002693 slab_err(s, page, "%s", text);
2694 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002695
Christoph Lameter5f80b132011-04-15 14:48:13 -05002696 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002697 for_each_object(p, s, addr, page->objects) {
2698
2699 if (!test_bit(slab_index(p, s, addr), map)) {
2700 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2701 p, p - addr);
2702 print_tracking(s, p);
2703 }
2704 }
2705 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002706 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002707#endif
2708}
2709
Christoph Lameter81819f02007-05-06 14:49:36 -07002710/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002711 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002712 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002713static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002714{
Christoph Lameter81819f02007-05-06 14:49:36 -07002715 unsigned long flags;
2716 struct page *page, *h;
2717
2718 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002719 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002720 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002721 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002722 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002723 } else {
2724 list_slab_objects(s, page,
2725 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002726 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002727 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002728 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002729}
2730
2731/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002732 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002733 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002734static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002735{
2736 int node;
2737
2738 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002739 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002740 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002741 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002742 struct kmem_cache_node *n = get_node(s, node);
2743
Christoph Lameter599870b2008-04-23 12:36:52 -07002744 free_partial(s, n);
2745 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002746 return 1;
2747 }
2748 free_kmem_cache_nodes(s);
2749 return 0;
2750}
2751
2752/*
2753 * Close a cache and release the kmem_cache structure
2754 * (must be used for caches created using kmem_cache_create)
2755 */
2756void kmem_cache_destroy(struct kmem_cache *s)
2757{
2758 down_write(&slub_lock);
2759 s->refcount--;
2760 if (!s->refcount) {
2761 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002762 if (kmem_cache_close(s)) {
2763 printk(KERN_ERR "SLUB %s: %s called for cache that "
2764 "still has objects.\n", s->name, __func__);
2765 dump_stack();
2766 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002767 if (s->flags & SLAB_DESTROY_BY_RCU)
2768 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002769 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002770 }
2771 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002772}
2773EXPORT_SYMBOL(kmem_cache_destroy);
2774
2775/********************************************************************
2776 * Kmalloc subsystem
2777 *******************************************************************/
2778
Christoph Lameter51df1142010-08-20 12:37:15 -05002779struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002780EXPORT_SYMBOL(kmalloc_caches);
2781
Christoph Lameter51df1142010-08-20 12:37:15 -05002782static struct kmem_cache *kmem_cache;
2783
Christoph Lameter55136592010-08-20 12:37:13 -05002784#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002785static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002786#endif
2787
Christoph Lameter81819f02007-05-06 14:49:36 -07002788static int __init setup_slub_min_order(char *str)
2789{
Pekka Enberg06428782008-01-07 23:20:27 -08002790 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002791
2792 return 1;
2793}
2794
2795__setup("slub_min_order=", setup_slub_min_order);
2796
2797static int __init setup_slub_max_order(char *str)
2798{
Pekka Enberg06428782008-01-07 23:20:27 -08002799 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002800 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002801
2802 return 1;
2803}
2804
2805__setup("slub_max_order=", setup_slub_max_order);
2806
2807static int __init setup_slub_min_objects(char *str)
2808{
Pekka Enberg06428782008-01-07 23:20:27 -08002809 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002810
2811 return 1;
2812}
2813
2814__setup("slub_min_objects=", setup_slub_min_objects);
2815
2816static int __init setup_slub_nomerge(char *str)
2817{
2818 slub_nomerge = 1;
2819 return 1;
2820}
2821
2822__setup("slub_nomerge", setup_slub_nomerge);
2823
Christoph Lameter51df1142010-08-20 12:37:15 -05002824static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2825 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002826{
Christoph Lameter51df1142010-08-20 12:37:15 -05002827 struct kmem_cache *s;
2828
2829 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2830
Pekka Enberg83b519e2009-06-10 19:40:04 +03002831 /*
2832 * This function is called with IRQs disabled during early-boot on
2833 * single CPU so there's no need to take slub_lock here.
2834 */
Christoph Lameter55136592010-08-20 12:37:13 -05002835 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002836 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002837 goto panic;
2838
2839 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002840 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002841
2842panic:
2843 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002844 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002845}
2846
Christoph Lameterf1b26332007-07-17 04:03:26 -07002847/*
2848 * Conversion table for small slabs sizes / 8 to the index in the
2849 * kmalloc array. This is necessary for slabs < 192 since we have non power
2850 * of two cache sizes there. The size of larger slabs can be determined using
2851 * fls.
2852 */
2853static s8 size_index[24] = {
2854 3, /* 8 */
2855 4, /* 16 */
2856 5, /* 24 */
2857 5, /* 32 */
2858 6, /* 40 */
2859 6, /* 48 */
2860 6, /* 56 */
2861 6, /* 64 */
2862 1, /* 72 */
2863 1, /* 80 */
2864 1, /* 88 */
2865 1, /* 96 */
2866 7, /* 104 */
2867 7, /* 112 */
2868 7, /* 120 */
2869 7, /* 128 */
2870 2, /* 136 */
2871 2, /* 144 */
2872 2, /* 152 */
2873 2, /* 160 */
2874 2, /* 168 */
2875 2, /* 176 */
2876 2, /* 184 */
2877 2 /* 192 */
2878};
2879
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002880static inline int size_index_elem(size_t bytes)
2881{
2882 return (bytes - 1) / 8;
2883}
2884
Christoph Lameter81819f02007-05-06 14:49:36 -07002885static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2886{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002887 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002888
Christoph Lameterf1b26332007-07-17 04:03:26 -07002889 if (size <= 192) {
2890 if (!size)
2891 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002892
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002893 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002894 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002895 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002896
2897#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002898 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002899 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002900
Christoph Lameter81819f02007-05-06 14:49:36 -07002901#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002902 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002903}
2904
2905void *__kmalloc(size_t size, gfp_t flags)
2906{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002907 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002908 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002909
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002910 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002911 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002912
2913 s = get_slab(size, flags);
2914
2915 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002916 return s;
2917
Christoph Lameter2154a332010-07-09 14:07:10 -05002918 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002919
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002920 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002921
2922 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002923}
2924EXPORT_SYMBOL(__kmalloc);
2925
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002926#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002927static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2928{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002929 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002930 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002931
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002932 flags |= __GFP_COMP | __GFP_NOTRACK;
2933 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002934 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002935 ptr = page_address(page);
2936
2937 kmemleak_alloc(ptr, size, 1, flags);
2938 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002939}
2940
Christoph Lameter81819f02007-05-06 14:49:36 -07002941void *__kmalloc_node(size_t size, gfp_t flags, int node)
2942{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002943 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002944 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002945
Ingo Molnar057685c2009-02-20 12:15:30 +01002946 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002947 ret = kmalloc_large_node(size, flags, node);
2948
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002949 trace_kmalloc_node(_RET_IP_, ret,
2950 size, PAGE_SIZE << get_order(size),
2951 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002952
2953 return ret;
2954 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002955
2956 s = get_slab(size, flags);
2957
2958 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002959 return s;
2960
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002961 ret = slab_alloc(s, flags, node, _RET_IP_);
2962
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002963 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002964
2965 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002966}
2967EXPORT_SYMBOL(__kmalloc_node);
2968#endif
2969
2970size_t ksize(const void *object)
2971{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002972 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002973
Christoph Lameteref8b4522007-10-16 01:24:46 -07002974 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002975 return 0;
2976
Vegard Nossum294a80a2007-12-04 23:45:30 -08002977 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002978
Pekka Enberg76994412008-05-22 19:22:25 +03002979 if (unlikely(!PageSlab(page))) {
2980 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002981 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002982 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002983
Eric Dumazetb3d41882011-02-14 18:35:22 +01002984 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002985}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002986EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002987
2988void kfree(const void *x)
2989{
Christoph Lameter81819f02007-05-06 14:49:36 -07002990 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002991 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002992
Pekka Enberg2121db72009-03-25 11:05:57 +02002993 trace_kfree(_RET_IP_, x);
2994
Satyam Sharma2408c552007-10-16 01:24:44 -07002995 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002996 return;
2997
Christoph Lameterb49af682007-05-06 14:49:41 -07002998 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002999 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003000 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003001 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003002 put_page(page);
3003 return;
3004 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003005 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003006}
3007EXPORT_SYMBOL(kfree);
3008
Christoph Lameter2086d262007-05-06 14:49:46 -07003009/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003010 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3011 * the remaining slabs by the number of items in use. The slabs with the
3012 * most items in use come first. New allocations will then fill those up
3013 * and thus they can be removed from the partial lists.
3014 *
3015 * The slabs with the least items are placed last. This results in them
3016 * being allocated from last increasing the chance that the last objects
3017 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003018 */
3019int kmem_cache_shrink(struct kmem_cache *s)
3020{
3021 int node;
3022 int i;
3023 struct kmem_cache_node *n;
3024 struct page *page;
3025 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003026 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003027 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003028 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003029 unsigned long flags;
3030
3031 if (!slabs_by_inuse)
3032 return -ENOMEM;
3033
3034 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003035 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003036 n = get_node(s, node);
3037
3038 if (!n->nr_partial)
3039 continue;
3040
Christoph Lameter834f3d12008-04-14 19:11:31 +03003041 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003042 INIT_LIST_HEAD(slabs_by_inuse + i);
3043
3044 spin_lock_irqsave(&n->list_lock, flags);
3045
3046 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003047 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003048 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003049 * Note that concurrent frees may occur while we hold the
3050 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003051 */
3052 list_for_each_entry_safe(page, t, &n->partial, lru) {
3053 if (!page->inuse && slab_trylock(page)) {
3054 /*
3055 * Must hold slab lock here because slab_free
3056 * may have freed the last object and be
3057 * waiting to release the slab.
3058 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003059 remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003060 slab_unlock(page);
3061 discard_slab(s, page);
3062 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07003063 list_move(&page->lru,
3064 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07003065 }
3066 }
3067
Christoph Lameter2086d262007-05-06 14:49:46 -07003068 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003069 * Rebuild the partial list with the slabs filled up most
3070 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003071 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003072 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003073 list_splice(slabs_by_inuse + i, n->partial.prev);
3074
Christoph Lameter2086d262007-05-06 14:49:46 -07003075 spin_unlock_irqrestore(&n->list_lock, flags);
3076 }
3077
3078 kfree(slabs_by_inuse);
3079 return 0;
3080}
3081EXPORT_SYMBOL(kmem_cache_shrink);
3082
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003083#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003084static int slab_mem_going_offline_callback(void *arg)
3085{
3086 struct kmem_cache *s;
3087
3088 down_read(&slub_lock);
3089 list_for_each_entry(s, &slab_caches, list)
3090 kmem_cache_shrink(s);
3091 up_read(&slub_lock);
3092
3093 return 0;
3094}
3095
3096static void slab_mem_offline_callback(void *arg)
3097{
3098 struct kmem_cache_node *n;
3099 struct kmem_cache *s;
3100 struct memory_notify *marg = arg;
3101 int offline_node;
3102
3103 offline_node = marg->status_change_nid;
3104
3105 /*
3106 * If the node still has available memory. we need kmem_cache_node
3107 * for it yet.
3108 */
3109 if (offline_node < 0)
3110 return;
3111
3112 down_read(&slub_lock);
3113 list_for_each_entry(s, &slab_caches, list) {
3114 n = get_node(s, offline_node);
3115 if (n) {
3116 /*
3117 * if n->nr_slabs > 0, slabs still exist on the node
3118 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003119 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003120 * callback. So, we must fail.
3121 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003122 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003123
3124 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003125 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003126 }
3127 }
3128 up_read(&slub_lock);
3129}
3130
3131static int slab_mem_going_online_callback(void *arg)
3132{
3133 struct kmem_cache_node *n;
3134 struct kmem_cache *s;
3135 struct memory_notify *marg = arg;
3136 int nid = marg->status_change_nid;
3137 int ret = 0;
3138
3139 /*
3140 * If the node's memory is already available, then kmem_cache_node is
3141 * already created. Nothing to do.
3142 */
3143 if (nid < 0)
3144 return 0;
3145
3146 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003147 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003148 * allocate a kmem_cache_node structure in order to bring the node
3149 * online.
3150 */
3151 down_read(&slub_lock);
3152 list_for_each_entry(s, &slab_caches, list) {
3153 /*
3154 * XXX: kmem_cache_alloc_node will fallback to other nodes
3155 * since memory is not yet available from the node that
3156 * is brought up.
3157 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003158 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003159 if (!n) {
3160 ret = -ENOMEM;
3161 goto out;
3162 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003163 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003164 s->node[nid] = n;
3165 }
3166out:
3167 up_read(&slub_lock);
3168 return ret;
3169}
3170
3171static int slab_memory_callback(struct notifier_block *self,
3172 unsigned long action, void *arg)
3173{
3174 int ret = 0;
3175
3176 switch (action) {
3177 case MEM_GOING_ONLINE:
3178 ret = slab_mem_going_online_callback(arg);
3179 break;
3180 case MEM_GOING_OFFLINE:
3181 ret = slab_mem_going_offline_callback(arg);
3182 break;
3183 case MEM_OFFLINE:
3184 case MEM_CANCEL_ONLINE:
3185 slab_mem_offline_callback(arg);
3186 break;
3187 case MEM_ONLINE:
3188 case MEM_CANCEL_OFFLINE:
3189 break;
3190 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003191 if (ret)
3192 ret = notifier_from_errno(ret);
3193 else
3194 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003195 return ret;
3196}
3197
3198#endif /* CONFIG_MEMORY_HOTPLUG */
3199
Christoph Lameter81819f02007-05-06 14:49:36 -07003200/********************************************************************
3201 * Basic setup of slabs
3202 *******************************************************************/
3203
Christoph Lameter51df1142010-08-20 12:37:15 -05003204/*
3205 * Used for early kmem_cache structures that were allocated using
3206 * the page allocator
3207 */
3208
3209static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3210{
3211 int node;
3212
3213 list_add(&s->list, &slab_caches);
3214 s->refcount = -1;
3215
3216 for_each_node_state(node, N_NORMAL_MEMORY) {
3217 struct kmem_cache_node *n = get_node(s, node);
3218 struct page *p;
3219
3220 if (n) {
3221 list_for_each_entry(p, &n->partial, lru)
3222 p->slab = s;
3223
Li Zefan607bf322011-04-12 15:22:26 +08003224#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003225 list_for_each_entry(p, &n->full, lru)
3226 p->slab = s;
3227#endif
3228 }
3229 }
3230}
3231
Christoph Lameter81819f02007-05-06 14:49:36 -07003232void __init kmem_cache_init(void)
3233{
3234 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003235 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003236 struct kmem_cache *temp_kmem_cache;
3237 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003238 struct kmem_cache *temp_kmem_cache_node;
3239 unsigned long kmalloc_size;
3240
3241 kmem_size = offsetof(struct kmem_cache, node) +
3242 nr_node_ids * sizeof(struct kmem_cache_node *);
3243
3244 /* Allocate two kmem_caches from the page allocator */
3245 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3246 order = get_order(2 * kmalloc_size);
3247 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3248
Christoph Lameter81819f02007-05-06 14:49:36 -07003249 /*
3250 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003251 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003252 * kmem_cache_open for slab_state == DOWN.
3253 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003254 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3255
3256 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3257 sizeof(struct kmem_cache_node),
3258 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003259
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003260 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003261
3262 /* Able to allocate the per node structures */
3263 slab_state = PARTIAL;
3264
Christoph Lameter51df1142010-08-20 12:37:15 -05003265 temp_kmem_cache = kmem_cache;
3266 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3267 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3268 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3269 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003270
Christoph Lameter51df1142010-08-20 12:37:15 -05003271 /*
3272 * Allocate kmem_cache_node properly from the kmem_cache slab.
3273 * kmem_cache_node is separately allocated so no need to
3274 * update any list pointers.
3275 */
3276 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003277
Christoph Lameter51df1142010-08-20 12:37:15 -05003278 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3279 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3280
3281 kmem_cache_bootstrap_fixup(kmem_cache_node);
3282
3283 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003284 kmem_cache_bootstrap_fixup(kmem_cache);
3285 caches++;
3286 /* Free temporary boot structure */
3287 free_pages((unsigned long)temp_kmem_cache, order);
3288
3289 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003290
3291 /*
3292 * Patch up the size_index table if we have strange large alignment
3293 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003294 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003295 *
3296 * Largest permitted alignment is 256 bytes due to the way we
3297 * handle the index determination for the smaller caches.
3298 *
3299 * Make sure that nothing crazy happens if someone starts tinkering
3300 * around with ARCH_KMALLOC_MINALIGN
3301 */
3302 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3303 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3304
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003305 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3306 int elem = size_index_elem(i);
3307 if (elem >= ARRAY_SIZE(size_index))
3308 break;
3309 size_index[elem] = KMALLOC_SHIFT_LOW;
3310 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003311
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003312 if (KMALLOC_MIN_SIZE == 64) {
3313 /*
3314 * The 96 byte size cache is not used if the alignment
3315 * is 64 byte.
3316 */
3317 for (i = 64 + 8; i <= 96; i += 8)
3318 size_index[size_index_elem(i)] = 7;
3319 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003320 /*
3321 * The 192 byte sized cache is not used if the alignment
3322 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3323 * instead.
3324 */
3325 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003326 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003327 }
3328
Christoph Lameter51df1142010-08-20 12:37:15 -05003329 /* Caches that are not of the two-to-the-power-of size */
3330 if (KMALLOC_MIN_SIZE <= 32) {
3331 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3332 caches++;
3333 }
3334
3335 if (KMALLOC_MIN_SIZE <= 64) {
3336 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3337 caches++;
3338 }
3339
3340 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3341 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3342 caches++;
3343 }
3344
Christoph Lameter81819f02007-05-06 14:49:36 -07003345 slab_state = UP;
3346
3347 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003348 if (KMALLOC_MIN_SIZE <= 32) {
3349 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3350 BUG_ON(!kmalloc_caches[1]->name);
3351 }
3352
3353 if (KMALLOC_MIN_SIZE <= 64) {
3354 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3355 BUG_ON(!kmalloc_caches[2]->name);
3356 }
3357
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003358 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3359 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3360
3361 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003362 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003363 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003364
3365#ifdef CONFIG_SMP
3366 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003367#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003368
Christoph Lameter55136592010-08-20 12:37:13 -05003369#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003370 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3371 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003372
Christoph Lameter51df1142010-08-20 12:37:15 -05003373 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003374 char *name = kasprintf(GFP_NOWAIT,
3375 "dma-kmalloc-%d", s->objsize);
3376
3377 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003378 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3379 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003380 }
3381 }
3382#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003383 printk(KERN_INFO
3384 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003385 " CPUs=%d, Nodes=%d\n",
3386 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003387 slub_min_order, slub_max_order, slub_min_objects,
3388 nr_cpu_ids, nr_node_ids);
3389}
3390
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003391void __init kmem_cache_init_late(void)
3392{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003393}
3394
Christoph Lameter81819f02007-05-06 14:49:36 -07003395/*
3396 * Find a mergeable slab cache
3397 */
3398static int slab_unmergeable(struct kmem_cache *s)
3399{
3400 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3401 return 1;
3402
Christoph Lameterc59def92007-05-16 22:10:50 -07003403 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003404 return 1;
3405
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003406 /*
3407 * We may have set a slab to be unmergeable during bootstrap.
3408 */
3409 if (s->refcount < 0)
3410 return 1;
3411
Christoph Lameter81819f02007-05-06 14:49:36 -07003412 return 0;
3413}
3414
3415static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003416 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003417 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003418{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003419 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003420
3421 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3422 return NULL;
3423
Christoph Lameterc59def92007-05-16 22:10:50 -07003424 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003425 return NULL;
3426
3427 size = ALIGN(size, sizeof(void *));
3428 align = calculate_alignment(flags, align, size);
3429 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003430 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003431
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003432 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003433 if (slab_unmergeable(s))
3434 continue;
3435
3436 if (size > s->size)
3437 continue;
3438
Christoph Lameterba0268a2007-09-11 15:24:11 -07003439 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003440 continue;
3441 /*
3442 * Check if alignment is compatible.
3443 * Courtesy of Adrian Drzewiecki
3444 */
Pekka Enberg06428782008-01-07 23:20:27 -08003445 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003446 continue;
3447
3448 if (s->size - size >= sizeof(void *))
3449 continue;
3450
3451 return s;
3452 }
3453 return NULL;
3454}
3455
3456struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003457 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003458{
3459 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003460 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003461
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003462 if (WARN_ON(!name))
3463 return NULL;
3464
Christoph Lameter81819f02007-05-06 14:49:36 -07003465 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003466 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003467 if (s) {
3468 s->refcount++;
3469 /*
3470 * Adjust the object sizes so that we clear
3471 * the complete object on kzalloc.
3472 */
3473 s->objsize = max(s->objsize, (int)size);
3474 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003475
David Rientjes7b8f3b62008-12-17 22:09:46 -08003476 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003477 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003478 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003479 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003480 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003481 return s;
3482 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003483
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003484 n = kstrdup(name, GFP_KERNEL);
3485 if (!n)
3486 goto err;
3487
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003488 s = kmalloc(kmem_size, GFP_KERNEL);
3489 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003490 if (kmem_cache_open(s, n,
Christoph Lameterc59def92007-05-16 22:10:50 -07003491 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003492 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003493 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003494 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003495 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003496 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003497 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003498 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003499 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003500 return s;
3501 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003502 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003503 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003504 }
Pavel Emelyanov68cee4f12010-10-28 13:50:37 +04003505err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003506 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003507
Christoph Lameter81819f02007-05-06 14:49:36 -07003508 if (flags & SLAB_PANIC)
3509 panic("Cannot create slabcache %s\n", name);
3510 else
3511 s = NULL;
3512 return s;
3513}
3514EXPORT_SYMBOL(kmem_cache_create);
3515
Christoph Lameter81819f02007-05-06 14:49:36 -07003516#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003517/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003518 * Use the cpu notifier to insure that the cpu slabs are flushed when
3519 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003520 */
3521static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3522 unsigned long action, void *hcpu)
3523{
3524 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003525 struct kmem_cache *s;
3526 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003527
3528 switch (action) {
3529 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003530 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003531 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003532 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003533 down_read(&slub_lock);
3534 list_for_each_entry(s, &slab_caches, list) {
3535 local_irq_save(flags);
3536 __flush_cpu_slab(s, cpu);
3537 local_irq_restore(flags);
3538 }
3539 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003540 break;
3541 default:
3542 break;
3543 }
3544 return NOTIFY_OK;
3545}
3546
Pekka Enberg06428782008-01-07 23:20:27 -08003547static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003548 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003549};
Christoph Lameter81819f02007-05-06 14:49:36 -07003550
3551#endif
3552
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003553void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003554{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003555 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003556 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003557
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003558 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003559 return kmalloc_large(size, gfpflags);
3560
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003561 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003562
Satyam Sharma2408c552007-10-16 01:24:44 -07003563 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003564 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003565
Christoph Lameter2154a332010-07-09 14:07:10 -05003566 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003567
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003568 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003569 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003570
3571 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003572}
3573
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003574#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003575void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003576 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003577{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003578 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003579 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003580
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003581 if (unlikely(size > SLUB_MAX_SIZE)) {
3582 ret = kmalloc_large_node(size, gfpflags, node);
3583
3584 trace_kmalloc_node(caller, ret,
3585 size, PAGE_SIZE << get_order(size),
3586 gfpflags, node);
3587
3588 return ret;
3589 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003590
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003591 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003592
Satyam Sharma2408c552007-10-16 01:24:44 -07003593 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003594 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003595
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003596 ret = slab_alloc(s, gfpflags, node, caller);
3597
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003598 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003599 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003600
3601 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003602}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003603#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003604
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003605#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003606static int count_inuse(struct page *page)
3607{
3608 return page->inuse;
3609}
3610
3611static int count_total(struct page *page)
3612{
3613 return page->objects;
3614}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003615#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003616
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003617#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003618static int validate_slab(struct kmem_cache *s, struct page *page,
3619 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003620{
3621 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003622 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003623
3624 if (!check_slab(s, page) ||
3625 !on_freelist(s, page, NULL))
3626 return 0;
3627
3628 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003629 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003630
Christoph Lameter5f80b132011-04-15 14:48:13 -05003631 get_map(s, page, map);
3632 for_each_object(p, s, addr, page->objects) {
3633 if (test_bit(slab_index(p, s, addr), map))
3634 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3635 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003636 }
3637
Christoph Lameter224a88b2008-04-14 19:11:31 +03003638 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003639 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003640 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003641 return 0;
3642 return 1;
3643}
3644
Christoph Lameter434e2452007-07-17 04:03:30 -07003645static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3646 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003647{
3648 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003649 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003650 slab_unlock(page);
3651 } else
3652 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3653 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003654}
3655
Christoph Lameter434e2452007-07-17 04:03:30 -07003656static int validate_slab_node(struct kmem_cache *s,
3657 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003658{
3659 unsigned long count = 0;
3660 struct page *page;
3661 unsigned long flags;
3662
3663 spin_lock_irqsave(&n->list_lock, flags);
3664
3665 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003666 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003667 count++;
3668 }
3669 if (count != n->nr_partial)
3670 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3671 "counter=%ld\n", s->name, count, n->nr_partial);
3672
3673 if (!(s->flags & SLAB_STORE_USER))
3674 goto out;
3675
3676 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003677 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003678 count++;
3679 }
3680 if (count != atomic_long_read(&n->nr_slabs))
3681 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3682 "counter=%ld\n", s->name, count,
3683 atomic_long_read(&n->nr_slabs));
3684
3685out:
3686 spin_unlock_irqrestore(&n->list_lock, flags);
3687 return count;
3688}
3689
Christoph Lameter434e2452007-07-17 04:03:30 -07003690static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003691{
3692 int node;
3693 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003694 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003695 sizeof(unsigned long), GFP_KERNEL);
3696
3697 if (!map)
3698 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003699
3700 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003701 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003702 struct kmem_cache_node *n = get_node(s, node);
3703
Christoph Lameter434e2452007-07-17 04:03:30 -07003704 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003705 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003706 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003707 return count;
3708}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003709/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003710 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003711 * and freed.
3712 */
3713
3714struct location {
3715 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003716 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003717 long long sum_time;
3718 long min_time;
3719 long max_time;
3720 long min_pid;
3721 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303722 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003723 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003724};
3725
3726struct loc_track {
3727 unsigned long max;
3728 unsigned long count;
3729 struct location *loc;
3730};
3731
3732static void free_loc_track(struct loc_track *t)
3733{
3734 if (t->max)
3735 free_pages((unsigned long)t->loc,
3736 get_order(sizeof(struct location) * t->max));
3737}
3738
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003739static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003740{
3741 struct location *l;
3742 int order;
3743
Christoph Lameter88a420e2007-05-06 14:49:45 -07003744 order = get_order(sizeof(struct location) * max);
3745
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003746 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003747 if (!l)
3748 return 0;
3749
3750 if (t->count) {
3751 memcpy(l, t->loc, sizeof(struct location) * t->count);
3752 free_loc_track(t);
3753 }
3754 t->max = max;
3755 t->loc = l;
3756 return 1;
3757}
3758
3759static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003760 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003761{
3762 long start, end, pos;
3763 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003764 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003765 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003766
3767 start = -1;
3768 end = t->count;
3769
3770 for ( ; ; ) {
3771 pos = start + (end - start + 1) / 2;
3772
3773 /*
3774 * There is nothing at "end". If we end up there
3775 * we need to add something to before end.
3776 */
3777 if (pos == end)
3778 break;
3779
3780 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003781 if (track->addr == caddr) {
3782
3783 l = &t->loc[pos];
3784 l->count++;
3785 if (track->when) {
3786 l->sum_time += age;
3787 if (age < l->min_time)
3788 l->min_time = age;
3789 if (age > l->max_time)
3790 l->max_time = age;
3791
3792 if (track->pid < l->min_pid)
3793 l->min_pid = track->pid;
3794 if (track->pid > l->max_pid)
3795 l->max_pid = track->pid;
3796
Rusty Russell174596a2009-01-01 10:12:29 +10303797 cpumask_set_cpu(track->cpu,
3798 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003799 }
3800 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003801 return 1;
3802 }
3803
Christoph Lameter45edfa52007-05-09 02:32:45 -07003804 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003805 end = pos;
3806 else
3807 start = pos;
3808 }
3809
3810 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003811 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003812 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003813 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003814 return 0;
3815
3816 l = t->loc + pos;
3817 if (pos < t->count)
3818 memmove(l + 1, l,
3819 (t->count - pos) * sizeof(struct location));
3820 t->count++;
3821 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003822 l->addr = track->addr;
3823 l->sum_time = age;
3824 l->min_time = age;
3825 l->max_time = age;
3826 l->min_pid = track->pid;
3827 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303828 cpumask_clear(to_cpumask(l->cpus));
3829 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003830 nodes_clear(l->nodes);
3831 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003832 return 1;
3833}
3834
3835static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003836 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003837 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003838{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003839 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003840 void *p;
3841
Christoph Lameter39b26462008-04-14 19:11:30 +03003842 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05003843 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003844
Christoph Lameter224a88b2008-04-14 19:11:31 +03003845 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003846 if (!test_bit(slab_index(p, s, addr), map))
3847 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003848}
3849
3850static int list_locations(struct kmem_cache *s, char *buf,
3851 enum track_item alloc)
3852{
Harvey Harrisone374d482008-01-31 15:20:50 -08003853 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003854 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003855 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003856 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003857 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3858 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003859
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003860 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3861 GFP_TEMPORARY)) {
3862 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003863 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003864 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003865 /* Push back cpu slabs */
3866 flush_all(s);
3867
Christoph Lameterf64dc582007-10-16 01:25:33 -07003868 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003869 struct kmem_cache_node *n = get_node(s, node);
3870 unsigned long flags;
3871 struct page *page;
3872
Christoph Lameter9e869432007-08-22 14:01:56 -07003873 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003874 continue;
3875
3876 spin_lock_irqsave(&n->list_lock, flags);
3877 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003878 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003879 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003880 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003881 spin_unlock_irqrestore(&n->list_lock, flags);
3882 }
3883
3884 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003885 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003886
Hugh Dickins9c246242008-12-09 13:14:27 -08003887 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003888 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003889 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003890
3891 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08003892 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003893 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003894 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003895
3896 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003897 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003898 l->min_time,
3899 (long)div_u64(l->sum_time, l->count),
3900 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003901 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003902 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003903 l->min_time);
3904
3905 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003906 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003907 l->min_pid, l->max_pid);
3908 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003909 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003910 l->min_pid);
3911
Rusty Russell174596a2009-01-01 10:12:29 +10303912 if (num_online_cpus() > 1 &&
3913 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003914 len < PAGE_SIZE - 60) {
3915 len += sprintf(buf + len, " cpus=");
3916 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303917 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003918 }
3919
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003920 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003921 len < PAGE_SIZE - 60) {
3922 len += sprintf(buf + len, " nodes=");
3923 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003924 l->nodes);
3925 }
3926
Harvey Harrisone374d482008-01-31 15:20:50 -08003927 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003928 }
3929
3930 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003931 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003932 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003933 len += sprintf(buf, "No data\n");
3934 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003935}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003936#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07003937
Christoph Lametera5a84752010-10-05 13:57:27 -05003938#ifdef SLUB_RESILIENCY_TEST
3939static void resiliency_test(void)
3940{
3941 u8 *p;
3942
3943 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3944
3945 printk(KERN_ERR "SLUB resiliency testing\n");
3946 printk(KERN_ERR "-----------------------\n");
3947 printk(KERN_ERR "A. Corruption after allocation\n");
3948
3949 p = kzalloc(16, GFP_KERNEL);
3950 p[16] = 0x12;
3951 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3952 " 0x12->0x%p\n\n", p + 16);
3953
3954 validate_slab_cache(kmalloc_caches[4]);
3955
3956 /* Hmmm... The next two are dangerous */
3957 p = kzalloc(32, GFP_KERNEL);
3958 p[32 + sizeof(void *)] = 0x34;
3959 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3960 " 0x34 -> -0x%p\n", p);
3961 printk(KERN_ERR
3962 "If allocated object is overwritten then not detectable\n\n");
3963
3964 validate_slab_cache(kmalloc_caches[5]);
3965 p = kzalloc(64, GFP_KERNEL);
3966 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3967 *p = 0x56;
3968 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3969 p);
3970 printk(KERN_ERR
3971 "If allocated object is overwritten then not detectable\n\n");
3972 validate_slab_cache(kmalloc_caches[6]);
3973
3974 printk(KERN_ERR "\nB. Corruption after free\n");
3975 p = kzalloc(128, GFP_KERNEL);
3976 kfree(p);
3977 *p = 0x78;
3978 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3979 validate_slab_cache(kmalloc_caches[7]);
3980
3981 p = kzalloc(256, GFP_KERNEL);
3982 kfree(p);
3983 p[50] = 0x9a;
3984 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3985 p);
3986 validate_slab_cache(kmalloc_caches[8]);
3987
3988 p = kzalloc(512, GFP_KERNEL);
3989 kfree(p);
3990 p[512] = 0xab;
3991 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3992 validate_slab_cache(kmalloc_caches[9]);
3993}
3994#else
3995#ifdef CONFIG_SYSFS
3996static void resiliency_test(void) {};
3997#endif
3998#endif
3999
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004000#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004001enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004002 SL_ALL, /* All slabs */
4003 SL_PARTIAL, /* Only partially allocated slabs */
4004 SL_CPU, /* Only slabs used for cpu caches */
4005 SL_OBJECTS, /* Determine allocated objects not slabs */
4006 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004007};
4008
Christoph Lameter205ab992008-04-14 19:11:40 +03004009#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004010#define SO_PARTIAL (1 << SL_PARTIAL)
4011#define SO_CPU (1 << SL_CPU)
4012#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004013#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004014
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004015static ssize_t show_slab_objects(struct kmem_cache *s,
4016 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004017{
4018 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004019 int node;
4020 int x;
4021 unsigned long *nodes;
4022 unsigned long *per_cpu;
4023
4024 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004025 if (!nodes)
4026 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004027 per_cpu = nodes + nr_node_ids;
4028
Christoph Lameter205ab992008-04-14 19:11:40 +03004029 if (flags & SO_CPU) {
4030 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004031
Christoph Lameter205ab992008-04-14 19:11:40 +03004032 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004033 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004034
Christoph Lameter205ab992008-04-14 19:11:40 +03004035 if (!c || c->node < 0)
4036 continue;
4037
4038 if (c->page) {
4039 if (flags & SO_TOTAL)
4040 x = c->page->objects;
4041 else if (flags & SO_OBJECTS)
4042 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07004043 else
4044 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03004045
Christoph Lameter81819f02007-05-06 14:49:36 -07004046 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03004047 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004048 }
Christoph Lameter205ab992008-04-14 19:11:40 +03004049 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004050 }
4051 }
4052
Christoph Lameter04d94872011-01-10 10:15:15 -06004053 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004054#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004055 if (flags & SO_ALL) {
4056 for_each_node_state(node, N_NORMAL_MEMORY) {
4057 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004058
Christoph Lameter205ab992008-04-14 19:11:40 +03004059 if (flags & SO_TOTAL)
4060 x = atomic_long_read(&n->total_objects);
4061 else if (flags & SO_OBJECTS)
4062 x = atomic_long_read(&n->total_objects) -
4063 count_partial(n, count_free);
4064
4065 else
4066 x = atomic_long_read(&n->nr_slabs);
4067 total += x;
4068 nodes[node] += x;
4069 }
4070
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004071 } else
4072#endif
4073 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004074 for_each_node_state(node, N_NORMAL_MEMORY) {
4075 struct kmem_cache_node *n = get_node(s, node);
4076
4077 if (flags & SO_TOTAL)
4078 x = count_partial(n, count_total);
4079 else if (flags & SO_OBJECTS)
4080 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004081 else
4082 x = n->nr_partial;
4083 total += x;
4084 nodes[node] += x;
4085 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004086 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004087 x = sprintf(buf, "%lu", total);
4088#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004089 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004090 if (nodes[node])
4091 x += sprintf(buf + x, " N%d=%lu",
4092 node, nodes[node]);
4093#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004094 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004095 kfree(nodes);
4096 return x + sprintf(buf + x, "\n");
4097}
4098
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004099#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004100static int any_slab_objects(struct kmem_cache *s)
4101{
4102 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004103
4104 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004105 struct kmem_cache_node *n = get_node(s, node);
4106
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004107 if (!n)
4108 continue;
4109
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004110 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004111 return 1;
4112 }
4113 return 0;
4114}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004115#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004116
4117#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4118#define to_slab(n) container_of(n, struct kmem_cache, kobj);
4119
4120struct slab_attribute {
4121 struct attribute attr;
4122 ssize_t (*show)(struct kmem_cache *s, char *buf);
4123 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4124};
4125
4126#define SLAB_ATTR_RO(_name) \
4127 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4128
4129#define SLAB_ATTR(_name) \
4130 static struct slab_attribute _name##_attr = \
4131 __ATTR(_name, 0644, _name##_show, _name##_store)
4132
Christoph Lameter81819f02007-05-06 14:49:36 -07004133static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4134{
4135 return sprintf(buf, "%d\n", s->size);
4136}
4137SLAB_ATTR_RO(slab_size);
4138
4139static ssize_t align_show(struct kmem_cache *s, char *buf)
4140{
4141 return sprintf(buf, "%d\n", s->align);
4142}
4143SLAB_ATTR_RO(align);
4144
4145static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4146{
4147 return sprintf(buf, "%d\n", s->objsize);
4148}
4149SLAB_ATTR_RO(object_size);
4150
4151static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4152{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004153 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004154}
4155SLAB_ATTR_RO(objs_per_slab);
4156
Christoph Lameter06b285d2008-04-14 19:11:41 +03004157static ssize_t order_store(struct kmem_cache *s,
4158 const char *buf, size_t length)
4159{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004160 unsigned long order;
4161 int err;
4162
4163 err = strict_strtoul(buf, 10, &order);
4164 if (err)
4165 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004166
4167 if (order > slub_max_order || order < slub_min_order)
4168 return -EINVAL;
4169
4170 calculate_sizes(s, order);
4171 return length;
4172}
4173
Christoph Lameter81819f02007-05-06 14:49:36 -07004174static ssize_t order_show(struct kmem_cache *s, char *buf)
4175{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004176 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004177}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004178SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004179
David Rientjes73d342b2009-02-22 17:40:09 -08004180static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4181{
4182 return sprintf(buf, "%lu\n", s->min_partial);
4183}
4184
4185static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4186 size_t length)
4187{
4188 unsigned long min;
4189 int err;
4190
4191 err = strict_strtoul(buf, 10, &min);
4192 if (err)
4193 return err;
4194
David Rientjesc0bdb232009-02-25 09:16:35 +02004195 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004196 return length;
4197}
4198SLAB_ATTR(min_partial);
4199
Christoph Lameter81819f02007-05-06 14:49:36 -07004200static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4201{
Joe Perches62c70bc2011-01-13 15:45:52 -08004202 if (!s->ctor)
4203 return 0;
4204 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004205}
4206SLAB_ATTR_RO(ctor);
4207
Christoph Lameter81819f02007-05-06 14:49:36 -07004208static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4209{
4210 return sprintf(buf, "%d\n", s->refcount - 1);
4211}
4212SLAB_ATTR_RO(aliases);
4213
Christoph Lameter81819f02007-05-06 14:49:36 -07004214static ssize_t partial_show(struct kmem_cache *s, char *buf)
4215{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004216 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004217}
4218SLAB_ATTR_RO(partial);
4219
4220static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4221{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004222 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004223}
4224SLAB_ATTR_RO(cpu_slabs);
4225
4226static ssize_t objects_show(struct kmem_cache *s, char *buf)
4227{
Christoph Lameter205ab992008-04-14 19:11:40 +03004228 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004229}
4230SLAB_ATTR_RO(objects);
4231
Christoph Lameter205ab992008-04-14 19:11:40 +03004232static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4233{
4234 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4235}
4236SLAB_ATTR_RO(objects_partial);
4237
Christoph Lameter81819f02007-05-06 14:49:36 -07004238static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4239{
4240 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4241}
4242
4243static ssize_t reclaim_account_store(struct kmem_cache *s,
4244 const char *buf, size_t length)
4245{
4246 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4247 if (buf[0] == '1')
4248 s->flags |= SLAB_RECLAIM_ACCOUNT;
4249 return length;
4250}
4251SLAB_ATTR(reclaim_account);
4252
4253static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4254{
Christoph Lameter5af60832007-05-06 14:49:56 -07004255 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004256}
4257SLAB_ATTR_RO(hwcache_align);
4258
4259#ifdef CONFIG_ZONE_DMA
4260static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4261{
4262 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4263}
4264SLAB_ATTR_RO(cache_dma);
4265#endif
4266
4267static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4268{
4269 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4270}
4271SLAB_ATTR_RO(destroy_by_rcu);
4272
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004273static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4274{
4275 return sprintf(buf, "%d\n", s->reserved);
4276}
4277SLAB_ATTR_RO(reserved);
4278
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004279#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004280static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4281{
4282 return show_slab_objects(s, buf, SO_ALL);
4283}
4284SLAB_ATTR_RO(slabs);
4285
4286static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4287{
4288 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4289}
4290SLAB_ATTR_RO(total_objects);
4291
4292static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4293{
4294 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4295}
4296
4297static ssize_t sanity_checks_store(struct kmem_cache *s,
4298 const char *buf, size_t length)
4299{
4300 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004301 if (buf[0] == '1') {
4302 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004303 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004304 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004305 return length;
4306}
4307SLAB_ATTR(sanity_checks);
4308
4309static ssize_t trace_show(struct kmem_cache *s, char *buf)
4310{
4311 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4312}
4313
4314static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4315 size_t length)
4316{
4317 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004318 if (buf[0] == '1') {
4319 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004320 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004321 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004322 return length;
4323}
4324SLAB_ATTR(trace);
4325
Christoph Lameter81819f02007-05-06 14:49:36 -07004326static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4327{
4328 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4329}
4330
4331static ssize_t red_zone_store(struct kmem_cache *s,
4332 const char *buf, size_t length)
4333{
4334 if (any_slab_objects(s))
4335 return -EBUSY;
4336
4337 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004338 if (buf[0] == '1') {
4339 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004340 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004341 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004342 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004343 return length;
4344}
4345SLAB_ATTR(red_zone);
4346
4347static ssize_t poison_show(struct kmem_cache *s, char *buf)
4348{
4349 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4350}
4351
4352static ssize_t poison_store(struct kmem_cache *s,
4353 const char *buf, size_t length)
4354{
4355 if (any_slab_objects(s))
4356 return -EBUSY;
4357
4358 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004359 if (buf[0] == '1') {
4360 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004361 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004362 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004363 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004364 return length;
4365}
4366SLAB_ATTR(poison);
4367
4368static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4369{
4370 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4371}
4372
4373static ssize_t store_user_store(struct kmem_cache *s,
4374 const char *buf, size_t length)
4375{
4376 if (any_slab_objects(s))
4377 return -EBUSY;
4378
4379 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004380 if (buf[0] == '1') {
4381 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004382 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004383 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004384 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004385 return length;
4386}
4387SLAB_ATTR(store_user);
4388
Christoph Lameter53e15af2007-05-06 14:49:43 -07004389static ssize_t validate_show(struct kmem_cache *s, char *buf)
4390{
4391 return 0;
4392}
4393
4394static ssize_t validate_store(struct kmem_cache *s,
4395 const char *buf, size_t length)
4396{
Christoph Lameter434e2452007-07-17 04:03:30 -07004397 int ret = -EINVAL;
4398
4399 if (buf[0] == '1') {
4400 ret = validate_slab_cache(s);
4401 if (ret >= 0)
4402 ret = length;
4403 }
4404 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004405}
4406SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004407
4408static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4409{
4410 if (!(s->flags & SLAB_STORE_USER))
4411 return -ENOSYS;
4412 return list_locations(s, buf, TRACK_ALLOC);
4413}
4414SLAB_ATTR_RO(alloc_calls);
4415
4416static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4417{
4418 if (!(s->flags & SLAB_STORE_USER))
4419 return -ENOSYS;
4420 return list_locations(s, buf, TRACK_FREE);
4421}
4422SLAB_ATTR_RO(free_calls);
4423#endif /* CONFIG_SLUB_DEBUG */
4424
4425#ifdef CONFIG_FAILSLAB
4426static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4427{
4428 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4429}
4430
4431static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4432 size_t length)
4433{
4434 s->flags &= ~SLAB_FAILSLAB;
4435 if (buf[0] == '1')
4436 s->flags |= SLAB_FAILSLAB;
4437 return length;
4438}
4439SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004440#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004441
Christoph Lameter2086d262007-05-06 14:49:46 -07004442static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4443{
4444 return 0;
4445}
4446
4447static ssize_t shrink_store(struct kmem_cache *s,
4448 const char *buf, size_t length)
4449{
4450 if (buf[0] == '1') {
4451 int rc = kmem_cache_shrink(s);
4452
4453 if (rc)
4454 return rc;
4455 } else
4456 return -EINVAL;
4457 return length;
4458}
4459SLAB_ATTR(shrink);
4460
Christoph Lameter81819f02007-05-06 14:49:36 -07004461#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004462static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004463{
Christoph Lameter98246012008-01-07 23:20:26 -08004464 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004465}
4466
Christoph Lameter98246012008-01-07 23:20:26 -08004467static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004468 const char *buf, size_t length)
4469{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004470 unsigned long ratio;
4471 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004472
Christoph Lameter0121c6192008-04-29 16:11:12 -07004473 err = strict_strtoul(buf, 10, &ratio);
4474 if (err)
4475 return err;
4476
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004477 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004478 s->remote_node_defrag_ratio = ratio * 10;
4479
Christoph Lameter81819f02007-05-06 14:49:36 -07004480 return length;
4481}
Christoph Lameter98246012008-01-07 23:20:26 -08004482SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004483#endif
4484
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004485#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004486static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4487{
4488 unsigned long sum = 0;
4489 int cpu;
4490 int len;
4491 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4492
4493 if (!data)
4494 return -ENOMEM;
4495
4496 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004497 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004498
4499 data[cpu] = x;
4500 sum += x;
4501 }
4502
4503 len = sprintf(buf, "%lu", sum);
4504
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004505#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004506 for_each_online_cpu(cpu) {
4507 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004508 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004509 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004510#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004511 kfree(data);
4512 return len + sprintf(buf + len, "\n");
4513}
4514
David Rientjes78eb00c2009-10-15 02:20:22 -07004515static void clear_stat(struct kmem_cache *s, enum stat_item si)
4516{
4517 int cpu;
4518
4519 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004520 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004521}
4522
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004523#define STAT_ATTR(si, text) \
4524static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4525{ \
4526 return show_stat(s, buf, si); \
4527} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004528static ssize_t text##_store(struct kmem_cache *s, \
4529 const char *buf, size_t length) \
4530{ \
4531 if (buf[0] != '0') \
4532 return -EINVAL; \
4533 clear_stat(s, si); \
4534 return length; \
4535} \
4536SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004537
4538STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4539STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4540STAT_ATTR(FREE_FASTPATH, free_fastpath);
4541STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4542STAT_ATTR(FREE_FROZEN, free_frozen);
4543STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4544STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4545STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4546STAT_ATTR(ALLOC_SLAB, alloc_slab);
4547STAT_ATTR(ALLOC_REFILL, alloc_refill);
4548STAT_ATTR(FREE_SLAB, free_slab);
4549STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4550STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4551STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4552STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4553STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4554STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004555STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004556STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4557STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004558#endif
4559
Pekka Enberg06428782008-01-07 23:20:27 -08004560static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004561 &slab_size_attr.attr,
4562 &object_size_attr.attr,
4563 &objs_per_slab_attr.attr,
4564 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004565 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004566 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004567 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004568 &partial_attr.attr,
4569 &cpu_slabs_attr.attr,
4570 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004571 &aliases_attr.attr,
4572 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004573 &hwcache_align_attr.attr,
4574 &reclaim_account_attr.attr,
4575 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004576 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004577 &reserved_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004578#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004579 &total_objects_attr.attr,
4580 &slabs_attr.attr,
4581 &sanity_checks_attr.attr,
4582 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004583 &red_zone_attr.attr,
4584 &poison_attr.attr,
4585 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004586 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004587 &alloc_calls_attr.attr,
4588 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004589#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004590#ifdef CONFIG_ZONE_DMA
4591 &cache_dma_attr.attr,
4592#endif
4593#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004594 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004595#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004596#ifdef CONFIG_SLUB_STATS
4597 &alloc_fastpath_attr.attr,
4598 &alloc_slowpath_attr.attr,
4599 &free_fastpath_attr.attr,
4600 &free_slowpath_attr.attr,
4601 &free_frozen_attr.attr,
4602 &free_add_partial_attr.attr,
4603 &free_remove_partial_attr.attr,
4604 &alloc_from_partial_attr.attr,
4605 &alloc_slab_attr.attr,
4606 &alloc_refill_attr.attr,
4607 &free_slab_attr.attr,
4608 &cpuslab_flush_attr.attr,
4609 &deactivate_full_attr.attr,
4610 &deactivate_empty_attr.attr,
4611 &deactivate_to_head_attr.attr,
4612 &deactivate_to_tail_attr.attr,
4613 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004614 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05004615 &cmpxchg_double_fail_attr.attr,
4616 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004617#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004618#ifdef CONFIG_FAILSLAB
4619 &failslab_attr.attr,
4620#endif
4621
Christoph Lameter81819f02007-05-06 14:49:36 -07004622 NULL
4623};
4624
4625static struct attribute_group slab_attr_group = {
4626 .attrs = slab_attrs,
4627};
4628
4629static ssize_t slab_attr_show(struct kobject *kobj,
4630 struct attribute *attr,
4631 char *buf)
4632{
4633 struct slab_attribute *attribute;
4634 struct kmem_cache *s;
4635 int err;
4636
4637 attribute = to_slab_attr(attr);
4638 s = to_slab(kobj);
4639
4640 if (!attribute->show)
4641 return -EIO;
4642
4643 err = attribute->show(s, buf);
4644
4645 return err;
4646}
4647
4648static ssize_t slab_attr_store(struct kobject *kobj,
4649 struct attribute *attr,
4650 const char *buf, size_t len)
4651{
4652 struct slab_attribute *attribute;
4653 struct kmem_cache *s;
4654 int err;
4655
4656 attribute = to_slab_attr(attr);
4657 s = to_slab(kobj);
4658
4659 if (!attribute->store)
4660 return -EIO;
4661
4662 err = attribute->store(s, buf, len);
4663
4664 return err;
4665}
4666
Christoph Lameter151c6022008-01-07 22:29:05 -08004667static void kmem_cache_release(struct kobject *kobj)
4668{
4669 struct kmem_cache *s = to_slab(kobj);
4670
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004671 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004672 kfree(s);
4673}
4674
Emese Revfy52cf25d2010-01-19 02:58:23 +01004675static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004676 .show = slab_attr_show,
4677 .store = slab_attr_store,
4678};
4679
4680static struct kobj_type slab_ktype = {
4681 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004682 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004683};
4684
4685static int uevent_filter(struct kset *kset, struct kobject *kobj)
4686{
4687 struct kobj_type *ktype = get_ktype(kobj);
4688
4689 if (ktype == &slab_ktype)
4690 return 1;
4691 return 0;
4692}
4693
Emese Revfy9cd43612009-12-31 14:52:51 +01004694static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004695 .filter = uevent_filter,
4696};
4697
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004698static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004699
4700#define ID_STR_LENGTH 64
4701
4702/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004703 *
4704 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004705 */
4706static char *create_unique_id(struct kmem_cache *s)
4707{
4708 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4709 char *p = name;
4710
4711 BUG_ON(!name);
4712
4713 *p++ = ':';
4714 /*
4715 * First flags affecting slabcache operations. We will only
4716 * get here for aliasable slabs so we do not need to support
4717 * too many flags. The flags here must cover all flags that
4718 * are matched during merging to guarantee that the id is
4719 * unique.
4720 */
4721 if (s->flags & SLAB_CACHE_DMA)
4722 *p++ = 'd';
4723 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4724 *p++ = 'a';
4725 if (s->flags & SLAB_DEBUG_FREE)
4726 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004727 if (!(s->flags & SLAB_NOTRACK))
4728 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004729 if (p != name + 1)
4730 *p++ = '-';
4731 p += sprintf(p, "%07d", s->size);
4732 BUG_ON(p > name + ID_STR_LENGTH - 1);
4733 return name;
4734}
4735
4736static int sysfs_slab_add(struct kmem_cache *s)
4737{
4738 int err;
4739 const char *name;
4740 int unmergeable;
4741
4742 if (slab_state < SYSFS)
4743 /* Defer until later */
4744 return 0;
4745
4746 unmergeable = slab_unmergeable(s);
4747 if (unmergeable) {
4748 /*
4749 * Slabcache can never be merged so we can use the name proper.
4750 * This is typically the case for debug situations. In that
4751 * case we can catch duplicate names easily.
4752 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004753 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004754 name = s->name;
4755 } else {
4756 /*
4757 * Create a unique name for the slab as a target
4758 * for the symlinks.
4759 */
4760 name = create_unique_id(s);
4761 }
4762
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004763 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004764 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4765 if (err) {
4766 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004767 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004768 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004769
4770 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004771 if (err) {
4772 kobject_del(&s->kobj);
4773 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004774 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004775 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004776 kobject_uevent(&s->kobj, KOBJ_ADD);
4777 if (!unmergeable) {
4778 /* Setup first alias */
4779 sysfs_slab_alias(s, s->name);
4780 kfree(name);
4781 }
4782 return 0;
4783}
4784
4785static void sysfs_slab_remove(struct kmem_cache *s)
4786{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004787 if (slab_state < SYSFS)
4788 /*
4789 * Sysfs has not been setup yet so no need to remove the
4790 * cache from sysfs.
4791 */
4792 return;
4793
Christoph Lameter81819f02007-05-06 14:49:36 -07004794 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4795 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004796 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004797}
4798
4799/*
4800 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004801 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004802 */
4803struct saved_alias {
4804 struct kmem_cache *s;
4805 const char *name;
4806 struct saved_alias *next;
4807};
4808
Adrian Bunk5af328a2007-07-17 04:03:27 -07004809static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004810
4811static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4812{
4813 struct saved_alias *al;
4814
4815 if (slab_state == SYSFS) {
4816 /*
4817 * If we have a leftover link then remove it.
4818 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004819 sysfs_remove_link(&slab_kset->kobj, name);
4820 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004821 }
4822
4823 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4824 if (!al)
4825 return -ENOMEM;
4826
4827 al->s = s;
4828 al->name = name;
4829 al->next = alias_list;
4830 alias_list = al;
4831 return 0;
4832}
4833
4834static int __init slab_sysfs_init(void)
4835{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004836 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004837 int err;
4838
Christoph Lameter2bce6482010-07-19 11:39:11 -05004839 down_write(&slub_lock);
4840
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004841 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004842 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004843 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004844 printk(KERN_ERR "Cannot register slab subsystem.\n");
4845 return -ENOSYS;
4846 }
4847
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004848 slab_state = SYSFS;
4849
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004850 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004851 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004852 if (err)
4853 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4854 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004855 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004856
4857 while (alias_list) {
4858 struct saved_alias *al = alias_list;
4859
4860 alias_list = alias_list->next;
4861 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004862 if (err)
4863 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4864 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004865 kfree(al);
4866 }
4867
Christoph Lameter2bce6482010-07-19 11:39:11 -05004868 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004869 resiliency_test();
4870 return 0;
4871}
4872
4873__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004874#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004875
4876/*
4877 * The /proc/slabinfo ABI
4878 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004879#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004880static void print_slabinfo_header(struct seq_file *m)
4881{
4882 seq_puts(m, "slabinfo - version: 2.1\n");
4883 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4884 "<objperslab> <pagesperslab>");
4885 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4886 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4887 seq_putc(m, '\n');
4888}
4889
4890static void *s_start(struct seq_file *m, loff_t *pos)
4891{
4892 loff_t n = *pos;
4893
4894 down_read(&slub_lock);
4895 if (!n)
4896 print_slabinfo_header(m);
4897
4898 return seq_list_start(&slab_caches, *pos);
4899}
4900
4901static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4902{
4903 return seq_list_next(p, &slab_caches, pos);
4904}
4905
4906static void s_stop(struct seq_file *m, void *p)
4907{
4908 up_read(&slub_lock);
4909}
4910
4911static int s_show(struct seq_file *m, void *p)
4912{
4913 unsigned long nr_partials = 0;
4914 unsigned long nr_slabs = 0;
4915 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004916 unsigned long nr_objs = 0;
4917 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004918 struct kmem_cache *s;
4919 int node;
4920
4921 s = list_entry(p, struct kmem_cache, list);
4922
4923 for_each_online_node(node) {
4924 struct kmem_cache_node *n = get_node(s, node);
4925
4926 if (!n)
4927 continue;
4928
4929 nr_partials += n->nr_partial;
4930 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004931 nr_objs += atomic_long_read(&n->total_objects);
4932 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004933 }
4934
Christoph Lameter205ab992008-04-14 19:11:40 +03004935 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004936
4937 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004938 nr_objs, s->size, oo_objects(s->oo),
4939 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004940 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4941 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4942 0UL);
4943 seq_putc(m, '\n');
4944 return 0;
4945}
4946
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004947static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004948 .start = s_start,
4949 .next = s_next,
4950 .stop = s_stop,
4951 .show = s_show,
4952};
4953
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004954static int slabinfo_open(struct inode *inode, struct file *file)
4955{
4956 return seq_open(file, &slabinfo_op);
4957}
4958
4959static const struct file_operations proc_slabinfo_operations = {
4960 .open = slabinfo_open,
4961 .read = seq_read,
4962 .llseek = seq_lseek,
4963 .release = seq_release,
4964};
4965
4966static int __init slab_proc_init(void)
4967{
WANG Congcf5d1132009-08-18 19:11:40 +03004968 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004969 return 0;
4970}
4971module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004972#endif /* CONFIG_SLABINFO */