blob: ffe13fdf8144adadfcc0a55b976f412009e87b38 [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
Christoph Lameter881db7f2011-06-01 12:25:53 -05005 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07007 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter881db7f2011-06-01 12:25:53 -05009 * (C) 2011 Linux Foundation, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -070010 */
11
12#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100013#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070014#include <linux/module.h>
15#include <linux/bit_spinlock.h>
16#include <linux/interrupt.h>
17#include <linux/bitops.h>
18#include <linux/slab.h>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040019#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070020#include <linux/seq_file.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020021#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070022#include <linux/cpu.h>
23#include <linux/cpuset.h>
24#include <linux/mempolicy.h>
25#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070026#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070027#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070028#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070029#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090030#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030031#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060032#include <linux/prefetch.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070033
Richard Kennedy4a923792010-10-21 10:29:19 +010034#include <trace/events/kmem.h>
35
Christoph Lameter81819f02007-05-06 14:49:36 -070036/*
37 * Lock order:
Christoph Lameter881db7f2011-06-01 12:25:53 -050038 * 1. slub_lock (Global Semaphore)
39 * 2. node->list_lock
40 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070041 *
Christoph Lameter881db7f2011-06-01 12:25:53 -050042 * slub_lock
43 *
44 * The role of the slub_lock is to protect the list of all the slabs
45 * and to synchronize major metadata changes to slab cache structures.
46 *
47 * The slab_lock is only used for debugging and on arches that do not
48 * have the ability to do a cmpxchg_double. It only protects the second
49 * double word in the page struct. Meaning
50 * A. page->freelist -> List of object free in a page
51 * B. page->counters -> Counters of objects
52 * C. page->frozen -> frozen state
53 *
54 * If a slab is frozen then it is exempt from list management. It is not
55 * on any list. The processor that froze the slab is the one who can
56 * perform list operations on the page. Other processors may put objects
57 * onto the freelist but the processor that froze the slab is the only
58 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070059 *
60 * The list_lock protects the partial and full list on each node and
61 * the partial slab counter. If taken then no new slabs may be added or
62 * removed from the lists nor make the number of partial slabs be modified.
63 * (Note that the total number of slabs is an atomic value that may be
64 * modified without taking the list lock).
65 *
66 * The list_lock is a centralized lock and thus we avoid taking it as
67 * much as possible. As long as SLUB does not have to handle partial
68 * slabs, operations can continue without any centralized lock. F.e.
69 * allocating a long series of objects that fill up slabs does not require
70 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070071 * Interrupts are disabled during allocation and deallocation in order to
72 * make the slab allocator safe to use in the context of an irq. In addition
73 * interrupts are disabled to ensure that the processor does not change
74 * while handling per_cpu slabs, due to kernel preemption.
75 *
76 * SLUB assigns one slab for allocation to each processor.
77 * Allocations only occur from these slabs called cpu slabs.
78 *
Christoph Lameter672bba32007-05-09 02:32:39 -070079 * Slabs with free elements are kept on a partial list and during regular
80 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070081 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070082 * We track full slabs for debugging purposes though because otherwise we
83 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070084 *
85 * Slabs are freed when they become empty. Teardown and setup is
86 * minimal so we rely on the page allocators per cpu caches for
87 * fast frees and allocs.
88 *
89 * Overloading of page flags that are otherwise used for LRU management.
90 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070091 * PageActive The slab is frozen and exempt from list processing.
92 * This means that the slab is dedicated to a purpose
93 * such as satisfying allocations for a specific
94 * processor. Objects may be freed in the slab while
95 * it is frozen but slab_free will then skip the usual
96 * list operations. It is up to the processor holding
97 * the slab to integrate the slab into the slab lists
98 * when the slab is no longer needed.
99 *
100 * One use of this flag is to mark slabs that are
101 * used for allocations. Then such a slab becomes a cpu
102 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700103 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700104 * free objects in addition to the regular freelist
105 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700106 *
107 * PageError Slab requires special handling due to debug
108 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700109 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700110 */
111
Christoph Lameteraf537b02010-07-09 14:07:14 -0500112#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
113 SLAB_TRACE | SLAB_DEBUG_FREE)
114
115static inline int kmem_cache_debug(struct kmem_cache *s)
116{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700117#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500118 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700119#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500120 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700121#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500122}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700123
Christoph Lameter81819f02007-05-06 14:49:36 -0700124/*
125 * Issues still to be resolved:
126 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700127 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700129 * - Variable sizing of the per node arrays
130 */
131
132/* Enable to test recovery from slab corruption on boot */
133#undef SLUB_RESILIENCY_TEST
134
Christoph Lameterb789ef52011-06-01 12:25:49 -0500135/* Enable to log cmpxchg failures */
136#undef SLUB_DEBUG_CMPXCHG
137
Christoph Lameter81819f02007-05-06 14:49:36 -0700138/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700139 * Mininum number of partial slabs. These will be left on the partial
140 * lists even if they are empty. kmem_cache_shrink may reclaim them.
141 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800142#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700143
Christoph Lameter2086d262007-05-06 14:49:46 -0700144/*
145 * Maximum number of desirable partial slabs.
146 * The existence of more partial slabs makes kmem_cache_shrink
147 * sort the partial list by the number of objects in the.
148 */
149#define MAX_PARTIAL 10
150
Christoph Lameter81819f02007-05-06 14:49:36 -0700151#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
152 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700153
Christoph Lameter81819f02007-05-06 14:49:36 -0700154/*
David Rientjes3de47212009-07-27 18:30:35 -0700155 * Debugging flags that require metadata to be stored in the slab. These get
156 * disabled when slub_debug=O is used and a cache's min order increases with
157 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700158 */
David Rientjes3de47212009-07-27 18:30:35 -0700159#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700160
161/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700162 * Set of flags that will prevent slab merging
163 */
164#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300165 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
166 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700167
168#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200169 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700170
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400171#define OO_SHIFT 16
172#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500173#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400174
Christoph Lameter81819f02007-05-06 14:49:36 -0700175/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500176#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500177#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700178
179static int kmem_size = sizeof(struct kmem_cache);
180
181#ifdef CONFIG_SMP
182static struct notifier_block slab_notifier;
183#endif
184
185static enum {
186 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500187 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700188 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700189 SYSFS /* Sysfs up */
190} slab_state = DOWN;
191
192/* A list of all slab caches on the system */
193static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700194static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700195
Christoph Lameter02cbc872007-05-09 02:32:43 -0700196/*
197 * Tracking user of a slab.
198 */
Ben Greeard6543e32011-07-07 11:36:36 -0700199#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700200struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300201 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700202#ifdef CONFIG_STACKTRACE
203 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
204#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700205 int cpu; /* Was running on cpu */
206 int pid; /* Pid context */
207 unsigned long when; /* When did the operation occur */
208};
209
210enum track_item { TRACK_ALLOC, TRACK_FREE };
211
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500212#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700213static int sysfs_slab_add(struct kmem_cache *);
214static int sysfs_slab_alias(struct kmem_cache *, const char *);
215static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800216
Christoph Lameter81819f02007-05-06 14:49:36 -0700217#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700218static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
219static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
220 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800221static inline void sysfs_slab_remove(struct kmem_cache *s)
222{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300223 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800224 kfree(s);
225}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800226
Christoph Lameter81819f02007-05-06 14:49:36 -0700227#endif
228
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500229static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800230{
231#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e62009-12-18 16:26:23 -0600232 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800233#endif
234}
235
Christoph Lameter81819f02007-05-06 14:49:36 -0700236/********************************************************************
237 * Core slab cache functions
238 *******************************************************************/
239
240int slab_is_available(void)
241{
242 return slab_state >= UP;
243}
244
245static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
246{
Christoph Lameter81819f02007-05-06 14:49:36 -0700247 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700248}
249
Christoph Lameter6446faa2008-02-15 23:45:26 -0800250/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700251static inline int check_valid_pointer(struct kmem_cache *s,
252 struct page *page, const void *object)
253{
254 void *base;
255
Christoph Lametera973e9d2008-03-01 13:40:44 -0800256 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700257 return 1;
258
Christoph Lametera973e9d2008-03-01 13:40:44 -0800259 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300260 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700261 (object - base) % s->size) {
262 return 0;
263 }
264
265 return 1;
266}
267
Christoph Lameter7656c722007-05-09 02:32:40 -0700268static inline void *get_freepointer(struct kmem_cache *s, void *object)
269{
270 return *(void **)(object + s->offset);
271}
272
Eric Dumazet0ad95002011-12-16 16:25:34 +0100273static void prefetch_freepointer(const struct kmem_cache *s, void *object)
274{
275 prefetch(object + s->offset);
276}
277
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500278static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
279{
280 void *p;
281
282#ifdef CONFIG_DEBUG_PAGEALLOC
283 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
284#else
285 p = get_freepointer(s, object);
286#endif
287 return p;
288}
289
Christoph Lameter7656c722007-05-09 02:32:40 -0700290static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
291{
292 *(void **)(object + s->offset) = fp;
293}
294
295/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300296#define for_each_object(__p, __s, __addr, __objects) \
297 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700298 __p += (__s)->size)
299
Christoph Lameter7656c722007-05-09 02:32:40 -0700300/* Determine object index from a given position */
301static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
302{
303 return (p - addr) / s->size;
304}
305
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100306static inline size_t slab_ksize(const struct kmem_cache *s)
307{
308#ifdef CONFIG_SLUB_DEBUG
309 /*
310 * Debugging requires use of the padding between object
311 * and whatever may come after it.
312 */
313 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
314 return s->objsize;
315
316#endif
317 /*
318 * If we have the need to store the freelist pointer
319 * back there or track user information then we can
320 * only use the space before that information.
321 */
322 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
323 return s->inuse;
324 /*
325 * Else we can use all the padding etc for the allocation
326 */
327 return s->size;
328}
329
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800330static inline int order_objects(int order, unsigned long size, int reserved)
331{
332 return ((PAGE_SIZE << order) - reserved) / size;
333}
334
Christoph Lameter834f3d12008-04-14 19:11:31 +0300335static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800336 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300337{
338 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800339 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300340 };
341
342 return x;
343}
344
345static inline int oo_order(struct kmem_cache_order_objects x)
346{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400347 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300348}
349
350static inline int oo_objects(struct kmem_cache_order_objects x)
351{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400352 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300353}
354
Christoph Lameter881db7f2011-06-01 12:25:53 -0500355/*
356 * Per slab locking using the pagelock
357 */
358static __always_inline void slab_lock(struct page *page)
359{
360 bit_spin_lock(PG_locked, &page->flags);
361}
362
363static __always_inline void slab_unlock(struct page *page)
364{
365 __bit_spin_unlock(PG_locked, &page->flags);
366}
367
Christoph Lameter1d071712011-07-14 12:49:12 -0500368/* Interrupts must be disabled (for the fallback code to work right) */
369static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500370 void *freelist_old, unsigned long counters_old,
371 void *freelist_new, unsigned long counters_new,
372 const char *n)
373{
Christoph Lameter1d071712011-07-14 12:49:12 -0500374 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800375#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
376 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500377 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000378 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500379 freelist_old, counters_old,
380 freelist_new, counters_new))
381 return 1;
382 } else
383#endif
384 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500385 slab_lock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500386 if (page->freelist == freelist_old && page->counters == counters_old) {
387 page->freelist = freelist_new;
388 page->counters = counters_new;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500389 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500390 return 1;
391 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500392 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500393 }
394
395 cpu_relax();
396 stat(s, CMPXCHG_DOUBLE_FAIL);
397
398#ifdef SLUB_DEBUG_CMPXCHG
399 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
400#endif
401
402 return 0;
403}
404
Christoph Lameter1d071712011-07-14 12:49:12 -0500405static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
406 void *freelist_old, unsigned long counters_old,
407 void *freelist_new, unsigned long counters_new,
408 const char *n)
409{
Heiko Carstens25654092012-01-12 17:17:33 -0800410#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
411 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500412 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000413 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameter1d071712011-07-14 12:49:12 -0500414 freelist_old, counters_old,
415 freelist_new, counters_new))
416 return 1;
417 } else
418#endif
419 {
420 unsigned long flags;
421
422 local_irq_save(flags);
423 slab_lock(page);
424 if (page->freelist == freelist_old && page->counters == counters_old) {
425 page->freelist = freelist_new;
426 page->counters = counters_new;
427 slab_unlock(page);
428 local_irq_restore(flags);
429 return 1;
430 }
431 slab_unlock(page);
432 local_irq_restore(flags);
433 }
434
435 cpu_relax();
436 stat(s, CMPXCHG_DOUBLE_FAIL);
437
438#ifdef SLUB_DEBUG_CMPXCHG
439 printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name);
440#endif
441
442 return 0;
443}
444
Christoph Lameter41ecc552007-05-09 02:32:44 -0700445#ifdef CONFIG_SLUB_DEBUG
446/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500447 * Determine a map of object in use on a page.
448 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500449 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500450 * not vanish from under us.
451 */
452static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
453{
454 void *p;
455 void *addr = page_address(page);
456
457 for (p = page->freelist; p; p = get_freepointer(s, p))
458 set_bit(slab_index(p, s, addr), map);
459}
460
Christoph Lameter41ecc552007-05-09 02:32:44 -0700461/*
462 * Debug settings:
463 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700464#ifdef CONFIG_SLUB_DEBUG_ON
465static int slub_debug = DEBUG_DEFAULT_FLAGS;
466#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700467static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700468#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700469
470static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700471static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700472
Christoph Lameter7656c722007-05-09 02:32:40 -0700473/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700474 * Object debugging
475 */
476static void print_section(char *text, u8 *addr, unsigned int length)
477{
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200478 print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
479 length, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700480}
481
Christoph Lameter81819f02007-05-06 14:49:36 -0700482static struct track *get_track(struct kmem_cache *s, void *object,
483 enum track_item alloc)
484{
485 struct track *p;
486
487 if (s->offset)
488 p = object + s->offset + sizeof(void *);
489 else
490 p = object + s->inuse;
491
492 return p + alloc;
493}
494
495static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300496 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700497{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900498 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700499
Christoph Lameter81819f02007-05-06 14:49:36 -0700500 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700501#ifdef CONFIG_STACKTRACE
502 struct stack_trace trace;
503 int i;
504
505 trace.nr_entries = 0;
506 trace.max_entries = TRACK_ADDRS_COUNT;
507 trace.entries = p->addrs;
508 trace.skip = 3;
509 save_stack_trace(&trace);
510
511 /* See rant in lockdep.c */
512 if (trace.nr_entries != 0 &&
513 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
514 trace.nr_entries--;
515
516 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
517 p->addrs[i] = 0;
518#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700519 p->addr = addr;
520 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400521 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700522 p->when = jiffies;
523 } else
524 memset(p, 0, sizeof(struct track));
525}
526
Christoph Lameter81819f02007-05-06 14:49:36 -0700527static void init_tracking(struct kmem_cache *s, void *object)
528{
Christoph Lameter24922682007-07-17 04:03:18 -0700529 if (!(s->flags & SLAB_STORE_USER))
530 return;
531
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300532 set_track(s, object, TRACK_FREE, 0UL);
533 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700534}
535
536static void print_track(const char *s, struct track *t)
537{
538 if (!t->addr)
539 return;
540
Linus Torvalds7daf7052008-07-14 12:12:53 -0700541 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300542 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700543#ifdef CONFIG_STACKTRACE
544 {
545 int i;
546 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
547 if (t->addrs[i])
548 printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
549 else
550 break;
551 }
552#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700553}
554
Christoph Lameter24922682007-07-17 04:03:18 -0700555static void print_tracking(struct kmem_cache *s, void *object)
556{
557 if (!(s->flags & SLAB_STORE_USER))
558 return;
559
560 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
561 print_track("Freed", get_track(s, object, TRACK_FREE));
562}
563
564static void print_page_info(struct page *page)
565{
Christoph Lameter39b26462008-04-14 19:11:30 +0300566 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
567 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700568
569}
570
571static void slab_bug(struct kmem_cache *s, char *fmt, ...)
572{
573 va_list args;
574 char buf[100];
575
576 va_start(args, fmt);
577 vsnprintf(buf, sizeof(buf), fmt, args);
578 va_end(args);
579 printk(KERN_ERR "========================================"
580 "=====================================\n");
Dave Jones265d47e2011-11-15 15:04:00 -0800581 printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700582 printk(KERN_ERR "----------------------------------------"
583 "-------------------------------------\n\n");
584}
585
586static void slab_fix(struct kmem_cache *s, char *fmt, ...)
587{
588 va_list args;
589 char buf[100];
590
591 va_start(args, fmt);
592 vsnprintf(buf, sizeof(buf), fmt, args);
593 va_end(args);
594 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
595}
596
597static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700598{
599 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800600 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700601
602 print_tracking(s, p);
603
604 print_page_info(page);
605
606 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
607 p, p - addr, get_freepointer(s, p));
608
609 if (p > addr + 16)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200610 print_section("Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700611
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200612 print_section("Object ", p, min_t(unsigned long, s->objsize,
613 PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700614 if (s->flags & SLAB_RED_ZONE)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200615 print_section("Redzone ", p + s->objsize,
Christoph Lameter81819f02007-05-06 14:49:36 -0700616 s->inuse - s->objsize);
617
Christoph Lameter81819f02007-05-06 14:49:36 -0700618 if (s->offset)
619 off = s->offset + sizeof(void *);
620 else
621 off = s->inuse;
622
Christoph Lameter24922682007-07-17 04:03:18 -0700623 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700624 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700625
626 if (off != s->size)
627 /* Beginning of the filler is the free pointer */
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200628 print_section("Padding ", p + off, s->size - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700629
630 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700631}
632
633static void object_err(struct kmem_cache *s, struct page *page,
634 u8 *object, char *reason)
635{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700636 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700637 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700638}
639
Christoph Lameter24922682007-07-17 04:03:18 -0700640static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700641{
642 va_list args;
643 char buf[100];
644
Christoph Lameter24922682007-07-17 04:03:18 -0700645 va_start(args, fmt);
646 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700647 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700648 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700649 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700650 dump_stack();
651}
652
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500653static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700654{
655 u8 *p = object;
656
657 if (s->flags & __OBJECT_POISON) {
658 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800659 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700660 }
661
662 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500663 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700664}
665
Christoph Lameter24922682007-07-17 04:03:18 -0700666static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
667 void *from, void *to)
668{
669 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
670 memset(from, data, to - from);
671}
672
673static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
674 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800675 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700676{
677 u8 *fault;
678 u8 *end;
679
Akinobu Mita798248202011-10-31 17:08:07 -0700680 fault = memchr_inv(start, value, bytes);
Christoph Lameter24922682007-07-17 04:03:18 -0700681 if (!fault)
682 return 1;
683
684 end = start + bytes;
685 while (end > fault && end[-1] == value)
686 end--;
687
688 slab_bug(s, "%s overwritten", what);
689 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
690 fault, end - 1, fault[0], value);
691 print_trailer(s, page, object);
692
693 restore_bytes(s, what, value, fault, end);
694 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700695}
696
Christoph Lameter81819f02007-05-06 14:49:36 -0700697/*
698 * Object layout:
699 *
700 * object address
701 * Bytes of the object to be managed.
702 * If the freepointer may overlay the object then the free
703 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700704 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700705 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
706 * 0xa5 (POISON_END)
707 *
708 * object + s->objsize
709 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700710 * Padding is extended by another word if Redzoning is enabled and
711 * objsize == inuse.
712 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700713 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
714 * 0xcc (RED_ACTIVE) for objects in use.
715 *
716 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700717 * Meta data starts here.
718 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700719 * A. Free pointer (if we cannot overwrite object on free)
720 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700721 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800722 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700723 * before the word boundary.
724 *
725 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700726 *
727 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700728 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700729 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700730 * If slabcaches are merged then the objsize and inuse boundaries are mostly
731 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700732 * may be used with merged slabcaches.
733 */
734
Christoph Lameter81819f02007-05-06 14:49:36 -0700735static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
736{
737 unsigned long off = s->inuse; /* The end of info */
738
739 if (s->offset)
740 /* Freepointer is placed after the object. */
741 off += sizeof(void *);
742
743 if (s->flags & SLAB_STORE_USER)
744 /* We also have user information there */
745 off += 2 * sizeof(struct track);
746
747 if (s->size == off)
748 return 1;
749
Christoph Lameter24922682007-07-17 04:03:18 -0700750 return check_bytes_and_report(s, page, p, "Object padding",
751 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700752}
753
Christoph Lameter39b26462008-04-14 19:11:30 +0300754/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700755static int slab_pad_check(struct kmem_cache *s, struct page *page)
756{
Christoph Lameter24922682007-07-17 04:03:18 -0700757 u8 *start;
758 u8 *fault;
759 u8 *end;
760 int length;
761 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700762
763 if (!(s->flags & SLAB_POISON))
764 return 1;
765
Christoph Lametera973e9d2008-03-01 13:40:44 -0800766 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800767 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300768 end = start + length;
769 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700770 if (!remainder)
771 return 1;
772
Akinobu Mita798248202011-10-31 17:08:07 -0700773 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700774 if (!fault)
775 return 1;
776 while (end > fault && end[-1] == POISON_INUSE)
777 end--;
778
779 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200780 print_section("Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700781
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200782 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700783 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700784}
785
786static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500787 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700788{
789 u8 *p = object;
790 u8 *endobject = object + s->objsize;
791
792 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700793 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500794 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700795 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700796 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800797 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
798 check_bytes_and_report(s, page, p, "Alignment padding",
799 endobject, POISON_INUSE, s->inuse - s->objsize);
800 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700801 }
802
803 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500804 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700805 (!check_bytes_and_report(s, page, p, "Poison", p,
806 POISON_FREE, s->objsize - 1) ||
807 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800808 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700809 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700810 /*
811 * check_pad_bytes cleans up on its own.
812 */
813 check_pad_bytes(s, page, p);
814 }
815
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500816 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700817 /*
818 * Object and freepointer overlap. Cannot check
819 * freepointer while object is allocated.
820 */
821 return 1;
822
823 /* Check free pointer validity */
824 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
825 object_err(s, page, p, "Freepointer corrupt");
826 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100827 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700828 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700829 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700830 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800831 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700832 return 0;
833 }
834 return 1;
835}
836
837static int check_slab(struct kmem_cache *s, struct page *page)
838{
Christoph Lameter39b26462008-04-14 19:11:30 +0300839 int maxobj;
840
Christoph Lameter81819f02007-05-06 14:49:36 -0700841 VM_BUG_ON(!irqs_disabled());
842
843 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700844 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700845 return 0;
846 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300847
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800848 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300849 if (page->objects > maxobj) {
850 slab_err(s, page, "objects %u > max %u",
851 s->name, page->objects, maxobj);
852 return 0;
853 }
854 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700855 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300856 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700857 return 0;
858 }
859 /* Slab_pad_check fixes things up after itself */
860 slab_pad_check(s, page);
861 return 1;
862}
863
864/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700865 * Determine if a certain object on a page is on the freelist. Must hold the
866 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700867 */
868static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
869{
870 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500871 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700872 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300873 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700874
Christoph Lameter881db7f2011-06-01 12:25:53 -0500875 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300876 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700877 if (fp == search)
878 return 1;
879 if (!check_valid_pointer(s, page, fp)) {
880 if (object) {
881 object_err(s, page, object,
882 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800883 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700884 break;
885 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700886 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800887 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300888 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700889 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700890 return 0;
891 }
892 break;
893 }
894 object = fp;
895 fp = get_freepointer(s, object);
896 nr++;
897 }
898
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800899 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400900 if (max_objects > MAX_OBJS_PER_PAGE)
901 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300902
903 if (page->objects != max_objects) {
904 slab_err(s, page, "Wrong number of objects. Found %d but "
905 "should be %d", page->objects, max_objects);
906 page->objects = max_objects;
907 slab_fix(s, "Number of objects adjusted.");
908 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300909 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700910 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300911 "counted were %d", page->inuse, page->objects - nr);
912 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700913 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700914 }
915 return search == NULL;
916}
917
Christoph Lameter0121c6192008-04-29 16:11:12 -0700918static void trace(struct kmem_cache *s, struct page *page, void *object,
919 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700920{
921 if (s->flags & SLAB_TRACE) {
922 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
923 s->name,
924 alloc ? "alloc" : "free",
925 object, page->inuse,
926 page->freelist);
927
928 if (!alloc)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200929 print_section("Object ", (void *)object, s->objsize);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700930
931 dump_stack();
932 }
933}
934
Christoph Lameter643b1132007-05-06 14:49:42 -0700935/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500936 * Hooks for other subsystems that check memory allocations. In a typical
937 * production configuration these hooks all should produce no code at all.
938 */
939static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
940{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500941 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500942 lockdep_trace_alloc(flags);
943 might_sleep_if(flags & __GFP_WAIT);
944
945 return should_failslab(s->objsize, flags, s->flags);
946}
947
948static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
949{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500950 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100951 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500952 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
953}
954
955static inline void slab_free_hook(struct kmem_cache *s, void *x)
956{
957 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500958
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600959 /*
960 * Trouble is that we may no longer disable interupts in the fast path
961 * So in order to make the debug calls that expect irqs to be
962 * disabled we need to disable interrupts temporarily.
963 */
964#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
965 {
966 unsigned long flags;
967
968 local_irq_save(flags);
969 kmemcheck_slab_free(s, x, s->objsize);
970 debug_check_no_locks_freed(x, s->objsize);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600971 local_irq_restore(flags);
972 }
973#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200974 if (!(s->flags & SLAB_DEBUG_OBJECTS))
975 debug_check_no_obj_freed(x, s->objsize);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500976}
977
978/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700979 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500980 *
981 * list_lock must be held.
Christoph Lameter643b1132007-05-06 14:49:42 -0700982 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500983static void add_full(struct kmem_cache *s,
984 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700985{
Christoph Lameter643b1132007-05-06 14:49:42 -0700986 if (!(s->flags & SLAB_STORE_USER))
987 return;
988
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500989 list_add(&page->lru, &n->full);
990}
Christoph Lameter643b1132007-05-06 14:49:42 -0700991
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500992/*
993 * list_lock must be held.
994 */
995static void remove_full(struct kmem_cache *s, struct page *page)
996{
997 if (!(s->flags & SLAB_STORE_USER))
998 return;
999
Christoph Lameter643b1132007-05-06 14:49:42 -07001000 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001001}
1002
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001003/* Tracking of the number of slabs for debugging purposes */
1004static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1005{
1006 struct kmem_cache_node *n = get_node(s, node);
1007
1008 return atomic_long_read(&n->nr_slabs);
1009}
1010
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001011static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1012{
1013 return atomic_long_read(&n->nr_slabs);
1014}
1015
Christoph Lameter205ab992008-04-14 19:11:40 +03001016static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001017{
1018 struct kmem_cache_node *n = get_node(s, node);
1019
1020 /*
1021 * May be called early in order to allocate a slab for the
1022 * kmem_cache_node structure. Solve the chicken-egg
1023 * dilemma by deferring the increment of the count during
1024 * bootstrap (see early_kmem_cache_node_alloc).
1025 */
Christoph Lameter7340cc82010-09-28 08:10:26 -05001026 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001027 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001028 atomic_long_add(objects, &n->total_objects);
1029 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001030}
Christoph Lameter205ab992008-04-14 19:11:40 +03001031static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001032{
1033 struct kmem_cache_node *n = get_node(s, node);
1034
1035 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001036 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001037}
1038
1039/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001040static void setup_object_debug(struct kmem_cache *s, struct page *page,
1041 void *object)
1042{
1043 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1044 return;
1045
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001046 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001047 init_tracking(s, object);
1048}
1049
Christoph Lameter15370662010-08-20 12:37:12 -05001050static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001051 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001052{
1053 if (!check_slab(s, page))
1054 goto bad;
1055
Christoph Lameter81819f02007-05-06 14:49:36 -07001056 if (!check_valid_pointer(s, page, object)) {
1057 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001058 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001059 }
1060
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001061 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001062 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001063
Christoph Lameter3ec09742007-05-16 22:11:00 -07001064 /* Success perform special debug activities for allocs */
1065 if (s->flags & SLAB_STORE_USER)
1066 set_track(s, object, TRACK_ALLOC, addr);
1067 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001068 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001069 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001070
Christoph Lameter81819f02007-05-06 14:49:36 -07001071bad:
1072 if (PageSlab(page)) {
1073 /*
1074 * If this is a slab page then lets do the best we can
1075 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001076 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001077 */
Christoph Lameter24922682007-07-17 04:03:18 -07001078 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001079 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001080 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001081 }
1082 return 0;
1083}
1084
Christoph Lameter15370662010-08-20 12:37:12 -05001085static noinline int free_debug_processing(struct kmem_cache *s,
1086 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001087{
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001088 unsigned long flags;
1089 int rc = 0;
1090
1091 local_irq_save(flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001092 slab_lock(page);
1093
Christoph Lameter81819f02007-05-06 14:49:36 -07001094 if (!check_slab(s, page))
1095 goto fail;
1096
1097 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001098 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001099 goto fail;
1100 }
1101
1102 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001103 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001104 goto fail;
1105 }
1106
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001107 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001108 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001109
1110 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001111 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001112 slab_err(s, page, "Attempt to free object(0x%p) "
1113 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001114 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001115 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -07001116 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -07001117 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001118 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001119 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001120 object_err(s, page, object,
1121 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001122 goto fail;
1123 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001124
Christoph Lameter3ec09742007-05-16 22:11:00 -07001125 if (s->flags & SLAB_STORE_USER)
1126 set_track(s, object, TRACK_FREE, addr);
1127 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001128 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001129 rc = 1;
1130out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001131 slab_unlock(page);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001132 local_irq_restore(flags);
1133 return rc;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001134
Christoph Lameter81819f02007-05-06 14:49:36 -07001135fail:
Christoph Lameter24922682007-07-17 04:03:18 -07001136 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001137 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001138}
1139
Christoph Lameter41ecc552007-05-09 02:32:44 -07001140static int __init setup_slub_debug(char *str)
1141{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001142 slub_debug = DEBUG_DEFAULT_FLAGS;
1143 if (*str++ != '=' || !*str)
1144 /*
1145 * No options specified. Switch on full debugging.
1146 */
1147 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001148
1149 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001150 /*
1151 * No options but restriction on slabs. This means full
1152 * debugging for slabs matching a pattern.
1153 */
1154 goto check_slabs;
1155
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001156 if (tolower(*str) == 'o') {
1157 /*
1158 * Avoid enabling debugging on caches if its minimum order
1159 * would increase as a result.
1160 */
1161 disable_higher_order_debug = 1;
1162 goto out;
1163 }
1164
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001165 slub_debug = 0;
1166 if (*str == '-')
1167 /*
1168 * Switch off all debugging measures.
1169 */
1170 goto out;
1171
1172 /*
1173 * Determine which debug features should be switched on
1174 */
Pekka Enberg06428782008-01-07 23:20:27 -08001175 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001176 switch (tolower(*str)) {
1177 case 'f':
1178 slub_debug |= SLAB_DEBUG_FREE;
1179 break;
1180 case 'z':
1181 slub_debug |= SLAB_RED_ZONE;
1182 break;
1183 case 'p':
1184 slub_debug |= SLAB_POISON;
1185 break;
1186 case 'u':
1187 slub_debug |= SLAB_STORE_USER;
1188 break;
1189 case 't':
1190 slub_debug |= SLAB_TRACE;
1191 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001192 case 'a':
1193 slub_debug |= SLAB_FAILSLAB;
1194 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001195 default:
1196 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001197 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001198 }
1199 }
1200
1201check_slabs:
1202 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001203 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001204out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001205 return 1;
1206}
1207
1208__setup("slub_debug", setup_slub_debug);
1209
Christoph Lameterba0268a2007-09-11 15:24:11 -07001210static unsigned long kmem_cache_flags(unsigned long objsize,
1211 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001212 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001213{
1214 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001215 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001216 */
Christoph Lametere1533622008-02-15 23:45:24 -08001217 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001218 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1219 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001220
1221 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001222}
1223#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001224static inline void setup_object_debug(struct kmem_cache *s,
1225 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001226
Christoph Lameter3ec09742007-05-16 22:11:00 -07001227static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001228 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001229
Christoph Lameter3ec09742007-05-16 22:11:00 -07001230static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001231 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001232
Christoph Lameter41ecc552007-05-09 02:32:44 -07001233static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1234 { return 1; }
1235static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001236 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001237static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1238 struct page *page) {}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001239static inline void remove_full(struct kmem_cache *s, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001240static inline unsigned long kmem_cache_flags(unsigned long objsize,
1241 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001242 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001243{
1244 return flags;
1245}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001246#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001247
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001248#define disable_higher_order_debug 0
1249
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001250static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1251 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001252static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1253 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001254static inline void inc_slabs_node(struct kmem_cache *s, int node,
1255 int objects) {}
1256static inline void dec_slabs_node(struct kmem_cache *s, int node,
1257 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001258
1259static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1260 { return 0; }
1261
1262static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1263 void *object) {}
1264
1265static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1266
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001267#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001268
Christoph Lameter81819f02007-05-06 14:49:36 -07001269/*
1270 * Slab allocation and freeing
1271 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001272static inline struct page *alloc_slab_page(gfp_t flags, int node,
1273 struct kmem_cache_order_objects oo)
1274{
1275 int order = oo_order(oo);
1276
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001277 flags |= __GFP_NOTRACK;
1278
Christoph Lameter2154a332010-07-09 14:07:10 -05001279 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001280 return alloc_pages(flags, order);
1281 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001282 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001283}
1284
Christoph Lameter81819f02007-05-06 14:49:36 -07001285static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1286{
Pekka Enberg06428782008-01-07 23:20:27 -08001287 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001288 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001289 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001290
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001291 flags &= gfp_allowed_mask;
1292
1293 if (flags & __GFP_WAIT)
1294 local_irq_enable();
1295
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001296 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001297
Pekka Enbergba522702009-06-24 21:59:51 +03001298 /*
1299 * Let the initial higher-order allocation fail under memory pressure
1300 * so we fall-back to the minimum order allocation.
1301 */
1302 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1303
1304 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001305 if (unlikely(!page)) {
1306 oo = s->min;
1307 /*
1308 * Allocation may have failed due to fragmentation.
1309 * Try a lower order alloc if possible
1310 */
1311 page = alloc_slab_page(flags, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001312
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001313 if (page)
1314 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001315 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001316
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001317 if (flags & __GFP_WAIT)
1318 local_irq_disable();
1319
1320 if (!page)
1321 return NULL;
1322
Vegard Nossum5a896d92008-04-04 00:54:48 +02001323 if (kmemcheck_enabled
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001324 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001325 int pages = 1 << oo_order(oo);
1326
1327 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1328
1329 /*
1330 * Objects from caches that have a constructor don't get
1331 * cleared when they're allocated, so we need to do it here.
1332 */
1333 if (s->ctor)
1334 kmemcheck_mark_uninitialized_pages(page, pages);
1335 else
1336 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001337 }
1338
Christoph Lameter834f3d12008-04-14 19:11:31 +03001339 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001340 mod_zone_page_state(page_zone(page),
1341 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1342 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001343 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001344
1345 return page;
1346}
1347
1348static void setup_object(struct kmem_cache *s, struct page *page,
1349 void *object)
1350{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001351 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001352 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001353 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001354}
1355
1356static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1357{
1358 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001359 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001360 void *last;
1361 void *p;
1362
Christoph Lameter6cb06222007-10-16 01:25:41 -07001363 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001364
Christoph Lameter6cb06222007-10-16 01:25:41 -07001365 page = allocate_slab(s,
1366 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001367 if (!page)
1368 goto out;
1369
Christoph Lameter205ab992008-04-14 19:11:40 +03001370 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001371 page->slab = s;
1372 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001373
1374 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001375
1376 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001377 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001378
1379 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001380 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001381 setup_object(s, page, last);
1382 set_freepointer(s, last, p);
1383 last = p;
1384 }
1385 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001386 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001387
1388 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001389 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001390 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001391out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001392 return page;
1393}
1394
1395static void __free_slab(struct kmem_cache *s, struct page *page)
1396{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001397 int order = compound_order(page);
1398 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001399
Christoph Lameteraf537b02010-07-09 14:07:14 -05001400 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001401 void *p;
1402
1403 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001404 for_each_object(p, s, page_address(page),
1405 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001406 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001407 }
1408
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001409 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001410
Christoph Lameter81819f02007-05-06 14:49:36 -07001411 mod_zone_page_state(page_zone(page),
1412 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1413 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001414 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001415
Christoph Lameter49bd5222008-04-14 18:52:18 +03001416 __ClearPageSlab(page);
1417 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001418 if (current->reclaim_state)
1419 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001420 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001421}
1422
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001423#define need_reserve_slab_rcu \
1424 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1425
Christoph Lameter81819f02007-05-06 14:49:36 -07001426static void rcu_free_slab(struct rcu_head *h)
1427{
1428 struct page *page;
1429
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001430 if (need_reserve_slab_rcu)
1431 page = virt_to_head_page(h);
1432 else
1433 page = container_of((struct list_head *)h, struct page, lru);
1434
Christoph Lameter81819f02007-05-06 14:49:36 -07001435 __free_slab(page->slab, page);
1436}
1437
1438static void free_slab(struct kmem_cache *s, struct page *page)
1439{
1440 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001441 struct rcu_head *head;
1442
1443 if (need_reserve_slab_rcu) {
1444 int order = compound_order(page);
1445 int offset = (PAGE_SIZE << order) - s->reserved;
1446
1447 VM_BUG_ON(s->reserved != sizeof(*head));
1448 head = page_address(page) + offset;
1449 } else {
1450 /*
1451 * RCU free overloads the RCU head over the LRU
1452 */
1453 head = (void *)&page->lru;
1454 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001455
1456 call_rcu(head, rcu_free_slab);
1457 } else
1458 __free_slab(s, page);
1459}
1460
1461static void discard_slab(struct kmem_cache *s, struct page *page)
1462{
Christoph Lameter205ab992008-04-14 19:11:40 +03001463 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001464 free_slab(s, page);
1465}
1466
1467/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001468 * Management of partially allocated slabs.
1469 *
1470 * list_lock must be held.
Christoph Lameter81819f02007-05-06 14:49:36 -07001471 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001472static inline void add_partial(struct kmem_cache_node *n,
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001473 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001474{
Christoph Lametere95eed52007-05-06 14:49:44 -07001475 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001476 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001477 list_add_tail(&page->lru, &n->partial);
1478 else
1479 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001480}
1481
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001482/*
1483 * list_lock must be held.
1484 */
1485static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001486 struct page *page)
1487{
1488 list_del(&page->lru);
1489 n->nr_partial--;
1490}
1491
Christoph Lameter81819f02007-05-06 14:49:36 -07001492/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001493 * Lock slab, remove from the partial list and put the object into the
1494 * per cpu freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001495 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001496 * Returns a list of objects or NULL if it fails.
1497 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001498 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001499 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001500static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001501 struct kmem_cache_node *n, struct page *page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001502 int mode)
Christoph Lameter81819f02007-05-06 14:49:36 -07001503{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001504 void *freelist;
1505 unsigned long counters;
1506 struct page new;
1507
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001508 /*
1509 * Zap the freelist and set the frozen bit.
1510 * The old freelist is the list of objects for the
1511 * per cpu allocation list.
1512 */
1513 do {
1514 freelist = page->freelist;
1515 counters = page->counters;
1516 new.counters = counters;
Christoph Lameter49e22582011-08-09 16:12:27 -05001517 if (mode)
1518 new.inuse = page->objects;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001519
1520 VM_BUG_ON(new.frozen);
1521 new.frozen = 1;
1522
Christoph Lameter1d071712011-07-14 12:49:12 -05001523 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001524 freelist, counters,
1525 NULL, new.counters,
1526 "lock and freeze"));
1527
1528 remove_partial(n, page);
Christoph Lameter49e22582011-08-09 16:12:27 -05001529 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001530}
1531
Christoph Lameter49e22582011-08-09 16:12:27 -05001532static int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1533
Christoph Lameter81819f02007-05-06 14:49:36 -07001534/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001535 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001536 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001537static void *get_partial_node(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001538 struct kmem_cache_node *n, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001539{
Christoph Lameter49e22582011-08-09 16:12:27 -05001540 struct page *page, *page2;
1541 void *object = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001542
1543 /*
1544 * Racy check. If we mistakenly see no partial slabs then we
1545 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001546 * partial slab and there is none available then get_partials()
1547 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001548 */
1549 if (!n || !n->nr_partial)
1550 return NULL;
1551
1552 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001553 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Alex,Shi12d79632011-09-07 10:26:36 +08001554 void *t = acquire_slab(s, n, page, object == NULL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001555 int available;
1556
1557 if (!t)
1558 break;
1559
Alex,Shi12d79632011-09-07 10:26:36 +08001560 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001561 c->page = page;
1562 c->node = page_to_nid(page);
1563 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001564 object = t;
1565 available = page->objects - page->inuse;
1566 } else {
1567 page->freelist = t;
1568 available = put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001569 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001570 }
1571 if (kmem_cache_debug(s) || available > s->cpu_partial / 2)
1572 break;
1573
Christoph Lameter497b66f2011-08-09 16:12:26 -05001574 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001575 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001576 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001577}
1578
1579/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001580 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001581 */
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001582static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags,
1583 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001584{
1585#ifdef CONFIG_NUMA
1586 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001587 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001588 struct zone *zone;
1589 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001590 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001591 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001592
1593 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001594 * The defrag ratio allows a configuration of the tradeoffs between
1595 * inter node defragmentation and node local allocations. A lower
1596 * defrag_ratio increases the tendency to do local allocations
1597 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001598 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001599 * If the defrag_ratio is set to 0 then kmalloc() always
1600 * returns node local objects. If the ratio is higher then kmalloc()
1601 * may return off node objects because partial slabs are obtained
1602 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001603 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001604 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001605 * defrag_ratio = 1000) then every (well almost) allocation will
1606 * first attempt to defrag slab caches on other nodes. This means
1607 * scanning over all nodes to look for partial slabs which may be
1608 * expensive if we do it every time we are trying to find a slab
1609 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001610 */
Christoph Lameter98246012008-01-07 23:20:26 -08001611 if (!s->remote_node_defrag_ratio ||
1612 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001613 return NULL;
1614
Mel Gormancc9a6c82012-03-21 16:34:11 -07001615 do {
1616 cpuset_mems_cookie = get_mems_allowed();
1617 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1618 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1619 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001620
Mel Gormancc9a6c82012-03-21 16:34:11 -07001621 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001622
Mel Gormancc9a6c82012-03-21 16:34:11 -07001623 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1624 n->nr_partial > s->min_partial) {
1625 object = get_partial_node(s, n, c);
1626 if (object) {
1627 /*
1628 * Return the object even if
1629 * put_mems_allowed indicated that
1630 * the cpuset mems_allowed was
1631 * updated in parallel. It's a
1632 * harmless race between the alloc
1633 * and the cpuset update.
1634 */
1635 put_mems_allowed(cpuset_mems_cookie);
1636 return object;
1637 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001638 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001639 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07001640 } while (!put_mems_allowed(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001641#endif
1642 return NULL;
1643}
1644
1645/*
1646 * Get a partial page, lock it and return it.
1647 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001648static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001649 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001650{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001651 void *object;
Christoph Lameter2154a332010-07-09 14:07:10 -05001652 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001653
Christoph Lameter497b66f2011-08-09 16:12:26 -05001654 object = get_partial_node(s, get_node(s, searchnode), c);
1655 if (object || node != NUMA_NO_NODE)
1656 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001657
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001658 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001659}
1660
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001661#ifdef CONFIG_PREEMPT
1662/*
1663 * Calculate the next globally unique transaction for disambiguiation
1664 * during cmpxchg. The transactions start with the cpu number and are then
1665 * incremented by CONFIG_NR_CPUS.
1666 */
1667#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1668#else
1669/*
1670 * No preemption supported therefore also no need to check for
1671 * different cpus.
1672 */
1673#define TID_STEP 1
1674#endif
1675
1676static inline unsigned long next_tid(unsigned long tid)
1677{
1678 return tid + TID_STEP;
1679}
1680
1681static inline unsigned int tid_to_cpu(unsigned long tid)
1682{
1683 return tid % TID_STEP;
1684}
1685
1686static inline unsigned long tid_to_event(unsigned long tid)
1687{
1688 return tid / TID_STEP;
1689}
1690
1691static inline unsigned int init_tid(int cpu)
1692{
1693 return cpu;
1694}
1695
1696static inline void note_cmpxchg_failure(const char *n,
1697 const struct kmem_cache *s, unsigned long tid)
1698{
1699#ifdef SLUB_DEBUG_CMPXCHG
1700 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1701
1702 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1703
1704#ifdef CONFIG_PREEMPT
1705 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1706 printk("due to cpu change %d -> %d\n",
1707 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1708 else
1709#endif
1710 if (tid_to_event(tid) != tid_to_event(actual_tid))
1711 printk("due to cpu running other code. Event %ld->%ld\n",
1712 tid_to_event(tid), tid_to_event(actual_tid));
1713 else
1714 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1715 actual_tid, tid, next_tid(tid));
1716#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001717 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001718}
1719
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001720void init_kmem_cache_cpus(struct kmem_cache *s)
1721{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001722 int cpu;
1723
1724 for_each_possible_cpu(cpu)
1725 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001726}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001727
1728/*
1729 * Remove the cpu slab
1730 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001731static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001732{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001733 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001734 struct page *page = c->page;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001735 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1736 int lock = 0;
1737 enum slab_modes l = M_NONE, m = M_NONE;
1738 void *freelist;
1739 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001740 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001741 struct page new;
1742 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001743
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001744 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001745 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001746 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001747 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001748
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001749 c->tid = next_tid(c->tid);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001750 c->page = NULL;
1751 freelist = c->freelist;
1752 c->freelist = NULL;
1753
1754 /*
1755 * Stage one: Free all available per cpu objects back
1756 * to the page freelist while it is still frozen. Leave the
1757 * last one.
1758 *
1759 * There is no need to take the list->lock because the page
1760 * is still frozen.
1761 */
1762 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1763 void *prior;
1764 unsigned long counters;
1765
1766 do {
1767 prior = page->freelist;
1768 counters = page->counters;
1769 set_freepointer(s, freelist, prior);
1770 new.counters = counters;
1771 new.inuse--;
1772 VM_BUG_ON(!new.frozen);
1773
Christoph Lameter1d071712011-07-14 12:49:12 -05001774 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001775 prior, counters,
1776 freelist, new.counters,
1777 "drain percpu freelist"));
1778
1779 freelist = nextfree;
1780 }
1781
1782 /*
1783 * Stage two: Ensure that the page is unfrozen while the
1784 * list presence reflects the actual number of objects
1785 * during unfreeze.
1786 *
1787 * We setup the list membership and then perform a cmpxchg
1788 * with the count. If there is a mismatch then the page
1789 * is not unfrozen but the page is on the wrong list.
1790 *
1791 * Then we restart the process which may have to remove
1792 * the page from the list that we just put it on again
1793 * because the number of objects in the slab may have
1794 * changed.
1795 */
1796redo:
1797
1798 old.freelist = page->freelist;
1799 old.counters = page->counters;
1800 VM_BUG_ON(!old.frozen);
1801
1802 /* Determine target state of the slab */
1803 new.counters = old.counters;
1804 if (freelist) {
1805 new.inuse--;
1806 set_freepointer(s, freelist, old.freelist);
1807 new.freelist = freelist;
1808 } else
1809 new.freelist = old.freelist;
1810
1811 new.frozen = 0;
1812
Christoph Lameter81107182011-08-09 13:01:32 -05001813 if (!new.inuse && n->nr_partial > s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001814 m = M_FREE;
1815 else if (new.freelist) {
1816 m = M_PARTIAL;
1817 if (!lock) {
1818 lock = 1;
1819 /*
1820 * Taking the spinlock removes the possiblity
1821 * that acquire_slab() will see a slab page that
1822 * is frozen
1823 */
1824 spin_lock(&n->list_lock);
1825 }
1826 } else {
1827 m = M_FULL;
1828 if (kmem_cache_debug(s) && !lock) {
1829 lock = 1;
1830 /*
1831 * This also ensures that the scanning of full
1832 * slabs from diagnostic functions will not see
1833 * any frozen slabs.
1834 */
1835 spin_lock(&n->list_lock);
1836 }
1837 }
1838
1839 if (l != m) {
1840
1841 if (l == M_PARTIAL)
1842
1843 remove_partial(n, page);
1844
1845 else if (l == M_FULL)
1846
1847 remove_full(s, page);
1848
1849 if (m == M_PARTIAL) {
1850
1851 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001852 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001853
1854 } else if (m == M_FULL) {
1855
1856 stat(s, DEACTIVATE_FULL);
1857 add_full(s, n, page);
1858
1859 }
1860 }
1861
1862 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001863 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001864 old.freelist, old.counters,
1865 new.freelist, new.counters,
1866 "unfreezing slab"))
1867 goto redo;
1868
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001869 if (lock)
1870 spin_unlock(&n->list_lock);
1871
1872 if (m == M_FREE) {
1873 stat(s, DEACTIVATE_EMPTY);
1874 discard_slab(s, page);
1875 stat(s, FREE_SLAB);
1876 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001877}
1878
Christoph Lameter49e22582011-08-09 16:12:27 -05001879/* Unfreeze all the cpu partial slabs */
1880static void unfreeze_partials(struct kmem_cache *s)
1881{
1882 struct kmem_cache_node *n = NULL;
1883 struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab);
Shaohua Li9ada1932011-11-14 13:34:13 +08001884 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001885
1886 while ((page = c->partial)) {
1887 enum slab_modes { M_PARTIAL, M_FREE };
1888 enum slab_modes l, m;
1889 struct page new;
1890 struct page old;
1891
1892 c->partial = page->next;
1893 l = M_FREE;
1894
1895 do {
1896
1897 old.freelist = page->freelist;
1898 old.counters = page->counters;
1899 VM_BUG_ON(!old.frozen);
1900
1901 new.counters = old.counters;
1902 new.freelist = old.freelist;
1903
1904 new.frozen = 0;
1905
Alex Shidcc3be62011-09-06 14:46:01 +08001906 if (!new.inuse && (!n || n->nr_partial > s->min_partial))
Christoph Lameter49e22582011-08-09 16:12:27 -05001907 m = M_FREE;
1908 else {
1909 struct kmem_cache_node *n2 = get_node(s,
1910 page_to_nid(page));
1911
1912 m = M_PARTIAL;
1913 if (n != n2) {
1914 if (n)
1915 spin_unlock(&n->list_lock);
1916
1917 n = n2;
1918 spin_lock(&n->list_lock);
1919 }
1920 }
1921
1922 if (l != m) {
Shaohua Li4c493a52011-11-11 14:54:14 +08001923 if (l == M_PARTIAL) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001924 remove_partial(n, page);
Shaohua Li4c493a52011-11-11 14:54:14 +08001925 stat(s, FREE_REMOVE_PARTIAL);
1926 } else {
Shaohua Lif64ae0422011-11-11 08:33:48 +08001927 add_partial(n, page,
1928 DEACTIVATE_TO_TAIL);
Shaohua Li4c493a52011-11-11 14:54:14 +08001929 stat(s, FREE_ADD_PARTIAL);
1930 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001931
1932 l = m;
1933 }
1934
1935 } while (!cmpxchg_double_slab(s, page,
1936 old.freelist, old.counters,
1937 new.freelist, new.counters,
1938 "unfreezing slab"));
1939
1940 if (m == M_FREE) {
Shaohua Li9ada1932011-11-14 13:34:13 +08001941 page->next = discard_page;
1942 discard_page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001943 }
1944 }
1945
1946 if (n)
1947 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08001948
1949 while (discard_page) {
1950 page = discard_page;
1951 discard_page = discard_page->next;
1952
1953 stat(s, DEACTIVATE_EMPTY);
1954 discard_slab(s, page);
1955 stat(s, FREE_SLAB);
1956 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001957}
1958
1959/*
1960 * Put a page that was just frozen (in __slab_free) into a partial page
1961 * slot if available. This is done without interrupts disabled and without
1962 * preemption disabled. The cmpxchg is racy and may put the partial page
1963 * onto a random cpus partial slot.
1964 *
1965 * If we did not find a slot then simply move all the partials to the
1966 * per node partial list.
1967 */
1968int put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
1969{
1970 struct page *oldpage;
1971 int pages;
1972 int pobjects;
1973
1974 do {
1975 pages = 0;
1976 pobjects = 0;
1977 oldpage = this_cpu_read(s->cpu_slab->partial);
1978
1979 if (oldpage) {
1980 pobjects = oldpage->pobjects;
1981 pages = oldpage->pages;
1982 if (drain && pobjects > s->cpu_partial) {
1983 unsigned long flags;
1984 /*
1985 * partial array is full. Move the existing
1986 * set to the per node partial list.
1987 */
1988 local_irq_save(flags);
1989 unfreeze_partials(s);
1990 local_irq_restore(flags);
1991 pobjects = 0;
1992 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08001993 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05001994 }
1995 }
1996
1997 pages++;
1998 pobjects += page->objects - page->inuse;
1999
2000 page->pages = pages;
2001 page->pobjects = pobjects;
2002 page->next = oldpage;
2003
Christoph Lameter933393f2011-12-22 11:58:51 -06002004 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) != oldpage);
Christoph Lameter49e22582011-08-09 16:12:27 -05002005 return pobjects;
2006}
2007
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002008static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002009{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002010 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002011 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07002012}
2013
2014/*
2015 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002016 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002017 * Called from IPI handler with interrupts disabled.
2018 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002019static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002020{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002021 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002022
Christoph Lameter49e22582011-08-09 16:12:27 -05002023 if (likely(c)) {
2024 if (c->page)
2025 flush_slab(s, c);
2026
2027 unfreeze_partials(s);
2028 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002029}
2030
2031static void flush_cpu_slab(void *d)
2032{
2033 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002034
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002035 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002036}
2037
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002038static bool has_cpu_slab(int cpu, void *info)
2039{
2040 struct kmem_cache *s = info;
2041 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2042
2043 return !!(c->page);
2044}
2045
Christoph Lameter81819f02007-05-06 14:49:36 -07002046static void flush_all(struct kmem_cache *s)
2047{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002048 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002049}
2050
2051/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002052 * Check if the objects in a per cpu structure fit numa
2053 * locality expectations.
2054 */
2055static inline int node_match(struct kmem_cache_cpu *c, int node)
2056{
2057#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05002058 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002059 return 0;
2060#endif
2061 return 1;
2062}
2063
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002064static int count_free(struct page *page)
2065{
2066 return page->objects - page->inuse;
2067}
2068
2069static unsigned long count_partial(struct kmem_cache_node *n,
2070 int (*get_count)(struct page *))
2071{
2072 unsigned long flags;
2073 unsigned long x = 0;
2074 struct page *page;
2075
2076 spin_lock_irqsave(&n->list_lock, flags);
2077 list_for_each_entry(page, &n->partial, lru)
2078 x += get_count(page);
2079 spin_unlock_irqrestore(&n->list_lock, flags);
2080 return x;
2081}
2082
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002083static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2084{
2085#ifdef CONFIG_SLUB_DEBUG
2086 return atomic_long_read(&n->total_objects);
2087#else
2088 return 0;
2089#endif
2090}
2091
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002092static noinline void
2093slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2094{
2095 int node;
2096
2097 printk(KERN_WARNING
2098 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2099 nid, gfpflags);
2100 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
2101 "default order: %d, min order: %d\n", s->name, s->objsize,
2102 s->size, oo_order(s->oo), oo_order(s->min));
2103
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002104 if (oo_order(s->min) > get_order(s->objsize))
2105 printk(KERN_WARNING " %s debugging increased min order, use "
2106 "slub_debug=O to disable.\n", s->name);
2107
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002108 for_each_online_node(node) {
2109 struct kmem_cache_node *n = get_node(s, node);
2110 unsigned long nr_slabs;
2111 unsigned long nr_objs;
2112 unsigned long nr_free;
2113
2114 if (!n)
2115 continue;
2116
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002117 nr_free = count_partial(n, count_free);
2118 nr_slabs = node_nr_slabs(n);
2119 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002120
2121 printk(KERN_WARNING
2122 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2123 node, nr_slabs, nr_objs, nr_free);
2124 }
2125}
2126
Christoph Lameter497b66f2011-08-09 16:12:26 -05002127static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2128 int node, struct kmem_cache_cpu **pc)
2129{
2130 void *object;
2131 struct kmem_cache_cpu *c;
2132 struct page *page = new_slab(s, flags, node);
2133
2134 if (page) {
2135 c = __this_cpu_ptr(s->cpu_slab);
2136 if (c->page)
2137 flush_slab(s, c);
2138
2139 /*
2140 * No other reference to the page yet so we can
2141 * muck around with it freely without cmpxchg
2142 */
2143 object = page->freelist;
2144 page->freelist = NULL;
2145
2146 stat(s, ALLOC_SLAB);
2147 c->node = page_to_nid(page);
2148 c->page = page;
2149 *pc = c;
2150 } else
2151 object = NULL;
2152
2153 return object;
2154}
2155
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002156/*
Christoph Lameter213eeb92011-11-11 14:07:14 -06002157 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2158 * or deactivate the page.
2159 *
2160 * The page is still frozen if the return value is not NULL.
2161 *
2162 * If this function returns NULL then the page has been unfrozen.
2163 */
2164static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2165{
2166 struct page new;
2167 unsigned long counters;
2168 void *freelist;
2169
2170 do {
2171 freelist = page->freelist;
2172 counters = page->counters;
2173 new.counters = counters;
2174 VM_BUG_ON(!new.frozen);
2175
2176 new.inuse = page->objects;
2177 new.frozen = freelist != NULL;
2178
2179 } while (!cmpxchg_double_slab(s, page,
2180 freelist, counters,
2181 NULL, new.counters,
2182 "get_freelist"));
2183
2184 return freelist;
2185}
2186
2187/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002188 * Slow path. The lockless freelist is empty or we need to perform
2189 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002190 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002191 * Processing is still very fast if new objects have been freed to the
2192 * regular freelist. In that case we simply take over the regular freelist
2193 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002194 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002195 * If that is not working then we fall back to the partial lists. We take the
2196 * first element of the freelist as the object to allocate now and move the
2197 * rest of the freelist to the lockless freelist.
2198 *
2199 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002200 * we need to allocate a new slab. This is the slowest path since it involves
2201 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002202 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002203static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2204 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002205{
Christoph Lameter81819f02007-05-06 14:49:36 -07002206 void **object;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002207 unsigned long flags;
2208
2209 local_irq_save(flags);
2210#ifdef CONFIG_PREEMPT
2211 /*
2212 * We may have been preempted and rescheduled on a different
2213 * cpu before disabling interrupts. Need to reload cpu area
2214 * pointer.
2215 */
2216 c = this_cpu_ptr(s->cpu_slab);
2217#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002218
Christoph Lameter497b66f2011-08-09 16:12:26 -05002219 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002220 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002221redo:
Christoph Lameterfc59c052011-06-01 12:25:56 -05002222 if (unlikely(!node_match(c, node))) {
Christoph Lametere36a2652011-06-01 12:25:57 -05002223 stat(s, ALLOC_NODE_MISMATCH);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002224 deactivate_slab(s, c);
2225 goto new_slab;
2226 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002227
Eric Dumazet73736e02011-12-13 04:57:06 +01002228 /* must check again c->freelist in case of cpu migration or IRQ */
2229 object = c->freelist;
2230 if (object)
2231 goto load_freelist;
2232
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002233 stat(s, ALLOC_SLOWPATH);
2234
Christoph Lameter213eeb92011-11-11 14:07:14 -06002235 object = get_freelist(s, c->page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002236
Christoph Lameter49e22582011-08-09 16:12:27 -05002237 if (!object) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002238 c->page = NULL;
2239 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002240 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002241 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002242
Christoph Lameter81819f02007-05-06 14:49:36 -07002243 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002244
Christoph Lameter894b8782007-05-10 03:15:16 -07002245load_freelist:
Christoph Lameterff120592009-12-18 16:26:22 -06002246 c->freelist = get_freepointer(s, object);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002247 c->tid = next_tid(c->tid);
2248 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002249 return object;
2250
Christoph Lameter81819f02007-05-06 14:49:36 -07002251new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002252
Christoph Lameter49e22582011-08-09 16:12:27 -05002253 if (c->partial) {
2254 c->page = c->partial;
2255 c->partial = c->page->next;
2256 c->node = page_to_nid(c->page);
2257 stat(s, CPU_PARTIAL_ALLOC);
2258 c->freelist = NULL;
2259 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002260 }
2261
Christoph Lameter49e22582011-08-09 16:12:27 -05002262 /* Then do expensive stuff like retrieving pages from the partial lists */
Christoph Lameter497b66f2011-08-09 16:12:26 -05002263 object = get_partial(s, gfpflags, node, c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002264
Christoph Lameter497b66f2011-08-09 16:12:26 -05002265 if (unlikely(!object)) {
Christoph Lameter01ad8a72011-04-15 14:48:14 -05002266
Christoph Lameter497b66f2011-08-09 16:12:26 -05002267 object = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002268
Christoph Lameter497b66f2011-08-09 16:12:26 -05002269 if (unlikely(!object)) {
2270 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
2271 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter9e577e82011-07-22 09:35:14 -05002272
Christoph Lameter497b66f2011-08-09 16:12:26 -05002273 local_irq_restore(flags);
2274 return NULL;
2275 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002276 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002277
Christoph Lameter497b66f2011-08-09 16:12:26 -05002278 if (likely(!kmem_cache_debug(s)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002279 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002280
Christoph Lameter497b66f2011-08-09 16:12:26 -05002281 /* Only entered in the debug case */
2282 if (!alloc_debug_processing(s, c->page, object, addr))
2283 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002284
2285 c->freelist = get_freepointer(s, object);
Christoph Lameter442b06b2011-05-17 16:29:31 -05002286 deactivate_slab(s, c);
Pekka Enberg15b7c512010-10-02 11:32:32 +03002287 c->node = NUMA_NO_NODE;
Christoph Lametera71ae472011-05-25 09:47:43 -05002288 local_irq_restore(flags);
2289 return object;
Christoph Lameter894b8782007-05-10 03:15:16 -07002290}
2291
2292/*
2293 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2294 * have the fastpath folded into their functions. So no function call
2295 * overhead for requests that can be satisfied on the fastpath.
2296 *
2297 * The fastpath works by first checking if the lockless freelist can be used.
2298 * If not then __slab_alloc is called for slow processing.
2299 *
2300 * Otherwise we can simply pick the next object from the lockless free list.
2301 */
Pekka Enberg06428782008-01-07 23:20:27 -08002302static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002303 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002304{
Christoph Lameter894b8782007-05-10 03:15:16 -07002305 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002306 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002307 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002308
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002309 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09002310 return NULL;
2311
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002312redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002313
2314 /*
2315 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2316 * enabled. We may switch back and forth between cpus while
2317 * reading from one cpu area. That does not matter as long
2318 * as we end up on the original cpu again when doing the cmpxchg.
2319 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002320 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002321
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002322 /*
2323 * The transaction ids are globally unique per cpu and per operation on
2324 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2325 * occurs on the right processor and that there was no operation on the
2326 * linked list in between.
2327 */
2328 tid = c->tid;
2329 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002330
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002331 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002332 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07002333
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002334 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07002335
2336 else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002337 void *next_object = get_freepointer_safe(s, object);
2338
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002339 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002340 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002341 * operation and if we are on the right processor.
2342 *
2343 * The cmpxchg does the following atomically (without lock semantics!)
2344 * 1. Relocate first pointer to the current per cpu area.
2345 * 2. Verify that tid and freelist have not been changed
2346 * 3. If they were not changed replace tid and freelist
2347 *
2348 * Since this is without lock semantics the protection is only against
2349 * code executing on this cpu *not* from access by other cpus.
2350 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002351 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002352 s->cpu_slab->freelist, s->cpu_slab->tid,
2353 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002354 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002355
2356 note_cmpxchg_failure("slab_alloc", s, tid);
2357 goto redo;
2358 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002359 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002360 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002361 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002362
Pekka Enberg74e21342009-11-25 20:14:48 +02002363 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06002364 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002365
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002366 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002367
Christoph Lameter894b8782007-05-10 03:15:16 -07002368 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002369}
2370
2371void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2372{
Christoph Lameter2154a332010-07-09 14:07:10 -05002373 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002374
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002375 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002376
2377 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002378}
2379EXPORT_SYMBOL(kmem_cache_alloc);
2380
Li Zefan0f24f122009-12-11 15:45:30 +08002381#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002382void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002383{
Richard Kennedy4a923792010-10-21 10:29:19 +01002384 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
2385 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2386 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002387}
Richard Kennedy4a923792010-10-21 10:29:19 +01002388EXPORT_SYMBOL(kmem_cache_alloc_trace);
2389
2390void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2391{
2392 void *ret = kmalloc_order(size, flags, order);
2393 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2394 return ret;
2395}
2396EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002397#endif
2398
Christoph Lameter81819f02007-05-06 14:49:36 -07002399#ifdef CONFIG_NUMA
2400void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2401{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002402 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2403
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002404 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2405 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002406
2407 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002408}
2409EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002410
Li Zefan0f24f122009-12-11 15:45:30 +08002411#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002412void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002413 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002414 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002415{
Richard Kennedy4a923792010-10-21 10:29:19 +01002416 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2417
2418 trace_kmalloc_node(_RET_IP_, ret,
2419 size, s->size, gfpflags, node);
2420 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002421}
Richard Kennedy4a923792010-10-21 10:29:19 +01002422EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002423#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002424#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002425
Christoph Lameter81819f02007-05-06 14:49:36 -07002426/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002427 * Slow patch handling. This may still be called frequently since objects
2428 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002429 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002430 * So we still attempt to reduce cache line usage. Just take the slab
2431 * lock and free the item. If there is no additional partial page
2432 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002433 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002434static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002435 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002436{
2437 void *prior;
2438 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002439 int was_frozen;
2440 int inuse;
2441 struct page new;
2442 unsigned long counters;
2443 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002444 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002445
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002446 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002447
Christoph Lameter8dc16c62011-04-15 14:48:16 -05002448 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002449 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002450
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002451 do {
2452 prior = page->freelist;
2453 counters = page->counters;
2454 set_freepointer(s, object, prior);
2455 new.counters = counters;
2456 was_frozen = new.frozen;
2457 new.inuse--;
2458 if ((!new.inuse || !prior) && !was_frozen && !n) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002459
2460 if (!kmem_cache_debug(s) && !prior)
2461
2462 /*
2463 * Slab was on no list before and will be partially empty
2464 * We can defer the list move and instead freeze it.
2465 */
2466 new.frozen = 1;
2467
2468 else { /* Needs to be taken off a list */
2469
2470 n = get_node(s, page_to_nid(page));
2471 /*
2472 * Speculatively acquire the list_lock.
2473 * If the cmpxchg does not succeed then we may
2474 * drop the list_lock without any processing.
2475 *
2476 * Otherwise the list_lock will synchronize with
2477 * other processors updating the list of slabs.
2478 */
2479 spin_lock_irqsave(&n->list_lock, flags);
2480
2481 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002482 }
2483 inuse = new.inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002484
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002485 } while (!cmpxchg_double_slab(s, page,
2486 prior, counters,
2487 object, new.counters,
2488 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002489
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002490 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002491
2492 /*
2493 * If we just froze the page then put it onto the
2494 * per cpu partial list.
2495 */
Alex Shi8028dce2012-02-03 23:34:56 +08002496 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002497 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002498 stat(s, CPU_PARTIAL_FREE);
2499 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002500 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002501 * The list lock was not taken therefore no list
2502 * activity can be necessary.
2503 */
2504 if (was_frozen)
2505 stat(s, FREE_FROZEN);
Christoph Lameter80f08c12011-06-01 12:25:55 -05002506 return;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002507 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002508
2509 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002510 * was_frozen may have been set after we acquired the list_lock in
2511 * an earlier loop. So we need to check it here again.
Christoph Lameter81819f02007-05-06 14:49:36 -07002512 */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002513 if (was_frozen)
2514 stat(s, FREE_FROZEN);
2515 else {
2516 if (unlikely(!inuse && n->nr_partial > s->min_partial))
2517 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002518
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002519 /*
2520 * Objects left in the slab. If it was not on the partial list before
2521 * then add it.
2522 */
2523 if (unlikely(!prior)) {
2524 remove_full(s, page);
Shaohua Li136333d2011-08-24 08:57:52 +08002525 add_partial(n, page, DEACTIVATE_TO_TAIL);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002526 stat(s, FREE_ADD_PARTIAL);
2527 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002528 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002529 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002530 return;
2531
2532slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002533 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002534 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002535 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002536 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002537 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002538 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002539 } else
2540 /* Slab must be on the full list */
2541 remove_full(s, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002542
Christoph Lameter80f08c12011-06-01 12:25:55 -05002543 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002544 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002545 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002546}
2547
Christoph Lameter894b8782007-05-10 03:15:16 -07002548/*
2549 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2550 * can perform fastpath freeing without additional function calls.
2551 *
2552 * The fastpath is only possible if we are freeing to the current cpu slab
2553 * of this processor. This typically the case if we have just allocated
2554 * the item before.
2555 *
2556 * If fastpath is not possible then fall back to __slab_free where we deal
2557 * with all sorts of special processing.
2558 */
Pekka Enberg06428782008-01-07 23:20:27 -08002559static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002560 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002561{
2562 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002563 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002564 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002565
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002566 slab_free_hook(s, x);
2567
Christoph Lametera24c5a02011-03-15 12:45:21 -05002568redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002569 /*
2570 * Determine the currently cpus per cpu slab.
2571 * The cpu may change afterward. However that does not matter since
2572 * data is retrieved via this pointer. If we are on the same cpu
2573 * during the cmpxchg then the free will succedd.
2574 */
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002575 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002576
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002577 tid = c->tid;
2578 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002579
Christoph Lameter442b06b2011-05-17 16:29:31 -05002580 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002581 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002582
Christoph Lameter933393f2011-12-22 11:58:51 -06002583 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002584 s->cpu_slab->freelist, s->cpu_slab->tid,
2585 c->freelist, tid,
2586 object, next_tid(tid)))) {
2587
2588 note_cmpxchg_failure("slab_free", s, tid);
2589 goto redo;
2590 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002591 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002592 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002593 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002594
Christoph Lameter894b8782007-05-10 03:15:16 -07002595}
2596
Christoph Lameter81819f02007-05-06 14:49:36 -07002597void kmem_cache_free(struct kmem_cache *s, void *x)
2598{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07002599 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002600
Christoph Lameterb49af682007-05-06 14:49:41 -07002601 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002602
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002603 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002604
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002605 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002606}
2607EXPORT_SYMBOL(kmem_cache_free);
2608
Christoph Lameter81819f02007-05-06 14:49:36 -07002609/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002610 * Object placement in a slab is made very easy because we always start at
2611 * offset 0. If we tune the size of the object to the alignment then we can
2612 * get the required alignment by putting one properly sized object after
2613 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002614 *
2615 * Notice that the allocation order determines the sizes of the per cpu
2616 * caches. Each processor has always one slab available for allocations.
2617 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002618 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002619 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002620 */
2621
2622/*
2623 * Mininum / Maximum order of slab pages. This influences locking overhead
2624 * and slab fragmentation. A higher order reduces the number of partial slabs
2625 * and increases the number of allocations possible without having to
2626 * take the list_lock.
2627 */
2628static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002629static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002630static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002631
2632/*
2633 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002634 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002635 */
2636static int slub_nomerge;
2637
2638/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002639 * Calculate the order of allocation given an slab object size.
2640 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002641 * The order of allocation has significant impact on performance and other
2642 * system components. Generally order 0 allocations should be preferred since
2643 * order 0 does not cause fragmentation in the page allocator. Larger objects
2644 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002645 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002646 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002647 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002648 * In order to reach satisfactory performance we must ensure that a minimum
2649 * number of objects is in one slab. Otherwise we may generate too much
2650 * activity on the partial lists which requires taking the list_lock. This is
2651 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002652 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002653 * slub_max_order specifies the order where we begin to stop considering the
2654 * number of objects in a slab as critical. If we reach slub_max_order then
2655 * we try to keep the page order as low as possible. So we accept more waste
2656 * of space in favor of a small page order.
2657 *
2658 * Higher order allocations also allow the placement of more objects in a
2659 * slab and thereby reduce object handling overhead. If the user has
2660 * requested a higher mininum order then we start with that one instead of
2661 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002662 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002663static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002664 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002665{
2666 int order;
2667 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002668 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002669
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002670 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002671 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002672
Christoph Lameter6300ea72007-07-17 04:03:20 -07002673 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002674 fls(min_objects * size - 1) - PAGE_SHIFT);
2675 order <= max_order; order++) {
2676
Christoph Lameter81819f02007-05-06 14:49:36 -07002677 unsigned long slab_size = PAGE_SIZE << order;
2678
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002679 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002680 continue;
2681
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002682 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002683
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002684 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002685 break;
2686
2687 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002688
Christoph Lameter81819f02007-05-06 14:49:36 -07002689 return order;
2690}
2691
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002692static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002693{
2694 int order;
2695 int min_objects;
2696 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002697 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002698
2699 /*
2700 * Attempt to find best configuration for a slab. This
2701 * works by first attempting to generate a layout with
2702 * the best configuration and backing off gradually.
2703 *
2704 * First we reduce the acceptable waste in a slab. Then
2705 * we reduce the minimum objects required in a slab.
2706 */
2707 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002708 if (!min_objects)
2709 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002710 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002711 min_objects = min(min_objects, max_objects);
2712
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002713 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002714 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002715 while (fraction >= 4) {
2716 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002717 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002718 if (order <= slub_max_order)
2719 return order;
2720 fraction /= 2;
2721 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002722 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002723 }
2724
2725 /*
2726 * We were unable to place multiple objects in a slab. Now
2727 * lets see if we can place a single object there.
2728 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002729 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002730 if (order <= slub_max_order)
2731 return order;
2732
2733 /*
2734 * Doh this slab cannot be placed using slub_max_order.
2735 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002736 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002737 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002738 return order;
2739 return -ENOSYS;
2740}
2741
Christoph Lameter81819f02007-05-06 14:49:36 -07002742/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002743 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002744 */
2745static unsigned long calculate_alignment(unsigned long flags,
2746 unsigned long align, unsigned long size)
2747{
2748 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002749 * If the user wants hardware cache aligned objects then follow that
2750 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002751 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002752 * The hardware cache alignment cannot override the specified
2753 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002754 */
Nick Pigginb6210382008-03-05 14:05:56 -08002755 if (flags & SLAB_HWCACHE_ALIGN) {
2756 unsigned long ralign = cache_line_size();
2757 while (size <= ralign / 2)
2758 ralign /= 2;
2759 align = max(align, ralign);
2760 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002761
2762 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002763 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002764
2765 return ALIGN(align, sizeof(void *));
2766}
2767
Pekka Enberg5595cff2008-08-05 09:28:47 +03002768static void
2769init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002770{
2771 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002772 spin_lock_init(&n->list_lock);
2773 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002774#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002775 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002776 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002777 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002778#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002779}
2780
Christoph Lameter55136592010-08-20 12:37:13 -05002781static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002782{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002783 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2784 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002785
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002786 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002787 * Must align to double word boundary for the double cmpxchg
2788 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002789 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002790 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2791 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002792
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002793 if (!s->cpu_slab)
2794 return 0;
2795
2796 init_kmem_cache_cpus(s);
2797
2798 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002799}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002800
Christoph Lameter51df1142010-08-20 12:37:15 -05002801static struct kmem_cache *kmem_cache_node;
2802
Christoph Lameter81819f02007-05-06 14:49:36 -07002803/*
2804 * No kmalloc_node yet so do it by hand. We know that this is the first
2805 * slab on the node for this slabcache. There are no concurrent accesses
2806 * possible.
2807 *
2808 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002809 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2810 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002811 */
Christoph Lameter55136592010-08-20 12:37:13 -05002812static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002813{
2814 struct page *page;
2815 struct kmem_cache_node *n;
2816
Christoph Lameter51df1142010-08-20 12:37:15 -05002817 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002818
Christoph Lameter51df1142010-08-20 12:37:15 -05002819 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002820
2821 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002822 if (page_to_nid(page) != node) {
2823 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2824 "node %d\n", node);
2825 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2826 "in order to be able to continue\n");
2827 }
2828
Christoph Lameter81819f02007-05-06 14:49:36 -07002829 n = page->freelist;
2830 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002831 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05002832 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002833 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002834 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002835#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002836 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002837 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002838#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002839 init_kmem_cache_node(n, kmem_cache_node);
2840 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002841
Shaohua Li136333d2011-08-24 08:57:52 +08002842 add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002843}
2844
2845static void free_kmem_cache_nodes(struct kmem_cache *s)
2846{
2847 int node;
2848
Christoph Lameterf64dc582007-10-16 01:25:33 -07002849 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002850 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002851
Alexander Duyck73367bd2010-05-21 14:41:35 -07002852 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002853 kmem_cache_free(kmem_cache_node, n);
2854
Christoph Lameter81819f02007-05-06 14:49:36 -07002855 s->node[node] = NULL;
2856 }
2857}
2858
Christoph Lameter55136592010-08-20 12:37:13 -05002859static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002860{
2861 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002862
Christoph Lameterf64dc582007-10-16 01:25:33 -07002863 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002864 struct kmem_cache_node *n;
2865
Alexander Duyck73367bd2010-05-21 14:41:35 -07002866 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002867 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002868 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002869 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002870 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002871 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002872
2873 if (!n) {
2874 free_kmem_cache_nodes(s);
2875 return 0;
2876 }
2877
Christoph Lameter81819f02007-05-06 14:49:36 -07002878 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002879 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002880 }
2881 return 1;
2882}
Christoph Lameter81819f02007-05-06 14:49:36 -07002883
David Rientjesc0bdb232009-02-25 09:16:35 +02002884static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002885{
2886 if (min < MIN_PARTIAL)
2887 min = MIN_PARTIAL;
2888 else if (min > MAX_PARTIAL)
2889 min = MAX_PARTIAL;
2890 s->min_partial = min;
2891}
2892
Christoph Lameter81819f02007-05-06 14:49:36 -07002893/*
2894 * calculate_sizes() determines the order and the distribution of data within
2895 * a slab object.
2896 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002897static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002898{
2899 unsigned long flags = s->flags;
2900 unsigned long size = s->objsize;
2901 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002902 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002903
2904 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002905 * Round up object size to the next word boundary. We can only
2906 * place the free pointer at word boundaries and this determines
2907 * the possible location of the free pointer.
2908 */
2909 size = ALIGN(size, sizeof(void *));
2910
2911#ifdef CONFIG_SLUB_DEBUG
2912 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002913 * Determine if we can poison the object itself. If the user of
2914 * the slab may touch the object after free or before allocation
2915 * then we should never poison the object itself.
2916 */
2917 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002918 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002919 s->flags |= __OBJECT_POISON;
2920 else
2921 s->flags &= ~__OBJECT_POISON;
2922
Christoph Lameter81819f02007-05-06 14:49:36 -07002923
2924 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002925 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002926 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002927 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002928 */
2929 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2930 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002931#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002932
2933 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002934 * With that we have determined the number of bytes in actual use
2935 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002936 */
2937 s->inuse = size;
2938
2939 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002940 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002941 /*
2942 * Relocate free pointer after the object if it is not
2943 * permitted to overwrite the first word of the object on
2944 * kmem_cache_free.
2945 *
2946 * This is the case if we do RCU, have a constructor or
2947 * destructor or are poisoning the objects.
2948 */
2949 s->offset = size;
2950 size += sizeof(void *);
2951 }
2952
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002953#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002954 if (flags & SLAB_STORE_USER)
2955 /*
2956 * Need to store information about allocs and frees after
2957 * the object.
2958 */
2959 size += 2 * sizeof(struct track);
2960
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002961 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002962 /*
2963 * Add some empty padding so that we can catch
2964 * overwrites from earlier objects rather than let
2965 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002966 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002967 * of the object.
2968 */
2969 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002970#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002971
Christoph Lameter81819f02007-05-06 14:49:36 -07002972 /*
2973 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002974 * user specified and the dynamic determination of cache line size
2975 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002976 */
2977 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002978 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002979
2980 /*
2981 * SLUB stores one object immediately after another beginning from
2982 * offset 0. In order to align the objects we have to simply size
2983 * each object to conform to the alignment.
2984 */
2985 size = ALIGN(size, align);
2986 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002987 if (forced_order >= 0)
2988 order = forced_order;
2989 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002990 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07002991
Christoph Lameter834f3d12008-04-14 19:11:31 +03002992 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002993 return 0;
2994
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002995 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002996 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002997 s->allocflags |= __GFP_COMP;
2998
2999 if (s->flags & SLAB_CACHE_DMA)
3000 s->allocflags |= SLUB_DMA;
3001
3002 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3003 s->allocflags |= __GFP_RECLAIMABLE;
3004
Christoph Lameter81819f02007-05-06 14:49:36 -07003005 /*
3006 * Determine the number of objects per slab
3007 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003008 s->oo = oo_make(order, size, s->reserved);
3009 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003010 if (oo_objects(s->oo) > oo_objects(s->max))
3011 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003012
Christoph Lameter834f3d12008-04-14 19:11:31 +03003013 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003014
3015}
3016
Christoph Lameter55136592010-08-20 12:37:13 -05003017static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07003018 const char *name, size_t size,
3019 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003020 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003021{
3022 memset(s, 0, kmem_size);
3023 s->name = name;
3024 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07003025 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07003026 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07003027 s->flags = kmem_cache_flags(size, flags, name, ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003028 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003029
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003030 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3031 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003032
Christoph Lameter06b285d2008-04-14 19:11:41 +03003033 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003034 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003035 if (disable_higher_order_debug) {
3036 /*
3037 * Disable debugging flags that store metadata if the min slab
3038 * order increased.
3039 */
3040 if (get_order(s->size) > get_order(s->objsize)) {
3041 s->flags &= ~DEBUG_METADATA_FLAGS;
3042 s->offset = 0;
3043 if (!calculate_sizes(s, -1))
3044 goto error;
3045 }
3046 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003047
Heiko Carstens25654092012-01-12 17:17:33 -08003048#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3049 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003050 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
3051 /* Enable fast mode */
3052 s->flags |= __CMPXCHG_DOUBLE;
3053#endif
3054
David Rientjes3b89d7d2009-02-22 17:40:07 -08003055 /*
3056 * The larger the object size is, the more pages we want on the partial
3057 * list to avoid pounding the page allocator excessively.
3058 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003059 set_min_partial(s, ilog2(s->size) / 2);
3060
3061 /*
3062 * cpu_partial determined the maximum number of objects kept in the
3063 * per cpu partial lists of a processor.
3064 *
3065 * Per cpu partial lists mainly contain slabs that just have one
3066 * object freed. If they are used for allocation then they can be
3067 * filled up again with minimal effort. The slab will never hit the
3068 * per node partial lists and therefore no locking will be required.
3069 *
3070 * This setting also determines
3071 *
3072 * A) The number of objects from per cpu partial slabs dumped to the
3073 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003074 * B) The number of objects in cpu partial slabs to extract from the
Christoph Lameter49e22582011-08-09 16:12:27 -05003075 * per node list when we run out of per cpu objects. We only fetch 50%
3076 * to keep some capacity around for frees.
3077 */
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003078 if (kmem_cache_debug(s))
3079 s->cpu_partial = 0;
3080 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003081 s->cpu_partial = 2;
3082 else if (s->size >= 1024)
3083 s->cpu_partial = 6;
3084 else if (s->size >= 256)
3085 s->cpu_partial = 13;
3086 else
3087 s->cpu_partial = 30;
3088
Christoph Lameter81819f02007-05-06 14:49:36 -07003089 s->refcount = 1;
3090#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003091 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003092#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003093 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003094 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003095
Christoph Lameter55136592010-08-20 12:37:13 -05003096 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07003097 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06003098
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003099 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003100error:
3101 if (flags & SLAB_PANIC)
3102 panic("Cannot create slab %s size=%lu realsize=%u "
3103 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03003104 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07003105 s->offset, flags);
3106 return 0;
3107}
Christoph Lameter81819f02007-05-06 14:49:36 -07003108
3109/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003110 * Determine the size of a slab object
3111 */
3112unsigned int kmem_cache_size(struct kmem_cache *s)
3113{
3114 return s->objsize;
3115}
3116EXPORT_SYMBOL(kmem_cache_size);
3117
Christoph Lameter33b12c32008-04-25 12:22:43 -07003118static void list_slab_objects(struct kmem_cache *s, struct page *page,
3119 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003120{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003121#ifdef CONFIG_SLUB_DEBUG
3122 void *addr = page_address(page);
3123 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003124 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3125 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003126 if (!map)
3127 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07003128 slab_err(s, page, "%s", text);
3129 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003130
Christoph Lameter5f80b132011-04-15 14:48:13 -05003131 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003132 for_each_object(p, s, addr, page->objects) {
3133
3134 if (!test_bit(slab_index(p, s, addr), map)) {
3135 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
3136 p, p - addr);
3137 print_tracking(s, p);
3138 }
3139 }
3140 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003141 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003142#endif
3143}
3144
Christoph Lameter81819f02007-05-06 14:49:36 -07003145/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003146 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003147 * This is called from kmem_cache_close(). We must be the last thread
3148 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003149 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003150static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003151{
Christoph Lameter81819f02007-05-06 14:49:36 -07003152 struct page *page, *h;
3153
Christoph Lameter33b12c32008-04-25 12:22:43 -07003154 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003155 if (!page->inuse) {
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003156 remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003157 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003158 } else {
3159 list_slab_objects(s, page,
3160 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003161 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003162 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003163}
3164
3165/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003166 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003167 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003168static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003169{
3170 int node;
3171
3172 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003173 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003174 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07003175 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003176 struct kmem_cache_node *n = get_node(s, node);
3177
Christoph Lameter599870b2008-04-23 12:36:52 -07003178 free_partial(s, n);
3179 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003180 return 1;
3181 }
3182 free_kmem_cache_nodes(s);
3183 return 0;
3184}
3185
3186/*
3187 * Close a cache and release the kmem_cache structure
3188 * (must be used for caches created using kmem_cache_create)
3189 */
3190void kmem_cache_destroy(struct kmem_cache *s)
3191{
3192 down_write(&slub_lock);
3193 s->refcount--;
3194 if (!s->refcount) {
3195 list_del(&s->list);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003196 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03003197 if (kmem_cache_close(s)) {
3198 printk(KERN_ERR "SLUB %s: %s called for cache that "
3199 "still has objects.\n", s->name, __func__);
3200 dump_stack();
3201 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03003202 if (s->flags & SLAB_DESTROY_BY_RCU)
3203 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07003204 sysfs_slab_remove(s);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003205 } else
3206 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003207}
3208EXPORT_SYMBOL(kmem_cache_destroy);
3209
3210/********************************************************************
3211 * Kmalloc subsystem
3212 *******************************************************************/
3213
Christoph Lameter51df1142010-08-20 12:37:15 -05003214struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07003215EXPORT_SYMBOL(kmalloc_caches);
3216
Christoph Lameter51df1142010-08-20 12:37:15 -05003217static struct kmem_cache *kmem_cache;
3218
Christoph Lameter55136592010-08-20 12:37:13 -05003219#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003220static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05003221#endif
3222
Christoph Lameter81819f02007-05-06 14:49:36 -07003223static int __init setup_slub_min_order(char *str)
3224{
Pekka Enberg06428782008-01-07 23:20:27 -08003225 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003226
3227 return 1;
3228}
3229
3230__setup("slub_min_order=", setup_slub_min_order);
3231
3232static int __init setup_slub_max_order(char *str)
3233{
Pekka Enberg06428782008-01-07 23:20:27 -08003234 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003235 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003236
3237 return 1;
3238}
3239
3240__setup("slub_max_order=", setup_slub_max_order);
3241
3242static int __init setup_slub_min_objects(char *str)
3243{
Pekka Enberg06428782008-01-07 23:20:27 -08003244 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003245
3246 return 1;
3247}
3248
3249__setup("slub_min_objects=", setup_slub_min_objects);
3250
3251static int __init setup_slub_nomerge(char *str)
3252{
3253 slub_nomerge = 1;
3254 return 1;
3255}
3256
3257__setup("slub_nomerge", setup_slub_nomerge);
3258
Christoph Lameter51df1142010-08-20 12:37:15 -05003259static struct kmem_cache *__init create_kmalloc_cache(const char *name,
3260 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003261{
Christoph Lameter51df1142010-08-20 12:37:15 -05003262 struct kmem_cache *s;
3263
3264 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3265
Pekka Enberg83b519e2009-06-10 19:40:04 +03003266 /*
3267 * This function is called with IRQs disabled during early-boot on
3268 * single CPU so there's no need to take slub_lock here.
3269 */
Christoph Lameter55136592010-08-20 12:37:13 -05003270 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03003271 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07003272 goto panic;
3273
3274 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05003275 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003276
3277panic:
3278 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003279 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003280}
3281
Christoph Lameterf1b26332007-07-17 04:03:26 -07003282/*
3283 * Conversion table for small slabs sizes / 8 to the index in the
3284 * kmalloc array. This is necessary for slabs < 192 since we have non power
3285 * of two cache sizes there. The size of larger slabs can be determined using
3286 * fls.
3287 */
3288static s8 size_index[24] = {
3289 3, /* 8 */
3290 4, /* 16 */
3291 5, /* 24 */
3292 5, /* 32 */
3293 6, /* 40 */
3294 6, /* 48 */
3295 6, /* 56 */
3296 6, /* 64 */
3297 1, /* 72 */
3298 1, /* 80 */
3299 1, /* 88 */
3300 1, /* 96 */
3301 7, /* 104 */
3302 7, /* 112 */
3303 7, /* 120 */
3304 7, /* 128 */
3305 2, /* 136 */
3306 2, /* 144 */
3307 2, /* 152 */
3308 2, /* 160 */
3309 2, /* 168 */
3310 2, /* 176 */
3311 2, /* 184 */
3312 2 /* 192 */
3313};
3314
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003315static inline int size_index_elem(size_t bytes)
3316{
3317 return (bytes - 1) / 8;
3318}
3319
Christoph Lameter81819f02007-05-06 14:49:36 -07003320static struct kmem_cache *get_slab(size_t size, gfp_t flags)
3321{
Christoph Lameterf1b26332007-07-17 04:03:26 -07003322 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07003323
Christoph Lameterf1b26332007-07-17 04:03:26 -07003324 if (size <= 192) {
3325 if (!size)
3326 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07003327
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003328 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003329 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07003330 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003331
3332#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07003333 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05003334 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07003335
Christoph Lameter81819f02007-05-06 14:49:36 -07003336#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003337 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07003338}
3339
3340void *__kmalloc(size_t size, gfp_t flags)
3341{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003342 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003343 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003344
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003345 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003346 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003347
3348 s = get_slab(size, flags);
3349
3350 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003351 return s;
3352
Christoph Lameter2154a332010-07-09 14:07:10 -05003353 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003354
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003355 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003356
3357 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003358}
3359EXPORT_SYMBOL(__kmalloc);
3360
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003361#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003362static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3363{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003364 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003365 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003366
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003367 flags |= __GFP_COMP | __GFP_NOTRACK;
3368 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003369 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003370 ptr = page_address(page);
3371
3372 kmemleak_alloc(ptr, size, 1, flags);
3373 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003374}
3375
Christoph Lameter81819f02007-05-06 14:49:36 -07003376void *__kmalloc_node(size_t size, gfp_t flags, int node)
3377{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003378 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003379 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003380
Ingo Molnar057685c2009-02-20 12:15:30 +01003381 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003382 ret = kmalloc_large_node(size, flags, node);
3383
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003384 trace_kmalloc_node(_RET_IP_, ret,
3385 size, PAGE_SIZE << get_order(size),
3386 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003387
3388 return ret;
3389 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003390
3391 s = get_slab(size, flags);
3392
3393 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003394 return s;
3395
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003396 ret = slab_alloc(s, flags, node, _RET_IP_);
3397
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003398 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003399
3400 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003401}
3402EXPORT_SYMBOL(__kmalloc_node);
3403#endif
3404
3405size_t ksize(const void *object)
3406{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003407 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003408
Christoph Lameteref8b4522007-10-16 01:24:46 -07003409 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003410 return 0;
3411
Vegard Nossum294a80a2007-12-04 23:45:30 -08003412 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003413
Pekka Enberg76994412008-05-22 19:22:25 +03003414 if (unlikely(!PageSlab(page))) {
3415 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003416 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003417 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003418
Eric Dumazetb3d41882011-02-14 18:35:22 +01003419 return slab_ksize(page->slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003420}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003421EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003422
Ben Greeard18a90d2011-07-07 11:36:37 -07003423#ifdef CONFIG_SLUB_DEBUG
3424bool verify_mem_not_deleted(const void *x)
3425{
3426 struct page *page;
3427 void *object = (void *)x;
3428 unsigned long flags;
3429 bool rv;
3430
3431 if (unlikely(ZERO_OR_NULL_PTR(x)))
3432 return false;
3433
3434 local_irq_save(flags);
3435
3436 page = virt_to_head_page(x);
3437 if (unlikely(!PageSlab(page))) {
3438 /* maybe it was from stack? */
3439 rv = true;
3440 goto out_unlock;
3441 }
3442
3443 slab_lock(page);
3444 if (on_freelist(page->slab, page, object)) {
3445 object_err(page->slab, page, object, "Object is on free-list");
3446 rv = false;
3447 } else {
3448 rv = true;
3449 }
3450 slab_unlock(page);
3451
3452out_unlock:
3453 local_irq_restore(flags);
3454 return rv;
3455}
3456EXPORT_SYMBOL(verify_mem_not_deleted);
3457#endif
3458
Christoph Lameter81819f02007-05-06 14:49:36 -07003459void kfree(const void *x)
3460{
Christoph Lameter81819f02007-05-06 14:49:36 -07003461 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003462 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003463
Pekka Enberg2121db72009-03-25 11:05:57 +02003464 trace_kfree(_RET_IP_, x);
3465
Satyam Sharma2408c552007-10-16 01:24:44 -07003466 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003467 return;
3468
Christoph Lameterb49af682007-05-06 14:49:41 -07003469 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003470 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003471 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003472 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003473 put_page(page);
3474 return;
3475 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003476 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003477}
3478EXPORT_SYMBOL(kfree);
3479
Christoph Lameter2086d262007-05-06 14:49:46 -07003480/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003481 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3482 * the remaining slabs by the number of items in use. The slabs with the
3483 * most items in use come first. New allocations will then fill those up
3484 * and thus they can be removed from the partial lists.
3485 *
3486 * The slabs with the least items are placed last. This results in them
3487 * being allocated from last increasing the chance that the last objects
3488 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003489 */
3490int kmem_cache_shrink(struct kmem_cache *s)
3491{
3492 int node;
3493 int i;
3494 struct kmem_cache_node *n;
3495 struct page *page;
3496 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003497 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003498 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003499 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003500 unsigned long flags;
3501
3502 if (!slabs_by_inuse)
3503 return -ENOMEM;
3504
3505 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003506 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003507 n = get_node(s, node);
3508
3509 if (!n->nr_partial)
3510 continue;
3511
Christoph Lameter834f3d12008-04-14 19:11:31 +03003512 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003513 INIT_LIST_HEAD(slabs_by_inuse + i);
3514
3515 spin_lock_irqsave(&n->list_lock, flags);
3516
3517 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003518 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003519 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003520 * Note that concurrent frees may occur while we hold the
3521 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003522 */
3523 list_for_each_entry_safe(page, t, &n->partial, lru) {
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003524 list_move(&page->lru, slabs_by_inuse + page->inuse);
3525 if (!page->inuse)
3526 n->nr_partial--;
Christoph Lameter2086d262007-05-06 14:49:46 -07003527 }
3528
Christoph Lameter2086d262007-05-06 14:49:46 -07003529 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003530 * Rebuild the partial list with the slabs filled up most
3531 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003532 */
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003533 for (i = objects - 1; i > 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003534 list_splice(slabs_by_inuse + i, n->partial.prev);
3535
Christoph Lameter2086d262007-05-06 14:49:46 -07003536 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003537
3538 /* Release empty slabs */
3539 list_for_each_entry_safe(page, t, slabs_by_inuse, lru)
3540 discard_slab(s, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003541 }
3542
3543 kfree(slabs_by_inuse);
3544 return 0;
3545}
3546EXPORT_SYMBOL(kmem_cache_shrink);
3547
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03003548#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07003549static int slab_mem_going_offline_callback(void *arg)
3550{
3551 struct kmem_cache *s;
3552
3553 down_read(&slub_lock);
3554 list_for_each_entry(s, &slab_caches, list)
3555 kmem_cache_shrink(s);
3556 up_read(&slub_lock);
3557
3558 return 0;
3559}
3560
3561static void slab_mem_offline_callback(void *arg)
3562{
3563 struct kmem_cache_node *n;
3564 struct kmem_cache *s;
3565 struct memory_notify *marg = arg;
3566 int offline_node;
3567
3568 offline_node = marg->status_change_nid;
3569
3570 /*
3571 * If the node still has available memory. we need kmem_cache_node
3572 * for it yet.
3573 */
3574 if (offline_node < 0)
3575 return;
3576
3577 down_read(&slub_lock);
3578 list_for_each_entry(s, &slab_caches, list) {
3579 n = get_node(s, offline_node);
3580 if (n) {
3581 /*
3582 * if n->nr_slabs > 0, slabs still exist on the node
3583 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003584 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003585 * callback. So, we must fail.
3586 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003587 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003588
3589 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003590 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003591 }
3592 }
3593 up_read(&slub_lock);
3594}
3595
3596static int slab_mem_going_online_callback(void *arg)
3597{
3598 struct kmem_cache_node *n;
3599 struct kmem_cache *s;
3600 struct memory_notify *marg = arg;
3601 int nid = marg->status_change_nid;
3602 int ret = 0;
3603
3604 /*
3605 * If the node's memory is already available, then kmem_cache_node is
3606 * already created. Nothing to do.
3607 */
3608 if (nid < 0)
3609 return 0;
3610
3611 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003612 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003613 * allocate a kmem_cache_node structure in order to bring the node
3614 * online.
3615 */
3616 down_read(&slub_lock);
3617 list_for_each_entry(s, &slab_caches, list) {
3618 /*
3619 * XXX: kmem_cache_alloc_node will fallback to other nodes
3620 * since memory is not yet available from the node that
3621 * is brought up.
3622 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003623 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003624 if (!n) {
3625 ret = -ENOMEM;
3626 goto out;
3627 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003628 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003629 s->node[nid] = n;
3630 }
3631out:
3632 up_read(&slub_lock);
3633 return ret;
3634}
3635
3636static int slab_memory_callback(struct notifier_block *self,
3637 unsigned long action, void *arg)
3638{
3639 int ret = 0;
3640
3641 switch (action) {
3642 case MEM_GOING_ONLINE:
3643 ret = slab_mem_going_online_callback(arg);
3644 break;
3645 case MEM_GOING_OFFLINE:
3646 ret = slab_mem_going_offline_callback(arg);
3647 break;
3648 case MEM_OFFLINE:
3649 case MEM_CANCEL_ONLINE:
3650 slab_mem_offline_callback(arg);
3651 break;
3652 case MEM_ONLINE:
3653 case MEM_CANCEL_OFFLINE:
3654 break;
3655 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003656 if (ret)
3657 ret = notifier_from_errno(ret);
3658 else
3659 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003660 return ret;
3661}
3662
3663#endif /* CONFIG_MEMORY_HOTPLUG */
3664
Christoph Lameter81819f02007-05-06 14:49:36 -07003665/********************************************************************
3666 * Basic setup of slabs
3667 *******************************************************************/
3668
Christoph Lameter51df1142010-08-20 12:37:15 -05003669/*
3670 * Used for early kmem_cache structures that were allocated using
3671 * the page allocator
3672 */
3673
3674static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3675{
3676 int node;
3677
3678 list_add(&s->list, &slab_caches);
3679 s->refcount = -1;
3680
3681 for_each_node_state(node, N_NORMAL_MEMORY) {
3682 struct kmem_cache_node *n = get_node(s, node);
3683 struct page *p;
3684
3685 if (n) {
3686 list_for_each_entry(p, &n->partial, lru)
3687 p->slab = s;
3688
Li Zefan607bf322011-04-12 15:22:26 +08003689#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003690 list_for_each_entry(p, &n->full, lru)
3691 p->slab = s;
3692#endif
3693 }
3694 }
3695}
3696
Christoph Lameter81819f02007-05-06 14:49:36 -07003697void __init kmem_cache_init(void)
3698{
3699 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003700 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003701 struct kmem_cache *temp_kmem_cache;
3702 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003703 struct kmem_cache *temp_kmem_cache_node;
3704 unsigned long kmalloc_size;
3705
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08003706 if (debug_guardpage_minorder())
3707 slub_max_order = 0;
3708
Christoph Lameter51df1142010-08-20 12:37:15 -05003709 kmem_size = offsetof(struct kmem_cache, node) +
3710 nr_node_ids * sizeof(struct kmem_cache_node *);
3711
3712 /* Allocate two kmem_caches from the page allocator */
3713 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3714 order = get_order(2 * kmalloc_size);
3715 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3716
Christoph Lameter81819f02007-05-06 14:49:36 -07003717 /*
3718 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003719 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003720 * kmem_cache_open for slab_state == DOWN.
3721 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003722 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3723
3724 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3725 sizeof(struct kmem_cache_node),
3726 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003727
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003728 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003729
3730 /* Able to allocate the per node structures */
3731 slab_state = PARTIAL;
3732
Christoph Lameter51df1142010-08-20 12:37:15 -05003733 temp_kmem_cache = kmem_cache;
3734 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3735 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3736 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3737 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003738
Christoph Lameter51df1142010-08-20 12:37:15 -05003739 /*
3740 * Allocate kmem_cache_node properly from the kmem_cache slab.
3741 * kmem_cache_node is separately allocated so no need to
3742 * update any list pointers.
3743 */
3744 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003745
Christoph Lameter51df1142010-08-20 12:37:15 -05003746 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3747 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3748
3749 kmem_cache_bootstrap_fixup(kmem_cache_node);
3750
3751 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003752 kmem_cache_bootstrap_fixup(kmem_cache);
3753 caches++;
3754 /* Free temporary boot structure */
3755 free_pages((unsigned long)temp_kmem_cache, order);
3756
3757 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003758
3759 /*
3760 * Patch up the size_index table if we have strange large alignment
3761 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003762 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003763 *
3764 * Largest permitted alignment is 256 bytes due to the way we
3765 * handle the index determination for the smaller caches.
3766 *
3767 * Make sure that nothing crazy happens if someone starts tinkering
3768 * around with ARCH_KMALLOC_MINALIGN
3769 */
3770 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3771 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3772
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003773 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3774 int elem = size_index_elem(i);
3775 if (elem >= ARRAY_SIZE(size_index))
3776 break;
3777 size_index[elem] = KMALLOC_SHIFT_LOW;
3778 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003779
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003780 if (KMALLOC_MIN_SIZE == 64) {
3781 /*
3782 * The 96 byte size cache is not used if the alignment
3783 * is 64 byte.
3784 */
3785 for (i = 64 + 8; i <= 96; i += 8)
3786 size_index[size_index_elem(i)] = 7;
3787 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003788 /*
3789 * The 192 byte sized cache is not used if the alignment
3790 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3791 * instead.
3792 */
3793 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003794 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003795 }
3796
Christoph Lameter51df1142010-08-20 12:37:15 -05003797 /* Caches that are not of the two-to-the-power-of size */
3798 if (KMALLOC_MIN_SIZE <= 32) {
3799 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3800 caches++;
3801 }
3802
3803 if (KMALLOC_MIN_SIZE <= 64) {
3804 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3805 caches++;
3806 }
3807
3808 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3809 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3810 caches++;
3811 }
3812
Christoph Lameter81819f02007-05-06 14:49:36 -07003813 slab_state = UP;
3814
3815 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003816 if (KMALLOC_MIN_SIZE <= 32) {
3817 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3818 BUG_ON(!kmalloc_caches[1]->name);
3819 }
3820
3821 if (KMALLOC_MIN_SIZE <= 64) {
3822 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3823 BUG_ON(!kmalloc_caches[2]->name);
3824 }
3825
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003826 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3827 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3828
3829 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003830 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003831 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003832
3833#ifdef CONFIG_SMP
3834 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003835#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003836
Christoph Lameter55136592010-08-20 12:37:13 -05003837#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003838 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3839 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003840
Christoph Lameter51df1142010-08-20 12:37:15 -05003841 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003842 char *name = kasprintf(GFP_NOWAIT,
3843 "dma-kmalloc-%d", s->objsize);
3844
3845 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003846 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3847 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003848 }
3849 }
3850#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003851 printk(KERN_INFO
3852 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003853 " CPUs=%d, Nodes=%d\n",
3854 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003855 slub_min_order, slub_max_order, slub_min_objects,
3856 nr_cpu_ids, nr_node_ids);
3857}
3858
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003859void __init kmem_cache_init_late(void)
3860{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003861}
3862
Christoph Lameter81819f02007-05-06 14:49:36 -07003863/*
3864 * Find a mergeable slab cache
3865 */
3866static int slab_unmergeable(struct kmem_cache *s)
3867{
3868 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3869 return 1;
3870
Christoph Lameterc59def92007-05-16 22:10:50 -07003871 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003872 return 1;
3873
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003874 /*
3875 * We may have set a slab to be unmergeable during bootstrap.
3876 */
3877 if (s->refcount < 0)
3878 return 1;
3879
Christoph Lameter81819f02007-05-06 14:49:36 -07003880 return 0;
3881}
3882
3883static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003884 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003885 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003886{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003887 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003888
3889 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3890 return NULL;
3891
Christoph Lameterc59def92007-05-16 22:10:50 -07003892 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003893 return NULL;
3894
3895 size = ALIGN(size, sizeof(void *));
3896 align = calculate_alignment(flags, align, size);
3897 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003898 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003899
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003900 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003901 if (slab_unmergeable(s))
3902 continue;
3903
3904 if (size > s->size)
3905 continue;
3906
Christoph Lameterba0268a2007-09-11 15:24:11 -07003907 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003908 continue;
3909 /*
3910 * Check if alignment is compatible.
3911 * Courtesy of Adrian Drzewiecki
3912 */
Pekka Enberg06428782008-01-07 23:20:27 -08003913 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003914 continue;
3915
3916 if (s->size - size >= sizeof(void *))
3917 continue;
3918
3919 return s;
3920 }
3921 return NULL;
3922}
3923
3924struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003925 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003926{
3927 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003928 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003929
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003930 if (WARN_ON(!name))
3931 return NULL;
3932
Christoph Lameter81819f02007-05-06 14:49:36 -07003933 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003934 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003935 if (s) {
3936 s->refcount++;
3937 /*
3938 * Adjust the object sizes so that we clear
3939 * the complete object on kzalloc.
3940 */
3941 s->objsize = max(s->objsize, (int)size);
3942 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003943
David Rientjes7b8f3b62008-12-17 22:09:46 -08003944 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003945 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003946 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003947 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003948 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003949 return s;
3950 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003951
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003952 n = kstrdup(name, GFP_KERNEL);
3953 if (!n)
3954 goto err;
3955
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003956 s = kmalloc(kmem_size, GFP_KERNEL);
3957 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003958 if (kmem_cache_open(s, n,
Christoph Lameterc59def92007-05-16 22:10:50 -07003959 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003960 list_add(&s->list, &slab_caches);
Christoph Lameter66c4c352012-01-17 09:27:31 -06003961 up_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003962 if (sysfs_slab_add(s)) {
Christoph Lameter66c4c352012-01-17 09:27:31 -06003963 down_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003964 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003965 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003966 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003967 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003968 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003969 return s;
3970 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003971 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003972 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003973 }
Pavel Emelyanov68cee4f12010-10-28 13:50:37 +04003974err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003975 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003976
Christoph Lameter81819f02007-05-06 14:49:36 -07003977 if (flags & SLAB_PANIC)
3978 panic("Cannot create slabcache %s\n", name);
3979 else
3980 s = NULL;
3981 return s;
3982}
3983EXPORT_SYMBOL(kmem_cache_create);
3984
Christoph Lameter81819f02007-05-06 14:49:36 -07003985#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003986/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003987 * Use the cpu notifier to insure that the cpu slabs are flushed when
3988 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003989 */
3990static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3991 unsigned long action, void *hcpu)
3992{
3993 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003994 struct kmem_cache *s;
3995 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003996
3997 switch (action) {
3998 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003999 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07004000 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07004001 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004002 down_read(&slub_lock);
4003 list_for_each_entry(s, &slab_caches, list) {
4004 local_irq_save(flags);
4005 __flush_cpu_slab(s, cpu);
4006 local_irq_restore(flags);
4007 }
4008 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004009 break;
4010 default:
4011 break;
4012 }
4013 return NOTIFY_OK;
4014}
4015
Pekka Enberg06428782008-01-07 23:20:27 -08004016static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08004017 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08004018};
Christoph Lameter81819f02007-05-06 14:49:36 -07004019
4020#endif
4021
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004022void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004023{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004024 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004025 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004026
Christoph Lameterffadd4d2009-02-17 12:05:07 -05004027 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004028 return kmalloc_large(size, gfpflags);
4029
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004030 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004031
Satyam Sharma2408c552007-10-16 01:24:44 -07004032 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004033 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004034
Christoph Lameter2154a332010-07-09 14:07:10 -05004035 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004036
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004037 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004038 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004039
4040 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004041}
4042
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004043#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004044void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004045 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004046{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004047 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004048 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004049
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004050 if (unlikely(size > SLUB_MAX_SIZE)) {
4051 ret = kmalloc_large_node(size, gfpflags, node);
4052
4053 trace_kmalloc_node(caller, ret,
4054 size, PAGE_SIZE << get_order(size),
4055 gfpflags, node);
4056
4057 return ret;
4058 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004059
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004060 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004061
Satyam Sharma2408c552007-10-16 01:24:44 -07004062 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004063 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004064
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004065 ret = slab_alloc(s, gfpflags, node, caller);
4066
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004067 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004068 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004069
4070 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004071}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004072#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004073
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004074#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004075static int count_inuse(struct page *page)
4076{
4077 return page->inuse;
4078}
4079
4080static int count_total(struct page *page)
4081{
4082 return page->objects;
4083}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004084#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004085
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004086#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004087static int validate_slab(struct kmem_cache *s, struct page *page,
4088 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004089{
4090 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004091 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004092
4093 if (!check_slab(s, page) ||
4094 !on_freelist(s, page, NULL))
4095 return 0;
4096
4097 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004098 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004099
Christoph Lameter5f80b132011-04-15 14:48:13 -05004100 get_map(s, page, map);
4101 for_each_object(p, s, addr, page->objects) {
4102 if (test_bit(slab_index(p, s, addr), map))
4103 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4104 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004105 }
4106
Christoph Lameter224a88b2008-04-14 19:11:31 +03004107 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004108 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004109 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004110 return 0;
4111 return 1;
4112}
4113
Christoph Lameter434e2452007-07-17 04:03:30 -07004114static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4115 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004116{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004117 slab_lock(page);
4118 validate_slab(s, page, map);
4119 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004120}
4121
Christoph Lameter434e2452007-07-17 04:03:30 -07004122static int validate_slab_node(struct kmem_cache *s,
4123 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004124{
4125 unsigned long count = 0;
4126 struct page *page;
4127 unsigned long flags;
4128
4129 spin_lock_irqsave(&n->list_lock, flags);
4130
4131 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004132 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004133 count++;
4134 }
4135 if (count != n->nr_partial)
4136 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
4137 "counter=%ld\n", s->name, count, n->nr_partial);
4138
4139 if (!(s->flags & SLAB_STORE_USER))
4140 goto out;
4141
4142 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004143 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004144 count++;
4145 }
4146 if (count != atomic_long_read(&n->nr_slabs))
4147 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
4148 "counter=%ld\n", s->name, count,
4149 atomic_long_read(&n->nr_slabs));
4150
4151out:
4152 spin_unlock_irqrestore(&n->list_lock, flags);
4153 return count;
4154}
4155
Christoph Lameter434e2452007-07-17 04:03:30 -07004156static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004157{
4158 int node;
4159 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004160 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004161 sizeof(unsigned long), GFP_KERNEL);
4162
4163 if (!map)
4164 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004165
4166 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07004167 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07004168 struct kmem_cache_node *n = get_node(s, node);
4169
Christoph Lameter434e2452007-07-17 04:03:30 -07004170 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004171 }
Christoph Lameter434e2452007-07-17 04:03:30 -07004172 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004173 return count;
4174}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004175/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004176 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004177 * and freed.
4178 */
4179
4180struct location {
4181 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004182 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004183 long long sum_time;
4184 long min_time;
4185 long max_time;
4186 long min_pid;
4187 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304188 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004189 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004190};
4191
4192struct loc_track {
4193 unsigned long max;
4194 unsigned long count;
4195 struct location *loc;
4196};
4197
4198static void free_loc_track(struct loc_track *t)
4199{
4200 if (t->max)
4201 free_pages((unsigned long)t->loc,
4202 get_order(sizeof(struct location) * t->max));
4203}
4204
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004205static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004206{
4207 struct location *l;
4208 int order;
4209
Christoph Lameter88a420e2007-05-06 14:49:45 -07004210 order = get_order(sizeof(struct location) * max);
4211
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004212 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004213 if (!l)
4214 return 0;
4215
4216 if (t->count) {
4217 memcpy(l, t->loc, sizeof(struct location) * t->count);
4218 free_loc_track(t);
4219 }
4220 t->max = max;
4221 t->loc = l;
4222 return 1;
4223}
4224
4225static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004226 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004227{
4228 long start, end, pos;
4229 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004230 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004231 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004232
4233 start = -1;
4234 end = t->count;
4235
4236 for ( ; ; ) {
4237 pos = start + (end - start + 1) / 2;
4238
4239 /*
4240 * There is nothing at "end". If we end up there
4241 * we need to add something to before end.
4242 */
4243 if (pos == end)
4244 break;
4245
4246 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004247 if (track->addr == caddr) {
4248
4249 l = &t->loc[pos];
4250 l->count++;
4251 if (track->when) {
4252 l->sum_time += age;
4253 if (age < l->min_time)
4254 l->min_time = age;
4255 if (age > l->max_time)
4256 l->max_time = age;
4257
4258 if (track->pid < l->min_pid)
4259 l->min_pid = track->pid;
4260 if (track->pid > l->max_pid)
4261 l->max_pid = track->pid;
4262
Rusty Russell174596a2009-01-01 10:12:29 +10304263 cpumask_set_cpu(track->cpu,
4264 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004265 }
4266 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004267 return 1;
4268 }
4269
Christoph Lameter45edfa52007-05-09 02:32:45 -07004270 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004271 end = pos;
4272 else
4273 start = pos;
4274 }
4275
4276 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004277 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004278 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004279 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004280 return 0;
4281
4282 l = t->loc + pos;
4283 if (pos < t->count)
4284 memmove(l + 1, l,
4285 (t->count - pos) * sizeof(struct location));
4286 t->count++;
4287 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004288 l->addr = track->addr;
4289 l->sum_time = age;
4290 l->min_time = age;
4291 l->max_time = age;
4292 l->min_pid = track->pid;
4293 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304294 cpumask_clear(to_cpumask(l->cpus));
4295 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004296 nodes_clear(l->nodes);
4297 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004298 return 1;
4299}
4300
4301static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004302 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004303 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004304{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004305 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004306 void *p;
4307
Christoph Lameter39b26462008-04-14 19:11:30 +03004308 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004309 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004310
Christoph Lameter224a88b2008-04-14 19:11:31 +03004311 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004312 if (!test_bit(slab_index(p, s, addr), map))
4313 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004314}
4315
4316static int list_locations(struct kmem_cache *s, char *buf,
4317 enum track_item alloc)
4318{
Harvey Harrisone374d482008-01-31 15:20:50 -08004319 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004320 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004321 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004322 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004323 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4324 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004325
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004326 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4327 GFP_TEMPORARY)) {
4328 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004329 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004330 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004331 /* Push back cpu slabs */
4332 flush_all(s);
4333
Christoph Lameterf64dc582007-10-16 01:25:33 -07004334 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004335 struct kmem_cache_node *n = get_node(s, node);
4336 unsigned long flags;
4337 struct page *page;
4338
Christoph Lameter9e869432007-08-22 14:01:56 -07004339 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004340 continue;
4341
4342 spin_lock_irqsave(&n->list_lock, flags);
4343 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004344 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004345 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004346 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004347 spin_unlock_irqrestore(&n->list_lock, flags);
4348 }
4349
4350 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004351 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004352
Hugh Dickins9c246242008-12-09 13:14:27 -08004353 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004354 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004355 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004356
4357 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004358 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004359 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004360 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004361
4362 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004363 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004364 l->min_time,
4365 (long)div_u64(l->sum_time, l->count),
4366 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004367 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004368 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004369 l->min_time);
4370
4371 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004372 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004373 l->min_pid, l->max_pid);
4374 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004375 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004376 l->min_pid);
4377
Rusty Russell174596a2009-01-01 10:12:29 +10304378 if (num_online_cpus() > 1 &&
4379 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004380 len < PAGE_SIZE - 60) {
4381 len += sprintf(buf + len, " cpus=");
4382 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10304383 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004384 }
4385
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004386 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004387 len < PAGE_SIZE - 60) {
4388 len += sprintf(buf + len, " nodes=");
4389 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004390 l->nodes);
4391 }
4392
Harvey Harrisone374d482008-01-31 15:20:50 -08004393 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004394 }
4395
4396 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004397 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004398 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004399 len += sprintf(buf, "No data\n");
4400 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004401}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004402#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004403
Christoph Lametera5a84752010-10-05 13:57:27 -05004404#ifdef SLUB_RESILIENCY_TEST
4405static void resiliency_test(void)
4406{
4407 u8 *p;
4408
4409 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
4410
4411 printk(KERN_ERR "SLUB resiliency testing\n");
4412 printk(KERN_ERR "-----------------------\n");
4413 printk(KERN_ERR "A. Corruption after allocation\n");
4414
4415 p = kzalloc(16, GFP_KERNEL);
4416 p[16] = 0x12;
4417 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
4418 " 0x12->0x%p\n\n", p + 16);
4419
4420 validate_slab_cache(kmalloc_caches[4]);
4421
4422 /* Hmmm... The next two are dangerous */
4423 p = kzalloc(32, GFP_KERNEL);
4424 p[32 + sizeof(void *)] = 0x34;
4425 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
4426 " 0x34 -> -0x%p\n", p);
4427 printk(KERN_ERR
4428 "If allocated object is overwritten then not detectable\n\n");
4429
4430 validate_slab_cache(kmalloc_caches[5]);
4431 p = kzalloc(64, GFP_KERNEL);
4432 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4433 *p = 0x56;
4434 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4435 p);
4436 printk(KERN_ERR
4437 "If allocated object is overwritten then not detectable\n\n");
4438 validate_slab_cache(kmalloc_caches[6]);
4439
4440 printk(KERN_ERR "\nB. Corruption after free\n");
4441 p = kzalloc(128, GFP_KERNEL);
4442 kfree(p);
4443 *p = 0x78;
4444 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4445 validate_slab_cache(kmalloc_caches[7]);
4446
4447 p = kzalloc(256, GFP_KERNEL);
4448 kfree(p);
4449 p[50] = 0x9a;
4450 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4451 p);
4452 validate_slab_cache(kmalloc_caches[8]);
4453
4454 p = kzalloc(512, GFP_KERNEL);
4455 kfree(p);
4456 p[512] = 0xab;
4457 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4458 validate_slab_cache(kmalloc_caches[9]);
4459}
4460#else
4461#ifdef CONFIG_SYSFS
4462static void resiliency_test(void) {};
4463#endif
4464#endif
4465
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004466#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004467enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004468 SL_ALL, /* All slabs */
4469 SL_PARTIAL, /* Only partially allocated slabs */
4470 SL_CPU, /* Only slabs used for cpu caches */
4471 SL_OBJECTS, /* Determine allocated objects not slabs */
4472 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004473};
4474
Christoph Lameter205ab992008-04-14 19:11:40 +03004475#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004476#define SO_PARTIAL (1 << SL_PARTIAL)
4477#define SO_CPU (1 << SL_CPU)
4478#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004479#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004480
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004481static ssize_t show_slab_objects(struct kmem_cache *s,
4482 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004483{
4484 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004485 int node;
4486 int x;
4487 unsigned long *nodes;
4488 unsigned long *per_cpu;
4489
4490 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004491 if (!nodes)
4492 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004493 per_cpu = nodes + nr_node_ids;
4494
Christoph Lameter205ab992008-04-14 19:11:40 +03004495 if (flags & SO_CPU) {
4496 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004497
Christoph Lameter205ab992008-04-14 19:11:40 +03004498 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004499 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004500 int node = ACCESS_ONCE(c->node);
Christoph Lameter49e22582011-08-09 16:12:27 -05004501 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004502
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004503 if (node < 0)
Christoph Lameter205ab992008-04-14 19:11:40 +03004504 continue;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004505 page = ACCESS_ONCE(c->page);
4506 if (page) {
4507 if (flags & SO_TOTAL)
4508 x = page->objects;
Christoph Lameter205ab992008-04-14 19:11:40 +03004509 else if (flags & SO_OBJECTS)
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004510 x = page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07004511 else
4512 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03004513
Christoph Lameter81819f02007-05-06 14:49:36 -07004514 total += x;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004515 nodes[node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004516 }
Christoph Lameter49e22582011-08-09 16:12:27 -05004517 page = c->partial;
4518
4519 if (page) {
4520 x = page->pobjects;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004521 total += x;
4522 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004523 }
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004524 per_cpu[node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07004525 }
4526 }
4527
Christoph Lameter04d94872011-01-10 10:15:15 -06004528 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004529#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004530 if (flags & SO_ALL) {
4531 for_each_node_state(node, N_NORMAL_MEMORY) {
4532 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004533
Christoph Lameter205ab992008-04-14 19:11:40 +03004534 if (flags & SO_TOTAL)
4535 x = atomic_long_read(&n->total_objects);
4536 else if (flags & SO_OBJECTS)
4537 x = atomic_long_read(&n->total_objects) -
4538 count_partial(n, count_free);
4539
4540 else
4541 x = atomic_long_read(&n->nr_slabs);
4542 total += x;
4543 nodes[node] += x;
4544 }
4545
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004546 } else
4547#endif
4548 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004549 for_each_node_state(node, N_NORMAL_MEMORY) {
4550 struct kmem_cache_node *n = get_node(s, node);
4551
4552 if (flags & SO_TOTAL)
4553 x = count_partial(n, count_total);
4554 else if (flags & SO_OBJECTS)
4555 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004556 else
4557 x = n->nr_partial;
4558 total += x;
4559 nodes[node] += x;
4560 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004561 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004562 x = sprintf(buf, "%lu", total);
4563#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004564 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004565 if (nodes[node])
4566 x += sprintf(buf + x, " N%d=%lu",
4567 node, nodes[node]);
4568#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06004569 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07004570 kfree(nodes);
4571 return x + sprintf(buf + x, "\n");
4572}
4573
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004574#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004575static int any_slab_objects(struct kmem_cache *s)
4576{
4577 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004578
4579 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004580 struct kmem_cache_node *n = get_node(s, node);
4581
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004582 if (!n)
4583 continue;
4584
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004585 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004586 return 1;
4587 }
4588 return 0;
4589}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004590#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004591
4592#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004593#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004594
4595struct slab_attribute {
4596 struct attribute attr;
4597 ssize_t (*show)(struct kmem_cache *s, char *buf);
4598 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4599};
4600
4601#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004602 static struct slab_attribute _name##_attr = \
4603 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004604
4605#define SLAB_ATTR(_name) \
4606 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004607 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004608
Christoph Lameter81819f02007-05-06 14:49:36 -07004609static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4610{
4611 return sprintf(buf, "%d\n", s->size);
4612}
4613SLAB_ATTR_RO(slab_size);
4614
4615static ssize_t align_show(struct kmem_cache *s, char *buf)
4616{
4617 return sprintf(buf, "%d\n", s->align);
4618}
4619SLAB_ATTR_RO(align);
4620
4621static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4622{
4623 return sprintf(buf, "%d\n", s->objsize);
4624}
4625SLAB_ATTR_RO(object_size);
4626
4627static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4628{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004629 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004630}
4631SLAB_ATTR_RO(objs_per_slab);
4632
Christoph Lameter06b285d2008-04-14 19:11:41 +03004633static ssize_t order_store(struct kmem_cache *s,
4634 const char *buf, size_t length)
4635{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004636 unsigned long order;
4637 int err;
4638
4639 err = strict_strtoul(buf, 10, &order);
4640 if (err)
4641 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004642
4643 if (order > slub_max_order || order < slub_min_order)
4644 return -EINVAL;
4645
4646 calculate_sizes(s, order);
4647 return length;
4648}
4649
Christoph Lameter81819f02007-05-06 14:49:36 -07004650static ssize_t order_show(struct kmem_cache *s, char *buf)
4651{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004652 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004653}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004654SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004655
David Rientjes73d342b2009-02-22 17:40:09 -08004656static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4657{
4658 return sprintf(buf, "%lu\n", s->min_partial);
4659}
4660
4661static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4662 size_t length)
4663{
4664 unsigned long min;
4665 int err;
4666
4667 err = strict_strtoul(buf, 10, &min);
4668 if (err)
4669 return err;
4670
David Rientjesc0bdb232009-02-25 09:16:35 +02004671 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004672 return length;
4673}
4674SLAB_ATTR(min_partial);
4675
Christoph Lameter49e22582011-08-09 16:12:27 -05004676static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4677{
4678 return sprintf(buf, "%u\n", s->cpu_partial);
4679}
4680
4681static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4682 size_t length)
4683{
4684 unsigned long objects;
4685 int err;
4686
4687 err = strict_strtoul(buf, 10, &objects);
4688 if (err)
4689 return err;
David Rientjes74ee4ef2012-01-09 13:19:45 -08004690 if (objects && kmem_cache_debug(s))
4691 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004692
4693 s->cpu_partial = objects;
4694 flush_all(s);
4695 return length;
4696}
4697SLAB_ATTR(cpu_partial);
4698
Christoph Lameter81819f02007-05-06 14:49:36 -07004699static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4700{
Joe Perches62c70bc2011-01-13 15:45:52 -08004701 if (!s->ctor)
4702 return 0;
4703 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004704}
4705SLAB_ATTR_RO(ctor);
4706
Christoph Lameter81819f02007-05-06 14:49:36 -07004707static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4708{
4709 return sprintf(buf, "%d\n", s->refcount - 1);
4710}
4711SLAB_ATTR_RO(aliases);
4712
Christoph Lameter81819f02007-05-06 14:49:36 -07004713static ssize_t partial_show(struct kmem_cache *s, char *buf)
4714{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004715 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004716}
4717SLAB_ATTR_RO(partial);
4718
4719static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4720{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004721 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004722}
4723SLAB_ATTR_RO(cpu_slabs);
4724
4725static ssize_t objects_show(struct kmem_cache *s, char *buf)
4726{
Christoph Lameter205ab992008-04-14 19:11:40 +03004727 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004728}
4729SLAB_ATTR_RO(objects);
4730
Christoph Lameter205ab992008-04-14 19:11:40 +03004731static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4732{
4733 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4734}
4735SLAB_ATTR_RO(objects_partial);
4736
Christoph Lameter49e22582011-08-09 16:12:27 -05004737static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4738{
4739 int objects = 0;
4740 int pages = 0;
4741 int cpu;
4742 int len;
4743
4744 for_each_online_cpu(cpu) {
4745 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4746
4747 if (page) {
4748 pages += page->pages;
4749 objects += page->pobjects;
4750 }
4751 }
4752
4753 len = sprintf(buf, "%d(%d)", objects, pages);
4754
4755#ifdef CONFIG_SMP
4756 for_each_online_cpu(cpu) {
4757 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4758
4759 if (page && len < PAGE_SIZE - 20)
4760 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4761 page->pobjects, page->pages);
4762 }
4763#endif
4764 return len + sprintf(buf + len, "\n");
4765}
4766SLAB_ATTR_RO(slabs_cpu_partial);
4767
Christoph Lameter81819f02007-05-06 14:49:36 -07004768static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4769{
4770 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4771}
4772
4773static ssize_t reclaim_account_store(struct kmem_cache *s,
4774 const char *buf, size_t length)
4775{
4776 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4777 if (buf[0] == '1')
4778 s->flags |= SLAB_RECLAIM_ACCOUNT;
4779 return length;
4780}
4781SLAB_ATTR(reclaim_account);
4782
4783static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4784{
Christoph Lameter5af60832007-05-06 14:49:56 -07004785 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004786}
4787SLAB_ATTR_RO(hwcache_align);
4788
4789#ifdef CONFIG_ZONE_DMA
4790static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4791{
4792 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4793}
4794SLAB_ATTR_RO(cache_dma);
4795#endif
4796
4797static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4798{
4799 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4800}
4801SLAB_ATTR_RO(destroy_by_rcu);
4802
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004803static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4804{
4805 return sprintf(buf, "%d\n", s->reserved);
4806}
4807SLAB_ATTR_RO(reserved);
4808
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004809#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004810static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4811{
4812 return show_slab_objects(s, buf, SO_ALL);
4813}
4814SLAB_ATTR_RO(slabs);
4815
4816static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4817{
4818 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4819}
4820SLAB_ATTR_RO(total_objects);
4821
4822static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4823{
4824 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4825}
4826
4827static ssize_t sanity_checks_store(struct kmem_cache *s,
4828 const char *buf, size_t length)
4829{
4830 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004831 if (buf[0] == '1') {
4832 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004833 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004834 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004835 return length;
4836}
4837SLAB_ATTR(sanity_checks);
4838
4839static ssize_t trace_show(struct kmem_cache *s, char *buf)
4840{
4841 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4842}
4843
4844static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4845 size_t length)
4846{
4847 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004848 if (buf[0] == '1') {
4849 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004850 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004851 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004852 return length;
4853}
4854SLAB_ATTR(trace);
4855
Christoph Lameter81819f02007-05-06 14:49:36 -07004856static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4857{
4858 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4859}
4860
4861static ssize_t red_zone_store(struct kmem_cache *s,
4862 const char *buf, size_t length)
4863{
4864 if (any_slab_objects(s))
4865 return -EBUSY;
4866
4867 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004868 if (buf[0] == '1') {
4869 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004870 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004871 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004872 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004873 return length;
4874}
4875SLAB_ATTR(red_zone);
4876
4877static ssize_t poison_show(struct kmem_cache *s, char *buf)
4878{
4879 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4880}
4881
4882static ssize_t poison_store(struct kmem_cache *s,
4883 const char *buf, size_t length)
4884{
4885 if (any_slab_objects(s))
4886 return -EBUSY;
4887
4888 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004889 if (buf[0] == '1') {
4890 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004891 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004892 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004893 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004894 return length;
4895}
4896SLAB_ATTR(poison);
4897
4898static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4899{
4900 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4901}
4902
4903static ssize_t store_user_store(struct kmem_cache *s,
4904 const char *buf, size_t length)
4905{
4906 if (any_slab_objects(s))
4907 return -EBUSY;
4908
4909 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004910 if (buf[0] == '1') {
4911 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004912 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004913 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004914 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004915 return length;
4916}
4917SLAB_ATTR(store_user);
4918
Christoph Lameter53e15af2007-05-06 14:49:43 -07004919static ssize_t validate_show(struct kmem_cache *s, char *buf)
4920{
4921 return 0;
4922}
4923
4924static ssize_t validate_store(struct kmem_cache *s,
4925 const char *buf, size_t length)
4926{
Christoph Lameter434e2452007-07-17 04:03:30 -07004927 int ret = -EINVAL;
4928
4929 if (buf[0] == '1') {
4930 ret = validate_slab_cache(s);
4931 if (ret >= 0)
4932 ret = length;
4933 }
4934 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004935}
4936SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004937
4938static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4939{
4940 if (!(s->flags & SLAB_STORE_USER))
4941 return -ENOSYS;
4942 return list_locations(s, buf, TRACK_ALLOC);
4943}
4944SLAB_ATTR_RO(alloc_calls);
4945
4946static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4947{
4948 if (!(s->flags & SLAB_STORE_USER))
4949 return -ENOSYS;
4950 return list_locations(s, buf, TRACK_FREE);
4951}
4952SLAB_ATTR_RO(free_calls);
4953#endif /* CONFIG_SLUB_DEBUG */
4954
4955#ifdef CONFIG_FAILSLAB
4956static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4957{
4958 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4959}
4960
4961static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4962 size_t length)
4963{
4964 s->flags &= ~SLAB_FAILSLAB;
4965 if (buf[0] == '1')
4966 s->flags |= SLAB_FAILSLAB;
4967 return length;
4968}
4969SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004970#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004971
Christoph Lameter2086d262007-05-06 14:49:46 -07004972static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4973{
4974 return 0;
4975}
4976
4977static ssize_t shrink_store(struct kmem_cache *s,
4978 const char *buf, size_t length)
4979{
4980 if (buf[0] == '1') {
4981 int rc = kmem_cache_shrink(s);
4982
4983 if (rc)
4984 return rc;
4985 } else
4986 return -EINVAL;
4987 return length;
4988}
4989SLAB_ATTR(shrink);
4990
Christoph Lameter81819f02007-05-06 14:49:36 -07004991#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004992static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004993{
Christoph Lameter98246012008-01-07 23:20:26 -08004994 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004995}
4996
Christoph Lameter98246012008-01-07 23:20:26 -08004997static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004998 const char *buf, size_t length)
4999{
Christoph Lameter0121c6192008-04-29 16:11:12 -07005000 unsigned long ratio;
5001 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07005002
Christoph Lameter0121c6192008-04-29 16:11:12 -07005003 err = strict_strtoul(buf, 10, &ratio);
5004 if (err)
5005 return err;
5006
Christoph Lametere2cb96b2008-08-19 08:51:22 -05005007 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07005008 s->remote_node_defrag_ratio = ratio * 10;
5009
Christoph Lameter81819f02007-05-06 14:49:36 -07005010 return length;
5011}
Christoph Lameter98246012008-01-07 23:20:26 -08005012SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07005013#endif
5014
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005015#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005016static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5017{
5018 unsigned long sum = 0;
5019 int cpu;
5020 int len;
5021 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
5022
5023 if (!data)
5024 return -ENOMEM;
5025
5026 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005027 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005028
5029 data[cpu] = x;
5030 sum += x;
5031 }
5032
5033 len = sprintf(buf, "%lu", sum);
5034
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005035#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005036 for_each_online_cpu(cpu) {
5037 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005038 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005039 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005040#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005041 kfree(data);
5042 return len + sprintf(buf + len, "\n");
5043}
5044
David Rientjes78eb00c2009-10-15 02:20:22 -07005045static void clear_stat(struct kmem_cache *s, enum stat_item si)
5046{
5047 int cpu;
5048
5049 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005050 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005051}
5052
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005053#define STAT_ATTR(si, text) \
5054static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5055{ \
5056 return show_stat(s, buf, si); \
5057} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005058static ssize_t text##_store(struct kmem_cache *s, \
5059 const char *buf, size_t length) \
5060{ \
5061 if (buf[0] != '0') \
5062 return -EINVAL; \
5063 clear_stat(s, si); \
5064 return length; \
5065} \
5066SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005067
5068STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5069STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5070STAT_ATTR(FREE_FASTPATH, free_fastpath);
5071STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5072STAT_ATTR(FREE_FROZEN, free_frozen);
5073STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5074STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5075STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5076STAT_ATTR(ALLOC_SLAB, alloc_slab);
5077STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005078STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005079STAT_ATTR(FREE_SLAB, free_slab);
5080STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5081STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5082STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5083STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5084STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5085STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005086STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005087STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005088STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5089STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005090STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5091STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08005092STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5093STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005094#endif
5095
Pekka Enberg06428782008-01-07 23:20:27 -08005096static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005097 &slab_size_attr.attr,
5098 &object_size_attr.attr,
5099 &objs_per_slab_attr.attr,
5100 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005101 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005102 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005103 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005104 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005105 &partial_attr.attr,
5106 &cpu_slabs_attr.attr,
5107 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005108 &aliases_attr.attr,
5109 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005110 &hwcache_align_attr.attr,
5111 &reclaim_account_attr.attr,
5112 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005113 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005114 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005115 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005116#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005117 &total_objects_attr.attr,
5118 &slabs_attr.attr,
5119 &sanity_checks_attr.attr,
5120 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005121 &red_zone_attr.attr,
5122 &poison_attr.attr,
5123 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005124 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005125 &alloc_calls_attr.attr,
5126 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005127#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005128#ifdef CONFIG_ZONE_DMA
5129 &cache_dma_attr.attr,
5130#endif
5131#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005132 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005133#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005134#ifdef CONFIG_SLUB_STATS
5135 &alloc_fastpath_attr.attr,
5136 &alloc_slowpath_attr.attr,
5137 &free_fastpath_attr.attr,
5138 &free_slowpath_attr.attr,
5139 &free_frozen_attr.attr,
5140 &free_add_partial_attr.attr,
5141 &free_remove_partial_attr.attr,
5142 &alloc_from_partial_attr.attr,
5143 &alloc_slab_attr.attr,
5144 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005145 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005146 &free_slab_attr.attr,
5147 &cpuslab_flush_attr.attr,
5148 &deactivate_full_attr.attr,
5149 &deactivate_empty_attr.attr,
5150 &deactivate_to_head_attr.attr,
5151 &deactivate_to_tail_attr.attr,
5152 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005153 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005154 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005155 &cmpxchg_double_fail_attr.attr,
5156 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005157 &cpu_partial_alloc_attr.attr,
5158 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005159 &cpu_partial_node_attr.attr,
5160 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005161#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005162#ifdef CONFIG_FAILSLAB
5163 &failslab_attr.attr,
5164#endif
5165
Christoph Lameter81819f02007-05-06 14:49:36 -07005166 NULL
5167};
5168
5169static struct attribute_group slab_attr_group = {
5170 .attrs = slab_attrs,
5171};
5172
5173static ssize_t slab_attr_show(struct kobject *kobj,
5174 struct attribute *attr,
5175 char *buf)
5176{
5177 struct slab_attribute *attribute;
5178 struct kmem_cache *s;
5179 int err;
5180
5181 attribute = to_slab_attr(attr);
5182 s = to_slab(kobj);
5183
5184 if (!attribute->show)
5185 return -EIO;
5186
5187 err = attribute->show(s, buf);
5188
5189 return err;
5190}
5191
5192static ssize_t slab_attr_store(struct kobject *kobj,
5193 struct attribute *attr,
5194 const char *buf, size_t len)
5195{
5196 struct slab_attribute *attribute;
5197 struct kmem_cache *s;
5198 int err;
5199
5200 attribute = to_slab_attr(attr);
5201 s = to_slab(kobj);
5202
5203 if (!attribute->store)
5204 return -EIO;
5205
5206 err = attribute->store(s, buf, len);
5207
5208 return err;
5209}
5210
Christoph Lameter151c6022008-01-07 22:29:05 -08005211static void kmem_cache_release(struct kobject *kobj)
5212{
5213 struct kmem_cache *s = to_slab(kobj);
5214
Pekka Enberg84c1cf62010-09-14 23:21:12 +03005215 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08005216 kfree(s);
5217}
5218
Emese Revfy52cf25d2010-01-19 02:58:23 +01005219static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005220 .show = slab_attr_show,
5221 .store = slab_attr_store,
5222};
5223
5224static struct kobj_type slab_ktype = {
5225 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08005226 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07005227};
5228
5229static int uevent_filter(struct kset *kset, struct kobject *kobj)
5230{
5231 struct kobj_type *ktype = get_ktype(kobj);
5232
5233 if (ktype == &slab_ktype)
5234 return 1;
5235 return 0;
5236}
5237
Emese Revfy9cd43612009-12-31 14:52:51 +01005238static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005239 .filter = uevent_filter,
5240};
5241
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005242static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005243
5244#define ID_STR_LENGTH 64
5245
5246/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005247 *
5248 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005249 */
5250static char *create_unique_id(struct kmem_cache *s)
5251{
5252 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5253 char *p = name;
5254
5255 BUG_ON(!name);
5256
5257 *p++ = ':';
5258 /*
5259 * First flags affecting slabcache operations. We will only
5260 * get here for aliasable slabs so we do not need to support
5261 * too many flags. The flags here must cover all flags that
5262 * are matched during merging to guarantee that the id is
5263 * unique.
5264 */
5265 if (s->flags & SLAB_CACHE_DMA)
5266 *p++ = 'd';
5267 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5268 *p++ = 'a';
5269 if (s->flags & SLAB_DEBUG_FREE)
5270 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005271 if (!(s->flags & SLAB_NOTRACK))
5272 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005273 if (p != name + 1)
5274 *p++ = '-';
5275 p += sprintf(p, "%07d", s->size);
5276 BUG_ON(p > name + ID_STR_LENGTH - 1);
5277 return name;
5278}
5279
5280static int sysfs_slab_add(struct kmem_cache *s)
5281{
5282 int err;
5283 const char *name;
5284 int unmergeable;
5285
5286 if (slab_state < SYSFS)
5287 /* Defer until later */
5288 return 0;
5289
5290 unmergeable = slab_unmergeable(s);
5291 if (unmergeable) {
5292 /*
5293 * Slabcache can never be merged so we can use the name proper.
5294 * This is typically the case for debug situations. In that
5295 * case we can catch duplicate names easily.
5296 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005297 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005298 name = s->name;
5299 } else {
5300 /*
5301 * Create a unique name for the slab as a target
5302 * for the symlinks.
5303 */
5304 name = create_unique_id(s);
5305 }
5306
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005307 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005308 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
5309 if (err) {
5310 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005311 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07005312 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005313
5314 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005315 if (err) {
5316 kobject_del(&s->kobj);
5317 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005318 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08005319 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005320 kobject_uevent(&s->kobj, KOBJ_ADD);
5321 if (!unmergeable) {
5322 /* Setup first alias */
5323 sysfs_slab_alias(s, s->name);
5324 kfree(name);
5325 }
5326 return 0;
5327}
5328
5329static void sysfs_slab_remove(struct kmem_cache *s)
5330{
Christoph Lameter2bce6482010-07-19 11:39:11 -05005331 if (slab_state < SYSFS)
5332 /*
5333 * Sysfs has not been setup yet so no need to remove the
5334 * cache from sysfs.
5335 */
5336 return;
5337
Christoph Lameter81819f02007-05-06 14:49:36 -07005338 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5339 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005340 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005341}
5342
5343/*
5344 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005345 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005346 */
5347struct saved_alias {
5348 struct kmem_cache *s;
5349 const char *name;
5350 struct saved_alias *next;
5351};
5352
Adrian Bunk5af328a2007-07-17 04:03:27 -07005353static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005354
5355static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5356{
5357 struct saved_alias *al;
5358
5359 if (slab_state == SYSFS) {
5360 /*
5361 * If we have a leftover link then remove it.
5362 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005363 sysfs_remove_link(&slab_kset->kobj, name);
5364 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005365 }
5366
5367 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5368 if (!al)
5369 return -ENOMEM;
5370
5371 al->s = s;
5372 al->name = name;
5373 al->next = alias_list;
5374 alias_list = al;
5375 return 0;
5376}
5377
5378static int __init slab_sysfs_init(void)
5379{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005380 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005381 int err;
5382
Christoph Lameter2bce6482010-07-19 11:39:11 -05005383 down_write(&slub_lock);
5384
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005385 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005386 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05005387 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005388 printk(KERN_ERR "Cannot register slab subsystem.\n");
5389 return -ENOSYS;
5390 }
5391
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005392 slab_state = SYSFS;
5393
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005394 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005395 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005396 if (err)
5397 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
5398 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005399 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005400
5401 while (alias_list) {
5402 struct saved_alias *al = alias_list;
5403
5404 alias_list = alias_list->next;
5405 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005406 if (err)
5407 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
5408 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005409 kfree(al);
5410 }
5411
Christoph Lameter2bce6482010-07-19 11:39:11 -05005412 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07005413 resiliency_test();
5414 return 0;
5415}
5416
5417__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005418#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005419
5420/*
5421 * The /proc/slabinfo ABI
5422 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005423#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005424static void print_slabinfo_header(struct seq_file *m)
5425{
5426 seq_puts(m, "slabinfo - version: 2.1\n");
5427 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
5428 "<objperslab> <pagesperslab>");
5429 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
5430 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5431 seq_putc(m, '\n');
5432}
5433
5434static void *s_start(struct seq_file *m, loff_t *pos)
5435{
5436 loff_t n = *pos;
5437
5438 down_read(&slub_lock);
5439 if (!n)
5440 print_slabinfo_header(m);
5441
5442 return seq_list_start(&slab_caches, *pos);
5443}
5444
5445static void *s_next(struct seq_file *m, void *p, loff_t *pos)
5446{
5447 return seq_list_next(p, &slab_caches, pos);
5448}
5449
5450static void s_stop(struct seq_file *m, void *p)
5451{
5452 up_read(&slub_lock);
5453}
5454
5455static int s_show(struct seq_file *m, void *p)
5456{
5457 unsigned long nr_partials = 0;
5458 unsigned long nr_slabs = 0;
5459 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005460 unsigned long nr_objs = 0;
5461 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005462 struct kmem_cache *s;
5463 int node;
5464
5465 s = list_entry(p, struct kmem_cache, list);
5466
5467 for_each_online_node(node) {
5468 struct kmem_cache_node *n = get_node(s, node);
5469
5470 if (!n)
5471 continue;
5472
5473 nr_partials += n->nr_partial;
5474 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03005475 nr_objs += atomic_long_read(&n->total_objects);
5476 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005477 }
5478
Christoph Lameter205ab992008-04-14 19:11:40 +03005479 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005480
5481 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03005482 nr_objs, s->size, oo_objects(s->oo),
5483 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005484 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
5485 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
5486 0UL);
5487 seq_putc(m, '\n');
5488 return 0;
5489}
5490
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005491static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005492 .start = s_start,
5493 .next = s_next,
5494 .stop = s_stop,
5495 .show = s_show,
5496};
5497
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005498static int slabinfo_open(struct inode *inode, struct file *file)
5499{
5500 return seq_open(file, &slabinfo_op);
5501}
5502
5503static const struct file_operations proc_slabinfo_operations = {
5504 .open = slabinfo_open,
5505 .read = seq_read,
5506 .llseek = seq_lseek,
5507 .release = seq_release,
5508};
5509
5510static int __init slab_proc_init(void)
5511{
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04005512 proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005513 return 0;
5514}
5515module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08005516#endif /* CONFIG_SLABINFO */