blob: a9201d83178b8d47d38897ff44814225e37b3102 [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
7 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -07009 */
10
11#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100012#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070013#include <linux/module.h>
14#include <linux/bit_spinlock.h>
15#include <linux/interrupt.h>
16#include <linux/bitops.h>
17#include <linux/slab.h>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040018#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070019#include <linux/seq_file.h>
Zhaolei02af61b2009-04-10 14:26:18 +080020#include <linux/kmemtrace.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>
Catalin Marinas06f22f12009-06-11 13:23:18 +010024#include <linux/kmemleak.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070025#include <linux/mempolicy.h>
26#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070027#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070028#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070029#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070030#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090031#include <linux/fault-inject.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070032
33/*
34 * Lock order:
35 * 1. slab_lock(page)
36 * 2. slab->list_lock
37 *
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
44 *
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
50 *
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
55 * the list lock.
56 *
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
69 *
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
74 *
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
77 *
Christoph Lameter672bba32007-05-09 02:32:39 -070078 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070080 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070081 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070083 *
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
87 *
88 * Overloading of page flags that are otherwise used for LRU management.
89 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070090 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
98 *
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700102 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700105 *
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700108 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700109 */
110
Christoph Lameter5577bd82007-05-16 22:10:56 -0700111#ifdef CONFIG_SLUB_DEBUG
Andy Whitcroft8a380822008-07-23 21:27:18 -0700112#define SLABDEBUG 1
Christoph Lameter5577bd82007-05-16 22:10:56 -0700113#else
114#define SLABDEBUG 0
115#endif
116
Christoph Lameter81819f02007-05-06 14:49:36 -0700117/*
118 * Issues still to be resolved:
119 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700120 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
121 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700122 * - Variable sizing of the per node arrays
123 */
124
125/* Enable to test recovery from slab corruption on boot */
126#undef SLUB_RESILIENCY_TEST
127
Christoph Lameter81819f02007-05-06 14:49:36 -0700128/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700129 * Mininum number of partial slabs. These will be left on the partial
130 * lists even if they are empty. kmem_cache_shrink may reclaim them.
131 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800132#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700133
Christoph Lameter2086d262007-05-06 14:49:46 -0700134/*
135 * Maximum number of desirable partial slabs.
136 * The existence of more partial slabs makes kmem_cache_shrink
137 * sort the partial list by the number of objects in the.
138 */
139#define MAX_PARTIAL 10
140
Christoph Lameter81819f02007-05-06 14:49:36 -0700141#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
142 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700143
Christoph Lameter81819f02007-05-06 14:49:36 -0700144/*
145 * Set of flags that will prevent slab merging
146 */
147#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Catalin Marinas06f22f12009-06-11 13:23:18 +0100148 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700149
150#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200151 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700152
153#ifndef ARCH_KMALLOC_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700154#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700155#endif
156
157#ifndef ARCH_SLAB_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700158#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700159#endif
160
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400161#define OO_SHIFT 16
162#define OO_MASK ((1 << OO_SHIFT) - 1)
163#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
164
Christoph Lameter81819f02007-05-06 14:49:36 -0700165/* Internal SLUB flags */
Christoph Lameter1ceef402007-08-07 15:11:48 -0700166#define __OBJECT_POISON 0x80000000 /* Poison object */
167#define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700168
169static int kmem_size = sizeof(struct kmem_cache);
170
171#ifdef CONFIG_SMP
172static struct notifier_block slab_notifier;
173#endif
174
175static enum {
176 DOWN, /* No slab functionality available */
177 PARTIAL, /* kmem_cache_open() works but kmalloc does not */
Christoph Lameter672bba32007-05-09 02:32:39 -0700178 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700179 SYSFS /* Sysfs up */
180} slab_state = DOWN;
181
182/* A list of all slab caches on the system */
183static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700184static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700185
Christoph Lameter02cbc872007-05-09 02:32:43 -0700186/*
187 * Tracking user of a slab.
188 */
189struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300190 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700191 int cpu; /* Was running on cpu */
192 int pid; /* Pid context */
193 unsigned long when; /* When did the operation occur */
194};
195
196enum track_item { TRACK_ALLOC, TRACK_FREE };
197
Christoph Lameterf6acb632008-04-29 16:16:06 -0700198#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -0700199static int sysfs_slab_add(struct kmem_cache *);
200static int sysfs_slab_alias(struct kmem_cache *, const char *);
201static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800202
Christoph Lameter81819f02007-05-06 14:49:36 -0700203#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700204static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
205static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
206 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800207static inline void sysfs_slab_remove(struct kmem_cache *s)
208{
209 kfree(s);
210}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800211
Christoph Lameter81819f02007-05-06 14:49:36 -0700212#endif
213
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800214static inline void stat(struct kmem_cache_cpu *c, enum stat_item si)
215{
216#ifdef CONFIG_SLUB_STATS
217 c->stat[si]++;
218#endif
219}
220
Christoph Lameter81819f02007-05-06 14:49:36 -0700221/********************************************************************
222 * Core slab cache functions
223 *******************************************************************/
224
225int slab_is_available(void)
226{
227 return slab_state >= UP;
228}
229
230static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
231{
232#ifdef CONFIG_NUMA
233 return s->node[node];
234#else
235 return &s->local_node;
236#endif
237}
238
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700239static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu)
240{
Christoph Lameter4c93c3552007-10-16 01:26:08 -0700241#ifdef CONFIG_SMP
242 return s->cpu_slab[cpu];
243#else
244 return &s->cpu_slab;
245#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700246}
247
Christoph Lameter6446faa2008-02-15 23:45:26 -0800248/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700249static inline int check_valid_pointer(struct kmem_cache *s,
250 struct page *page, const void *object)
251{
252 void *base;
253
Christoph Lametera973e9d2008-03-01 13:40:44 -0800254 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700255 return 1;
256
Christoph Lametera973e9d2008-03-01 13:40:44 -0800257 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300258 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700259 (object - base) % s->size) {
260 return 0;
261 }
262
263 return 1;
264}
265
Christoph Lameter81819f02007-05-06 14:49:36 -0700266/*
Christoph Lameter7656c722007-05-09 02:32:40 -0700267 * Slow version of get and set free pointer.
268 *
269 * This version requires touching the cache lines of kmem_cache which
270 * we avoid to do in the fast alloc free paths. There we obtain the offset
271 * from the page struct.
272 */
273static inline void *get_freepointer(struct kmem_cache *s, void *object)
274{
275 return *(void **)(object + s->offset);
276}
277
278static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
279{
280 *(void **)(object + s->offset) = fp;
281}
282
283/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300284#define for_each_object(__p, __s, __addr, __objects) \
285 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700286 __p += (__s)->size)
287
288/* Scan freelist */
289#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800290 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700291
292/* Determine object index from a given position */
293static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
294{
295 return (p - addr) / s->size;
296}
297
Christoph Lameter834f3d12008-04-14 19:11:31 +0300298static inline struct kmem_cache_order_objects oo_make(int order,
299 unsigned long size)
300{
301 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400302 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300303 };
304
305 return x;
306}
307
308static inline int oo_order(struct kmem_cache_order_objects x)
309{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400310 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300311}
312
313static inline int oo_objects(struct kmem_cache_order_objects x)
314{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400315 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300316}
317
Christoph Lameter41ecc552007-05-09 02:32:44 -0700318#ifdef CONFIG_SLUB_DEBUG
319/*
320 * Debug settings:
321 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700322#ifdef CONFIG_SLUB_DEBUG_ON
323static int slub_debug = DEBUG_DEFAULT_FLAGS;
324#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700325static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700326#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700327
328static char *slub_debug_slabs;
329
Christoph Lameter7656c722007-05-09 02:32:40 -0700330/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700331 * Object debugging
332 */
333static void print_section(char *text, u8 *addr, unsigned int length)
334{
335 int i, offset;
336 int newline = 1;
337 char ascii[17];
338
339 ascii[16] = 0;
340
341 for (i = 0; i < length; i++) {
342 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700343 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700344 newline = 0;
345 }
Pekka Enberg06428782008-01-07 23:20:27 -0800346 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700347 offset = i % 16;
348 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
349 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800350 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700351 newline = 1;
352 }
353 }
354 if (!newline) {
355 i %= 16;
356 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800357 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700358 ascii[i] = ' ';
359 i++;
360 }
Pekka Enberg06428782008-01-07 23:20:27 -0800361 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700362 }
363}
364
Christoph Lameter81819f02007-05-06 14:49:36 -0700365static struct track *get_track(struct kmem_cache *s, void *object,
366 enum track_item alloc)
367{
368 struct track *p;
369
370 if (s->offset)
371 p = object + s->offset + sizeof(void *);
372 else
373 p = object + s->inuse;
374
375 return p + alloc;
376}
377
378static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300379 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700380{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900381 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700382
Christoph Lameter81819f02007-05-06 14:49:36 -0700383 if (addr) {
384 p->addr = addr;
385 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400386 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700387 p->when = jiffies;
388 } else
389 memset(p, 0, sizeof(struct track));
390}
391
Christoph Lameter81819f02007-05-06 14:49:36 -0700392static void init_tracking(struct kmem_cache *s, void *object)
393{
Christoph Lameter24922682007-07-17 04:03:18 -0700394 if (!(s->flags & SLAB_STORE_USER))
395 return;
396
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300397 set_track(s, object, TRACK_FREE, 0UL);
398 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700399}
400
401static void print_track(const char *s, struct track *t)
402{
403 if (!t->addr)
404 return;
405
Linus Torvalds7daf7052008-07-14 12:12:53 -0700406 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300407 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700408}
409
Christoph Lameter24922682007-07-17 04:03:18 -0700410static void print_tracking(struct kmem_cache *s, void *object)
411{
412 if (!(s->flags & SLAB_STORE_USER))
413 return;
414
415 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
416 print_track("Freed", get_track(s, object, TRACK_FREE));
417}
418
419static void print_page_info(struct page *page)
420{
Christoph Lameter39b26462008-04-14 19:11:30 +0300421 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
422 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700423
424}
425
426static void slab_bug(struct kmem_cache *s, char *fmt, ...)
427{
428 va_list args;
429 char buf[100];
430
431 va_start(args, fmt);
432 vsnprintf(buf, sizeof(buf), fmt, args);
433 va_end(args);
434 printk(KERN_ERR "========================================"
435 "=====================================\n");
436 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
437 printk(KERN_ERR "----------------------------------------"
438 "-------------------------------------\n\n");
439}
440
441static void slab_fix(struct kmem_cache *s, char *fmt, ...)
442{
443 va_list args;
444 char buf[100];
445
446 va_start(args, fmt);
447 vsnprintf(buf, sizeof(buf), fmt, args);
448 va_end(args);
449 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
450}
451
452static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700453{
454 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800455 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700456
457 print_tracking(s, p);
458
459 print_page_info(page);
460
461 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
462 p, p - addr, get_freepointer(s, p));
463
464 if (p > addr + 16)
465 print_section("Bytes b4", p - 16, 16);
466
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300467 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700468
469 if (s->flags & SLAB_RED_ZONE)
470 print_section("Redzone", p + s->objsize,
471 s->inuse - s->objsize);
472
Christoph Lameter81819f02007-05-06 14:49:36 -0700473 if (s->offset)
474 off = s->offset + sizeof(void *);
475 else
476 off = s->inuse;
477
Christoph Lameter24922682007-07-17 04:03:18 -0700478 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700479 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700480
481 if (off != s->size)
482 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700483 print_section("Padding", p + off, s->size - off);
484
485 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700486}
487
488static void object_err(struct kmem_cache *s, struct page *page,
489 u8 *object, char *reason)
490{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700491 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700492 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700493}
494
Christoph Lameter24922682007-07-17 04:03:18 -0700495static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700496{
497 va_list args;
498 char buf[100];
499
Christoph Lameter24922682007-07-17 04:03:18 -0700500 va_start(args, fmt);
501 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700502 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700503 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700504 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700505 dump_stack();
506}
507
508static void init_object(struct kmem_cache *s, void *object, int active)
509{
510 u8 *p = object;
511
512 if (s->flags & __OBJECT_POISON) {
513 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800514 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700515 }
516
517 if (s->flags & SLAB_RED_ZONE)
518 memset(p + s->objsize,
519 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
520 s->inuse - s->objsize);
521}
522
Christoph Lameter24922682007-07-17 04:03:18 -0700523static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700524{
525 while (bytes) {
526 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700527 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700528 start++;
529 bytes--;
530 }
Christoph Lameter24922682007-07-17 04:03:18 -0700531 return NULL;
532}
533
534static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
535 void *from, void *to)
536{
537 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
538 memset(from, data, to - from);
539}
540
541static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
542 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800543 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700544{
545 u8 *fault;
546 u8 *end;
547
548 fault = check_bytes(start, value, bytes);
549 if (!fault)
550 return 1;
551
552 end = start + bytes;
553 while (end > fault && end[-1] == value)
554 end--;
555
556 slab_bug(s, "%s overwritten", what);
557 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
558 fault, end - 1, fault[0], value);
559 print_trailer(s, page, object);
560
561 restore_bytes(s, what, value, fault, end);
562 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700563}
564
Christoph Lameter81819f02007-05-06 14:49:36 -0700565/*
566 * Object layout:
567 *
568 * object address
569 * Bytes of the object to be managed.
570 * If the freepointer may overlay the object then the free
571 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700572 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700573 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
574 * 0xa5 (POISON_END)
575 *
576 * object + s->objsize
577 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700578 * Padding is extended by another word if Redzoning is enabled and
579 * objsize == inuse.
580 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700581 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
582 * 0xcc (RED_ACTIVE) for objects in use.
583 *
584 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700585 * Meta data starts here.
586 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700587 * A. Free pointer (if we cannot overwrite object on free)
588 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700589 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800590 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700591 * before the word boundary.
592 *
593 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700594 *
595 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700596 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700597 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700598 * If slabcaches are merged then the objsize and inuse boundaries are mostly
599 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700600 * may be used with merged slabcaches.
601 */
602
Christoph Lameter81819f02007-05-06 14:49:36 -0700603static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
604{
605 unsigned long off = s->inuse; /* The end of info */
606
607 if (s->offset)
608 /* Freepointer is placed after the object. */
609 off += sizeof(void *);
610
611 if (s->flags & SLAB_STORE_USER)
612 /* We also have user information there */
613 off += 2 * sizeof(struct track);
614
615 if (s->size == off)
616 return 1;
617
Christoph Lameter24922682007-07-17 04:03:18 -0700618 return check_bytes_and_report(s, page, p, "Object padding",
619 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700620}
621
Christoph Lameter39b26462008-04-14 19:11:30 +0300622/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700623static int slab_pad_check(struct kmem_cache *s, struct page *page)
624{
Christoph Lameter24922682007-07-17 04:03:18 -0700625 u8 *start;
626 u8 *fault;
627 u8 *end;
628 int length;
629 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700630
631 if (!(s->flags & SLAB_POISON))
632 return 1;
633
Christoph Lametera973e9d2008-03-01 13:40:44 -0800634 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300635 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300636 end = start + length;
637 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700638 if (!remainder)
639 return 1;
640
Christoph Lameter39b26462008-04-14 19:11:30 +0300641 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700642 if (!fault)
643 return 1;
644 while (end > fault && end[-1] == POISON_INUSE)
645 end--;
646
647 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300648 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700649
650 restore_bytes(s, "slab padding", POISON_INUSE, start, end);
651 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700652}
653
654static int check_object(struct kmem_cache *s, struct page *page,
655 void *object, int active)
656{
657 u8 *p = object;
658 u8 *endobject = object + s->objsize;
659
660 if (s->flags & SLAB_RED_ZONE) {
661 unsigned int red =
662 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
663
Christoph Lameter24922682007-07-17 04:03:18 -0700664 if (!check_bytes_and_report(s, page, object, "Redzone",
665 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700666 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700667 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800668 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
669 check_bytes_and_report(s, page, p, "Alignment padding",
670 endobject, POISON_INUSE, s->inuse - s->objsize);
671 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700672 }
673
674 if (s->flags & SLAB_POISON) {
675 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700676 (!check_bytes_and_report(s, page, p, "Poison", p,
677 POISON_FREE, s->objsize - 1) ||
678 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800679 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700680 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700681 /*
682 * check_pad_bytes cleans up on its own.
683 */
684 check_pad_bytes(s, page, p);
685 }
686
687 if (!s->offset && active)
688 /*
689 * Object and freepointer overlap. Cannot check
690 * freepointer while object is allocated.
691 */
692 return 1;
693
694 /* Check free pointer validity */
695 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
696 object_err(s, page, p, "Freepointer corrupt");
697 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100698 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700699 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700700 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700701 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800702 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700703 return 0;
704 }
705 return 1;
706}
707
708static int check_slab(struct kmem_cache *s, struct page *page)
709{
Christoph Lameter39b26462008-04-14 19:11:30 +0300710 int maxobj;
711
Christoph Lameter81819f02007-05-06 14:49:36 -0700712 VM_BUG_ON(!irqs_disabled());
713
714 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700715 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700716 return 0;
717 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300718
719 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
720 if (page->objects > maxobj) {
721 slab_err(s, page, "objects %u > max %u",
722 s->name, page->objects, maxobj);
723 return 0;
724 }
725 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700726 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300727 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700728 return 0;
729 }
730 /* Slab_pad_check fixes things up after itself */
731 slab_pad_check(s, page);
732 return 1;
733}
734
735/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700736 * Determine if a certain object on a page is on the freelist. Must hold the
737 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700738 */
739static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
740{
741 int nr = 0;
742 void *fp = page->freelist;
743 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300744 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700745
Christoph Lameter39b26462008-04-14 19:11:30 +0300746 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700747 if (fp == search)
748 return 1;
749 if (!check_valid_pointer(s, page, fp)) {
750 if (object) {
751 object_err(s, page, object,
752 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800753 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700754 break;
755 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700756 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800757 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300758 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700759 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700760 return 0;
761 }
762 break;
763 }
764 object = fp;
765 fp = get_freepointer(s, object);
766 nr++;
767 }
768
Christoph Lameter224a88b2008-04-14 19:11:31 +0300769 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400770 if (max_objects > MAX_OBJS_PER_PAGE)
771 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300772
773 if (page->objects != max_objects) {
774 slab_err(s, page, "Wrong number of objects. Found %d but "
775 "should be %d", page->objects, max_objects);
776 page->objects = max_objects;
777 slab_fix(s, "Number of objects adjusted.");
778 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300779 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700780 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300781 "counted were %d", page->inuse, page->objects - nr);
782 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700783 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700784 }
785 return search == NULL;
786}
787
Christoph Lameter0121c6192008-04-29 16:11:12 -0700788static void trace(struct kmem_cache *s, struct page *page, void *object,
789 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700790{
791 if (s->flags & SLAB_TRACE) {
792 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
793 s->name,
794 alloc ? "alloc" : "free",
795 object, page->inuse,
796 page->freelist);
797
798 if (!alloc)
799 print_section("Object", (void *)object, s->objsize);
800
801 dump_stack();
802 }
803}
804
Christoph Lameter643b1132007-05-06 14:49:42 -0700805/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700806 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700807 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700808static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700809{
Christoph Lameter643b1132007-05-06 14:49:42 -0700810 spin_lock(&n->list_lock);
811 list_add(&page->lru, &n->full);
812 spin_unlock(&n->list_lock);
813}
814
815static void remove_full(struct kmem_cache *s, struct page *page)
816{
817 struct kmem_cache_node *n;
818
819 if (!(s->flags & SLAB_STORE_USER))
820 return;
821
822 n = get_node(s, page_to_nid(page));
823
824 spin_lock(&n->list_lock);
825 list_del(&page->lru);
826 spin_unlock(&n->list_lock);
827}
828
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300829/* Tracking of the number of slabs for debugging purposes */
830static inline unsigned long slabs_node(struct kmem_cache *s, int node)
831{
832 struct kmem_cache_node *n = get_node(s, node);
833
834 return atomic_long_read(&n->nr_slabs);
835}
836
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400837static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
838{
839 return atomic_long_read(&n->nr_slabs);
840}
841
Christoph Lameter205ab992008-04-14 19:11:40 +0300842static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300843{
844 struct kmem_cache_node *n = get_node(s, node);
845
846 /*
847 * May be called early in order to allocate a slab for the
848 * kmem_cache_node structure. Solve the chicken-egg
849 * dilemma by deferring the increment of the count during
850 * bootstrap (see early_kmem_cache_node_alloc).
851 */
Christoph Lameter205ab992008-04-14 19:11:40 +0300852 if (!NUMA_BUILD || n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300853 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300854 atomic_long_add(objects, &n->total_objects);
855 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300856}
Christoph Lameter205ab992008-04-14 19:11:40 +0300857static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300858{
859 struct kmem_cache_node *n = get_node(s, node);
860
861 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300862 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300863}
864
865/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700866static void setup_object_debug(struct kmem_cache *s, struct page *page,
867 void *object)
868{
869 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
870 return;
871
872 init_object(s, object, 0);
873 init_tracking(s, object);
874}
875
876static int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300877 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700878{
879 if (!check_slab(s, page))
880 goto bad;
881
Christoph Lameterd692ef62008-02-15 23:45:24 -0800882 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700883 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700884 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700885 }
886
887 if (!check_valid_pointer(s, page, object)) {
888 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700889 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700890 }
891
Christoph Lameterd692ef62008-02-15 23:45:24 -0800892 if (!check_object(s, page, object, 0))
Christoph Lameter81819f02007-05-06 14:49:36 -0700893 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700894
Christoph Lameter3ec09742007-05-16 22:11:00 -0700895 /* Success perform special debug activities for allocs */
896 if (s->flags & SLAB_STORE_USER)
897 set_track(s, object, TRACK_ALLOC, addr);
898 trace(s, page, object, 1);
899 init_object(s, object, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700900 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700901
Christoph Lameter81819f02007-05-06 14:49:36 -0700902bad:
903 if (PageSlab(page)) {
904 /*
905 * If this is a slab page then lets do the best we can
906 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700907 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700908 */
Christoph Lameter24922682007-07-17 04:03:18 -0700909 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300910 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800911 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700912 }
913 return 0;
914}
915
Christoph Lameter3ec09742007-05-16 22:11:00 -0700916static int free_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300917 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700918{
919 if (!check_slab(s, page))
920 goto fail;
921
922 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700923 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700924 goto fail;
925 }
926
927 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700928 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700929 goto fail;
930 }
931
932 if (!check_object(s, page, object, 1))
933 return 0;
934
935 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800936 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700937 slab_err(s, page, "Attempt to free object(0x%p) "
938 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800939 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700940 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700941 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700942 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700943 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800944 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700945 object_err(s, page, object,
946 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700947 goto fail;
948 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700949
950 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700951 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700952 remove_full(s, page);
953 if (s->flags & SLAB_STORE_USER)
954 set_track(s, object, TRACK_FREE, addr);
955 trace(s, page, object, 0);
956 init_object(s, object, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -0700957 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700958
Christoph Lameter81819f02007-05-06 14:49:36 -0700959fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700960 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700961 return 0;
962}
963
Christoph Lameter41ecc552007-05-09 02:32:44 -0700964static int __init setup_slub_debug(char *str)
965{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700966 slub_debug = DEBUG_DEFAULT_FLAGS;
967 if (*str++ != '=' || !*str)
968 /*
969 * No options specified. Switch on full debugging.
970 */
971 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700972
973 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700974 /*
975 * No options but restriction on slabs. This means full
976 * debugging for slabs matching a pattern.
977 */
978 goto check_slabs;
979
980 slub_debug = 0;
981 if (*str == '-')
982 /*
983 * Switch off all debugging measures.
984 */
985 goto out;
986
987 /*
988 * Determine which debug features should be switched on
989 */
Pekka Enberg06428782008-01-07 23:20:27 -0800990 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700991 switch (tolower(*str)) {
992 case 'f':
993 slub_debug |= SLAB_DEBUG_FREE;
994 break;
995 case 'z':
996 slub_debug |= SLAB_RED_ZONE;
997 break;
998 case 'p':
999 slub_debug |= SLAB_POISON;
1000 break;
1001 case 'u':
1002 slub_debug |= SLAB_STORE_USER;
1003 break;
1004 case 't':
1005 slub_debug |= SLAB_TRACE;
1006 break;
1007 default:
1008 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001009 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001010 }
1011 }
1012
1013check_slabs:
1014 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001015 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001016out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001017 return 1;
1018}
1019
1020__setup("slub_debug", setup_slub_debug);
1021
Christoph Lameterba0268a2007-09-11 15:24:11 -07001022static unsigned long kmem_cache_flags(unsigned long objsize,
1023 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001024 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001025{
1026 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001027 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001028 */
Christoph Lametere1533622008-02-15 23:45:24 -08001029 if (slub_debug && (!slub_debug_slabs ||
1030 strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0))
1031 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001032
1033 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001034}
1035#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001036static inline void setup_object_debug(struct kmem_cache *s,
1037 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001038
Christoph Lameter3ec09742007-05-16 22:11:00 -07001039static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001040 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001041
Christoph Lameter3ec09742007-05-16 22:11:00 -07001042static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001043 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001044
Christoph Lameter41ecc552007-05-09 02:32:44 -07001045static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1046 { return 1; }
1047static inline int check_object(struct kmem_cache *s, struct page *page,
1048 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001049static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001050static inline unsigned long kmem_cache_flags(unsigned long objsize,
1051 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001052 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001053{
1054 return flags;
1055}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001056#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001057
1058static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1059 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001060static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1061 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001062static inline void inc_slabs_node(struct kmem_cache *s, int node,
1063 int objects) {}
1064static inline void dec_slabs_node(struct kmem_cache *s, int node,
1065 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001066#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001067
Christoph Lameter81819f02007-05-06 14:49:36 -07001068/*
1069 * Slab allocation and freeing
1070 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001071static inline struct page *alloc_slab_page(gfp_t flags, int node,
1072 struct kmem_cache_order_objects oo)
1073{
1074 int order = oo_order(oo);
1075
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001076 flags |= __GFP_NOTRACK;
1077
Christoph Lameter65c33762008-04-14 19:11:40 +03001078 if (node == -1)
1079 return alloc_pages(flags, order);
1080 else
1081 return alloc_pages_node(node, flags, order);
1082}
1083
Christoph Lameter81819f02007-05-06 14:49:36 -07001084static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1085{
Pekka Enberg06428782008-01-07 23:20:27 -08001086 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001087 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001088 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001089
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001090 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001091
Pekka Enbergba522702009-06-24 21:59:51 +03001092 /*
1093 * Let the initial higher-order allocation fail under memory pressure
1094 * so we fall-back to the minimum order allocation.
1095 */
1096 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1097
1098 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001099 if (unlikely(!page)) {
1100 oo = s->min;
1101 /*
1102 * Allocation may have failed due to fragmentation.
1103 * Try a lower order alloc if possible
1104 */
1105 page = alloc_slab_page(flags, node, oo);
1106 if (!page)
1107 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001108
Christoph Lameter65c33762008-04-14 19:11:40 +03001109 stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK);
1110 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001111
1112 if (kmemcheck_enabled
1113 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS)))
1114 {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001115 int pages = 1 << oo_order(oo);
1116
1117 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1118
1119 /*
1120 * Objects from caches that have a constructor don't get
1121 * cleared when they're allocated, so we need to do it here.
1122 */
1123 if (s->ctor)
1124 kmemcheck_mark_uninitialized_pages(page, pages);
1125 else
1126 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001127 }
1128
Christoph Lameter834f3d12008-04-14 19:11:31 +03001129 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001130 mod_zone_page_state(page_zone(page),
1131 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1132 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001133 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001134
1135 return page;
1136}
1137
1138static void setup_object(struct kmem_cache *s, struct page *page,
1139 void *object)
1140{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001141 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001142 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001143 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001144}
1145
1146static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1147{
1148 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001149 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001150 void *last;
1151 void *p;
1152
Christoph Lameter6cb06222007-10-16 01:25:41 -07001153 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001154
Christoph Lameter6cb06222007-10-16 01:25:41 -07001155 page = allocate_slab(s,
1156 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001157 if (!page)
1158 goto out;
1159
Christoph Lameter205ab992008-04-14 19:11:40 +03001160 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001161 page->slab = s;
1162 page->flags |= 1 << PG_slab;
1163 if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
1164 SLAB_STORE_USER | SLAB_TRACE))
Andy Whitcroft8a380822008-07-23 21:27:18 -07001165 __SetPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001166
1167 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001168
1169 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001170 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001171
1172 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001173 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001174 setup_object(s, page, last);
1175 set_freepointer(s, last, p);
1176 last = p;
1177 }
1178 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001179 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001180
1181 page->freelist = start;
1182 page->inuse = 0;
1183out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001184 return page;
1185}
1186
1187static void __free_slab(struct kmem_cache *s, struct page *page)
1188{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001189 int order = compound_order(page);
1190 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001191
Andy Whitcroft8a380822008-07-23 21:27:18 -07001192 if (unlikely(SLABDEBUG && PageSlubDebug(page))) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001193 void *p;
1194
1195 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001196 for_each_object(p, s, page_address(page),
1197 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001198 check_object(s, page, p, 0);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001199 __ClearPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001200 }
1201
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001202 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001203
Christoph Lameter81819f02007-05-06 14:49:36 -07001204 mod_zone_page_state(page_zone(page),
1205 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1206 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001207 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001208
Christoph Lameter49bd5222008-04-14 18:52:18 +03001209 __ClearPageSlab(page);
1210 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001211 if (current->reclaim_state)
1212 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001213 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001214}
1215
1216static void rcu_free_slab(struct rcu_head *h)
1217{
1218 struct page *page;
1219
1220 page = container_of((struct list_head *)h, struct page, lru);
1221 __free_slab(page->slab, page);
1222}
1223
1224static void free_slab(struct kmem_cache *s, struct page *page)
1225{
1226 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1227 /*
1228 * RCU free overloads the RCU head over the LRU
1229 */
1230 struct rcu_head *head = (void *)&page->lru;
1231
1232 call_rcu(head, rcu_free_slab);
1233 } else
1234 __free_slab(s, page);
1235}
1236
1237static void discard_slab(struct kmem_cache *s, struct page *page)
1238{
Christoph Lameter205ab992008-04-14 19:11:40 +03001239 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001240 free_slab(s, page);
1241}
1242
1243/*
1244 * Per slab locking using the pagelock
1245 */
1246static __always_inline void slab_lock(struct page *page)
1247{
1248 bit_spin_lock(PG_locked, &page->flags);
1249}
1250
1251static __always_inline void slab_unlock(struct page *page)
1252{
Nick Piggina76d3542008-01-07 23:20:27 -08001253 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001254}
1255
1256static __always_inline int slab_trylock(struct page *page)
1257{
1258 int rc = 1;
1259
1260 rc = bit_spin_trylock(PG_locked, &page->flags);
1261 return rc;
1262}
1263
1264/*
1265 * Management of partially allocated slabs
1266 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001267static void add_partial(struct kmem_cache_node *n,
1268 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001269{
Christoph Lametere95eed52007-05-06 14:49:44 -07001270 spin_lock(&n->list_lock);
1271 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001272 if (tail)
1273 list_add_tail(&page->lru, &n->partial);
1274 else
1275 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001276 spin_unlock(&n->list_lock);
1277}
1278
Christoph Lameter0121c6192008-04-29 16:11:12 -07001279static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001280{
1281 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1282
1283 spin_lock(&n->list_lock);
1284 list_del(&page->lru);
1285 n->nr_partial--;
1286 spin_unlock(&n->list_lock);
1287}
1288
1289/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001290 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001291 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001292 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001293 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001294static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1295 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001296{
1297 if (slab_trylock(page)) {
1298 list_del(&page->lru);
1299 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001300 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001301 return 1;
1302 }
1303 return 0;
1304}
1305
1306/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001307 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001308 */
1309static struct page *get_partial_node(struct kmem_cache_node *n)
1310{
1311 struct page *page;
1312
1313 /*
1314 * Racy check. If we mistakenly see no partial slabs then we
1315 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001316 * partial slab and there is none available then get_partials()
1317 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001318 */
1319 if (!n || !n->nr_partial)
1320 return NULL;
1321
1322 spin_lock(&n->list_lock);
1323 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001324 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001325 goto out;
1326 page = NULL;
1327out:
1328 spin_unlock(&n->list_lock);
1329 return page;
1330}
1331
1332/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001333 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001334 */
1335static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1336{
1337#ifdef CONFIG_NUMA
1338 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001339 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001340 struct zone *zone;
1341 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001342 struct page *page;
1343
1344 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001345 * The defrag ratio allows a configuration of the tradeoffs between
1346 * inter node defragmentation and node local allocations. A lower
1347 * defrag_ratio increases the tendency to do local allocations
1348 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001349 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001350 * If the defrag_ratio is set to 0 then kmalloc() always
1351 * returns node local objects. If the ratio is higher then kmalloc()
1352 * may return off node objects because partial slabs are obtained
1353 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001354 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001355 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001356 * defrag_ratio = 1000) then every (well almost) allocation will
1357 * first attempt to defrag slab caches on other nodes. This means
1358 * scanning over all nodes to look for partial slabs which may be
1359 * expensive if we do it every time we are trying to find a slab
1360 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001361 */
Christoph Lameter98246012008-01-07 23:20:26 -08001362 if (!s->remote_node_defrag_ratio ||
1363 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001364 return NULL;
1365
Mel Gorman0e884602008-04-28 02:12:14 -07001366 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001367 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001368 struct kmem_cache_node *n;
1369
Mel Gorman54a6eb52008-04-28 02:12:16 -07001370 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001371
Mel Gorman54a6eb52008-04-28 02:12:16 -07001372 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001373 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001374 page = get_partial_node(n);
1375 if (page)
1376 return page;
1377 }
1378 }
1379#endif
1380 return NULL;
1381}
1382
1383/*
1384 * Get a partial page, lock it and return it.
1385 */
1386static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1387{
1388 struct page *page;
1389 int searchnode = (node == -1) ? numa_node_id() : node;
1390
1391 page = get_partial_node(get_node(s, searchnode));
1392 if (page || (flags & __GFP_THISNODE))
1393 return page;
1394
1395 return get_any_partial(s, flags);
1396}
1397
1398/*
1399 * Move a page back to the lists.
1400 *
1401 * Must be called with the slab lock held.
1402 *
1403 * On exit the slab lock will have been dropped.
1404 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001405static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001406{
Christoph Lametere95eed52007-05-06 14:49:44 -07001407 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001408 struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id());
Christoph Lametere95eed52007-05-06 14:49:44 -07001409
Andy Whitcroft8a380822008-07-23 21:27:18 -07001410 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001411 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001412
Christoph Lametera973e9d2008-03-01 13:40:44 -08001413 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001414 add_partial(n, page, tail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001415 stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1416 } else {
1417 stat(c, DEACTIVATE_FULL);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001418 if (SLABDEBUG && PageSlubDebug(page) &&
1419 (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001420 add_full(n, page);
1421 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001422 slab_unlock(page);
1423 } else {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001424 stat(c, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001425 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001426 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001427 * Adding an empty slab to the partial slabs in order
1428 * to avoid page allocator overhead. This slab needs
1429 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001430 * so that the others get filled first. That way the
1431 * size of the partial list stays small.
1432 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001433 * kmem_cache_shrink can reclaim any empty slabs from
1434 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001435 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001436 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001437 slab_unlock(page);
1438 } else {
1439 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001440 stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001441 discard_slab(s, page);
1442 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001443 }
1444}
1445
1446/*
1447 * Remove the cpu slab
1448 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001449static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001450{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001451 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001452 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001453
Christoph Lameterb773ad72008-03-04 11:10:17 -08001454 if (page->freelist)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001455 stat(c, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001456 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001457 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001458 * because both freelists are empty. So this is unlikely
1459 * to occur.
1460 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001461 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001462 void **object;
1463
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001464 tail = 0; /* Hot objects. Put the slab first */
1465
Christoph Lameter894b8782007-05-10 03:15:16 -07001466 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001467 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001468 c->freelist = c->freelist[c->offset];
Christoph Lameter894b8782007-05-10 03:15:16 -07001469
1470 /* And put onto the regular freelist */
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001471 object[c->offset] = page->freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07001472 page->freelist = object;
1473 page->inuse--;
1474 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001475 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001476 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001477}
1478
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001479static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001480{
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001481 stat(c, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001482 slab_lock(c->page);
1483 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001484}
1485
1486/*
1487 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001488 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001489 * Called from IPI handler with interrupts disabled.
1490 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001491static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001492{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001493 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001494
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001495 if (likely(c && c->page))
1496 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001497}
1498
1499static void flush_cpu_slab(void *d)
1500{
1501 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001502
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001503 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001504}
1505
1506static void flush_all(struct kmem_cache *s)
1507{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001508 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001509}
1510
1511/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001512 * Check if the objects in a per cpu structure fit numa
1513 * locality expectations.
1514 */
1515static inline int node_match(struct kmem_cache_cpu *c, int node)
1516{
1517#ifdef CONFIG_NUMA
1518 if (node != -1 && c->node != node)
1519 return 0;
1520#endif
1521 return 1;
1522}
1523
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001524static int count_free(struct page *page)
1525{
1526 return page->objects - page->inuse;
1527}
1528
1529static unsigned long count_partial(struct kmem_cache_node *n,
1530 int (*get_count)(struct page *))
1531{
1532 unsigned long flags;
1533 unsigned long x = 0;
1534 struct page *page;
1535
1536 spin_lock_irqsave(&n->list_lock, flags);
1537 list_for_each_entry(page, &n->partial, lru)
1538 x += get_count(page);
1539 spin_unlock_irqrestore(&n->list_lock, flags);
1540 return x;
1541}
1542
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001543static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1544{
1545#ifdef CONFIG_SLUB_DEBUG
1546 return atomic_long_read(&n->total_objects);
1547#else
1548 return 0;
1549#endif
1550}
1551
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001552static noinline void
1553slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1554{
1555 int node;
1556
1557 printk(KERN_WARNING
1558 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1559 nid, gfpflags);
1560 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1561 "default order: %d, min order: %d\n", s->name, s->objsize,
1562 s->size, oo_order(s->oo), oo_order(s->min));
1563
1564 for_each_online_node(node) {
1565 struct kmem_cache_node *n = get_node(s, node);
1566 unsigned long nr_slabs;
1567 unsigned long nr_objs;
1568 unsigned long nr_free;
1569
1570 if (!n)
1571 continue;
1572
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001573 nr_free = count_partial(n, count_free);
1574 nr_slabs = node_nr_slabs(n);
1575 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001576
1577 printk(KERN_WARNING
1578 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1579 node, nr_slabs, nr_objs, nr_free);
1580 }
1581}
1582
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001583/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001584 * Slow path. The lockless freelist is empty or we need to perform
1585 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001586 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001587 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001588 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001589 * Processing is still very fast if new objects have been freed to the
1590 * regular freelist. In that case we simply take over the regular freelist
1591 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001592 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001593 * If that is not working then we fall back to the partial lists. We take the
1594 * first element of the freelist as the object to allocate now and move the
1595 * rest of the freelist to the lockless freelist.
1596 *
1597 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001598 * we need to allocate a new slab. This is the slowest path since it involves
1599 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001600 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001601static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1602 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001603{
Christoph Lameter81819f02007-05-06 14:49:36 -07001604 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001605 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001606
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001607 /* We handle __GFP_ZERO in the caller */
1608 gfpflags &= ~__GFP_ZERO;
1609
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001610 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001611 goto new_slab;
1612
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001613 slab_lock(c->page);
1614 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001615 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001616
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001617 stat(c, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001618
Christoph Lameter894b8782007-05-10 03:15:16 -07001619load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001620 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001621 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001622 goto another_slab;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001623 if (unlikely(SLABDEBUG && PageSlubDebug(c->page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001624 goto debug;
1625
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001626 c->freelist = object[c->offset];
Christoph Lameter39b26462008-04-14 19:11:30 +03001627 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001628 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001629 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001630unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001631 slab_unlock(c->page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001632 stat(c, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001633 return object;
1634
1635another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001636 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001637
1638new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001639 new = get_partial(s, gfpflags, node);
1640 if (new) {
1641 c->page = new;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001642 stat(c, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001643 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001644 }
1645
Christoph Lameterb811c202007-10-16 23:25:51 -07001646 if (gfpflags & __GFP_WAIT)
1647 local_irq_enable();
1648
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001649 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001650
1651 if (gfpflags & __GFP_WAIT)
1652 local_irq_disable();
1653
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001654 if (new) {
1655 c = get_cpu_slab(s, smp_processor_id());
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001656 stat(c, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001657 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001658 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001659 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001660 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001661 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001662 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001663 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001664 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1665 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001666 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001667debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001668 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001669 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001670
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001671 c->page->inuse++;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001672 c->page->freelist = object[c->offset];
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001673 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001674 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001675}
1676
1677/*
1678 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1679 * have the fastpath folded into their functions. So no function call
1680 * overhead for requests that can be satisfied on the fastpath.
1681 *
1682 * The fastpath works by first checking if the lockless freelist can be used.
1683 * If not then __slab_alloc is called for slow processing.
1684 *
1685 * Otherwise we can simply pick the next object from the lockless free list.
1686 */
Pekka Enberg06428782008-01-07 23:20:27 -08001687static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001688 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001689{
Christoph Lameter894b8782007-05-10 03:15:16 -07001690 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001691 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001692 unsigned long flags;
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001693 unsigned int objsize;
Christoph Lameter1f842602008-01-07 23:20:30 -08001694
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001695 gfpflags &= gfp_allowed_mask;
Pekka Enberg7e85ee02009-06-12 14:03:06 +03001696
Nick Piggincf40bd12009-01-21 08:12:39 +01001697 lockdep_trace_alloc(gfpflags);
OGAWA Hirofumi89124d72008-11-19 21:23:59 +09001698 might_sleep_if(gfpflags & __GFP_WAIT);
Pekka Enberg3c506ef2008-12-29 11:47:05 +02001699
Akinobu Mita773ff602008-12-23 19:37:01 +09001700 if (should_failslab(s->objsize, gfpflags))
1701 return NULL;
1702
Christoph Lameter894b8782007-05-10 03:15:16 -07001703 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001704 c = get_cpu_slab(s, smp_processor_id());
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001705 objsize = c->objsize;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001706 if (unlikely(!c->freelist || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001707
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001708 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001709
1710 else {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001711 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001712 c->freelist = object[c->offset];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001713 stat(c, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001714 }
1715 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001716
1717 if (unlikely((gfpflags & __GFP_ZERO) && object))
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001718 memset(object, 0, objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001719
Vegard Nossum5a896d92008-04-04 00:54:48 +02001720 kmemcheck_slab_alloc(s, gfpflags, object, c->objsize);
Catalin Marinas06f22f12009-06-11 13:23:18 +01001721 kmemleak_alloc_recursive(object, objsize, 1, s->flags, gfpflags);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001722
Christoph Lameter894b8782007-05-10 03:15:16 -07001723 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001724}
1725
1726void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1727{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001728 void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_);
1729
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02001730 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001731
1732 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001733}
1734EXPORT_SYMBOL(kmem_cache_alloc);
1735
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001736#ifdef CONFIG_KMEMTRACE
1737void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1738{
1739 return slab_alloc(s, gfpflags, -1, _RET_IP_);
1740}
1741EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1742#endif
1743
Christoph Lameter81819f02007-05-06 14:49:36 -07001744#ifdef CONFIG_NUMA
1745void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1746{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001747 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1748
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02001749 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1750 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001751
1752 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001753}
1754EXPORT_SYMBOL(kmem_cache_alloc_node);
1755#endif
1756
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001757#ifdef CONFIG_KMEMTRACE
1758void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1759 gfp_t gfpflags,
1760 int node)
1761{
1762 return slab_alloc(s, gfpflags, node, _RET_IP_);
1763}
1764EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1765#endif
1766
Christoph Lameter81819f02007-05-06 14:49:36 -07001767/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001768 * Slow patch handling. This may still be called frequently since objects
1769 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001770 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001771 * So we still attempt to reduce cache line usage. Just take the slab
1772 * lock and free the item. If there is no additional partial page
1773 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001774 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001775static void __slab_free(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001776 void *x, unsigned long addr, unsigned int offset)
Christoph Lameter81819f02007-05-06 14:49:36 -07001777{
1778 void *prior;
1779 void **object = (void *)x;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001780 struct kmem_cache_cpu *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07001781
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001782 c = get_cpu_slab(s, raw_smp_processor_id());
1783 stat(c, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001784 slab_lock(page);
1785
Andy Whitcroft8a380822008-07-23 21:27:18 -07001786 if (unlikely(SLABDEBUG && PageSlubDebug(page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001787 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001788
Christoph Lameter81819f02007-05-06 14:49:36 -07001789checks_ok:
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001790 prior = object[offset] = page->freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001791 page->freelist = object;
1792 page->inuse--;
1793
Andy Whitcroft8a380822008-07-23 21:27:18 -07001794 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001795 stat(c, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001796 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001797 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001798
1799 if (unlikely(!page->inuse))
1800 goto slab_empty;
1801
1802 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001803 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001804 * then add it.
1805 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001806 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001807 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001808 stat(c, FREE_ADD_PARTIAL);
1809 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001810
1811out_unlock:
1812 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001813 return;
1814
1815slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001816 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001817 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001818 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001819 */
1820 remove_partial(s, page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001821 stat(c, FREE_REMOVE_PARTIAL);
1822 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001823 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001824 stat(c, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001825 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001826 return;
1827
1828debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001829 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001830 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001831 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001832}
1833
Christoph Lameter894b8782007-05-10 03:15:16 -07001834/*
1835 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1836 * can perform fastpath freeing without additional function calls.
1837 *
1838 * The fastpath is only possible if we are freeing to the current cpu slab
1839 * of this processor. This typically the case if we have just allocated
1840 * the item before.
1841 *
1842 * If fastpath is not possible then fall back to __slab_free where we deal
1843 * with all sorts of special processing.
1844 */
Pekka Enberg06428782008-01-07 23:20:27 -08001845static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001846 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001847{
1848 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001849 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001850 unsigned long flags;
1851
Catalin Marinas06f22f12009-06-11 13:23:18 +01001852 kmemleak_free_recursive(x, s->flags);
Christoph Lameter894b8782007-05-10 03:15:16 -07001853 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001854 c = get_cpu_slab(s, smp_processor_id());
Vegard Nossum5a896d92008-04-04 00:54:48 +02001855 kmemcheck_slab_free(s, object, c->objsize);
Christoph Lameter27d9e4e2008-02-15 23:45:25 -08001856 debug_check_no_locks_freed(object, c->objsize);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001857 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Pekka Enberg6047a002009-01-14 12:22:25 +02001858 debug_check_no_obj_freed(object, c->objsize);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001859 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001860 object[c->offset] = c->freelist;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001861 c->freelist = object;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001862 stat(c, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001863 } else
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001864 __slab_free(s, page, x, addr, c->offset);
Christoph Lameter894b8782007-05-10 03:15:16 -07001865
1866 local_irq_restore(flags);
1867}
1868
Christoph Lameter81819f02007-05-06 14:49:36 -07001869void kmem_cache_free(struct kmem_cache *s, void *x)
1870{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001871 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001872
Christoph Lameterb49af682007-05-06 14:49:41 -07001873 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001874
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001875 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001876
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02001877 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001878}
1879EXPORT_SYMBOL(kmem_cache_free);
1880
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001881/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001882static struct page *get_object_page(const void *x)
1883{
Christoph Lameterb49af682007-05-06 14:49:41 -07001884 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001885
1886 if (!PageSlab(page))
1887 return NULL;
1888
1889 return page;
1890}
1891
1892/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001893 * Object placement in a slab is made very easy because we always start at
1894 * offset 0. If we tune the size of the object to the alignment then we can
1895 * get the required alignment by putting one properly sized object after
1896 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001897 *
1898 * Notice that the allocation order determines the sizes of the per cpu
1899 * caches. Each processor has always one slab available for allocations.
1900 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001901 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001902 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001903 */
1904
1905/*
1906 * Mininum / Maximum order of slab pages. This influences locking overhead
1907 * and slab fragmentation. A higher order reduces the number of partial slabs
1908 * and increases the number of allocations possible without having to
1909 * take the list_lock.
1910 */
1911static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001912static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001913static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001914
1915/*
1916 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001917 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001918 */
1919static int slub_nomerge;
1920
1921/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001922 * Calculate the order of allocation given an slab object size.
1923 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001924 * The order of allocation has significant impact on performance and other
1925 * system components. Generally order 0 allocations should be preferred since
1926 * order 0 does not cause fragmentation in the page allocator. Larger objects
1927 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001928 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001929 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001930 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001931 * In order to reach satisfactory performance we must ensure that a minimum
1932 * number of objects is in one slab. Otherwise we may generate too much
1933 * activity on the partial lists which requires taking the list_lock. This is
1934 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001935 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001936 * slub_max_order specifies the order where we begin to stop considering the
1937 * number of objects in a slab as critical. If we reach slub_max_order then
1938 * we try to keep the page order as low as possible. So we accept more waste
1939 * of space in favor of a small page order.
1940 *
1941 * Higher order allocations also allow the placement of more objects in a
1942 * slab and thereby reduce object handling overhead. If the user has
1943 * requested a higher mininum order then we start with that one instead of
1944 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001945 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001946static inline int slab_order(int size, int min_objects,
1947 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001948{
1949 int order;
1950 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001951 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001952
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001953 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1954 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001955
Christoph Lameter6300ea72007-07-17 04:03:20 -07001956 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001957 fls(min_objects * size - 1) - PAGE_SHIFT);
1958 order <= max_order; order++) {
1959
Christoph Lameter81819f02007-05-06 14:49:36 -07001960 unsigned long slab_size = PAGE_SIZE << order;
1961
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001962 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001963 continue;
1964
Christoph Lameter81819f02007-05-06 14:49:36 -07001965 rem = slab_size % size;
1966
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001967 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001968 break;
1969
1970 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001971
Christoph Lameter81819f02007-05-06 14:49:36 -07001972 return order;
1973}
1974
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001975static inline int calculate_order(int size)
1976{
1977 int order;
1978 int min_objects;
1979 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001980 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001981
1982 /*
1983 * Attempt to find best configuration for a slab. This
1984 * works by first attempting to generate a layout with
1985 * the best configuration and backing off gradually.
1986 *
1987 * First we reduce the acceptable waste in a slab. Then
1988 * we reduce the minimum objects required in a slab.
1989 */
1990 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001991 if (!min_objects)
1992 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001993 max_objects = (PAGE_SIZE << slub_max_order)/size;
1994 min_objects = min(min_objects, max_objects);
1995
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001996 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001997 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001998 while (fraction >= 4) {
1999 order = slab_order(size, min_objects,
2000 slub_max_order, fraction);
2001 if (order <= slub_max_order)
2002 return order;
2003 fraction /= 2;
2004 }
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002005 min_objects --;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002006 }
2007
2008 /*
2009 * We were unable to place multiple objects in a slab. Now
2010 * lets see if we can place a single object there.
2011 */
2012 order = slab_order(size, 1, slub_max_order, 1);
2013 if (order <= slub_max_order)
2014 return order;
2015
2016 /*
2017 * Doh this slab cannot be placed using slub_max_order.
2018 */
2019 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002020 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002021 return order;
2022 return -ENOSYS;
2023}
2024
Christoph Lameter81819f02007-05-06 14:49:36 -07002025/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002026 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002027 */
2028static unsigned long calculate_alignment(unsigned long flags,
2029 unsigned long align, unsigned long size)
2030{
2031 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002032 * If the user wants hardware cache aligned objects then follow that
2033 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002034 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002035 * The hardware cache alignment cannot override the specified
2036 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002037 */
Nick Pigginb6210382008-03-05 14:05:56 -08002038 if (flags & SLAB_HWCACHE_ALIGN) {
2039 unsigned long ralign = cache_line_size();
2040 while (size <= ralign / 2)
2041 ralign /= 2;
2042 align = max(align, ralign);
2043 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002044
2045 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002046 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002047
2048 return ALIGN(align, sizeof(void *));
2049}
2050
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002051static void init_kmem_cache_cpu(struct kmem_cache *s,
2052 struct kmem_cache_cpu *c)
2053{
2054 c->page = NULL;
Christoph Lametera973e9d2008-03-01 13:40:44 -08002055 c->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002056 c->node = 0;
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07002057 c->offset = s->offset / sizeof(void *);
2058 c->objsize = s->objsize;
Pekka Enberg62f75532008-04-14 18:50:44 +03002059#ifdef CONFIG_SLUB_STATS
2060 memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned));
2061#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002062}
2063
Pekka Enberg5595cff2008-08-05 09:28:47 +03002064static void
2065init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002066{
2067 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002068 spin_lock_init(&n->list_lock);
2069 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002070#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002071 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002072 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002073 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002074#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002075}
2076
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002077#ifdef CONFIG_SMP
2078/*
2079 * Per cpu array for per cpu structures.
2080 *
2081 * The per cpu array places all kmem_cache_cpu structures from one processor
2082 * close together meaning that it becomes possible that multiple per cpu
2083 * structures are contained in one cacheline. This may be particularly
2084 * beneficial for the kmalloc caches.
2085 *
2086 * A desktop system typically has around 60-80 slabs. With 100 here we are
2087 * likely able to get per cpu structures for all caches from the array defined
2088 * here. We must be able to cover all kmalloc caches during bootstrap.
2089 *
2090 * If the per cpu array is exhausted then fall back to kmalloc
2091 * of individual cachelines. No sharing is possible then.
2092 */
2093#define NR_KMEM_CACHE_CPU 100
2094
2095static DEFINE_PER_CPU(struct kmem_cache_cpu,
2096 kmem_cache_cpu)[NR_KMEM_CACHE_CPU];
2097
2098static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free);
Rusty Russell174596a2009-01-01 10:12:29 +10302099static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002100
2101static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s,
2102 int cpu, gfp_t flags)
2103{
2104 struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu);
2105
2106 if (c)
2107 per_cpu(kmem_cache_cpu_free, cpu) =
2108 (void *)c->freelist;
2109 else {
2110 /* Table overflow: So allocate ourselves */
2111 c = kmalloc_node(
2112 ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()),
2113 flags, cpu_to_node(cpu));
2114 if (!c)
2115 return NULL;
2116 }
2117
2118 init_kmem_cache_cpu(s, c);
2119 return c;
2120}
2121
2122static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu)
2123{
2124 if (c < per_cpu(kmem_cache_cpu, cpu) ||
David Rientjes37189092009-01-27 18:59:46 -08002125 c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002126 kfree(c);
2127 return;
2128 }
2129 c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu);
2130 per_cpu(kmem_cache_cpu_free, cpu) = c;
2131}
2132
2133static void free_kmem_cache_cpus(struct kmem_cache *s)
2134{
2135 int cpu;
2136
2137 for_each_online_cpu(cpu) {
2138 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2139
2140 if (c) {
2141 s->cpu_slab[cpu] = NULL;
2142 free_kmem_cache_cpu(c, cpu);
2143 }
2144 }
2145}
2146
2147static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2148{
2149 int cpu;
2150
2151 for_each_online_cpu(cpu) {
2152 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2153
2154 if (c)
2155 continue;
2156
2157 c = alloc_kmem_cache_cpu(s, cpu, flags);
2158 if (!c) {
2159 free_kmem_cache_cpus(s);
2160 return 0;
2161 }
2162 s->cpu_slab[cpu] = c;
2163 }
2164 return 1;
2165}
2166
2167/*
2168 * Initialize the per cpu array.
2169 */
2170static void init_alloc_cpu_cpu(int cpu)
2171{
2172 int i;
2173
Rusty Russell174596a2009-01-01 10:12:29 +10302174 if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)))
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002175 return;
2176
2177 for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--)
2178 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu);
2179
Rusty Russell174596a2009-01-01 10:12:29 +10302180 cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once));
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002181}
2182
2183static void __init init_alloc_cpu(void)
2184{
2185 int cpu;
2186
2187 for_each_online_cpu(cpu)
2188 init_alloc_cpu_cpu(cpu);
2189 }
2190
2191#else
2192static inline void free_kmem_cache_cpus(struct kmem_cache *s) {}
2193static inline void init_alloc_cpu(void) {}
2194
2195static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2196{
2197 init_kmem_cache_cpu(s, &s->cpu_slab);
2198 return 1;
2199}
2200#endif
2201
Christoph Lameter81819f02007-05-06 14:49:36 -07002202#ifdef CONFIG_NUMA
2203/*
2204 * No kmalloc_node yet so do it by hand. We know that this is the first
2205 * slab on the node for this slabcache. There are no concurrent accesses
2206 * possible.
2207 *
2208 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002209 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2210 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002211 */
David Rientjes0094de92008-11-25 19:14:19 -08002212static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002213{
2214 struct page *page;
2215 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002216 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002217
2218 BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2219
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002220 page = new_slab(kmalloc_caches, gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002221
2222 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002223 if (page_to_nid(page) != node) {
2224 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2225 "node %d\n", node);
2226 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2227 "in order to be able to continue\n");
2228 }
2229
Christoph Lameter81819f02007-05-06 14:49:36 -07002230 n = page->freelist;
2231 BUG_ON(!n);
2232 page->freelist = get_freepointer(kmalloc_caches, n);
2233 page->inuse++;
2234 kmalloc_caches->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002235#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterd45f39c2007-07-17 04:03:21 -07002236 init_object(kmalloc_caches, n, 1);
2237 init_tracking(kmalloc_caches, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002238#endif
Pekka Enberg5595cff2008-08-05 09:28:47 +03002239 init_kmem_cache_node(n, kmalloc_caches);
Christoph Lameter205ab992008-04-14 19:11:40 +03002240 inc_slabs_node(kmalloc_caches, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002241
rootba84c732008-01-07 23:20:28 -08002242 /*
2243 * lockdep requires consistent irq usage for each lock
2244 * so even though there cannot be a race this early in
2245 * the boot sequence, we still disable irqs.
2246 */
2247 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002248 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002249 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002250}
2251
2252static void free_kmem_cache_nodes(struct kmem_cache *s)
2253{
2254 int node;
2255
Christoph Lameterf64dc582007-10-16 01:25:33 -07002256 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002257 struct kmem_cache_node *n = s->node[node];
2258 if (n && n != &s->local_node)
2259 kmem_cache_free(kmalloc_caches, n);
2260 s->node[node] = NULL;
2261 }
2262}
2263
2264static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2265{
2266 int node;
2267 int local_node;
2268
2269 if (slab_state >= UP)
2270 local_node = page_to_nid(virt_to_page(s));
2271 else
2272 local_node = 0;
2273
Christoph Lameterf64dc582007-10-16 01:25:33 -07002274 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002275 struct kmem_cache_node *n;
2276
2277 if (local_node == node)
2278 n = &s->local_node;
2279 else {
2280 if (slab_state == DOWN) {
David Rientjes0094de92008-11-25 19:14:19 -08002281 early_kmem_cache_node_alloc(gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002282 continue;
2283 }
2284 n = kmem_cache_alloc_node(kmalloc_caches,
2285 gfpflags, node);
2286
2287 if (!n) {
2288 free_kmem_cache_nodes(s);
2289 return 0;
2290 }
2291
2292 }
2293 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002294 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002295 }
2296 return 1;
2297}
2298#else
2299static void free_kmem_cache_nodes(struct kmem_cache *s)
2300{
2301}
2302
2303static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2304{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002305 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002306 return 1;
2307}
2308#endif
2309
David Rientjesc0bdb232009-02-25 09:16:35 +02002310static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002311{
2312 if (min < MIN_PARTIAL)
2313 min = MIN_PARTIAL;
2314 else if (min > MAX_PARTIAL)
2315 min = MAX_PARTIAL;
2316 s->min_partial = min;
2317}
2318
Christoph Lameter81819f02007-05-06 14:49:36 -07002319/*
2320 * calculate_sizes() determines the order and the distribution of data within
2321 * a slab object.
2322 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002323static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002324{
2325 unsigned long flags = s->flags;
2326 unsigned long size = s->objsize;
2327 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002328 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002329
2330 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002331 * Round up object size to the next word boundary. We can only
2332 * place the free pointer at word boundaries and this determines
2333 * the possible location of the free pointer.
2334 */
2335 size = ALIGN(size, sizeof(void *));
2336
2337#ifdef CONFIG_SLUB_DEBUG
2338 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002339 * Determine if we can poison the object itself. If the user of
2340 * the slab may touch the object after free or before allocation
2341 * then we should never poison the object itself.
2342 */
2343 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002344 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002345 s->flags |= __OBJECT_POISON;
2346 else
2347 s->flags &= ~__OBJECT_POISON;
2348
Christoph Lameter81819f02007-05-06 14:49:36 -07002349
2350 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002351 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002352 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002353 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002354 */
2355 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2356 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002357#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002358
2359 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002360 * With that we have determined the number of bytes in actual use
2361 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002362 */
2363 s->inuse = size;
2364
2365 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002366 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002367 /*
2368 * Relocate free pointer after the object if it is not
2369 * permitted to overwrite the first word of the object on
2370 * kmem_cache_free.
2371 *
2372 * This is the case if we do RCU, have a constructor or
2373 * destructor or are poisoning the objects.
2374 */
2375 s->offset = size;
2376 size += sizeof(void *);
2377 }
2378
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002379#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002380 if (flags & SLAB_STORE_USER)
2381 /*
2382 * Need to store information about allocs and frees after
2383 * the object.
2384 */
2385 size += 2 * sizeof(struct track);
2386
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002387 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002388 /*
2389 * Add some empty padding so that we can catch
2390 * overwrites from earlier objects rather than let
2391 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002392 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002393 * of the object.
2394 */
2395 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002396#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002397
Christoph Lameter81819f02007-05-06 14:49:36 -07002398 /*
2399 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002400 * user specified and the dynamic determination of cache line size
2401 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002402 */
2403 align = calculate_alignment(flags, align, s->objsize);
2404
2405 /*
2406 * SLUB stores one object immediately after another beginning from
2407 * offset 0. In order to align the objects we have to simply size
2408 * each object to conform to the alignment.
2409 */
2410 size = ALIGN(size, align);
2411 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002412 if (forced_order >= 0)
2413 order = forced_order;
2414 else
2415 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002416
Christoph Lameter834f3d12008-04-14 19:11:31 +03002417 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002418 return 0;
2419
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002420 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002421 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002422 s->allocflags |= __GFP_COMP;
2423
2424 if (s->flags & SLAB_CACHE_DMA)
2425 s->allocflags |= SLUB_DMA;
2426
2427 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2428 s->allocflags |= __GFP_RECLAIMABLE;
2429
Christoph Lameter81819f02007-05-06 14:49:36 -07002430 /*
2431 * Determine the number of objects per slab
2432 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002433 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002434 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002435 if (oo_objects(s->oo) > oo_objects(s->max))
2436 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002437
Christoph Lameter834f3d12008-04-14 19:11:31 +03002438 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002439
2440}
2441
Christoph Lameter81819f02007-05-06 14:49:36 -07002442static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2443 const char *name, size_t size,
2444 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002445 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002446{
2447 memset(s, 0, kmem_size);
2448 s->name = name;
2449 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002450 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002451 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002452 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002453
Christoph Lameter06b285d2008-04-14 19:11:41 +03002454 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002455 goto error;
2456
David Rientjes3b89d7d2009-02-22 17:40:07 -08002457 /*
2458 * The larger the object size is, the more pages we want on the partial
2459 * list to avoid pounding the page allocator excessively.
2460 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002461 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002462 s->refcount = 1;
2463#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002464 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002465#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002466 if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2467 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002468
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002469 if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
Christoph Lameter81819f02007-05-06 14:49:36 -07002470 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002471 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002472error:
2473 if (flags & SLAB_PANIC)
2474 panic("Cannot create slab %s size=%lu realsize=%u "
2475 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002476 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002477 s->offset, flags);
2478 return 0;
2479}
Christoph Lameter81819f02007-05-06 14:49:36 -07002480
2481/*
2482 * Check if a given pointer is valid
2483 */
2484int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2485{
Pekka Enberg06428782008-01-07 23:20:27 -08002486 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002487
2488 page = get_object_page(object);
2489
2490 if (!page || s != page->slab)
2491 /* No slab or wrong slab */
2492 return 0;
2493
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002494 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002495 return 0;
2496
2497 /*
2498 * We could also check if the object is on the slabs freelist.
2499 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002500 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002501 * to a certain slab.
2502 */
2503 return 1;
2504}
2505EXPORT_SYMBOL(kmem_ptr_validate);
2506
2507/*
2508 * Determine the size of a slab object
2509 */
2510unsigned int kmem_cache_size(struct kmem_cache *s)
2511{
2512 return s->objsize;
2513}
2514EXPORT_SYMBOL(kmem_cache_size);
2515
2516const char *kmem_cache_name(struct kmem_cache *s)
2517{
2518 return s->name;
2519}
2520EXPORT_SYMBOL(kmem_cache_name);
2521
Christoph Lameter33b12c32008-04-25 12:22:43 -07002522static void list_slab_objects(struct kmem_cache *s, struct page *page,
2523 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002524{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002525#ifdef CONFIG_SLUB_DEBUG
2526 void *addr = page_address(page);
2527 void *p;
2528 DECLARE_BITMAP(map, page->objects);
2529
2530 bitmap_zero(map, page->objects);
2531 slab_err(s, page, "%s", text);
2532 slab_lock(page);
2533 for_each_free_object(p, s, page->freelist)
2534 set_bit(slab_index(p, s, addr), map);
2535
2536 for_each_object(p, s, addr, page->objects) {
2537
2538 if (!test_bit(slab_index(p, s, addr), map)) {
2539 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2540 p, p - addr);
2541 print_tracking(s, p);
2542 }
2543 }
2544 slab_unlock(page);
2545#endif
2546}
2547
Christoph Lameter81819f02007-05-06 14:49:36 -07002548/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002549 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002550 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002551static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002552{
Christoph Lameter81819f02007-05-06 14:49:36 -07002553 unsigned long flags;
2554 struct page *page, *h;
2555
2556 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002557 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002558 if (!page->inuse) {
2559 list_del(&page->lru);
2560 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002561 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002562 } else {
2563 list_slab_objects(s, page,
2564 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002565 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002566 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002567 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002568}
2569
2570/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002571 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002572 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002573static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002574{
2575 int node;
2576
2577 flush_all(s);
2578
2579 /* Attempt to free all objects */
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002580 free_kmem_cache_cpus(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002581 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002582 struct kmem_cache_node *n = get_node(s, node);
2583
Christoph Lameter599870b2008-04-23 12:36:52 -07002584 free_partial(s, n);
2585 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002586 return 1;
2587 }
2588 free_kmem_cache_nodes(s);
2589 return 0;
2590}
2591
2592/*
2593 * Close a cache and release the kmem_cache structure
2594 * (must be used for caches created using kmem_cache_create)
2595 */
2596void kmem_cache_destroy(struct kmem_cache *s)
2597{
Paul E. McKenney7ed9f7e2009-06-25 12:31:37 -07002598 if (s->flags & SLAB_DESTROY_BY_RCU)
2599 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002600 down_write(&slub_lock);
2601 s->refcount--;
2602 if (!s->refcount) {
2603 list_del(&s->list);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002604 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03002605 if (kmem_cache_close(s)) {
2606 printk(KERN_ERR "SLUB %s: %s called for cache that "
2607 "still has objects.\n", s->name, __func__);
2608 dump_stack();
2609 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002610 sysfs_slab_remove(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002611 } else
2612 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002613}
2614EXPORT_SYMBOL(kmem_cache_destroy);
2615
2616/********************************************************************
2617 * Kmalloc subsystem
2618 *******************************************************************/
2619
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002620struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned;
Christoph Lameter81819f02007-05-06 14:49:36 -07002621EXPORT_SYMBOL(kmalloc_caches);
2622
Christoph Lameter81819f02007-05-06 14:49:36 -07002623static int __init setup_slub_min_order(char *str)
2624{
Pekka Enberg06428782008-01-07 23:20:27 -08002625 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002626
2627 return 1;
2628}
2629
2630__setup("slub_min_order=", setup_slub_min_order);
2631
2632static int __init setup_slub_max_order(char *str)
2633{
Pekka Enberg06428782008-01-07 23:20:27 -08002634 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002635 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002636
2637 return 1;
2638}
2639
2640__setup("slub_max_order=", setup_slub_max_order);
2641
2642static int __init setup_slub_min_objects(char *str)
2643{
Pekka Enberg06428782008-01-07 23:20:27 -08002644 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002645
2646 return 1;
2647}
2648
2649__setup("slub_min_objects=", setup_slub_min_objects);
2650
2651static int __init setup_slub_nomerge(char *str)
2652{
2653 slub_nomerge = 1;
2654 return 1;
2655}
2656
2657__setup("slub_nomerge", setup_slub_nomerge);
2658
Christoph Lameter81819f02007-05-06 14:49:36 -07002659static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2660 const char *name, int size, gfp_t gfp_flags)
2661{
2662 unsigned int flags = 0;
2663
2664 if (gfp_flags & SLUB_DMA)
2665 flags = SLAB_CACHE_DMA;
2666
Pekka Enberg83b519e2009-06-10 19:40:04 +03002667 /*
2668 * This function is called with IRQs disabled during early-boot on
2669 * single CPU so there's no need to take slub_lock here.
2670 */
Christoph Lameter81819f02007-05-06 14:49:36 -07002671 if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002672 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002673 goto panic;
2674
2675 list_add(&s->list, &slab_caches);
Pekka Enberg83b519e2009-06-10 19:40:04 +03002676
Christoph Lameter81819f02007-05-06 14:49:36 -07002677 if (sysfs_slab_add(s))
2678 goto panic;
2679 return s;
2680
2681panic:
2682 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2683}
2684
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002685#ifdef CONFIG_ZONE_DMA
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002686static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
Christoph Lameter1ceef402007-08-07 15:11:48 -07002687
2688static void sysfs_add_func(struct work_struct *w)
2689{
2690 struct kmem_cache *s;
2691
2692 down_write(&slub_lock);
2693 list_for_each_entry(s, &slab_caches, list) {
2694 if (s->flags & __SYSFS_ADD_DEFERRED) {
2695 s->flags &= ~__SYSFS_ADD_DEFERRED;
2696 sysfs_slab_add(s);
2697 }
2698 }
2699 up_write(&slub_lock);
2700}
2701
2702static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2703
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002704static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2705{
2706 struct kmem_cache *s;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002707 char *text;
2708 size_t realsize;
Nick Piggin964cf352009-06-15 13:35:10 +03002709 unsigned long slabflags;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002710
2711 s = kmalloc_caches_dma[index];
2712 if (s)
2713 return s;
2714
2715 /* Dynamically create dma cache */
Christoph Lameter1ceef402007-08-07 15:11:48 -07002716 if (flags & __GFP_WAIT)
2717 down_write(&slub_lock);
2718 else {
2719 if (!down_write_trylock(&slub_lock))
2720 goto out;
2721 }
2722
2723 if (kmalloc_caches_dma[index])
2724 goto unlock_out;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002725
Christoph Lameter7b55f622007-07-17 04:03:27 -07002726 realsize = kmalloc_caches[index].objsize;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08002727 text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2728 (unsigned int)realsize);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002729 s = kmalloc(kmem_size, flags & ~SLUB_DMA);
2730
Nick Piggin964cf352009-06-15 13:35:10 +03002731 /*
2732 * Must defer sysfs creation to a workqueue because we don't know
2733 * what context we are called from. Before sysfs comes up, we don't
2734 * need to do anything because our sysfs initcall will start by
2735 * adding all existing slabs to sysfs.
2736 */
Pekka Enberg5caf5c72009-06-17 08:30:54 +03002737 slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK;
Nick Piggin964cf352009-06-15 13:35:10 +03002738 if (slab_state >= SYSFS)
2739 slabflags |= __SYSFS_ADD_DEFERRED;
2740
Christoph Lameter1ceef402007-08-07 15:11:48 -07002741 if (!s || !text || !kmem_cache_open(s, flags, text,
Nick Piggin964cf352009-06-15 13:35:10 +03002742 realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) {
Christoph Lameter1ceef402007-08-07 15:11:48 -07002743 kfree(s);
2744 kfree(text);
2745 goto unlock_out;
Christoph Lameterdfce8642007-07-17 04:03:25 -07002746 }
Christoph Lameter1ceef402007-08-07 15:11:48 -07002747
2748 list_add(&s->list, &slab_caches);
2749 kmalloc_caches_dma[index] = s;
2750
Nick Piggin964cf352009-06-15 13:35:10 +03002751 if (slab_state >= SYSFS)
2752 schedule_work(&sysfs_add_work);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002753
2754unlock_out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002755 up_write(&slub_lock);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002756out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002757 return kmalloc_caches_dma[index];
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002758}
2759#endif
2760
Christoph Lameterf1b26332007-07-17 04:03:26 -07002761/*
2762 * Conversion table for small slabs sizes / 8 to the index in the
2763 * kmalloc array. This is necessary for slabs < 192 since we have non power
2764 * of two cache sizes there. The size of larger slabs can be determined using
2765 * fls.
2766 */
2767static s8 size_index[24] = {
2768 3, /* 8 */
2769 4, /* 16 */
2770 5, /* 24 */
2771 5, /* 32 */
2772 6, /* 40 */
2773 6, /* 48 */
2774 6, /* 56 */
2775 6, /* 64 */
2776 1, /* 72 */
2777 1, /* 80 */
2778 1, /* 88 */
2779 1, /* 96 */
2780 7, /* 104 */
2781 7, /* 112 */
2782 7, /* 120 */
2783 7, /* 128 */
2784 2, /* 136 */
2785 2, /* 144 */
2786 2, /* 152 */
2787 2, /* 160 */
2788 2, /* 168 */
2789 2, /* 176 */
2790 2, /* 184 */
2791 2 /* 192 */
2792};
2793
Christoph Lameter81819f02007-05-06 14:49:36 -07002794static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2795{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002796 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002797
Christoph Lameterf1b26332007-07-17 04:03:26 -07002798 if (size <= 192) {
2799 if (!size)
2800 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002801
Christoph Lameterf1b26332007-07-17 04:03:26 -07002802 index = size_index[(size - 1) / 8];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002803 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002804 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002805
2806#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002807 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002808 return dma_kmalloc_cache(index, flags);
Christoph Lameterf1b26332007-07-17 04:03:26 -07002809
Christoph Lameter81819f02007-05-06 14:49:36 -07002810#endif
2811 return &kmalloc_caches[index];
2812}
2813
2814void *__kmalloc(size_t size, gfp_t flags)
2815{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002816 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002817 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002818
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002819 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002820 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002821
2822 s = get_slab(size, flags);
2823
2824 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002825 return s;
2826
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002827 ret = slab_alloc(s, flags, -1, _RET_IP_);
2828
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002829 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002830
2831 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002832}
2833EXPORT_SYMBOL(__kmalloc);
2834
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002835static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2836{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002837 struct page *page;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002838
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002839 flags |= __GFP_COMP | __GFP_NOTRACK;
2840 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002841 if (page)
2842 return page_address(page);
2843 else
2844 return NULL;
2845}
2846
Christoph Lameter81819f02007-05-06 14:49:36 -07002847#ifdef CONFIG_NUMA
2848void *__kmalloc_node(size_t size, gfp_t flags, int node)
2849{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002850 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002851 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002852
Ingo Molnar057685c2009-02-20 12:15:30 +01002853 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002854 ret = kmalloc_large_node(size, flags, node);
2855
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002856 trace_kmalloc_node(_RET_IP_, ret,
2857 size, PAGE_SIZE << get_order(size),
2858 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002859
2860 return ret;
2861 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002862
2863 s = get_slab(size, flags);
2864
2865 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002866 return s;
2867
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002868 ret = slab_alloc(s, flags, node, _RET_IP_);
2869
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002870 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002871
2872 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002873}
2874EXPORT_SYMBOL(__kmalloc_node);
2875#endif
2876
2877size_t ksize(const void *object)
2878{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002879 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002880 struct kmem_cache *s;
2881
Christoph Lameteref8b4522007-10-16 01:24:46 -07002882 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002883 return 0;
2884
Vegard Nossum294a80a2007-12-04 23:45:30 -08002885 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002886
Pekka Enberg76994412008-05-22 19:22:25 +03002887 if (unlikely(!PageSlab(page))) {
2888 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002889 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002890 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002891 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002892
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002893#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002894 /*
2895 * Debugging requires use of the padding between object
2896 * and whatever may come after it.
2897 */
2898 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2899 return s->objsize;
2900
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002901#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002902 /*
2903 * If we have the need to store the freelist pointer
2904 * back there or track user information then we can
2905 * only use the space before that information.
2906 */
2907 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2908 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002909 /*
2910 * Else we can use all the padding etc for the allocation
2911 */
2912 return s->size;
2913}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002914EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002915
2916void kfree(const void *x)
2917{
Christoph Lameter81819f02007-05-06 14:49:36 -07002918 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002919 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002920
Pekka Enberg2121db72009-03-25 11:05:57 +02002921 trace_kfree(_RET_IP_, x);
2922
Satyam Sharma2408c552007-10-16 01:24:44 -07002923 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002924 return;
2925
Christoph Lameterb49af682007-05-06 14:49:41 -07002926 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002927 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002928 BUG_ON(!PageCompound(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002929 put_page(page);
2930 return;
2931 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002932 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002933}
2934EXPORT_SYMBOL(kfree);
2935
Christoph Lameter2086d262007-05-06 14:49:46 -07002936/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002937 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2938 * the remaining slabs by the number of items in use. The slabs with the
2939 * most items in use come first. New allocations will then fill those up
2940 * and thus they can be removed from the partial lists.
2941 *
2942 * The slabs with the least items are placed last. This results in them
2943 * being allocated from last increasing the chance that the last objects
2944 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002945 */
2946int kmem_cache_shrink(struct kmem_cache *s)
2947{
2948 int node;
2949 int i;
2950 struct kmem_cache_node *n;
2951 struct page *page;
2952 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002953 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002954 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002955 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002956 unsigned long flags;
2957
2958 if (!slabs_by_inuse)
2959 return -ENOMEM;
2960
2961 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002962 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002963 n = get_node(s, node);
2964
2965 if (!n->nr_partial)
2966 continue;
2967
Christoph Lameter834f3d12008-04-14 19:11:31 +03002968 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002969 INIT_LIST_HEAD(slabs_by_inuse + i);
2970
2971 spin_lock_irqsave(&n->list_lock, flags);
2972
2973 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002974 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002975 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002976 * Note that concurrent frees may occur while we hold the
2977 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002978 */
2979 list_for_each_entry_safe(page, t, &n->partial, lru) {
2980 if (!page->inuse && slab_trylock(page)) {
2981 /*
2982 * Must hold slab lock here because slab_free
2983 * may have freed the last object and be
2984 * waiting to release the slab.
2985 */
2986 list_del(&page->lru);
2987 n->nr_partial--;
2988 slab_unlock(page);
2989 discard_slab(s, page);
2990 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002991 list_move(&page->lru,
2992 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002993 }
2994 }
2995
Christoph Lameter2086d262007-05-06 14:49:46 -07002996 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002997 * Rebuild the partial list with the slabs filled up most
2998 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002999 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003000 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003001 list_splice(slabs_by_inuse + i, n->partial.prev);
3002
Christoph Lameter2086d262007-05-06 14:49:46 -07003003 spin_unlock_irqrestore(&n->list_lock, flags);
3004 }
3005
3006 kfree(slabs_by_inuse);
3007 return 0;
3008}
3009EXPORT_SYMBOL(kmem_cache_shrink);
3010
Yasunori Gotob9049e22007-10-21 16:41:37 -07003011#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
3012static int slab_mem_going_offline_callback(void *arg)
3013{
3014 struct kmem_cache *s;
3015
3016 down_read(&slub_lock);
3017 list_for_each_entry(s, &slab_caches, list)
3018 kmem_cache_shrink(s);
3019 up_read(&slub_lock);
3020
3021 return 0;
3022}
3023
3024static void slab_mem_offline_callback(void *arg)
3025{
3026 struct kmem_cache_node *n;
3027 struct kmem_cache *s;
3028 struct memory_notify *marg = arg;
3029 int offline_node;
3030
3031 offline_node = marg->status_change_nid;
3032
3033 /*
3034 * If the node still has available memory. we need kmem_cache_node
3035 * for it yet.
3036 */
3037 if (offline_node < 0)
3038 return;
3039
3040 down_read(&slub_lock);
3041 list_for_each_entry(s, &slab_caches, list) {
3042 n = get_node(s, offline_node);
3043 if (n) {
3044 /*
3045 * if n->nr_slabs > 0, slabs still exist on the node
3046 * that is going down. We were unable to free them,
3047 * and offline_pages() function shoudn't call this
3048 * callback. So, we must fail.
3049 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003050 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003051
3052 s->node[offline_node] = NULL;
3053 kmem_cache_free(kmalloc_caches, n);
3054 }
3055 }
3056 up_read(&slub_lock);
3057}
3058
3059static int slab_mem_going_online_callback(void *arg)
3060{
3061 struct kmem_cache_node *n;
3062 struct kmem_cache *s;
3063 struct memory_notify *marg = arg;
3064 int nid = marg->status_change_nid;
3065 int ret = 0;
3066
3067 /*
3068 * If the node's memory is already available, then kmem_cache_node is
3069 * already created. Nothing to do.
3070 */
3071 if (nid < 0)
3072 return 0;
3073
3074 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003075 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003076 * allocate a kmem_cache_node structure in order to bring the node
3077 * online.
3078 */
3079 down_read(&slub_lock);
3080 list_for_each_entry(s, &slab_caches, list) {
3081 /*
3082 * XXX: kmem_cache_alloc_node will fallback to other nodes
3083 * since memory is not yet available from the node that
3084 * is brought up.
3085 */
3086 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
3087 if (!n) {
3088 ret = -ENOMEM;
3089 goto out;
3090 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003091 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003092 s->node[nid] = n;
3093 }
3094out:
3095 up_read(&slub_lock);
3096 return ret;
3097}
3098
3099static int slab_memory_callback(struct notifier_block *self,
3100 unsigned long action, void *arg)
3101{
3102 int ret = 0;
3103
3104 switch (action) {
3105 case MEM_GOING_ONLINE:
3106 ret = slab_mem_going_online_callback(arg);
3107 break;
3108 case MEM_GOING_OFFLINE:
3109 ret = slab_mem_going_offline_callback(arg);
3110 break;
3111 case MEM_OFFLINE:
3112 case MEM_CANCEL_ONLINE:
3113 slab_mem_offline_callback(arg);
3114 break;
3115 case MEM_ONLINE:
3116 case MEM_CANCEL_OFFLINE:
3117 break;
3118 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003119 if (ret)
3120 ret = notifier_from_errno(ret);
3121 else
3122 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003123 return ret;
3124}
3125
3126#endif /* CONFIG_MEMORY_HOTPLUG */
3127
Christoph Lameter81819f02007-05-06 14:49:36 -07003128/********************************************************************
3129 * Basic setup of slabs
3130 *******************************************************************/
3131
3132void __init kmem_cache_init(void)
3133{
3134 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003135 int caches = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003136
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003137 init_alloc_cpu();
3138
Christoph Lameter81819f02007-05-06 14:49:36 -07003139#ifdef CONFIG_NUMA
3140 /*
3141 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003142 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003143 * kmem_cache_open for slab_state == DOWN.
3144 */
3145 create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
Pekka Enberg83b519e2009-06-10 19:40:04 +03003146 sizeof(struct kmem_cache_node), GFP_NOWAIT);
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003147 kmalloc_caches[0].refcount = -1;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003148 caches++;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003149
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003150 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003151#endif
3152
3153 /* Able to allocate the per node structures */
3154 slab_state = PARTIAL;
3155
3156 /* Caches that are not of the two-to-the-power-of size */
Christoph Lameter4b356be2007-06-16 10:16:13 -07003157 if (KMALLOC_MIN_SIZE <= 64) {
3158 create_kmalloc_cache(&kmalloc_caches[1],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003159 "kmalloc-96", 96, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003160 caches++;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003161 create_kmalloc_cache(&kmalloc_caches[2],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003162 "kmalloc-192", 192, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003163 caches++;
3164 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003165
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003166 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003167 create_kmalloc_cache(&kmalloc_caches[i],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003168 "kmalloc", 1 << i, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003169 caches++;
3170 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003171
Christoph Lameterf1b26332007-07-17 04:03:26 -07003172
3173 /*
3174 * Patch up the size_index table if we have strange large alignment
3175 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003176 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003177 *
3178 * Largest permitted alignment is 256 bytes due to the way we
3179 * handle the index determination for the smaller caches.
3180 *
3181 * Make sure that nothing crazy happens if someone starts tinkering
3182 * around with ARCH_KMALLOC_MINALIGN
3183 */
3184 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3185 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3186
Christoph Lameter12ad6842007-07-17 04:03:28 -07003187 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8)
Christoph Lameterf1b26332007-07-17 04:03:26 -07003188 size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW;
3189
Christoph Lameter41d54d32008-07-03 09:14:26 -05003190 if (KMALLOC_MIN_SIZE == 128) {
3191 /*
3192 * The 192 byte sized cache is not used if the alignment
3193 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3194 * instead.
3195 */
3196 for (i = 128 + 8; i <= 192; i += 8)
3197 size_index[(i - 1) / 8] = 8;
3198 }
3199
Christoph Lameter81819f02007-05-06 14:49:36 -07003200 slab_state = UP;
3201
3202 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003203 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++)
Christoph Lameter81819f02007-05-06 14:49:36 -07003204 kmalloc_caches[i]. name =
Pekka Enberg83b519e2009-06-10 19:40:04 +03003205 kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
Christoph Lameter81819f02007-05-06 14:49:36 -07003206
3207#ifdef CONFIG_SMP
3208 register_cpu_notifier(&slab_notifier);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003209 kmem_size = offsetof(struct kmem_cache, cpu_slab) +
3210 nr_cpu_ids * sizeof(struct kmem_cache_cpu *);
3211#else
3212 kmem_size = sizeof(struct kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003213#endif
3214
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003215 printk(KERN_INFO
3216 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003217 " CPUs=%d, Nodes=%d\n",
3218 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003219 slub_min_order, slub_max_order, slub_min_objects,
3220 nr_cpu_ids, nr_node_ids);
3221}
3222
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003223void __init kmem_cache_init_late(void)
3224{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003225}
3226
Christoph Lameter81819f02007-05-06 14:49:36 -07003227/*
3228 * Find a mergeable slab cache
3229 */
3230static int slab_unmergeable(struct kmem_cache *s)
3231{
3232 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3233 return 1;
3234
Christoph Lameterc59def92007-05-16 22:10:50 -07003235 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003236 return 1;
3237
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003238 /*
3239 * We may have set a slab to be unmergeable during bootstrap.
3240 */
3241 if (s->refcount < 0)
3242 return 1;
3243
Christoph Lameter81819f02007-05-06 14:49:36 -07003244 return 0;
3245}
3246
3247static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003248 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003249 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003250{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003251 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003252
3253 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3254 return NULL;
3255
Christoph Lameterc59def92007-05-16 22:10:50 -07003256 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003257 return NULL;
3258
3259 size = ALIGN(size, sizeof(void *));
3260 align = calculate_alignment(flags, align, size);
3261 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003262 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003263
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003264 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003265 if (slab_unmergeable(s))
3266 continue;
3267
3268 if (size > s->size)
3269 continue;
3270
Christoph Lameterba0268a2007-09-11 15:24:11 -07003271 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003272 continue;
3273 /*
3274 * Check if alignment is compatible.
3275 * Courtesy of Adrian Drzewiecki
3276 */
Pekka Enberg06428782008-01-07 23:20:27 -08003277 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003278 continue;
3279
3280 if (s->size - size >= sizeof(void *))
3281 continue;
3282
3283 return s;
3284 }
3285 return NULL;
3286}
3287
3288struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003289 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003290{
3291 struct kmem_cache *s;
3292
3293 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003294 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003295 if (s) {
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003296 int cpu;
3297
Christoph Lameter81819f02007-05-06 14:49:36 -07003298 s->refcount++;
3299 /*
3300 * Adjust the object sizes so that we clear
3301 * the complete object on kzalloc.
3302 */
3303 s->objsize = max(s->objsize, (int)size);
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003304
3305 /*
3306 * And then we need to update the object size in the
3307 * per cpu structures
3308 */
3309 for_each_online_cpu(cpu)
3310 get_cpu_slab(s, cpu)->objsize = s->objsize;
Christoph Lameter6446faa2008-02-15 23:45:26 -08003311
Christoph Lameter81819f02007-05-06 14:49:36 -07003312 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003313 up_write(&slub_lock);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003314
David Rientjes7b8f3b62008-12-17 22:09:46 -08003315 if (sysfs_slab_alias(s, name)) {
3316 down_write(&slub_lock);
3317 s->refcount--;
3318 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003319 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003320 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003321 return s;
3322 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003323
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003324 s = kmalloc(kmem_size, GFP_KERNEL);
3325 if (s) {
3326 if (kmem_cache_open(s, GFP_KERNEL, name,
Christoph Lameterc59def92007-05-16 22:10:50 -07003327 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003328 list_add(&s->list, &slab_caches);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003329 up_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003330 if (sysfs_slab_add(s)) {
3331 down_write(&slub_lock);
3332 list_del(&s->list);
3333 up_write(&slub_lock);
3334 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003335 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003336 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003337 return s;
3338 }
3339 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003340 }
3341 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003342
3343err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003344 if (flags & SLAB_PANIC)
3345 panic("Cannot create slabcache %s\n", name);
3346 else
3347 s = NULL;
3348 return s;
3349}
3350EXPORT_SYMBOL(kmem_cache_create);
3351
Christoph Lameter81819f02007-05-06 14:49:36 -07003352#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003353/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003354 * Use the cpu notifier to insure that the cpu slabs are flushed when
3355 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003356 */
3357static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3358 unsigned long action, void *hcpu)
3359{
3360 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003361 struct kmem_cache *s;
3362 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003363
3364 switch (action) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003365 case CPU_UP_PREPARE:
3366 case CPU_UP_PREPARE_FROZEN:
3367 init_alloc_cpu_cpu(cpu);
3368 down_read(&slub_lock);
3369 list_for_each_entry(s, &slab_caches, list)
3370 s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu,
3371 GFP_KERNEL);
3372 up_read(&slub_lock);
3373 break;
3374
Christoph Lameter81819f02007-05-06 14:49:36 -07003375 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003376 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003377 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003378 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003379 down_read(&slub_lock);
3380 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003381 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
3382
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003383 local_irq_save(flags);
3384 __flush_cpu_slab(s, cpu);
3385 local_irq_restore(flags);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003386 free_kmem_cache_cpu(c, cpu);
3387 s->cpu_slab[cpu] = NULL;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003388 }
3389 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003390 break;
3391 default:
3392 break;
3393 }
3394 return NOTIFY_OK;
3395}
3396
Pekka Enberg06428782008-01-07 23:20:27 -08003397static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003398 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003399};
Christoph Lameter81819f02007-05-06 14:49:36 -07003400
3401#endif
3402
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003403void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003404{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003405 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003406 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003407
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003408 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003409 return kmalloc_large(size, gfpflags);
3410
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003411 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003412
Satyam Sharma2408c552007-10-16 01:24:44 -07003413 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003414 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003415
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003416 ret = slab_alloc(s, gfpflags, -1, caller);
3417
3418 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003419 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003420
3421 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003422}
3423
3424void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003425 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003426{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003427 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003428 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003429
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003430 if (unlikely(size > SLUB_MAX_SIZE))
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003431 return kmalloc_large_node(size, gfpflags, node);
Pekka Enbergeada35e2008-02-11 22:47:46 +02003432
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003433 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003434
Satyam Sharma2408c552007-10-16 01:24:44 -07003435 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003436 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003437
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003438 ret = slab_alloc(s, gfpflags, node, caller);
3439
3440 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003441 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003442
3443 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003444}
3445
Christoph Lameterf6acb632008-04-29 16:16:06 -07003446#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003447static int count_inuse(struct page *page)
3448{
3449 return page->inuse;
3450}
3451
3452static int count_total(struct page *page)
3453{
3454 return page->objects;
3455}
3456
Christoph Lameter434e2452007-07-17 04:03:30 -07003457static int validate_slab(struct kmem_cache *s, struct page *page,
3458 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003459{
3460 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003461 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003462
3463 if (!check_slab(s, page) ||
3464 !on_freelist(s, page, NULL))
3465 return 0;
3466
3467 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003468 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003469
Christoph Lameter7656c722007-05-09 02:32:40 -07003470 for_each_free_object(p, s, page->freelist) {
3471 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003472 if (!check_object(s, page, p, 0))
3473 return 0;
3474 }
3475
Christoph Lameter224a88b2008-04-14 19:11:31 +03003476 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003477 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003478 if (!check_object(s, page, p, 1))
3479 return 0;
3480 return 1;
3481}
3482
Christoph Lameter434e2452007-07-17 04:03:30 -07003483static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3484 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003485{
3486 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003487 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003488 slab_unlock(page);
3489 } else
3490 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3491 s->name, page);
3492
3493 if (s->flags & DEBUG_DEFAULT_FLAGS) {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003494 if (!PageSlubDebug(page))
3495 printk(KERN_ERR "SLUB %s: SlubDebug not set "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003496 "on slab 0x%p\n", s->name, page);
3497 } else {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003498 if (PageSlubDebug(page))
3499 printk(KERN_ERR "SLUB %s: SlubDebug set on "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003500 "slab 0x%p\n", s->name, page);
3501 }
3502}
3503
Christoph Lameter434e2452007-07-17 04:03:30 -07003504static int validate_slab_node(struct kmem_cache *s,
3505 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003506{
3507 unsigned long count = 0;
3508 struct page *page;
3509 unsigned long flags;
3510
3511 spin_lock_irqsave(&n->list_lock, flags);
3512
3513 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003514 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003515 count++;
3516 }
3517 if (count != n->nr_partial)
3518 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3519 "counter=%ld\n", s->name, count, n->nr_partial);
3520
3521 if (!(s->flags & SLAB_STORE_USER))
3522 goto out;
3523
3524 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003525 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003526 count++;
3527 }
3528 if (count != atomic_long_read(&n->nr_slabs))
3529 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3530 "counter=%ld\n", s->name, count,
3531 atomic_long_read(&n->nr_slabs));
3532
3533out:
3534 spin_unlock_irqrestore(&n->list_lock, flags);
3535 return count;
3536}
3537
Christoph Lameter434e2452007-07-17 04:03:30 -07003538static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003539{
3540 int node;
3541 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003542 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003543 sizeof(unsigned long), GFP_KERNEL);
3544
3545 if (!map)
3546 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003547
3548 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003549 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003550 struct kmem_cache_node *n = get_node(s, node);
3551
Christoph Lameter434e2452007-07-17 04:03:30 -07003552 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003553 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003554 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003555 return count;
3556}
3557
Christoph Lameterb3459702007-05-09 02:32:41 -07003558#ifdef SLUB_RESILIENCY_TEST
3559static void resiliency_test(void)
3560{
3561 u8 *p;
3562
3563 printk(KERN_ERR "SLUB resiliency testing\n");
3564 printk(KERN_ERR "-----------------------\n");
3565 printk(KERN_ERR "A. Corruption after allocation\n");
3566
3567 p = kzalloc(16, GFP_KERNEL);
3568 p[16] = 0x12;
3569 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3570 " 0x12->0x%p\n\n", p + 16);
3571
3572 validate_slab_cache(kmalloc_caches + 4);
3573
3574 /* Hmmm... The next two are dangerous */
3575 p = kzalloc(32, GFP_KERNEL);
3576 p[32 + sizeof(void *)] = 0x34;
3577 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003578 " 0x34 -> -0x%p\n", p);
3579 printk(KERN_ERR
3580 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003581
3582 validate_slab_cache(kmalloc_caches + 5);
3583 p = kzalloc(64, GFP_KERNEL);
3584 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3585 *p = 0x56;
3586 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3587 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003588 printk(KERN_ERR
3589 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003590 validate_slab_cache(kmalloc_caches + 6);
3591
3592 printk(KERN_ERR "\nB. Corruption after free\n");
3593 p = kzalloc(128, GFP_KERNEL);
3594 kfree(p);
3595 *p = 0x78;
3596 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3597 validate_slab_cache(kmalloc_caches + 7);
3598
3599 p = kzalloc(256, GFP_KERNEL);
3600 kfree(p);
3601 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003602 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3603 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003604 validate_slab_cache(kmalloc_caches + 8);
3605
3606 p = kzalloc(512, GFP_KERNEL);
3607 kfree(p);
3608 p[512] = 0xab;
3609 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3610 validate_slab_cache(kmalloc_caches + 9);
3611}
3612#else
3613static void resiliency_test(void) {};
3614#endif
3615
Christoph Lameter88a420e2007-05-06 14:49:45 -07003616/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003617 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003618 * and freed.
3619 */
3620
3621struct location {
3622 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003623 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003624 long long sum_time;
3625 long min_time;
3626 long max_time;
3627 long min_pid;
3628 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303629 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003630 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003631};
3632
3633struct loc_track {
3634 unsigned long max;
3635 unsigned long count;
3636 struct location *loc;
3637};
3638
3639static void free_loc_track(struct loc_track *t)
3640{
3641 if (t->max)
3642 free_pages((unsigned long)t->loc,
3643 get_order(sizeof(struct location) * t->max));
3644}
3645
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003646static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003647{
3648 struct location *l;
3649 int order;
3650
Christoph Lameter88a420e2007-05-06 14:49:45 -07003651 order = get_order(sizeof(struct location) * max);
3652
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003653 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003654 if (!l)
3655 return 0;
3656
3657 if (t->count) {
3658 memcpy(l, t->loc, sizeof(struct location) * t->count);
3659 free_loc_track(t);
3660 }
3661 t->max = max;
3662 t->loc = l;
3663 return 1;
3664}
3665
3666static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003667 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003668{
3669 long start, end, pos;
3670 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003671 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003672 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003673
3674 start = -1;
3675 end = t->count;
3676
3677 for ( ; ; ) {
3678 pos = start + (end - start + 1) / 2;
3679
3680 /*
3681 * There is nothing at "end". If we end up there
3682 * we need to add something to before end.
3683 */
3684 if (pos == end)
3685 break;
3686
3687 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003688 if (track->addr == caddr) {
3689
3690 l = &t->loc[pos];
3691 l->count++;
3692 if (track->when) {
3693 l->sum_time += age;
3694 if (age < l->min_time)
3695 l->min_time = age;
3696 if (age > l->max_time)
3697 l->max_time = age;
3698
3699 if (track->pid < l->min_pid)
3700 l->min_pid = track->pid;
3701 if (track->pid > l->max_pid)
3702 l->max_pid = track->pid;
3703
Rusty Russell174596a2009-01-01 10:12:29 +10303704 cpumask_set_cpu(track->cpu,
3705 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003706 }
3707 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003708 return 1;
3709 }
3710
Christoph Lameter45edfa52007-05-09 02:32:45 -07003711 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003712 end = pos;
3713 else
3714 start = pos;
3715 }
3716
3717 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003718 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003719 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003720 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003721 return 0;
3722
3723 l = t->loc + pos;
3724 if (pos < t->count)
3725 memmove(l + 1, l,
3726 (t->count - pos) * sizeof(struct location));
3727 t->count++;
3728 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003729 l->addr = track->addr;
3730 l->sum_time = age;
3731 l->min_time = age;
3732 l->max_time = age;
3733 l->min_pid = track->pid;
3734 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303735 cpumask_clear(to_cpumask(l->cpus));
3736 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003737 nodes_clear(l->nodes);
3738 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003739 return 1;
3740}
3741
3742static void process_slab(struct loc_track *t, struct kmem_cache *s,
3743 struct page *page, enum track_item alloc)
3744{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003745 void *addr = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +03003746 DECLARE_BITMAP(map, page->objects);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003747 void *p;
3748
Christoph Lameter39b26462008-04-14 19:11:30 +03003749 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003750 for_each_free_object(p, s, page->freelist)
3751 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003752
Christoph Lameter224a88b2008-04-14 19:11:31 +03003753 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003754 if (!test_bit(slab_index(p, s, addr), map))
3755 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003756}
3757
3758static int list_locations(struct kmem_cache *s, char *buf,
3759 enum track_item alloc)
3760{
Harvey Harrisone374d482008-01-31 15:20:50 -08003761 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003762 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003763 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003764 int node;
3765
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003766 if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
Andrew Mortonea3061d2007-10-16 01:26:09 -07003767 GFP_TEMPORARY))
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003768 return sprintf(buf, "Out of memory\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003769
3770 /* Push back cpu slabs */
3771 flush_all(s);
3772
Christoph Lameterf64dc582007-10-16 01:25:33 -07003773 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003774 struct kmem_cache_node *n = get_node(s, node);
3775 unsigned long flags;
3776 struct page *page;
3777
Christoph Lameter9e869432007-08-22 14:01:56 -07003778 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003779 continue;
3780
3781 spin_lock_irqsave(&n->list_lock, flags);
3782 list_for_each_entry(page, &n->partial, lru)
3783 process_slab(&t, s, page, alloc);
3784 list_for_each_entry(page, &n->full, lru)
3785 process_slab(&t, s, page, alloc);
3786 spin_unlock_irqrestore(&n->list_lock, flags);
3787 }
3788
3789 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003790 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003791
Hugh Dickins9c246242008-12-09 13:14:27 -08003792 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003793 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003794 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003795
3796 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003797 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003798 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003799 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003800
3801 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003802 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003803 l->min_time,
3804 (long)div_u64(l->sum_time, l->count),
3805 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003806 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003807 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003808 l->min_time);
3809
3810 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003811 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003812 l->min_pid, l->max_pid);
3813 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003814 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003815 l->min_pid);
3816
Rusty Russell174596a2009-01-01 10:12:29 +10303817 if (num_online_cpus() > 1 &&
3818 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003819 len < PAGE_SIZE - 60) {
3820 len += sprintf(buf + len, " cpus=");
3821 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303822 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003823 }
3824
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003825 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003826 len < PAGE_SIZE - 60) {
3827 len += sprintf(buf + len, " nodes=");
3828 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003829 l->nodes);
3830 }
3831
Harvey Harrisone374d482008-01-31 15:20:50 -08003832 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003833 }
3834
3835 free_loc_track(&t);
3836 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003837 len += sprintf(buf, "No data\n");
3838 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003839}
3840
Christoph Lameter81819f02007-05-06 14:49:36 -07003841enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003842 SL_ALL, /* All slabs */
3843 SL_PARTIAL, /* Only partially allocated slabs */
3844 SL_CPU, /* Only slabs used for cpu caches */
3845 SL_OBJECTS, /* Determine allocated objects not slabs */
3846 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003847};
3848
Christoph Lameter205ab992008-04-14 19:11:40 +03003849#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003850#define SO_PARTIAL (1 << SL_PARTIAL)
3851#define SO_CPU (1 << SL_CPU)
3852#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003853#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003854
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003855static ssize_t show_slab_objects(struct kmem_cache *s,
3856 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003857{
3858 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003859 int node;
3860 int x;
3861 unsigned long *nodes;
3862 unsigned long *per_cpu;
3863
3864 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003865 if (!nodes)
3866 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003867 per_cpu = nodes + nr_node_ids;
3868
Christoph Lameter205ab992008-04-14 19:11:40 +03003869 if (flags & SO_CPU) {
3870 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003871
Christoph Lameter205ab992008-04-14 19:11:40 +03003872 for_each_possible_cpu(cpu) {
3873 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003874
Christoph Lameter205ab992008-04-14 19:11:40 +03003875 if (!c || c->node < 0)
3876 continue;
3877
3878 if (c->page) {
3879 if (flags & SO_TOTAL)
3880 x = c->page->objects;
3881 else if (flags & SO_OBJECTS)
3882 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003883 else
3884 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003885
Christoph Lameter81819f02007-05-06 14:49:36 -07003886 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003887 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003888 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003889 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003890 }
3891 }
3892
Christoph Lameter205ab992008-04-14 19:11:40 +03003893 if (flags & SO_ALL) {
3894 for_each_node_state(node, N_NORMAL_MEMORY) {
3895 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003896
Christoph Lameter205ab992008-04-14 19:11:40 +03003897 if (flags & SO_TOTAL)
3898 x = atomic_long_read(&n->total_objects);
3899 else if (flags & SO_OBJECTS)
3900 x = atomic_long_read(&n->total_objects) -
3901 count_partial(n, count_free);
3902
3903 else
3904 x = atomic_long_read(&n->nr_slabs);
3905 total += x;
3906 nodes[node] += x;
3907 }
3908
3909 } else if (flags & SO_PARTIAL) {
3910 for_each_node_state(node, N_NORMAL_MEMORY) {
3911 struct kmem_cache_node *n = get_node(s, node);
3912
3913 if (flags & SO_TOTAL)
3914 x = count_partial(n, count_total);
3915 else if (flags & SO_OBJECTS)
3916 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003917 else
3918 x = n->nr_partial;
3919 total += x;
3920 nodes[node] += x;
3921 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003922 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003923 x = sprintf(buf, "%lu", total);
3924#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003925 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003926 if (nodes[node])
3927 x += sprintf(buf + x, " N%d=%lu",
3928 node, nodes[node]);
3929#endif
3930 kfree(nodes);
3931 return x + sprintf(buf + x, "\n");
3932}
3933
3934static int any_slab_objects(struct kmem_cache *s)
3935{
3936 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003937
3938 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003939 struct kmem_cache_node *n = get_node(s, node);
3940
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003941 if (!n)
3942 continue;
3943
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003944 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003945 return 1;
3946 }
3947 return 0;
3948}
3949
3950#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3951#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3952
3953struct slab_attribute {
3954 struct attribute attr;
3955 ssize_t (*show)(struct kmem_cache *s, char *buf);
3956 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3957};
3958
3959#define SLAB_ATTR_RO(_name) \
3960 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3961
3962#define SLAB_ATTR(_name) \
3963 static struct slab_attribute _name##_attr = \
3964 __ATTR(_name, 0644, _name##_show, _name##_store)
3965
Christoph Lameter81819f02007-05-06 14:49:36 -07003966static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3967{
3968 return sprintf(buf, "%d\n", s->size);
3969}
3970SLAB_ATTR_RO(slab_size);
3971
3972static ssize_t align_show(struct kmem_cache *s, char *buf)
3973{
3974 return sprintf(buf, "%d\n", s->align);
3975}
3976SLAB_ATTR_RO(align);
3977
3978static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3979{
3980 return sprintf(buf, "%d\n", s->objsize);
3981}
3982SLAB_ATTR_RO(object_size);
3983
3984static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3985{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003986 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003987}
3988SLAB_ATTR_RO(objs_per_slab);
3989
Christoph Lameter06b285d2008-04-14 19:11:41 +03003990static ssize_t order_store(struct kmem_cache *s,
3991 const char *buf, size_t length)
3992{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003993 unsigned long order;
3994 int err;
3995
3996 err = strict_strtoul(buf, 10, &order);
3997 if (err)
3998 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003999
4000 if (order > slub_max_order || order < slub_min_order)
4001 return -EINVAL;
4002
4003 calculate_sizes(s, order);
4004 return length;
4005}
4006
Christoph Lameter81819f02007-05-06 14:49:36 -07004007static ssize_t order_show(struct kmem_cache *s, char *buf)
4008{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004009 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004010}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004011SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004012
David Rientjes73d342b2009-02-22 17:40:09 -08004013static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4014{
4015 return sprintf(buf, "%lu\n", s->min_partial);
4016}
4017
4018static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4019 size_t length)
4020{
4021 unsigned long min;
4022 int err;
4023
4024 err = strict_strtoul(buf, 10, &min);
4025 if (err)
4026 return err;
4027
David Rientjesc0bdb232009-02-25 09:16:35 +02004028 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004029 return length;
4030}
4031SLAB_ATTR(min_partial);
4032
Christoph Lameter81819f02007-05-06 14:49:36 -07004033static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4034{
4035 if (s->ctor) {
4036 int n = sprint_symbol(buf, (unsigned long)s->ctor);
4037
4038 return n + sprintf(buf + n, "\n");
4039 }
4040 return 0;
4041}
4042SLAB_ATTR_RO(ctor);
4043
Christoph Lameter81819f02007-05-06 14:49:36 -07004044static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4045{
4046 return sprintf(buf, "%d\n", s->refcount - 1);
4047}
4048SLAB_ATTR_RO(aliases);
4049
4050static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4051{
Christoph Lameter205ab992008-04-14 19:11:40 +03004052 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004053}
4054SLAB_ATTR_RO(slabs);
4055
4056static ssize_t partial_show(struct kmem_cache *s, char *buf)
4057{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004058 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004059}
4060SLAB_ATTR_RO(partial);
4061
4062static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4063{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004064 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004065}
4066SLAB_ATTR_RO(cpu_slabs);
4067
4068static ssize_t objects_show(struct kmem_cache *s, char *buf)
4069{
Christoph Lameter205ab992008-04-14 19:11:40 +03004070 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004071}
4072SLAB_ATTR_RO(objects);
4073
Christoph Lameter205ab992008-04-14 19:11:40 +03004074static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4075{
4076 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4077}
4078SLAB_ATTR_RO(objects_partial);
4079
4080static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4081{
4082 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4083}
4084SLAB_ATTR_RO(total_objects);
4085
Christoph Lameter81819f02007-05-06 14:49:36 -07004086static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4087{
4088 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4089}
4090
4091static ssize_t sanity_checks_store(struct kmem_cache *s,
4092 const char *buf, size_t length)
4093{
4094 s->flags &= ~SLAB_DEBUG_FREE;
4095 if (buf[0] == '1')
4096 s->flags |= SLAB_DEBUG_FREE;
4097 return length;
4098}
4099SLAB_ATTR(sanity_checks);
4100
4101static ssize_t trace_show(struct kmem_cache *s, char *buf)
4102{
4103 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4104}
4105
4106static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4107 size_t length)
4108{
4109 s->flags &= ~SLAB_TRACE;
4110 if (buf[0] == '1')
4111 s->flags |= SLAB_TRACE;
4112 return length;
4113}
4114SLAB_ATTR(trace);
4115
4116static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4117{
4118 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4119}
4120
4121static ssize_t reclaim_account_store(struct kmem_cache *s,
4122 const char *buf, size_t length)
4123{
4124 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4125 if (buf[0] == '1')
4126 s->flags |= SLAB_RECLAIM_ACCOUNT;
4127 return length;
4128}
4129SLAB_ATTR(reclaim_account);
4130
4131static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4132{
Christoph Lameter5af60832007-05-06 14:49:56 -07004133 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004134}
4135SLAB_ATTR_RO(hwcache_align);
4136
4137#ifdef CONFIG_ZONE_DMA
4138static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4139{
4140 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4141}
4142SLAB_ATTR_RO(cache_dma);
4143#endif
4144
4145static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4146{
4147 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4148}
4149SLAB_ATTR_RO(destroy_by_rcu);
4150
4151static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4152{
4153 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4154}
4155
4156static ssize_t red_zone_store(struct kmem_cache *s,
4157 const char *buf, size_t length)
4158{
4159 if (any_slab_objects(s))
4160 return -EBUSY;
4161
4162 s->flags &= ~SLAB_RED_ZONE;
4163 if (buf[0] == '1')
4164 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004165 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004166 return length;
4167}
4168SLAB_ATTR(red_zone);
4169
4170static ssize_t poison_show(struct kmem_cache *s, char *buf)
4171{
4172 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4173}
4174
4175static ssize_t poison_store(struct kmem_cache *s,
4176 const char *buf, size_t length)
4177{
4178 if (any_slab_objects(s))
4179 return -EBUSY;
4180
4181 s->flags &= ~SLAB_POISON;
4182 if (buf[0] == '1')
4183 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004184 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004185 return length;
4186}
4187SLAB_ATTR(poison);
4188
4189static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4190{
4191 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4192}
4193
4194static ssize_t store_user_store(struct kmem_cache *s,
4195 const char *buf, size_t length)
4196{
4197 if (any_slab_objects(s))
4198 return -EBUSY;
4199
4200 s->flags &= ~SLAB_STORE_USER;
4201 if (buf[0] == '1')
4202 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004203 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004204 return length;
4205}
4206SLAB_ATTR(store_user);
4207
Christoph Lameter53e15af2007-05-06 14:49:43 -07004208static ssize_t validate_show(struct kmem_cache *s, char *buf)
4209{
4210 return 0;
4211}
4212
4213static ssize_t validate_store(struct kmem_cache *s,
4214 const char *buf, size_t length)
4215{
Christoph Lameter434e2452007-07-17 04:03:30 -07004216 int ret = -EINVAL;
4217
4218 if (buf[0] == '1') {
4219 ret = validate_slab_cache(s);
4220 if (ret >= 0)
4221 ret = length;
4222 }
4223 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004224}
4225SLAB_ATTR(validate);
4226
Christoph Lameter2086d262007-05-06 14:49:46 -07004227static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4228{
4229 return 0;
4230}
4231
4232static ssize_t shrink_store(struct kmem_cache *s,
4233 const char *buf, size_t length)
4234{
4235 if (buf[0] == '1') {
4236 int rc = kmem_cache_shrink(s);
4237
4238 if (rc)
4239 return rc;
4240 } else
4241 return -EINVAL;
4242 return length;
4243}
4244SLAB_ATTR(shrink);
4245
Christoph Lameter88a420e2007-05-06 14:49:45 -07004246static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4247{
4248 if (!(s->flags & SLAB_STORE_USER))
4249 return -ENOSYS;
4250 return list_locations(s, buf, TRACK_ALLOC);
4251}
4252SLAB_ATTR_RO(alloc_calls);
4253
4254static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4255{
4256 if (!(s->flags & SLAB_STORE_USER))
4257 return -ENOSYS;
4258 return list_locations(s, buf, TRACK_FREE);
4259}
4260SLAB_ATTR_RO(free_calls);
4261
Christoph Lameter81819f02007-05-06 14:49:36 -07004262#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004263static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004264{
Christoph Lameter98246012008-01-07 23:20:26 -08004265 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004266}
4267
Christoph Lameter98246012008-01-07 23:20:26 -08004268static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004269 const char *buf, size_t length)
4270{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004271 unsigned long ratio;
4272 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004273
Christoph Lameter0121c6192008-04-29 16:11:12 -07004274 err = strict_strtoul(buf, 10, &ratio);
4275 if (err)
4276 return err;
4277
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004278 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004279 s->remote_node_defrag_ratio = ratio * 10;
4280
Christoph Lameter81819f02007-05-06 14:49:36 -07004281 return length;
4282}
Christoph Lameter98246012008-01-07 23:20:26 -08004283SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004284#endif
4285
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004286#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004287static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4288{
4289 unsigned long sum = 0;
4290 int cpu;
4291 int len;
4292 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4293
4294 if (!data)
4295 return -ENOMEM;
4296
4297 for_each_online_cpu(cpu) {
4298 unsigned x = get_cpu_slab(s, cpu)->stat[si];
4299
4300 data[cpu] = x;
4301 sum += x;
4302 }
4303
4304 len = sprintf(buf, "%lu", sum);
4305
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004306#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004307 for_each_online_cpu(cpu) {
4308 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004309 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004310 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004311#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004312 kfree(data);
4313 return len + sprintf(buf + len, "\n");
4314}
4315
4316#define STAT_ATTR(si, text) \
4317static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4318{ \
4319 return show_stat(s, buf, si); \
4320} \
4321SLAB_ATTR_RO(text); \
4322
4323STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4324STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4325STAT_ATTR(FREE_FASTPATH, free_fastpath);
4326STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4327STAT_ATTR(FREE_FROZEN, free_frozen);
4328STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4329STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4330STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4331STAT_ATTR(ALLOC_SLAB, alloc_slab);
4332STAT_ATTR(ALLOC_REFILL, alloc_refill);
4333STAT_ATTR(FREE_SLAB, free_slab);
4334STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4335STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4336STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4337STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4338STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4339STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004340STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004341#endif
4342
Pekka Enberg06428782008-01-07 23:20:27 -08004343static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004344 &slab_size_attr.attr,
4345 &object_size_attr.attr,
4346 &objs_per_slab_attr.attr,
4347 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004348 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004349 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004350 &objects_partial_attr.attr,
4351 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004352 &slabs_attr.attr,
4353 &partial_attr.attr,
4354 &cpu_slabs_attr.attr,
4355 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004356 &aliases_attr.attr,
4357 &align_attr.attr,
4358 &sanity_checks_attr.attr,
4359 &trace_attr.attr,
4360 &hwcache_align_attr.attr,
4361 &reclaim_account_attr.attr,
4362 &destroy_by_rcu_attr.attr,
4363 &red_zone_attr.attr,
4364 &poison_attr.attr,
4365 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004366 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004367 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004368 &alloc_calls_attr.attr,
4369 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004370#ifdef CONFIG_ZONE_DMA
4371 &cache_dma_attr.attr,
4372#endif
4373#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004374 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004375#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004376#ifdef CONFIG_SLUB_STATS
4377 &alloc_fastpath_attr.attr,
4378 &alloc_slowpath_attr.attr,
4379 &free_fastpath_attr.attr,
4380 &free_slowpath_attr.attr,
4381 &free_frozen_attr.attr,
4382 &free_add_partial_attr.attr,
4383 &free_remove_partial_attr.attr,
4384 &alloc_from_partial_attr.attr,
4385 &alloc_slab_attr.attr,
4386 &alloc_refill_attr.attr,
4387 &free_slab_attr.attr,
4388 &cpuslab_flush_attr.attr,
4389 &deactivate_full_attr.attr,
4390 &deactivate_empty_attr.attr,
4391 &deactivate_to_head_attr.attr,
4392 &deactivate_to_tail_attr.attr,
4393 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004394 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004395#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004396 NULL
4397};
4398
4399static struct attribute_group slab_attr_group = {
4400 .attrs = slab_attrs,
4401};
4402
4403static ssize_t slab_attr_show(struct kobject *kobj,
4404 struct attribute *attr,
4405 char *buf)
4406{
4407 struct slab_attribute *attribute;
4408 struct kmem_cache *s;
4409 int err;
4410
4411 attribute = to_slab_attr(attr);
4412 s = to_slab(kobj);
4413
4414 if (!attribute->show)
4415 return -EIO;
4416
4417 err = attribute->show(s, buf);
4418
4419 return err;
4420}
4421
4422static ssize_t slab_attr_store(struct kobject *kobj,
4423 struct attribute *attr,
4424 const char *buf, size_t len)
4425{
4426 struct slab_attribute *attribute;
4427 struct kmem_cache *s;
4428 int err;
4429
4430 attribute = to_slab_attr(attr);
4431 s = to_slab(kobj);
4432
4433 if (!attribute->store)
4434 return -EIO;
4435
4436 err = attribute->store(s, buf, len);
4437
4438 return err;
4439}
4440
Christoph Lameter151c6022008-01-07 22:29:05 -08004441static void kmem_cache_release(struct kobject *kobj)
4442{
4443 struct kmem_cache *s = to_slab(kobj);
4444
4445 kfree(s);
4446}
4447
Christoph Lameter81819f02007-05-06 14:49:36 -07004448static struct sysfs_ops slab_sysfs_ops = {
4449 .show = slab_attr_show,
4450 .store = slab_attr_store,
4451};
4452
4453static struct kobj_type slab_ktype = {
4454 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004455 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004456};
4457
4458static int uevent_filter(struct kset *kset, struct kobject *kobj)
4459{
4460 struct kobj_type *ktype = get_ktype(kobj);
4461
4462 if (ktype == &slab_ktype)
4463 return 1;
4464 return 0;
4465}
4466
4467static struct kset_uevent_ops slab_uevent_ops = {
4468 .filter = uevent_filter,
4469};
4470
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004471static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004472
4473#define ID_STR_LENGTH 64
4474
4475/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004476 *
4477 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004478 */
4479static char *create_unique_id(struct kmem_cache *s)
4480{
4481 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4482 char *p = name;
4483
4484 BUG_ON(!name);
4485
4486 *p++ = ':';
4487 /*
4488 * First flags affecting slabcache operations. We will only
4489 * get here for aliasable slabs so we do not need to support
4490 * too many flags. The flags here must cover all flags that
4491 * are matched during merging to guarantee that the id is
4492 * unique.
4493 */
4494 if (s->flags & SLAB_CACHE_DMA)
4495 *p++ = 'd';
4496 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4497 *p++ = 'a';
4498 if (s->flags & SLAB_DEBUG_FREE)
4499 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004500 if (!(s->flags & SLAB_NOTRACK))
4501 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004502 if (p != name + 1)
4503 *p++ = '-';
4504 p += sprintf(p, "%07d", s->size);
4505 BUG_ON(p > name + ID_STR_LENGTH - 1);
4506 return name;
4507}
4508
4509static int sysfs_slab_add(struct kmem_cache *s)
4510{
4511 int err;
4512 const char *name;
4513 int unmergeable;
4514
4515 if (slab_state < SYSFS)
4516 /* Defer until later */
4517 return 0;
4518
4519 unmergeable = slab_unmergeable(s);
4520 if (unmergeable) {
4521 /*
4522 * Slabcache can never be merged so we can use the name proper.
4523 * This is typically the case for debug situations. In that
4524 * case we can catch duplicate names easily.
4525 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004526 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004527 name = s->name;
4528 } else {
4529 /*
4530 * Create a unique name for the slab as a target
4531 * for the symlinks.
4532 */
4533 name = create_unique_id(s);
4534 }
4535
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004536 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004537 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4538 if (err) {
4539 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004540 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004541 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004542
4543 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4544 if (err)
4545 return err;
4546 kobject_uevent(&s->kobj, KOBJ_ADD);
4547 if (!unmergeable) {
4548 /* Setup first alias */
4549 sysfs_slab_alias(s, s->name);
4550 kfree(name);
4551 }
4552 return 0;
4553}
4554
4555static void sysfs_slab_remove(struct kmem_cache *s)
4556{
4557 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4558 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004559 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004560}
4561
4562/*
4563 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004564 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004565 */
4566struct saved_alias {
4567 struct kmem_cache *s;
4568 const char *name;
4569 struct saved_alias *next;
4570};
4571
Adrian Bunk5af328a2007-07-17 04:03:27 -07004572static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004573
4574static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4575{
4576 struct saved_alias *al;
4577
4578 if (slab_state == SYSFS) {
4579 /*
4580 * If we have a leftover link then remove it.
4581 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004582 sysfs_remove_link(&slab_kset->kobj, name);
4583 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004584 }
4585
4586 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4587 if (!al)
4588 return -ENOMEM;
4589
4590 al->s = s;
4591 al->name = name;
4592 al->next = alias_list;
4593 alias_list = al;
4594 return 0;
4595}
4596
4597static int __init slab_sysfs_init(void)
4598{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004599 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004600 int err;
4601
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004602 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004603 if (!slab_kset) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004604 printk(KERN_ERR "Cannot register slab subsystem.\n");
4605 return -ENOSYS;
4606 }
4607
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004608 slab_state = SYSFS;
4609
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004610 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004611 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004612 if (err)
4613 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4614 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004615 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004616
4617 while (alias_list) {
4618 struct saved_alias *al = alias_list;
4619
4620 alias_list = alias_list->next;
4621 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004622 if (err)
4623 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4624 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004625 kfree(al);
4626 }
4627
4628 resiliency_test();
4629 return 0;
4630}
4631
4632__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004633#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004634
4635/*
4636 * The /proc/slabinfo ABI
4637 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004638#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004639static void print_slabinfo_header(struct seq_file *m)
4640{
4641 seq_puts(m, "slabinfo - version: 2.1\n");
4642 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4643 "<objperslab> <pagesperslab>");
4644 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4645 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4646 seq_putc(m, '\n');
4647}
4648
4649static void *s_start(struct seq_file *m, loff_t *pos)
4650{
4651 loff_t n = *pos;
4652
4653 down_read(&slub_lock);
4654 if (!n)
4655 print_slabinfo_header(m);
4656
4657 return seq_list_start(&slab_caches, *pos);
4658}
4659
4660static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4661{
4662 return seq_list_next(p, &slab_caches, pos);
4663}
4664
4665static void s_stop(struct seq_file *m, void *p)
4666{
4667 up_read(&slub_lock);
4668}
4669
4670static int s_show(struct seq_file *m, void *p)
4671{
4672 unsigned long nr_partials = 0;
4673 unsigned long nr_slabs = 0;
4674 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004675 unsigned long nr_objs = 0;
4676 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004677 struct kmem_cache *s;
4678 int node;
4679
4680 s = list_entry(p, struct kmem_cache, list);
4681
4682 for_each_online_node(node) {
4683 struct kmem_cache_node *n = get_node(s, node);
4684
4685 if (!n)
4686 continue;
4687
4688 nr_partials += n->nr_partial;
4689 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004690 nr_objs += atomic_long_read(&n->total_objects);
4691 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004692 }
4693
Christoph Lameter205ab992008-04-14 19:11:40 +03004694 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004695
4696 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004697 nr_objs, s->size, oo_objects(s->oo),
4698 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004699 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4700 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4701 0UL);
4702 seq_putc(m, '\n');
4703 return 0;
4704}
4705
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004706static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004707 .start = s_start,
4708 .next = s_next,
4709 .stop = s_stop,
4710 .show = s_show,
4711};
4712
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004713static int slabinfo_open(struct inode *inode, struct file *file)
4714{
4715 return seq_open(file, &slabinfo_op);
4716}
4717
4718static const struct file_operations proc_slabinfo_operations = {
4719 .open = slabinfo_open,
4720 .read = seq_read,
4721 .llseek = seq_lseek,
4722 .release = seq_release,
4723};
4724
4725static int __init slab_proc_init(void)
4726{
4727 proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations);
4728 return 0;
4729}
4730module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004731#endif /* CONFIG_SLABINFO */