blob: 65ffda5934b09b8220e9a00332945dc19ba88de6 [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>
Frederic Weisbecker36994e52008-12-29 13:42:23 -080020#include <trace/kmemtrace.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070021#include <linux/cpu.h>
22#include <linux/cpuset.h>
23#include <linux/mempolicy.h>
24#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070025#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070026#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070027#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070028#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090029#include <linux/fault-inject.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070030
31/*
32 * Lock order:
33 * 1. slab_lock(page)
34 * 2. slab->list_lock
35 *
36 * The slab_lock protects operations on the object of a particular
37 * slab and its metadata in the page struct. If the slab lock
38 * has been taken then no allocations nor frees can be performed
39 * on the objects in the slab nor can the slab be added or removed
40 * from the partial or full lists since this would mean modifying
41 * the page_struct of the slab.
42 *
43 * The list_lock protects the partial and full list on each node and
44 * the partial slab counter. If taken then no new slabs may be added or
45 * removed from the lists nor make the number of partial slabs be modified.
46 * (Note that the total number of slabs is an atomic value that may be
47 * modified without taking the list lock).
48 *
49 * The list_lock is a centralized lock and thus we avoid taking it as
50 * much as possible. As long as SLUB does not have to handle partial
51 * slabs, operations can continue without any centralized lock. F.e.
52 * allocating a long series of objects that fill up slabs does not require
53 * the list lock.
54 *
55 * The lock order is sometimes inverted when we are trying to get a slab
56 * off a list. We take the list_lock and then look for a page on the list
57 * to use. While we do that objects in the slabs may be freed. We can
58 * only operate on the slab if we have also taken the slab_lock. So we use
59 * a slab_trylock() on the slab. If trylock was successful then no frees
60 * can occur anymore and we can use the slab for allocations etc. If the
61 * slab_trylock() does not succeed then frees are in progress in the slab and
62 * we must stay away from it for a while since we may cause a bouncing
63 * cacheline if we try to acquire the lock. So go onto the next slab.
64 * If all pages are busy then we may allocate a new slab instead of reusing
65 * a partial slab. A new slab has noone operating on it and thus there is
66 * no danger of cacheline contention.
67 *
68 * Interrupts are disabled during allocation and deallocation in order to
69 * make the slab allocator safe to use in the context of an irq. In addition
70 * interrupts are disabled to ensure that the processor does not change
71 * while handling per_cpu slabs, due to kernel preemption.
72 *
73 * SLUB assigns one slab for allocation to each processor.
74 * Allocations only occur from these slabs called cpu slabs.
75 *
Christoph Lameter672bba32007-05-09 02:32:39 -070076 * Slabs with free elements are kept on a partial list and during regular
77 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070078 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070079 * We track full slabs for debugging purposes though because otherwise we
80 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070081 *
82 * Slabs are freed when they become empty. Teardown and setup is
83 * minimal so we rely on the page allocators per cpu caches for
84 * fast frees and allocs.
85 *
86 * Overloading of page flags that are otherwise used for LRU management.
87 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070088 * PageActive The slab is frozen and exempt from list processing.
89 * This means that the slab is dedicated to a purpose
90 * such as satisfying allocations for a specific
91 * processor. Objects may be freed in the slab while
92 * it is frozen but slab_free will then skip the usual
93 * list operations. It is up to the processor holding
94 * the slab to integrate the slab into the slab lists
95 * when the slab is no longer needed.
96 *
97 * One use of this flag is to mark slabs that are
98 * used for allocations. Then such a slab becomes a cpu
99 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700100 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700101 * free objects in addition to the regular freelist
102 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700103 *
104 * PageError Slab requires special handling due to debug
105 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700106 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700107 */
108
Christoph Lameter5577bd82007-05-16 22:10:56 -0700109#ifdef CONFIG_SLUB_DEBUG
Andy Whitcroft8a380822008-07-23 21:27:18 -0700110#define SLABDEBUG 1
Christoph Lameter5577bd82007-05-16 22:10:56 -0700111#else
112#define SLABDEBUG 0
113#endif
114
Christoph Lameter81819f02007-05-06 14:49:36 -0700115/*
116 * Issues still to be resolved:
117 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700118 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
119 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700120 * - Variable sizing of the per node arrays
121 */
122
123/* Enable to test recovery from slab corruption on boot */
124#undef SLUB_RESILIENCY_TEST
125
Christoph Lameter81819f02007-05-06 14:49:36 -0700126/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700127 * Mininum number of partial slabs. These will be left on the partial
128 * lists even if they are empty. kmem_cache_shrink may reclaim them.
129 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800130#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700131
Christoph Lameter2086d262007-05-06 14:49:46 -0700132/*
133 * Maximum number of desirable partial slabs.
134 * The existence of more partial slabs makes kmem_cache_shrink
135 * sort the partial list by the number of objects in the.
136 */
137#define MAX_PARTIAL 10
138
Christoph Lameter81819f02007-05-06 14:49:36 -0700139#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
140 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700141
Christoph Lameter81819f02007-05-06 14:49:36 -0700142/*
143 * Set of flags that will prevent slab merging
144 */
145#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
146 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
147
148#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
149 SLAB_CACHE_DMA)
150
151#ifndef ARCH_KMALLOC_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700152#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700153#endif
154
155#ifndef ARCH_SLAB_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700156#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700157#endif
158
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400159#define OO_SHIFT 16
160#define OO_MASK ((1 << OO_SHIFT) - 1)
161#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
162
Christoph Lameter81819f02007-05-06 14:49:36 -0700163/* Internal SLUB flags */
Christoph Lameter1ceef402007-08-07 15:11:48 -0700164#define __OBJECT_POISON 0x80000000 /* Poison object */
165#define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700166
167static int kmem_size = sizeof(struct kmem_cache);
168
169#ifdef CONFIG_SMP
170static struct notifier_block slab_notifier;
171#endif
172
173static enum {
174 DOWN, /* No slab functionality available */
175 PARTIAL, /* kmem_cache_open() works but kmalloc does not */
Christoph Lameter672bba32007-05-09 02:32:39 -0700176 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700177 SYSFS /* Sysfs up */
178} slab_state = DOWN;
179
180/* A list of all slab caches on the system */
181static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700182static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700183
Christoph Lameter02cbc872007-05-09 02:32:43 -0700184/*
185 * Tracking user of a slab.
186 */
187struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300188 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700189 int cpu; /* Was running on cpu */
190 int pid; /* Pid context */
191 unsigned long when; /* When did the operation occur */
192};
193
194enum track_item { TRACK_ALLOC, TRACK_FREE };
195
Christoph Lameterf6acb632008-04-29 16:16:06 -0700196#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -0700197static int sysfs_slab_add(struct kmem_cache *);
198static int sysfs_slab_alias(struct kmem_cache *, const char *);
199static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800200
Christoph Lameter81819f02007-05-06 14:49:36 -0700201#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700202static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
203static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
204 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800205static inline void sysfs_slab_remove(struct kmem_cache *s)
206{
207 kfree(s);
208}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800209
Christoph Lameter81819f02007-05-06 14:49:36 -0700210#endif
211
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800212static inline void stat(struct kmem_cache_cpu *c, enum stat_item si)
213{
214#ifdef CONFIG_SLUB_STATS
215 c->stat[si]++;
216#endif
217}
218
Christoph Lameter81819f02007-05-06 14:49:36 -0700219/********************************************************************
220 * Core slab cache functions
221 *******************************************************************/
222
223int slab_is_available(void)
224{
225 return slab_state >= UP;
226}
227
228static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
229{
230#ifdef CONFIG_NUMA
231 return s->node[node];
232#else
233 return &s->local_node;
234#endif
235}
236
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700237static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu)
238{
Christoph Lameter4c93c3552007-10-16 01:26:08 -0700239#ifdef CONFIG_SMP
240 return s->cpu_slab[cpu];
241#else
242 return &s->cpu_slab;
243#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700244}
245
Christoph Lameter6446faa2008-02-15 23:45:26 -0800246/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700247static inline int check_valid_pointer(struct kmem_cache *s,
248 struct page *page, const void *object)
249{
250 void *base;
251
Christoph Lametera973e9d2008-03-01 13:40:44 -0800252 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700253 return 1;
254
Christoph Lametera973e9d2008-03-01 13:40:44 -0800255 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300256 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700257 (object - base) % s->size) {
258 return 0;
259 }
260
261 return 1;
262}
263
Christoph Lameter81819f02007-05-06 14:49:36 -0700264/*
Christoph Lameter7656c722007-05-09 02:32:40 -0700265 * Slow version of get and set free pointer.
266 *
267 * This version requires touching the cache lines of kmem_cache which
268 * we avoid to do in the fast alloc free paths. There we obtain the offset
269 * from the page struct.
270 */
271static inline void *get_freepointer(struct kmem_cache *s, void *object)
272{
273 return *(void **)(object + s->offset);
274}
275
276static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
277{
278 *(void **)(object + s->offset) = fp;
279}
280
281/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300282#define for_each_object(__p, __s, __addr, __objects) \
283 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700284 __p += (__s)->size)
285
286/* Scan freelist */
287#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800288 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700289
290/* Determine object index from a given position */
291static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
292{
293 return (p - addr) / s->size;
294}
295
Christoph Lameter834f3d12008-04-14 19:11:31 +0300296static inline struct kmem_cache_order_objects oo_make(int order,
297 unsigned long size)
298{
299 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400300 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300301 };
302
303 return x;
304}
305
306static inline int oo_order(struct kmem_cache_order_objects x)
307{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400308 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300309}
310
311static inline int oo_objects(struct kmem_cache_order_objects x)
312{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400313 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300314}
315
Christoph Lameter41ecc552007-05-09 02:32:44 -0700316#ifdef CONFIG_SLUB_DEBUG
317/*
318 * Debug settings:
319 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700320#ifdef CONFIG_SLUB_DEBUG_ON
321static int slub_debug = DEBUG_DEFAULT_FLAGS;
322#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700323static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700324#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700325
326static char *slub_debug_slabs;
327
Christoph Lameter7656c722007-05-09 02:32:40 -0700328/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700329 * Object debugging
330 */
331static void print_section(char *text, u8 *addr, unsigned int length)
332{
333 int i, offset;
334 int newline = 1;
335 char ascii[17];
336
337 ascii[16] = 0;
338
339 for (i = 0; i < length; i++) {
340 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700341 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700342 newline = 0;
343 }
Pekka Enberg06428782008-01-07 23:20:27 -0800344 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700345 offset = i % 16;
346 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
347 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800348 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700349 newline = 1;
350 }
351 }
352 if (!newline) {
353 i %= 16;
354 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800355 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700356 ascii[i] = ' ';
357 i++;
358 }
Pekka Enberg06428782008-01-07 23:20:27 -0800359 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700360 }
361}
362
Christoph Lameter81819f02007-05-06 14:49:36 -0700363static struct track *get_track(struct kmem_cache *s, void *object,
364 enum track_item alloc)
365{
366 struct track *p;
367
368 if (s->offset)
369 p = object + s->offset + sizeof(void *);
370 else
371 p = object + s->inuse;
372
373 return p + alloc;
374}
375
376static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300377 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700378{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900379 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700380
Christoph Lameter81819f02007-05-06 14:49:36 -0700381 if (addr) {
382 p->addr = addr;
383 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400384 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700385 p->when = jiffies;
386 } else
387 memset(p, 0, sizeof(struct track));
388}
389
Christoph Lameter81819f02007-05-06 14:49:36 -0700390static void init_tracking(struct kmem_cache *s, void *object)
391{
Christoph Lameter24922682007-07-17 04:03:18 -0700392 if (!(s->flags & SLAB_STORE_USER))
393 return;
394
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300395 set_track(s, object, TRACK_FREE, 0UL);
396 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700397}
398
399static void print_track(const char *s, struct track *t)
400{
401 if (!t->addr)
402 return;
403
Linus Torvalds7daf7052008-07-14 12:12:53 -0700404 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300405 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700406}
407
Christoph Lameter24922682007-07-17 04:03:18 -0700408static void print_tracking(struct kmem_cache *s, void *object)
409{
410 if (!(s->flags & SLAB_STORE_USER))
411 return;
412
413 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
414 print_track("Freed", get_track(s, object, TRACK_FREE));
415}
416
417static void print_page_info(struct page *page)
418{
Christoph Lameter39b26462008-04-14 19:11:30 +0300419 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
420 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700421
422}
423
424static void slab_bug(struct kmem_cache *s, char *fmt, ...)
425{
426 va_list args;
427 char buf[100];
428
429 va_start(args, fmt);
430 vsnprintf(buf, sizeof(buf), fmt, args);
431 va_end(args);
432 printk(KERN_ERR "========================================"
433 "=====================================\n");
434 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
435 printk(KERN_ERR "----------------------------------------"
436 "-------------------------------------\n\n");
437}
438
439static void slab_fix(struct kmem_cache *s, char *fmt, ...)
440{
441 va_list args;
442 char buf[100];
443
444 va_start(args, fmt);
445 vsnprintf(buf, sizeof(buf), fmt, args);
446 va_end(args);
447 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
448}
449
450static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700451{
452 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800453 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700454
455 print_tracking(s, p);
456
457 print_page_info(page);
458
459 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
460 p, p - addr, get_freepointer(s, p));
461
462 if (p > addr + 16)
463 print_section("Bytes b4", p - 16, 16);
464
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300465 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700466
467 if (s->flags & SLAB_RED_ZONE)
468 print_section("Redzone", p + s->objsize,
469 s->inuse - s->objsize);
470
Christoph Lameter81819f02007-05-06 14:49:36 -0700471 if (s->offset)
472 off = s->offset + sizeof(void *);
473 else
474 off = s->inuse;
475
Christoph Lameter24922682007-07-17 04:03:18 -0700476 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700477 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700478
479 if (off != s->size)
480 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700481 print_section("Padding", p + off, s->size - off);
482
483 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700484}
485
486static void object_err(struct kmem_cache *s, struct page *page,
487 u8 *object, char *reason)
488{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700489 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700490 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700491}
492
Christoph Lameter24922682007-07-17 04:03:18 -0700493static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700494{
495 va_list args;
496 char buf[100];
497
Christoph Lameter24922682007-07-17 04:03:18 -0700498 va_start(args, fmt);
499 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700500 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700501 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700502 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700503 dump_stack();
504}
505
506static void init_object(struct kmem_cache *s, void *object, int active)
507{
508 u8 *p = object;
509
510 if (s->flags & __OBJECT_POISON) {
511 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800512 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700513 }
514
515 if (s->flags & SLAB_RED_ZONE)
516 memset(p + s->objsize,
517 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
518 s->inuse - s->objsize);
519}
520
Christoph Lameter24922682007-07-17 04:03:18 -0700521static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700522{
523 while (bytes) {
524 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700525 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700526 start++;
527 bytes--;
528 }
Christoph Lameter24922682007-07-17 04:03:18 -0700529 return NULL;
530}
531
532static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
533 void *from, void *to)
534{
535 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
536 memset(from, data, to - from);
537}
538
539static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
540 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800541 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700542{
543 u8 *fault;
544 u8 *end;
545
546 fault = check_bytes(start, value, bytes);
547 if (!fault)
548 return 1;
549
550 end = start + bytes;
551 while (end > fault && end[-1] == value)
552 end--;
553
554 slab_bug(s, "%s overwritten", what);
555 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
556 fault, end - 1, fault[0], value);
557 print_trailer(s, page, object);
558
559 restore_bytes(s, what, value, fault, end);
560 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700561}
562
Christoph Lameter81819f02007-05-06 14:49:36 -0700563/*
564 * Object layout:
565 *
566 * object address
567 * Bytes of the object to be managed.
568 * If the freepointer may overlay the object then the free
569 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700570 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700571 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
572 * 0xa5 (POISON_END)
573 *
574 * object + s->objsize
575 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700576 * Padding is extended by another word if Redzoning is enabled and
577 * objsize == inuse.
578 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700579 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
580 * 0xcc (RED_ACTIVE) for objects in use.
581 *
582 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700583 * Meta data starts here.
584 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700585 * A. Free pointer (if we cannot overwrite object on free)
586 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700587 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800588 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700589 * before the word boundary.
590 *
591 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700592 *
593 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700594 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700595 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700596 * If slabcaches are merged then the objsize and inuse boundaries are mostly
597 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700598 * may be used with merged slabcaches.
599 */
600
Christoph Lameter81819f02007-05-06 14:49:36 -0700601static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
602{
603 unsigned long off = s->inuse; /* The end of info */
604
605 if (s->offset)
606 /* Freepointer is placed after the object. */
607 off += sizeof(void *);
608
609 if (s->flags & SLAB_STORE_USER)
610 /* We also have user information there */
611 off += 2 * sizeof(struct track);
612
613 if (s->size == off)
614 return 1;
615
Christoph Lameter24922682007-07-17 04:03:18 -0700616 return check_bytes_and_report(s, page, p, "Object padding",
617 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700618}
619
Christoph Lameter39b26462008-04-14 19:11:30 +0300620/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700621static int slab_pad_check(struct kmem_cache *s, struct page *page)
622{
Christoph Lameter24922682007-07-17 04:03:18 -0700623 u8 *start;
624 u8 *fault;
625 u8 *end;
626 int length;
627 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700628
629 if (!(s->flags & SLAB_POISON))
630 return 1;
631
Christoph Lametera973e9d2008-03-01 13:40:44 -0800632 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300633 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300634 end = start + length;
635 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700636 if (!remainder)
637 return 1;
638
Christoph Lameter39b26462008-04-14 19:11:30 +0300639 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700640 if (!fault)
641 return 1;
642 while (end > fault && end[-1] == POISON_INUSE)
643 end--;
644
645 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300646 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700647
648 restore_bytes(s, "slab padding", POISON_INUSE, start, end);
649 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700650}
651
652static int check_object(struct kmem_cache *s, struct page *page,
653 void *object, int active)
654{
655 u8 *p = object;
656 u8 *endobject = object + s->objsize;
657
658 if (s->flags & SLAB_RED_ZONE) {
659 unsigned int red =
660 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
661
Christoph Lameter24922682007-07-17 04:03:18 -0700662 if (!check_bytes_and_report(s, page, object, "Redzone",
663 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700664 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700665 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800666 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
667 check_bytes_and_report(s, page, p, "Alignment padding",
668 endobject, POISON_INUSE, s->inuse - s->objsize);
669 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700670 }
671
672 if (s->flags & SLAB_POISON) {
673 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700674 (!check_bytes_and_report(s, page, p, "Poison", p,
675 POISON_FREE, s->objsize - 1) ||
676 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800677 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700678 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700679 /*
680 * check_pad_bytes cleans up on its own.
681 */
682 check_pad_bytes(s, page, p);
683 }
684
685 if (!s->offset && active)
686 /*
687 * Object and freepointer overlap. Cannot check
688 * freepointer while object is allocated.
689 */
690 return 1;
691
692 /* Check free pointer validity */
693 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
694 object_err(s, page, p, "Freepointer corrupt");
695 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100696 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700697 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700698 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700699 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800700 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700701 return 0;
702 }
703 return 1;
704}
705
706static int check_slab(struct kmem_cache *s, struct page *page)
707{
Christoph Lameter39b26462008-04-14 19:11:30 +0300708 int maxobj;
709
Christoph Lameter81819f02007-05-06 14:49:36 -0700710 VM_BUG_ON(!irqs_disabled());
711
712 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700713 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700714 return 0;
715 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300716
717 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
718 if (page->objects > maxobj) {
719 slab_err(s, page, "objects %u > max %u",
720 s->name, page->objects, maxobj);
721 return 0;
722 }
723 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700724 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300725 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700726 return 0;
727 }
728 /* Slab_pad_check fixes things up after itself */
729 slab_pad_check(s, page);
730 return 1;
731}
732
733/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700734 * Determine if a certain object on a page is on the freelist. Must hold the
735 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700736 */
737static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
738{
739 int nr = 0;
740 void *fp = page->freelist;
741 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300742 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700743
Christoph Lameter39b26462008-04-14 19:11:30 +0300744 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700745 if (fp == search)
746 return 1;
747 if (!check_valid_pointer(s, page, fp)) {
748 if (object) {
749 object_err(s, page, object,
750 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800751 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700752 break;
753 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700754 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800755 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300756 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700757 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700758 return 0;
759 }
760 break;
761 }
762 object = fp;
763 fp = get_freepointer(s, object);
764 nr++;
765 }
766
Christoph Lameter224a88b2008-04-14 19:11:31 +0300767 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400768 if (max_objects > MAX_OBJS_PER_PAGE)
769 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300770
771 if (page->objects != max_objects) {
772 slab_err(s, page, "Wrong number of objects. Found %d but "
773 "should be %d", page->objects, max_objects);
774 page->objects = max_objects;
775 slab_fix(s, "Number of objects adjusted.");
776 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300777 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700778 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300779 "counted were %d", page->inuse, page->objects - nr);
780 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700781 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700782 }
783 return search == NULL;
784}
785
Christoph Lameter0121c6192008-04-29 16:11:12 -0700786static void trace(struct kmem_cache *s, struct page *page, void *object,
787 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700788{
789 if (s->flags & SLAB_TRACE) {
790 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
791 s->name,
792 alloc ? "alloc" : "free",
793 object, page->inuse,
794 page->freelist);
795
796 if (!alloc)
797 print_section("Object", (void *)object, s->objsize);
798
799 dump_stack();
800 }
801}
802
Christoph Lameter643b1132007-05-06 14:49:42 -0700803/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700804 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700805 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700806static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700807{
Christoph Lameter643b1132007-05-06 14:49:42 -0700808 spin_lock(&n->list_lock);
809 list_add(&page->lru, &n->full);
810 spin_unlock(&n->list_lock);
811}
812
813static void remove_full(struct kmem_cache *s, struct page *page)
814{
815 struct kmem_cache_node *n;
816
817 if (!(s->flags & SLAB_STORE_USER))
818 return;
819
820 n = get_node(s, page_to_nid(page));
821
822 spin_lock(&n->list_lock);
823 list_del(&page->lru);
824 spin_unlock(&n->list_lock);
825}
826
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300827/* Tracking of the number of slabs for debugging purposes */
828static inline unsigned long slabs_node(struct kmem_cache *s, int node)
829{
830 struct kmem_cache_node *n = get_node(s, node);
831
832 return atomic_long_read(&n->nr_slabs);
833}
834
Christoph Lameter205ab992008-04-14 19:11:40 +0300835static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300836{
837 struct kmem_cache_node *n = get_node(s, node);
838
839 /*
840 * May be called early in order to allocate a slab for the
841 * kmem_cache_node structure. Solve the chicken-egg
842 * dilemma by deferring the increment of the count during
843 * bootstrap (see early_kmem_cache_node_alloc).
844 */
Christoph Lameter205ab992008-04-14 19:11:40 +0300845 if (!NUMA_BUILD || n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300846 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300847 atomic_long_add(objects, &n->total_objects);
848 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300849}
Christoph Lameter205ab992008-04-14 19:11:40 +0300850static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300851{
852 struct kmem_cache_node *n = get_node(s, node);
853
854 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300855 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300856}
857
858/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700859static void setup_object_debug(struct kmem_cache *s, struct page *page,
860 void *object)
861{
862 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
863 return;
864
865 init_object(s, object, 0);
866 init_tracking(s, object);
867}
868
869static int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300870 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700871{
872 if (!check_slab(s, page))
873 goto bad;
874
Christoph Lameterd692ef62008-02-15 23:45:24 -0800875 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700876 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700877 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700878 }
879
880 if (!check_valid_pointer(s, page, object)) {
881 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700882 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700883 }
884
Christoph Lameterd692ef62008-02-15 23:45:24 -0800885 if (!check_object(s, page, object, 0))
Christoph Lameter81819f02007-05-06 14:49:36 -0700886 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700887
Christoph Lameter3ec09742007-05-16 22:11:00 -0700888 /* Success perform special debug activities for allocs */
889 if (s->flags & SLAB_STORE_USER)
890 set_track(s, object, TRACK_ALLOC, addr);
891 trace(s, page, object, 1);
892 init_object(s, object, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700893 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700894
Christoph Lameter81819f02007-05-06 14:49:36 -0700895bad:
896 if (PageSlab(page)) {
897 /*
898 * If this is a slab page then lets do the best we can
899 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700900 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700901 */
Christoph Lameter24922682007-07-17 04:03:18 -0700902 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300903 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800904 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700905 }
906 return 0;
907}
908
Christoph Lameter3ec09742007-05-16 22:11:00 -0700909static int free_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300910 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700911{
912 if (!check_slab(s, page))
913 goto fail;
914
915 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700916 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700917 goto fail;
918 }
919
920 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700921 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700922 goto fail;
923 }
924
925 if (!check_object(s, page, object, 1))
926 return 0;
927
928 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800929 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700930 slab_err(s, page, "Attempt to free object(0x%p) "
931 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800932 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700933 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700934 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700935 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700936 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800937 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700938 object_err(s, page, object,
939 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700940 goto fail;
941 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700942
943 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700944 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700945 remove_full(s, page);
946 if (s->flags & SLAB_STORE_USER)
947 set_track(s, object, TRACK_FREE, addr);
948 trace(s, page, object, 0);
949 init_object(s, object, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -0700950 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700951
Christoph Lameter81819f02007-05-06 14:49:36 -0700952fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700953 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700954 return 0;
955}
956
Christoph Lameter41ecc552007-05-09 02:32:44 -0700957static int __init setup_slub_debug(char *str)
958{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700959 slub_debug = DEBUG_DEFAULT_FLAGS;
960 if (*str++ != '=' || !*str)
961 /*
962 * No options specified. Switch on full debugging.
963 */
964 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700965
966 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700967 /*
968 * No options but restriction on slabs. This means full
969 * debugging for slabs matching a pattern.
970 */
971 goto check_slabs;
972
973 slub_debug = 0;
974 if (*str == '-')
975 /*
976 * Switch off all debugging measures.
977 */
978 goto out;
979
980 /*
981 * Determine which debug features should be switched on
982 */
Pekka Enberg06428782008-01-07 23:20:27 -0800983 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700984 switch (tolower(*str)) {
985 case 'f':
986 slub_debug |= SLAB_DEBUG_FREE;
987 break;
988 case 'z':
989 slub_debug |= SLAB_RED_ZONE;
990 break;
991 case 'p':
992 slub_debug |= SLAB_POISON;
993 break;
994 case 'u':
995 slub_debug |= SLAB_STORE_USER;
996 break;
997 case 't':
998 slub_debug |= SLAB_TRACE;
999 break;
1000 default:
1001 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001002 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001003 }
1004 }
1005
1006check_slabs:
1007 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001008 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001009out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001010 return 1;
1011}
1012
1013__setup("slub_debug", setup_slub_debug);
1014
Christoph Lameterba0268a2007-09-11 15:24:11 -07001015static unsigned long kmem_cache_flags(unsigned long objsize,
1016 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001017 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001018{
1019 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001020 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001021 */
Christoph Lametere1533622008-02-15 23:45:24 -08001022 if (slub_debug && (!slub_debug_slabs ||
1023 strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0))
1024 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001025
1026 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001027}
1028#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001029static inline void setup_object_debug(struct kmem_cache *s,
1030 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001031
Christoph Lameter3ec09742007-05-16 22:11:00 -07001032static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001033 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001034
Christoph Lameter3ec09742007-05-16 22:11:00 -07001035static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001036 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001037
Christoph Lameter41ecc552007-05-09 02:32:44 -07001038static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1039 { return 1; }
1040static inline int check_object(struct kmem_cache *s, struct page *page,
1041 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001042static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001043static inline unsigned long kmem_cache_flags(unsigned long objsize,
1044 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001045 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001046{
1047 return flags;
1048}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001049#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001050
1051static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1052 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001053static inline void inc_slabs_node(struct kmem_cache *s, int node,
1054 int objects) {}
1055static inline void dec_slabs_node(struct kmem_cache *s, int node,
1056 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001057#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001058
Christoph Lameter81819f02007-05-06 14:49:36 -07001059/*
1060 * Slab allocation and freeing
1061 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001062static inline struct page *alloc_slab_page(gfp_t flags, int node,
1063 struct kmem_cache_order_objects oo)
1064{
1065 int order = oo_order(oo);
1066
1067 if (node == -1)
1068 return alloc_pages(flags, order);
1069 else
1070 return alloc_pages_node(node, flags, order);
1071}
1072
Christoph Lameter81819f02007-05-06 14:49:36 -07001073static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1074{
Pekka Enberg06428782008-01-07 23:20:27 -08001075 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001076 struct kmem_cache_order_objects oo = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07001077
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001078 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001079
Christoph Lameter65c33762008-04-14 19:11:40 +03001080 page = alloc_slab_page(flags | __GFP_NOWARN | __GFP_NORETRY, node,
1081 oo);
1082 if (unlikely(!page)) {
1083 oo = s->min;
1084 /*
1085 * Allocation may have failed due to fragmentation.
1086 * Try a lower order alloc if possible
1087 */
1088 page = alloc_slab_page(flags, node, oo);
1089 if (!page)
1090 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001091
Christoph Lameter65c33762008-04-14 19:11:40 +03001092 stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK);
1093 }
Christoph Lameter834f3d12008-04-14 19:11:31 +03001094 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001095 mod_zone_page_state(page_zone(page),
1096 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1097 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001098 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001099
1100 return page;
1101}
1102
1103static void setup_object(struct kmem_cache *s, struct page *page,
1104 void *object)
1105{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001106 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001107 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001108 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001109}
1110
1111static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1112{
1113 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001114 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001115 void *last;
1116 void *p;
1117
Christoph Lameter6cb06222007-10-16 01:25:41 -07001118 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001119
Christoph Lameter6cb06222007-10-16 01:25:41 -07001120 page = allocate_slab(s,
1121 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001122 if (!page)
1123 goto out;
1124
Christoph Lameter205ab992008-04-14 19:11:40 +03001125 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001126 page->slab = s;
1127 page->flags |= 1 << PG_slab;
1128 if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
1129 SLAB_STORE_USER | SLAB_TRACE))
Andy Whitcroft8a380822008-07-23 21:27:18 -07001130 __SetPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001131
1132 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001133
1134 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001135 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001136
1137 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001138 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001139 setup_object(s, page, last);
1140 set_freepointer(s, last, p);
1141 last = p;
1142 }
1143 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001144 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001145
1146 page->freelist = start;
1147 page->inuse = 0;
1148out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001149 return page;
1150}
1151
1152static void __free_slab(struct kmem_cache *s, struct page *page)
1153{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001154 int order = compound_order(page);
1155 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001156
Andy Whitcroft8a380822008-07-23 21:27:18 -07001157 if (unlikely(SLABDEBUG && PageSlubDebug(page))) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001158 void *p;
1159
1160 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001161 for_each_object(p, s, page_address(page),
1162 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001163 check_object(s, page, p, 0);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001164 __ClearPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001165 }
1166
1167 mod_zone_page_state(page_zone(page),
1168 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1169 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001170 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001171
Christoph Lameter49bd5222008-04-14 18:52:18 +03001172 __ClearPageSlab(page);
1173 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001174 if (current->reclaim_state)
1175 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001176 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001177}
1178
1179static void rcu_free_slab(struct rcu_head *h)
1180{
1181 struct page *page;
1182
1183 page = container_of((struct list_head *)h, struct page, lru);
1184 __free_slab(page->slab, page);
1185}
1186
1187static void free_slab(struct kmem_cache *s, struct page *page)
1188{
1189 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1190 /*
1191 * RCU free overloads the RCU head over the LRU
1192 */
1193 struct rcu_head *head = (void *)&page->lru;
1194
1195 call_rcu(head, rcu_free_slab);
1196 } else
1197 __free_slab(s, page);
1198}
1199
1200static void discard_slab(struct kmem_cache *s, struct page *page)
1201{
Christoph Lameter205ab992008-04-14 19:11:40 +03001202 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001203 free_slab(s, page);
1204}
1205
1206/*
1207 * Per slab locking using the pagelock
1208 */
1209static __always_inline void slab_lock(struct page *page)
1210{
1211 bit_spin_lock(PG_locked, &page->flags);
1212}
1213
1214static __always_inline void slab_unlock(struct page *page)
1215{
Nick Piggina76d3542008-01-07 23:20:27 -08001216 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001217}
1218
1219static __always_inline int slab_trylock(struct page *page)
1220{
1221 int rc = 1;
1222
1223 rc = bit_spin_trylock(PG_locked, &page->flags);
1224 return rc;
1225}
1226
1227/*
1228 * Management of partially allocated slabs
1229 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001230static void add_partial(struct kmem_cache_node *n,
1231 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001232{
Christoph Lametere95eed52007-05-06 14:49:44 -07001233 spin_lock(&n->list_lock);
1234 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001235 if (tail)
1236 list_add_tail(&page->lru, &n->partial);
1237 else
1238 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001239 spin_unlock(&n->list_lock);
1240}
1241
Christoph Lameter0121c6192008-04-29 16:11:12 -07001242static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001243{
1244 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1245
1246 spin_lock(&n->list_lock);
1247 list_del(&page->lru);
1248 n->nr_partial--;
1249 spin_unlock(&n->list_lock);
1250}
1251
1252/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001253 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001254 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001255 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001256 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001257static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1258 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001259{
1260 if (slab_trylock(page)) {
1261 list_del(&page->lru);
1262 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001263 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001264 return 1;
1265 }
1266 return 0;
1267}
1268
1269/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001270 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001271 */
1272static struct page *get_partial_node(struct kmem_cache_node *n)
1273{
1274 struct page *page;
1275
1276 /*
1277 * Racy check. If we mistakenly see no partial slabs then we
1278 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001279 * partial slab and there is none available then get_partials()
1280 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001281 */
1282 if (!n || !n->nr_partial)
1283 return NULL;
1284
1285 spin_lock(&n->list_lock);
1286 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001287 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001288 goto out;
1289 page = NULL;
1290out:
1291 spin_unlock(&n->list_lock);
1292 return page;
1293}
1294
1295/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001296 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001297 */
1298static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1299{
1300#ifdef CONFIG_NUMA
1301 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001302 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001303 struct zone *zone;
1304 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001305 struct page *page;
1306
1307 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001308 * The defrag ratio allows a configuration of the tradeoffs between
1309 * inter node defragmentation and node local allocations. A lower
1310 * defrag_ratio increases the tendency to do local allocations
1311 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001312 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001313 * If the defrag_ratio is set to 0 then kmalloc() always
1314 * returns node local objects. If the ratio is higher then kmalloc()
1315 * may return off node objects because partial slabs are obtained
1316 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001317 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001318 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001319 * defrag_ratio = 1000) then every (well almost) allocation will
1320 * first attempt to defrag slab caches on other nodes. This means
1321 * scanning over all nodes to look for partial slabs which may be
1322 * expensive if we do it every time we are trying to find a slab
1323 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001324 */
Christoph Lameter98246012008-01-07 23:20:26 -08001325 if (!s->remote_node_defrag_ratio ||
1326 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001327 return NULL;
1328
Mel Gorman0e884602008-04-28 02:12:14 -07001329 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001330 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001331 struct kmem_cache_node *n;
1332
Mel Gorman54a6eb52008-04-28 02:12:16 -07001333 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001334
Mel Gorman54a6eb52008-04-28 02:12:16 -07001335 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001336 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001337 page = get_partial_node(n);
1338 if (page)
1339 return page;
1340 }
1341 }
1342#endif
1343 return NULL;
1344}
1345
1346/*
1347 * Get a partial page, lock it and return it.
1348 */
1349static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1350{
1351 struct page *page;
1352 int searchnode = (node == -1) ? numa_node_id() : node;
1353
1354 page = get_partial_node(get_node(s, searchnode));
1355 if (page || (flags & __GFP_THISNODE))
1356 return page;
1357
1358 return get_any_partial(s, flags);
1359}
1360
1361/*
1362 * Move a page back to the lists.
1363 *
1364 * Must be called with the slab lock held.
1365 *
1366 * On exit the slab lock will have been dropped.
1367 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001368static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001369{
Christoph Lametere95eed52007-05-06 14:49:44 -07001370 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001371 struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id());
Christoph Lametere95eed52007-05-06 14:49:44 -07001372
Andy Whitcroft8a380822008-07-23 21:27:18 -07001373 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001374 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001375
Christoph Lametera973e9d2008-03-01 13:40:44 -08001376 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001377 add_partial(n, page, tail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001378 stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1379 } else {
1380 stat(c, DEACTIVATE_FULL);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001381 if (SLABDEBUG && PageSlubDebug(page) &&
1382 (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001383 add_full(n, page);
1384 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001385 slab_unlock(page);
1386 } else {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001387 stat(c, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001388 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001389 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001390 * Adding an empty slab to the partial slabs in order
1391 * to avoid page allocator overhead. This slab needs
1392 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001393 * so that the others get filled first. That way the
1394 * size of the partial list stays small.
1395 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001396 * kmem_cache_shrink can reclaim any empty slabs from
1397 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001398 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001399 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001400 slab_unlock(page);
1401 } else {
1402 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001403 stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001404 discard_slab(s, page);
1405 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001406 }
1407}
1408
1409/*
1410 * Remove the cpu slab
1411 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001412static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001413{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001414 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001415 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001416
Christoph Lameterb773ad72008-03-04 11:10:17 -08001417 if (page->freelist)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001418 stat(c, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001419 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001420 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001421 * because both freelists are empty. So this is unlikely
1422 * to occur.
1423 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001424 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001425 void **object;
1426
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001427 tail = 0; /* Hot objects. Put the slab first */
1428
Christoph Lameter894b8782007-05-10 03:15:16 -07001429 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001430 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001431 c->freelist = c->freelist[c->offset];
Christoph Lameter894b8782007-05-10 03:15:16 -07001432
1433 /* And put onto the regular freelist */
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001434 object[c->offset] = page->freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07001435 page->freelist = object;
1436 page->inuse--;
1437 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001438 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001439 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001440}
1441
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001442static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001443{
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001444 stat(c, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001445 slab_lock(c->page);
1446 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001447}
1448
1449/*
1450 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001451 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001452 * Called from IPI handler with interrupts disabled.
1453 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001454static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001455{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001456 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001457
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001458 if (likely(c && c->page))
1459 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001460}
1461
1462static void flush_cpu_slab(void *d)
1463{
1464 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001465
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001466 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001467}
1468
1469static void flush_all(struct kmem_cache *s)
1470{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001471 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001472}
1473
1474/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001475 * Check if the objects in a per cpu structure fit numa
1476 * locality expectations.
1477 */
1478static inline int node_match(struct kmem_cache_cpu *c, int node)
1479{
1480#ifdef CONFIG_NUMA
1481 if (node != -1 && c->node != node)
1482 return 0;
1483#endif
1484 return 1;
1485}
1486
1487/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001488 * Slow path. The lockless freelist is empty or we need to perform
1489 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001490 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001491 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001492 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001493 * Processing is still very fast if new objects have been freed to the
1494 * regular freelist. In that case we simply take over the regular freelist
1495 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001496 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001497 * If that is not working then we fall back to the partial lists. We take the
1498 * first element of the freelist as the object to allocate now and move the
1499 * rest of the freelist to the lockless freelist.
1500 *
1501 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001502 * we need to allocate a new slab. This is the slowest path since it involves
1503 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001504 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001505static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1506 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001507{
Christoph Lameter81819f02007-05-06 14:49:36 -07001508 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001509 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001510
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001511 /* We handle __GFP_ZERO in the caller */
1512 gfpflags &= ~__GFP_ZERO;
1513
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001514 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001515 goto new_slab;
1516
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001517 slab_lock(c->page);
1518 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001519 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001520
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001521 stat(c, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001522
Christoph Lameter894b8782007-05-10 03:15:16 -07001523load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001524 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001525 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001526 goto another_slab;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001527 if (unlikely(SLABDEBUG && PageSlubDebug(c->page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001528 goto debug;
1529
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001530 c->freelist = object[c->offset];
Christoph Lameter39b26462008-04-14 19:11:30 +03001531 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001532 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001533 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001534unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001535 slab_unlock(c->page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001536 stat(c, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001537 return object;
1538
1539another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001540 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001541
1542new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001543 new = get_partial(s, gfpflags, node);
1544 if (new) {
1545 c->page = new;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001546 stat(c, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001547 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001548 }
1549
Christoph Lameterb811c202007-10-16 23:25:51 -07001550 if (gfpflags & __GFP_WAIT)
1551 local_irq_enable();
1552
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001553 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001554
1555 if (gfpflags & __GFP_WAIT)
1556 local_irq_disable();
1557
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001558 if (new) {
1559 c = get_cpu_slab(s, smp_processor_id());
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001560 stat(c, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001561 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001562 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001563 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001564 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001565 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001566 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001567 }
Christoph Lameter71c7a062008-02-14 14:28:01 -08001568 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001569debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001570 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001571 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001572
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001573 c->page->inuse++;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001574 c->page->freelist = object[c->offset];
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001575 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001576 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001577}
1578
1579/*
1580 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1581 * have the fastpath folded into their functions. So no function call
1582 * overhead for requests that can be satisfied on the fastpath.
1583 *
1584 * The fastpath works by first checking if the lockless freelist can be used.
1585 * If not then __slab_alloc is called for slow processing.
1586 *
1587 * Otherwise we can simply pick the next object from the lockless free list.
1588 */
Pekka Enberg06428782008-01-07 23:20:27 -08001589static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001590 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001591{
Christoph Lameter894b8782007-05-10 03:15:16 -07001592 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001593 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001594 unsigned long flags;
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001595 unsigned int objsize;
Christoph Lameter1f842602008-01-07 23:20:30 -08001596
Nick Piggincf40bd12009-01-21 08:12:39 +01001597 lockdep_trace_alloc(gfpflags);
OGAWA Hirofumi89124d72008-11-19 21:23:59 +09001598 might_sleep_if(gfpflags & __GFP_WAIT);
Pekka Enberg3c506ef2008-12-29 11:47:05 +02001599
Akinobu Mita773ff602008-12-23 19:37:01 +09001600 if (should_failslab(s->objsize, gfpflags))
1601 return NULL;
1602
Christoph Lameter894b8782007-05-10 03:15:16 -07001603 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001604 c = get_cpu_slab(s, smp_processor_id());
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001605 objsize = c->objsize;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001606 if (unlikely(!c->freelist || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001607
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001608 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001609
1610 else {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001611 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001612 c->freelist = object[c->offset];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001613 stat(c, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001614 }
1615 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001616
1617 if (unlikely((gfpflags & __GFP_ZERO) && object))
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001618 memset(object, 0, objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001619
Christoph Lameter894b8782007-05-10 03:15:16 -07001620 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001621}
1622
1623void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1624{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001625 void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_);
1626
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001627 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001628
1629 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001630}
1631EXPORT_SYMBOL(kmem_cache_alloc);
1632
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001633#ifdef CONFIG_KMEMTRACE
1634void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1635{
1636 return slab_alloc(s, gfpflags, -1, _RET_IP_);
1637}
1638EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1639#endif
1640
Christoph Lameter81819f02007-05-06 14:49:36 -07001641#ifdef CONFIG_NUMA
1642void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1643{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001644 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1645
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001646 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1647 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001648
1649 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001650}
1651EXPORT_SYMBOL(kmem_cache_alloc_node);
1652#endif
1653
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001654#ifdef CONFIG_KMEMTRACE
1655void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1656 gfp_t gfpflags,
1657 int node)
1658{
1659 return slab_alloc(s, gfpflags, node, _RET_IP_);
1660}
1661EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1662#endif
1663
Christoph Lameter81819f02007-05-06 14:49:36 -07001664/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001665 * Slow patch handling. This may still be called frequently since objects
1666 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001667 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001668 * So we still attempt to reduce cache line usage. Just take the slab
1669 * lock and free the item. If there is no additional partial page
1670 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001671 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001672static void __slab_free(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001673 void *x, unsigned long addr, unsigned int offset)
Christoph Lameter81819f02007-05-06 14:49:36 -07001674{
1675 void *prior;
1676 void **object = (void *)x;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001677 struct kmem_cache_cpu *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07001678
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001679 c = get_cpu_slab(s, raw_smp_processor_id());
1680 stat(c, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001681 slab_lock(page);
1682
Andy Whitcroft8a380822008-07-23 21:27:18 -07001683 if (unlikely(SLABDEBUG && PageSlubDebug(page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001684 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001685
Christoph Lameter81819f02007-05-06 14:49:36 -07001686checks_ok:
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001687 prior = object[offset] = page->freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001688 page->freelist = object;
1689 page->inuse--;
1690
Andy Whitcroft8a380822008-07-23 21:27:18 -07001691 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001692 stat(c, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001693 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001694 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001695
1696 if (unlikely(!page->inuse))
1697 goto slab_empty;
1698
1699 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001700 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001701 * then add it.
1702 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001703 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001704 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001705 stat(c, FREE_ADD_PARTIAL);
1706 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001707
1708out_unlock:
1709 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001710 return;
1711
1712slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001713 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001714 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001715 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001716 */
1717 remove_partial(s, page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001718 stat(c, FREE_REMOVE_PARTIAL);
1719 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001720 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001721 stat(c, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001722 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001723 return;
1724
1725debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001726 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001727 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001728 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001729}
1730
Christoph Lameter894b8782007-05-10 03:15:16 -07001731/*
1732 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1733 * can perform fastpath freeing without additional function calls.
1734 *
1735 * The fastpath is only possible if we are freeing to the current cpu slab
1736 * of this processor. This typically the case if we have just allocated
1737 * the item before.
1738 *
1739 * If fastpath is not possible then fall back to __slab_free where we deal
1740 * with all sorts of special processing.
1741 */
Pekka Enberg06428782008-01-07 23:20:27 -08001742static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001743 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001744{
1745 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001746 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001747 unsigned long flags;
1748
Christoph Lameter894b8782007-05-10 03:15:16 -07001749 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001750 c = get_cpu_slab(s, smp_processor_id());
Christoph Lameter27d9e4e2008-02-15 23:45:25 -08001751 debug_check_no_locks_freed(object, c->objsize);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001752 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Pekka Enberg6047a002009-01-14 12:22:25 +02001753 debug_check_no_obj_freed(object, c->objsize);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001754 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001755 object[c->offset] = c->freelist;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001756 c->freelist = object;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001757 stat(c, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001758 } else
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001759 __slab_free(s, page, x, addr, c->offset);
Christoph Lameter894b8782007-05-10 03:15:16 -07001760
1761 local_irq_restore(flags);
1762}
1763
Christoph Lameter81819f02007-05-06 14:49:36 -07001764void kmem_cache_free(struct kmem_cache *s, void *x)
1765{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001766 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001767
Christoph Lameterb49af682007-05-06 14:49:41 -07001768 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001769
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001770 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001771
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001772 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001773}
1774EXPORT_SYMBOL(kmem_cache_free);
1775
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001776/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001777static struct page *get_object_page(const void *x)
1778{
Christoph Lameterb49af682007-05-06 14:49:41 -07001779 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001780
1781 if (!PageSlab(page))
1782 return NULL;
1783
1784 return page;
1785}
1786
1787/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001788 * Object placement in a slab is made very easy because we always start at
1789 * offset 0. If we tune the size of the object to the alignment then we can
1790 * get the required alignment by putting one properly sized object after
1791 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001792 *
1793 * Notice that the allocation order determines the sizes of the per cpu
1794 * caches. Each processor has always one slab available for allocations.
1795 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001796 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001797 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001798 */
1799
1800/*
1801 * Mininum / Maximum order of slab pages. This influences locking overhead
1802 * and slab fragmentation. A higher order reduces the number of partial slabs
1803 * and increases the number of allocations possible without having to
1804 * take the list_lock.
1805 */
1806static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001807static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001808static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001809
1810/*
1811 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001812 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001813 */
1814static int slub_nomerge;
1815
1816/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001817 * Calculate the order of allocation given an slab object size.
1818 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001819 * The order of allocation has significant impact on performance and other
1820 * system components. Generally order 0 allocations should be preferred since
1821 * order 0 does not cause fragmentation in the page allocator. Larger objects
1822 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001823 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001824 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001825 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001826 * In order to reach satisfactory performance we must ensure that a minimum
1827 * number of objects is in one slab. Otherwise we may generate too much
1828 * activity on the partial lists which requires taking the list_lock. This is
1829 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001830 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001831 * slub_max_order specifies the order where we begin to stop considering the
1832 * number of objects in a slab as critical. If we reach slub_max_order then
1833 * we try to keep the page order as low as possible. So we accept more waste
1834 * of space in favor of a small page order.
1835 *
1836 * Higher order allocations also allow the placement of more objects in a
1837 * slab and thereby reduce object handling overhead. If the user has
1838 * requested a higher mininum order then we start with that one instead of
1839 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001840 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001841static inline int slab_order(int size, int min_objects,
1842 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001843{
1844 int order;
1845 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001846 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001847
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001848 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1849 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001850
Christoph Lameter6300ea72007-07-17 04:03:20 -07001851 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001852 fls(min_objects * size - 1) - PAGE_SHIFT);
1853 order <= max_order; order++) {
1854
Christoph Lameter81819f02007-05-06 14:49:36 -07001855 unsigned long slab_size = PAGE_SIZE << order;
1856
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001857 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001858 continue;
1859
Christoph Lameter81819f02007-05-06 14:49:36 -07001860 rem = slab_size % size;
1861
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001862 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001863 break;
1864
1865 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001866
Christoph Lameter81819f02007-05-06 14:49:36 -07001867 return order;
1868}
1869
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001870static inline int calculate_order(int size)
1871{
1872 int order;
1873 int min_objects;
1874 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001875 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001876
1877 /*
1878 * Attempt to find best configuration for a slab. This
1879 * works by first attempting to generate a layout with
1880 * the best configuration and backing off gradually.
1881 *
1882 * First we reduce the acceptable waste in a slab. Then
1883 * we reduce the minimum objects required in a slab.
1884 */
1885 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001886 if (!min_objects)
1887 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001888 max_objects = (PAGE_SIZE << slub_max_order)/size;
1889 min_objects = min(min_objects, max_objects);
1890
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001891 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001892 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001893 while (fraction >= 4) {
1894 order = slab_order(size, min_objects,
1895 slub_max_order, fraction);
1896 if (order <= slub_max_order)
1897 return order;
1898 fraction /= 2;
1899 }
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001900 min_objects --;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001901 }
1902
1903 /*
1904 * We were unable to place multiple objects in a slab. Now
1905 * lets see if we can place a single object there.
1906 */
1907 order = slab_order(size, 1, slub_max_order, 1);
1908 if (order <= slub_max_order)
1909 return order;
1910
1911 /*
1912 * Doh this slab cannot be placed using slub_max_order.
1913 */
1914 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03001915 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001916 return order;
1917 return -ENOSYS;
1918}
1919
Christoph Lameter81819f02007-05-06 14:49:36 -07001920/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001921 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07001922 */
1923static unsigned long calculate_alignment(unsigned long flags,
1924 unsigned long align, unsigned long size)
1925{
1926 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001927 * If the user wants hardware cache aligned objects then follow that
1928 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07001929 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001930 * The hardware cache alignment cannot override the specified
1931 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07001932 */
Nick Pigginb6210382008-03-05 14:05:56 -08001933 if (flags & SLAB_HWCACHE_ALIGN) {
1934 unsigned long ralign = cache_line_size();
1935 while (size <= ralign / 2)
1936 ralign /= 2;
1937 align = max(align, ralign);
1938 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001939
1940 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08001941 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07001942
1943 return ALIGN(align, sizeof(void *));
1944}
1945
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001946static void init_kmem_cache_cpu(struct kmem_cache *s,
1947 struct kmem_cache_cpu *c)
1948{
1949 c->page = NULL;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001950 c->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001951 c->node = 0;
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07001952 c->offset = s->offset / sizeof(void *);
1953 c->objsize = s->objsize;
Pekka Enberg62f75532008-04-14 18:50:44 +03001954#ifdef CONFIG_SLUB_STATS
1955 memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned));
1956#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001957}
1958
Pekka Enberg5595cff2008-08-05 09:28:47 +03001959static void
1960init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07001961{
1962 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001963 spin_lock_init(&n->list_lock);
1964 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07001965#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001966 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07001967 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07001968 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07001969#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07001970}
1971
Christoph Lameter4c93c3552007-10-16 01:26:08 -07001972#ifdef CONFIG_SMP
1973/*
1974 * Per cpu array for per cpu structures.
1975 *
1976 * The per cpu array places all kmem_cache_cpu structures from one processor
1977 * close together meaning that it becomes possible that multiple per cpu
1978 * structures are contained in one cacheline. This may be particularly
1979 * beneficial for the kmalloc caches.
1980 *
1981 * A desktop system typically has around 60-80 slabs. With 100 here we are
1982 * likely able to get per cpu structures for all caches from the array defined
1983 * here. We must be able to cover all kmalloc caches during bootstrap.
1984 *
1985 * If the per cpu array is exhausted then fall back to kmalloc
1986 * of individual cachelines. No sharing is possible then.
1987 */
1988#define NR_KMEM_CACHE_CPU 100
1989
1990static DEFINE_PER_CPU(struct kmem_cache_cpu,
1991 kmem_cache_cpu)[NR_KMEM_CACHE_CPU];
1992
1993static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free);
Rusty Russell174596a2009-01-01 10:12:29 +10301994static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07001995
1996static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s,
1997 int cpu, gfp_t flags)
1998{
1999 struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu);
2000
2001 if (c)
2002 per_cpu(kmem_cache_cpu_free, cpu) =
2003 (void *)c->freelist;
2004 else {
2005 /* Table overflow: So allocate ourselves */
2006 c = kmalloc_node(
2007 ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()),
2008 flags, cpu_to_node(cpu));
2009 if (!c)
2010 return NULL;
2011 }
2012
2013 init_kmem_cache_cpu(s, c);
2014 return c;
2015}
2016
2017static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu)
2018{
2019 if (c < per_cpu(kmem_cache_cpu, cpu) ||
David Rientjes37189092009-01-27 18:59:46 -08002020 c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002021 kfree(c);
2022 return;
2023 }
2024 c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu);
2025 per_cpu(kmem_cache_cpu_free, cpu) = c;
2026}
2027
2028static void free_kmem_cache_cpus(struct kmem_cache *s)
2029{
2030 int cpu;
2031
2032 for_each_online_cpu(cpu) {
2033 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2034
2035 if (c) {
2036 s->cpu_slab[cpu] = NULL;
2037 free_kmem_cache_cpu(c, cpu);
2038 }
2039 }
2040}
2041
2042static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2043{
2044 int cpu;
2045
2046 for_each_online_cpu(cpu) {
2047 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2048
2049 if (c)
2050 continue;
2051
2052 c = alloc_kmem_cache_cpu(s, cpu, flags);
2053 if (!c) {
2054 free_kmem_cache_cpus(s);
2055 return 0;
2056 }
2057 s->cpu_slab[cpu] = c;
2058 }
2059 return 1;
2060}
2061
2062/*
2063 * Initialize the per cpu array.
2064 */
2065static void init_alloc_cpu_cpu(int cpu)
2066{
2067 int i;
2068
Rusty Russell174596a2009-01-01 10:12:29 +10302069 if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)))
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002070 return;
2071
2072 for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--)
2073 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu);
2074
Rusty Russell174596a2009-01-01 10:12:29 +10302075 cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once));
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002076}
2077
2078static void __init init_alloc_cpu(void)
2079{
2080 int cpu;
2081
2082 for_each_online_cpu(cpu)
2083 init_alloc_cpu_cpu(cpu);
2084 }
2085
2086#else
2087static inline void free_kmem_cache_cpus(struct kmem_cache *s) {}
2088static inline void init_alloc_cpu(void) {}
2089
2090static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2091{
2092 init_kmem_cache_cpu(s, &s->cpu_slab);
2093 return 1;
2094}
2095#endif
2096
Christoph Lameter81819f02007-05-06 14:49:36 -07002097#ifdef CONFIG_NUMA
2098/*
2099 * No kmalloc_node yet so do it by hand. We know that this is the first
2100 * slab on the node for this slabcache. There are no concurrent accesses
2101 * possible.
2102 *
2103 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002104 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2105 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002106 */
David Rientjes0094de92008-11-25 19:14:19 -08002107static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002108{
2109 struct page *page;
2110 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002111 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002112
2113 BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2114
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002115 page = new_slab(kmalloc_caches, gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002116
2117 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002118 if (page_to_nid(page) != node) {
2119 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2120 "node %d\n", node);
2121 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2122 "in order to be able to continue\n");
2123 }
2124
Christoph Lameter81819f02007-05-06 14:49:36 -07002125 n = page->freelist;
2126 BUG_ON(!n);
2127 page->freelist = get_freepointer(kmalloc_caches, n);
2128 page->inuse++;
2129 kmalloc_caches->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002130#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterd45f39c2007-07-17 04:03:21 -07002131 init_object(kmalloc_caches, n, 1);
2132 init_tracking(kmalloc_caches, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002133#endif
Pekka Enberg5595cff2008-08-05 09:28:47 +03002134 init_kmem_cache_node(n, kmalloc_caches);
Christoph Lameter205ab992008-04-14 19:11:40 +03002135 inc_slabs_node(kmalloc_caches, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002136
rootba84c732008-01-07 23:20:28 -08002137 /*
2138 * lockdep requires consistent irq usage for each lock
2139 * so even though there cannot be a race this early in
2140 * the boot sequence, we still disable irqs.
2141 */
2142 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002143 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002144 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002145}
2146
2147static void free_kmem_cache_nodes(struct kmem_cache *s)
2148{
2149 int node;
2150
Christoph Lameterf64dc582007-10-16 01:25:33 -07002151 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002152 struct kmem_cache_node *n = s->node[node];
2153 if (n && n != &s->local_node)
2154 kmem_cache_free(kmalloc_caches, n);
2155 s->node[node] = NULL;
2156 }
2157}
2158
2159static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2160{
2161 int node;
2162 int local_node;
2163
2164 if (slab_state >= UP)
2165 local_node = page_to_nid(virt_to_page(s));
2166 else
2167 local_node = 0;
2168
Christoph Lameterf64dc582007-10-16 01:25:33 -07002169 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002170 struct kmem_cache_node *n;
2171
2172 if (local_node == node)
2173 n = &s->local_node;
2174 else {
2175 if (slab_state == DOWN) {
David Rientjes0094de92008-11-25 19:14:19 -08002176 early_kmem_cache_node_alloc(gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002177 continue;
2178 }
2179 n = kmem_cache_alloc_node(kmalloc_caches,
2180 gfpflags, node);
2181
2182 if (!n) {
2183 free_kmem_cache_nodes(s);
2184 return 0;
2185 }
2186
2187 }
2188 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002189 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002190 }
2191 return 1;
2192}
2193#else
2194static void free_kmem_cache_nodes(struct kmem_cache *s)
2195{
2196}
2197
2198static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2199{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002200 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002201 return 1;
2202}
2203#endif
2204
David Rientjesc0bdb232009-02-25 09:16:35 +02002205static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002206{
2207 if (min < MIN_PARTIAL)
2208 min = MIN_PARTIAL;
2209 else if (min > MAX_PARTIAL)
2210 min = MAX_PARTIAL;
2211 s->min_partial = min;
2212}
2213
Christoph Lameter81819f02007-05-06 14:49:36 -07002214/*
2215 * calculate_sizes() determines the order and the distribution of data within
2216 * a slab object.
2217 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002218static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002219{
2220 unsigned long flags = s->flags;
2221 unsigned long size = s->objsize;
2222 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002223 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002224
2225 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002226 * Round up object size to the next word boundary. We can only
2227 * place the free pointer at word boundaries and this determines
2228 * the possible location of the free pointer.
2229 */
2230 size = ALIGN(size, sizeof(void *));
2231
2232#ifdef CONFIG_SLUB_DEBUG
2233 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002234 * Determine if we can poison the object itself. If the user of
2235 * the slab may touch the object after free or before allocation
2236 * then we should never poison the object itself.
2237 */
2238 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002239 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002240 s->flags |= __OBJECT_POISON;
2241 else
2242 s->flags &= ~__OBJECT_POISON;
2243
Christoph Lameter81819f02007-05-06 14:49:36 -07002244
2245 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002246 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002247 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002248 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002249 */
2250 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2251 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002252#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002253
2254 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002255 * With that we have determined the number of bytes in actual use
2256 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002257 */
2258 s->inuse = size;
2259
2260 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002261 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002262 /*
2263 * Relocate free pointer after the object if it is not
2264 * permitted to overwrite the first word of the object on
2265 * kmem_cache_free.
2266 *
2267 * This is the case if we do RCU, have a constructor or
2268 * destructor or are poisoning the objects.
2269 */
2270 s->offset = size;
2271 size += sizeof(void *);
2272 }
2273
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002274#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002275 if (flags & SLAB_STORE_USER)
2276 /*
2277 * Need to store information about allocs and frees after
2278 * the object.
2279 */
2280 size += 2 * sizeof(struct track);
2281
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002282 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002283 /*
2284 * Add some empty padding so that we can catch
2285 * overwrites from earlier objects rather than let
2286 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002287 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002288 * of the object.
2289 */
2290 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002291#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002292
Christoph Lameter81819f02007-05-06 14:49:36 -07002293 /*
2294 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002295 * user specified and the dynamic determination of cache line size
2296 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002297 */
2298 align = calculate_alignment(flags, align, s->objsize);
2299
2300 /*
2301 * SLUB stores one object immediately after another beginning from
2302 * offset 0. In order to align the objects we have to simply size
2303 * each object to conform to the alignment.
2304 */
2305 size = ALIGN(size, align);
2306 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002307 if (forced_order >= 0)
2308 order = forced_order;
2309 else
2310 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002311
Christoph Lameter834f3d12008-04-14 19:11:31 +03002312 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002313 return 0;
2314
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002315 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002316 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002317 s->allocflags |= __GFP_COMP;
2318
2319 if (s->flags & SLAB_CACHE_DMA)
2320 s->allocflags |= SLUB_DMA;
2321
2322 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2323 s->allocflags |= __GFP_RECLAIMABLE;
2324
Christoph Lameter81819f02007-05-06 14:49:36 -07002325 /*
2326 * Determine the number of objects per slab
2327 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002328 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002329 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002330 if (oo_objects(s->oo) > oo_objects(s->max))
2331 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002332
Christoph Lameter834f3d12008-04-14 19:11:31 +03002333 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002334
2335}
2336
Christoph Lameter81819f02007-05-06 14:49:36 -07002337static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2338 const char *name, size_t size,
2339 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002340 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002341{
2342 memset(s, 0, kmem_size);
2343 s->name = name;
2344 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002345 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002346 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002347 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002348
Christoph Lameter06b285d2008-04-14 19:11:41 +03002349 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002350 goto error;
2351
David Rientjes3b89d7d2009-02-22 17:40:07 -08002352 /*
2353 * The larger the object size is, the more pages we want on the partial
2354 * list to avoid pounding the page allocator excessively.
2355 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002356 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002357 s->refcount = 1;
2358#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002359 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002360#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002361 if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2362 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002363
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002364 if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
Christoph Lameter81819f02007-05-06 14:49:36 -07002365 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002366 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002367error:
2368 if (flags & SLAB_PANIC)
2369 panic("Cannot create slab %s size=%lu realsize=%u "
2370 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002371 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002372 s->offset, flags);
2373 return 0;
2374}
Christoph Lameter81819f02007-05-06 14:49:36 -07002375
2376/*
2377 * Check if a given pointer is valid
2378 */
2379int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2380{
Pekka Enberg06428782008-01-07 23:20:27 -08002381 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002382
2383 page = get_object_page(object);
2384
2385 if (!page || s != page->slab)
2386 /* No slab or wrong slab */
2387 return 0;
2388
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002389 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002390 return 0;
2391
2392 /*
2393 * We could also check if the object is on the slabs freelist.
2394 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002395 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002396 * to a certain slab.
2397 */
2398 return 1;
2399}
2400EXPORT_SYMBOL(kmem_ptr_validate);
2401
2402/*
2403 * Determine the size of a slab object
2404 */
2405unsigned int kmem_cache_size(struct kmem_cache *s)
2406{
2407 return s->objsize;
2408}
2409EXPORT_SYMBOL(kmem_cache_size);
2410
2411const char *kmem_cache_name(struct kmem_cache *s)
2412{
2413 return s->name;
2414}
2415EXPORT_SYMBOL(kmem_cache_name);
2416
Christoph Lameter33b12c32008-04-25 12:22:43 -07002417static void list_slab_objects(struct kmem_cache *s, struct page *page,
2418 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002419{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002420#ifdef CONFIG_SLUB_DEBUG
2421 void *addr = page_address(page);
2422 void *p;
2423 DECLARE_BITMAP(map, page->objects);
2424
2425 bitmap_zero(map, page->objects);
2426 slab_err(s, page, "%s", text);
2427 slab_lock(page);
2428 for_each_free_object(p, s, page->freelist)
2429 set_bit(slab_index(p, s, addr), map);
2430
2431 for_each_object(p, s, addr, page->objects) {
2432
2433 if (!test_bit(slab_index(p, s, addr), map)) {
2434 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2435 p, p - addr);
2436 print_tracking(s, p);
2437 }
2438 }
2439 slab_unlock(page);
2440#endif
2441}
2442
Christoph Lameter81819f02007-05-06 14:49:36 -07002443/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002444 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002445 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002446static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002447{
Christoph Lameter81819f02007-05-06 14:49:36 -07002448 unsigned long flags;
2449 struct page *page, *h;
2450
2451 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002452 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002453 if (!page->inuse) {
2454 list_del(&page->lru);
2455 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002456 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002457 } else {
2458 list_slab_objects(s, page,
2459 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002460 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002461 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002462 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002463}
2464
2465/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002466 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002467 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002468static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002469{
2470 int node;
2471
2472 flush_all(s);
2473
2474 /* Attempt to free all objects */
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002475 free_kmem_cache_cpus(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002476 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002477 struct kmem_cache_node *n = get_node(s, node);
2478
Christoph Lameter599870b2008-04-23 12:36:52 -07002479 free_partial(s, n);
2480 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002481 return 1;
2482 }
2483 free_kmem_cache_nodes(s);
2484 return 0;
2485}
2486
2487/*
2488 * Close a cache and release the kmem_cache structure
2489 * (must be used for caches created using kmem_cache_create)
2490 */
2491void kmem_cache_destroy(struct kmem_cache *s)
2492{
2493 down_write(&slub_lock);
2494 s->refcount--;
2495 if (!s->refcount) {
2496 list_del(&s->list);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002497 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03002498 if (kmem_cache_close(s)) {
2499 printk(KERN_ERR "SLUB %s: %s called for cache that "
2500 "still has objects.\n", s->name, __func__);
2501 dump_stack();
2502 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002503 sysfs_slab_remove(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002504 } else
2505 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002506}
2507EXPORT_SYMBOL(kmem_cache_destroy);
2508
2509/********************************************************************
2510 * Kmalloc subsystem
2511 *******************************************************************/
2512
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002513struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned;
Christoph Lameter81819f02007-05-06 14:49:36 -07002514EXPORT_SYMBOL(kmalloc_caches);
2515
Christoph Lameter81819f02007-05-06 14:49:36 -07002516static int __init setup_slub_min_order(char *str)
2517{
Pekka Enberg06428782008-01-07 23:20:27 -08002518 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002519
2520 return 1;
2521}
2522
2523__setup("slub_min_order=", setup_slub_min_order);
2524
2525static int __init setup_slub_max_order(char *str)
2526{
Pekka Enberg06428782008-01-07 23:20:27 -08002527 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002528 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002529
2530 return 1;
2531}
2532
2533__setup("slub_max_order=", setup_slub_max_order);
2534
2535static int __init setup_slub_min_objects(char *str)
2536{
Pekka Enberg06428782008-01-07 23:20:27 -08002537 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002538
2539 return 1;
2540}
2541
2542__setup("slub_min_objects=", setup_slub_min_objects);
2543
2544static int __init setup_slub_nomerge(char *str)
2545{
2546 slub_nomerge = 1;
2547 return 1;
2548}
2549
2550__setup("slub_nomerge", setup_slub_nomerge);
2551
Christoph Lameter81819f02007-05-06 14:49:36 -07002552static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2553 const char *name, int size, gfp_t gfp_flags)
2554{
2555 unsigned int flags = 0;
2556
2557 if (gfp_flags & SLUB_DMA)
2558 flags = SLAB_CACHE_DMA;
2559
2560 down_write(&slub_lock);
2561 if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002562 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002563 goto panic;
2564
2565 list_add(&s->list, &slab_caches);
2566 up_write(&slub_lock);
2567 if (sysfs_slab_add(s))
2568 goto panic;
2569 return s;
2570
2571panic:
2572 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2573}
2574
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002575#ifdef CONFIG_ZONE_DMA
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002576static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
Christoph Lameter1ceef402007-08-07 15:11:48 -07002577
2578static void sysfs_add_func(struct work_struct *w)
2579{
2580 struct kmem_cache *s;
2581
2582 down_write(&slub_lock);
2583 list_for_each_entry(s, &slab_caches, list) {
2584 if (s->flags & __SYSFS_ADD_DEFERRED) {
2585 s->flags &= ~__SYSFS_ADD_DEFERRED;
2586 sysfs_slab_add(s);
2587 }
2588 }
2589 up_write(&slub_lock);
2590}
2591
2592static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2593
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002594static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2595{
2596 struct kmem_cache *s;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002597 char *text;
2598 size_t realsize;
2599
2600 s = kmalloc_caches_dma[index];
2601 if (s)
2602 return s;
2603
2604 /* Dynamically create dma cache */
Christoph Lameter1ceef402007-08-07 15:11:48 -07002605 if (flags & __GFP_WAIT)
2606 down_write(&slub_lock);
2607 else {
2608 if (!down_write_trylock(&slub_lock))
2609 goto out;
2610 }
2611
2612 if (kmalloc_caches_dma[index])
2613 goto unlock_out;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002614
Christoph Lameter7b55f622007-07-17 04:03:27 -07002615 realsize = kmalloc_caches[index].objsize;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08002616 text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2617 (unsigned int)realsize);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002618 s = kmalloc(kmem_size, flags & ~SLUB_DMA);
2619
2620 if (!s || !text || !kmem_cache_open(s, flags, text,
2621 realsize, ARCH_KMALLOC_MINALIGN,
2622 SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) {
2623 kfree(s);
2624 kfree(text);
2625 goto unlock_out;
Christoph Lameterdfce8642007-07-17 04:03:25 -07002626 }
Christoph Lameter1ceef402007-08-07 15:11:48 -07002627
2628 list_add(&s->list, &slab_caches);
2629 kmalloc_caches_dma[index] = s;
2630
2631 schedule_work(&sysfs_add_work);
2632
2633unlock_out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002634 up_write(&slub_lock);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002635out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002636 return kmalloc_caches_dma[index];
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002637}
2638#endif
2639
Christoph Lameterf1b26332007-07-17 04:03:26 -07002640/*
2641 * Conversion table for small slabs sizes / 8 to the index in the
2642 * kmalloc array. This is necessary for slabs < 192 since we have non power
2643 * of two cache sizes there. The size of larger slabs can be determined using
2644 * fls.
2645 */
2646static s8 size_index[24] = {
2647 3, /* 8 */
2648 4, /* 16 */
2649 5, /* 24 */
2650 5, /* 32 */
2651 6, /* 40 */
2652 6, /* 48 */
2653 6, /* 56 */
2654 6, /* 64 */
2655 1, /* 72 */
2656 1, /* 80 */
2657 1, /* 88 */
2658 1, /* 96 */
2659 7, /* 104 */
2660 7, /* 112 */
2661 7, /* 120 */
2662 7, /* 128 */
2663 2, /* 136 */
2664 2, /* 144 */
2665 2, /* 152 */
2666 2, /* 160 */
2667 2, /* 168 */
2668 2, /* 176 */
2669 2, /* 184 */
2670 2 /* 192 */
2671};
2672
Christoph Lameter81819f02007-05-06 14:49:36 -07002673static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2674{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002675 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002676
Christoph Lameterf1b26332007-07-17 04:03:26 -07002677 if (size <= 192) {
2678 if (!size)
2679 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002680
Christoph Lameterf1b26332007-07-17 04:03:26 -07002681 index = size_index[(size - 1) / 8];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002682 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002683 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002684
2685#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002686 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002687 return dma_kmalloc_cache(index, flags);
Christoph Lameterf1b26332007-07-17 04:03:26 -07002688
Christoph Lameter81819f02007-05-06 14:49:36 -07002689#endif
2690 return &kmalloc_caches[index];
2691}
2692
2693void *__kmalloc(size_t size, gfp_t flags)
2694{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002695 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002696 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002697
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002698 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002699 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002700
2701 s = get_slab(size, flags);
2702
2703 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002704 return s;
2705
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002706 ret = slab_alloc(s, flags, -1, _RET_IP_);
2707
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002708 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002709
2710 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002711}
2712EXPORT_SYMBOL(__kmalloc);
2713
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002714static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2715{
2716 struct page *page = alloc_pages_node(node, flags | __GFP_COMP,
2717 get_order(size));
2718
2719 if (page)
2720 return page_address(page);
2721 else
2722 return NULL;
2723}
2724
Christoph Lameter81819f02007-05-06 14:49:36 -07002725#ifdef CONFIG_NUMA
2726void *__kmalloc_node(size_t size, gfp_t flags, int node)
2727{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002728 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002729 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002730
Ingo Molnar057685c2009-02-20 12:15:30 +01002731 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002732 ret = kmalloc_large_node(size, flags, node);
2733
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002734 trace_kmalloc_node(_RET_IP_, ret,
2735 size, PAGE_SIZE << get_order(size),
2736 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002737
2738 return ret;
2739 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002740
2741 s = get_slab(size, flags);
2742
2743 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002744 return s;
2745
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002746 ret = slab_alloc(s, flags, node, _RET_IP_);
2747
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002748 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002749
2750 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002751}
2752EXPORT_SYMBOL(__kmalloc_node);
2753#endif
2754
2755size_t ksize(const void *object)
2756{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002757 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002758 struct kmem_cache *s;
2759
Christoph Lameteref8b4522007-10-16 01:24:46 -07002760 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002761 return 0;
2762
Vegard Nossum294a80a2007-12-04 23:45:30 -08002763 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002764
Pekka Enberg76994412008-05-22 19:22:25 +03002765 if (unlikely(!PageSlab(page))) {
2766 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002767 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002768 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002769 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002770
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002771#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002772 /*
2773 * Debugging requires use of the padding between object
2774 * and whatever may come after it.
2775 */
2776 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2777 return s->objsize;
2778
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002779#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002780 /*
2781 * If we have the need to store the freelist pointer
2782 * back there or track user information then we can
2783 * only use the space before that information.
2784 */
2785 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2786 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002787 /*
2788 * Else we can use all the padding etc for the allocation
2789 */
2790 return s->size;
2791}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002792EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002793
2794void kfree(const void *x)
2795{
Christoph Lameter81819f02007-05-06 14:49:36 -07002796 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002797 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002798
Pekka Enberg2121db72009-03-25 11:05:57 +02002799 trace_kfree(_RET_IP_, x);
2800
Satyam Sharma2408c552007-10-16 01:24:44 -07002801 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002802 return;
2803
Christoph Lameterb49af682007-05-06 14:49:41 -07002804 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002805 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002806 BUG_ON(!PageCompound(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002807 put_page(page);
2808 return;
2809 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002810 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002811}
2812EXPORT_SYMBOL(kfree);
2813
Christoph Lameter2086d262007-05-06 14:49:46 -07002814/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002815 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2816 * the remaining slabs by the number of items in use. The slabs with the
2817 * most items in use come first. New allocations will then fill those up
2818 * and thus they can be removed from the partial lists.
2819 *
2820 * The slabs with the least items are placed last. This results in them
2821 * being allocated from last increasing the chance that the last objects
2822 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002823 */
2824int kmem_cache_shrink(struct kmem_cache *s)
2825{
2826 int node;
2827 int i;
2828 struct kmem_cache_node *n;
2829 struct page *page;
2830 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002831 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002832 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002833 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002834 unsigned long flags;
2835
2836 if (!slabs_by_inuse)
2837 return -ENOMEM;
2838
2839 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002840 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002841 n = get_node(s, node);
2842
2843 if (!n->nr_partial)
2844 continue;
2845
Christoph Lameter834f3d12008-04-14 19:11:31 +03002846 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002847 INIT_LIST_HEAD(slabs_by_inuse + i);
2848
2849 spin_lock_irqsave(&n->list_lock, flags);
2850
2851 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002852 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002853 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002854 * Note that concurrent frees may occur while we hold the
2855 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002856 */
2857 list_for_each_entry_safe(page, t, &n->partial, lru) {
2858 if (!page->inuse && slab_trylock(page)) {
2859 /*
2860 * Must hold slab lock here because slab_free
2861 * may have freed the last object and be
2862 * waiting to release the slab.
2863 */
2864 list_del(&page->lru);
2865 n->nr_partial--;
2866 slab_unlock(page);
2867 discard_slab(s, page);
2868 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002869 list_move(&page->lru,
2870 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002871 }
2872 }
2873
Christoph Lameter2086d262007-05-06 14:49:46 -07002874 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002875 * Rebuild the partial list with the slabs filled up most
2876 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002877 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002878 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002879 list_splice(slabs_by_inuse + i, n->partial.prev);
2880
Christoph Lameter2086d262007-05-06 14:49:46 -07002881 spin_unlock_irqrestore(&n->list_lock, flags);
2882 }
2883
2884 kfree(slabs_by_inuse);
2885 return 0;
2886}
2887EXPORT_SYMBOL(kmem_cache_shrink);
2888
Yasunori Gotob9049e22007-10-21 16:41:37 -07002889#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2890static int slab_mem_going_offline_callback(void *arg)
2891{
2892 struct kmem_cache *s;
2893
2894 down_read(&slub_lock);
2895 list_for_each_entry(s, &slab_caches, list)
2896 kmem_cache_shrink(s);
2897 up_read(&slub_lock);
2898
2899 return 0;
2900}
2901
2902static void slab_mem_offline_callback(void *arg)
2903{
2904 struct kmem_cache_node *n;
2905 struct kmem_cache *s;
2906 struct memory_notify *marg = arg;
2907 int offline_node;
2908
2909 offline_node = marg->status_change_nid;
2910
2911 /*
2912 * If the node still has available memory. we need kmem_cache_node
2913 * for it yet.
2914 */
2915 if (offline_node < 0)
2916 return;
2917
2918 down_read(&slub_lock);
2919 list_for_each_entry(s, &slab_caches, list) {
2920 n = get_node(s, offline_node);
2921 if (n) {
2922 /*
2923 * if n->nr_slabs > 0, slabs still exist on the node
2924 * that is going down. We were unable to free them,
2925 * and offline_pages() function shoudn't call this
2926 * callback. So, we must fail.
2927 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002928 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002929
2930 s->node[offline_node] = NULL;
2931 kmem_cache_free(kmalloc_caches, n);
2932 }
2933 }
2934 up_read(&slub_lock);
2935}
2936
2937static int slab_mem_going_online_callback(void *arg)
2938{
2939 struct kmem_cache_node *n;
2940 struct kmem_cache *s;
2941 struct memory_notify *marg = arg;
2942 int nid = marg->status_change_nid;
2943 int ret = 0;
2944
2945 /*
2946 * If the node's memory is already available, then kmem_cache_node is
2947 * already created. Nothing to do.
2948 */
2949 if (nid < 0)
2950 return 0;
2951
2952 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002953 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002954 * allocate a kmem_cache_node structure in order to bring the node
2955 * online.
2956 */
2957 down_read(&slub_lock);
2958 list_for_each_entry(s, &slab_caches, list) {
2959 /*
2960 * XXX: kmem_cache_alloc_node will fallback to other nodes
2961 * since memory is not yet available from the node that
2962 * is brought up.
2963 */
2964 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
2965 if (!n) {
2966 ret = -ENOMEM;
2967 goto out;
2968 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03002969 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002970 s->node[nid] = n;
2971 }
2972out:
2973 up_read(&slub_lock);
2974 return ret;
2975}
2976
2977static int slab_memory_callback(struct notifier_block *self,
2978 unsigned long action, void *arg)
2979{
2980 int ret = 0;
2981
2982 switch (action) {
2983 case MEM_GOING_ONLINE:
2984 ret = slab_mem_going_online_callback(arg);
2985 break;
2986 case MEM_GOING_OFFLINE:
2987 ret = slab_mem_going_offline_callback(arg);
2988 break;
2989 case MEM_OFFLINE:
2990 case MEM_CANCEL_ONLINE:
2991 slab_mem_offline_callback(arg);
2992 break;
2993 case MEM_ONLINE:
2994 case MEM_CANCEL_OFFLINE:
2995 break;
2996 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08002997 if (ret)
2998 ret = notifier_from_errno(ret);
2999 else
3000 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003001 return ret;
3002}
3003
3004#endif /* CONFIG_MEMORY_HOTPLUG */
3005
Christoph Lameter81819f02007-05-06 14:49:36 -07003006/********************************************************************
3007 * Basic setup of slabs
3008 *******************************************************************/
3009
3010void __init kmem_cache_init(void)
3011{
3012 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003013 int caches = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003014
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003015 init_alloc_cpu();
3016
Christoph Lameter81819f02007-05-06 14:49:36 -07003017#ifdef CONFIG_NUMA
3018 /*
3019 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003020 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003021 * kmem_cache_open for slab_state == DOWN.
3022 */
3023 create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
3024 sizeof(struct kmem_cache_node), GFP_KERNEL);
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003025 kmalloc_caches[0].refcount = -1;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003026 caches++;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003027
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003028 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003029#endif
3030
3031 /* Able to allocate the per node structures */
3032 slab_state = PARTIAL;
3033
3034 /* Caches that are not of the two-to-the-power-of size */
Christoph Lameter4b356be2007-06-16 10:16:13 -07003035 if (KMALLOC_MIN_SIZE <= 64) {
3036 create_kmalloc_cache(&kmalloc_caches[1],
Christoph Lameter81819f02007-05-06 14:49:36 -07003037 "kmalloc-96", 96, GFP_KERNEL);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003038 caches++;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003039 create_kmalloc_cache(&kmalloc_caches[2],
Christoph Lameter81819f02007-05-06 14:49:36 -07003040 "kmalloc-192", 192, GFP_KERNEL);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003041 caches++;
3042 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003043
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003044 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003045 create_kmalloc_cache(&kmalloc_caches[i],
3046 "kmalloc", 1 << i, GFP_KERNEL);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003047 caches++;
3048 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003049
Christoph Lameterf1b26332007-07-17 04:03:26 -07003050
3051 /*
3052 * Patch up the size_index table if we have strange large alignment
3053 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003054 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003055 *
3056 * Largest permitted alignment is 256 bytes due to the way we
3057 * handle the index determination for the smaller caches.
3058 *
3059 * Make sure that nothing crazy happens if someone starts tinkering
3060 * around with ARCH_KMALLOC_MINALIGN
3061 */
3062 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3063 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3064
Christoph Lameter12ad6842007-07-17 04:03:28 -07003065 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8)
Christoph Lameterf1b26332007-07-17 04:03:26 -07003066 size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW;
3067
Christoph Lameter41d54d32008-07-03 09:14:26 -05003068 if (KMALLOC_MIN_SIZE == 128) {
3069 /*
3070 * The 192 byte sized cache is not used if the alignment
3071 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3072 * instead.
3073 */
3074 for (i = 128 + 8; i <= 192; i += 8)
3075 size_index[(i - 1) / 8] = 8;
3076 }
3077
Christoph Lameter81819f02007-05-06 14:49:36 -07003078 slab_state = UP;
3079
3080 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003081 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++)
Christoph Lameter81819f02007-05-06 14:49:36 -07003082 kmalloc_caches[i]. name =
3083 kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i);
3084
3085#ifdef CONFIG_SMP
3086 register_cpu_notifier(&slab_notifier);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003087 kmem_size = offsetof(struct kmem_cache, cpu_slab) +
3088 nr_cpu_ids * sizeof(struct kmem_cache_cpu *);
3089#else
3090 kmem_size = sizeof(struct kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003091#endif
3092
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003093 printk(KERN_INFO
3094 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003095 " CPUs=%d, Nodes=%d\n",
3096 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003097 slub_min_order, slub_max_order, slub_min_objects,
3098 nr_cpu_ids, nr_node_ids);
3099}
3100
3101/*
3102 * Find a mergeable slab cache
3103 */
3104static int slab_unmergeable(struct kmem_cache *s)
3105{
3106 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3107 return 1;
3108
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003109 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003110 return 1;
3111
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003112 /*
3113 * We may have set a slab to be unmergeable during bootstrap.
3114 */
3115 if (s->refcount < 0)
3116 return 1;
3117
Christoph Lameter81819f02007-05-06 14:49:36 -07003118 return 0;
3119}
3120
3121static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003122 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003123 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003124{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003125 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003126
3127 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3128 return NULL;
3129
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003130 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003131 return NULL;
3132
3133 size = ALIGN(size, sizeof(void *));
3134 align = calculate_alignment(flags, align, size);
3135 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003136 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003137
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003138 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003139 if (slab_unmergeable(s))
3140 continue;
3141
3142 if (size > s->size)
3143 continue;
3144
Christoph Lameterba0268a2007-09-11 15:24:11 -07003145 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003146 continue;
3147 /*
3148 * Check if alignment is compatible.
3149 * Courtesy of Adrian Drzewiecki
3150 */
Pekka Enberg06428782008-01-07 23:20:27 -08003151 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003152 continue;
3153
3154 if (s->size - size >= sizeof(void *))
3155 continue;
3156
3157 return s;
3158 }
3159 return NULL;
3160}
3161
3162struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003163 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003164{
3165 struct kmem_cache *s;
3166
3167 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003168 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003169 if (s) {
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003170 int cpu;
3171
Christoph Lameter81819f02007-05-06 14:49:36 -07003172 s->refcount++;
3173 /*
3174 * Adjust the object sizes so that we clear
3175 * the complete object on kzalloc.
3176 */
3177 s->objsize = max(s->objsize, (int)size);
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003178
3179 /*
3180 * And then we need to update the object size in the
3181 * per cpu structures
3182 */
3183 for_each_online_cpu(cpu)
3184 get_cpu_slab(s, cpu)->objsize = s->objsize;
Christoph Lameter6446faa2008-02-15 23:45:26 -08003185
Christoph Lameter81819f02007-05-06 14:49:36 -07003186 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003187 up_write(&slub_lock);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003188
David Rientjes7b8f3b62008-12-17 22:09:46 -08003189 if (sysfs_slab_alias(s, name)) {
3190 down_write(&slub_lock);
3191 s->refcount--;
3192 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003193 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003194 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003195 return s;
3196 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003197
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003198 s = kmalloc(kmem_size, GFP_KERNEL);
3199 if (s) {
3200 if (kmem_cache_open(s, GFP_KERNEL, name,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003201 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003202 list_add(&s->list, &slab_caches);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003203 up_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003204 if (sysfs_slab_add(s)) {
3205 down_write(&slub_lock);
3206 list_del(&s->list);
3207 up_write(&slub_lock);
3208 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003209 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003210 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003211 return s;
3212 }
3213 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003214 }
3215 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003216
3217err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003218 if (flags & SLAB_PANIC)
3219 panic("Cannot create slabcache %s\n", name);
3220 else
3221 s = NULL;
3222 return s;
3223}
3224EXPORT_SYMBOL(kmem_cache_create);
3225
Christoph Lameter81819f02007-05-06 14:49:36 -07003226#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003227/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003228 * Use the cpu notifier to insure that the cpu slabs are flushed when
3229 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003230 */
3231static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3232 unsigned long action, void *hcpu)
3233{
3234 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003235 struct kmem_cache *s;
3236 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003237
3238 switch (action) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003239 case CPU_UP_PREPARE:
3240 case CPU_UP_PREPARE_FROZEN:
3241 init_alloc_cpu_cpu(cpu);
3242 down_read(&slub_lock);
3243 list_for_each_entry(s, &slab_caches, list)
3244 s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu,
3245 GFP_KERNEL);
3246 up_read(&slub_lock);
3247 break;
3248
Christoph Lameter81819f02007-05-06 14:49:36 -07003249 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003250 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003251 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003252 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003253 down_read(&slub_lock);
3254 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003255 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
3256
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003257 local_irq_save(flags);
3258 __flush_cpu_slab(s, cpu);
3259 local_irq_restore(flags);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003260 free_kmem_cache_cpu(c, cpu);
3261 s->cpu_slab[cpu] = NULL;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003262 }
3263 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003264 break;
3265 default:
3266 break;
3267 }
3268 return NOTIFY_OK;
3269}
3270
Pekka Enberg06428782008-01-07 23:20:27 -08003271static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003272 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003273};
Christoph Lameter81819f02007-05-06 14:49:36 -07003274
3275#endif
3276
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003277void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003278{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003279 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003280 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003281
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003282 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003283 return kmalloc_large(size, gfpflags);
3284
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003285 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003286
Satyam Sharma2408c552007-10-16 01:24:44 -07003287 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003288 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003289
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003290 ret = slab_alloc(s, gfpflags, -1, caller);
3291
3292 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003293 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003294
3295 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003296}
3297
3298void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003299 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003300{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003301 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003302 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003303
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003304 if (unlikely(size > SLUB_MAX_SIZE))
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003305 return kmalloc_large_node(size, gfpflags, node);
Pekka Enbergeada35e2008-02-11 22:47:46 +02003306
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003307 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003308
Satyam Sharma2408c552007-10-16 01:24:44 -07003309 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003310 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003311
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003312 ret = slab_alloc(s, gfpflags, node, caller);
3313
3314 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003315 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003316
3317 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003318}
3319
Christoph Lameterf6acb632008-04-29 16:16:06 -07003320#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003321static unsigned long count_partial(struct kmem_cache_node *n,
3322 int (*get_count)(struct page *))
Christoph Lameter5b06c8532008-04-14 18:51:34 +03003323{
3324 unsigned long flags;
3325 unsigned long x = 0;
3326 struct page *page;
3327
3328 spin_lock_irqsave(&n->list_lock, flags);
3329 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter205ab992008-04-14 19:11:40 +03003330 x += get_count(page);
Christoph Lameter5b06c8532008-04-14 18:51:34 +03003331 spin_unlock_irqrestore(&n->list_lock, flags);
3332 return x;
3333}
Christoph Lameter205ab992008-04-14 19:11:40 +03003334
3335static int count_inuse(struct page *page)
3336{
3337 return page->inuse;
3338}
3339
3340static int count_total(struct page *page)
3341{
3342 return page->objects;
3343}
3344
3345static int count_free(struct page *page)
3346{
3347 return page->objects - page->inuse;
3348}
Christoph Lameter5b06c8532008-04-14 18:51:34 +03003349
Christoph Lameter434e2452007-07-17 04:03:30 -07003350static int validate_slab(struct kmem_cache *s, struct page *page,
3351 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003352{
3353 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003354 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003355
3356 if (!check_slab(s, page) ||
3357 !on_freelist(s, page, NULL))
3358 return 0;
3359
3360 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003361 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003362
Christoph Lameter7656c722007-05-09 02:32:40 -07003363 for_each_free_object(p, s, page->freelist) {
3364 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003365 if (!check_object(s, page, p, 0))
3366 return 0;
3367 }
3368
Christoph Lameter224a88b2008-04-14 19:11:31 +03003369 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003370 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003371 if (!check_object(s, page, p, 1))
3372 return 0;
3373 return 1;
3374}
3375
Christoph Lameter434e2452007-07-17 04:03:30 -07003376static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3377 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003378{
3379 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003380 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003381 slab_unlock(page);
3382 } else
3383 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3384 s->name, page);
3385
3386 if (s->flags & DEBUG_DEFAULT_FLAGS) {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003387 if (!PageSlubDebug(page))
3388 printk(KERN_ERR "SLUB %s: SlubDebug not set "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003389 "on slab 0x%p\n", s->name, page);
3390 } else {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003391 if (PageSlubDebug(page))
3392 printk(KERN_ERR "SLUB %s: SlubDebug set on "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003393 "slab 0x%p\n", s->name, page);
3394 }
3395}
3396
Christoph Lameter434e2452007-07-17 04:03:30 -07003397static int validate_slab_node(struct kmem_cache *s,
3398 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003399{
3400 unsigned long count = 0;
3401 struct page *page;
3402 unsigned long flags;
3403
3404 spin_lock_irqsave(&n->list_lock, flags);
3405
3406 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003407 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003408 count++;
3409 }
3410 if (count != n->nr_partial)
3411 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3412 "counter=%ld\n", s->name, count, n->nr_partial);
3413
3414 if (!(s->flags & SLAB_STORE_USER))
3415 goto out;
3416
3417 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003418 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003419 count++;
3420 }
3421 if (count != atomic_long_read(&n->nr_slabs))
3422 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3423 "counter=%ld\n", s->name, count,
3424 atomic_long_read(&n->nr_slabs));
3425
3426out:
3427 spin_unlock_irqrestore(&n->list_lock, flags);
3428 return count;
3429}
3430
Christoph Lameter434e2452007-07-17 04:03:30 -07003431static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003432{
3433 int node;
3434 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003435 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003436 sizeof(unsigned long), GFP_KERNEL);
3437
3438 if (!map)
3439 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003440
3441 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003442 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003443 struct kmem_cache_node *n = get_node(s, node);
3444
Christoph Lameter434e2452007-07-17 04:03:30 -07003445 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003446 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003447 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003448 return count;
3449}
3450
Christoph Lameterb3459702007-05-09 02:32:41 -07003451#ifdef SLUB_RESILIENCY_TEST
3452static void resiliency_test(void)
3453{
3454 u8 *p;
3455
3456 printk(KERN_ERR "SLUB resiliency testing\n");
3457 printk(KERN_ERR "-----------------------\n");
3458 printk(KERN_ERR "A. Corruption after allocation\n");
3459
3460 p = kzalloc(16, GFP_KERNEL);
3461 p[16] = 0x12;
3462 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3463 " 0x12->0x%p\n\n", p + 16);
3464
3465 validate_slab_cache(kmalloc_caches + 4);
3466
3467 /* Hmmm... The next two are dangerous */
3468 p = kzalloc(32, GFP_KERNEL);
3469 p[32 + sizeof(void *)] = 0x34;
3470 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003471 " 0x34 -> -0x%p\n", p);
3472 printk(KERN_ERR
3473 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003474
3475 validate_slab_cache(kmalloc_caches + 5);
3476 p = kzalloc(64, GFP_KERNEL);
3477 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3478 *p = 0x56;
3479 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3480 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003481 printk(KERN_ERR
3482 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003483 validate_slab_cache(kmalloc_caches + 6);
3484
3485 printk(KERN_ERR "\nB. Corruption after free\n");
3486 p = kzalloc(128, GFP_KERNEL);
3487 kfree(p);
3488 *p = 0x78;
3489 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3490 validate_slab_cache(kmalloc_caches + 7);
3491
3492 p = kzalloc(256, GFP_KERNEL);
3493 kfree(p);
3494 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003495 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3496 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003497 validate_slab_cache(kmalloc_caches + 8);
3498
3499 p = kzalloc(512, GFP_KERNEL);
3500 kfree(p);
3501 p[512] = 0xab;
3502 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3503 validate_slab_cache(kmalloc_caches + 9);
3504}
3505#else
3506static void resiliency_test(void) {};
3507#endif
3508
Christoph Lameter88a420e2007-05-06 14:49:45 -07003509/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003510 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003511 * and freed.
3512 */
3513
3514struct location {
3515 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003516 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003517 long long sum_time;
3518 long min_time;
3519 long max_time;
3520 long min_pid;
3521 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303522 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003523 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003524};
3525
3526struct loc_track {
3527 unsigned long max;
3528 unsigned long count;
3529 struct location *loc;
3530};
3531
3532static void free_loc_track(struct loc_track *t)
3533{
3534 if (t->max)
3535 free_pages((unsigned long)t->loc,
3536 get_order(sizeof(struct location) * t->max));
3537}
3538
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003539static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003540{
3541 struct location *l;
3542 int order;
3543
Christoph Lameter88a420e2007-05-06 14:49:45 -07003544 order = get_order(sizeof(struct location) * max);
3545
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003546 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003547 if (!l)
3548 return 0;
3549
3550 if (t->count) {
3551 memcpy(l, t->loc, sizeof(struct location) * t->count);
3552 free_loc_track(t);
3553 }
3554 t->max = max;
3555 t->loc = l;
3556 return 1;
3557}
3558
3559static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003560 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003561{
3562 long start, end, pos;
3563 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003564 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003565 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003566
3567 start = -1;
3568 end = t->count;
3569
3570 for ( ; ; ) {
3571 pos = start + (end - start + 1) / 2;
3572
3573 /*
3574 * There is nothing at "end". If we end up there
3575 * we need to add something to before end.
3576 */
3577 if (pos == end)
3578 break;
3579
3580 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003581 if (track->addr == caddr) {
3582
3583 l = &t->loc[pos];
3584 l->count++;
3585 if (track->when) {
3586 l->sum_time += age;
3587 if (age < l->min_time)
3588 l->min_time = age;
3589 if (age > l->max_time)
3590 l->max_time = age;
3591
3592 if (track->pid < l->min_pid)
3593 l->min_pid = track->pid;
3594 if (track->pid > l->max_pid)
3595 l->max_pid = track->pid;
3596
Rusty Russell174596a2009-01-01 10:12:29 +10303597 cpumask_set_cpu(track->cpu,
3598 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003599 }
3600 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003601 return 1;
3602 }
3603
Christoph Lameter45edfa52007-05-09 02:32:45 -07003604 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003605 end = pos;
3606 else
3607 start = pos;
3608 }
3609
3610 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003611 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003612 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003613 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003614 return 0;
3615
3616 l = t->loc + pos;
3617 if (pos < t->count)
3618 memmove(l + 1, l,
3619 (t->count - pos) * sizeof(struct location));
3620 t->count++;
3621 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003622 l->addr = track->addr;
3623 l->sum_time = age;
3624 l->min_time = age;
3625 l->max_time = age;
3626 l->min_pid = track->pid;
3627 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303628 cpumask_clear(to_cpumask(l->cpus));
3629 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003630 nodes_clear(l->nodes);
3631 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003632 return 1;
3633}
3634
3635static void process_slab(struct loc_track *t, struct kmem_cache *s,
3636 struct page *page, enum track_item alloc)
3637{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003638 void *addr = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +03003639 DECLARE_BITMAP(map, page->objects);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003640 void *p;
3641
Christoph Lameter39b26462008-04-14 19:11:30 +03003642 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003643 for_each_free_object(p, s, page->freelist)
3644 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003645
Christoph Lameter224a88b2008-04-14 19:11:31 +03003646 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003647 if (!test_bit(slab_index(p, s, addr), map))
3648 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003649}
3650
3651static int list_locations(struct kmem_cache *s, char *buf,
3652 enum track_item alloc)
3653{
Harvey Harrisone374d482008-01-31 15:20:50 -08003654 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003655 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003656 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003657 int node;
3658
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003659 if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
Andrew Mortonea3061d2007-10-16 01:26:09 -07003660 GFP_TEMPORARY))
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003661 return sprintf(buf, "Out of memory\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003662
3663 /* Push back cpu slabs */
3664 flush_all(s);
3665
Christoph Lameterf64dc582007-10-16 01:25:33 -07003666 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003667 struct kmem_cache_node *n = get_node(s, node);
3668 unsigned long flags;
3669 struct page *page;
3670
Christoph Lameter9e869432007-08-22 14:01:56 -07003671 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003672 continue;
3673
3674 spin_lock_irqsave(&n->list_lock, flags);
3675 list_for_each_entry(page, &n->partial, lru)
3676 process_slab(&t, s, page, alloc);
3677 list_for_each_entry(page, &n->full, lru)
3678 process_slab(&t, s, page, alloc);
3679 spin_unlock_irqrestore(&n->list_lock, flags);
3680 }
3681
3682 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003683 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003684
Hugh Dickins9c246242008-12-09 13:14:27 -08003685 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003686 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003687 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003688
3689 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003690 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003691 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003692 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003693
3694 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003695 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003696 l->min_time,
3697 (long)div_u64(l->sum_time, l->count),
3698 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003699 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003700 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003701 l->min_time);
3702
3703 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003704 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003705 l->min_pid, l->max_pid);
3706 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003707 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003708 l->min_pid);
3709
Rusty Russell174596a2009-01-01 10:12:29 +10303710 if (num_online_cpus() > 1 &&
3711 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003712 len < PAGE_SIZE - 60) {
3713 len += sprintf(buf + len, " cpus=");
3714 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303715 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003716 }
3717
Christoph Lameter84966342007-06-23 17:16:32 -07003718 if (num_online_nodes() > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003719 len < PAGE_SIZE - 60) {
3720 len += sprintf(buf + len, " nodes=");
3721 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003722 l->nodes);
3723 }
3724
Harvey Harrisone374d482008-01-31 15:20:50 -08003725 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003726 }
3727
3728 free_loc_track(&t);
3729 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003730 len += sprintf(buf, "No data\n");
3731 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003732}
3733
Christoph Lameter81819f02007-05-06 14:49:36 -07003734enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003735 SL_ALL, /* All slabs */
3736 SL_PARTIAL, /* Only partially allocated slabs */
3737 SL_CPU, /* Only slabs used for cpu caches */
3738 SL_OBJECTS, /* Determine allocated objects not slabs */
3739 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003740};
3741
Christoph Lameter205ab992008-04-14 19:11:40 +03003742#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003743#define SO_PARTIAL (1 << SL_PARTIAL)
3744#define SO_CPU (1 << SL_CPU)
3745#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003746#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003747
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003748static ssize_t show_slab_objects(struct kmem_cache *s,
3749 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003750{
3751 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003752 int node;
3753 int x;
3754 unsigned long *nodes;
3755 unsigned long *per_cpu;
3756
3757 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003758 if (!nodes)
3759 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003760 per_cpu = nodes + nr_node_ids;
3761
Christoph Lameter205ab992008-04-14 19:11:40 +03003762 if (flags & SO_CPU) {
3763 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003764
Christoph Lameter205ab992008-04-14 19:11:40 +03003765 for_each_possible_cpu(cpu) {
3766 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003767
Christoph Lameter205ab992008-04-14 19:11:40 +03003768 if (!c || c->node < 0)
3769 continue;
3770
3771 if (c->page) {
3772 if (flags & SO_TOTAL)
3773 x = c->page->objects;
3774 else if (flags & SO_OBJECTS)
3775 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003776 else
3777 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003778
Christoph Lameter81819f02007-05-06 14:49:36 -07003779 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003780 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003781 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003782 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003783 }
3784 }
3785
Christoph Lameter205ab992008-04-14 19:11:40 +03003786 if (flags & SO_ALL) {
3787 for_each_node_state(node, N_NORMAL_MEMORY) {
3788 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003789
Christoph Lameter205ab992008-04-14 19:11:40 +03003790 if (flags & SO_TOTAL)
3791 x = atomic_long_read(&n->total_objects);
3792 else if (flags & SO_OBJECTS)
3793 x = atomic_long_read(&n->total_objects) -
3794 count_partial(n, count_free);
3795
3796 else
3797 x = atomic_long_read(&n->nr_slabs);
3798 total += x;
3799 nodes[node] += x;
3800 }
3801
3802 } else if (flags & SO_PARTIAL) {
3803 for_each_node_state(node, N_NORMAL_MEMORY) {
3804 struct kmem_cache_node *n = get_node(s, node);
3805
3806 if (flags & SO_TOTAL)
3807 x = count_partial(n, count_total);
3808 else if (flags & SO_OBJECTS)
3809 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003810 else
3811 x = n->nr_partial;
3812 total += x;
3813 nodes[node] += x;
3814 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003815 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003816 x = sprintf(buf, "%lu", total);
3817#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003818 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003819 if (nodes[node])
3820 x += sprintf(buf + x, " N%d=%lu",
3821 node, nodes[node]);
3822#endif
3823 kfree(nodes);
3824 return x + sprintf(buf + x, "\n");
3825}
3826
3827static int any_slab_objects(struct kmem_cache *s)
3828{
3829 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003830
3831 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003832 struct kmem_cache_node *n = get_node(s, node);
3833
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003834 if (!n)
3835 continue;
3836
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003837 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003838 return 1;
3839 }
3840 return 0;
3841}
3842
3843#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3844#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3845
3846struct slab_attribute {
3847 struct attribute attr;
3848 ssize_t (*show)(struct kmem_cache *s, char *buf);
3849 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3850};
3851
3852#define SLAB_ATTR_RO(_name) \
3853 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3854
3855#define SLAB_ATTR(_name) \
3856 static struct slab_attribute _name##_attr = \
3857 __ATTR(_name, 0644, _name##_show, _name##_store)
3858
Christoph Lameter81819f02007-05-06 14:49:36 -07003859static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3860{
3861 return sprintf(buf, "%d\n", s->size);
3862}
3863SLAB_ATTR_RO(slab_size);
3864
3865static ssize_t align_show(struct kmem_cache *s, char *buf)
3866{
3867 return sprintf(buf, "%d\n", s->align);
3868}
3869SLAB_ATTR_RO(align);
3870
3871static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3872{
3873 return sprintf(buf, "%d\n", s->objsize);
3874}
3875SLAB_ATTR_RO(object_size);
3876
3877static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3878{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003879 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003880}
3881SLAB_ATTR_RO(objs_per_slab);
3882
Christoph Lameter06b285d2008-04-14 19:11:41 +03003883static ssize_t order_store(struct kmem_cache *s,
3884 const char *buf, size_t length)
3885{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003886 unsigned long order;
3887 int err;
3888
3889 err = strict_strtoul(buf, 10, &order);
3890 if (err)
3891 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003892
3893 if (order > slub_max_order || order < slub_min_order)
3894 return -EINVAL;
3895
3896 calculate_sizes(s, order);
3897 return length;
3898}
3899
Christoph Lameter81819f02007-05-06 14:49:36 -07003900static ssize_t order_show(struct kmem_cache *s, char *buf)
3901{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003902 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003903}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003904SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003905
David Rientjes73d342b2009-02-22 17:40:09 -08003906static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3907{
3908 return sprintf(buf, "%lu\n", s->min_partial);
3909}
3910
3911static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3912 size_t length)
3913{
3914 unsigned long min;
3915 int err;
3916
3917 err = strict_strtoul(buf, 10, &min);
3918 if (err)
3919 return err;
3920
David Rientjesc0bdb232009-02-25 09:16:35 +02003921 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003922 return length;
3923}
3924SLAB_ATTR(min_partial);
3925
Christoph Lameter81819f02007-05-06 14:49:36 -07003926static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3927{
3928 if (s->ctor) {
3929 int n = sprint_symbol(buf, (unsigned long)s->ctor);
3930
3931 return n + sprintf(buf + n, "\n");
3932 }
3933 return 0;
3934}
3935SLAB_ATTR_RO(ctor);
3936
Christoph Lameter81819f02007-05-06 14:49:36 -07003937static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3938{
3939 return sprintf(buf, "%d\n", s->refcount - 1);
3940}
3941SLAB_ATTR_RO(aliases);
3942
3943static ssize_t slabs_show(struct kmem_cache *s, char *buf)
3944{
Christoph Lameter205ab992008-04-14 19:11:40 +03003945 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003946}
3947SLAB_ATTR_RO(slabs);
3948
3949static ssize_t partial_show(struct kmem_cache *s, char *buf)
3950{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003951 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003952}
3953SLAB_ATTR_RO(partial);
3954
3955static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3956{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003957 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003958}
3959SLAB_ATTR_RO(cpu_slabs);
3960
3961static ssize_t objects_show(struct kmem_cache *s, char *buf)
3962{
Christoph Lameter205ab992008-04-14 19:11:40 +03003963 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07003964}
3965SLAB_ATTR_RO(objects);
3966
Christoph Lameter205ab992008-04-14 19:11:40 +03003967static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
3968{
3969 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
3970}
3971SLAB_ATTR_RO(objects_partial);
3972
3973static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
3974{
3975 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
3976}
3977SLAB_ATTR_RO(total_objects);
3978
Christoph Lameter81819f02007-05-06 14:49:36 -07003979static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
3980{
3981 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
3982}
3983
3984static ssize_t sanity_checks_store(struct kmem_cache *s,
3985 const char *buf, size_t length)
3986{
3987 s->flags &= ~SLAB_DEBUG_FREE;
3988 if (buf[0] == '1')
3989 s->flags |= SLAB_DEBUG_FREE;
3990 return length;
3991}
3992SLAB_ATTR(sanity_checks);
3993
3994static ssize_t trace_show(struct kmem_cache *s, char *buf)
3995{
3996 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
3997}
3998
3999static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4000 size_t length)
4001{
4002 s->flags &= ~SLAB_TRACE;
4003 if (buf[0] == '1')
4004 s->flags |= SLAB_TRACE;
4005 return length;
4006}
4007SLAB_ATTR(trace);
4008
4009static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4010{
4011 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4012}
4013
4014static ssize_t reclaim_account_store(struct kmem_cache *s,
4015 const char *buf, size_t length)
4016{
4017 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4018 if (buf[0] == '1')
4019 s->flags |= SLAB_RECLAIM_ACCOUNT;
4020 return length;
4021}
4022SLAB_ATTR(reclaim_account);
4023
4024static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4025{
Christoph Lameter5af60832007-05-06 14:49:56 -07004026 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004027}
4028SLAB_ATTR_RO(hwcache_align);
4029
4030#ifdef CONFIG_ZONE_DMA
4031static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4032{
4033 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4034}
4035SLAB_ATTR_RO(cache_dma);
4036#endif
4037
4038static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4039{
4040 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4041}
4042SLAB_ATTR_RO(destroy_by_rcu);
4043
4044static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4045{
4046 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4047}
4048
4049static ssize_t red_zone_store(struct kmem_cache *s,
4050 const char *buf, size_t length)
4051{
4052 if (any_slab_objects(s))
4053 return -EBUSY;
4054
4055 s->flags &= ~SLAB_RED_ZONE;
4056 if (buf[0] == '1')
4057 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004058 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004059 return length;
4060}
4061SLAB_ATTR(red_zone);
4062
4063static ssize_t poison_show(struct kmem_cache *s, char *buf)
4064{
4065 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4066}
4067
4068static ssize_t poison_store(struct kmem_cache *s,
4069 const char *buf, size_t length)
4070{
4071 if (any_slab_objects(s))
4072 return -EBUSY;
4073
4074 s->flags &= ~SLAB_POISON;
4075 if (buf[0] == '1')
4076 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004077 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004078 return length;
4079}
4080SLAB_ATTR(poison);
4081
4082static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4083{
4084 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4085}
4086
4087static ssize_t store_user_store(struct kmem_cache *s,
4088 const char *buf, size_t length)
4089{
4090 if (any_slab_objects(s))
4091 return -EBUSY;
4092
4093 s->flags &= ~SLAB_STORE_USER;
4094 if (buf[0] == '1')
4095 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004096 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004097 return length;
4098}
4099SLAB_ATTR(store_user);
4100
Christoph Lameter53e15af2007-05-06 14:49:43 -07004101static ssize_t validate_show(struct kmem_cache *s, char *buf)
4102{
4103 return 0;
4104}
4105
4106static ssize_t validate_store(struct kmem_cache *s,
4107 const char *buf, size_t length)
4108{
Christoph Lameter434e2452007-07-17 04:03:30 -07004109 int ret = -EINVAL;
4110
4111 if (buf[0] == '1') {
4112 ret = validate_slab_cache(s);
4113 if (ret >= 0)
4114 ret = length;
4115 }
4116 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004117}
4118SLAB_ATTR(validate);
4119
Christoph Lameter2086d262007-05-06 14:49:46 -07004120static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4121{
4122 return 0;
4123}
4124
4125static ssize_t shrink_store(struct kmem_cache *s,
4126 const char *buf, size_t length)
4127{
4128 if (buf[0] == '1') {
4129 int rc = kmem_cache_shrink(s);
4130
4131 if (rc)
4132 return rc;
4133 } else
4134 return -EINVAL;
4135 return length;
4136}
4137SLAB_ATTR(shrink);
4138
Christoph Lameter88a420e2007-05-06 14:49:45 -07004139static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4140{
4141 if (!(s->flags & SLAB_STORE_USER))
4142 return -ENOSYS;
4143 return list_locations(s, buf, TRACK_ALLOC);
4144}
4145SLAB_ATTR_RO(alloc_calls);
4146
4147static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4148{
4149 if (!(s->flags & SLAB_STORE_USER))
4150 return -ENOSYS;
4151 return list_locations(s, buf, TRACK_FREE);
4152}
4153SLAB_ATTR_RO(free_calls);
4154
Christoph Lameter81819f02007-05-06 14:49:36 -07004155#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004156static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004157{
Christoph Lameter98246012008-01-07 23:20:26 -08004158 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004159}
4160
Christoph Lameter98246012008-01-07 23:20:26 -08004161static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004162 const char *buf, size_t length)
4163{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004164 unsigned long ratio;
4165 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004166
Christoph Lameter0121c6192008-04-29 16:11:12 -07004167 err = strict_strtoul(buf, 10, &ratio);
4168 if (err)
4169 return err;
4170
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004171 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004172 s->remote_node_defrag_ratio = ratio * 10;
4173
Christoph Lameter81819f02007-05-06 14:49:36 -07004174 return length;
4175}
Christoph Lameter98246012008-01-07 23:20:26 -08004176SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004177#endif
4178
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004179#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004180static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4181{
4182 unsigned long sum = 0;
4183 int cpu;
4184 int len;
4185 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4186
4187 if (!data)
4188 return -ENOMEM;
4189
4190 for_each_online_cpu(cpu) {
4191 unsigned x = get_cpu_slab(s, cpu)->stat[si];
4192
4193 data[cpu] = x;
4194 sum += x;
4195 }
4196
4197 len = sprintf(buf, "%lu", sum);
4198
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004199#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004200 for_each_online_cpu(cpu) {
4201 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004202 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004203 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004204#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004205 kfree(data);
4206 return len + sprintf(buf + len, "\n");
4207}
4208
4209#define STAT_ATTR(si, text) \
4210static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4211{ \
4212 return show_stat(s, buf, si); \
4213} \
4214SLAB_ATTR_RO(text); \
4215
4216STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4217STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4218STAT_ATTR(FREE_FASTPATH, free_fastpath);
4219STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4220STAT_ATTR(FREE_FROZEN, free_frozen);
4221STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4222STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4223STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4224STAT_ATTR(ALLOC_SLAB, alloc_slab);
4225STAT_ATTR(ALLOC_REFILL, alloc_refill);
4226STAT_ATTR(FREE_SLAB, free_slab);
4227STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4228STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4229STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4230STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4231STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4232STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004233STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004234#endif
4235
Pekka Enberg06428782008-01-07 23:20:27 -08004236static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004237 &slab_size_attr.attr,
4238 &object_size_attr.attr,
4239 &objs_per_slab_attr.attr,
4240 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004241 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004242 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004243 &objects_partial_attr.attr,
4244 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004245 &slabs_attr.attr,
4246 &partial_attr.attr,
4247 &cpu_slabs_attr.attr,
4248 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004249 &aliases_attr.attr,
4250 &align_attr.attr,
4251 &sanity_checks_attr.attr,
4252 &trace_attr.attr,
4253 &hwcache_align_attr.attr,
4254 &reclaim_account_attr.attr,
4255 &destroy_by_rcu_attr.attr,
4256 &red_zone_attr.attr,
4257 &poison_attr.attr,
4258 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004259 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004260 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004261 &alloc_calls_attr.attr,
4262 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004263#ifdef CONFIG_ZONE_DMA
4264 &cache_dma_attr.attr,
4265#endif
4266#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004267 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004268#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004269#ifdef CONFIG_SLUB_STATS
4270 &alloc_fastpath_attr.attr,
4271 &alloc_slowpath_attr.attr,
4272 &free_fastpath_attr.attr,
4273 &free_slowpath_attr.attr,
4274 &free_frozen_attr.attr,
4275 &free_add_partial_attr.attr,
4276 &free_remove_partial_attr.attr,
4277 &alloc_from_partial_attr.attr,
4278 &alloc_slab_attr.attr,
4279 &alloc_refill_attr.attr,
4280 &free_slab_attr.attr,
4281 &cpuslab_flush_attr.attr,
4282 &deactivate_full_attr.attr,
4283 &deactivate_empty_attr.attr,
4284 &deactivate_to_head_attr.attr,
4285 &deactivate_to_tail_attr.attr,
4286 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004287 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004288#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004289 NULL
4290};
4291
4292static struct attribute_group slab_attr_group = {
4293 .attrs = slab_attrs,
4294};
4295
4296static ssize_t slab_attr_show(struct kobject *kobj,
4297 struct attribute *attr,
4298 char *buf)
4299{
4300 struct slab_attribute *attribute;
4301 struct kmem_cache *s;
4302 int err;
4303
4304 attribute = to_slab_attr(attr);
4305 s = to_slab(kobj);
4306
4307 if (!attribute->show)
4308 return -EIO;
4309
4310 err = attribute->show(s, buf);
4311
4312 return err;
4313}
4314
4315static ssize_t slab_attr_store(struct kobject *kobj,
4316 struct attribute *attr,
4317 const char *buf, size_t len)
4318{
4319 struct slab_attribute *attribute;
4320 struct kmem_cache *s;
4321 int err;
4322
4323 attribute = to_slab_attr(attr);
4324 s = to_slab(kobj);
4325
4326 if (!attribute->store)
4327 return -EIO;
4328
4329 err = attribute->store(s, buf, len);
4330
4331 return err;
4332}
4333
Christoph Lameter151c6022008-01-07 22:29:05 -08004334static void kmem_cache_release(struct kobject *kobj)
4335{
4336 struct kmem_cache *s = to_slab(kobj);
4337
4338 kfree(s);
4339}
4340
Christoph Lameter81819f02007-05-06 14:49:36 -07004341static struct sysfs_ops slab_sysfs_ops = {
4342 .show = slab_attr_show,
4343 .store = slab_attr_store,
4344};
4345
4346static struct kobj_type slab_ktype = {
4347 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004348 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004349};
4350
4351static int uevent_filter(struct kset *kset, struct kobject *kobj)
4352{
4353 struct kobj_type *ktype = get_ktype(kobj);
4354
4355 if (ktype == &slab_ktype)
4356 return 1;
4357 return 0;
4358}
4359
4360static struct kset_uevent_ops slab_uevent_ops = {
4361 .filter = uevent_filter,
4362};
4363
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004364static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004365
4366#define ID_STR_LENGTH 64
4367
4368/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004369 *
4370 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004371 */
4372static char *create_unique_id(struct kmem_cache *s)
4373{
4374 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4375 char *p = name;
4376
4377 BUG_ON(!name);
4378
4379 *p++ = ':';
4380 /*
4381 * First flags affecting slabcache operations. We will only
4382 * get here for aliasable slabs so we do not need to support
4383 * too many flags. The flags here must cover all flags that
4384 * are matched during merging to guarantee that the id is
4385 * unique.
4386 */
4387 if (s->flags & SLAB_CACHE_DMA)
4388 *p++ = 'd';
4389 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4390 *p++ = 'a';
4391 if (s->flags & SLAB_DEBUG_FREE)
4392 *p++ = 'F';
4393 if (p != name + 1)
4394 *p++ = '-';
4395 p += sprintf(p, "%07d", s->size);
4396 BUG_ON(p > name + ID_STR_LENGTH - 1);
4397 return name;
4398}
4399
4400static int sysfs_slab_add(struct kmem_cache *s)
4401{
4402 int err;
4403 const char *name;
4404 int unmergeable;
4405
4406 if (slab_state < SYSFS)
4407 /* Defer until later */
4408 return 0;
4409
4410 unmergeable = slab_unmergeable(s);
4411 if (unmergeable) {
4412 /*
4413 * Slabcache can never be merged so we can use the name proper.
4414 * This is typically the case for debug situations. In that
4415 * case we can catch duplicate names easily.
4416 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004417 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004418 name = s->name;
4419 } else {
4420 /*
4421 * Create a unique name for the slab as a target
4422 * for the symlinks.
4423 */
4424 name = create_unique_id(s);
4425 }
4426
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004427 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004428 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4429 if (err) {
4430 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004431 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004432 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004433
4434 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4435 if (err)
4436 return err;
4437 kobject_uevent(&s->kobj, KOBJ_ADD);
4438 if (!unmergeable) {
4439 /* Setup first alias */
4440 sysfs_slab_alias(s, s->name);
4441 kfree(name);
4442 }
4443 return 0;
4444}
4445
4446static void sysfs_slab_remove(struct kmem_cache *s)
4447{
4448 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4449 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004450 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004451}
4452
4453/*
4454 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004455 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004456 */
4457struct saved_alias {
4458 struct kmem_cache *s;
4459 const char *name;
4460 struct saved_alias *next;
4461};
4462
Adrian Bunk5af328a2007-07-17 04:03:27 -07004463static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004464
4465static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4466{
4467 struct saved_alias *al;
4468
4469 if (slab_state == SYSFS) {
4470 /*
4471 * If we have a leftover link then remove it.
4472 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004473 sysfs_remove_link(&slab_kset->kobj, name);
4474 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004475 }
4476
4477 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4478 if (!al)
4479 return -ENOMEM;
4480
4481 al->s = s;
4482 al->name = name;
4483 al->next = alias_list;
4484 alias_list = al;
4485 return 0;
4486}
4487
4488static int __init slab_sysfs_init(void)
4489{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004490 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004491 int err;
4492
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004493 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004494 if (!slab_kset) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004495 printk(KERN_ERR "Cannot register slab subsystem.\n");
4496 return -ENOSYS;
4497 }
4498
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004499 slab_state = SYSFS;
4500
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004501 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004502 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004503 if (err)
4504 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4505 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004506 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004507
4508 while (alias_list) {
4509 struct saved_alias *al = alias_list;
4510
4511 alias_list = alias_list->next;
4512 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004513 if (err)
4514 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4515 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004516 kfree(al);
4517 }
4518
4519 resiliency_test();
4520 return 0;
4521}
4522
4523__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004524#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004525
4526/*
4527 * The /proc/slabinfo ABI
4528 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004529#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004530static void print_slabinfo_header(struct seq_file *m)
4531{
4532 seq_puts(m, "slabinfo - version: 2.1\n");
4533 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4534 "<objperslab> <pagesperslab>");
4535 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4536 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4537 seq_putc(m, '\n');
4538}
4539
4540static void *s_start(struct seq_file *m, loff_t *pos)
4541{
4542 loff_t n = *pos;
4543
4544 down_read(&slub_lock);
4545 if (!n)
4546 print_slabinfo_header(m);
4547
4548 return seq_list_start(&slab_caches, *pos);
4549}
4550
4551static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4552{
4553 return seq_list_next(p, &slab_caches, pos);
4554}
4555
4556static void s_stop(struct seq_file *m, void *p)
4557{
4558 up_read(&slub_lock);
4559}
4560
4561static int s_show(struct seq_file *m, void *p)
4562{
4563 unsigned long nr_partials = 0;
4564 unsigned long nr_slabs = 0;
4565 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004566 unsigned long nr_objs = 0;
4567 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004568 struct kmem_cache *s;
4569 int node;
4570
4571 s = list_entry(p, struct kmem_cache, list);
4572
4573 for_each_online_node(node) {
4574 struct kmem_cache_node *n = get_node(s, node);
4575
4576 if (!n)
4577 continue;
4578
4579 nr_partials += n->nr_partial;
4580 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004581 nr_objs += atomic_long_read(&n->total_objects);
4582 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004583 }
4584
Christoph Lameter205ab992008-04-14 19:11:40 +03004585 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004586
4587 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004588 nr_objs, s->size, oo_objects(s->oo),
4589 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004590 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4591 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4592 0UL);
4593 seq_putc(m, '\n');
4594 return 0;
4595}
4596
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004597static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004598 .start = s_start,
4599 .next = s_next,
4600 .stop = s_stop,
4601 .show = s_show,
4602};
4603
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004604static int slabinfo_open(struct inode *inode, struct file *file)
4605{
4606 return seq_open(file, &slabinfo_op);
4607}
4608
4609static const struct file_operations proc_slabinfo_operations = {
4610 .open = slabinfo_open,
4611 .read = seq_read,
4612 .llseek = seq_lseek,
4613 .release = seq_release,
4614};
4615
4616static int __init slab_proc_init(void)
4617{
4618 proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations);
4619 return 0;
4620}
4621module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004622#endif /* CONFIG_SLABINFO */