blob: 7ab54ecbd3f3a5abe50eba2c0a5dc035bf4d5efc [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>
12#include <linux/module.h>
13#include <linux/bit_spinlock.h>
14#include <linux/interrupt.h>
15#include <linux/bitops.h>
16#include <linux/slab.h>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040017#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070018#include <linux/seq_file.h>
Frederic Weisbecker36994e52008-12-29 13:42:23 -080019#include <trace/kmemtrace.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070020#include <linux/cpu.h>
21#include <linux/cpuset.h>
22#include <linux/mempolicy.h>
23#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070024#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070025#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070026#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070027#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090028#include <linux/fault-inject.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070029
30/*
31 * Lock order:
32 * 1. slab_lock(page)
33 * 2. slab->list_lock
34 *
35 * The slab_lock protects operations on the object of a particular
36 * slab and its metadata in the page struct. If the slab lock
37 * has been taken then no allocations nor frees can be performed
38 * on the objects in the slab nor can the slab be added or removed
39 * from the partial or full lists since this would mean modifying
40 * the page_struct of the slab.
41 *
42 * The list_lock protects the partial and full list on each node and
43 * the partial slab counter. If taken then no new slabs may be added or
44 * removed from the lists nor make the number of partial slabs be modified.
45 * (Note that the total number of slabs is an atomic value that may be
46 * modified without taking the list lock).
47 *
48 * The list_lock is a centralized lock and thus we avoid taking it as
49 * much as possible. As long as SLUB does not have to handle partial
50 * slabs, operations can continue without any centralized lock. F.e.
51 * allocating a long series of objects that fill up slabs does not require
52 * the list lock.
53 *
54 * The lock order is sometimes inverted when we are trying to get a slab
55 * off a list. We take the list_lock and then look for a page on the list
56 * to use. While we do that objects in the slabs may be freed. We can
57 * only operate on the slab if we have also taken the slab_lock. So we use
58 * a slab_trylock() on the slab. If trylock was successful then no frees
59 * can occur anymore and we can use the slab for allocations etc. If the
60 * slab_trylock() does not succeed then frees are in progress in the slab and
61 * we must stay away from it for a while since we may cause a bouncing
62 * cacheline if we try to acquire the lock. So go onto the next slab.
63 * If all pages are busy then we may allocate a new slab instead of reusing
64 * a partial slab. A new slab has noone operating on it and thus there is
65 * no danger of cacheline contention.
66 *
67 * Interrupts are disabled during allocation and deallocation in order to
68 * make the slab allocator safe to use in the context of an irq. In addition
69 * interrupts are disabled to ensure that the processor does not change
70 * while handling per_cpu slabs, due to kernel preemption.
71 *
72 * SLUB assigns one slab for allocation to each processor.
73 * Allocations only occur from these slabs called cpu slabs.
74 *
Christoph Lameter672bba32007-05-09 02:32:39 -070075 * Slabs with free elements are kept on a partial list and during regular
76 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070077 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070078 * We track full slabs for debugging purposes though because otherwise we
79 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070080 *
81 * Slabs are freed when they become empty. Teardown and setup is
82 * minimal so we rely on the page allocators per cpu caches for
83 * fast frees and allocs.
84 *
85 * Overloading of page flags that are otherwise used for LRU management.
86 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070087 * PageActive The slab is frozen and exempt from list processing.
88 * This means that the slab is dedicated to a purpose
89 * such as satisfying allocations for a specific
90 * processor. Objects may be freed in the slab while
91 * it is frozen but slab_free will then skip the usual
92 * list operations. It is up to the processor holding
93 * the slab to integrate the slab into the slab lists
94 * when the slab is no longer needed.
95 *
96 * One use of this flag is to mark slabs that are
97 * used for allocations. Then such a slab becomes a cpu
98 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -070099 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700100 * free objects in addition to the regular freelist
101 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700102 *
103 * PageError Slab requires special handling due to debug
104 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700105 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700106 */
107
Christoph Lameter5577bd82007-05-16 22:10:56 -0700108#ifdef CONFIG_SLUB_DEBUG
Andy Whitcroft8a380822008-07-23 21:27:18 -0700109#define SLABDEBUG 1
Christoph Lameter5577bd82007-05-16 22:10:56 -0700110#else
111#define SLABDEBUG 0
112#endif
113
Christoph Lameter81819f02007-05-06 14:49:36 -0700114/*
115 * Issues still to be resolved:
116 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700117 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
118 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700119 * - Variable sizing of the per node arrays
120 */
121
122/* Enable to test recovery from slab corruption on boot */
123#undef SLUB_RESILIENCY_TEST
124
Christoph Lameter81819f02007-05-06 14:49:36 -0700125/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700126 * Mininum number of partial slabs. These will be left on the partial
127 * lists even if they are empty. kmem_cache_shrink may reclaim them.
128 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800129#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700130
Christoph Lameter2086d262007-05-06 14:49:46 -0700131/*
132 * Maximum number of desirable partial slabs.
133 * The existence of more partial slabs makes kmem_cache_shrink
134 * sort the partial list by the number of objects in the.
135 */
136#define MAX_PARTIAL 10
137
Christoph Lameter81819f02007-05-06 14:49:36 -0700138#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
139 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700140
Christoph Lameter81819f02007-05-06 14:49:36 -0700141/*
142 * Set of flags that will prevent slab merging
143 */
144#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
145 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
146
147#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
148 SLAB_CACHE_DMA)
149
150#ifndef ARCH_KMALLOC_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700151#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700152#endif
153
154#ifndef ARCH_SLAB_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700155#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700156#endif
157
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400158#define OO_SHIFT 16
159#define OO_MASK ((1 << OO_SHIFT) - 1)
160#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
161
Christoph Lameter81819f02007-05-06 14:49:36 -0700162/* Internal SLUB flags */
Christoph Lameter1ceef402007-08-07 15:11:48 -0700163#define __OBJECT_POISON 0x80000000 /* Poison object */
164#define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700165
166static int kmem_size = sizeof(struct kmem_cache);
167
168#ifdef CONFIG_SMP
169static struct notifier_block slab_notifier;
170#endif
171
172static enum {
173 DOWN, /* No slab functionality available */
174 PARTIAL, /* kmem_cache_open() works but kmalloc does not */
Christoph Lameter672bba32007-05-09 02:32:39 -0700175 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700176 SYSFS /* Sysfs up */
177} slab_state = DOWN;
178
179/* A list of all slab caches on the system */
180static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700181static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700182
Christoph Lameter02cbc872007-05-09 02:32:43 -0700183/*
184 * Tracking user of a slab.
185 */
186struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300187 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700188 int cpu; /* Was running on cpu */
189 int pid; /* Pid context */
190 unsigned long when; /* When did the operation occur */
191};
192
193enum track_item { TRACK_ALLOC, TRACK_FREE };
194
Christoph Lameterf6acb632008-04-29 16:16:06 -0700195#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -0700196static int sysfs_slab_add(struct kmem_cache *);
197static int sysfs_slab_alias(struct kmem_cache *, const char *);
198static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800199
Christoph Lameter81819f02007-05-06 14:49:36 -0700200#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700201static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
202static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
203 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800204static inline void sysfs_slab_remove(struct kmem_cache *s)
205{
206 kfree(s);
207}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800208
Christoph Lameter81819f02007-05-06 14:49:36 -0700209#endif
210
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800211static inline void stat(struct kmem_cache_cpu *c, enum stat_item si)
212{
213#ifdef CONFIG_SLUB_STATS
214 c->stat[si]++;
215#endif
216}
217
Christoph Lameter81819f02007-05-06 14:49:36 -0700218/********************************************************************
219 * Core slab cache functions
220 *******************************************************************/
221
222int slab_is_available(void)
223{
224 return slab_state >= UP;
225}
226
227static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
228{
229#ifdef CONFIG_NUMA
230 return s->node[node];
231#else
232 return &s->local_node;
233#endif
234}
235
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700236static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu)
237{
Christoph Lameter4c93c3552007-10-16 01:26:08 -0700238#ifdef CONFIG_SMP
239 return s->cpu_slab[cpu];
240#else
241 return &s->cpu_slab;
242#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700243}
244
Christoph Lameter6446faa2008-02-15 23:45:26 -0800245/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700246static inline int check_valid_pointer(struct kmem_cache *s,
247 struct page *page, const void *object)
248{
249 void *base;
250
Christoph Lametera973e9d2008-03-01 13:40:44 -0800251 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700252 return 1;
253
Christoph Lametera973e9d2008-03-01 13:40:44 -0800254 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300255 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700256 (object - base) % s->size) {
257 return 0;
258 }
259
260 return 1;
261}
262
Christoph Lameter81819f02007-05-06 14:49:36 -0700263/*
Christoph Lameter7656c722007-05-09 02:32:40 -0700264 * Slow version of get and set free pointer.
265 *
266 * This version requires touching the cache lines of kmem_cache which
267 * we avoid to do in the fast alloc free paths. There we obtain the offset
268 * from the page struct.
269 */
270static inline void *get_freepointer(struct kmem_cache *s, void *object)
271{
272 return *(void **)(object + s->offset);
273}
274
275static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
276{
277 *(void **)(object + s->offset) = fp;
278}
279
280/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300281#define for_each_object(__p, __s, __addr, __objects) \
282 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700283 __p += (__s)->size)
284
285/* Scan freelist */
286#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800287 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700288
289/* Determine object index from a given position */
290static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
291{
292 return (p - addr) / s->size;
293}
294
Christoph Lameter834f3d12008-04-14 19:11:31 +0300295static inline struct kmem_cache_order_objects oo_make(int order,
296 unsigned long size)
297{
298 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400299 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300300 };
301
302 return x;
303}
304
305static inline int oo_order(struct kmem_cache_order_objects x)
306{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400307 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300308}
309
310static inline int oo_objects(struct kmem_cache_order_objects x)
311{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400312 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300313}
314
Christoph Lameter41ecc552007-05-09 02:32:44 -0700315#ifdef CONFIG_SLUB_DEBUG
316/*
317 * Debug settings:
318 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700319#ifdef CONFIG_SLUB_DEBUG_ON
320static int slub_debug = DEBUG_DEFAULT_FLAGS;
321#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700322static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700323#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700324
325static char *slub_debug_slabs;
326
Christoph Lameter7656c722007-05-09 02:32:40 -0700327/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700328 * Object debugging
329 */
330static void print_section(char *text, u8 *addr, unsigned int length)
331{
332 int i, offset;
333 int newline = 1;
334 char ascii[17];
335
336 ascii[16] = 0;
337
338 for (i = 0; i < length; i++) {
339 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700340 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700341 newline = 0;
342 }
Pekka Enberg06428782008-01-07 23:20:27 -0800343 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700344 offset = i % 16;
345 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
346 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800347 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700348 newline = 1;
349 }
350 }
351 if (!newline) {
352 i %= 16;
353 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800354 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700355 ascii[i] = ' ';
356 i++;
357 }
Pekka Enberg06428782008-01-07 23:20:27 -0800358 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700359 }
360}
361
Christoph Lameter81819f02007-05-06 14:49:36 -0700362static struct track *get_track(struct kmem_cache *s, void *object,
363 enum track_item alloc)
364{
365 struct track *p;
366
367 if (s->offset)
368 p = object + s->offset + sizeof(void *);
369 else
370 p = object + s->inuse;
371
372 return p + alloc;
373}
374
375static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300376 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700377{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900378 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700379
Christoph Lameter81819f02007-05-06 14:49:36 -0700380 if (addr) {
381 p->addr = addr;
382 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400383 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700384 p->when = jiffies;
385 } else
386 memset(p, 0, sizeof(struct track));
387}
388
Christoph Lameter81819f02007-05-06 14:49:36 -0700389static void init_tracking(struct kmem_cache *s, void *object)
390{
Christoph Lameter24922682007-07-17 04:03:18 -0700391 if (!(s->flags & SLAB_STORE_USER))
392 return;
393
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300394 set_track(s, object, TRACK_FREE, 0UL);
395 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700396}
397
398static void print_track(const char *s, struct track *t)
399{
400 if (!t->addr)
401 return;
402
Linus Torvalds7daf7052008-07-14 12:12:53 -0700403 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300404 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700405}
406
Christoph Lameter24922682007-07-17 04:03:18 -0700407static void print_tracking(struct kmem_cache *s, void *object)
408{
409 if (!(s->flags & SLAB_STORE_USER))
410 return;
411
412 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
413 print_track("Freed", get_track(s, object, TRACK_FREE));
414}
415
416static void print_page_info(struct page *page)
417{
Christoph Lameter39b26462008-04-14 19:11:30 +0300418 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
419 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700420
421}
422
423static void slab_bug(struct kmem_cache *s, char *fmt, ...)
424{
425 va_list args;
426 char buf[100];
427
428 va_start(args, fmt);
429 vsnprintf(buf, sizeof(buf), fmt, args);
430 va_end(args);
431 printk(KERN_ERR "========================================"
432 "=====================================\n");
433 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
434 printk(KERN_ERR "----------------------------------------"
435 "-------------------------------------\n\n");
436}
437
438static void slab_fix(struct kmem_cache *s, char *fmt, ...)
439{
440 va_list args;
441 char buf[100];
442
443 va_start(args, fmt);
444 vsnprintf(buf, sizeof(buf), fmt, args);
445 va_end(args);
446 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
447}
448
449static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700450{
451 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800452 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700453
454 print_tracking(s, p);
455
456 print_page_info(page);
457
458 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
459 p, p - addr, get_freepointer(s, p));
460
461 if (p > addr + 16)
462 print_section("Bytes b4", p - 16, 16);
463
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300464 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700465
466 if (s->flags & SLAB_RED_ZONE)
467 print_section("Redzone", p + s->objsize,
468 s->inuse - s->objsize);
469
Christoph Lameter81819f02007-05-06 14:49:36 -0700470 if (s->offset)
471 off = s->offset + sizeof(void *);
472 else
473 off = s->inuse;
474
Christoph Lameter24922682007-07-17 04:03:18 -0700475 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700476 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700477
478 if (off != s->size)
479 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700480 print_section("Padding", p + off, s->size - off);
481
482 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700483}
484
485static void object_err(struct kmem_cache *s, struct page *page,
486 u8 *object, char *reason)
487{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700488 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700489 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700490}
491
Christoph Lameter24922682007-07-17 04:03:18 -0700492static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700493{
494 va_list args;
495 char buf[100];
496
Christoph Lameter24922682007-07-17 04:03:18 -0700497 va_start(args, fmt);
498 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700499 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700500 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700501 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700502 dump_stack();
503}
504
505static void init_object(struct kmem_cache *s, void *object, int active)
506{
507 u8 *p = object;
508
509 if (s->flags & __OBJECT_POISON) {
510 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800511 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700512 }
513
514 if (s->flags & SLAB_RED_ZONE)
515 memset(p + s->objsize,
516 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
517 s->inuse - s->objsize);
518}
519
Christoph Lameter24922682007-07-17 04:03:18 -0700520static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700521{
522 while (bytes) {
523 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700524 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700525 start++;
526 bytes--;
527 }
Christoph Lameter24922682007-07-17 04:03:18 -0700528 return NULL;
529}
530
531static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
532 void *from, void *to)
533{
534 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
535 memset(from, data, to - from);
536}
537
538static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
539 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800540 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700541{
542 u8 *fault;
543 u8 *end;
544
545 fault = check_bytes(start, value, bytes);
546 if (!fault)
547 return 1;
548
549 end = start + bytes;
550 while (end > fault && end[-1] == value)
551 end--;
552
553 slab_bug(s, "%s overwritten", what);
554 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
555 fault, end - 1, fault[0], value);
556 print_trailer(s, page, object);
557
558 restore_bytes(s, what, value, fault, end);
559 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700560}
561
Christoph Lameter81819f02007-05-06 14:49:36 -0700562/*
563 * Object layout:
564 *
565 * object address
566 * Bytes of the object to be managed.
567 * If the freepointer may overlay the object then the free
568 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700569 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700570 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
571 * 0xa5 (POISON_END)
572 *
573 * object + s->objsize
574 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700575 * Padding is extended by another word if Redzoning is enabled and
576 * objsize == inuse.
577 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700578 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
579 * 0xcc (RED_ACTIVE) for objects in use.
580 *
581 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700582 * Meta data starts here.
583 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700584 * A. Free pointer (if we cannot overwrite object on free)
585 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700586 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800587 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700588 * before the word boundary.
589 *
590 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700591 *
592 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700593 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700594 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700595 * If slabcaches are merged then the objsize and inuse boundaries are mostly
596 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700597 * may be used with merged slabcaches.
598 */
599
Christoph Lameter81819f02007-05-06 14:49:36 -0700600static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
601{
602 unsigned long off = s->inuse; /* The end of info */
603
604 if (s->offset)
605 /* Freepointer is placed after the object. */
606 off += sizeof(void *);
607
608 if (s->flags & SLAB_STORE_USER)
609 /* We also have user information there */
610 off += 2 * sizeof(struct track);
611
612 if (s->size == off)
613 return 1;
614
Christoph Lameter24922682007-07-17 04:03:18 -0700615 return check_bytes_and_report(s, page, p, "Object padding",
616 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700617}
618
Christoph Lameter39b26462008-04-14 19:11:30 +0300619/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700620static int slab_pad_check(struct kmem_cache *s, struct page *page)
621{
Christoph Lameter24922682007-07-17 04:03:18 -0700622 u8 *start;
623 u8 *fault;
624 u8 *end;
625 int length;
626 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700627
628 if (!(s->flags & SLAB_POISON))
629 return 1;
630
Christoph Lametera973e9d2008-03-01 13:40:44 -0800631 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300632 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300633 end = start + length;
634 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700635 if (!remainder)
636 return 1;
637
Christoph Lameter39b26462008-04-14 19:11:30 +0300638 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700639 if (!fault)
640 return 1;
641 while (end > fault && end[-1] == POISON_INUSE)
642 end--;
643
644 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300645 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700646
647 restore_bytes(s, "slab padding", POISON_INUSE, start, end);
648 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700649}
650
651static int check_object(struct kmem_cache *s, struct page *page,
652 void *object, int active)
653{
654 u8 *p = object;
655 u8 *endobject = object + s->objsize;
656
657 if (s->flags & SLAB_RED_ZONE) {
658 unsigned int red =
659 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
660
Christoph Lameter24922682007-07-17 04:03:18 -0700661 if (!check_bytes_and_report(s, page, object, "Redzone",
662 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700663 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700664 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800665 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
666 check_bytes_and_report(s, page, p, "Alignment padding",
667 endobject, POISON_INUSE, s->inuse - s->objsize);
668 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700669 }
670
671 if (s->flags & SLAB_POISON) {
672 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700673 (!check_bytes_and_report(s, page, p, "Poison", p,
674 POISON_FREE, s->objsize - 1) ||
675 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800676 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700677 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700678 /*
679 * check_pad_bytes cleans up on its own.
680 */
681 check_pad_bytes(s, page, p);
682 }
683
684 if (!s->offset && active)
685 /*
686 * Object and freepointer overlap. Cannot check
687 * freepointer while object is allocated.
688 */
689 return 1;
690
691 /* Check free pointer validity */
692 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
693 object_err(s, page, p, "Freepointer corrupt");
694 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100695 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700696 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700697 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700698 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800699 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700700 return 0;
701 }
702 return 1;
703}
704
705static int check_slab(struct kmem_cache *s, struct page *page)
706{
Christoph Lameter39b26462008-04-14 19:11:30 +0300707 int maxobj;
708
Christoph Lameter81819f02007-05-06 14:49:36 -0700709 VM_BUG_ON(!irqs_disabled());
710
711 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700712 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700713 return 0;
714 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300715
716 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
717 if (page->objects > maxobj) {
718 slab_err(s, page, "objects %u > max %u",
719 s->name, page->objects, maxobj);
720 return 0;
721 }
722 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700723 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300724 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700725 return 0;
726 }
727 /* Slab_pad_check fixes things up after itself */
728 slab_pad_check(s, page);
729 return 1;
730}
731
732/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700733 * Determine if a certain object on a page is on the freelist. Must hold the
734 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700735 */
736static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
737{
738 int nr = 0;
739 void *fp = page->freelist;
740 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300741 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700742
Christoph Lameter39b26462008-04-14 19:11:30 +0300743 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700744 if (fp == search)
745 return 1;
746 if (!check_valid_pointer(s, page, fp)) {
747 if (object) {
748 object_err(s, page, object,
749 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800750 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700751 break;
752 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700753 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800754 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300755 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700756 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700757 return 0;
758 }
759 break;
760 }
761 object = fp;
762 fp = get_freepointer(s, object);
763 nr++;
764 }
765
Christoph Lameter224a88b2008-04-14 19:11:31 +0300766 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400767 if (max_objects > MAX_OBJS_PER_PAGE)
768 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300769
770 if (page->objects != max_objects) {
771 slab_err(s, page, "Wrong number of objects. Found %d but "
772 "should be %d", page->objects, max_objects);
773 page->objects = max_objects;
774 slab_fix(s, "Number of objects adjusted.");
775 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300776 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700777 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300778 "counted were %d", page->inuse, page->objects - nr);
779 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700780 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700781 }
782 return search == NULL;
783}
784
Christoph Lameter0121c6192008-04-29 16:11:12 -0700785static void trace(struct kmem_cache *s, struct page *page, void *object,
786 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700787{
788 if (s->flags & SLAB_TRACE) {
789 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
790 s->name,
791 alloc ? "alloc" : "free",
792 object, page->inuse,
793 page->freelist);
794
795 if (!alloc)
796 print_section("Object", (void *)object, s->objsize);
797
798 dump_stack();
799 }
800}
801
Christoph Lameter643b1132007-05-06 14:49:42 -0700802/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700803 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700804 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700805static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700806{
Christoph Lameter643b1132007-05-06 14:49:42 -0700807 spin_lock(&n->list_lock);
808 list_add(&page->lru, &n->full);
809 spin_unlock(&n->list_lock);
810}
811
812static void remove_full(struct kmem_cache *s, struct page *page)
813{
814 struct kmem_cache_node *n;
815
816 if (!(s->flags & SLAB_STORE_USER))
817 return;
818
819 n = get_node(s, page_to_nid(page));
820
821 spin_lock(&n->list_lock);
822 list_del(&page->lru);
823 spin_unlock(&n->list_lock);
824}
825
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300826/* Tracking of the number of slabs for debugging purposes */
827static inline unsigned long slabs_node(struct kmem_cache *s, int node)
828{
829 struct kmem_cache_node *n = get_node(s, node);
830
831 return atomic_long_read(&n->nr_slabs);
832}
833
Christoph Lameter205ab992008-04-14 19:11:40 +0300834static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300835{
836 struct kmem_cache_node *n = get_node(s, node);
837
838 /*
839 * May be called early in order to allocate a slab for the
840 * kmem_cache_node structure. Solve the chicken-egg
841 * dilemma by deferring the increment of the count during
842 * bootstrap (see early_kmem_cache_node_alloc).
843 */
Christoph Lameter205ab992008-04-14 19:11:40 +0300844 if (!NUMA_BUILD || n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300845 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300846 atomic_long_add(objects, &n->total_objects);
847 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300848}
Christoph Lameter205ab992008-04-14 19:11:40 +0300849static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300850{
851 struct kmem_cache_node *n = get_node(s, node);
852
853 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300854 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300855}
856
857/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700858static void setup_object_debug(struct kmem_cache *s, struct page *page,
859 void *object)
860{
861 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
862 return;
863
864 init_object(s, object, 0);
865 init_tracking(s, object);
866}
867
868static int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300869 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700870{
871 if (!check_slab(s, page))
872 goto bad;
873
Christoph Lameterd692ef62008-02-15 23:45:24 -0800874 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700875 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700876 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700877 }
878
879 if (!check_valid_pointer(s, page, object)) {
880 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700881 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700882 }
883
Christoph Lameterd692ef62008-02-15 23:45:24 -0800884 if (!check_object(s, page, object, 0))
Christoph Lameter81819f02007-05-06 14:49:36 -0700885 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700886
Christoph Lameter3ec09742007-05-16 22:11:00 -0700887 /* Success perform special debug activities for allocs */
888 if (s->flags & SLAB_STORE_USER)
889 set_track(s, object, TRACK_ALLOC, addr);
890 trace(s, page, object, 1);
891 init_object(s, object, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700892 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700893
Christoph Lameter81819f02007-05-06 14:49:36 -0700894bad:
895 if (PageSlab(page)) {
896 /*
897 * If this is a slab page then lets do the best we can
898 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700899 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700900 */
Christoph Lameter24922682007-07-17 04:03:18 -0700901 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300902 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800903 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700904 }
905 return 0;
906}
907
Christoph Lameter3ec09742007-05-16 22:11:00 -0700908static int free_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300909 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700910{
911 if (!check_slab(s, page))
912 goto fail;
913
914 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700915 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700916 goto fail;
917 }
918
919 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700920 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700921 goto fail;
922 }
923
924 if (!check_object(s, page, object, 1))
925 return 0;
926
927 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800928 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700929 slab_err(s, page, "Attempt to free object(0x%p) "
930 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800931 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700932 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700933 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700934 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700935 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800936 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700937 object_err(s, page, object,
938 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700939 goto fail;
940 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700941
942 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700943 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700944 remove_full(s, page);
945 if (s->flags & SLAB_STORE_USER)
946 set_track(s, object, TRACK_FREE, addr);
947 trace(s, page, object, 0);
948 init_object(s, object, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -0700949 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700950
Christoph Lameter81819f02007-05-06 14:49:36 -0700951fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700952 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700953 return 0;
954}
955
Christoph Lameter41ecc552007-05-09 02:32:44 -0700956static int __init setup_slub_debug(char *str)
957{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700958 slub_debug = DEBUG_DEFAULT_FLAGS;
959 if (*str++ != '=' || !*str)
960 /*
961 * No options specified. Switch on full debugging.
962 */
963 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700964
965 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700966 /*
967 * No options but restriction on slabs. This means full
968 * debugging for slabs matching a pattern.
969 */
970 goto check_slabs;
971
972 slub_debug = 0;
973 if (*str == '-')
974 /*
975 * Switch off all debugging measures.
976 */
977 goto out;
978
979 /*
980 * Determine which debug features should be switched on
981 */
Pekka Enberg06428782008-01-07 23:20:27 -0800982 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700983 switch (tolower(*str)) {
984 case 'f':
985 slub_debug |= SLAB_DEBUG_FREE;
986 break;
987 case 'z':
988 slub_debug |= SLAB_RED_ZONE;
989 break;
990 case 'p':
991 slub_debug |= SLAB_POISON;
992 break;
993 case 'u':
994 slub_debug |= SLAB_STORE_USER;
995 break;
996 case 't':
997 slub_debug |= SLAB_TRACE;
998 break;
999 default:
1000 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001001 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001002 }
1003 }
1004
1005check_slabs:
1006 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001007 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001008out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001009 return 1;
1010}
1011
1012__setup("slub_debug", setup_slub_debug);
1013
Christoph Lameterba0268a2007-09-11 15:24:11 -07001014static unsigned long kmem_cache_flags(unsigned long objsize,
1015 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001016 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001017{
1018 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001019 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001020 */
Christoph Lametere1533622008-02-15 23:45:24 -08001021 if (slub_debug && (!slub_debug_slabs ||
1022 strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0))
1023 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001024
1025 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001026}
1027#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001028static inline void setup_object_debug(struct kmem_cache *s,
1029 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001030
Christoph Lameter3ec09742007-05-16 22:11:00 -07001031static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001032 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001033
Christoph Lameter3ec09742007-05-16 22:11:00 -07001034static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001035 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001036
Christoph Lameter41ecc552007-05-09 02:32:44 -07001037static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1038 { return 1; }
1039static inline int check_object(struct kmem_cache *s, struct page *page,
1040 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001041static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001042static inline unsigned long kmem_cache_flags(unsigned long objsize,
1043 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001044 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001045{
1046 return flags;
1047}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001048#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001049
1050static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1051 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001052static inline void inc_slabs_node(struct kmem_cache *s, int node,
1053 int objects) {}
1054static inline void dec_slabs_node(struct kmem_cache *s, int node,
1055 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001056#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001057
Christoph Lameter81819f02007-05-06 14:49:36 -07001058/*
1059 * Slab allocation and freeing
1060 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001061static inline struct page *alloc_slab_page(gfp_t flags, int node,
1062 struct kmem_cache_order_objects oo)
1063{
1064 int order = oo_order(oo);
1065
1066 if (node == -1)
1067 return alloc_pages(flags, order);
1068 else
1069 return alloc_pages_node(node, flags, order);
1070}
1071
Christoph Lameter81819f02007-05-06 14:49:36 -07001072static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1073{
Pekka Enberg06428782008-01-07 23:20:27 -08001074 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001075 struct kmem_cache_order_objects oo = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07001076
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001077 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001078
Christoph Lameter65c33762008-04-14 19:11:40 +03001079 page = alloc_slab_page(flags | __GFP_NOWARN | __GFP_NORETRY, node,
1080 oo);
1081 if (unlikely(!page)) {
1082 oo = s->min;
1083 /*
1084 * Allocation may have failed due to fragmentation.
1085 * Try a lower order alloc if possible
1086 */
1087 page = alloc_slab_page(flags, node, oo);
1088 if (!page)
1089 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001090
Christoph Lameter65c33762008-04-14 19:11:40 +03001091 stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK);
1092 }
Christoph Lameter834f3d12008-04-14 19:11:31 +03001093 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001094 mod_zone_page_state(page_zone(page),
1095 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1096 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001097 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001098
1099 return page;
1100}
1101
1102static void setup_object(struct kmem_cache *s, struct page *page,
1103 void *object)
1104{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001105 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001106 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001107 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001108}
1109
1110static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1111{
1112 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001113 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001114 void *last;
1115 void *p;
1116
Christoph Lameter6cb06222007-10-16 01:25:41 -07001117 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001118
Christoph Lameter6cb06222007-10-16 01:25:41 -07001119 page = allocate_slab(s,
1120 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001121 if (!page)
1122 goto out;
1123
Christoph Lameter205ab992008-04-14 19:11:40 +03001124 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001125 page->slab = s;
1126 page->flags |= 1 << PG_slab;
1127 if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
1128 SLAB_STORE_USER | SLAB_TRACE))
Andy Whitcroft8a380822008-07-23 21:27:18 -07001129 __SetPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001130
1131 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001132
1133 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001134 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001135
1136 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001137 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001138 setup_object(s, page, last);
1139 set_freepointer(s, last, p);
1140 last = p;
1141 }
1142 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001143 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001144
1145 page->freelist = start;
1146 page->inuse = 0;
1147out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001148 return page;
1149}
1150
1151static void __free_slab(struct kmem_cache *s, struct page *page)
1152{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001153 int order = compound_order(page);
1154 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001155
Andy Whitcroft8a380822008-07-23 21:27:18 -07001156 if (unlikely(SLABDEBUG && PageSlubDebug(page))) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001157 void *p;
1158
1159 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001160 for_each_object(p, s, page_address(page),
1161 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001162 check_object(s, page, p, 0);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001163 __ClearPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001164 }
1165
1166 mod_zone_page_state(page_zone(page),
1167 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1168 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001169 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001170
Christoph Lameter49bd5222008-04-14 18:52:18 +03001171 __ClearPageSlab(page);
1172 reset_page_mapcount(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +03001173 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001174}
1175
1176static void rcu_free_slab(struct rcu_head *h)
1177{
1178 struct page *page;
1179
1180 page = container_of((struct list_head *)h, struct page, lru);
1181 __free_slab(page->slab, page);
1182}
1183
1184static void free_slab(struct kmem_cache *s, struct page *page)
1185{
1186 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1187 /*
1188 * RCU free overloads the RCU head over the LRU
1189 */
1190 struct rcu_head *head = (void *)&page->lru;
1191
1192 call_rcu(head, rcu_free_slab);
1193 } else
1194 __free_slab(s, page);
1195}
1196
1197static void discard_slab(struct kmem_cache *s, struct page *page)
1198{
Christoph Lameter205ab992008-04-14 19:11:40 +03001199 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001200 free_slab(s, page);
1201}
1202
1203/*
1204 * Per slab locking using the pagelock
1205 */
1206static __always_inline void slab_lock(struct page *page)
1207{
1208 bit_spin_lock(PG_locked, &page->flags);
1209}
1210
1211static __always_inline void slab_unlock(struct page *page)
1212{
Nick Piggina76d3542008-01-07 23:20:27 -08001213 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001214}
1215
1216static __always_inline int slab_trylock(struct page *page)
1217{
1218 int rc = 1;
1219
1220 rc = bit_spin_trylock(PG_locked, &page->flags);
1221 return rc;
1222}
1223
1224/*
1225 * Management of partially allocated slabs
1226 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001227static void add_partial(struct kmem_cache_node *n,
1228 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001229{
Christoph Lametere95eed52007-05-06 14:49:44 -07001230 spin_lock(&n->list_lock);
1231 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001232 if (tail)
1233 list_add_tail(&page->lru, &n->partial);
1234 else
1235 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001236 spin_unlock(&n->list_lock);
1237}
1238
Christoph Lameter0121c6192008-04-29 16:11:12 -07001239static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001240{
1241 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1242
1243 spin_lock(&n->list_lock);
1244 list_del(&page->lru);
1245 n->nr_partial--;
1246 spin_unlock(&n->list_lock);
1247}
1248
1249/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001250 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001251 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001252 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001253 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001254static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1255 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001256{
1257 if (slab_trylock(page)) {
1258 list_del(&page->lru);
1259 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001260 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001261 return 1;
1262 }
1263 return 0;
1264}
1265
1266/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001267 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001268 */
1269static struct page *get_partial_node(struct kmem_cache_node *n)
1270{
1271 struct page *page;
1272
1273 /*
1274 * Racy check. If we mistakenly see no partial slabs then we
1275 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001276 * partial slab and there is none available then get_partials()
1277 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001278 */
1279 if (!n || !n->nr_partial)
1280 return NULL;
1281
1282 spin_lock(&n->list_lock);
1283 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001284 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001285 goto out;
1286 page = NULL;
1287out:
1288 spin_unlock(&n->list_lock);
1289 return page;
1290}
1291
1292/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001293 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001294 */
1295static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1296{
1297#ifdef CONFIG_NUMA
1298 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001299 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001300 struct zone *zone;
1301 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001302 struct page *page;
1303
1304 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001305 * The defrag ratio allows a configuration of the tradeoffs between
1306 * inter node defragmentation and node local allocations. A lower
1307 * defrag_ratio increases the tendency to do local allocations
1308 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001309 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001310 * If the defrag_ratio is set to 0 then kmalloc() always
1311 * returns node local objects. If the ratio is higher then kmalloc()
1312 * may return off node objects because partial slabs are obtained
1313 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001314 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001315 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001316 * defrag_ratio = 1000) then every (well almost) allocation will
1317 * first attempt to defrag slab caches on other nodes. This means
1318 * scanning over all nodes to look for partial slabs which may be
1319 * expensive if we do it every time we are trying to find a slab
1320 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001321 */
Christoph Lameter98246012008-01-07 23:20:26 -08001322 if (!s->remote_node_defrag_ratio ||
1323 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001324 return NULL;
1325
Mel Gorman0e884602008-04-28 02:12:14 -07001326 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001327 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001328 struct kmem_cache_node *n;
1329
Mel Gorman54a6eb52008-04-28 02:12:16 -07001330 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001331
Mel Gorman54a6eb52008-04-28 02:12:16 -07001332 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001333 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001334 page = get_partial_node(n);
1335 if (page)
1336 return page;
1337 }
1338 }
1339#endif
1340 return NULL;
1341}
1342
1343/*
1344 * Get a partial page, lock it and return it.
1345 */
1346static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1347{
1348 struct page *page;
1349 int searchnode = (node == -1) ? numa_node_id() : node;
1350
1351 page = get_partial_node(get_node(s, searchnode));
1352 if (page || (flags & __GFP_THISNODE))
1353 return page;
1354
1355 return get_any_partial(s, flags);
1356}
1357
1358/*
1359 * Move a page back to the lists.
1360 *
1361 * Must be called with the slab lock held.
1362 *
1363 * On exit the slab lock will have been dropped.
1364 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001365static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001366{
Christoph Lametere95eed52007-05-06 14:49:44 -07001367 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001368 struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id());
Christoph Lametere95eed52007-05-06 14:49:44 -07001369
Andy Whitcroft8a380822008-07-23 21:27:18 -07001370 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001371 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001372
Christoph Lametera973e9d2008-03-01 13:40:44 -08001373 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001374 add_partial(n, page, tail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001375 stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1376 } else {
1377 stat(c, DEACTIVATE_FULL);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001378 if (SLABDEBUG && PageSlubDebug(page) &&
1379 (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001380 add_full(n, page);
1381 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001382 slab_unlock(page);
1383 } else {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001384 stat(c, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001385 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001386 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001387 * Adding an empty slab to the partial slabs in order
1388 * to avoid page allocator overhead. This slab needs
1389 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001390 * so that the others get filled first. That way the
1391 * size of the partial list stays small.
1392 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001393 * kmem_cache_shrink can reclaim any empty slabs from
1394 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001395 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001396 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001397 slab_unlock(page);
1398 } else {
1399 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001400 stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001401 discard_slab(s, page);
1402 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001403 }
1404}
1405
1406/*
1407 * Remove the cpu slab
1408 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001409static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001410{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001411 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001412 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001413
Christoph Lameterb773ad72008-03-04 11:10:17 -08001414 if (page->freelist)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001415 stat(c, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001416 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001417 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001418 * because both freelists are empty. So this is unlikely
1419 * to occur.
1420 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001421 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001422 void **object;
1423
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001424 tail = 0; /* Hot objects. Put the slab first */
1425
Christoph Lameter894b8782007-05-10 03:15:16 -07001426 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001427 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001428 c->freelist = c->freelist[c->offset];
Christoph Lameter894b8782007-05-10 03:15:16 -07001429
1430 /* And put onto the regular freelist */
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001431 object[c->offset] = page->freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07001432 page->freelist = object;
1433 page->inuse--;
1434 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001435 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001436 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001437}
1438
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001439static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001440{
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001441 stat(c, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001442 slab_lock(c->page);
1443 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001444}
1445
1446/*
1447 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001448 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001449 * Called from IPI handler with interrupts disabled.
1450 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001451static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001452{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001453 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001454
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001455 if (likely(c && c->page))
1456 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001457}
1458
1459static void flush_cpu_slab(void *d)
1460{
1461 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001462
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001463 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001464}
1465
1466static void flush_all(struct kmem_cache *s)
1467{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001468 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001469}
1470
1471/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001472 * Check if the objects in a per cpu structure fit numa
1473 * locality expectations.
1474 */
1475static inline int node_match(struct kmem_cache_cpu *c, int node)
1476{
1477#ifdef CONFIG_NUMA
1478 if (node != -1 && c->node != node)
1479 return 0;
1480#endif
1481 return 1;
1482}
1483
1484/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001485 * Slow path. The lockless freelist is empty or we need to perform
1486 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001487 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001488 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001489 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001490 * Processing is still very fast if new objects have been freed to the
1491 * regular freelist. In that case we simply take over the regular freelist
1492 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001493 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001494 * If that is not working then we fall back to the partial lists. We take the
1495 * first element of the freelist as the object to allocate now and move the
1496 * rest of the freelist to the lockless freelist.
1497 *
1498 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001499 * we need to allocate a new slab. This is the slowest path since it involves
1500 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001501 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001502static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1503 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001504{
Christoph Lameter81819f02007-05-06 14:49:36 -07001505 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001506 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001507
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001508 /* We handle __GFP_ZERO in the caller */
1509 gfpflags &= ~__GFP_ZERO;
1510
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001511 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001512 goto new_slab;
1513
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001514 slab_lock(c->page);
1515 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001516 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001517
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001518 stat(c, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001519
Christoph Lameter894b8782007-05-10 03:15:16 -07001520load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001521 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001522 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001523 goto another_slab;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001524 if (unlikely(SLABDEBUG && PageSlubDebug(c->page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001525 goto debug;
1526
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001527 c->freelist = object[c->offset];
Christoph Lameter39b26462008-04-14 19:11:30 +03001528 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001529 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001530 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001531unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001532 slab_unlock(c->page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001533 stat(c, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001534 return object;
1535
1536another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001537 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001538
1539new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001540 new = get_partial(s, gfpflags, node);
1541 if (new) {
1542 c->page = new;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001543 stat(c, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001544 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001545 }
1546
Christoph Lameterb811c202007-10-16 23:25:51 -07001547 if (gfpflags & __GFP_WAIT)
1548 local_irq_enable();
1549
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001550 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001551
1552 if (gfpflags & __GFP_WAIT)
1553 local_irq_disable();
1554
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001555 if (new) {
1556 c = get_cpu_slab(s, smp_processor_id());
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001557 stat(c, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001558 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001559 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001560 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001561 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001562 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001563 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001564 }
Christoph Lameter71c7a062008-02-14 14:28:01 -08001565 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001566debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001567 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001568 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001569
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001570 c->page->inuse++;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001571 c->page->freelist = object[c->offset];
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001572 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001573 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001574}
1575
1576/*
1577 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1578 * have the fastpath folded into their functions. So no function call
1579 * overhead for requests that can be satisfied on the fastpath.
1580 *
1581 * The fastpath works by first checking if the lockless freelist can be used.
1582 * If not then __slab_alloc is called for slow processing.
1583 *
1584 * Otherwise we can simply pick the next object from the lockless free list.
1585 */
Pekka Enberg06428782008-01-07 23:20:27 -08001586static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001587 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001588{
Christoph Lameter894b8782007-05-10 03:15:16 -07001589 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001590 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001591 unsigned long flags;
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001592 unsigned int objsize;
Christoph Lameter1f842602008-01-07 23:20:30 -08001593
Nick Piggincf40bd12009-01-21 08:12:39 +01001594 lockdep_trace_alloc(gfpflags);
OGAWA Hirofumi89124d72008-11-19 21:23:59 +09001595 might_sleep_if(gfpflags & __GFP_WAIT);
Pekka Enberg3c506ef2008-12-29 11:47:05 +02001596
Akinobu Mita773ff602008-12-23 19:37:01 +09001597 if (should_failslab(s->objsize, gfpflags))
1598 return NULL;
1599
Christoph Lameter894b8782007-05-10 03:15:16 -07001600 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001601 c = get_cpu_slab(s, smp_processor_id());
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001602 objsize = c->objsize;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001603 if (unlikely(!c->freelist || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001604
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001605 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001606
1607 else {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001608 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001609 c->freelist = object[c->offset];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001610 stat(c, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001611 }
1612 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001613
1614 if (unlikely((gfpflags & __GFP_ZERO) && object))
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001615 memset(object, 0, objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001616
Christoph Lameter894b8782007-05-10 03:15:16 -07001617 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001618}
1619
1620void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1621{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001622 void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_);
1623
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001624 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001625
1626 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001627}
1628EXPORT_SYMBOL(kmem_cache_alloc);
1629
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001630#ifdef CONFIG_KMEMTRACE
1631void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1632{
1633 return slab_alloc(s, gfpflags, -1, _RET_IP_);
1634}
1635EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1636#endif
1637
Christoph Lameter81819f02007-05-06 14:49:36 -07001638#ifdef CONFIG_NUMA
1639void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1640{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001641 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1642
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001643 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1644 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001645
1646 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001647}
1648EXPORT_SYMBOL(kmem_cache_alloc_node);
1649#endif
1650
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001651#ifdef CONFIG_KMEMTRACE
1652void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1653 gfp_t gfpflags,
1654 int node)
1655{
1656 return slab_alloc(s, gfpflags, node, _RET_IP_);
1657}
1658EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1659#endif
1660
Christoph Lameter81819f02007-05-06 14:49:36 -07001661/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001662 * Slow patch handling. This may still be called frequently since objects
1663 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001664 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001665 * So we still attempt to reduce cache line usage. Just take the slab
1666 * lock and free the item. If there is no additional partial page
1667 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001668 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001669static void __slab_free(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001670 void *x, unsigned long addr, unsigned int offset)
Christoph Lameter81819f02007-05-06 14:49:36 -07001671{
1672 void *prior;
1673 void **object = (void *)x;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001674 struct kmem_cache_cpu *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07001675
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001676 c = get_cpu_slab(s, raw_smp_processor_id());
1677 stat(c, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001678 slab_lock(page);
1679
Andy Whitcroft8a380822008-07-23 21:27:18 -07001680 if (unlikely(SLABDEBUG && PageSlubDebug(page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001681 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001682
Christoph Lameter81819f02007-05-06 14:49:36 -07001683checks_ok:
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001684 prior = object[offset] = page->freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001685 page->freelist = object;
1686 page->inuse--;
1687
Andy Whitcroft8a380822008-07-23 21:27:18 -07001688 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001689 stat(c, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001690 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001691 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001692
1693 if (unlikely(!page->inuse))
1694 goto slab_empty;
1695
1696 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001697 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001698 * then add it.
1699 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001700 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001701 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001702 stat(c, FREE_ADD_PARTIAL);
1703 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001704
1705out_unlock:
1706 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001707 return;
1708
1709slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001710 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001711 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001712 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001713 */
1714 remove_partial(s, page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001715 stat(c, FREE_REMOVE_PARTIAL);
1716 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001717 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001718 stat(c, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001719 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001720 return;
1721
1722debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001723 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001724 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001725 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001726}
1727
Christoph Lameter894b8782007-05-10 03:15:16 -07001728/*
1729 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1730 * can perform fastpath freeing without additional function calls.
1731 *
1732 * The fastpath is only possible if we are freeing to the current cpu slab
1733 * of this processor. This typically the case if we have just allocated
1734 * the item before.
1735 *
1736 * If fastpath is not possible then fall back to __slab_free where we deal
1737 * with all sorts of special processing.
1738 */
Pekka Enberg06428782008-01-07 23:20:27 -08001739static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001740 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001741{
1742 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001743 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001744 unsigned long flags;
1745
Christoph Lameter894b8782007-05-10 03:15:16 -07001746 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001747 c = get_cpu_slab(s, smp_processor_id());
Christoph Lameter27d9e4e2008-02-15 23:45:25 -08001748 debug_check_no_locks_freed(object, c->objsize);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001749 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Pekka Enberg6047a002009-01-14 12:22:25 +02001750 debug_check_no_obj_freed(object, c->objsize);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001751 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001752 object[c->offset] = c->freelist;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001753 c->freelist = object;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001754 stat(c, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001755 } else
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001756 __slab_free(s, page, x, addr, c->offset);
Christoph Lameter894b8782007-05-10 03:15:16 -07001757
1758 local_irq_restore(flags);
1759}
1760
Christoph Lameter81819f02007-05-06 14:49:36 -07001761void kmem_cache_free(struct kmem_cache *s, void *x)
1762{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001763 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001764
Christoph Lameterb49af682007-05-06 14:49:41 -07001765 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001766
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001767 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001768
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001769 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001770}
1771EXPORT_SYMBOL(kmem_cache_free);
1772
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001773/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001774static struct page *get_object_page(const void *x)
1775{
Christoph Lameterb49af682007-05-06 14:49:41 -07001776 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001777
1778 if (!PageSlab(page))
1779 return NULL;
1780
1781 return page;
1782}
1783
1784/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001785 * Object placement in a slab is made very easy because we always start at
1786 * offset 0. If we tune the size of the object to the alignment then we can
1787 * get the required alignment by putting one properly sized object after
1788 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001789 *
1790 * Notice that the allocation order determines the sizes of the per cpu
1791 * caches. Each processor has always one slab available for allocations.
1792 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001793 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001794 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001795 */
1796
1797/*
1798 * Mininum / Maximum order of slab pages. This influences locking overhead
1799 * and slab fragmentation. A higher order reduces the number of partial slabs
1800 * and increases the number of allocations possible without having to
1801 * take the list_lock.
1802 */
1803static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001804static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001805static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001806
1807/*
1808 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001809 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001810 */
1811static int slub_nomerge;
1812
1813/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001814 * Calculate the order of allocation given an slab object size.
1815 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001816 * The order of allocation has significant impact on performance and other
1817 * system components. Generally order 0 allocations should be preferred since
1818 * order 0 does not cause fragmentation in the page allocator. Larger objects
1819 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001820 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001821 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001822 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001823 * In order to reach satisfactory performance we must ensure that a minimum
1824 * number of objects is in one slab. Otherwise we may generate too much
1825 * activity on the partial lists which requires taking the list_lock. This is
1826 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001827 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001828 * slub_max_order specifies the order where we begin to stop considering the
1829 * number of objects in a slab as critical. If we reach slub_max_order then
1830 * we try to keep the page order as low as possible. So we accept more waste
1831 * of space in favor of a small page order.
1832 *
1833 * Higher order allocations also allow the placement of more objects in a
1834 * slab and thereby reduce object handling overhead. If the user has
1835 * requested a higher mininum order then we start with that one instead of
1836 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001837 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001838static inline int slab_order(int size, int min_objects,
1839 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001840{
1841 int order;
1842 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001843 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001844
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001845 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1846 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001847
Christoph Lameter6300ea72007-07-17 04:03:20 -07001848 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001849 fls(min_objects * size - 1) - PAGE_SHIFT);
1850 order <= max_order; order++) {
1851
Christoph Lameter81819f02007-05-06 14:49:36 -07001852 unsigned long slab_size = PAGE_SIZE << order;
1853
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001854 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001855 continue;
1856
Christoph Lameter81819f02007-05-06 14:49:36 -07001857 rem = slab_size % size;
1858
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001859 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001860 break;
1861
1862 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001863
Christoph Lameter81819f02007-05-06 14:49:36 -07001864 return order;
1865}
1866
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001867static inline int calculate_order(int size)
1868{
1869 int order;
1870 int min_objects;
1871 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001872 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001873
1874 /*
1875 * Attempt to find best configuration for a slab. This
1876 * works by first attempting to generate a layout with
1877 * the best configuration and backing off gradually.
1878 *
1879 * First we reduce the acceptable waste in a slab. Then
1880 * we reduce the minimum objects required in a slab.
1881 */
1882 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001883 if (!min_objects)
1884 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001885 max_objects = (PAGE_SIZE << slub_max_order)/size;
1886 min_objects = min(min_objects, max_objects);
1887
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001888 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001889 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001890 while (fraction >= 4) {
1891 order = slab_order(size, min_objects,
1892 slub_max_order, fraction);
1893 if (order <= slub_max_order)
1894 return order;
1895 fraction /= 2;
1896 }
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001897 min_objects --;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001898 }
1899
1900 /*
1901 * We were unable to place multiple objects in a slab. Now
1902 * lets see if we can place a single object there.
1903 */
1904 order = slab_order(size, 1, slub_max_order, 1);
1905 if (order <= slub_max_order)
1906 return order;
1907
1908 /*
1909 * Doh this slab cannot be placed using slub_max_order.
1910 */
1911 order = slab_order(size, 1, MAX_ORDER, 1);
1912 if (order <= MAX_ORDER)
1913 return order;
1914 return -ENOSYS;
1915}
1916
Christoph Lameter81819f02007-05-06 14:49:36 -07001917/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001918 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07001919 */
1920static unsigned long calculate_alignment(unsigned long flags,
1921 unsigned long align, unsigned long size)
1922{
1923 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001924 * If the user wants hardware cache aligned objects then follow that
1925 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07001926 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001927 * The hardware cache alignment cannot override the specified
1928 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07001929 */
Nick Pigginb6210382008-03-05 14:05:56 -08001930 if (flags & SLAB_HWCACHE_ALIGN) {
1931 unsigned long ralign = cache_line_size();
1932 while (size <= ralign / 2)
1933 ralign /= 2;
1934 align = max(align, ralign);
1935 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001936
1937 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08001938 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07001939
1940 return ALIGN(align, sizeof(void *));
1941}
1942
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001943static void init_kmem_cache_cpu(struct kmem_cache *s,
1944 struct kmem_cache_cpu *c)
1945{
1946 c->page = NULL;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001947 c->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001948 c->node = 0;
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07001949 c->offset = s->offset / sizeof(void *);
1950 c->objsize = s->objsize;
Pekka Enberg62f75532008-04-14 18:50:44 +03001951#ifdef CONFIG_SLUB_STATS
1952 memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned));
1953#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001954}
1955
Pekka Enberg5595cff2008-08-05 09:28:47 +03001956static void
1957init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07001958{
1959 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001960 spin_lock_init(&n->list_lock);
1961 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07001962#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001963 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07001964 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07001965 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07001966#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07001967}
1968
Christoph Lameter4c93c3552007-10-16 01:26:08 -07001969#ifdef CONFIG_SMP
1970/*
1971 * Per cpu array for per cpu structures.
1972 *
1973 * The per cpu array places all kmem_cache_cpu structures from one processor
1974 * close together meaning that it becomes possible that multiple per cpu
1975 * structures are contained in one cacheline. This may be particularly
1976 * beneficial for the kmalloc caches.
1977 *
1978 * A desktop system typically has around 60-80 slabs. With 100 here we are
1979 * likely able to get per cpu structures for all caches from the array defined
1980 * here. We must be able to cover all kmalloc caches during bootstrap.
1981 *
1982 * If the per cpu array is exhausted then fall back to kmalloc
1983 * of individual cachelines. No sharing is possible then.
1984 */
1985#define NR_KMEM_CACHE_CPU 100
1986
1987static DEFINE_PER_CPU(struct kmem_cache_cpu,
1988 kmem_cache_cpu)[NR_KMEM_CACHE_CPU];
1989
1990static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free);
Rusty Russell174596a2009-01-01 10:12:29 +10301991static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07001992
1993static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s,
1994 int cpu, gfp_t flags)
1995{
1996 struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu);
1997
1998 if (c)
1999 per_cpu(kmem_cache_cpu_free, cpu) =
2000 (void *)c->freelist;
2001 else {
2002 /* Table overflow: So allocate ourselves */
2003 c = kmalloc_node(
2004 ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()),
2005 flags, cpu_to_node(cpu));
2006 if (!c)
2007 return NULL;
2008 }
2009
2010 init_kmem_cache_cpu(s, c);
2011 return c;
2012}
2013
2014static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu)
2015{
2016 if (c < per_cpu(kmem_cache_cpu, cpu) ||
David Rientjes37189092009-01-27 18:59:46 -08002017 c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002018 kfree(c);
2019 return;
2020 }
2021 c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu);
2022 per_cpu(kmem_cache_cpu_free, cpu) = c;
2023}
2024
2025static void free_kmem_cache_cpus(struct kmem_cache *s)
2026{
2027 int cpu;
2028
2029 for_each_online_cpu(cpu) {
2030 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2031
2032 if (c) {
2033 s->cpu_slab[cpu] = NULL;
2034 free_kmem_cache_cpu(c, cpu);
2035 }
2036 }
2037}
2038
2039static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2040{
2041 int cpu;
2042
2043 for_each_online_cpu(cpu) {
2044 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2045
2046 if (c)
2047 continue;
2048
2049 c = alloc_kmem_cache_cpu(s, cpu, flags);
2050 if (!c) {
2051 free_kmem_cache_cpus(s);
2052 return 0;
2053 }
2054 s->cpu_slab[cpu] = c;
2055 }
2056 return 1;
2057}
2058
2059/*
2060 * Initialize the per cpu array.
2061 */
2062static void init_alloc_cpu_cpu(int cpu)
2063{
2064 int i;
2065
Rusty Russell174596a2009-01-01 10:12:29 +10302066 if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)))
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002067 return;
2068
2069 for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--)
2070 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu);
2071
Rusty Russell174596a2009-01-01 10:12:29 +10302072 cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once));
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002073}
2074
2075static void __init init_alloc_cpu(void)
2076{
2077 int cpu;
2078
2079 for_each_online_cpu(cpu)
2080 init_alloc_cpu_cpu(cpu);
2081 }
2082
2083#else
2084static inline void free_kmem_cache_cpus(struct kmem_cache *s) {}
2085static inline void init_alloc_cpu(void) {}
2086
2087static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2088{
2089 init_kmem_cache_cpu(s, &s->cpu_slab);
2090 return 1;
2091}
2092#endif
2093
Christoph Lameter81819f02007-05-06 14:49:36 -07002094#ifdef CONFIG_NUMA
2095/*
2096 * No kmalloc_node yet so do it by hand. We know that this is the first
2097 * slab on the node for this slabcache. There are no concurrent accesses
2098 * possible.
2099 *
2100 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002101 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2102 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002103 */
David Rientjes0094de92008-11-25 19:14:19 -08002104static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002105{
2106 struct page *page;
2107 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002108 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002109
2110 BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2111
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002112 page = new_slab(kmalloc_caches, gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002113
2114 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002115 if (page_to_nid(page) != node) {
2116 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2117 "node %d\n", node);
2118 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2119 "in order to be able to continue\n");
2120 }
2121
Christoph Lameter81819f02007-05-06 14:49:36 -07002122 n = page->freelist;
2123 BUG_ON(!n);
2124 page->freelist = get_freepointer(kmalloc_caches, n);
2125 page->inuse++;
2126 kmalloc_caches->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002127#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterd45f39c2007-07-17 04:03:21 -07002128 init_object(kmalloc_caches, n, 1);
2129 init_tracking(kmalloc_caches, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002130#endif
Pekka Enberg5595cff2008-08-05 09:28:47 +03002131 init_kmem_cache_node(n, kmalloc_caches);
Christoph Lameter205ab992008-04-14 19:11:40 +03002132 inc_slabs_node(kmalloc_caches, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002133
rootba84c732008-01-07 23:20:28 -08002134 /*
2135 * lockdep requires consistent irq usage for each lock
2136 * so even though there cannot be a race this early in
2137 * the boot sequence, we still disable irqs.
2138 */
2139 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002140 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002141 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002142}
2143
2144static void free_kmem_cache_nodes(struct kmem_cache *s)
2145{
2146 int node;
2147
Christoph Lameterf64dc582007-10-16 01:25:33 -07002148 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002149 struct kmem_cache_node *n = s->node[node];
2150 if (n && n != &s->local_node)
2151 kmem_cache_free(kmalloc_caches, n);
2152 s->node[node] = NULL;
2153 }
2154}
2155
2156static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2157{
2158 int node;
2159 int local_node;
2160
2161 if (slab_state >= UP)
2162 local_node = page_to_nid(virt_to_page(s));
2163 else
2164 local_node = 0;
2165
Christoph Lameterf64dc582007-10-16 01:25:33 -07002166 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002167 struct kmem_cache_node *n;
2168
2169 if (local_node == node)
2170 n = &s->local_node;
2171 else {
2172 if (slab_state == DOWN) {
David Rientjes0094de92008-11-25 19:14:19 -08002173 early_kmem_cache_node_alloc(gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002174 continue;
2175 }
2176 n = kmem_cache_alloc_node(kmalloc_caches,
2177 gfpflags, node);
2178
2179 if (!n) {
2180 free_kmem_cache_nodes(s);
2181 return 0;
2182 }
2183
2184 }
2185 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002186 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002187 }
2188 return 1;
2189}
2190#else
2191static void free_kmem_cache_nodes(struct kmem_cache *s)
2192{
2193}
2194
2195static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2196{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002197 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002198 return 1;
2199}
2200#endif
2201
David Rientjesc0bdb232009-02-25 09:16:35 +02002202static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002203{
2204 if (min < MIN_PARTIAL)
2205 min = MIN_PARTIAL;
2206 else if (min > MAX_PARTIAL)
2207 min = MAX_PARTIAL;
2208 s->min_partial = min;
2209}
2210
Christoph Lameter81819f02007-05-06 14:49:36 -07002211/*
2212 * calculate_sizes() determines the order and the distribution of data within
2213 * a slab object.
2214 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002215static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002216{
2217 unsigned long flags = s->flags;
2218 unsigned long size = s->objsize;
2219 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002220 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002221
2222 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002223 * Round up object size to the next word boundary. We can only
2224 * place the free pointer at word boundaries and this determines
2225 * the possible location of the free pointer.
2226 */
2227 size = ALIGN(size, sizeof(void *));
2228
2229#ifdef CONFIG_SLUB_DEBUG
2230 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002231 * Determine if we can poison the object itself. If the user of
2232 * the slab may touch the object after free or before allocation
2233 * then we should never poison the object itself.
2234 */
2235 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002236 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002237 s->flags |= __OBJECT_POISON;
2238 else
2239 s->flags &= ~__OBJECT_POISON;
2240
Christoph Lameter81819f02007-05-06 14:49:36 -07002241
2242 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002243 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002244 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002245 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002246 */
2247 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2248 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002249#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002250
2251 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002252 * With that we have determined the number of bytes in actual use
2253 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002254 */
2255 s->inuse = size;
2256
2257 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002258 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002259 /*
2260 * Relocate free pointer after the object if it is not
2261 * permitted to overwrite the first word of the object on
2262 * kmem_cache_free.
2263 *
2264 * This is the case if we do RCU, have a constructor or
2265 * destructor or are poisoning the objects.
2266 */
2267 s->offset = size;
2268 size += sizeof(void *);
2269 }
2270
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002271#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002272 if (flags & SLAB_STORE_USER)
2273 /*
2274 * Need to store information about allocs and frees after
2275 * the object.
2276 */
2277 size += 2 * sizeof(struct track);
2278
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002279 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002280 /*
2281 * Add some empty padding so that we can catch
2282 * overwrites from earlier objects rather than let
2283 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002284 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002285 * of the object.
2286 */
2287 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002288#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002289
Christoph Lameter81819f02007-05-06 14:49:36 -07002290 /*
2291 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002292 * user specified and the dynamic determination of cache line size
2293 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002294 */
2295 align = calculate_alignment(flags, align, s->objsize);
2296
2297 /*
2298 * SLUB stores one object immediately after another beginning from
2299 * offset 0. In order to align the objects we have to simply size
2300 * each object to conform to the alignment.
2301 */
2302 size = ALIGN(size, align);
2303 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002304 if (forced_order >= 0)
2305 order = forced_order;
2306 else
2307 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002308
Christoph Lameter834f3d12008-04-14 19:11:31 +03002309 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002310 return 0;
2311
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002312 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002313 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002314 s->allocflags |= __GFP_COMP;
2315
2316 if (s->flags & SLAB_CACHE_DMA)
2317 s->allocflags |= SLUB_DMA;
2318
2319 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2320 s->allocflags |= __GFP_RECLAIMABLE;
2321
Christoph Lameter81819f02007-05-06 14:49:36 -07002322 /*
2323 * Determine the number of objects per slab
2324 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002325 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002326 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002327 if (oo_objects(s->oo) > oo_objects(s->max))
2328 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002329
Christoph Lameter834f3d12008-04-14 19:11:31 +03002330 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002331
2332}
2333
Christoph Lameter81819f02007-05-06 14:49:36 -07002334static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2335 const char *name, size_t size,
2336 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002337 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002338{
2339 memset(s, 0, kmem_size);
2340 s->name = name;
2341 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002342 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002343 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002344 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002345
Christoph Lameter06b285d2008-04-14 19:11:41 +03002346 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002347 goto error;
2348
David Rientjes3b89d7d2009-02-22 17:40:07 -08002349 /*
2350 * The larger the object size is, the more pages we want on the partial
2351 * list to avoid pounding the page allocator excessively.
2352 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002353 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002354 s->refcount = 1;
2355#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002356 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002357#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002358 if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2359 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002360
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002361 if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
Christoph Lameter81819f02007-05-06 14:49:36 -07002362 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002363 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002364error:
2365 if (flags & SLAB_PANIC)
2366 panic("Cannot create slab %s size=%lu realsize=%u "
2367 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002368 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002369 s->offset, flags);
2370 return 0;
2371}
Christoph Lameter81819f02007-05-06 14:49:36 -07002372
2373/*
2374 * Check if a given pointer is valid
2375 */
2376int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2377{
Pekka Enberg06428782008-01-07 23:20:27 -08002378 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002379
2380 page = get_object_page(object);
2381
2382 if (!page || s != page->slab)
2383 /* No slab or wrong slab */
2384 return 0;
2385
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002386 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002387 return 0;
2388
2389 /*
2390 * We could also check if the object is on the slabs freelist.
2391 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002392 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002393 * to a certain slab.
2394 */
2395 return 1;
2396}
2397EXPORT_SYMBOL(kmem_ptr_validate);
2398
2399/*
2400 * Determine the size of a slab object
2401 */
2402unsigned int kmem_cache_size(struct kmem_cache *s)
2403{
2404 return s->objsize;
2405}
2406EXPORT_SYMBOL(kmem_cache_size);
2407
2408const char *kmem_cache_name(struct kmem_cache *s)
2409{
2410 return s->name;
2411}
2412EXPORT_SYMBOL(kmem_cache_name);
2413
Christoph Lameter33b12c32008-04-25 12:22:43 -07002414static void list_slab_objects(struct kmem_cache *s, struct page *page,
2415 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002416{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002417#ifdef CONFIG_SLUB_DEBUG
2418 void *addr = page_address(page);
2419 void *p;
2420 DECLARE_BITMAP(map, page->objects);
2421
2422 bitmap_zero(map, page->objects);
2423 slab_err(s, page, "%s", text);
2424 slab_lock(page);
2425 for_each_free_object(p, s, page->freelist)
2426 set_bit(slab_index(p, s, addr), map);
2427
2428 for_each_object(p, s, addr, page->objects) {
2429
2430 if (!test_bit(slab_index(p, s, addr), map)) {
2431 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2432 p, p - addr);
2433 print_tracking(s, p);
2434 }
2435 }
2436 slab_unlock(page);
2437#endif
2438}
2439
Christoph Lameter81819f02007-05-06 14:49:36 -07002440/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002441 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002442 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002443static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002444{
Christoph Lameter81819f02007-05-06 14:49:36 -07002445 unsigned long flags;
2446 struct page *page, *h;
2447
2448 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002449 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002450 if (!page->inuse) {
2451 list_del(&page->lru);
2452 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002453 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002454 } else {
2455 list_slab_objects(s, page,
2456 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002457 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002458 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002459 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002460}
2461
2462/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002463 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002464 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002465static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002466{
2467 int node;
2468
2469 flush_all(s);
2470
2471 /* Attempt to free all objects */
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002472 free_kmem_cache_cpus(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002473 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002474 struct kmem_cache_node *n = get_node(s, node);
2475
Christoph Lameter599870b2008-04-23 12:36:52 -07002476 free_partial(s, n);
2477 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002478 return 1;
2479 }
2480 free_kmem_cache_nodes(s);
2481 return 0;
2482}
2483
2484/*
2485 * Close a cache and release the kmem_cache structure
2486 * (must be used for caches created using kmem_cache_create)
2487 */
2488void kmem_cache_destroy(struct kmem_cache *s)
2489{
2490 down_write(&slub_lock);
2491 s->refcount--;
2492 if (!s->refcount) {
2493 list_del(&s->list);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002494 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03002495 if (kmem_cache_close(s)) {
2496 printk(KERN_ERR "SLUB %s: %s called for cache that "
2497 "still has objects.\n", s->name, __func__);
2498 dump_stack();
2499 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002500 sysfs_slab_remove(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002501 } else
2502 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002503}
2504EXPORT_SYMBOL(kmem_cache_destroy);
2505
2506/********************************************************************
2507 * Kmalloc subsystem
2508 *******************************************************************/
2509
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002510struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned;
Christoph Lameter81819f02007-05-06 14:49:36 -07002511EXPORT_SYMBOL(kmalloc_caches);
2512
Christoph Lameter81819f02007-05-06 14:49:36 -07002513static int __init setup_slub_min_order(char *str)
2514{
Pekka Enberg06428782008-01-07 23:20:27 -08002515 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002516
2517 return 1;
2518}
2519
2520__setup("slub_min_order=", setup_slub_min_order);
2521
2522static int __init setup_slub_max_order(char *str)
2523{
Pekka Enberg06428782008-01-07 23:20:27 -08002524 get_option(&str, &slub_max_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002525
2526 return 1;
2527}
2528
2529__setup("slub_max_order=", setup_slub_max_order);
2530
2531static int __init setup_slub_min_objects(char *str)
2532{
Pekka Enberg06428782008-01-07 23:20:27 -08002533 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002534
2535 return 1;
2536}
2537
2538__setup("slub_min_objects=", setup_slub_min_objects);
2539
2540static int __init setup_slub_nomerge(char *str)
2541{
2542 slub_nomerge = 1;
2543 return 1;
2544}
2545
2546__setup("slub_nomerge", setup_slub_nomerge);
2547
Christoph Lameter81819f02007-05-06 14:49:36 -07002548static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2549 const char *name, int size, gfp_t gfp_flags)
2550{
2551 unsigned int flags = 0;
2552
2553 if (gfp_flags & SLUB_DMA)
2554 flags = SLAB_CACHE_DMA;
2555
2556 down_write(&slub_lock);
2557 if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002558 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002559 goto panic;
2560
2561 list_add(&s->list, &slab_caches);
2562 up_write(&slub_lock);
2563 if (sysfs_slab_add(s))
2564 goto panic;
2565 return s;
2566
2567panic:
2568 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2569}
2570
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002571#ifdef CONFIG_ZONE_DMA
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002572static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
Christoph Lameter1ceef402007-08-07 15:11:48 -07002573
2574static void sysfs_add_func(struct work_struct *w)
2575{
2576 struct kmem_cache *s;
2577
2578 down_write(&slub_lock);
2579 list_for_each_entry(s, &slab_caches, list) {
2580 if (s->flags & __SYSFS_ADD_DEFERRED) {
2581 s->flags &= ~__SYSFS_ADD_DEFERRED;
2582 sysfs_slab_add(s);
2583 }
2584 }
2585 up_write(&slub_lock);
2586}
2587
2588static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2589
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002590static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2591{
2592 struct kmem_cache *s;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002593 char *text;
2594 size_t realsize;
2595
2596 s = kmalloc_caches_dma[index];
2597 if (s)
2598 return s;
2599
2600 /* Dynamically create dma cache */
Christoph Lameter1ceef402007-08-07 15:11:48 -07002601 if (flags & __GFP_WAIT)
2602 down_write(&slub_lock);
2603 else {
2604 if (!down_write_trylock(&slub_lock))
2605 goto out;
2606 }
2607
2608 if (kmalloc_caches_dma[index])
2609 goto unlock_out;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002610
Christoph Lameter7b55f622007-07-17 04:03:27 -07002611 realsize = kmalloc_caches[index].objsize;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08002612 text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2613 (unsigned int)realsize);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002614 s = kmalloc(kmem_size, flags & ~SLUB_DMA);
2615
2616 if (!s || !text || !kmem_cache_open(s, flags, text,
2617 realsize, ARCH_KMALLOC_MINALIGN,
2618 SLAB_CACHE_DMA|__SYSFS_ADD_DEFERRED, NULL)) {
2619 kfree(s);
2620 kfree(text);
2621 goto unlock_out;
Christoph Lameterdfce8642007-07-17 04:03:25 -07002622 }
Christoph Lameter1ceef402007-08-07 15:11:48 -07002623
2624 list_add(&s->list, &slab_caches);
2625 kmalloc_caches_dma[index] = s;
2626
2627 schedule_work(&sysfs_add_work);
2628
2629unlock_out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002630 up_write(&slub_lock);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002631out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002632 return kmalloc_caches_dma[index];
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002633}
2634#endif
2635
Christoph Lameterf1b26332007-07-17 04:03:26 -07002636/*
2637 * Conversion table for small slabs sizes / 8 to the index in the
2638 * kmalloc array. This is necessary for slabs < 192 since we have non power
2639 * of two cache sizes there. The size of larger slabs can be determined using
2640 * fls.
2641 */
2642static s8 size_index[24] = {
2643 3, /* 8 */
2644 4, /* 16 */
2645 5, /* 24 */
2646 5, /* 32 */
2647 6, /* 40 */
2648 6, /* 48 */
2649 6, /* 56 */
2650 6, /* 64 */
2651 1, /* 72 */
2652 1, /* 80 */
2653 1, /* 88 */
2654 1, /* 96 */
2655 7, /* 104 */
2656 7, /* 112 */
2657 7, /* 120 */
2658 7, /* 128 */
2659 2, /* 136 */
2660 2, /* 144 */
2661 2, /* 152 */
2662 2, /* 160 */
2663 2, /* 168 */
2664 2, /* 176 */
2665 2, /* 184 */
2666 2 /* 192 */
2667};
2668
Christoph Lameter81819f02007-05-06 14:49:36 -07002669static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2670{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002671 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002672
Christoph Lameterf1b26332007-07-17 04:03:26 -07002673 if (size <= 192) {
2674 if (!size)
2675 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002676
Christoph Lameterf1b26332007-07-17 04:03:26 -07002677 index = size_index[(size - 1) / 8];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002678 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002679 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002680
2681#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002682 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002683 return dma_kmalloc_cache(index, flags);
Christoph Lameterf1b26332007-07-17 04:03:26 -07002684
Christoph Lameter81819f02007-05-06 14:49:36 -07002685#endif
2686 return &kmalloc_caches[index];
2687}
2688
2689void *__kmalloc(size_t size, gfp_t flags)
2690{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002691 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002692 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002693
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002694 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002695 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002696
2697 s = get_slab(size, flags);
2698
2699 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002700 return s;
2701
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002702 ret = slab_alloc(s, flags, -1, _RET_IP_);
2703
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002704 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002705
2706 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002707}
2708EXPORT_SYMBOL(__kmalloc);
2709
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002710static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2711{
2712 struct page *page = alloc_pages_node(node, flags | __GFP_COMP,
2713 get_order(size));
2714
2715 if (page)
2716 return page_address(page);
2717 else
2718 return NULL;
2719}
2720
Christoph Lameter81819f02007-05-06 14:49:36 -07002721#ifdef CONFIG_NUMA
2722void *__kmalloc_node(size_t size, gfp_t flags, int node)
2723{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002724 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002725 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002726
Ingo Molnar057685c2009-02-20 12:15:30 +01002727 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002728 ret = kmalloc_large_node(size, flags, node);
2729
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002730 trace_kmalloc_node(_RET_IP_, ret,
2731 size, PAGE_SIZE << get_order(size),
2732 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002733
2734 return ret;
2735 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002736
2737 s = get_slab(size, flags);
2738
2739 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002740 return s;
2741
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002742 ret = slab_alloc(s, flags, node, _RET_IP_);
2743
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002744 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002745
2746 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002747}
2748EXPORT_SYMBOL(__kmalloc_node);
2749#endif
2750
2751size_t ksize(const void *object)
2752{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002753 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002754 struct kmem_cache *s;
2755
Christoph Lameteref8b4522007-10-16 01:24:46 -07002756 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002757 return 0;
2758
Vegard Nossum294a80a2007-12-04 23:45:30 -08002759 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002760
Pekka Enberg76994412008-05-22 19:22:25 +03002761 if (unlikely(!PageSlab(page))) {
2762 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002763 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002764 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002765 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002766
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002767#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002768 /*
2769 * Debugging requires use of the padding between object
2770 * and whatever may come after it.
2771 */
2772 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2773 return s->objsize;
2774
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002775#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002776 /*
2777 * If we have the need to store the freelist pointer
2778 * back there or track user information then we can
2779 * only use the space before that information.
2780 */
2781 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2782 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002783 /*
2784 * Else we can use all the padding etc for the allocation
2785 */
2786 return s->size;
2787}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002788EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002789
2790void kfree(const void *x)
2791{
Christoph Lameter81819f02007-05-06 14:49:36 -07002792 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002793 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002794
Pekka Enberg2121db72009-03-25 11:05:57 +02002795 trace_kfree(_RET_IP_, x);
2796
Satyam Sharma2408c552007-10-16 01:24:44 -07002797 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002798 return;
2799
Christoph Lameterb49af682007-05-06 14:49:41 -07002800 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002801 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002802 BUG_ON(!PageCompound(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002803 put_page(page);
2804 return;
2805 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002806 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002807}
2808EXPORT_SYMBOL(kfree);
2809
Christoph Lameter2086d262007-05-06 14:49:46 -07002810/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002811 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2812 * the remaining slabs by the number of items in use. The slabs with the
2813 * most items in use come first. New allocations will then fill those up
2814 * and thus they can be removed from the partial lists.
2815 *
2816 * The slabs with the least items are placed last. This results in them
2817 * being allocated from last increasing the chance that the last objects
2818 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002819 */
2820int kmem_cache_shrink(struct kmem_cache *s)
2821{
2822 int node;
2823 int i;
2824 struct kmem_cache_node *n;
2825 struct page *page;
2826 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002827 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002828 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002829 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002830 unsigned long flags;
2831
2832 if (!slabs_by_inuse)
2833 return -ENOMEM;
2834
2835 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002836 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002837 n = get_node(s, node);
2838
2839 if (!n->nr_partial)
2840 continue;
2841
Christoph Lameter834f3d12008-04-14 19:11:31 +03002842 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002843 INIT_LIST_HEAD(slabs_by_inuse + i);
2844
2845 spin_lock_irqsave(&n->list_lock, flags);
2846
2847 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002848 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002849 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002850 * Note that concurrent frees may occur while we hold the
2851 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002852 */
2853 list_for_each_entry_safe(page, t, &n->partial, lru) {
2854 if (!page->inuse && slab_trylock(page)) {
2855 /*
2856 * Must hold slab lock here because slab_free
2857 * may have freed the last object and be
2858 * waiting to release the slab.
2859 */
2860 list_del(&page->lru);
2861 n->nr_partial--;
2862 slab_unlock(page);
2863 discard_slab(s, page);
2864 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002865 list_move(&page->lru,
2866 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002867 }
2868 }
2869
Christoph Lameter2086d262007-05-06 14:49:46 -07002870 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002871 * Rebuild the partial list with the slabs filled up most
2872 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002873 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002874 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002875 list_splice(slabs_by_inuse + i, n->partial.prev);
2876
Christoph Lameter2086d262007-05-06 14:49:46 -07002877 spin_unlock_irqrestore(&n->list_lock, flags);
2878 }
2879
2880 kfree(slabs_by_inuse);
2881 return 0;
2882}
2883EXPORT_SYMBOL(kmem_cache_shrink);
2884
Yasunori Gotob9049e22007-10-21 16:41:37 -07002885#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2886static int slab_mem_going_offline_callback(void *arg)
2887{
2888 struct kmem_cache *s;
2889
2890 down_read(&slub_lock);
2891 list_for_each_entry(s, &slab_caches, list)
2892 kmem_cache_shrink(s);
2893 up_read(&slub_lock);
2894
2895 return 0;
2896}
2897
2898static void slab_mem_offline_callback(void *arg)
2899{
2900 struct kmem_cache_node *n;
2901 struct kmem_cache *s;
2902 struct memory_notify *marg = arg;
2903 int offline_node;
2904
2905 offline_node = marg->status_change_nid;
2906
2907 /*
2908 * If the node still has available memory. we need kmem_cache_node
2909 * for it yet.
2910 */
2911 if (offline_node < 0)
2912 return;
2913
2914 down_read(&slub_lock);
2915 list_for_each_entry(s, &slab_caches, list) {
2916 n = get_node(s, offline_node);
2917 if (n) {
2918 /*
2919 * if n->nr_slabs > 0, slabs still exist on the node
2920 * that is going down. We were unable to free them,
2921 * and offline_pages() function shoudn't call this
2922 * callback. So, we must fail.
2923 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002924 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002925
2926 s->node[offline_node] = NULL;
2927 kmem_cache_free(kmalloc_caches, n);
2928 }
2929 }
2930 up_read(&slub_lock);
2931}
2932
2933static int slab_mem_going_online_callback(void *arg)
2934{
2935 struct kmem_cache_node *n;
2936 struct kmem_cache *s;
2937 struct memory_notify *marg = arg;
2938 int nid = marg->status_change_nid;
2939 int ret = 0;
2940
2941 /*
2942 * If the node's memory is already available, then kmem_cache_node is
2943 * already created. Nothing to do.
2944 */
2945 if (nid < 0)
2946 return 0;
2947
2948 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002949 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002950 * allocate a kmem_cache_node structure in order to bring the node
2951 * online.
2952 */
2953 down_read(&slub_lock);
2954 list_for_each_entry(s, &slab_caches, list) {
2955 /*
2956 * XXX: kmem_cache_alloc_node will fallback to other nodes
2957 * since memory is not yet available from the node that
2958 * is brought up.
2959 */
2960 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
2961 if (!n) {
2962 ret = -ENOMEM;
2963 goto out;
2964 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03002965 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002966 s->node[nid] = n;
2967 }
2968out:
2969 up_read(&slub_lock);
2970 return ret;
2971}
2972
2973static int slab_memory_callback(struct notifier_block *self,
2974 unsigned long action, void *arg)
2975{
2976 int ret = 0;
2977
2978 switch (action) {
2979 case MEM_GOING_ONLINE:
2980 ret = slab_mem_going_online_callback(arg);
2981 break;
2982 case MEM_GOING_OFFLINE:
2983 ret = slab_mem_going_offline_callback(arg);
2984 break;
2985 case MEM_OFFLINE:
2986 case MEM_CANCEL_ONLINE:
2987 slab_mem_offline_callback(arg);
2988 break;
2989 case MEM_ONLINE:
2990 case MEM_CANCEL_OFFLINE:
2991 break;
2992 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08002993 if (ret)
2994 ret = notifier_from_errno(ret);
2995 else
2996 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07002997 return ret;
2998}
2999
3000#endif /* CONFIG_MEMORY_HOTPLUG */
3001
Christoph Lameter81819f02007-05-06 14:49:36 -07003002/********************************************************************
3003 * Basic setup of slabs
3004 *******************************************************************/
3005
3006void __init kmem_cache_init(void)
3007{
3008 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003009 int caches = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003010
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003011 init_alloc_cpu();
3012
Christoph Lameter81819f02007-05-06 14:49:36 -07003013#ifdef CONFIG_NUMA
3014 /*
3015 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003016 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003017 * kmem_cache_open for slab_state == DOWN.
3018 */
3019 create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
3020 sizeof(struct kmem_cache_node), GFP_KERNEL);
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003021 kmalloc_caches[0].refcount = -1;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003022 caches++;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003023
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003024 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003025#endif
3026
3027 /* Able to allocate the per node structures */
3028 slab_state = PARTIAL;
3029
3030 /* Caches that are not of the two-to-the-power-of size */
Christoph Lameter4b356be2007-06-16 10:16:13 -07003031 if (KMALLOC_MIN_SIZE <= 64) {
3032 create_kmalloc_cache(&kmalloc_caches[1],
Christoph Lameter81819f02007-05-06 14:49:36 -07003033 "kmalloc-96", 96, GFP_KERNEL);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003034 caches++;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003035 create_kmalloc_cache(&kmalloc_caches[2],
Christoph Lameter81819f02007-05-06 14:49:36 -07003036 "kmalloc-192", 192, GFP_KERNEL);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003037 caches++;
3038 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003039
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003040 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003041 create_kmalloc_cache(&kmalloc_caches[i],
3042 "kmalloc", 1 << i, GFP_KERNEL);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003043 caches++;
3044 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003045
Christoph Lameterf1b26332007-07-17 04:03:26 -07003046
3047 /*
3048 * Patch up the size_index table if we have strange large alignment
3049 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003050 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003051 *
3052 * Largest permitted alignment is 256 bytes due to the way we
3053 * handle the index determination for the smaller caches.
3054 *
3055 * Make sure that nothing crazy happens if someone starts tinkering
3056 * around with ARCH_KMALLOC_MINALIGN
3057 */
3058 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3059 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3060
Christoph Lameter12ad6842007-07-17 04:03:28 -07003061 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8)
Christoph Lameterf1b26332007-07-17 04:03:26 -07003062 size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW;
3063
Christoph Lameter41d54d32008-07-03 09:14:26 -05003064 if (KMALLOC_MIN_SIZE == 128) {
3065 /*
3066 * The 192 byte sized cache is not used if the alignment
3067 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3068 * instead.
3069 */
3070 for (i = 128 + 8; i <= 192; i += 8)
3071 size_index[(i - 1) / 8] = 8;
3072 }
3073
Christoph Lameter81819f02007-05-06 14:49:36 -07003074 slab_state = UP;
3075
3076 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003077 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++)
Christoph Lameter81819f02007-05-06 14:49:36 -07003078 kmalloc_caches[i]. name =
3079 kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i);
3080
3081#ifdef CONFIG_SMP
3082 register_cpu_notifier(&slab_notifier);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003083 kmem_size = offsetof(struct kmem_cache, cpu_slab) +
3084 nr_cpu_ids * sizeof(struct kmem_cache_cpu *);
3085#else
3086 kmem_size = sizeof(struct kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003087#endif
3088
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003089 printk(KERN_INFO
3090 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003091 " CPUs=%d, Nodes=%d\n",
3092 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003093 slub_min_order, slub_max_order, slub_min_objects,
3094 nr_cpu_ids, nr_node_ids);
3095}
3096
3097/*
3098 * Find a mergeable slab cache
3099 */
3100static int slab_unmergeable(struct kmem_cache *s)
3101{
3102 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3103 return 1;
3104
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003105 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003106 return 1;
3107
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003108 /*
3109 * We may have set a slab to be unmergeable during bootstrap.
3110 */
3111 if (s->refcount < 0)
3112 return 1;
3113
Christoph Lameter81819f02007-05-06 14:49:36 -07003114 return 0;
3115}
3116
3117static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003118 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003119 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003120{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003121 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003122
3123 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3124 return NULL;
3125
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003126 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003127 return NULL;
3128
3129 size = ALIGN(size, sizeof(void *));
3130 align = calculate_alignment(flags, align, size);
3131 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003132 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003133
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003134 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003135 if (slab_unmergeable(s))
3136 continue;
3137
3138 if (size > s->size)
3139 continue;
3140
Christoph Lameterba0268a2007-09-11 15:24:11 -07003141 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003142 continue;
3143 /*
3144 * Check if alignment is compatible.
3145 * Courtesy of Adrian Drzewiecki
3146 */
Pekka Enberg06428782008-01-07 23:20:27 -08003147 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003148 continue;
3149
3150 if (s->size - size >= sizeof(void *))
3151 continue;
3152
3153 return s;
3154 }
3155 return NULL;
3156}
3157
3158struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003159 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003160{
3161 struct kmem_cache *s;
3162
3163 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003164 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003165 if (s) {
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003166 int cpu;
3167
Christoph Lameter81819f02007-05-06 14:49:36 -07003168 s->refcount++;
3169 /*
3170 * Adjust the object sizes so that we clear
3171 * the complete object on kzalloc.
3172 */
3173 s->objsize = max(s->objsize, (int)size);
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003174
3175 /*
3176 * And then we need to update the object size in the
3177 * per cpu structures
3178 */
3179 for_each_online_cpu(cpu)
3180 get_cpu_slab(s, cpu)->objsize = s->objsize;
Christoph Lameter6446faa2008-02-15 23:45:26 -08003181
Christoph Lameter81819f02007-05-06 14:49:36 -07003182 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003183 up_write(&slub_lock);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003184
David Rientjes7b8f3b62008-12-17 22:09:46 -08003185 if (sysfs_slab_alias(s, name)) {
3186 down_write(&slub_lock);
3187 s->refcount--;
3188 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003189 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003190 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003191 return s;
3192 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003193
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003194 s = kmalloc(kmem_size, GFP_KERNEL);
3195 if (s) {
3196 if (kmem_cache_open(s, GFP_KERNEL, name,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003197 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003198 list_add(&s->list, &slab_caches);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003199 up_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003200 if (sysfs_slab_add(s)) {
3201 down_write(&slub_lock);
3202 list_del(&s->list);
3203 up_write(&slub_lock);
3204 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003205 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003206 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003207 return s;
3208 }
3209 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003210 }
3211 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003212
3213err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003214 if (flags & SLAB_PANIC)
3215 panic("Cannot create slabcache %s\n", name);
3216 else
3217 s = NULL;
3218 return s;
3219}
3220EXPORT_SYMBOL(kmem_cache_create);
3221
Christoph Lameter81819f02007-05-06 14:49:36 -07003222#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003223/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003224 * Use the cpu notifier to insure that the cpu slabs are flushed when
3225 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003226 */
3227static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3228 unsigned long action, void *hcpu)
3229{
3230 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003231 struct kmem_cache *s;
3232 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003233
3234 switch (action) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003235 case CPU_UP_PREPARE:
3236 case CPU_UP_PREPARE_FROZEN:
3237 init_alloc_cpu_cpu(cpu);
3238 down_read(&slub_lock);
3239 list_for_each_entry(s, &slab_caches, list)
3240 s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu,
3241 GFP_KERNEL);
3242 up_read(&slub_lock);
3243 break;
3244
Christoph Lameter81819f02007-05-06 14:49:36 -07003245 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003246 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003247 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003248 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003249 down_read(&slub_lock);
3250 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003251 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
3252
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003253 local_irq_save(flags);
3254 __flush_cpu_slab(s, cpu);
3255 local_irq_restore(flags);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003256 free_kmem_cache_cpu(c, cpu);
3257 s->cpu_slab[cpu] = NULL;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003258 }
3259 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003260 break;
3261 default:
3262 break;
3263 }
3264 return NOTIFY_OK;
3265}
3266
Pekka Enberg06428782008-01-07 23:20:27 -08003267static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003268 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003269};
Christoph Lameter81819f02007-05-06 14:49:36 -07003270
3271#endif
3272
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003273void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003274{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003275 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003276 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003277
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003278 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003279 return kmalloc_large(size, gfpflags);
3280
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003281 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003282
Satyam Sharma2408c552007-10-16 01:24:44 -07003283 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003284 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003285
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003286 ret = slab_alloc(s, gfpflags, -1, caller);
3287
3288 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003289 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003290
3291 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003292}
3293
3294void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003295 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003296{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003297 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003298 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003299
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003300 if (unlikely(size > SLUB_MAX_SIZE))
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003301 return kmalloc_large_node(size, gfpflags, node);
Pekka Enbergeada35e2008-02-11 22:47:46 +02003302
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003303 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003304
Satyam Sharma2408c552007-10-16 01:24:44 -07003305 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003306 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003307
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003308 ret = slab_alloc(s, gfpflags, node, caller);
3309
3310 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003311 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003312
3313 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003314}
3315
Christoph Lameterf6acb632008-04-29 16:16:06 -07003316#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003317static unsigned long count_partial(struct kmem_cache_node *n,
3318 int (*get_count)(struct page *))
Christoph Lameter5b06c8532008-04-14 18:51:34 +03003319{
3320 unsigned long flags;
3321 unsigned long x = 0;
3322 struct page *page;
3323
3324 spin_lock_irqsave(&n->list_lock, flags);
3325 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter205ab992008-04-14 19:11:40 +03003326 x += get_count(page);
Christoph Lameter5b06c8532008-04-14 18:51:34 +03003327 spin_unlock_irqrestore(&n->list_lock, flags);
3328 return x;
3329}
Christoph Lameter205ab992008-04-14 19:11:40 +03003330
3331static int count_inuse(struct page *page)
3332{
3333 return page->inuse;
3334}
3335
3336static int count_total(struct page *page)
3337{
3338 return page->objects;
3339}
3340
3341static int count_free(struct page *page)
3342{
3343 return page->objects - page->inuse;
3344}
Christoph Lameter5b06c8532008-04-14 18:51:34 +03003345
Christoph Lameter434e2452007-07-17 04:03:30 -07003346static int validate_slab(struct kmem_cache *s, struct page *page,
3347 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003348{
3349 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003350 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003351
3352 if (!check_slab(s, page) ||
3353 !on_freelist(s, page, NULL))
3354 return 0;
3355
3356 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003357 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003358
Christoph Lameter7656c722007-05-09 02:32:40 -07003359 for_each_free_object(p, s, page->freelist) {
3360 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003361 if (!check_object(s, page, p, 0))
3362 return 0;
3363 }
3364
Christoph Lameter224a88b2008-04-14 19:11:31 +03003365 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003366 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003367 if (!check_object(s, page, p, 1))
3368 return 0;
3369 return 1;
3370}
3371
Christoph Lameter434e2452007-07-17 04:03:30 -07003372static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3373 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003374{
3375 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003376 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003377 slab_unlock(page);
3378 } else
3379 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3380 s->name, page);
3381
3382 if (s->flags & DEBUG_DEFAULT_FLAGS) {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003383 if (!PageSlubDebug(page))
3384 printk(KERN_ERR "SLUB %s: SlubDebug not set "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003385 "on slab 0x%p\n", s->name, page);
3386 } else {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003387 if (PageSlubDebug(page))
3388 printk(KERN_ERR "SLUB %s: SlubDebug set on "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003389 "slab 0x%p\n", s->name, page);
3390 }
3391}
3392
Christoph Lameter434e2452007-07-17 04:03:30 -07003393static int validate_slab_node(struct kmem_cache *s,
3394 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003395{
3396 unsigned long count = 0;
3397 struct page *page;
3398 unsigned long flags;
3399
3400 spin_lock_irqsave(&n->list_lock, flags);
3401
3402 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003403 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003404 count++;
3405 }
3406 if (count != n->nr_partial)
3407 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3408 "counter=%ld\n", s->name, count, n->nr_partial);
3409
3410 if (!(s->flags & SLAB_STORE_USER))
3411 goto out;
3412
3413 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003414 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003415 count++;
3416 }
3417 if (count != atomic_long_read(&n->nr_slabs))
3418 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3419 "counter=%ld\n", s->name, count,
3420 atomic_long_read(&n->nr_slabs));
3421
3422out:
3423 spin_unlock_irqrestore(&n->list_lock, flags);
3424 return count;
3425}
3426
Christoph Lameter434e2452007-07-17 04:03:30 -07003427static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003428{
3429 int node;
3430 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003431 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003432 sizeof(unsigned long), GFP_KERNEL);
3433
3434 if (!map)
3435 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003436
3437 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003438 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003439 struct kmem_cache_node *n = get_node(s, node);
3440
Christoph Lameter434e2452007-07-17 04:03:30 -07003441 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003442 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003443 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003444 return count;
3445}
3446
Christoph Lameterb3459702007-05-09 02:32:41 -07003447#ifdef SLUB_RESILIENCY_TEST
3448static void resiliency_test(void)
3449{
3450 u8 *p;
3451
3452 printk(KERN_ERR "SLUB resiliency testing\n");
3453 printk(KERN_ERR "-----------------------\n");
3454 printk(KERN_ERR "A. Corruption after allocation\n");
3455
3456 p = kzalloc(16, GFP_KERNEL);
3457 p[16] = 0x12;
3458 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3459 " 0x12->0x%p\n\n", p + 16);
3460
3461 validate_slab_cache(kmalloc_caches + 4);
3462
3463 /* Hmmm... The next two are dangerous */
3464 p = kzalloc(32, GFP_KERNEL);
3465 p[32 + sizeof(void *)] = 0x34;
3466 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003467 " 0x34 -> -0x%p\n", p);
3468 printk(KERN_ERR
3469 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003470
3471 validate_slab_cache(kmalloc_caches + 5);
3472 p = kzalloc(64, GFP_KERNEL);
3473 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3474 *p = 0x56;
3475 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3476 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003477 printk(KERN_ERR
3478 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003479 validate_slab_cache(kmalloc_caches + 6);
3480
3481 printk(KERN_ERR "\nB. Corruption after free\n");
3482 p = kzalloc(128, GFP_KERNEL);
3483 kfree(p);
3484 *p = 0x78;
3485 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3486 validate_slab_cache(kmalloc_caches + 7);
3487
3488 p = kzalloc(256, GFP_KERNEL);
3489 kfree(p);
3490 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003491 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3492 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003493 validate_slab_cache(kmalloc_caches + 8);
3494
3495 p = kzalloc(512, GFP_KERNEL);
3496 kfree(p);
3497 p[512] = 0xab;
3498 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3499 validate_slab_cache(kmalloc_caches + 9);
3500}
3501#else
3502static void resiliency_test(void) {};
3503#endif
3504
Christoph Lameter88a420e2007-05-06 14:49:45 -07003505/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003506 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003507 * and freed.
3508 */
3509
3510struct location {
3511 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003512 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003513 long long sum_time;
3514 long min_time;
3515 long max_time;
3516 long min_pid;
3517 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303518 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003519 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003520};
3521
3522struct loc_track {
3523 unsigned long max;
3524 unsigned long count;
3525 struct location *loc;
3526};
3527
3528static void free_loc_track(struct loc_track *t)
3529{
3530 if (t->max)
3531 free_pages((unsigned long)t->loc,
3532 get_order(sizeof(struct location) * t->max));
3533}
3534
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003535static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003536{
3537 struct location *l;
3538 int order;
3539
Christoph Lameter88a420e2007-05-06 14:49:45 -07003540 order = get_order(sizeof(struct location) * max);
3541
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003542 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003543 if (!l)
3544 return 0;
3545
3546 if (t->count) {
3547 memcpy(l, t->loc, sizeof(struct location) * t->count);
3548 free_loc_track(t);
3549 }
3550 t->max = max;
3551 t->loc = l;
3552 return 1;
3553}
3554
3555static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003556 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003557{
3558 long start, end, pos;
3559 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003560 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003561 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003562
3563 start = -1;
3564 end = t->count;
3565
3566 for ( ; ; ) {
3567 pos = start + (end - start + 1) / 2;
3568
3569 /*
3570 * There is nothing at "end". If we end up there
3571 * we need to add something to before end.
3572 */
3573 if (pos == end)
3574 break;
3575
3576 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003577 if (track->addr == caddr) {
3578
3579 l = &t->loc[pos];
3580 l->count++;
3581 if (track->when) {
3582 l->sum_time += age;
3583 if (age < l->min_time)
3584 l->min_time = age;
3585 if (age > l->max_time)
3586 l->max_time = age;
3587
3588 if (track->pid < l->min_pid)
3589 l->min_pid = track->pid;
3590 if (track->pid > l->max_pid)
3591 l->max_pid = track->pid;
3592
Rusty Russell174596a2009-01-01 10:12:29 +10303593 cpumask_set_cpu(track->cpu,
3594 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003595 }
3596 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003597 return 1;
3598 }
3599
Christoph Lameter45edfa52007-05-09 02:32:45 -07003600 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003601 end = pos;
3602 else
3603 start = pos;
3604 }
3605
3606 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003607 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003608 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003609 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003610 return 0;
3611
3612 l = t->loc + pos;
3613 if (pos < t->count)
3614 memmove(l + 1, l,
3615 (t->count - pos) * sizeof(struct location));
3616 t->count++;
3617 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003618 l->addr = track->addr;
3619 l->sum_time = age;
3620 l->min_time = age;
3621 l->max_time = age;
3622 l->min_pid = track->pid;
3623 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303624 cpumask_clear(to_cpumask(l->cpus));
3625 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003626 nodes_clear(l->nodes);
3627 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003628 return 1;
3629}
3630
3631static void process_slab(struct loc_track *t, struct kmem_cache *s,
3632 struct page *page, enum track_item alloc)
3633{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003634 void *addr = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +03003635 DECLARE_BITMAP(map, page->objects);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003636 void *p;
3637
Christoph Lameter39b26462008-04-14 19:11:30 +03003638 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003639 for_each_free_object(p, s, page->freelist)
3640 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003641
Christoph Lameter224a88b2008-04-14 19:11:31 +03003642 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003643 if (!test_bit(slab_index(p, s, addr), map))
3644 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003645}
3646
3647static int list_locations(struct kmem_cache *s, char *buf,
3648 enum track_item alloc)
3649{
Harvey Harrisone374d482008-01-31 15:20:50 -08003650 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003651 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003652 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003653 int node;
3654
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003655 if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
Andrew Mortonea3061d2007-10-16 01:26:09 -07003656 GFP_TEMPORARY))
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003657 return sprintf(buf, "Out of memory\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003658
3659 /* Push back cpu slabs */
3660 flush_all(s);
3661
Christoph Lameterf64dc582007-10-16 01:25:33 -07003662 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003663 struct kmem_cache_node *n = get_node(s, node);
3664 unsigned long flags;
3665 struct page *page;
3666
Christoph Lameter9e869432007-08-22 14:01:56 -07003667 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003668 continue;
3669
3670 spin_lock_irqsave(&n->list_lock, flags);
3671 list_for_each_entry(page, &n->partial, lru)
3672 process_slab(&t, s, page, alloc);
3673 list_for_each_entry(page, &n->full, lru)
3674 process_slab(&t, s, page, alloc);
3675 spin_unlock_irqrestore(&n->list_lock, flags);
3676 }
3677
3678 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003679 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003680
Hugh Dickins9c246242008-12-09 13:14:27 -08003681 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003682 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003683 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003684
3685 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003686 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003687 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003688 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003689
3690 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003691 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003692 l->min_time,
3693 (long)div_u64(l->sum_time, l->count),
3694 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003695 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003696 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003697 l->min_time);
3698
3699 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003700 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003701 l->min_pid, l->max_pid);
3702 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003703 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003704 l->min_pid);
3705
Rusty Russell174596a2009-01-01 10:12:29 +10303706 if (num_online_cpus() > 1 &&
3707 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003708 len < PAGE_SIZE - 60) {
3709 len += sprintf(buf + len, " cpus=");
3710 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303711 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003712 }
3713
Christoph Lameter84966342007-06-23 17:16:32 -07003714 if (num_online_nodes() > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003715 len < PAGE_SIZE - 60) {
3716 len += sprintf(buf + len, " nodes=");
3717 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003718 l->nodes);
3719 }
3720
Harvey Harrisone374d482008-01-31 15:20:50 -08003721 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003722 }
3723
3724 free_loc_track(&t);
3725 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003726 len += sprintf(buf, "No data\n");
3727 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003728}
3729
Christoph Lameter81819f02007-05-06 14:49:36 -07003730enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003731 SL_ALL, /* All slabs */
3732 SL_PARTIAL, /* Only partially allocated slabs */
3733 SL_CPU, /* Only slabs used for cpu caches */
3734 SL_OBJECTS, /* Determine allocated objects not slabs */
3735 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003736};
3737
Christoph Lameter205ab992008-04-14 19:11:40 +03003738#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003739#define SO_PARTIAL (1 << SL_PARTIAL)
3740#define SO_CPU (1 << SL_CPU)
3741#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003742#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003743
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003744static ssize_t show_slab_objects(struct kmem_cache *s,
3745 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003746{
3747 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003748 int node;
3749 int x;
3750 unsigned long *nodes;
3751 unsigned long *per_cpu;
3752
3753 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003754 if (!nodes)
3755 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003756 per_cpu = nodes + nr_node_ids;
3757
Christoph Lameter205ab992008-04-14 19:11:40 +03003758 if (flags & SO_CPU) {
3759 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003760
Christoph Lameter205ab992008-04-14 19:11:40 +03003761 for_each_possible_cpu(cpu) {
3762 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003763
Christoph Lameter205ab992008-04-14 19:11:40 +03003764 if (!c || c->node < 0)
3765 continue;
3766
3767 if (c->page) {
3768 if (flags & SO_TOTAL)
3769 x = c->page->objects;
3770 else if (flags & SO_OBJECTS)
3771 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003772 else
3773 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003774
Christoph Lameter81819f02007-05-06 14:49:36 -07003775 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003776 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003777 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003778 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003779 }
3780 }
3781
Christoph Lameter205ab992008-04-14 19:11:40 +03003782 if (flags & SO_ALL) {
3783 for_each_node_state(node, N_NORMAL_MEMORY) {
3784 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003785
Christoph Lameter205ab992008-04-14 19:11:40 +03003786 if (flags & SO_TOTAL)
3787 x = atomic_long_read(&n->total_objects);
3788 else if (flags & SO_OBJECTS)
3789 x = atomic_long_read(&n->total_objects) -
3790 count_partial(n, count_free);
3791
3792 else
3793 x = atomic_long_read(&n->nr_slabs);
3794 total += x;
3795 nodes[node] += x;
3796 }
3797
3798 } else if (flags & SO_PARTIAL) {
3799 for_each_node_state(node, N_NORMAL_MEMORY) {
3800 struct kmem_cache_node *n = get_node(s, node);
3801
3802 if (flags & SO_TOTAL)
3803 x = count_partial(n, count_total);
3804 else if (flags & SO_OBJECTS)
3805 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003806 else
3807 x = n->nr_partial;
3808 total += x;
3809 nodes[node] += x;
3810 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003811 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003812 x = sprintf(buf, "%lu", total);
3813#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003814 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003815 if (nodes[node])
3816 x += sprintf(buf + x, " N%d=%lu",
3817 node, nodes[node]);
3818#endif
3819 kfree(nodes);
3820 return x + sprintf(buf + x, "\n");
3821}
3822
3823static int any_slab_objects(struct kmem_cache *s)
3824{
3825 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003826
3827 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003828 struct kmem_cache_node *n = get_node(s, node);
3829
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003830 if (!n)
3831 continue;
3832
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003833 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003834 return 1;
3835 }
3836 return 0;
3837}
3838
3839#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3840#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3841
3842struct slab_attribute {
3843 struct attribute attr;
3844 ssize_t (*show)(struct kmem_cache *s, char *buf);
3845 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3846};
3847
3848#define SLAB_ATTR_RO(_name) \
3849 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3850
3851#define SLAB_ATTR(_name) \
3852 static struct slab_attribute _name##_attr = \
3853 __ATTR(_name, 0644, _name##_show, _name##_store)
3854
Christoph Lameter81819f02007-05-06 14:49:36 -07003855static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3856{
3857 return sprintf(buf, "%d\n", s->size);
3858}
3859SLAB_ATTR_RO(slab_size);
3860
3861static ssize_t align_show(struct kmem_cache *s, char *buf)
3862{
3863 return sprintf(buf, "%d\n", s->align);
3864}
3865SLAB_ATTR_RO(align);
3866
3867static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3868{
3869 return sprintf(buf, "%d\n", s->objsize);
3870}
3871SLAB_ATTR_RO(object_size);
3872
3873static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3874{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003875 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003876}
3877SLAB_ATTR_RO(objs_per_slab);
3878
Christoph Lameter06b285d2008-04-14 19:11:41 +03003879static ssize_t order_store(struct kmem_cache *s,
3880 const char *buf, size_t length)
3881{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003882 unsigned long order;
3883 int err;
3884
3885 err = strict_strtoul(buf, 10, &order);
3886 if (err)
3887 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003888
3889 if (order > slub_max_order || order < slub_min_order)
3890 return -EINVAL;
3891
3892 calculate_sizes(s, order);
3893 return length;
3894}
3895
Christoph Lameter81819f02007-05-06 14:49:36 -07003896static ssize_t order_show(struct kmem_cache *s, char *buf)
3897{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003898 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003899}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003900SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003901
David Rientjes73d342b2009-02-22 17:40:09 -08003902static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3903{
3904 return sprintf(buf, "%lu\n", s->min_partial);
3905}
3906
3907static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3908 size_t length)
3909{
3910 unsigned long min;
3911 int err;
3912
3913 err = strict_strtoul(buf, 10, &min);
3914 if (err)
3915 return err;
3916
David Rientjesc0bdb232009-02-25 09:16:35 +02003917 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003918 return length;
3919}
3920SLAB_ATTR(min_partial);
3921
Christoph Lameter81819f02007-05-06 14:49:36 -07003922static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3923{
3924 if (s->ctor) {
3925 int n = sprint_symbol(buf, (unsigned long)s->ctor);
3926
3927 return n + sprintf(buf + n, "\n");
3928 }
3929 return 0;
3930}
3931SLAB_ATTR_RO(ctor);
3932
Christoph Lameter81819f02007-05-06 14:49:36 -07003933static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3934{
3935 return sprintf(buf, "%d\n", s->refcount - 1);
3936}
3937SLAB_ATTR_RO(aliases);
3938
3939static ssize_t slabs_show(struct kmem_cache *s, char *buf)
3940{
Christoph Lameter205ab992008-04-14 19:11:40 +03003941 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003942}
3943SLAB_ATTR_RO(slabs);
3944
3945static ssize_t partial_show(struct kmem_cache *s, char *buf)
3946{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003947 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003948}
3949SLAB_ATTR_RO(partial);
3950
3951static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3952{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003953 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003954}
3955SLAB_ATTR_RO(cpu_slabs);
3956
3957static ssize_t objects_show(struct kmem_cache *s, char *buf)
3958{
Christoph Lameter205ab992008-04-14 19:11:40 +03003959 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07003960}
3961SLAB_ATTR_RO(objects);
3962
Christoph Lameter205ab992008-04-14 19:11:40 +03003963static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
3964{
3965 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
3966}
3967SLAB_ATTR_RO(objects_partial);
3968
3969static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
3970{
3971 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
3972}
3973SLAB_ATTR_RO(total_objects);
3974
Christoph Lameter81819f02007-05-06 14:49:36 -07003975static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
3976{
3977 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
3978}
3979
3980static ssize_t sanity_checks_store(struct kmem_cache *s,
3981 const char *buf, size_t length)
3982{
3983 s->flags &= ~SLAB_DEBUG_FREE;
3984 if (buf[0] == '1')
3985 s->flags |= SLAB_DEBUG_FREE;
3986 return length;
3987}
3988SLAB_ATTR(sanity_checks);
3989
3990static ssize_t trace_show(struct kmem_cache *s, char *buf)
3991{
3992 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
3993}
3994
3995static ssize_t trace_store(struct kmem_cache *s, const char *buf,
3996 size_t length)
3997{
3998 s->flags &= ~SLAB_TRACE;
3999 if (buf[0] == '1')
4000 s->flags |= SLAB_TRACE;
4001 return length;
4002}
4003SLAB_ATTR(trace);
4004
4005static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4006{
4007 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4008}
4009
4010static ssize_t reclaim_account_store(struct kmem_cache *s,
4011 const char *buf, size_t length)
4012{
4013 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4014 if (buf[0] == '1')
4015 s->flags |= SLAB_RECLAIM_ACCOUNT;
4016 return length;
4017}
4018SLAB_ATTR(reclaim_account);
4019
4020static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4021{
Christoph Lameter5af60832007-05-06 14:49:56 -07004022 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004023}
4024SLAB_ATTR_RO(hwcache_align);
4025
4026#ifdef CONFIG_ZONE_DMA
4027static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4028{
4029 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4030}
4031SLAB_ATTR_RO(cache_dma);
4032#endif
4033
4034static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4035{
4036 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4037}
4038SLAB_ATTR_RO(destroy_by_rcu);
4039
4040static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4041{
4042 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4043}
4044
4045static ssize_t red_zone_store(struct kmem_cache *s,
4046 const char *buf, size_t length)
4047{
4048 if (any_slab_objects(s))
4049 return -EBUSY;
4050
4051 s->flags &= ~SLAB_RED_ZONE;
4052 if (buf[0] == '1')
4053 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004054 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004055 return length;
4056}
4057SLAB_ATTR(red_zone);
4058
4059static ssize_t poison_show(struct kmem_cache *s, char *buf)
4060{
4061 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4062}
4063
4064static ssize_t poison_store(struct kmem_cache *s,
4065 const char *buf, size_t length)
4066{
4067 if (any_slab_objects(s))
4068 return -EBUSY;
4069
4070 s->flags &= ~SLAB_POISON;
4071 if (buf[0] == '1')
4072 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004073 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004074 return length;
4075}
4076SLAB_ATTR(poison);
4077
4078static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4079{
4080 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4081}
4082
4083static ssize_t store_user_store(struct kmem_cache *s,
4084 const char *buf, size_t length)
4085{
4086 if (any_slab_objects(s))
4087 return -EBUSY;
4088
4089 s->flags &= ~SLAB_STORE_USER;
4090 if (buf[0] == '1')
4091 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004092 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004093 return length;
4094}
4095SLAB_ATTR(store_user);
4096
Christoph Lameter53e15af2007-05-06 14:49:43 -07004097static ssize_t validate_show(struct kmem_cache *s, char *buf)
4098{
4099 return 0;
4100}
4101
4102static ssize_t validate_store(struct kmem_cache *s,
4103 const char *buf, size_t length)
4104{
Christoph Lameter434e2452007-07-17 04:03:30 -07004105 int ret = -EINVAL;
4106
4107 if (buf[0] == '1') {
4108 ret = validate_slab_cache(s);
4109 if (ret >= 0)
4110 ret = length;
4111 }
4112 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004113}
4114SLAB_ATTR(validate);
4115
Christoph Lameter2086d262007-05-06 14:49:46 -07004116static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4117{
4118 return 0;
4119}
4120
4121static ssize_t shrink_store(struct kmem_cache *s,
4122 const char *buf, size_t length)
4123{
4124 if (buf[0] == '1') {
4125 int rc = kmem_cache_shrink(s);
4126
4127 if (rc)
4128 return rc;
4129 } else
4130 return -EINVAL;
4131 return length;
4132}
4133SLAB_ATTR(shrink);
4134
Christoph Lameter88a420e2007-05-06 14:49:45 -07004135static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4136{
4137 if (!(s->flags & SLAB_STORE_USER))
4138 return -ENOSYS;
4139 return list_locations(s, buf, TRACK_ALLOC);
4140}
4141SLAB_ATTR_RO(alloc_calls);
4142
4143static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4144{
4145 if (!(s->flags & SLAB_STORE_USER))
4146 return -ENOSYS;
4147 return list_locations(s, buf, TRACK_FREE);
4148}
4149SLAB_ATTR_RO(free_calls);
4150
Christoph Lameter81819f02007-05-06 14:49:36 -07004151#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004152static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004153{
Christoph Lameter98246012008-01-07 23:20:26 -08004154 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004155}
4156
Christoph Lameter98246012008-01-07 23:20:26 -08004157static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004158 const char *buf, size_t length)
4159{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004160 unsigned long ratio;
4161 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004162
Christoph Lameter0121c6192008-04-29 16:11:12 -07004163 err = strict_strtoul(buf, 10, &ratio);
4164 if (err)
4165 return err;
4166
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004167 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004168 s->remote_node_defrag_ratio = ratio * 10;
4169
Christoph Lameter81819f02007-05-06 14:49:36 -07004170 return length;
4171}
Christoph Lameter98246012008-01-07 23:20:26 -08004172SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004173#endif
4174
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004175#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004176static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4177{
4178 unsigned long sum = 0;
4179 int cpu;
4180 int len;
4181 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4182
4183 if (!data)
4184 return -ENOMEM;
4185
4186 for_each_online_cpu(cpu) {
4187 unsigned x = get_cpu_slab(s, cpu)->stat[si];
4188
4189 data[cpu] = x;
4190 sum += x;
4191 }
4192
4193 len = sprintf(buf, "%lu", sum);
4194
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004195#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004196 for_each_online_cpu(cpu) {
4197 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004198 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004199 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004200#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004201 kfree(data);
4202 return len + sprintf(buf + len, "\n");
4203}
4204
4205#define STAT_ATTR(si, text) \
4206static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4207{ \
4208 return show_stat(s, buf, si); \
4209} \
4210SLAB_ATTR_RO(text); \
4211
4212STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4213STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4214STAT_ATTR(FREE_FASTPATH, free_fastpath);
4215STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4216STAT_ATTR(FREE_FROZEN, free_frozen);
4217STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4218STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4219STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4220STAT_ATTR(ALLOC_SLAB, alloc_slab);
4221STAT_ATTR(ALLOC_REFILL, alloc_refill);
4222STAT_ATTR(FREE_SLAB, free_slab);
4223STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4224STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4225STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4226STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4227STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4228STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004229STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004230#endif
4231
Pekka Enberg06428782008-01-07 23:20:27 -08004232static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004233 &slab_size_attr.attr,
4234 &object_size_attr.attr,
4235 &objs_per_slab_attr.attr,
4236 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004237 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004238 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004239 &objects_partial_attr.attr,
4240 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004241 &slabs_attr.attr,
4242 &partial_attr.attr,
4243 &cpu_slabs_attr.attr,
4244 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004245 &aliases_attr.attr,
4246 &align_attr.attr,
4247 &sanity_checks_attr.attr,
4248 &trace_attr.attr,
4249 &hwcache_align_attr.attr,
4250 &reclaim_account_attr.attr,
4251 &destroy_by_rcu_attr.attr,
4252 &red_zone_attr.attr,
4253 &poison_attr.attr,
4254 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004255 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004256 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004257 &alloc_calls_attr.attr,
4258 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004259#ifdef CONFIG_ZONE_DMA
4260 &cache_dma_attr.attr,
4261#endif
4262#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004263 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004264#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004265#ifdef CONFIG_SLUB_STATS
4266 &alloc_fastpath_attr.attr,
4267 &alloc_slowpath_attr.attr,
4268 &free_fastpath_attr.attr,
4269 &free_slowpath_attr.attr,
4270 &free_frozen_attr.attr,
4271 &free_add_partial_attr.attr,
4272 &free_remove_partial_attr.attr,
4273 &alloc_from_partial_attr.attr,
4274 &alloc_slab_attr.attr,
4275 &alloc_refill_attr.attr,
4276 &free_slab_attr.attr,
4277 &cpuslab_flush_attr.attr,
4278 &deactivate_full_attr.attr,
4279 &deactivate_empty_attr.attr,
4280 &deactivate_to_head_attr.attr,
4281 &deactivate_to_tail_attr.attr,
4282 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004283 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004284#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004285 NULL
4286};
4287
4288static struct attribute_group slab_attr_group = {
4289 .attrs = slab_attrs,
4290};
4291
4292static ssize_t slab_attr_show(struct kobject *kobj,
4293 struct attribute *attr,
4294 char *buf)
4295{
4296 struct slab_attribute *attribute;
4297 struct kmem_cache *s;
4298 int err;
4299
4300 attribute = to_slab_attr(attr);
4301 s = to_slab(kobj);
4302
4303 if (!attribute->show)
4304 return -EIO;
4305
4306 err = attribute->show(s, buf);
4307
4308 return err;
4309}
4310
4311static ssize_t slab_attr_store(struct kobject *kobj,
4312 struct attribute *attr,
4313 const char *buf, size_t len)
4314{
4315 struct slab_attribute *attribute;
4316 struct kmem_cache *s;
4317 int err;
4318
4319 attribute = to_slab_attr(attr);
4320 s = to_slab(kobj);
4321
4322 if (!attribute->store)
4323 return -EIO;
4324
4325 err = attribute->store(s, buf, len);
4326
4327 return err;
4328}
4329
Christoph Lameter151c6022008-01-07 22:29:05 -08004330static void kmem_cache_release(struct kobject *kobj)
4331{
4332 struct kmem_cache *s = to_slab(kobj);
4333
4334 kfree(s);
4335}
4336
Christoph Lameter81819f02007-05-06 14:49:36 -07004337static struct sysfs_ops slab_sysfs_ops = {
4338 .show = slab_attr_show,
4339 .store = slab_attr_store,
4340};
4341
4342static struct kobj_type slab_ktype = {
4343 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004344 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004345};
4346
4347static int uevent_filter(struct kset *kset, struct kobject *kobj)
4348{
4349 struct kobj_type *ktype = get_ktype(kobj);
4350
4351 if (ktype == &slab_ktype)
4352 return 1;
4353 return 0;
4354}
4355
4356static struct kset_uevent_ops slab_uevent_ops = {
4357 .filter = uevent_filter,
4358};
4359
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004360static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004361
4362#define ID_STR_LENGTH 64
4363
4364/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004365 *
4366 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004367 */
4368static char *create_unique_id(struct kmem_cache *s)
4369{
4370 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4371 char *p = name;
4372
4373 BUG_ON(!name);
4374
4375 *p++ = ':';
4376 /*
4377 * First flags affecting slabcache operations. We will only
4378 * get here for aliasable slabs so we do not need to support
4379 * too many flags. The flags here must cover all flags that
4380 * are matched during merging to guarantee that the id is
4381 * unique.
4382 */
4383 if (s->flags & SLAB_CACHE_DMA)
4384 *p++ = 'd';
4385 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4386 *p++ = 'a';
4387 if (s->flags & SLAB_DEBUG_FREE)
4388 *p++ = 'F';
4389 if (p != name + 1)
4390 *p++ = '-';
4391 p += sprintf(p, "%07d", s->size);
4392 BUG_ON(p > name + ID_STR_LENGTH - 1);
4393 return name;
4394}
4395
4396static int sysfs_slab_add(struct kmem_cache *s)
4397{
4398 int err;
4399 const char *name;
4400 int unmergeable;
4401
4402 if (slab_state < SYSFS)
4403 /* Defer until later */
4404 return 0;
4405
4406 unmergeable = slab_unmergeable(s);
4407 if (unmergeable) {
4408 /*
4409 * Slabcache can never be merged so we can use the name proper.
4410 * This is typically the case for debug situations. In that
4411 * case we can catch duplicate names easily.
4412 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004413 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004414 name = s->name;
4415 } else {
4416 /*
4417 * Create a unique name for the slab as a target
4418 * for the symlinks.
4419 */
4420 name = create_unique_id(s);
4421 }
4422
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004423 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004424 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4425 if (err) {
4426 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004427 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004428 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004429
4430 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4431 if (err)
4432 return err;
4433 kobject_uevent(&s->kobj, KOBJ_ADD);
4434 if (!unmergeable) {
4435 /* Setup first alias */
4436 sysfs_slab_alias(s, s->name);
4437 kfree(name);
4438 }
4439 return 0;
4440}
4441
4442static void sysfs_slab_remove(struct kmem_cache *s)
4443{
4444 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4445 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004446 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004447}
4448
4449/*
4450 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004451 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004452 */
4453struct saved_alias {
4454 struct kmem_cache *s;
4455 const char *name;
4456 struct saved_alias *next;
4457};
4458
Adrian Bunk5af328a2007-07-17 04:03:27 -07004459static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004460
4461static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4462{
4463 struct saved_alias *al;
4464
4465 if (slab_state == SYSFS) {
4466 /*
4467 * If we have a leftover link then remove it.
4468 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004469 sysfs_remove_link(&slab_kset->kobj, name);
4470 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004471 }
4472
4473 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4474 if (!al)
4475 return -ENOMEM;
4476
4477 al->s = s;
4478 al->name = name;
4479 al->next = alias_list;
4480 alias_list = al;
4481 return 0;
4482}
4483
4484static int __init slab_sysfs_init(void)
4485{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004486 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004487 int err;
4488
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004489 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004490 if (!slab_kset) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004491 printk(KERN_ERR "Cannot register slab subsystem.\n");
4492 return -ENOSYS;
4493 }
4494
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004495 slab_state = SYSFS;
4496
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004497 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004498 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004499 if (err)
4500 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4501 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004502 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004503
4504 while (alias_list) {
4505 struct saved_alias *al = alias_list;
4506
4507 alias_list = alias_list->next;
4508 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004509 if (err)
4510 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4511 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004512 kfree(al);
4513 }
4514
4515 resiliency_test();
4516 return 0;
4517}
4518
4519__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004520#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004521
4522/*
4523 * The /proc/slabinfo ABI
4524 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004525#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004526static void print_slabinfo_header(struct seq_file *m)
4527{
4528 seq_puts(m, "slabinfo - version: 2.1\n");
4529 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4530 "<objperslab> <pagesperslab>");
4531 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4532 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4533 seq_putc(m, '\n');
4534}
4535
4536static void *s_start(struct seq_file *m, loff_t *pos)
4537{
4538 loff_t n = *pos;
4539
4540 down_read(&slub_lock);
4541 if (!n)
4542 print_slabinfo_header(m);
4543
4544 return seq_list_start(&slab_caches, *pos);
4545}
4546
4547static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4548{
4549 return seq_list_next(p, &slab_caches, pos);
4550}
4551
4552static void s_stop(struct seq_file *m, void *p)
4553{
4554 up_read(&slub_lock);
4555}
4556
4557static int s_show(struct seq_file *m, void *p)
4558{
4559 unsigned long nr_partials = 0;
4560 unsigned long nr_slabs = 0;
4561 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004562 unsigned long nr_objs = 0;
4563 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004564 struct kmem_cache *s;
4565 int node;
4566
4567 s = list_entry(p, struct kmem_cache, list);
4568
4569 for_each_online_node(node) {
4570 struct kmem_cache_node *n = get_node(s, node);
4571
4572 if (!n)
4573 continue;
4574
4575 nr_partials += n->nr_partial;
4576 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004577 nr_objs += atomic_long_read(&n->total_objects);
4578 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004579 }
4580
Christoph Lameter205ab992008-04-14 19:11:40 +03004581 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004582
4583 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004584 nr_objs, s->size, oo_objects(s->oo),
4585 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004586 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4587 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4588 0UL);
4589 seq_putc(m, '\n');
4590 return 0;
4591}
4592
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004593static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004594 .start = s_start,
4595 .next = s_next,
4596 .stop = s_stop,
4597 .show = s_show,
4598};
4599
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004600static int slabinfo_open(struct inode *inode, struct file *file)
4601{
4602 return seq_open(file, &slabinfo_op);
4603}
4604
4605static const struct file_operations proc_slabinfo_operations = {
4606 .open = slabinfo_open,
4607 .read = seq_read,
4608 .llseek = seq_lseek,
4609 .release = seq_release,
4610};
4611
4612static int __init slab_proc_init(void)
4613{
4614 proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations);
4615 return 0;
4616}
4617module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004618#endif /* CONFIG_SLABINFO */