blob: d2a54fe71ea22519feada61f181e874685c0a51b [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
7 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -07009 */
10
11#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100012#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070013#include <linux/module.h>
14#include <linux/bit_spinlock.h>
15#include <linux/interrupt.h>
16#include <linux/bitops.h>
17#include <linux/slab.h>
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040018#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070019#include <linux/seq_file.h>
Zhaolei02af61b2009-04-10 14:26:18 +080020#include <linux/kmemtrace.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020021#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070022#include <linux/cpu.h>
23#include <linux/cpuset.h>
24#include <linux/mempolicy.h>
25#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070026#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070027#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070028#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070029#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090030#include <linux/fault-inject.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070031
32/*
33 * Lock order:
34 * 1. slab_lock(page)
35 * 2. slab->list_lock
36 *
37 * The slab_lock protects operations on the object of a particular
38 * slab and its metadata in the page struct. If the slab lock
39 * has been taken then no allocations nor frees can be performed
40 * on the objects in the slab nor can the slab be added or removed
41 * from the partial or full lists since this would mean modifying
42 * the page_struct of the slab.
43 *
44 * The list_lock protects the partial and full list on each node and
45 * the partial slab counter. If taken then no new slabs may be added or
46 * removed from the lists nor make the number of partial slabs be modified.
47 * (Note that the total number of slabs is an atomic value that may be
48 * modified without taking the list lock).
49 *
50 * The list_lock is a centralized lock and thus we avoid taking it as
51 * much as possible. As long as SLUB does not have to handle partial
52 * slabs, operations can continue without any centralized lock. F.e.
53 * allocating a long series of objects that fill up slabs does not require
54 * the list lock.
55 *
56 * The lock order is sometimes inverted when we are trying to get a slab
57 * off a list. We take the list_lock and then look for a page on the list
58 * to use. While we do that objects in the slabs may be freed. We can
59 * only operate on the slab if we have also taken the slab_lock. So we use
60 * a slab_trylock() on the slab. If trylock was successful then no frees
61 * can occur anymore and we can use the slab for allocations etc. If the
62 * slab_trylock() does not succeed then frees are in progress in the slab and
63 * we must stay away from it for a while since we may cause a bouncing
64 * cacheline if we try to acquire the lock. So go onto the next slab.
65 * If all pages are busy then we may allocate a new slab instead of reusing
66 * a partial slab. A new slab has noone operating on it and thus there is
67 * no danger of cacheline contention.
68 *
69 * Interrupts are disabled during allocation and deallocation in order to
70 * make the slab allocator safe to use in the context of an irq. In addition
71 * interrupts are disabled to ensure that the processor does not change
72 * while handling per_cpu slabs, due to kernel preemption.
73 *
74 * SLUB assigns one slab for allocation to each processor.
75 * Allocations only occur from these slabs called cpu slabs.
76 *
Christoph Lameter672bba32007-05-09 02:32:39 -070077 * Slabs with free elements are kept on a partial list and during regular
78 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070079 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070080 * We track full slabs for debugging purposes though because otherwise we
81 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070082 *
83 * Slabs are freed when they become empty. Teardown and setup is
84 * minimal so we rely on the page allocators per cpu caches for
85 * fast frees and allocs.
86 *
87 * Overloading of page flags that are otherwise used for LRU management.
88 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070089 * PageActive The slab is frozen and exempt from list processing.
90 * This means that the slab is dedicated to a purpose
91 * such as satisfying allocations for a specific
92 * processor. Objects may be freed in the slab while
93 * it is frozen but slab_free will then skip the usual
94 * list operations. It is up to the processor holding
95 * the slab to integrate the slab into the slab lists
96 * when the slab is no longer needed.
97 *
98 * One use of this flag is to mark slabs that are
99 * used for allocations. Then such a slab becomes a cpu
100 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700101 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700102 * free objects in addition to the regular freelist
103 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700104 *
105 * PageError Slab requires special handling due to debug
106 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700107 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700108 */
109
Christoph Lameter5577bd82007-05-16 22:10:56 -0700110#ifdef CONFIG_SLUB_DEBUG
Andy Whitcroft8a380822008-07-23 21:27:18 -0700111#define SLABDEBUG 1
Christoph Lameter5577bd82007-05-16 22:10:56 -0700112#else
113#define SLABDEBUG 0
114#endif
115
Christoph Lameter81819f02007-05-06 14:49:36 -0700116/*
117 * Issues still to be resolved:
118 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700119 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
120 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700121 * - Variable sizing of the per node arrays
122 */
123
124/* Enable to test recovery from slab corruption on boot */
125#undef SLUB_RESILIENCY_TEST
126
Christoph Lameter81819f02007-05-06 14:49:36 -0700127/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700128 * Mininum number of partial slabs. These will be left on the partial
129 * lists even if they are empty. kmem_cache_shrink may reclaim them.
130 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800131#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700132
Christoph Lameter2086d262007-05-06 14:49:46 -0700133/*
134 * Maximum number of desirable partial slabs.
135 * The existence of more partial slabs makes kmem_cache_shrink
136 * sort the partial list by the number of objects in the.
137 */
138#define MAX_PARTIAL 10
139
Christoph Lameter81819f02007-05-06 14:49:36 -0700140#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
141 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700142
Christoph Lameter81819f02007-05-06 14:49:36 -0700143/*
David Rientjes3de47212009-07-27 18:30:35 -0700144 * Debugging flags that require metadata to be stored in the slab. These get
145 * disabled when slub_debug=O is used and a cache's min order increases with
146 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700147 */
David Rientjes3de47212009-07-27 18:30:35 -0700148#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700149
150/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700151 * Set of flags that will prevent slab merging
152 */
153#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300154 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
155 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700156
157#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200158 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700159
160#ifndef ARCH_KMALLOC_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700161#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700162#endif
163
164#ifndef ARCH_SLAB_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700165#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166#endif
167
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400168#define OO_SHIFT 16
169#define OO_MASK ((1 << OO_SHIFT) - 1)
170#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
171
Christoph Lameter81819f02007-05-06 14:49:36 -0700172/* Internal SLUB flags */
Christoph Lameter1ceef402007-08-07 15:11:48 -0700173#define __OBJECT_POISON 0x80000000 /* Poison object */
174#define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700175
176static int kmem_size = sizeof(struct kmem_cache);
177
178#ifdef CONFIG_SMP
179static struct notifier_block slab_notifier;
180#endif
181
182static enum {
183 DOWN, /* No slab functionality available */
184 PARTIAL, /* kmem_cache_open() works but kmalloc does not */
Christoph Lameter672bba32007-05-09 02:32:39 -0700185 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700186 SYSFS /* Sysfs up */
187} slab_state = DOWN;
188
189/* A list of all slab caches on the system */
190static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700191static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700192
Christoph Lameter02cbc872007-05-09 02:32:43 -0700193/*
194 * Tracking user of a slab.
195 */
196struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300197 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700198 int cpu; /* Was running on cpu */
199 int pid; /* Pid context */
200 unsigned long when; /* When did the operation occur */
201};
202
203enum track_item { TRACK_ALLOC, TRACK_FREE };
204
Christoph Lameterf6acb632008-04-29 16:16:06 -0700205#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -0700206static int sysfs_slab_add(struct kmem_cache *);
207static int sysfs_slab_alias(struct kmem_cache *, const char *);
208static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800209
Christoph Lameter81819f02007-05-06 14:49:36 -0700210#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700211static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
212static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
213 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800214static inline void sysfs_slab_remove(struct kmem_cache *s)
215{
216 kfree(s);
217}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800218
Christoph Lameter81819f02007-05-06 14:49:36 -0700219#endif
220
Christoph Lameter84e554e2009-12-18 16:26:23 -0600221static inline void stat(struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800222{
223#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600224 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800225#endif
226}
227
Christoph Lameter81819f02007-05-06 14:49:36 -0700228/********************************************************************
229 * Core slab cache functions
230 *******************************************************************/
231
232int slab_is_available(void)
233{
234 return slab_state >= UP;
235}
236
237static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
238{
239#ifdef CONFIG_NUMA
240 return s->node[node];
241#else
242 return &s->local_node;
243#endif
244}
245
Christoph Lameter6446faa2008-02-15 23:45:26 -0800246/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700247static inline int check_valid_pointer(struct kmem_cache *s,
248 struct page *page, const void *object)
249{
250 void *base;
251
Christoph Lametera973e9d2008-03-01 13:40:44 -0800252 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700253 return 1;
254
Christoph Lametera973e9d2008-03-01 13:40:44 -0800255 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300256 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700257 (object - base) % s->size) {
258 return 0;
259 }
260
261 return 1;
262}
263
Christoph Lameter7656c722007-05-09 02:32:40 -0700264static inline void *get_freepointer(struct kmem_cache *s, void *object)
265{
266 return *(void **)(object + s->offset);
267}
268
269static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
270{
271 *(void **)(object + s->offset) = fp;
272}
273
274/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300275#define for_each_object(__p, __s, __addr, __objects) \
276 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700277 __p += (__s)->size)
278
279/* Scan freelist */
280#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800281 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700282
283/* Determine object index from a given position */
284static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
285{
286 return (p - addr) / s->size;
287}
288
Christoph Lameter834f3d12008-04-14 19:11:31 +0300289static inline struct kmem_cache_order_objects oo_make(int order,
290 unsigned long size)
291{
292 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400293 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300294 };
295
296 return x;
297}
298
299static inline int oo_order(struct kmem_cache_order_objects x)
300{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400301 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300302}
303
304static inline int oo_objects(struct kmem_cache_order_objects x)
305{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400306 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300307}
308
Christoph Lameter41ecc552007-05-09 02:32:44 -0700309#ifdef CONFIG_SLUB_DEBUG
310/*
311 * Debug settings:
312 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700313#ifdef CONFIG_SLUB_DEBUG_ON
314static int slub_debug = DEBUG_DEFAULT_FLAGS;
315#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700316static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700317#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700318
319static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700320static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700321
Christoph Lameter7656c722007-05-09 02:32:40 -0700322/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700323 * Object debugging
324 */
325static void print_section(char *text, u8 *addr, unsigned int length)
326{
327 int i, offset;
328 int newline = 1;
329 char ascii[17];
330
331 ascii[16] = 0;
332
333 for (i = 0; i < length; i++) {
334 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700335 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700336 newline = 0;
337 }
Pekka Enberg06428782008-01-07 23:20:27 -0800338 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700339 offset = i % 16;
340 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
341 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800342 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700343 newline = 1;
344 }
345 }
346 if (!newline) {
347 i %= 16;
348 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800349 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700350 ascii[i] = ' ';
351 i++;
352 }
Pekka Enberg06428782008-01-07 23:20:27 -0800353 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700354 }
355}
356
Christoph Lameter81819f02007-05-06 14:49:36 -0700357static struct track *get_track(struct kmem_cache *s, void *object,
358 enum track_item alloc)
359{
360 struct track *p;
361
362 if (s->offset)
363 p = object + s->offset + sizeof(void *);
364 else
365 p = object + s->inuse;
366
367 return p + alloc;
368}
369
370static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300371 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700372{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900373 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700374
Christoph Lameter81819f02007-05-06 14:49:36 -0700375 if (addr) {
376 p->addr = addr;
377 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400378 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700379 p->when = jiffies;
380 } else
381 memset(p, 0, sizeof(struct track));
382}
383
Christoph Lameter81819f02007-05-06 14:49:36 -0700384static void init_tracking(struct kmem_cache *s, void *object)
385{
Christoph Lameter24922682007-07-17 04:03:18 -0700386 if (!(s->flags & SLAB_STORE_USER))
387 return;
388
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300389 set_track(s, object, TRACK_FREE, 0UL);
390 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700391}
392
393static void print_track(const char *s, struct track *t)
394{
395 if (!t->addr)
396 return;
397
Linus Torvalds7daf7052008-07-14 12:12:53 -0700398 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300399 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700400}
401
Christoph Lameter24922682007-07-17 04:03:18 -0700402static void print_tracking(struct kmem_cache *s, void *object)
403{
404 if (!(s->flags & SLAB_STORE_USER))
405 return;
406
407 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
408 print_track("Freed", get_track(s, object, TRACK_FREE));
409}
410
411static void print_page_info(struct page *page)
412{
Christoph Lameter39b26462008-04-14 19:11:30 +0300413 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
414 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700415
416}
417
418static void slab_bug(struct kmem_cache *s, char *fmt, ...)
419{
420 va_list args;
421 char buf[100];
422
423 va_start(args, fmt);
424 vsnprintf(buf, sizeof(buf), fmt, args);
425 va_end(args);
426 printk(KERN_ERR "========================================"
427 "=====================================\n");
428 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
429 printk(KERN_ERR "----------------------------------------"
430 "-------------------------------------\n\n");
431}
432
433static void slab_fix(struct kmem_cache *s, char *fmt, ...)
434{
435 va_list args;
436 char buf[100];
437
438 va_start(args, fmt);
439 vsnprintf(buf, sizeof(buf), fmt, args);
440 va_end(args);
441 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
442}
443
444static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700445{
446 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800447 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700448
449 print_tracking(s, p);
450
451 print_page_info(page);
452
453 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
454 p, p - addr, get_freepointer(s, p));
455
456 if (p > addr + 16)
457 print_section("Bytes b4", p - 16, 16);
458
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300459 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700460
461 if (s->flags & SLAB_RED_ZONE)
462 print_section("Redzone", p + s->objsize,
463 s->inuse - s->objsize);
464
Christoph Lameter81819f02007-05-06 14:49:36 -0700465 if (s->offset)
466 off = s->offset + sizeof(void *);
467 else
468 off = s->inuse;
469
Christoph Lameter24922682007-07-17 04:03:18 -0700470 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700471 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700472
473 if (off != s->size)
474 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700475 print_section("Padding", p + off, s->size - off);
476
477 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700478}
479
480static void object_err(struct kmem_cache *s, struct page *page,
481 u8 *object, char *reason)
482{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700483 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700484 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700485}
486
Christoph Lameter24922682007-07-17 04:03:18 -0700487static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700488{
489 va_list args;
490 char buf[100];
491
Christoph Lameter24922682007-07-17 04:03:18 -0700492 va_start(args, fmt);
493 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700494 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700495 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700496 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700497 dump_stack();
498}
499
500static void init_object(struct kmem_cache *s, void *object, int active)
501{
502 u8 *p = object;
503
504 if (s->flags & __OBJECT_POISON) {
505 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800506 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700507 }
508
509 if (s->flags & SLAB_RED_ZONE)
510 memset(p + s->objsize,
511 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
512 s->inuse - s->objsize);
513}
514
Christoph Lameter24922682007-07-17 04:03:18 -0700515static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700516{
517 while (bytes) {
518 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700519 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700520 start++;
521 bytes--;
522 }
Christoph Lameter24922682007-07-17 04:03:18 -0700523 return NULL;
524}
525
526static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
527 void *from, void *to)
528{
529 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
530 memset(from, data, to - from);
531}
532
533static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
534 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800535 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700536{
537 u8 *fault;
538 u8 *end;
539
540 fault = check_bytes(start, value, bytes);
541 if (!fault)
542 return 1;
543
544 end = start + bytes;
545 while (end > fault && end[-1] == value)
546 end--;
547
548 slab_bug(s, "%s overwritten", what);
549 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
550 fault, end - 1, fault[0], value);
551 print_trailer(s, page, object);
552
553 restore_bytes(s, what, value, fault, end);
554 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700555}
556
Christoph Lameter81819f02007-05-06 14:49:36 -0700557/*
558 * Object layout:
559 *
560 * object address
561 * Bytes of the object to be managed.
562 * If the freepointer may overlay the object then the free
563 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700564 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700565 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
566 * 0xa5 (POISON_END)
567 *
568 * object + s->objsize
569 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700570 * Padding is extended by another word if Redzoning is enabled and
571 * objsize == inuse.
572 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700573 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
574 * 0xcc (RED_ACTIVE) for objects in use.
575 *
576 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700577 * Meta data starts here.
578 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700579 * A. Free pointer (if we cannot overwrite object on free)
580 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700581 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800582 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700583 * before the word boundary.
584 *
585 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700586 *
587 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700588 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700589 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700590 * If slabcaches are merged then the objsize and inuse boundaries are mostly
591 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700592 * may be used with merged slabcaches.
593 */
594
Christoph Lameter81819f02007-05-06 14:49:36 -0700595static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
596{
597 unsigned long off = s->inuse; /* The end of info */
598
599 if (s->offset)
600 /* Freepointer is placed after the object. */
601 off += sizeof(void *);
602
603 if (s->flags & SLAB_STORE_USER)
604 /* We also have user information there */
605 off += 2 * sizeof(struct track);
606
607 if (s->size == off)
608 return 1;
609
Christoph Lameter24922682007-07-17 04:03:18 -0700610 return check_bytes_and_report(s, page, p, "Object padding",
611 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700612}
613
Christoph Lameter39b26462008-04-14 19:11:30 +0300614/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700615static int slab_pad_check(struct kmem_cache *s, struct page *page)
616{
Christoph Lameter24922682007-07-17 04:03:18 -0700617 u8 *start;
618 u8 *fault;
619 u8 *end;
620 int length;
621 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700622
623 if (!(s->flags & SLAB_POISON))
624 return 1;
625
Christoph Lametera973e9d2008-03-01 13:40:44 -0800626 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300627 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300628 end = start + length;
629 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700630 if (!remainder)
631 return 1;
632
Christoph Lameter39b26462008-04-14 19:11:30 +0300633 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700634 if (!fault)
635 return 1;
636 while (end > fault && end[-1] == POISON_INUSE)
637 end--;
638
639 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300640 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700641
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200642 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700643 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700644}
645
646static int check_object(struct kmem_cache *s, struct page *page,
647 void *object, int active)
648{
649 u8 *p = object;
650 u8 *endobject = object + s->objsize;
651
652 if (s->flags & SLAB_RED_ZONE) {
653 unsigned int red =
654 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
655
Christoph Lameter24922682007-07-17 04:03:18 -0700656 if (!check_bytes_and_report(s, page, object, "Redzone",
657 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700658 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700659 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800660 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
661 check_bytes_and_report(s, page, p, "Alignment padding",
662 endobject, POISON_INUSE, s->inuse - s->objsize);
663 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700664 }
665
666 if (s->flags & SLAB_POISON) {
667 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700668 (!check_bytes_and_report(s, page, p, "Poison", p,
669 POISON_FREE, s->objsize - 1) ||
670 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800671 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700672 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700673 /*
674 * check_pad_bytes cleans up on its own.
675 */
676 check_pad_bytes(s, page, p);
677 }
678
679 if (!s->offset && active)
680 /*
681 * Object and freepointer overlap. Cannot check
682 * freepointer while object is allocated.
683 */
684 return 1;
685
686 /* Check free pointer validity */
687 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
688 object_err(s, page, p, "Freepointer corrupt");
689 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100690 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700691 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700692 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700693 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800694 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700695 return 0;
696 }
697 return 1;
698}
699
700static int check_slab(struct kmem_cache *s, struct page *page)
701{
Christoph Lameter39b26462008-04-14 19:11:30 +0300702 int maxobj;
703
Christoph Lameter81819f02007-05-06 14:49:36 -0700704 VM_BUG_ON(!irqs_disabled());
705
706 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700707 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700708 return 0;
709 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300710
711 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
712 if (page->objects > maxobj) {
713 slab_err(s, page, "objects %u > max %u",
714 s->name, page->objects, maxobj);
715 return 0;
716 }
717 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700718 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300719 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700720 return 0;
721 }
722 /* Slab_pad_check fixes things up after itself */
723 slab_pad_check(s, page);
724 return 1;
725}
726
727/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700728 * Determine if a certain object on a page is on the freelist. Must hold the
729 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700730 */
731static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
732{
733 int nr = 0;
734 void *fp = page->freelist;
735 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300736 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700737
Christoph Lameter39b26462008-04-14 19:11:30 +0300738 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700739 if (fp == search)
740 return 1;
741 if (!check_valid_pointer(s, page, fp)) {
742 if (object) {
743 object_err(s, page, object,
744 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800745 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700746 break;
747 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700748 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800749 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300750 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700751 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700752 return 0;
753 }
754 break;
755 }
756 object = fp;
757 fp = get_freepointer(s, object);
758 nr++;
759 }
760
Christoph Lameter224a88b2008-04-14 19:11:31 +0300761 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400762 if (max_objects > MAX_OBJS_PER_PAGE)
763 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300764
765 if (page->objects != max_objects) {
766 slab_err(s, page, "Wrong number of objects. Found %d but "
767 "should be %d", page->objects, max_objects);
768 page->objects = max_objects;
769 slab_fix(s, "Number of objects adjusted.");
770 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300771 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700772 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300773 "counted were %d", page->inuse, page->objects - nr);
774 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700775 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700776 }
777 return search == NULL;
778}
779
Christoph Lameter0121c6192008-04-29 16:11:12 -0700780static void trace(struct kmem_cache *s, struct page *page, void *object,
781 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700782{
783 if (s->flags & SLAB_TRACE) {
784 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
785 s->name,
786 alloc ? "alloc" : "free",
787 object, page->inuse,
788 page->freelist);
789
790 if (!alloc)
791 print_section("Object", (void *)object, s->objsize);
792
793 dump_stack();
794 }
795}
796
Christoph Lameter643b1132007-05-06 14:49:42 -0700797/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700798 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700799 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700800static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700801{
Christoph Lameter643b1132007-05-06 14:49:42 -0700802 spin_lock(&n->list_lock);
803 list_add(&page->lru, &n->full);
804 spin_unlock(&n->list_lock);
805}
806
807static void remove_full(struct kmem_cache *s, struct page *page)
808{
809 struct kmem_cache_node *n;
810
811 if (!(s->flags & SLAB_STORE_USER))
812 return;
813
814 n = get_node(s, page_to_nid(page));
815
816 spin_lock(&n->list_lock);
817 list_del(&page->lru);
818 spin_unlock(&n->list_lock);
819}
820
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300821/* Tracking of the number of slabs for debugging purposes */
822static inline unsigned long slabs_node(struct kmem_cache *s, int node)
823{
824 struct kmem_cache_node *n = get_node(s, node);
825
826 return atomic_long_read(&n->nr_slabs);
827}
828
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400829static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
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
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700972 if (tolower(*str) == 'o') {
973 /*
974 * Avoid enabling debugging on caches if its minimum order
975 * would increase as a result.
976 */
977 disable_higher_order_debug = 1;
978 goto out;
979 }
980
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700981 slub_debug = 0;
982 if (*str == '-')
983 /*
984 * Switch off all debugging measures.
985 */
986 goto out;
987
988 /*
989 * Determine which debug features should be switched on
990 */
Pekka Enberg06428782008-01-07 23:20:27 -0800991 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700992 switch (tolower(*str)) {
993 case 'f':
994 slub_debug |= SLAB_DEBUG_FREE;
995 break;
996 case 'z':
997 slub_debug |= SLAB_RED_ZONE;
998 break;
999 case 'p':
1000 slub_debug |= SLAB_POISON;
1001 break;
1002 case 'u':
1003 slub_debug |= SLAB_STORE_USER;
1004 break;
1005 case 't':
1006 slub_debug |= SLAB_TRACE;
1007 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001008 case 'a':
1009 slub_debug |= SLAB_FAILSLAB;
1010 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001011 default:
1012 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001013 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001014 }
1015 }
1016
1017check_slabs:
1018 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001019 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001020out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001021 return 1;
1022}
1023
1024__setup("slub_debug", setup_slub_debug);
1025
Christoph Lameterba0268a2007-09-11 15:24:11 -07001026static unsigned long kmem_cache_flags(unsigned long objsize,
1027 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001028 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001029{
1030 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001031 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001032 */
Christoph Lametere1533622008-02-15 23:45:24 -08001033 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001034 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1035 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001036
1037 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001038}
1039#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001040static inline void setup_object_debug(struct kmem_cache *s,
1041 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001042
Christoph Lameter3ec09742007-05-16 22:11:00 -07001043static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001044 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001045
Christoph Lameter3ec09742007-05-16 22:11:00 -07001046static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001047 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001048
Christoph Lameter41ecc552007-05-09 02:32:44 -07001049static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1050 { return 1; }
1051static inline int check_object(struct kmem_cache *s, struct page *page,
1052 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001053static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001054static inline unsigned long kmem_cache_flags(unsigned long objsize,
1055 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001056 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001057{
1058 return flags;
1059}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001060#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001061
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001062#define disable_higher_order_debug 0
1063
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001064static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1065 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001066static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1067 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001068static inline void inc_slabs_node(struct kmem_cache *s, int node,
1069 int objects) {}
1070static inline void dec_slabs_node(struct kmem_cache *s, int node,
1071 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001072#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001073
Christoph Lameter81819f02007-05-06 14:49:36 -07001074/*
1075 * Slab allocation and freeing
1076 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001077static inline struct page *alloc_slab_page(gfp_t flags, int node,
1078 struct kmem_cache_order_objects oo)
1079{
1080 int order = oo_order(oo);
1081
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001082 flags |= __GFP_NOTRACK;
1083
Christoph Lameter65c33762008-04-14 19:11:40 +03001084 if (node == -1)
1085 return alloc_pages(flags, order);
1086 else
1087 return alloc_pages_node(node, flags, order);
1088}
1089
Christoph Lameter81819f02007-05-06 14:49:36 -07001090static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1091{
Pekka Enberg06428782008-01-07 23:20:27 -08001092 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001093 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001094 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001095
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001096 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001097
Pekka Enbergba522702009-06-24 21:59:51 +03001098 /*
1099 * Let the initial higher-order allocation fail under memory pressure
1100 * so we fall-back to the minimum order allocation.
1101 */
1102 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1103
1104 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001105 if (unlikely(!page)) {
1106 oo = s->min;
1107 /*
1108 * Allocation may have failed due to fragmentation.
1109 * Try a lower order alloc if possible
1110 */
1111 page = alloc_slab_page(flags, node, oo);
1112 if (!page)
1113 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001114
Christoph Lameter84e554e2009-12-18 16:26:23 -06001115 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001116 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001117
1118 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001119 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001120 int pages = 1 << oo_order(oo);
1121
1122 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1123
1124 /*
1125 * Objects from caches that have a constructor don't get
1126 * cleared when they're allocated, so we need to do it here.
1127 */
1128 if (s->ctor)
1129 kmemcheck_mark_uninitialized_pages(page, pages);
1130 else
1131 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001132 }
1133
Christoph Lameter834f3d12008-04-14 19:11:31 +03001134 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001135 mod_zone_page_state(page_zone(page),
1136 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1137 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001138 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001139
1140 return page;
1141}
1142
1143static void setup_object(struct kmem_cache *s, struct page *page,
1144 void *object)
1145{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001146 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001147 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001148 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001149}
1150
1151static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1152{
1153 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001154 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001155 void *last;
1156 void *p;
1157
Christoph Lameter6cb06222007-10-16 01:25:41 -07001158 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001159
Christoph Lameter6cb06222007-10-16 01:25:41 -07001160 page = allocate_slab(s,
1161 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001162 if (!page)
1163 goto out;
1164
Christoph Lameter205ab992008-04-14 19:11:40 +03001165 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001166 page->slab = s;
1167 page->flags |= 1 << PG_slab;
1168 if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
1169 SLAB_STORE_USER | SLAB_TRACE))
Andy Whitcroft8a380822008-07-23 21:27:18 -07001170 __SetPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001171
1172 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001173
1174 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001175 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001176
1177 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001178 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001179 setup_object(s, page, last);
1180 set_freepointer(s, last, p);
1181 last = p;
1182 }
1183 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001184 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001185
1186 page->freelist = start;
1187 page->inuse = 0;
1188out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001189 return page;
1190}
1191
1192static void __free_slab(struct kmem_cache *s, struct page *page)
1193{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001194 int order = compound_order(page);
1195 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001196
Andy Whitcroft8a380822008-07-23 21:27:18 -07001197 if (unlikely(SLABDEBUG && PageSlubDebug(page))) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001198 void *p;
1199
1200 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001201 for_each_object(p, s, page_address(page),
1202 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001203 check_object(s, page, p, 0);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001204 __ClearPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001205 }
1206
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001207 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001208
Christoph Lameter81819f02007-05-06 14:49:36 -07001209 mod_zone_page_state(page_zone(page),
1210 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1211 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001212 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001213
Christoph Lameter49bd5222008-04-14 18:52:18 +03001214 __ClearPageSlab(page);
1215 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001216 if (current->reclaim_state)
1217 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001218 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001219}
1220
1221static void rcu_free_slab(struct rcu_head *h)
1222{
1223 struct page *page;
1224
1225 page = container_of((struct list_head *)h, struct page, lru);
1226 __free_slab(page->slab, page);
1227}
1228
1229static void free_slab(struct kmem_cache *s, struct page *page)
1230{
1231 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1232 /*
1233 * RCU free overloads the RCU head over the LRU
1234 */
1235 struct rcu_head *head = (void *)&page->lru;
1236
1237 call_rcu(head, rcu_free_slab);
1238 } else
1239 __free_slab(s, page);
1240}
1241
1242static void discard_slab(struct kmem_cache *s, struct page *page)
1243{
Christoph Lameter205ab992008-04-14 19:11:40 +03001244 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001245 free_slab(s, page);
1246}
1247
1248/*
1249 * Per slab locking using the pagelock
1250 */
1251static __always_inline void slab_lock(struct page *page)
1252{
1253 bit_spin_lock(PG_locked, &page->flags);
1254}
1255
1256static __always_inline void slab_unlock(struct page *page)
1257{
Nick Piggina76d3542008-01-07 23:20:27 -08001258 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001259}
1260
1261static __always_inline int slab_trylock(struct page *page)
1262{
1263 int rc = 1;
1264
1265 rc = bit_spin_trylock(PG_locked, &page->flags);
1266 return rc;
1267}
1268
1269/*
1270 * Management of partially allocated slabs
1271 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001272static void add_partial(struct kmem_cache_node *n,
1273 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001274{
Christoph Lametere95eed52007-05-06 14:49:44 -07001275 spin_lock(&n->list_lock);
1276 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001277 if (tail)
1278 list_add_tail(&page->lru, &n->partial);
1279 else
1280 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001281 spin_unlock(&n->list_lock);
1282}
1283
Christoph Lameter0121c6192008-04-29 16:11:12 -07001284static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001285{
1286 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1287
1288 spin_lock(&n->list_lock);
1289 list_del(&page->lru);
1290 n->nr_partial--;
1291 spin_unlock(&n->list_lock);
1292}
1293
1294/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001295 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001296 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001297 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001298 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001299static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1300 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001301{
1302 if (slab_trylock(page)) {
1303 list_del(&page->lru);
1304 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001305 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001306 return 1;
1307 }
1308 return 0;
1309}
1310
1311/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001312 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001313 */
1314static struct page *get_partial_node(struct kmem_cache_node *n)
1315{
1316 struct page *page;
1317
1318 /*
1319 * Racy check. If we mistakenly see no partial slabs then we
1320 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001321 * partial slab and there is none available then get_partials()
1322 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001323 */
1324 if (!n || !n->nr_partial)
1325 return NULL;
1326
1327 spin_lock(&n->list_lock);
1328 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001329 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001330 goto out;
1331 page = NULL;
1332out:
1333 spin_unlock(&n->list_lock);
1334 return page;
1335}
1336
1337/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001338 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001339 */
1340static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1341{
1342#ifdef CONFIG_NUMA
1343 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001344 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001345 struct zone *zone;
1346 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001347 struct page *page;
1348
1349 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001350 * The defrag ratio allows a configuration of the tradeoffs between
1351 * inter node defragmentation and node local allocations. A lower
1352 * defrag_ratio increases the tendency to do local allocations
1353 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001354 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001355 * If the defrag_ratio is set to 0 then kmalloc() always
1356 * returns node local objects. If the ratio is higher then kmalloc()
1357 * may return off node objects because partial slabs are obtained
1358 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001359 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001360 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001361 * defrag_ratio = 1000) then every (well almost) allocation will
1362 * first attempt to defrag slab caches on other nodes. This means
1363 * scanning over all nodes to look for partial slabs which may be
1364 * expensive if we do it every time we are trying to find a slab
1365 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001366 */
Christoph Lameter98246012008-01-07 23:20:26 -08001367 if (!s->remote_node_defrag_ratio ||
1368 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001369 return NULL;
1370
Mel Gorman0e884602008-04-28 02:12:14 -07001371 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001372 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001373 struct kmem_cache_node *n;
1374
Mel Gorman54a6eb52008-04-28 02:12:16 -07001375 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001376
Mel Gorman54a6eb52008-04-28 02:12:16 -07001377 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001378 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001379 page = get_partial_node(n);
1380 if (page)
1381 return page;
1382 }
1383 }
1384#endif
1385 return NULL;
1386}
1387
1388/*
1389 * Get a partial page, lock it and return it.
1390 */
1391static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1392{
1393 struct page *page;
1394 int searchnode = (node == -1) ? numa_node_id() : node;
1395
1396 page = get_partial_node(get_node(s, searchnode));
1397 if (page || (flags & __GFP_THISNODE))
1398 return page;
1399
1400 return get_any_partial(s, flags);
1401}
1402
1403/*
1404 * Move a page back to the lists.
1405 *
1406 * Must be called with the slab lock held.
1407 *
1408 * On exit the slab lock will have been dropped.
1409 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001410static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001411{
Christoph Lametere95eed52007-05-06 14:49:44 -07001412 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1413
Andy Whitcroft8a380822008-07-23 21:27:18 -07001414 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001415 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001416
Christoph Lametera973e9d2008-03-01 13:40:44 -08001417 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001418 add_partial(n, page, tail);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001419 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001420 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001421 stat(s, DEACTIVATE_FULL);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001422 if (SLABDEBUG && PageSlubDebug(page) &&
1423 (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001424 add_full(n, page);
1425 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001426 slab_unlock(page);
1427 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001428 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001429 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001430 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001431 * Adding an empty slab to the partial slabs in order
1432 * to avoid page allocator overhead. This slab needs
1433 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001434 * so that the others get filled first. That way the
1435 * size of the partial list stays small.
1436 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001437 * kmem_cache_shrink can reclaim any empty slabs from
1438 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001439 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001440 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001441 slab_unlock(page);
1442 } else {
1443 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001444 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001445 discard_slab(s, page);
1446 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001447 }
1448}
1449
1450/*
1451 * Remove the cpu slab
1452 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001453static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001454{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001455 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001456 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001457
Christoph Lameterb773ad72008-03-04 11:10:17 -08001458 if (page->freelist)
Christoph Lameter84e554e2009-12-18 16:26:23 -06001459 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001460 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001461 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001462 * because both freelists are empty. So this is unlikely
1463 * to occur.
1464 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001465 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001466 void **object;
1467
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001468 tail = 0; /* Hot objects. Put the slab first */
1469
Christoph Lameter894b8782007-05-10 03:15:16 -07001470 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001471 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001472 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001473
1474 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001475 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001476 page->freelist = object;
1477 page->inuse--;
1478 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001479 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001480 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001481}
1482
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001483static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001484{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001485 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001486 slab_lock(c->page);
1487 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001488}
1489
1490/*
1491 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001492 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001493 * Called from IPI handler with interrupts disabled.
1494 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001495static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001496{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001497 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001498
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001499 if (likely(c && c->page))
1500 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001501}
1502
1503static void flush_cpu_slab(void *d)
1504{
1505 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001506
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001507 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001508}
1509
1510static void flush_all(struct kmem_cache *s)
1511{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001512 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001513}
1514
1515/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001516 * Check if the objects in a per cpu structure fit numa
1517 * locality expectations.
1518 */
1519static inline int node_match(struct kmem_cache_cpu *c, int node)
1520{
1521#ifdef CONFIG_NUMA
1522 if (node != -1 && c->node != node)
1523 return 0;
1524#endif
1525 return 1;
1526}
1527
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001528static int count_free(struct page *page)
1529{
1530 return page->objects - page->inuse;
1531}
1532
1533static unsigned long count_partial(struct kmem_cache_node *n,
1534 int (*get_count)(struct page *))
1535{
1536 unsigned long flags;
1537 unsigned long x = 0;
1538 struct page *page;
1539
1540 spin_lock_irqsave(&n->list_lock, flags);
1541 list_for_each_entry(page, &n->partial, lru)
1542 x += get_count(page);
1543 spin_unlock_irqrestore(&n->list_lock, flags);
1544 return x;
1545}
1546
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001547static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1548{
1549#ifdef CONFIG_SLUB_DEBUG
1550 return atomic_long_read(&n->total_objects);
1551#else
1552 return 0;
1553#endif
1554}
1555
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001556static noinline void
1557slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1558{
1559 int node;
1560
1561 printk(KERN_WARNING
1562 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1563 nid, gfpflags);
1564 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1565 "default order: %d, min order: %d\n", s->name, s->objsize,
1566 s->size, oo_order(s->oo), oo_order(s->min));
1567
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001568 if (oo_order(s->min) > get_order(s->objsize))
1569 printk(KERN_WARNING " %s debugging increased min order, use "
1570 "slub_debug=O to disable.\n", s->name);
1571
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001572 for_each_online_node(node) {
1573 struct kmem_cache_node *n = get_node(s, node);
1574 unsigned long nr_slabs;
1575 unsigned long nr_objs;
1576 unsigned long nr_free;
1577
1578 if (!n)
1579 continue;
1580
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001581 nr_free = count_partial(n, count_free);
1582 nr_slabs = node_nr_slabs(n);
1583 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001584
1585 printk(KERN_WARNING
1586 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1587 node, nr_slabs, nr_objs, nr_free);
1588 }
1589}
1590
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001591/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001592 * Slow path. The lockless freelist is empty or we need to perform
1593 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001594 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001595 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001596 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001597 * Processing is still very fast if new objects have been freed to the
1598 * regular freelist. In that case we simply take over the regular freelist
1599 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001600 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001601 * If that is not working then we fall back to the partial lists. We take the
1602 * first element of the freelist as the object to allocate now and move the
1603 * rest of the freelist to the lockless freelist.
1604 *
1605 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001606 * we need to allocate a new slab. This is the slowest path since it involves
1607 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001608 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001609static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1610 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001611{
Christoph Lameter81819f02007-05-06 14:49:36 -07001612 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001613 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001614
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001615 /* We handle __GFP_ZERO in the caller */
1616 gfpflags &= ~__GFP_ZERO;
1617
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001618 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001619 goto new_slab;
1620
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001621 slab_lock(c->page);
1622 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001623 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001624
Christoph Lameter84e554e2009-12-18 16:26:23 -06001625 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001626
Christoph Lameter894b8782007-05-10 03:15:16 -07001627load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001628 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001629 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001630 goto another_slab;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001631 if (unlikely(SLABDEBUG && PageSlubDebug(c->page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001632 goto debug;
1633
Christoph Lameterff120592009-12-18 16:26:22 -06001634 c->freelist = get_freepointer(s, object);
Christoph Lameter39b26462008-04-14 19:11:30 +03001635 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001636 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001637 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001638unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001639 slab_unlock(c->page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001640 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001641 return object;
1642
1643another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001644 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001645
1646new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001647 new = get_partial(s, gfpflags, node);
1648 if (new) {
1649 c->page = new;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001650 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001651 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001652 }
1653
Christoph Lameterb811c202007-10-16 23:25:51 -07001654 if (gfpflags & __GFP_WAIT)
1655 local_irq_enable();
1656
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001657 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001658
1659 if (gfpflags & __GFP_WAIT)
1660 local_irq_disable();
1661
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001662 if (new) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001663 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001664 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001665 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001666 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001667 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001668 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001669 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001670 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001671 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001672 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1673 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001674 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001675debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001676 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001677 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001678
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001679 c->page->inuse++;
Christoph Lameterff120592009-12-18 16:26:22 -06001680 c->page->freelist = get_freepointer(s, object);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001681 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001682 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001683}
1684
1685/*
1686 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1687 * have the fastpath folded into their functions. So no function call
1688 * overhead for requests that can be satisfied on the fastpath.
1689 *
1690 * The fastpath works by first checking if the lockless freelist can be used.
1691 * If not then __slab_alloc is called for slow processing.
1692 *
1693 * Otherwise we can simply pick the next object from the lockless free list.
1694 */
Pekka Enberg06428782008-01-07 23:20:27 -08001695static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001696 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001697{
Christoph Lameter894b8782007-05-10 03:15:16 -07001698 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001699 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001700 unsigned long flags;
1701
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001702 gfpflags &= gfp_allowed_mask;
Pekka Enberg7e85ee02009-06-12 14:03:06 +03001703
Nick Piggincf40bd12009-01-21 08:12:39 +01001704 lockdep_trace_alloc(gfpflags);
OGAWA Hirofumi89124d72008-11-19 21:23:59 +09001705 might_sleep_if(gfpflags & __GFP_WAIT);
Pekka Enberg3c506ef2008-12-29 11:47:05 +02001706
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001707 if (should_failslab(s->objsize, gfpflags, s->flags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001708 return NULL;
1709
Christoph Lameter894b8782007-05-10 03:15:16 -07001710 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001711 c = __this_cpu_ptr(s->cpu_slab);
1712 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001713 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001714
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001715 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001716
1717 else {
Christoph Lameterff120592009-12-18 16:26:22 -06001718 c->freelist = get_freepointer(s, object);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001719 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001720 }
1721 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001722
Pekka Enberg74e21342009-11-25 20:14:48 +02001723 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001724 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001725
Christoph Lameterff120592009-12-18 16:26:22 -06001726 kmemcheck_slab_alloc(s, gfpflags, object, s->objsize);
1727 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, gfpflags);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001728
Christoph Lameter894b8782007-05-10 03:15:16 -07001729 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001730}
1731
1732void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1733{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001734 void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_);
1735
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001736 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001737
1738 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001739}
1740EXPORT_SYMBOL(kmem_cache_alloc);
1741
Li Zefan0f24f122009-12-11 15:45:30 +08001742#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001743void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1744{
1745 return slab_alloc(s, gfpflags, -1, _RET_IP_);
1746}
1747EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1748#endif
1749
Christoph Lameter81819f02007-05-06 14:49:36 -07001750#ifdef CONFIG_NUMA
1751void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1752{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001753 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1754
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001755 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1756 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001757
1758 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001759}
1760EXPORT_SYMBOL(kmem_cache_alloc_node);
1761#endif
1762
Li Zefan0f24f122009-12-11 15:45:30 +08001763#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001764void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1765 gfp_t gfpflags,
1766 int node)
1767{
1768 return slab_alloc(s, gfpflags, node, _RET_IP_);
1769}
1770EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1771#endif
1772
Christoph Lameter81819f02007-05-06 14:49:36 -07001773/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001774 * Slow patch handling. This may still be called frequently since objects
1775 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001776 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001777 * So we still attempt to reduce cache line usage. Just take the slab
1778 * lock and free the item. If there is no additional partial page
1779 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001780 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001781static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06001782 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001783{
1784 void *prior;
1785 void **object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07001786
Christoph Lameter84e554e2009-12-18 16:26:23 -06001787 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001788 slab_lock(page);
1789
Andy Whitcroft8a380822008-07-23 21:27:18 -07001790 if (unlikely(SLABDEBUG && PageSlubDebug(page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001791 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001792
Christoph Lameter81819f02007-05-06 14:49:36 -07001793checks_ok:
Christoph Lameterff120592009-12-18 16:26:22 -06001794 prior = page->freelist;
1795 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07001796 page->freelist = object;
1797 page->inuse--;
1798
Andy Whitcroft8a380822008-07-23 21:27:18 -07001799 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001800 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001801 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001802 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001803
1804 if (unlikely(!page->inuse))
1805 goto slab_empty;
1806
1807 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001808 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001809 * then add it.
1810 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001811 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001812 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001813 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001814 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001815
1816out_unlock:
1817 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001818 return;
1819
1820slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001821 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001822 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001823 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001824 */
1825 remove_partial(s, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001826 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001827 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001828 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001829 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001830 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001831 return;
1832
1833debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001834 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001835 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001836 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001837}
1838
Christoph Lameter894b8782007-05-10 03:15:16 -07001839/*
1840 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1841 * can perform fastpath freeing without additional function calls.
1842 *
1843 * The fastpath is only possible if we are freeing to the current cpu slab
1844 * of this processor. This typically the case if we have just allocated
1845 * the item before.
1846 *
1847 * If fastpath is not possible then fall back to __slab_free where we deal
1848 * with all sorts of special processing.
1849 */
Pekka Enberg06428782008-01-07 23:20:27 -08001850static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001851 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001852{
1853 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001854 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001855 unsigned long flags;
1856
Catalin Marinas06f22f12009-06-11 13:23:18 +01001857 kmemleak_free_recursive(x, s->flags);
Christoph Lameter894b8782007-05-10 03:15:16 -07001858 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001859 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterff120592009-12-18 16:26:22 -06001860 kmemcheck_slab_free(s, object, s->objsize);
1861 debug_check_no_locks_freed(object, s->objsize);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001862 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameterff120592009-12-18 16:26:22 -06001863 debug_check_no_obj_freed(object, s->objsize);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001864 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterff120592009-12-18 16:26:22 -06001865 set_freepointer(s, object, c->freelist);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001866 c->freelist = object;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001867 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001868 } else
Christoph Lameterff120592009-12-18 16:26:22 -06001869 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07001870
1871 local_irq_restore(flags);
1872}
1873
Christoph Lameter81819f02007-05-06 14:49:36 -07001874void kmem_cache_free(struct kmem_cache *s, void *x)
1875{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001876 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001877
Christoph Lameterb49af682007-05-06 14:49:41 -07001878 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001879
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001880 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001881
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001882 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001883}
1884EXPORT_SYMBOL(kmem_cache_free);
1885
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001886/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001887static struct page *get_object_page(const void *x)
1888{
Christoph Lameterb49af682007-05-06 14:49:41 -07001889 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001890
1891 if (!PageSlab(page))
1892 return NULL;
1893
1894 return page;
1895}
1896
1897/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001898 * Object placement in a slab is made very easy because we always start at
1899 * offset 0. If we tune the size of the object to the alignment then we can
1900 * get the required alignment by putting one properly sized object after
1901 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001902 *
1903 * Notice that the allocation order determines the sizes of the per cpu
1904 * caches. Each processor has always one slab available for allocations.
1905 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001906 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001907 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001908 */
1909
1910/*
1911 * Mininum / Maximum order of slab pages. This influences locking overhead
1912 * and slab fragmentation. A higher order reduces the number of partial slabs
1913 * and increases the number of allocations possible without having to
1914 * take the list_lock.
1915 */
1916static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001917static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001918static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001919
1920/*
1921 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001922 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001923 */
1924static int slub_nomerge;
1925
1926/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001927 * Calculate the order of allocation given an slab object size.
1928 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001929 * The order of allocation has significant impact on performance and other
1930 * system components. Generally order 0 allocations should be preferred since
1931 * order 0 does not cause fragmentation in the page allocator. Larger objects
1932 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001933 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001934 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001935 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001936 * In order to reach satisfactory performance we must ensure that a minimum
1937 * number of objects is in one slab. Otherwise we may generate too much
1938 * activity on the partial lists which requires taking the list_lock. This is
1939 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001940 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001941 * slub_max_order specifies the order where we begin to stop considering the
1942 * number of objects in a slab as critical. If we reach slub_max_order then
1943 * we try to keep the page order as low as possible. So we accept more waste
1944 * of space in favor of a small page order.
1945 *
1946 * Higher order allocations also allow the placement of more objects in a
1947 * slab and thereby reduce object handling overhead. If the user has
1948 * requested a higher mininum order then we start with that one instead of
1949 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001950 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001951static inline int slab_order(int size, int min_objects,
1952 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001953{
1954 int order;
1955 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001956 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001957
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001958 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1959 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001960
Christoph Lameter6300ea72007-07-17 04:03:20 -07001961 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001962 fls(min_objects * size - 1) - PAGE_SHIFT);
1963 order <= max_order; order++) {
1964
Christoph Lameter81819f02007-05-06 14:49:36 -07001965 unsigned long slab_size = PAGE_SIZE << order;
1966
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001967 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001968 continue;
1969
Christoph Lameter81819f02007-05-06 14:49:36 -07001970 rem = slab_size % size;
1971
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001972 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001973 break;
1974
1975 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001976
Christoph Lameter81819f02007-05-06 14:49:36 -07001977 return order;
1978}
1979
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001980static inline int calculate_order(int size)
1981{
1982 int order;
1983 int min_objects;
1984 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001985 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001986
1987 /*
1988 * Attempt to find best configuration for a slab. This
1989 * works by first attempting to generate a layout with
1990 * the best configuration and backing off gradually.
1991 *
1992 * First we reduce the acceptable waste in a slab. Then
1993 * we reduce the minimum objects required in a slab.
1994 */
1995 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001996 if (!min_objects)
1997 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001998 max_objects = (PAGE_SIZE << slub_max_order)/size;
1999 min_objects = min(min_objects, max_objects);
2000
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002001 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002002 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002003 while (fraction >= 4) {
2004 order = slab_order(size, min_objects,
2005 slub_max_order, fraction);
2006 if (order <= slub_max_order)
2007 return order;
2008 fraction /= 2;
2009 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002010 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002011 }
2012
2013 /*
2014 * We were unable to place multiple objects in a slab. Now
2015 * lets see if we can place a single object there.
2016 */
2017 order = slab_order(size, 1, slub_max_order, 1);
2018 if (order <= slub_max_order)
2019 return order;
2020
2021 /*
2022 * Doh this slab cannot be placed using slub_max_order.
2023 */
2024 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002025 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002026 return order;
2027 return -ENOSYS;
2028}
2029
Christoph Lameter81819f02007-05-06 14:49:36 -07002030/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002031 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002032 */
2033static unsigned long calculate_alignment(unsigned long flags,
2034 unsigned long align, unsigned long size)
2035{
2036 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002037 * If the user wants hardware cache aligned objects then follow that
2038 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002039 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002040 * The hardware cache alignment cannot override the specified
2041 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002042 */
Nick Pigginb6210382008-03-05 14:05:56 -08002043 if (flags & SLAB_HWCACHE_ALIGN) {
2044 unsigned long ralign = cache_line_size();
2045 while (size <= ralign / 2)
2046 ralign /= 2;
2047 align = max(align, ralign);
2048 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002049
2050 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002051 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002052
2053 return ALIGN(align, sizeof(void *));
2054}
2055
Pekka Enberg5595cff2008-08-05 09:28:47 +03002056static void
2057init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002058{
2059 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002060 spin_lock_init(&n->list_lock);
2061 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002062#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002063 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002064 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002065 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002066#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002067}
2068
Christoph Lameter91efd772010-01-21 17:43:35 -06002069static DEFINE_PER_CPU(struct kmem_cache_cpu, kmalloc_percpu[KMALLOC_CACHES]);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002070
2071static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2072{
Christoph Lameter756dee72009-12-18 16:26:21 -06002073 if (s < kmalloc_caches + KMALLOC_CACHES && s >= kmalloc_caches)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002074 /*
2075 * Boot time creation of the kmalloc array. Use static per cpu data
2076 * since the per cpu allocator is not available yet.
2077 */
Stephen Rothwell1154fab2010-03-01 16:04:45 +11002078 s->cpu_slab = kmalloc_percpu + (s - kmalloc_caches);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002079 else
2080 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2081
2082 if (!s->cpu_slab)
2083 return 0;
2084
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002085 return 1;
2086}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002087
Christoph Lameter81819f02007-05-06 14:49:36 -07002088#ifdef CONFIG_NUMA
2089/*
2090 * No kmalloc_node yet so do it by hand. We know that this is the first
2091 * slab on the node for this slabcache. There are no concurrent accesses
2092 * possible.
2093 *
2094 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002095 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2096 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002097 */
David Rientjes0094de92008-11-25 19:14:19 -08002098static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002099{
2100 struct page *page;
2101 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002102 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002103
2104 BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2105
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002106 page = new_slab(kmalloc_caches, gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002107
2108 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002109 if (page_to_nid(page) != node) {
2110 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2111 "node %d\n", node);
2112 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2113 "in order to be able to continue\n");
2114 }
2115
Christoph Lameter81819f02007-05-06 14:49:36 -07002116 n = page->freelist;
2117 BUG_ON(!n);
2118 page->freelist = get_freepointer(kmalloc_caches, n);
2119 page->inuse++;
2120 kmalloc_caches->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002121#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterd45f39c2007-07-17 04:03:21 -07002122 init_object(kmalloc_caches, n, 1);
2123 init_tracking(kmalloc_caches, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002124#endif
Pekka Enberg5595cff2008-08-05 09:28:47 +03002125 init_kmem_cache_node(n, kmalloc_caches);
Christoph Lameter205ab992008-04-14 19:11:40 +03002126 inc_slabs_node(kmalloc_caches, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002127
rootba84c732008-01-07 23:20:28 -08002128 /*
2129 * lockdep requires consistent irq usage for each lock
2130 * so even though there cannot be a race this early in
2131 * the boot sequence, we still disable irqs.
2132 */
2133 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002134 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002135 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002136}
2137
2138static void free_kmem_cache_nodes(struct kmem_cache *s)
2139{
2140 int node;
2141
Christoph Lameterf64dc582007-10-16 01:25:33 -07002142 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002143 struct kmem_cache_node *n = s->node[node];
2144 if (n && n != &s->local_node)
2145 kmem_cache_free(kmalloc_caches, n);
2146 s->node[node] = NULL;
2147 }
2148}
2149
2150static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2151{
2152 int node;
2153 int local_node;
2154
Christoph Lameter91efd772010-01-21 17:43:35 -06002155 if (slab_state >= UP && (s < kmalloc_caches ||
Zhang, Yanmin111c7d82010-04-01 17:32:30 +08002156 s >= kmalloc_caches + KMALLOC_CACHES))
Christoph Lameter81819f02007-05-06 14:49:36 -07002157 local_node = page_to_nid(virt_to_page(s));
2158 else
2159 local_node = 0;
2160
Christoph Lameterf64dc582007-10-16 01:25:33 -07002161 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002162 struct kmem_cache_node *n;
2163
2164 if (local_node == node)
2165 n = &s->local_node;
2166 else {
2167 if (slab_state == DOWN) {
David Rientjes0094de92008-11-25 19:14:19 -08002168 early_kmem_cache_node_alloc(gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002169 continue;
2170 }
2171 n = kmem_cache_alloc_node(kmalloc_caches,
2172 gfpflags, node);
2173
2174 if (!n) {
2175 free_kmem_cache_nodes(s);
2176 return 0;
2177 }
2178
2179 }
2180 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002181 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002182 }
2183 return 1;
2184}
2185#else
2186static void free_kmem_cache_nodes(struct kmem_cache *s)
2187{
2188}
2189
2190static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2191{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002192 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002193 return 1;
2194}
2195#endif
2196
David Rientjesc0bdb232009-02-25 09:16:35 +02002197static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002198{
2199 if (min < MIN_PARTIAL)
2200 min = MIN_PARTIAL;
2201 else if (min > MAX_PARTIAL)
2202 min = MAX_PARTIAL;
2203 s->min_partial = min;
2204}
2205
Christoph Lameter81819f02007-05-06 14:49:36 -07002206/*
2207 * calculate_sizes() determines the order and the distribution of data within
2208 * a slab object.
2209 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002210static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002211{
2212 unsigned long flags = s->flags;
2213 unsigned long size = s->objsize;
2214 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002215 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002216
2217 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002218 * Round up object size to the next word boundary. We can only
2219 * place the free pointer at word boundaries and this determines
2220 * the possible location of the free pointer.
2221 */
2222 size = ALIGN(size, sizeof(void *));
2223
2224#ifdef CONFIG_SLUB_DEBUG
2225 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002226 * Determine if we can poison the object itself. If the user of
2227 * the slab may touch the object after free or before allocation
2228 * then we should never poison the object itself.
2229 */
2230 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002231 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002232 s->flags |= __OBJECT_POISON;
2233 else
2234 s->flags &= ~__OBJECT_POISON;
2235
Christoph Lameter81819f02007-05-06 14:49:36 -07002236
2237 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002238 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002239 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002240 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002241 */
2242 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2243 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002244#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002245
2246 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002247 * With that we have determined the number of bytes in actual use
2248 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002249 */
2250 s->inuse = size;
2251
2252 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002253 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002254 /*
2255 * Relocate free pointer after the object if it is not
2256 * permitted to overwrite the first word of the object on
2257 * kmem_cache_free.
2258 *
2259 * This is the case if we do RCU, have a constructor or
2260 * destructor or are poisoning the objects.
2261 */
2262 s->offset = size;
2263 size += sizeof(void *);
2264 }
2265
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002266#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002267 if (flags & SLAB_STORE_USER)
2268 /*
2269 * Need to store information about allocs and frees after
2270 * the object.
2271 */
2272 size += 2 * sizeof(struct track);
2273
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002274 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002275 /*
2276 * Add some empty padding so that we can catch
2277 * overwrites from earlier objects rather than let
2278 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002279 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002280 * of the object.
2281 */
2282 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002283#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002284
Christoph Lameter81819f02007-05-06 14:49:36 -07002285 /*
2286 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002287 * user specified and the dynamic determination of cache line size
2288 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002289 */
2290 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002291 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002292
2293 /*
2294 * SLUB stores one object immediately after another beginning from
2295 * offset 0. In order to align the objects we have to simply size
2296 * each object to conform to the alignment.
2297 */
2298 size = ALIGN(size, align);
2299 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002300 if (forced_order >= 0)
2301 order = forced_order;
2302 else
2303 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002304
Christoph Lameter834f3d12008-04-14 19:11:31 +03002305 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002306 return 0;
2307
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002308 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002309 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002310 s->allocflags |= __GFP_COMP;
2311
2312 if (s->flags & SLAB_CACHE_DMA)
2313 s->allocflags |= SLUB_DMA;
2314
2315 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2316 s->allocflags |= __GFP_RECLAIMABLE;
2317
Christoph Lameter81819f02007-05-06 14:49:36 -07002318 /*
2319 * Determine the number of objects per slab
2320 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002321 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002322 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002323 if (oo_objects(s->oo) > oo_objects(s->max))
2324 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002325
Christoph Lameter834f3d12008-04-14 19:11:31 +03002326 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002327
2328}
2329
Christoph Lameter81819f02007-05-06 14:49:36 -07002330static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2331 const char *name, size_t size,
2332 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002333 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002334{
2335 memset(s, 0, kmem_size);
2336 s->name = name;
2337 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002338 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002339 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002340 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002341
Christoph Lameter06b285d2008-04-14 19:11:41 +03002342 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002343 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002344 if (disable_higher_order_debug) {
2345 /*
2346 * Disable debugging flags that store metadata if the min slab
2347 * order increased.
2348 */
2349 if (get_order(s->size) > get_order(s->objsize)) {
2350 s->flags &= ~DEBUG_METADATA_FLAGS;
2351 s->offset = 0;
2352 if (!calculate_sizes(s, -1))
2353 goto error;
2354 }
2355 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002356
David Rientjes3b89d7d2009-02-22 17:40:07 -08002357 /*
2358 * The larger the object size is, the more pages we want on the partial
2359 * list to avoid pounding the page allocator excessively.
2360 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002361 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002362 s->refcount = 1;
2363#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002364 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002365#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002366 if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2367 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002368
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002369 if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
Christoph Lameter81819f02007-05-06 14:49:36 -07002370 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002371
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002372 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002373error:
2374 if (flags & SLAB_PANIC)
2375 panic("Cannot create slab %s size=%lu realsize=%u "
2376 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002377 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002378 s->offset, flags);
2379 return 0;
2380}
Christoph Lameter81819f02007-05-06 14:49:36 -07002381
2382/*
2383 * Check if a given pointer is valid
2384 */
2385int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2386{
Pekka Enberg06428782008-01-07 23:20:27 -08002387 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002388
Pekka Enbergd3e06e22010-04-07 19:23:41 +03002389 if (!kern_ptr_validate(object, s->size))
2390 return 0;
2391
Christoph Lameter81819f02007-05-06 14:49:36 -07002392 page = get_object_page(object);
2393
2394 if (!page || s != page->slab)
2395 /* No slab or wrong slab */
2396 return 0;
2397
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002398 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002399 return 0;
2400
2401 /*
2402 * We could also check if the object is on the slabs freelist.
2403 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002404 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002405 * to a certain slab.
2406 */
2407 return 1;
2408}
2409EXPORT_SYMBOL(kmem_ptr_validate);
2410
2411/*
2412 * Determine the size of a slab object
2413 */
2414unsigned int kmem_cache_size(struct kmem_cache *s)
2415{
2416 return s->objsize;
2417}
2418EXPORT_SYMBOL(kmem_cache_size);
2419
2420const char *kmem_cache_name(struct kmem_cache *s)
2421{
2422 return s->name;
2423}
2424EXPORT_SYMBOL(kmem_cache_name);
2425
Christoph Lameter33b12c32008-04-25 12:22:43 -07002426static void list_slab_objects(struct kmem_cache *s, struct page *page,
2427 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002428{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002429#ifdef CONFIG_SLUB_DEBUG
2430 void *addr = page_address(page);
2431 void *p;
2432 DECLARE_BITMAP(map, page->objects);
2433
2434 bitmap_zero(map, page->objects);
2435 slab_err(s, page, "%s", text);
2436 slab_lock(page);
2437 for_each_free_object(p, s, page->freelist)
2438 set_bit(slab_index(p, s, addr), map);
2439
2440 for_each_object(p, s, addr, page->objects) {
2441
2442 if (!test_bit(slab_index(p, s, addr), map)) {
2443 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2444 p, p - addr);
2445 print_tracking(s, p);
2446 }
2447 }
2448 slab_unlock(page);
2449#endif
2450}
2451
Christoph Lameter81819f02007-05-06 14:49:36 -07002452/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002453 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002454 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002455static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002456{
Christoph Lameter81819f02007-05-06 14:49:36 -07002457 unsigned long flags;
2458 struct page *page, *h;
2459
2460 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002461 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002462 if (!page->inuse) {
2463 list_del(&page->lru);
2464 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002465 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002466 } else {
2467 list_slab_objects(s, page,
2468 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002469 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002470 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002471 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002472}
2473
2474/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002475 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002476 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002477static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002478{
2479 int node;
2480
2481 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002482 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002483 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002484 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002485 struct kmem_cache_node *n = get_node(s, node);
2486
Christoph Lameter599870b2008-04-23 12:36:52 -07002487 free_partial(s, n);
2488 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002489 return 1;
2490 }
2491 free_kmem_cache_nodes(s);
2492 return 0;
2493}
2494
2495/*
2496 * Close a cache and release the kmem_cache structure
2497 * (must be used for caches created using kmem_cache_create)
2498 */
2499void kmem_cache_destroy(struct kmem_cache *s)
2500{
2501 down_write(&slub_lock);
2502 s->refcount--;
2503 if (!s->refcount) {
2504 list_del(&s->list);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002505 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03002506 if (kmem_cache_close(s)) {
2507 printk(KERN_ERR "SLUB %s: %s called for cache that "
2508 "still has objects.\n", s->name, __func__);
2509 dump_stack();
2510 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002511 if (s->flags & SLAB_DESTROY_BY_RCU)
2512 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002513 sysfs_slab_remove(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002514 } else
2515 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002516}
2517EXPORT_SYMBOL(kmem_cache_destroy);
2518
2519/********************************************************************
2520 * Kmalloc subsystem
2521 *******************************************************************/
2522
Christoph Lameter756dee72009-12-18 16:26:21 -06002523struct kmem_cache kmalloc_caches[KMALLOC_CACHES] __cacheline_aligned;
Christoph Lameter81819f02007-05-06 14:49:36 -07002524EXPORT_SYMBOL(kmalloc_caches);
2525
Christoph Lameter81819f02007-05-06 14:49:36 -07002526static int __init setup_slub_min_order(char *str)
2527{
Pekka Enberg06428782008-01-07 23:20:27 -08002528 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002529
2530 return 1;
2531}
2532
2533__setup("slub_min_order=", setup_slub_min_order);
2534
2535static int __init setup_slub_max_order(char *str)
2536{
Pekka Enberg06428782008-01-07 23:20:27 -08002537 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002538 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002539
2540 return 1;
2541}
2542
2543__setup("slub_max_order=", setup_slub_max_order);
2544
2545static int __init setup_slub_min_objects(char *str)
2546{
Pekka Enberg06428782008-01-07 23:20:27 -08002547 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002548
2549 return 1;
2550}
2551
2552__setup("slub_min_objects=", setup_slub_min_objects);
2553
2554static int __init setup_slub_nomerge(char *str)
2555{
2556 slub_nomerge = 1;
2557 return 1;
2558}
2559
2560__setup("slub_nomerge", setup_slub_nomerge);
2561
Christoph Lameter81819f02007-05-06 14:49:36 -07002562static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2563 const char *name, int size, gfp_t gfp_flags)
2564{
2565 unsigned int flags = 0;
2566
2567 if (gfp_flags & SLUB_DMA)
2568 flags = SLAB_CACHE_DMA;
2569
Pekka Enberg83b519e2009-06-10 19:40:04 +03002570 /*
2571 * This function is called with IRQs disabled during early-boot on
2572 * single CPU so there's no need to take slub_lock here.
2573 */
Christoph Lameter81819f02007-05-06 14:49:36 -07002574 if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002575 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002576 goto panic;
2577
2578 list_add(&s->list, &slab_caches);
Pekka Enberg83b519e2009-06-10 19:40:04 +03002579
Christoph Lameter81819f02007-05-06 14:49:36 -07002580 if (sysfs_slab_add(s))
2581 goto panic;
2582 return s;
2583
2584panic:
2585 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2586}
2587
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002588#ifdef CONFIG_ZONE_DMA
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002589static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
Christoph Lameter1ceef402007-08-07 15:11:48 -07002590
2591static void sysfs_add_func(struct work_struct *w)
2592{
2593 struct kmem_cache *s;
2594
2595 down_write(&slub_lock);
2596 list_for_each_entry(s, &slab_caches, list) {
2597 if (s->flags & __SYSFS_ADD_DEFERRED) {
2598 s->flags &= ~__SYSFS_ADD_DEFERRED;
2599 sysfs_slab_add(s);
2600 }
2601 }
2602 up_write(&slub_lock);
2603}
2604
2605static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2606
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002607static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2608{
2609 struct kmem_cache *s;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002610 char *text;
2611 size_t realsize;
Nick Piggin964cf352009-06-15 13:35:10 +03002612 unsigned long slabflags;
Christoph Lameter756dee72009-12-18 16:26:21 -06002613 int i;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002614
2615 s = kmalloc_caches_dma[index];
2616 if (s)
2617 return s;
2618
2619 /* Dynamically create dma cache */
Christoph Lameter1ceef402007-08-07 15:11:48 -07002620 if (flags & __GFP_WAIT)
2621 down_write(&slub_lock);
2622 else {
2623 if (!down_write_trylock(&slub_lock))
2624 goto out;
2625 }
2626
2627 if (kmalloc_caches_dma[index])
2628 goto unlock_out;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002629
Christoph Lameter7b55f622007-07-17 04:03:27 -07002630 realsize = kmalloc_caches[index].objsize;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08002631 text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2632 (unsigned int)realsize);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002633
Christoph Lameter756dee72009-12-18 16:26:21 -06002634 s = NULL;
2635 for (i = 0; i < KMALLOC_CACHES; i++)
2636 if (!kmalloc_caches[i].size)
2637 break;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002638
Christoph Lameter756dee72009-12-18 16:26:21 -06002639 BUG_ON(i >= KMALLOC_CACHES);
2640 s = kmalloc_caches + i;
Christoph Lameter1ceef402007-08-07 15:11:48 -07002641
Nick Piggin964cf352009-06-15 13:35:10 +03002642 /*
2643 * Must defer sysfs creation to a workqueue because we don't know
2644 * what context we are called from. Before sysfs comes up, we don't
2645 * need to do anything because our sysfs initcall will start by
2646 * adding all existing slabs to sysfs.
2647 */
Pekka Enberg5caf5c72009-06-17 08:30:54 +03002648 slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK;
Nick Piggin964cf352009-06-15 13:35:10 +03002649 if (slab_state >= SYSFS)
2650 slabflags |= __SYSFS_ADD_DEFERRED;
2651
David Rientjes7738dd92010-01-15 12:49:56 -08002652 if (!text || !kmem_cache_open(s, flags, text,
Nick Piggin964cf352009-06-15 13:35:10 +03002653 realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) {
Christoph Lameter756dee72009-12-18 16:26:21 -06002654 s->size = 0;
Christoph Lameter1ceef402007-08-07 15:11:48 -07002655 kfree(text);
2656 goto unlock_out;
Christoph Lameterdfce8642007-07-17 04:03:25 -07002657 }
Christoph Lameter1ceef402007-08-07 15:11:48 -07002658
2659 list_add(&s->list, &slab_caches);
2660 kmalloc_caches_dma[index] = s;
2661
Nick Piggin964cf352009-06-15 13:35:10 +03002662 if (slab_state >= SYSFS)
2663 schedule_work(&sysfs_add_work);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002664
2665unlock_out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002666 up_write(&slub_lock);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002667out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002668 return kmalloc_caches_dma[index];
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002669}
2670#endif
2671
Christoph Lameterf1b26332007-07-17 04:03:26 -07002672/*
2673 * Conversion table for small slabs sizes / 8 to the index in the
2674 * kmalloc array. This is necessary for slabs < 192 since we have non power
2675 * of two cache sizes there. The size of larger slabs can be determined using
2676 * fls.
2677 */
2678static s8 size_index[24] = {
2679 3, /* 8 */
2680 4, /* 16 */
2681 5, /* 24 */
2682 5, /* 32 */
2683 6, /* 40 */
2684 6, /* 48 */
2685 6, /* 56 */
2686 6, /* 64 */
2687 1, /* 72 */
2688 1, /* 80 */
2689 1, /* 88 */
2690 1, /* 96 */
2691 7, /* 104 */
2692 7, /* 112 */
2693 7, /* 120 */
2694 7, /* 128 */
2695 2, /* 136 */
2696 2, /* 144 */
2697 2, /* 152 */
2698 2, /* 160 */
2699 2, /* 168 */
2700 2, /* 176 */
2701 2, /* 184 */
2702 2 /* 192 */
2703};
2704
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002705static inline int size_index_elem(size_t bytes)
2706{
2707 return (bytes - 1) / 8;
2708}
2709
Christoph Lameter81819f02007-05-06 14:49:36 -07002710static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2711{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002712 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002713
Christoph Lameterf1b26332007-07-17 04:03:26 -07002714 if (size <= 192) {
2715 if (!size)
2716 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002717
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002718 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002719 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002720 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002721
2722#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002723 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002724 return dma_kmalloc_cache(index, flags);
Christoph Lameterf1b26332007-07-17 04:03:26 -07002725
Christoph Lameter81819f02007-05-06 14:49:36 -07002726#endif
2727 return &kmalloc_caches[index];
2728}
2729
2730void *__kmalloc(size_t size, gfp_t flags)
2731{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002732 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002733 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002734
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002735 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002736 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002737
2738 s = get_slab(size, flags);
2739
2740 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002741 return s;
2742
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002743 ret = slab_alloc(s, flags, -1, _RET_IP_);
2744
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002745 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002746
2747 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002748}
2749EXPORT_SYMBOL(__kmalloc);
2750
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002751static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2752{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002753 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002754 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002755
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002756 flags |= __GFP_COMP | __GFP_NOTRACK;
2757 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002758 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002759 ptr = page_address(page);
2760
2761 kmemleak_alloc(ptr, size, 1, flags);
2762 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002763}
2764
Christoph Lameter81819f02007-05-06 14:49:36 -07002765#ifdef CONFIG_NUMA
2766void *__kmalloc_node(size_t size, gfp_t flags, int node)
2767{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002768 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002769 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002770
Ingo Molnar057685c2009-02-20 12:15:30 +01002771 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002772 ret = kmalloc_large_node(size, flags, node);
2773
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002774 trace_kmalloc_node(_RET_IP_, ret,
2775 size, PAGE_SIZE << get_order(size),
2776 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002777
2778 return ret;
2779 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002780
2781 s = get_slab(size, flags);
2782
2783 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002784 return s;
2785
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002786 ret = slab_alloc(s, flags, node, _RET_IP_);
2787
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002788 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002789
2790 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002791}
2792EXPORT_SYMBOL(__kmalloc_node);
2793#endif
2794
2795size_t ksize(const void *object)
2796{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002797 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002798 struct kmem_cache *s;
2799
Christoph Lameteref8b4522007-10-16 01:24:46 -07002800 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002801 return 0;
2802
Vegard Nossum294a80a2007-12-04 23:45:30 -08002803 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002804
Pekka Enberg76994412008-05-22 19:22:25 +03002805 if (unlikely(!PageSlab(page))) {
2806 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002807 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002808 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002809 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002810
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002811#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002812 /*
2813 * Debugging requires use of the padding between object
2814 * and whatever may come after it.
2815 */
2816 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2817 return s->objsize;
2818
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002819#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002820 /*
2821 * If we have the need to store the freelist pointer
2822 * back there or track user information then we can
2823 * only use the space before that information.
2824 */
2825 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2826 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002827 /*
2828 * Else we can use all the padding etc for the allocation
2829 */
2830 return s->size;
2831}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002832EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002833
2834void kfree(const void *x)
2835{
Christoph Lameter81819f02007-05-06 14:49:36 -07002836 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002837 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002838
Pekka Enberg2121db72009-03-25 11:05:57 +02002839 trace_kfree(_RET_IP_, x);
2840
Satyam Sharma2408c552007-10-16 01:24:44 -07002841 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002842 return;
2843
Christoph Lameterb49af682007-05-06 14:49:41 -07002844 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002845 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002846 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002847 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002848 put_page(page);
2849 return;
2850 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002851 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002852}
2853EXPORT_SYMBOL(kfree);
2854
Christoph Lameter2086d262007-05-06 14:49:46 -07002855/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002856 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2857 * the remaining slabs by the number of items in use. The slabs with the
2858 * most items in use come first. New allocations will then fill those up
2859 * and thus they can be removed from the partial lists.
2860 *
2861 * The slabs with the least items are placed last. This results in them
2862 * being allocated from last increasing the chance that the last objects
2863 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002864 */
2865int kmem_cache_shrink(struct kmem_cache *s)
2866{
2867 int node;
2868 int i;
2869 struct kmem_cache_node *n;
2870 struct page *page;
2871 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002872 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002873 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002874 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002875 unsigned long flags;
2876
2877 if (!slabs_by_inuse)
2878 return -ENOMEM;
2879
2880 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002881 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002882 n = get_node(s, node);
2883
2884 if (!n->nr_partial)
2885 continue;
2886
Christoph Lameter834f3d12008-04-14 19:11:31 +03002887 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002888 INIT_LIST_HEAD(slabs_by_inuse + i);
2889
2890 spin_lock_irqsave(&n->list_lock, flags);
2891
2892 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002893 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002894 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002895 * Note that concurrent frees may occur while we hold the
2896 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002897 */
2898 list_for_each_entry_safe(page, t, &n->partial, lru) {
2899 if (!page->inuse && slab_trylock(page)) {
2900 /*
2901 * Must hold slab lock here because slab_free
2902 * may have freed the last object and be
2903 * waiting to release the slab.
2904 */
2905 list_del(&page->lru);
2906 n->nr_partial--;
2907 slab_unlock(page);
2908 discard_slab(s, page);
2909 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002910 list_move(&page->lru,
2911 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002912 }
2913 }
2914
Christoph Lameter2086d262007-05-06 14:49:46 -07002915 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002916 * Rebuild the partial list with the slabs filled up most
2917 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002918 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002919 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002920 list_splice(slabs_by_inuse + i, n->partial.prev);
2921
Christoph Lameter2086d262007-05-06 14:49:46 -07002922 spin_unlock_irqrestore(&n->list_lock, flags);
2923 }
2924
2925 kfree(slabs_by_inuse);
2926 return 0;
2927}
2928EXPORT_SYMBOL(kmem_cache_shrink);
2929
Yasunori Gotob9049e22007-10-21 16:41:37 -07002930#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2931static int slab_mem_going_offline_callback(void *arg)
2932{
2933 struct kmem_cache *s;
2934
2935 down_read(&slub_lock);
2936 list_for_each_entry(s, &slab_caches, list)
2937 kmem_cache_shrink(s);
2938 up_read(&slub_lock);
2939
2940 return 0;
2941}
2942
2943static void slab_mem_offline_callback(void *arg)
2944{
2945 struct kmem_cache_node *n;
2946 struct kmem_cache *s;
2947 struct memory_notify *marg = arg;
2948 int offline_node;
2949
2950 offline_node = marg->status_change_nid;
2951
2952 /*
2953 * If the node still has available memory. we need kmem_cache_node
2954 * for it yet.
2955 */
2956 if (offline_node < 0)
2957 return;
2958
2959 down_read(&slub_lock);
2960 list_for_each_entry(s, &slab_caches, list) {
2961 n = get_node(s, offline_node);
2962 if (n) {
2963 /*
2964 * if n->nr_slabs > 0, slabs still exist on the node
2965 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05002966 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07002967 * callback. So, we must fail.
2968 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002969 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002970
2971 s->node[offline_node] = NULL;
2972 kmem_cache_free(kmalloc_caches, n);
2973 }
2974 }
2975 up_read(&slub_lock);
2976}
2977
2978static int slab_mem_going_online_callback(void *arg)
2979{
2980 struct kmem_cache_node *n;
2981 struct kmem_cache *s;
2982 struct memory_notify *marg = arg;
2983 int nid = marg->status_change_nid;
2984 int ret = 0;
2985
2986 /*
2987 * If the node's memory is already available, then kmem_cache_node is
2988 * already created. Nothing to do.
2989 */
2990 if (nid < 0)
2991 return 0;
2992
2993 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002994 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002995 * allocate a kmem_cache_node structure in order to bring the node
2996 * online.
2997 */
2998 down_read(&slub_lock);
2999 list_for_each_entry(s, &slab_caches, list) {
3000 /*
3001 * XXX: kmem_cache_alloc_node will fallback to other nodes
3002 * since memory is not yet available from the node that
3003 * is brought up.
3004 */
3005 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
3006 if (!n) {
3007 ret = -ENOMEM;
3008 goto out;
3009 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003010 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003011 s->node[nid] = n;
3012 }
3013out:
3014 up_read(&slub_lock);
3015 return ret;
3016}
3017
3018static int slab_memory_callback(struct notifier_block *self,
3019 unsigned long action, void *arg)
3020{
3021 int ret = 0;
3022
3023 switch (action) {
3024 case MEM_GOING_ONLINE:
3025 ret = slab_mem_going_online_callback(arg);
3026 break;
3027 case MEM_GOING_OFFLINE:
3028 ret = slab_mem_going_offline_callback(arg);
3029 break;
3030 case MEM_OFFLINE:
3031 case MEM_CANCEL_ONLINE:
3032 slab_mem_offline_callback(arg);
3033 break;
3034 case MEM_ONLINE:
3035 case MEM_CANCEL_OFFLINE:
3036 break;
3037 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003038 if (ret)
3039 ret = notifier_from_errno(ret);
3040 else
3041 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003042 return ret;
3043}
3044
3045#endif /* CONFIG_MEMORY_HOTPLUG */
3046
Christoph Lameter81819f02007-05-06 14:49:36 -07003047/********************************************************************
3048 * Basic setup of slabs
3049 *******************************************************************/
3050
3051void __init kmem_cache_init(void)
3052{
3053 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003054 int caches = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003055
3056#ifdef CONFIG_NUMA
3057 /*
3058 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003059 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003060 * kmem_cache_open for slab_state == DOWN.
3061 */
3062 create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
Pekka Enberg83b519e2009-06-10 19:40:04 +03003063 sizeof(struct kmem_cache_node), GFP_NOWAIT);
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003064 kmalloc_caches[0].refcount = -1;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003065 caches++;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003066
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003067 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003068#endif
3069
3070 /* Able to allocate the per node structures */
3071 slab_state = PARTIAL;
3072
3073 /* Caches that are not of the two-to-the-power-of size */
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003074 if (KMALLOC_MIN_SIZE <= 32) {
Christoph Lameter4b356be2007-06-16 10:16:13 -07003075 create_kmalloc_cache(&kmalloc_caches[1],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003076 "kmalloc-96", 96, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003077 caches++;
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003078 }
3079 if (KMALLOC_MIN_SIZE <= 64) {
Christoph Lameter4b356be2007-06-16 10:16:13 -07003080 create_kmalloc_cache(&kmalloc_caches[2],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003081 "kmalloc-192", 192, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003082 caches++;
3083 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003084
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003085 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003086 create_kmalloc_cache(&kmalloc_caches[i],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003087 "kmalloc", 1 << i, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003088 caches++;
3089 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003090
Christoph Lameterf1b26332007-07-17 04:03:26 -07003091
3092 /*
3093 * Patch up the size_index table if we have strange large alignment
3094 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003095 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003096 *
3097 * Largest permitted alignment is 256 bytes due to the way we
3098 * handle the index determination for the smaller caches.
3099 *
3100 * Make sure that nothing crazy happens if someone starts tinkering
3101 * around with ARCH_KMALLOC_MINALIGN
3102 */
3103 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3104 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3105
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003106 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3107 int elem = size_index_elem(i);
3108 if (elem >= ARRAY_SIZE(size_index))
3109 break;
3110 size_index[elem] = KMALLOC_SHIFT_LOW;
3111 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003112
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003113 if (KMALLOC_MIN_SIZE == 64) {
3114 /*
3115 * The 96 byte size cache is not used if the alignment
3116 * is 64 byte.
3117 */
3118 for (i = 64 + 8; i <= 96; i += 8)
3119 size_index[size_index_elem(i)] = 7;
3120 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003121 /*
3122 * The 192 byte sized cache is not used if the alignment
3123 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3124 * instead.
3125 */
3126 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003127 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003128 }
3129
Christoph Lameter81819f02007-05-06 14:49:36 -07003130 slab_state = UP;
3131
3132 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003133 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++)
Christoph Lameter81819f02007-05-06 14:49:36 -07003134 kmalloc_caches[i]. name =
Pekka Enberg83b519e2009-06-10 19:40:04 +03003135 kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
Christoph Lameter81819f02007-05-06 14:49:36 -07003136
3137#ifdef CONFIG_SMP
3138 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003139#endif
3140#ifdef CONFIG_NUMA
3141 kmem_size = offsetof(struct kmem_cache, node) +
3142 nr_node_ids * sizeof(struct kmem_cache_node *);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003143#else
3144 kmem_size = sizeof(struct kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003145#endif
3146
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003147 printk(KERN_INFO
3148 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003149 " CPUs=%d, Nodes=%d\n",
3150 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003151 slub_min_order, slub_max_order, slub_min_objects,
3152 nr_cpu_ids, nr_node_ids);
3153}
3154
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003155void __init kmem_cache_init_late(void)
3156{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003157}
3158
Christoph Lameter81819f02007-05-06 14:49:36 -07003159/*
3160 * Find a mergeable slab cache
3161 */
3162static int slab_unmergeable(struct kmem_cache *s)
3163{
3164 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3165 return 1;
3166
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003167 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003168 return 1;
3169
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003170 /*
3171 * We may have set a slab to be unmergeable during bootstrap.
3172 */
3173 if (s->refcount < 0)
3174 return 1;
3175
Christoph Lameter81819f02007-05-06 14:49:36 -07003176 return 0;
3177}
3178
3179static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003180 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003181 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003182{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003183 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003184
3185 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3186 return NULL;
3187
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003188 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003189 return NULL;
3190
3191 size = ALIGN(size, sizeof(void *));
3192 align = calculate_alignment(flags, align, size);
3193 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003194 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003195
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003196 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003197 if (slab_unmergeable(s))
3198 continue;
3199
3200 if (size > s->size)
3201 continue;
3202
Christoph Lameterba0268a2007-09-11 15:24:11 -07003203 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003204 continue;
3205 /*
3206 * Check if alignment is compatible.
3207 * Courtesy of Adrian Drzewiecki
3208 */
Pekka Enberg06428782008-01-07 23:20:27 -08003209 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003210 continue;
3211
3212 if (s->size - size >= sizeof(void *))
3213 continue;
3214
3215 return s;
3216 }
3217 return NULL;
3218}
3219
3220struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003221 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003222{
3223 struct kmem_cache *s;
3224
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003225 if (WARN_ON(!name))
3226 return NULL;
3227
Christoph Lameter81819f02007-05-06 14:49:36 -07003228 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003229 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003230 if (s) {
3231 s->refcount++;
3232 /*
3233 * Adjust the object sizes so that we clear
3234 * the complete object on kzalloc.
3235 */
3236 s->objsize = max(s->objsize, (int)size);
3237 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003238 up_write(&slub_lock);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003239
David Rientjes7b8f3b62008-12-17 22:09:46 -08003240 if (sysfs_slab_alias(s, name)) {
3241 down_write(&slub_lock);
3242 s->refcount--;
3243 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003244 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003245 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003246 return s;
3247 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003248
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003249 s = kmalloc(kmem_size, GFP_KERNEL);
3250 if (s) {
3251 if (kmem_cache_open(s, GFP_KERNEL, name,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003252 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003253 list_add(&s->list, &slab_caches);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003254 up_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003255 if (sysfs_slab_add(s)) {
3256 down_write(&slub_lock);
3257 list_del(&s->list);
3258 up_write(&slub_lock);
3259 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003260 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003261 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003262 return s;
3263 }
3264 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003265 }
3266 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003267
3268err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003269 if (flags & SLAB_PANIC)
3270 panic("Cannot create slabcache %s\n", name);
3271 else
3272 s = NULL;
3273 return s;
3274}
3275EXPORT_SYMBOL(kmem_cache_create);
3276
Christoph Lameter81819f02007-05-06 14:49:36 -07003277#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003278/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003279 * Use the cpu notifier to insure that the cpu slabs are flushed when
3280 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003281 */
3282static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3283 unsigned long action, void *hcpu)
3284{
3285 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003286 struct kmem_cache *s;
3287 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003288
3289 switch (action) {
3290 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003291 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003292 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003293 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003294 down_read(&slub_lock);
3295 list_for_each_entry(s, &slab_caches, list) {
3296 local_irq_save(flags);
3297 __flush_cpu_slab(s, cpu);
3298 local_irq_restore(flags);
3299 }
3300 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003301 break;
3302 default:
3303 break;
3304 }
3305 return NOTIFY_OK;
3306}
3307
Pekka Enberg06428782008-01-07 23:20:27 -08003308static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003309 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003310};
Christoph Lameter81819f02007-05-06 14:49:36 -07003311
3312#endif
3313
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003314void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003315{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003316 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003317 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003318
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003319 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003320 return kmalloc_large(size, gfpflags);
3321
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003322 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003323
Satyam Sharma2408c552007-10-16 01:24:44 -07003324 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003325 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003326
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003327 ret = slab_alloc(s, gfpflags, -1, caller);
3328
3329 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003330 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003331
3332 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003333}
3334
3335void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003336 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003337{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003338 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003339 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003340
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003341 if (unlikely(size > SLUB_MAX_SIZE))
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003342 return kmalloc_large_node(size, gfpflags, node);
Pekka Enbergeada35e2008-02-11 22:47:46 +02003343
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003344 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003345
Satyam Sharma2408c552007-10-16 01:24:44 -07003346 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003347 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003348
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003349 ret = slab_alloc(s, gfpflags, node, caller);
3350
3351 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003352 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003353
3354 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003355}
3356
Christoph Lameterf6acb632008-04-29 16:16:06 -07003357#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003358static int count_inuse(struct page *page)
3359{
3360 return page->inuse;
3361}
3362
3363static int count_total(struct page *page)
3364{
3365 return page->objects;
3366}
3367
Christoph Lameter434e2452007-07-17 04:03:30 -07003368static int validate_slab(struct kmem_cache *s, struct page *page,
3369 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003370{
3371 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003372 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003373
3374 if (!check_slab(s, page) ||
3375 !on_freelist(s, page, NULL))
3376 return 0;
3377
3378 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003379 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003380
Christoph Lameter7656c722007-05-09 02:32:40 -07003381 for_each_free_object(p, s, page->freelist) {
3382 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003383 if (!check_object(s, page, p, 0))
3384 return 0;
3385 }
3386
Christoph Lameter224a88b2008-04-14 19:11:31 +03003387 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003388 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003389 if (!check_object(s, page, p, 1))
3390 return 0;
3391 return 1;
3392}
3393
Christoph Lameter434e2452007-07-17 04:03:30 -07003394static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3395 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003396{
3397 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003398 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003399 slab_unlock(page);
3400 } else
3401 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3402 s->name, page);
3403
3404 if (s->flags & DEBUG_DEFAULT_FLAGS) {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003405 if (!PageSlubDebug(page))
3406 printk(KERN_ERR "SLUB %s: SlubDebug not set "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003407 "on slab 0x%p\n", s->name, page);
3408 } else {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003409 if (PageSlubDebug(page))
3410 printk(KERN_ERR "SLUB %s: SlubDebug set on "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003411 "slab 0x%p\n", s->name, page);
3412 }
3413}
3414
Christoph Lameter434e2452007-07-17 04:03:30 -07003415static int validate_slab_node(struct kmem_cache *s,
3416 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003417{
3418 unsigned long count = 0;
3419 struct page *page;
3420 unsigned long flags;
3421
3422 spin_lock_irqsave(&n->list_lock, flags);
3423
3424 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003425 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003426 count++;
3427 }
3428 if (count != n->nr_partial)
3429 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3430 "counter=%ld\n", s->name, count, n->nr_partial);
3431
3432 if (!(s->flags & SLAB_STORE_USER))
3433 goto out;
3434
3435 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003436 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003437 count++;
3438 }
3439 if (count != atomic_long_read(&n->nr_slabs))
3440 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3441 "counter=%ld\n", s->name, count,
3442 atomic_long_read(&n->nr_slabs));
3443
3444out:
3445 spin_unlock_irqrestore(&n->list_lock, flags);
3446 return count;
3447}
3448
Christoph Lameter434e2452007-07-17 04:03:30 -07003449static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003450{
3451 int node;
3452 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003453 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003454 sizeof(unsigned long), GFP_KERNEL);
3455
3456 if (!map)
3457 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003458
3459 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003460 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003461 struct kmem_cache_node *n = get_node(s, node);
3462
Christoph Lameter434e2452007-07-17 04:03:30 -07003463 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003464 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003465 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003466 return count;
3467}
3468
Christoph Lameterb3459702007-05-09 02:32:41 -07003469#ifdef SLUB_RESILIENCY_TEST
3470static void resiliency_test(void)
3471{
3472 u8 *p;
3473
3474 printk(KERN_ERR "SLUB resiliency testing\n");
3475 printk(KERN_ERR "-----------------------\n");
3476 printk(KERN_ERR "A. Corruption after allocation\n");
3477
3478 p = kzalloc(16, GFP_KERNEL);
3479 p[16] = 0x12;
3480 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3481 " 0x12->0x%p\n\n", p + 16);
3482
3483 validate_slab_cache(kmalloc_caches + 4);
3484
3485 /* Hmmm... The next two are dangerous */
3486 p = kzalloc(32, GFP_KERNEL);
3487 p[32 + sizeof(void *)] = 0x34;
3488 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003489 " 0x34 -> -0x%p\n", p);
3490 printk(KERN_ERR
3491 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003492
3493 validate_slab_cache(kmalloc_caches + 5);
3494 p = kzalloc(64, GFP_KERNEL);
3495 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3496 *p = 0x56;
3497 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3498 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003499 printk(KERN_ERR
3500 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003501 validate_slab_cache(kmalloc_caches + 6);
3502
3503 printk(KERN_ERR "\nB. Corruption after free\n");
3504 p = kzalloc(128, GFP_KERNEL);
3505 kfree(p);
3506 *p = 0x78;
3507 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3508 validate_slab_cache(kmalloc_caches + 7);
3509
3510 p = kzalloc(256, GFP_KERNEL);
3511 kfree(p);
3512 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003513 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3514 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003515 validate_slab_cache(kmalloc_caches + 8);
3516
3517 p = kzalloc(512, GFP_KERNEL);
3518 kfree(p);
3519 p[512] = 0xab;
3520 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3521 validate_slab_cache(kmalloc_caches + 9);
3522}
3523#else
3524static void resiliency_test(void) {};
3525#endif
3526
Christoph Lameter88a420e2007-05-06 14:49:45 -07003527/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003528 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003529 * and freed.
3530 */
3531
3532struct location {
3533 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003534 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003535 long long sum_time;
3536 long min_time;
3537 long max_time;
3538 long min_pid;
3539 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303540 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003541 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003542};
3543
3544struct loc_track {
3545 unsigned long max;
3546 unsigned long count;
3547 struct location *loc;
3548};
3549
3550static void free_loc_track(struct loc_track *t)
3551{
3552 if (t->max)
3553 free_pages((unsigned long)t->loc,
3554 get_order(sizeof(struct location) * t->max));
3555}
3556
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003557static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003558{
3559 struct location *l;
3560 int order;
3561
Christoph Lameter88a420e2007-05-06 14:49:45 -07003562 order = get_order(sizeof(struct location) * max);
3563
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003564 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003565 if (!l)
3566 return 0;
3567
3568 if (t->count) {
3569 memcpy(l, t->loc, sizeof(struct location) * t->count);
3570 free_loc_track(t);
3571 }
3572 t->max = max;
3573 t->loc = l;
3574 return 1;
3575}
3576
3577static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003578 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003579{
3580 long start, end, pos;
3581 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003582 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003583 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003584
3585 start = -1;
3586 end = t->count;
3587
3588 for ( ; ; ) {
3589 pos = start + (end - start + 1) / 2;
3590
3591 /*
3592 * There is nothing at "end". If we end up there
3593 * we need to add something to before end.
3594 */
3595 if (pos == end)
3596 break;
3597
3598 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003599 if (track->addr == caddr) {
3600
3601 l = &t->loc[pos];
3602 l->count++;
3603 if (track->when) {
3604 l->sum_time += age;
3605 if (age < l->min_time)
3606 l->min_time = age;
3607 if (age > l->max_time)
3608 l->max_time = age;
3609
3610 if (track->pid < l->min_pid)
3611 l->min_pid = track->pid;
3612 if (track->pid > l->max_pid)
3613 l->max_pid = track->pid;
3614
Rusty Russell174596a2009-01-01 10:12:29 +10303615 cpumask_set_cpu(track->cpu,
3616 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003617 }
3618 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003619 return 1;
3620 }
3621
Christoph Lameter45edfa52007-05-09 02:32:45 -07003622 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003623 end = pos;
3624 else
3625 start = pos;
3626 }
3627
3628 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003629 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003630 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003631 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003632 return 0;
3633
3634 l = t->loc + pos;
3635 if (pos < t->count)
3636 memmove(l + 1, l,
3637 (t->count - pos) * sizeof(struct location));
3638 t->count++;
3639 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003640 l->addr = track->addr;
3641 l->sum_time = age;
3642 l->min_time = age;
3643 l->max_time = age;
3644 l->min_pid = track->pid;
3645 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303646 cpumask_clear(to_cpumask(l->cpus));
3647 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003648 nodes_clear(l->nodes);
3649 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003650 return 1;
3651}
3652
3653static void process_slab(struct loc_track *t, struct kmem_cache *s,
3654 struct page *page, enum track_item alloc)
3655{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003656 void *addr = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +03003657 DECLARE_BITMAP(map, page->objects);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003658 void *p;
3659
Christoph Lameter39b26462008-04-14 19:11:30 +03003660 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003661 for_each_free_object(p, s, page->freelist)
3662 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003663
Christoph Lameter224a88b2008-04-14 19:11:31 +03003664 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003665 if (!test_bit(slab_index(p, s, addr), map))
3666 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003667}
3668
3669static int list_locations(struct kmem_cache *s, char *buf,
3670 enum track_item alloc)
3671{
Harvey Harrisone374d482008-01-31 15:20:50 -08003672 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003673 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003674 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003675 int node;
3676
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003677 if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
Andrew Mortonea3061d2007-10-16 01:26:09 -07003678 GFP_TEMPORARY))
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003679 return sprintf(buf, "Out of memory\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003680
3681 /* Push back cpu slabs */
3682 flush_all(s);
3683
Christoph Lameterf64dc582007-10-16 01:25:33 -07003684 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003685 struct kmem_cache_node *n = get_node(s, node);
3686 unsigned long flags;
3687 struct page *page;
3688
Christoph Lameter9e869432007-08-22 14:01:56 -07003689 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003690 continue;
3691
3692 spin_lock_irqsave(&n->list_lock, flags);
3693 list_for_each_entry(page, &n->partial, lru)
3694 process_slab(&t, s, page, alloc);
3695 list_for_each_entry(page, &n->full, lru)
3696 process_slab(&t, s, page, alloc);
3697 spin_unlock_irqrestore(&n->list_lock, flags);
3698 }
3699
3700 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003701 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003702
Hugh Dickins9c246242008-12-09 13:14:27 -08003703 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003704 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003705 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003706
3707 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003708 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003709 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003710 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003711
3712 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003713 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003714 l->min_time,
3715 (long)div_u64(l->sum_time, l->count),
3716 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003717 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003718 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003719 l->min_time);
3720
3721 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003722 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003723 l->min_pid, l->max_pid);
3724 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003725 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003726 l->min_pid);
3727
Rusty Russell174596a2009-01-01 10:12:29 +10303728 if (num_online_cpus() > 1 &&
3729 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003730 len < PAGE_SIZE - 60) {
3731 len += sprintf(buf + len, " cpus=");
3732 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303733 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003734 }
3735
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003736 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003737 len < PAGE_SIZE - 60) {
3738 len += sprintf(buf + len, " nodes=");
3739 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003740 l->nodes);
3741 }
3742
Harvey Harrisone374d482008-01-31 15:20:50 -08003743 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003744 }
3745
3746 free_loc_track(&t);
3747 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003748 len += sprintf(buf, "No data\n");
3749 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003750}
3751
Christoph Lameter81819f02007-05-06 14:49:36 -07003752enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003753 SL_ALL, /* All slabs */
3754 SL_PARTIAL, /* Only partially allocated slabs */
3755 SL_CPU, /* Only slabs used for cpu caches */
3756 SL_OBJECTS, /* Determine allocated objects not slabs */
3757 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003758};
3759
Christoph Lameter205ab992008-04-14 19:11:40 +03003760#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003761#define SO_PARTIAL (1 << SL_PARTIAL)
3762#define SO_CPU (1 << SL_CPU)
3763#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003764#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003765
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003766static ssize_t show_slab_objects(struct kmem_cache *s,
3767 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003768{
3769 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003770 int node;
3771 int x;
3772 unsigned long *nodes;
3773 unsigned long *per_cpu;
3774
3775 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003776 if (!nodes)
3777 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003778 per_cpu = nodes + nr_node_ids;
3779
Christoph Lameter205ab992008-04-14 19:11:40 +03003780 if (flags & SO_CPU) {
3781 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003782
Christoph Lameter205ab992008-04-14 19:11:40 +03003783 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003784 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003785
Christoph Lameter205ab992008-04-14 19:11:40 +03003786 if (!c || c->node < 0)
3787 continue;
3788
3789 if (c->page) {
3790 if (flags & SO_TOTAL)
3791 x = c->page->objects;
3792 else if (flags & SO_OBJECTS)
3793 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003794 else
3795 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003796
Christoph Lameter81819f02007-05-06 14:49:36 -07003797 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003798 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003799 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003800 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003801 }
3802 }
3803
Christoph Lameter205ab992008-04-14 19:11:40 +03003804 if (flags & SO_ALL) {
3805 for_each_node_state(node, N_NORMAL_MEMORY) {
3806 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003807
Christoph Lameter205ab992008-04-14 19:11:40 +03003808 if (flags & SO_TOTAL)
3809 x = atomic_long_read(&n->total_objects);
3810 else if (flags & SO_OBJECTS)
3811 x = atomic_long_read(&n->total_objects) -
3812 count_partial(n, count_free);
3813
3814 else
3815 x = atomic_long_read(&n->nr_slabs);
3816 total += x;
3817 nodes[node] += x;
3818 }
3819
3820 } else if (flags & SO_PARTIAL) {
3821 for_each_node_state(node, N_NORMAL_MEMORY) {
3822 struct kmem_cache_node *n = get_node(s, node);
3823
3824 if (flags & SO_TOTAL)
3825 x = count_partial(n, count_total);
3826 else if (flags & SO_OBJECTS)
3827 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003828 else
3829 x = n->nr_partial;
3830 total += x;
3831 nodes[node] += x;
3832 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003833 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003834 x = sprintf(buf, "%lu", total);
3835#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003836 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003837 if (nodes[node])
3838 x += sprintf(buf + x, " N%d=%lu",
3839 node, nodes[node]);
3840#endif
3841 kfree(nodes);
3842 return x + sprintf(buf + x, "\n");
3843}
3844
3845static int any_slab_objects(struct kmem_cache *s)
3846{
3847 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003848
3849 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003850 struct kmem_cache_node *n = get_node(s, node);
3851
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003852 if (!n)
3853 continue;
3854
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003855 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003856 return 1;
3857 }
3858 return 0;
3859}
3860
3861#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3862#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3863
3864struct slab_attribute {
3865 struct attribute attr;
3866 ssize_t (*show)(struct kmem_cache *s, char *buf);
3867 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3868};
3869
3870#define SLAB_ATTR_RO(_name) \
3871 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3872
3873#define SLAB_ATTR(_name) \
3874 static struct slab_attribute _name##_attr = \
3875 __ATTR(_name, 0644, _name##_show, _name##_store)
3876
Christoph Lameter81819f02007-05-06 14:49:36 -07003877static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3878{
3879 return sprintf(buf, "%d\n", s->size);
3880}
3881SLAB_ATTR_RO(slab_size);
3882
3883static ssize_t align_show(struct kmem_cache *s, char *buf)
3884{
3885 return sprintf(buf, "%d\n", s->align);
3886}
3887SLAB_ATTR_RO(align);
3888
3889static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3890{
3891 return sprintf(buf, "%d\n", s->objsize);
3892}
3893SLAB_ATTR_RO(object_size);
3894
3895static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3896{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003897 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003898}
3899SLAB_ATTR_RO(objs_per_slab);
3900
Christoph Lameter06b285d2008-04-14 19:11:41 +03003901static ssize_t order_store(struct kmem_cache *s,
3902 const char *buf, size_t length)
3903{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003904 unsigned long order;
3905 int err;
3906
3907 err = strict_strtoul(buf, 10, &order);
3908 if (err)
3909 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003910
3911 if (order > slub_max_order || order < slub_min_order)
3912 return -EINVAL;
3913
3914 calculate_sizes(s, order);
3915 return length;
3916}
3917
Christoph Lameter81819f02007-05-06 14:49:36 -07003918static ssize_t order_show(struct kmem_cache *s, char *buf)
3919{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003920 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003921}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003922SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003923
David Rientjes73d342b2009-02-22 17:40:09 -08003924static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3925{
3926 return sprintf(buf, "%lu\n", s->min_partial);
3927}
3928
3929static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3930 size_t length)
3931{
3932 unsigned long min;
3933 int err;
3934
3935 err = strict_strtoul(buf, 10, &min);
3936 if (err)
3937 return err;
3938
David Rientjesc0bdb232009-02-25 09:16:35 +02003939 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003940 return length;
3941}
3942SLAB_ATTR(min_partial);
3943
Christoph Lameter81819f02007-05-06 14:49:36 -07003944static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3945{
3946 if (s->ctor) {
3947 int n = sprint_symbol(buf, (unsigned long)s->ctor);
3948
3949 return n + sprintf(buf + n, "\n");
3950 }
3951 return 0;
3952}
3953SLAB_ATTR_RO(ctor);
3954
Christoph Lameter81819f02007-05-06 14:49:36 -07003955static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3956{
3957 return sprintf(buf, "%d\n", s->refcount - 1);
3958}
3959SLAB_ATTR_RO(aliases);
3960
3961static ssize_t slabs_show(struct kmem_cache *s, char *buf)
3962{
Christoph Lameter205ab992008-04-14 19:11:40 +03003963 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003964}
3965SLAB_ATTR_RO(slabs);
3966
3967static ssize_t partial_show(struct kmem_cache *s, char *buf)
3968{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003969 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003970}
3971SLAB_ATTR_RO(partial);
3972
3973static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3974{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003975 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003976}
3977SLAB_ATTR_RO(cpu_slabs);
3978
3979static ssize_t objects_show(struct kmem_cache *s, char *buf)
3980{
Christoph Lameter205ab992008-04-14 19:11:40 +03003981 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07003982}
3983SLAB_ATTR_RO(objects);
3984
Christoph Lameter205ab992008-04-14 19:11:40 +03003985static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
3986{
3987 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
3988}
3989SLAB_ATTR_RO(objects_partial);
3990
3991static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
3992{
3993 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
3994}
3995SLAB_ATTR_RO(total_objects);
3996
Christoph Lameter81819f02007-05-06 14:49:36 -07003997static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
3998{
3999 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4000}
4001
4002static ssize_t sanity_checks_store(struct kmem_cache *s,
4003 const char *buf, size_t length)
4004{
4005 s->flags &= ~SLAB_DEBUG_FREE;
4006 if (buf[0] == '1')
4007 s->flags |= SLAB_DEBUG_FREE;
4008 return length;
4009}
4010SLAB_ATTR(sanity_checks);
4011
4012static ssize_t trace_show(struct kmem_cache *s, char *buf)
4013{
4014 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4015}
4016
4017static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4018 size_t length)
4019{
4020 s->flags &= ~SLAB_TRACE;
4021 if (buf[0] == '1')
4022 s->flags |= SLAB_TRACE;
4023 return length;
4024}
4025SLAB_ATTR(trace);
4026
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004027#ifdef CONFIG_FAILSLAB
4028static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4029{
4030 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4031}
4032
4033static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4034 size_t length)
4035{
4036 s->flags &= ~SLAB_FAILSLAB;
4037 if (buf[0] == '1')
4038 s->flags |= SLAB_FAILSLAB;
4039 return length;
4040}
4041SLAB_ATTR(failslab);
4042#endif
4043
Christoph Lameter81819f02007-05-06 14:49:36 -07004044static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4045{
4046 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4047}
4048
4049static ssize_t reclaim_account_store(struct kmem_cache *s,
4050 const char *buf, size_t length)
4051{
4052 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4053 if (buf[0] == '1')
4054 s->flags |= SLAB_RECLAIM_ACCOUNT;
4055 return length;
4056}
4057SLAB_ATTR(reclaim_account);
4058
4059static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4060{
Christoph Lameter5af60832007-05-06 14:49:56 -07004061 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004062}
4063SLAB_ATTR_RO(hwcache_align);
4064
4065#ifdef CONFIG_ZONE_DMA
4066static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4067{
4068 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4069}
4070SLAB_ATTR_RO(cache_dma);
4071#endif
4072
4073static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4074{
4075 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4076}
4077SLAB_ATTR_RO(destroy_by_rcu);
4078
4079static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4080{
4081 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4082}
4083
4084static ssize_t red_zone_store(struct kmem_cache *s,
4085 const char *buf, size_t length)
4086{
4087 if (any_slab_objects(s))
4088 return -EBUSY;
4089
4090 s->flags &= ~SLAB_RED_ZONE;
4091 if (buf[0] == '1')
4092 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004093 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004094 return length;
4095}
4096SLAB_ATTR(red_zone);
4097
4098static ssize_t poison_show(struct kmem_cache *s, char *buf)
4099{
4100 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4101}
4102
4103static ssize_t poison_store(struct kmem_cache *s,
4104 const char *buf, size_t length)
4105{
4106 if (any_slab_objects(s))
4107 return -EBUSY;
4108
4109 s->flags &= ~SLAB_POISON;
4110 if (buf[0] == '1')
4111 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004112 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004113 return length;
4114}
4115SLAB_ATTR(poison);
4116
4117static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4118{
4119 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4120}
4121
4122static ssize_t store_user_store(struct kmem_cache *s,
4123 const char *buf, size_t length)
4124{
4125 if (any_slab_objects(s))
4126 return -EBUSY;
4127
4128 s->flags &= ~SLAB_STORE_USER;
4129 if (buf[0] == '1')
4130 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004131 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004132 return length;
4133}
4134SLAB_ATTR(store_user);
4135
Christoph Lameter53e15af2007-05-06 14:49:43 -07004136static ssize_t validate_show(struct kmem_cache *s, char *buf)
4137{
4138 return 0;
4139}
4140
4141static ssize_t validate_store(struct kmem_cache *s,
4142 const char *buf, size_t length)
4143{
Christoph Lameter434e2452007-07-17 04:03:30 -07004144 int ret = -EINVAL;
4145
4146 if (buf[0] == '1') {
4147 ret = validate_slab_cache(s);
4148 if (ret >= 0)
4149 ret = length;
4150 }
4151 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004152}
4153SLAB_ATTR(validate);
4154
Christoph Lameter2086d262007-05-06 14:49:46 -07004155static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4156{
4157 return 0;
4158}
4159
4160static ssize_t shrink_store(struct kmem_cache *s,
4161 const char *buf, size_t length)
4162{
4163 if (buf[0] == '1') {
4164 int rc = kmem_cache_shrink(s);
4165
4166 if (rc)
4167 return rc;
4168 } else
4169 return -EINVAL;
4170 return length;
4171}
4172SLAB_ATTR(shrink);
4173
Christoph Lameter88a420e2007-05-06 14:49:45 -07004174static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4175{
4176 if (!(s->flags & SLAB_STORE_USER))
4177 return -ENOSYS;
4178 return list_locations(s, buf, TRACK_ALLOC);
4179}
4180SLAB_ATTR_RO(alloc_calls);
4181
4182static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4183{
4184 if (!(s->flags & SLAB_STORE_USER))
4185 return -ENOSYS;
4186 return list_locations(s, buf, TRACK_FREE);
4187}
4188SLAB_ATTR_RO(free_calls);
4189
Christoph Lameter81819f02007-05-06 14:49:36 -07004190#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004191static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004192{
Christoph Lameter98246012008-01-07 23:20:26 -08004193 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004194}
4195
Christoph Lameter98246012008-01-07 23:20:26 -08004196static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004197 const char *buf, size_t length)
4198{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004199 unsigned long ratio;
4200 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004201
Christoph Lameter0121c6192008-04-29 16:11:12 -07004202 err = strict_strtoul(buf, 10, &ratio);
4203 if (err)
4204 return err;
4205
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004206 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004207 s->remote_node_defrag_ratio = ratio * 10;
4208
Christoph Lameter81819f02007-05-06 14:49:36 -07004209 return length;
4210}
Christoph Lameter98246012008-01-07 23:20:26 -08004211SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004212#endif
4213
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004214#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004215static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4216{
4217 unsigned long sum = 0;
4218 int cpu;
4219 int len;
4220 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4221
4222 if (!data)
4223 return -ENOMEM;
4224
4225 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004226 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004227
4228 data[cpu] = x;
4229 sum += x;
4230 }
4231
4232 len = sprintf(buf, "%lu", sum);
4233
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004234#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004235 for_each_online_cpu(cpu) {
4236 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004237 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004238 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004239#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004240 kfree(data);
4241 return len + sprintf(buf + len, "\n");
4242}
4243
David Rientjes78eb00c2009-10-15 02:20:22 -07004244static void clear_stat(struct kmem_cache *s, enum stat_item si)
4245{
4246 int cpu;
4247
4248 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004249 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004250}
4251
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004252#define STAT_ATTR(si, text) \
4253static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4254{ \
4255 return show_stat(s, buf, si); \
4256} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004257static ssize_t text##_store(struct kmem_cache *s, \
4258 const char *buf, size_t length) \
4259{ \
4260 if (buf[0] != '0') \
4261 return -EINVAL; \
4262 clear_stat(s, si); \
4263 return length; \
4264} \
4265SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004266
4267STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4268STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4269STAT_ATTR(FREE_FASTPATH, free_fastpath);
4270STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4271STAT_ATTR(FREE_FROZEN, free_frozen);
4272STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4273STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4274STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4275STAT_ATTR(ALLOC_SLAB, alloc_slab);
4276STAT_ATTR(ALLOC_REFILL, alloc_refill);
4277STAT_ATTR(FREE_SLAB, free_slab);
4278STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4279STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4280STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4281STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4282STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4283STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004284STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004285#endif
4286
Pekka Enberg06428782008-01-07 23:20:27 -08004287static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004288 &slab_size_attr.attr,
4289 &object_size_attr.attr,
4290 &objs_per_slab_attr.attr,
4291 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004292 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004293 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004294 &objects_partial_attr.attr,
4295 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004296 &slabs_attr.attr,
4297 &partial_attr.attr,
4298 &cpu_slabs_attr.attr,
4299 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004300 &aliases_attr.attr,
4301 &align_attr.attr,
4302 &sanity_checks_attr.attr,
4303 &trace_attr.attr,
4304 &hwcache_align_attr.attr,
4305 &reclaim_account_attr.attr,
4306 &destroy_by_rcu_attr.attr,
4307 &red_zone_attr.attr,
4308 &poison_attr.attr,
4309 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004310 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004311 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004312 &alloc_calls_attr.attr,
4313 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004314#ifdef CONFIG_ZONE_DMA
4315 &cache_dma_attr.attr,
4316#endif
4317#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004318 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004319#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004320#ifdef CONFIG_SLUB_STATS
4321 &alloc_fastpath_attr.attr,
4322 &alloc_slowpath_attr.attr,
4323 &free_fastpath_attr.attr,
4324 &free_slowpath_attr.attr,
4325 &free_frozen_attr.attr,
4326 &free_add_partial_attr.attr,
4327 &free_remove_partial_attr.attr,
4328 &alloc_from_partial_attr.attr,
4329 &alloc_slab_attr.attr,
4330 &alloc_refill_attr.attr,
4331 &free_slab_attr.attr,
4332 &cpuslab_flush_attr.attr,
4333 &deactivate_full_attr.attr,
4334 &deactivate_empty_attr.attr,
4335 &deactivate_to_head_attr.attr,
4336 &deactivate_to_tail_attr.attr,
4337 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004338 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004339#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004340#ifdef CONFIG_FAILSLAB
4341 &failslab_attr.attr,
4342#endif
4343
Christoph Lameter81819f02007-05-06 14:49:36 -07004344 NULL
4345};
4346
4347static struct attribute_group slab_attr_group = {
4348 .attrs = slab_attrs,
4349};
4350
4351static ssize_t slab_attr_show(struct kobject *kobj,
4352 struct attribute *attr,
4353 char *buf)
4354{
4355 struct slab_attribute *attribute;
4356 struct kmem_cache *s;
4357 int err;
4358
4359 attribute = to_slab_attr(attr);
4360 s = to_slab(kobj);
4361
4362 if (!attribute->show)
4363 return -EIO;
4364
4365 err = attribute->show(s, buf);
4366
4367 return err;
4368}
4369
4370static ssize_t slab_attr_store(struct kobject *kobj,
4371 struct attribute *attr,
4372 const char *buf, size_t len)
4373{
4374 struct slab_attribute *attribute;
4375 struct kmem_cache *s;
4376 int err;
4377
4378 attribute = to_slab_attr(attr);
4379 s = to_slab(kobj);
4380
4381 if (!attribute->store)
4382 return -EIO;
4383
4384 err = attribute->store(s, buf, len);
4385
4386 return err;
4387}
4388
Christoph Lameter151c6022008-01-07 22:29:05 -08004389static void kmem_cache_release(struct kobject *kobj)
4390{
4391 struct kmem_cache *s = to_slab(kobj);
4392
4393 kfree(s);
4394}
4395
Emese Revfy52cf25d2010-01-19 02:58:23 +01004396static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004397 .show = slab_attr_show,
4398 .store = slab_attr_store,
4399};
4400
4401static struct kobj_type slab_ktype = {
4402 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004403 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004404};
4405
4406static int uevent_filter(struct kset *kset, struct kobject *kobj)
4407{
4408 struct kobj_type *ktype = get_ktype(kobj);
4409
4410 if (ktype == &slab_ktype)
4411 return 1;
4412 return 0;
4413}
4414
Emese Revfy9cd43612009-12-31 14:52:51 +01004415static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004416 .filter = uevent_filter,
4417};
4418
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004419static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004420
4421#define ID_STR_LENGTH 64
4422
4423/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004424 *
4425 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004426 */
4427static char *create_unique_id(struct kmem_cache *s)
4428{
4429 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4430 char *p = name;
4431
4432 BUG_ON(!name);
4433
4434 *p++ = ':';
4435 /*
4436 * First flags affecting slabcache operations. We will only
4437 * get here for aliasable slabs so we do not need to support
4438 * too many flags. The flags here must cover all flags that
4439 * are matched during merging to guarantee that the id is
4440 * unique.
4441 */
4442 if (s->flags & SLAB_CACHE_DMA)
4443 *p++ = 'd';
4444 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4445 *p++ = 'a';
4446 if (s->flags & SLAB_DEBUG_FREE)
4447 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004448 if (!(s->flags & SLAB_NOTRACK))
4449 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004450 if (p != name + 1)
4451 *p++ = '-';
4452 p += sprintf(p, "%07d", s->size);
4453 BUG_ON(p > name + ID_STR_LENGTH - 1);
4454 return name;
4455}
4456
4457static int sysfs_slab_add(struct kmem_cache *s)
4458{
4459 int err;
4460 const char *name;
4461 int unmergeable;
4462
4463 if (slab_state < SYSFS)
4464 /* Defer until later */
4465 return 0;
4466
4467 unmergeable = slab_unmergeable(s);
4468 if (unmergeable) {
4469 /*
4470 * Slabcache can never be merged so we can use the name proper.
4471 * This is typically the case for debug situations. In that
4472 * case we can catch duplicate names easily.
4473 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004474 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004475 name = s->name;
4476 } else {
4477 /*
4478 * Create a unique name for the slab as a target
4479 * for the symlinks.
4480 */
4481 name = create_unique_id(s);
4482 }
4483
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004484 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004485 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4486 if (err) {
4487 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004488 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004489 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004490
4491 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004492 if (err) {
4493 kobject_del(&s->kobj);
4494 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004495 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004496 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004497 kobject_uevent(&s->kobj, KOBJ_ADD);
4498 if (!unmergeable) {
4499 /* Setup first alias */
4500 sysfs_slab_alias(s, s->name);
4501 kfree(name);
4502 }
4503 return 0;
4504}
4505
4506static void sysfs_slab_remove(struct kmem_cache *s)
4507{
4508 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4509 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004510 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004511}
4512
4513/*
4514 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004515 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004516 */
4517struct saved_alias {
4518 struct kmem_cache *s;
4519 const char *name;
4520 struct saved_alias *next;
4521};
4522
Adrian Bunk5af328a2007-07-17 04:03:27 -07004523static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004524
4525static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4526{
4527 struct saved_alias *al;
4528
4529 if (slab_state == SYSFS) {
4530 /*
4531 * If we have a leftover link then remove it.
4532 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004533 sysfs_remove_link(&slab_kset->kobj, name);
4534 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004535 }
4536
4537 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4538 if (!al)
4539 return -ENOMEM;
4540
4541 al->s = s;
4542 al->name = name;
4543 al->next = alias_list;
4544 alias_list = al;
4545 return 0;
4546}
4547
4548static int __init slab_sysfs_init(void)
4549{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004550 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004551 int err;
4552
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004553 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004554 if (!slab_kset) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004555 printk(KERN_ERR "Cannot register slab subsystem.\n");
4556 return -ENOSYS;
4557 }
4558
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004559 slab_state = SYSFS;
4560
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004561 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004562 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004563 if (err)
4564 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4565 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004566 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004567
4568 while (alias_list) {
4569 struct saved_alias *al = alias_list;
4570
4571 alias_list = alias_list->next;
4572 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004573 if (err)
4574 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4575 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004576 kfree(al);
4577 }
4578
4579 resiliency_test();
4580 return 0;
4581}
4582
4583__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004584#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004585
4586/*
4587 * The /proc/slabinfo ABI
4588 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004589#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004590static void print_slabinfo_header(struct seq_file *m)
4591{
4592 seq_puts(m, "slabinfo - version: 2.1\n");
4593 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4594 "<objperslab> <pagesperslab>");
4595 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4596 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4597 seq_putc(m, '\n');
4598}
4599
4600static void *s_start(struct seq_file *m, loff_t *pos)
4601{
4602 loff_t n = *pos;
4603
4604 down_read(&slub_lock);
4605 if (!n)
4606 print_slabinfo_header(m);
4607
4608 return seq_list_start(&slab_caches, *pos);
4609}
4610
4611static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4612{
4613 return seq_list_next(p, &slab_caches, pos);
4614}
4615
4616static void s_stop(struct seq_file *m, void *p)
4617{
4618 up_read(&slub_lock);
4619}
4620
4621static int s_show(struct seq_file *m, void *p)
4622{
4623 unsigned long nr_partials = 0;
4624 unsigned long nr_slabs = 0;
4625 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004626 unsigned long nr_objs = 0;
4627 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004628 struct kmem_cache *s;
4629 int node;
4630
4631 s = list_entry(p, struct kmem_cache, list);
4632
4633 for_each_online_node(node) {
4634 struct kmem_cache_node *n = get_node(s, node);
4635
4636 if (!n)
4637 continue;
4638
4639 nr_partials += n->nr_partial;
4640 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004641 nr_objs += atomic_long_read(&n->total_objects);
4642 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004643 }
4644
Christoph Lameter205ab992008-04-14 19:11:40 +03004645 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004646
4647 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004648 nr_objs, s->size, oo_objects(s->oo),
4649 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004650 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4651 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4652 0UL);
4653 seq_putc(m, '\n');
4654 return 0;
4655}
4656
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004657static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004658 .start = s_start,
4659 .next = s_next,
4660 .stop = s_stop,
4661 .show = s_show,
4662};
4663
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004664static int slabinfo_open(struct inode *inode, struct file *file)
4665{
4666 return seq_open(file, &slabinfo_op);
4667}
4668
4669static const struct file_operations proc_slabinfo_operations = {
4670 .open = slabinfo_open,
4671 .read = seq_read,
4672 .llseek = seq_lseek,
4673 .release = seq_release,
4674};
4675
4676static int __init slab_proc_init(void)
4677{
WANG Congcf5d1132009-08-18 19:11:40 +03004678 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004679 return 0;
4680}
4681module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004682#endif /* CONFIG_SLABINFO */