blob: b6276753626e9ade2e8bec661e980fcfc3c95374 [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/*
144 * Set of flags that will prevent slab merging
145 */
146#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Catalin Marinas06f22f12009-06-11 13:23:18 +0100147 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700148
149#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200150 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700151
152#ifndef ARCH_KMALLOC_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700153#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700154#endif
155
156#ifndef ARCH_SLAB_MINALIGN
Christoph Lameter47bfdc02007-05-06 14:49:37 -0700157#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
Christoph Lameter81819f02007-05-06 14:49:36 -0700158#endif
159
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400160#define OO_SHIFT 16
161#define OO_MASK ((1 << OO_SHIFT) - 1)
162#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
163
Christoph Lameter81819f02007-05-06 14:49:36 -0700164/* Internal SLUB flags */
Christoph Lameter1ceef402007-08-07 15:11:48 -0700165#define __OBJECT_POISON 0x80000000 /* Poison object */
166#define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700167
168static int kmem_size = sizeof(struct kmem_cache);
169
170#ifdef CONFIG_SMP
171static struct notifier_block slab_notifier;
172#endif
173
174static enum {
175 DOWN, /* No slab functionality available */
176 PARTIAL, /* kmem_cache_open() works but kmalloc does not */
Christoph Lameter672bba32007-05-09 02:32:39 -0700177 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700178 SYSFS /* Sysfs up */
179} slab_state = DOWN;
180
181/* A list of all slab caches on the system */
182static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700183static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700184
Christoph Lameter02cbc872007-05-09 02:32:43 -0700185/*
186 * Tracking user of a slab.
187 */
188struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300189 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700190 int cpu; /* Was running on cpu */
191 int pid; /* Pid context */
192 unsigned long when; /* When did the operation occur */
193};
194
195enum track_item { TRACK_ALLOC, TRACK_FREE };
196
Christoph Lameterf6acb632008-04-29 16:16:06 -0700197#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -0700198static int sysfs_slab_add(struct kmem_cache *);
199static int sysfs_slab_alias(struct kmem_cache *, const char *);
200static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800201
Christoph Lameter81819f02007-05-06 14:49:36 -0700202#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700203static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
204static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
205 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800206static inline void sysfs_slab_remove(struct kmem_cache *s)
207{
208 kfree(s);
209}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800210
Christoph Lameter81819f02007-05-06 14:49:36 -0700211#endif
212
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800213static inline void stat(struct kmem_cache_cpu *c, enum stat_item si)
214{
215#ifdef CONFIG_SLUB_STATS
216 c->stat[si]++;
217#endif
218}
219
Christoph Lameter81819f02007-05-06 14:49:36 -0700220/********************************************************************
221 * Core slab cache functions
222 *******************************************************************/
223
224int slab_is_available(void)
225{
226 return slab_state >= UP;
227}
228
229static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
230{
231#ifdef CONFIG_NUMA
232 return s->node[node];
233#else
234 return &s->local_node;
235#endif
236}
237
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700238static inline struct kmem_cache_cpu *get_cpu_slab(struct kmem_cache *s, int cpu)
239{
Christoph Lameter4c93c3552007-10-16 01:26:08 -0700240#ifdef CONFIG_SMP
241 return s->cpu_slab[cpu];
242#else
243 return &s->cpu_slab;
244#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700245}
246
Christoph Lameter6446faa2008-02-15 23:45:26 -0800247/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700248static inline int check_valid_pointer(struct kmem_cache *s,
249 struct page *page, const void *object)
250{
251 void *base;
252
Christoph Lametera973e9d2008-03-01 13:40:44 -0800253 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700254 return 1;
255
Christoph Lametera973e9d2008-03-01 13:40:44 -0800256 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300257 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700258 (object - base) % s->size) {
259 return 0;
260 }
261
262 return 1;
263}
264
Christoph Lameter81819f02007-05-06 14:49:36 -0700265/*
Christoph Lameter7656c722007-05-09 02:32:40 -0700266 * Slow version of get and set free pointer.
267 *
268 * This version requires touching the cache lines of kmem_cache which
269 * we avoid to do in the fast alloc free paths. There we obtain the offset
270 * from the page struct.
271 */
272static inline void *get_freepointer(struct kmem_cache *s, void *object)
273{
274 return *(void **)(object + s->offset);
275}
276
277static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
278{
279 *(void **)(object + s->offset) = fp;
280}
281
282/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300283#define for_each_object(__p, __s, __addr, __objects) \
284 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700285 __p += (__s)->size)
286
287/* Scan freelist */
288#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800289 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700290
291/* Determine object index from a given position */
292static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
293{
294 return (p - addr) / s->size;
295}
296
Christoph Lameter834f3d12008-04-14 19:11:31 +0300297static inline struct kmem_cache_order_objects oo_make(int order,
298 unsigned long size)
299{
300 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400301 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300302 };
303
304 return x;
305}
306
307static inline int oo_order(struct kmem_cache_order_objects x)
308{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400309 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300310}
311
312static inline int oo_objects(struct kmem_cache_order_objects x)
313{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400314 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300315}
316
Christoph Lameter41ecc552007-05-09 02:32:44 -0700317#ifdef CONFIG_SLUB_DEBUG
318/*
319 * Debug settings:
320 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700321#ifdef CONFIG_SLUB_DEBUG_ON
322static int slub_debug = DEBUG_DEFAULT_FLAGS;
323#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700324static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700325#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700326
327static char *slub_debug_slabs;
328
Christoph Lameter7656c722007-05-09 02:32:40 -0700329/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700330 * Object debugging
331 */
332static void print_section(char *text, u8 *addr, unsigned int length)
333{
334 int i, offset;
335 int newline = 1;
336 char ascii[17];
337
338 ascii[16] = 0;
339
340 for (i = 0; i < length; i++) {
341 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700342 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700343 newline = 0;
344 }
Pekka Enberg06428782008-01-07 23:20:27 -0800345 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700346 offset = i % 16;
347 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
348 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800349 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700350 newline = 1;
351 }
352 }
353 if (!newline) {
354 i %= 16;
355 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800356 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700357 ascii[i] = ' ';
358 i++;
359 }
Pekka Enberg06428782008-01-07 23:20:27 -0800360 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700361 }
362}
363
Christoph Lameter81819f02007-05-06 14:49:36 -0700364static struct track *get_track(struct kmem_cache *s, void *object,
365 enum track_item alloc)
366{
367 struct track *p;
368
369 if (s->offset)
370 p = object + s->offset + sizeof(void *);
371 else
372 p = object + s->inuse;
373
374 return p + alloc;
375}
376
377static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300378 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700379{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900380 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700381
Christoph Lameter81819f02007-05-06 14:49:36 -0700382 if (addr) {
383 p->addr = addr;
384 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400385 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700386 p->when = jiffies;
387 } else
388 memset(p, 0, sizeof(struct track));
389}
390
Christoph Lameter81819f02007-05-06 14:49:36 -0700391static void init_tracking(struct kmem_cache *s, void *object)
392{
Christoph Lameter24922682007-07-17 04:03:18 -0700393 if (!(s->flags & SLAB_STORE_USER))
394 return;
395
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300396 set_track(s, object, TRACK_FREE, 0UL);
397 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700398}
399
400static void print_track(const char *s, struct track *t)
401{
402 if (!t->addr)
403 return;
404
Linus Torvalds7daf7052008-07-14 12:12:53 -0700405 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300406 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700407}
408
Christoph Lameter24922682007-07-17 04:03:18 -0700409static void print_tracking(struct kmem_cache *s, void *object)
410{
411 if (!(s->flags & SLAB_STORE_USER))
412 return;
413
414 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
415 print_track("Freed", get_track(s, object, TRACK_FREE));
416}
417
418static void print_page_info(struct page *page)
419{
Christoph Lameter39b26462008-04-14 19:11:30 +0300420 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
421 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700422
423}
424
425static void slab_bug(struct kmem_cache *s, char *fmt, ...)
426{
427 va_list args;
428 char buf[100];
429
430 va_start(args, fmt);
431 vsnprintf(buf, sizeof(buf), fmt, args);
432 va_end(args);
433 printk(KERN_ERR "========================================"
434 "=====================================\n");
435 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
436 printk(KERN_ERR "----------------------------------------"
437 "-------------------------------------\n\n");
438}
439
440static void slab_fix(struct kmem_cache *s, char *fmt, ...)
441{
442 va_list args;
443 char buf[100];
444
445 va_start(args, fmt);
446 vsnprintf(buf, sizeof(buf), fmt, args);
447 va_end(args);
448 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
449}
450
451static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700452{
453 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800454 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700455
456 print_tracking(s, p);
457
458 print_page_info(page);
459
460 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
461 p, p - addr, get_freepointer(s, p));
462
463 if (p > addr + 16)
464 print_section("Bytes b4", p - 16, 16);
465
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300466 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700467
468 if (s->flags & SLAB_RED_ZONE)
469 print_section("Redzone", p + s->objsize,
470 s->inuse - s->objsize);
471
Christoph Lameter81819f02007-05-06 14:49:36 -0700472 if (s->offset)
473 off = s->offset + sizeof(void *);
474 else
475 off = s->inuse;
476
Christoph Lameter24922682007-07-17 04:03:18 -0700477 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700478 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700479
480 if (off != s->size)
481 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700482 print_section("Padding", p + off, s->size - off);
483
484 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700485}
486
487static void object_err(struct kmem_cache *s, struct page *page,
488 u8 *object, char *reason)
489{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700490 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700491 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700492}
493
Christoph Lameter24922682007-07-17 04:03:18 -0700494static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700495{
496 va_list args;
497 char buf[100];
498
Christoph Lameter24922682007-07-17 04:03:18 -0700499 va_start(args, fmt);
500 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700501 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700502 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700503 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700504 dump_stack();
505}
506
507static void init_object(struct kmem_cache *s, void *object, int active)
508{
509 u8 *p = object;
510
511 if (s->flags & __OBJECT_POISON) {
512 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800513 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700514 }
515
516 if (s->flags & SLAB_RED_ZONE)
517 memset(p + s->objsize,
518 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
519 s->inuse - s->objsize);
520}
521
Christoph Lameter24922682007-07-17 04:03:18 -0700522static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700523{
524 while (bytes) {
525 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700526 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700527 start++;
528 bytes--;
529 }
Christoph Lameter24922682007-07-17 04:03:18 -0700530 return NULL;
531}
532
533static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
534 void *from, void *to)
535{
536 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
537 memset(from, data, to - from);
538}
539
540static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
541 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800542 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700543{
544 u8 *fault;
545 u8 *end;
546
547 fault = check_bytes(start, value, bytes);
548 if (!fault)
549 return 1;
550
551 end = start + bytes;
552 while (end > fault && end[-1] == value)
553 end--;
554
555 slab_bug(s, "%s overwritten", what);
556 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
557 fault, end - 1, fault[0], value);
558 print_trailer(s, page, object);
559
560 restore_bytes(s, what, value, fault, end);
561 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700562}
563
Christoph Lameter81819f02007-05-06 14:49:36 -0700564/*
565 * Object layout:
566 *
567 * object address
568 * Bytes of the object to be managed.
569 * If the freepointer may overlay the object then the free
570 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700571 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700572 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
573 * 0xa5 (POISON_END)
574 *
575 * object + s->objsize
576 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700577 * Padding is extended by another word if Redzoning is enabled and
578 * objsize == inuse.
579 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700580 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
581 * 0xcc (RED_ACTIVE) for objects in use.
582 *
583 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700584 * Meta data starts here.
585 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700586 * A. Free pointer (if we cannot overwrite object on free)
587 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700588 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800589 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700590 * before the word boundary.
591 *
592 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700593 *
594 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700595 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700596 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700597 * If slabcaches are merged then the objsize and inuse boundaries are mostly
598 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700599 * may be used with merged slabcaches.
600 */
601
Christoph Lameter81819f02007-05-06 14:49:36 -0700602static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
603{
604 unsigned long off = s->inuse; /* The end of info */
605
606 if (s->offset)
607 /* Freepointer is placed after the object. */
608 off += sizeof(void *);
609
610 if (s->flags & SLAB_STORE_USER)
611 /* We also have user information there */
612 off += 2 * sizeof(struct track);
613
614 if (s->size == off)
615 return 1;
616
Christoph Lameter24922682007-07-17 04:03:18 -0700617 return check_bytes_and_report(s, page, p, "Object padding",
618 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700619}
620
Christoph Lameter39b26462008-04-14 19:11:30 +0300621/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700622static int slab_pad_check(struct kmem_cache *s, struct page *page)
623{
Christoph Lameter24922682007-07-17 04:03:18 -0700624 u8 *start;
625 u8 *fault;
626 u8 *end;
627 int length;
628 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700629
630 if (!(s->flags & SLAB_POISON))
631 return 1;
632
Christoph Lametera973e9d2008-03-01 13:40:44 -0800633 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300634 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300635 end = start + length;
636 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700637 if (!remainder)
638 return 1;
639
Christoph Lameter39b26462008-04-14 19:11:30 +0300640 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700641 if (!fault)
642 return 1;
643 while (end > fault && end[-1] == POISON_INUSE)
644 end--;
645
646 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300647 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700648
649 restore_bytes(s, "slab padding", POISON_INUSE, start, end);
650 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700651}
652
653static int check_object(struct kmem_cache *s, struct page *page,
654 void *object, int active)
655{
656 u8 *p = object;
657 u8 *endobject = object + s->objsize;
658
659 if (s->flags & SLAB_RED_ZONE) {
660 unsigned int red =
661 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
662
Christoph Lameter24922682007-07-17 04:03:18 -0700663 if (!check_bytes_and_report(s, page, object, "Redzone",
664 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700665 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700666 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800667 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
668 check_bytes_and_report(s, page, p, "Alignment padding",
669 endobject, POISON_INUSE, s->inuse - s->objsize);
670 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700671 }
672
673 if (s->flags & SLAB_POISON) {
674 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700675 (!check_bytes_and_report(s, page, p, "Poison", p,
676 POISON_FREE, s->objsize - 1) ||
677 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800678 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700679 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700680 /*
681 * check_pad_bytes cleans up on its own.
682 */
683 check_pad_bytes(s, page, p);
684 }
685
686 if (!s->offset && active)
687 /*
688 * Object and freepointer overlap. Cannot check
689 * freepointer while object is allocated.
690 */
691 return 1;
692
693 /* Check free pointer validity */
694 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
695 object_err(s, page, p, "Freepointer corrupt");
696 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100697 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700698 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700699 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700700 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800701 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700702 return 0;
703 }
704 return 1;
705}
706
707static int check_slab(struct kmem_cache *s, struct page *page)
708{
Christoph Lameter39b26462008-04-14 19:11:30 +0300709 int maxobj;
710
Christoph Lameter81819f02007-05-06 14:49:36 -0700711 VM_BUG_ON(!irqs_disabled());
712
713 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700714 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700715 return 0;
716 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300717
718 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
719 if (page->objects > maxobj) {
720 slab_err(s, page, "objects %u > max %u",
721 s->name, page->objects, maxobj);
722 return 0;
723 }
724 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700725 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300726 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700727 return 0;
728 }
729 /* Slab_pad_check fixes things up after itself */
730 slab_pad_check(s, page);
731 return 1;
732}
733
734/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700735 * Determine if a certain object on a page is on the freelist. Must hold the
736 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700737 */
738static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
739{
740 int nr = 0;
741 void *fp = page->freelist;
742 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300743 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700744
Christoph Lameter39b26462008-04-14 19:11:30 +0300745 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700746 if (fp == search)
747 return 1;
748 if (!check_valid_pointer(s, page, fp)) {
749 if (object) {
750 object_err(s, page, object,
751 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800752 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700753 break;
754 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700755 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800756 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300757 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700758 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700759 return 0;
760 }
761 break;
762 }
763 object = fp;
764 fp = get_freepointer(s, object);
765 nr++;
766 }
767
Christoph Lameter224a88b2008-04-14 19:11:31 +0300768 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400769 if (max_objects > MAX_OBJS_PER_PAGE)
770 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300771
772 if (page->objects != max_objects) {
773 slab_err(s, page, "Wrong number of objects. Found %d but "
774 "should be %d", page->objects, max_objects);
775 page->objects = max_objects;
776 slab_fix(s, "Number of objects adjusted.");
777 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300778 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700779 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300780 "counted were %d", page->inuse, page->objects - nr);
781 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700782 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700783 }
784 return search == NULL;
785}
786
Christoph Lameter0121c6192008-04-29 16:11:12 -0700787static void trace(struct kmem_cache *s, struct page *page, void *object,
788 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700789{
790 if (s->flags & SLAB_TRACE) {
791 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
792 s->name,
793 alloc ? "alloc" : "free",
794 object, page->inuse,
795 page->freelist);
796
797 if (!alloc)
798 print_section("Object", (void *)object, s->objsize);
799
800 dump_stack();
801 }
802}
803
Christoph Lameter643b1132007-05-06 14:49:42 -0700804/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700805 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700806 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700807static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700808{
Christoph Lameter643b1132007-05-06 14:49:42 -0700809 spin_lock(&n->list_lock);
810 list_add(&page->lru, &n->full);
811 spin_unlock(&n->list_lock);
812}
813
814static void remove_full(struct kmem_cache *s, struct page *page)
815{
816 struct kmem_cache_node *n;
817
818 if (!(s->flags & SLAB_STORE_USER))
819 return;
820
821 n = get_node(s, page_to_nid(page));
822
823 spin_lock(&n->list_lock);
824 list_del(&page->lru);
825 spin_unlock(&n->list_lock);
826}
827
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300828/* Tracking of the number of slabs for debugging purposes */
829static inline unsigned long slabs_node(struct kmem_cache *s, int node)
830{
831 struct kmem_cache_node *n = get_node(s, node);
832
833 return atomic_long_read(&n->nr_slabs);
834}
835
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400836static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
837{
838 return atomic_long_read(&n->nr_slabs);
839}
840
Christoph Lameter205ab992008-04-14 19:11:40 +0300841static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300842{
843 struct kmem_cache_node *n = get_node(s, node);
844
845 /*
846 * May be called early in order to allocate a slab for the
847 * kmem_cache_node structure. Solve the chicken-egg
848 * dilemma by deferring the increment of the count during
849 * bootstrap (see early_kmem_cache_node_alloc).
850 */
Christoph Lameter205ab992008-04-14 19:11:40 +0300851 if (!NUMA_BUILD || n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300852 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300853 atomic_long_add(objects, &n->total_objects);
854 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300855}
Christoph Lameter205ab992008-04-14 19:11:40 +0300856static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300857{
858 struct kmem_cache_node *n = get_node(s, node);
859
860 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300861 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300862}
863
864/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700865static void setup_object_debug(struct kmem_cache *s, struct page *page,
866 void *object)
867{
868 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
869 return;
870
871 init_object(s, object, 0);
872 init_tracking(s, object);
873}
874
875static int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300876 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700877{
878 if (!check_slab(s, page))
879 goto bad;
880
Christoph Lameterd692ef62008-02-15 23:45:24 -0800881 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700882 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700883 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700884 }
885
886 if (!check_valid_pointer(s, page, object)) {
887 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700888 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700889 }
890
Christoph Lameterd692ef62008-02-15 23:45:24 -0800891 if (!check_object(s, page, object, 0))
Christoph Lameter81819f02007-05-06 14:49:36 -0700892 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700893
Christoph Lameter3ec09742007-05-16 22:11:00 -0700894 /* Success perform special debug activities for allocs */
895 if (s->flags & SLAB_STORE_USER)
896 set_track(s, object, TRACK_ALLOC, addr);
897 trace(s, page, object, 1);
898 init_object(s, object, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700899 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700900
Christoph Lameter81819f02007-05-06 14:49:36 -0700901bad:
902 if (PageSlab(page)) {
903 /*
904 * If this is a slab page then lets do the best we can
905 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700906 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700907 */
Christoph Lameter24922682007-07-17 04:03:18 -0700908 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300909 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800910 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700911 }
912 return 0;
913}
914
Christoph Lameter3ec09742007-05-16 22:11:00 -0700915static int free_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300916 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700917{
918 if (!check_slab(s, page))
919 goto fail;
920
921 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700922 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700923 goto fail;
924 }
925
926 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700927 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700928 goto fail;
929 }
930
931 if (!check_object(s, page, object, 1))
932 return 0;
933
934 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800935 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700936 slab_err(s, page, "Attempt to free object(0x%p) "
937 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800938 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700939 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700940 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700941 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700942 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800943 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700944 object_err(s, page, object,
945 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700946 goto fail;
947 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700948
949 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700950 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700951 remove_full(s, page);
952 if (s->flags & SLAB_STORE_USER)
953 set_track(s, object, TRACK_FREE, addr);
954 trace(s, page, object, 0);
955 init_object(s, object, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -0700956 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700957
Christoph Lameter81819f02007-05-06 14:49:36 -0700958fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700959 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700960 return 0;
961}
962
Christoph Lameter41ecc552007-05-09 02:32:44 -0700963static int __init setup_slub_debug(char *str)
964{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700965 slub_debug = DEBUG_DEFAULT_FLAGS;
966 if (*str++ != '=' || !*str)
967 /*
968 * No options specified. Switch on full debugging.
969 */
970 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700971
972 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700973 /*
974 * No options but restriction on slabs. This means full
975 * debugging for slabs matching a pattern.
976 */
977 goto check_slabs;
978
979 slub_debug = 0;
980 if (*str == '-')
981 /*
982 * Switch off all debugging measures.
983 */
984 goto out;
985
986 /*
987 * Determine which debug features should be switched on
988 */
Pekka Enberg06428782008-01-07 23:20:27 -0800989 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700990 switch (tolower(*str)) {
991 case 'f':
992 slub_debug |= SLAB_DEBUG_FREE;
993 break;
994 case 'z':
995 slub_debug |= SLAB_RED_ZONE;
996 break;
997 case 'p':
998 slub_debug |= SLAB_POISON;
999 break;
1000 case 'u':
1001 slub_debug |= SLAB_STORE_USER;
1002 break;
1003 case 't':
1004 slub_debug |= SLAB_TRACE;
1005 break;
1006 default:
1007 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001008 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001009 }
1010 }
1011
1012check_slabs:
1013 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001014 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001015out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001016 return 1;
1017}
1018
1019__setup("slub_debug", setup_slub_debug);
1020
Christoph Lameterba0268a2007-09-11 15:24:11 -07001021static unsigned long kmem_cache_flags(unsigned long objsize,
1022 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001023 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001024{
1025 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001026 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001027 */
Christoph Lametere1533622008-02-15 23:45:24 -08001028 if (slub_debug && (!slub_debug_slabs ||
1029 strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)) == 0))
1030 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001031
1032 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001033}
1034#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001035static inline void setup_object_debug(struct kmem_cache *s,
1036 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001037
Christoph Lameter3ec09742007-05-16 22:11:00 -07001038static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001039 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001040
Christoph Lameter3ec09742007-05-16 22:11:00 -07001041static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001042 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001043
Christoph Lameter41ecc552007-05-09 02:32:44 -07001044static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1045 { return 1; }
1046static inline int check_object(struct kmem_cache *s, struct page *page,
1047 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001048static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001049static inline unsigned long kmem_cache_flags(unsigned long objsize,
1050 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001051 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001052{
1053 return flags;
1054}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001055#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001056
1057static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1058 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001059static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1060 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001061static inline void inc_slabs_node(struct kmem_cache *s, int node,
1062 int objects) {}
1063static inline void dec_slabs_node(struct kmem_cache *s, int node,
1064 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001065#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001066
Christoph Lameter81819f02007-05-06 14:49:36 -07001067/*
1068 * Slab allocation and freeing
1069 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001070static inline struct page *alloc_slab_page(gfp_t flags, int node,
1071 struct kmem_cache_order_objects oo)
1072{
1073 int order = oo_order(oo);
1074
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001075 flags |= __GFP_NOTRACK;
1076
Christoph Lameter65c33762008-04-14 19:11:40 +03001077 if (node == -1)
1078 return alloc_pages(flags, order);
1079 else
1080 return alloc_pages_node(node, flags, order);
1081}
1082
Christoph Lameter81819f02007-05-06 14:49:36 -07001083static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1084{
Pekka Enberg06428782008-01-07 23:20:27 -08001085 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001086 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001087 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001088
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001089 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001090
Pekka Enbergba522702009-06-24 21:59:51 +03001091 /*
1092 * Let the initial higher-order allocation fail under memory pressure
1093 * so we fall-back to the minimum order allocation.
1094 */
1095 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1096
1097 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001098 if (unlikely(!page)) {
1099 oo = s->min;
1100 /*
1101 * Allocation may have failed due to fragmentation.
1102 * Try a lower order alloc if possible
1103 */
1104 page = alloc_slab_page(flags, node, oo);
1105 if (!page)
1106 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001107
Christoph Lameter65c33762008-04-14 19:11:40 +03001108 stat(get_cpu_slab(s, raw_smp_processor_id()), ORDER_FALLBACK);
1109 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001110
1111 if (kmemcheck_enabled
1112 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS)))
1113 {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001114 int pages = 1 << oo_order(oo);
1115
1116 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1117
1118 /*
1119 * Objects from caches that have a constructor don't get
1120 * cleared when they're allocated, so we need to do it here.
1121 */
1122 if (s->ctor)
1123 kmemcheck_mark_uninitialized_pages(page, pages);
1124 else
1125 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001126 }
1127
Christoph Lameter834f3d12008-04-14 19:11:31 +03001128 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001129 mod_zone_page_state(page_zone(page),
1130 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1131 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001132 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001133
1134 return page;
1135}
1136
1137static void setup_object(struct kmem_cache *s, struct page *page,
1138 void *object)
1139{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001140 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001141 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001142 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001143}
1144
1145static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1146{
1147 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001148 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001149 void *last;
1150 void *p;
1151
Christoph Lameter6cb06222007-10-16 01:25:41 -07001152 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001153
Christoph Lameter6cb06222007-10-16 01:25:41 -07001154 page = allocate_slab(s,
1155 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001156 if (!page)
1157 goto out;
1158
Christoph Lameter205ab992008-04-14 19:11:40 +03001159 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001160 page->slab = s;
1161 page->flags |= 1 << PG_slab;
1162 if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
1163 SLAB_STORE_USER | SLAB_TRACE))
Andy Whitcroft8a380822008-07-23 21:27:18 -07001164 __SetPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001165
1166 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001167
1168 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001169 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001170
1171 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001172 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001173 setup_object(s, page, last);
1174 set_freepointer(s, last, p);
1175 last = p;
1176 }
1177 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001178 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001179
1180 page->freelist = start;
1181 page->inuse = 0;
1182out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001183 return page;
1184}
1185
1186static void __free_slab(struct kmem_cache *s, struct page *page)
1187{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001188 int order = compound_order(page);
1189 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001190
Andy Whitcroft8a380822008-07-23 21:27:18 -07001191 if (unlikely(SLABDEBUG && PageSlubDebug(page))) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001192 void *p;
1193
1194 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001195 for_each_object(p, s, page_address(page),
1196 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001197 check_object(s, page, p, 0);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001198 __ClearPageSlubDebug(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001199 }
1200
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001201 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001202
Christoph Lameter81819f02007-05-06 14:49:36 -07001203 mod_zone_page_state(page_zone(page),
1204 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1205 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001206 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001207
Christoph Lameter49bd5222008-04-14 18:52:18 +03001208 __ClearPageSlab(page);
1209 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001210 if (current->reclaim_state)
1211 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001212 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001213}
1214
1215static void rcu_free_slab(struct rcu_head *h)
1216{
1217 struct page *page;
1218
1219 page = container_of((struct list_head *)h, struct page, lru);
1220 __free_slab(page->slab, page);
1221}
1222
1223static void free_slab(struct kmem_cache *s, struct page *page)
1224{
1225 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1226 /*
1227 * RCU free overloads the RCU head over the LRU
1228 */
1229 struct rcu_head *head = (void *)&page->lru;
1230
1231 call_rcu(head, rcu_free_slab);
1232 } else
1233 __free_slab(s, page);
1234}
1235
1236static void discard_slab(struct kmem_cache *s, struct page *page)
1237{
Christoph Lameter205ab992008-04-14 19:11:40 +03001238 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001239 free_slab(s, page);
1240}
1241
1242/*
1243 * Per slab locking using the pagelock
1244 */
1245static __always_inline void slab_lock(struct page *page)
1246{
1247 bit_spin_lock(PG_locked, &page->flags);
1248}
1249
1250static __always_inline void slab_unlock(struct page *page)
1251{
Nick Piggina76d3542008-01-07 23:20:27 -08001252 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001253}
1254
1255static __always_inline int slab_trylock(struct page *page)
1256{
1257 int rc = 1;
1258
1259 rc = bit_spin_trylock(PG_locked, &page->flags);
1260 return rc;
1261}
1262
1263/*
1264 * Management of partially allocated slabs
1265 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001266static void add_partial(struct kmem_cache_node *n,
1267 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001268{
Christoph Lametere95eed52007-05-06 14:49:44 -07001269 spin_lock(&n->list_lock);
1270 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001271 if (tail)
1272 list_add_tail(&page->lru, &n->partial);
1273 else
1274 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001275 spin_unlock(&n->list_lock);
1276}
1277
Christoph Lameter0121c6192008-04-29 16:11:12 -07001278static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001279{
1280 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1281
1282 spin_lock(&n->list_lock);
1283 list_del(&page->lru);
1284 n->nr_partial--;
1285 spin_unlock(&n->list_lock);
1286}
1287
1288/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001289 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001290 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001291 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001292 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001293static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1294 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001295{
1296 if (slab_trylock(page)) {
1297 list_del(&page->lru);
1298 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001299 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001300 return 1;
1301 }
1302 return 0;
1303}
1304
1305/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001306 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001307 */
1308static struct page *get_partial_node(struct kmem_cache_node *n)
1309{
1310 struct page *page;
1311
1312 /*
1313 * Racy check. If we mistakenly see no partial slabs then we
1314 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001315 * partial slab and there is none available then get_partials()
1316 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001317 */
1318 if (!n || !n->nr_partial)
1319 return NULL;
1320
1321 spin_lock(&n->list_lock);
1322 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001323 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001324 goto out;
1325 page = NULL;
1326out:
1327 spin_unlock(&n->list_lock);
1328 return page;
1329}
1330
1331/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001332 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001333 */
1334static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1335{
1336#ifdef CONFIG_NUMA
1337 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001338 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001339 struct zone *zone;
1340 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001341 struct page *page;
1342
1343 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001344 * The defrag ratio allows a configuration of the tradeoffs between
1345 * inter node defragmentation and node local allocations. A lower
1346 * defrag_ratio increases the tendency to do local allocations
1347 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001348 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001349 * If the defrag_ratio is set to 0 then kmalloc() always
1350 * returns node local objects. If the ratio is higher then kmalloc()
1351 * may return off node objects because partial slabs are obtained
1352 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001353 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001354 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001355 * defrag_ratio = 1000) then every (well almost) allocation will
1356 * first attempt to defrag slab caches on other nodes. This means
1357 * scanning over all nodes to look for partial slabs which may be
1358 * expensive if we do it every time we are trying to find a slab
1359 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001360 */
Christoph Lameter98246012008-01-07 23:20:26 -08001361 if (!s->remote_node_defrag_ratio ||
1362 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001363 return NULL;
1364
Mel Gorman0e884602008-04-28 02:12:14 -07001365 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001366 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001367 struct kmem_cache_node *n;
1368
Mel Gorman54a6eb52008-04-28 02:12:16 -07001369 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001370
Mel Gorman54a6eb52008-04-28 02:12:16 -07001371 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001372 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001373 page = get_partial_node(n);
1374 if (page)
1375 return page;
1376 }
1377 }
1378#endif
1379 return NULL;
1380}
1381
1382/*
1383 * Get a partial page, lock it and return it.
1384 */
1385static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1386{
1387 struct page *page;
1388 int searchnode = (node == -1) ? numa_node_id() : node;
1389
1390 page = get_partial_node(get_node(s, searchnode));
1391 if (page || (flags & __GFP_THISNODE))
1392 return page;
1393
1394 return get_any_partial(s, flags);
1395}
1396
1397/*
1398 * Move a page back to the lists.
1399 *
1400 * Must be called with the slab lock held.
1401 *
1402 * On exit the slab lock will have been dropped.
1403 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001404static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001405{
Christoph Lametere95eed52007-05-06 14:49:44 -07001406 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001407 struct kmem_cache_cpu *c = get_cpu_slab(s, smp_processor_id());
Christoph Lametere95eed52007-05-06 14:49:44 -07001408
Andy Whitcroft8a380822008-07-23 21:27:18 -07001409 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001410 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001411
Christoph Lametera973e9d2008-03-01 13:40:44 -08001412 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001413 add_partial(n, page, tail);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001414 stat(c, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1415 } else {
1416 stat(c, DEACTIVATE_FULL);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001417 if (SLABDEBUG && PageSlubDebug(page) &&
1418 (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001419 add_full(n, page);
1420 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001421 slab_unlock(page);
1422 } else {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001423 stat(c, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001424 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001425 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001426 * Adding an empty slab to the partial slabs in order
1427 * to avoid page allocator overhead. This slab needs
1428 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001429 * so that the others get filled first. That way the
1430 * size of the partial list stays small.
1431 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001432 * kmem_cache_shrink can reclaim any empty slabs from
1433 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001434 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001435 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001436 slab_unlock(page);
1437 } else {
1438 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001439 stat(get_cpu_slab(s, raw_smp_processor_id()), FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001440 discard_slab(s, page);
1441 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001442 }
1443}
1444
1445/*
1446 * Remove the cpu slab
1447 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001448static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001449{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001450 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001451 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001452
Christoph Lameterb773ad72008-03-04 11:10:17 -08001453 if (page->freelist)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001454 stat(c, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001455 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001456 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001457 * because both freelists are empty. So this is unlikely
1458 * to occur.
1459 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001460 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001461 void **object;
1462
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001463 tail = 0; /* Hot objects. Put the slab first */
1464
Christoph Lameter894b8782007-05-10 03:15:16 -07001465 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001466 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001467 c->freelist = c->freelist[c->offset];
Christoph Lameter894b8782007-05-10 03:15:16 -07001468
1469 /* And put onto the regular freelist */
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001470 object[c->offset] = page->freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07001471 page->freelist = object;
1472 page->inuse--;
1473 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001474 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001475 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001476}
1477
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001478static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001479{
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001480 stat(c, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001481 slab_lock(c->page);
1482 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001483}
1484
1485/*
1486 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001487 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001488 * Called from IPI handler with interrupts disabled.
1489 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001490static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001491{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001492 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001493
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001494 if (likely(c && c->page))
1495 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001496}
1497
1498static void flush_cpu_slab(void *d)
1499{
1500 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001501
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001502 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001503}
1504
1505static void flush_all(struct kmem_cache *s)
1506{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001507 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001508}
1509
1510/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001511 * Check if the objects in a per cpu structure fit numa
1512 * locality expectations.
1513 */
1514static inline int node_match(struct kmem_cache_cpu *c, int node)
1515{
1516#ifdef CONFIG_NUMA
1517 if (node != -1 && c->node != node)
1518 return 0;
1519#endif
1520 return 1;
1521}
1522
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001523static int count_free(struct page *page)
1524{
1525 return page->objects - page->inuse;
1526}
1527
1528static unsigned long count_partial(struct kmem_cache_node *n,
1529 int (*get_count)(struct page *))
1530{
1531 unsigned long flags;
1532 unsigned long x = 0;
1533 struct page *page;
1534
1535 spin_lock_irqsave(&n->list_lock, flags);
1536 list_for_each_entry(page, &n->partial, lru)
1537 x += get_count(page);
1538 spin_unlock_irqrestore(&n->list_lock, flags);
1539 return x;
1540}
1541
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001542static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1543{
1544#ifdef CONFIG_SLUB_DEBUG
1545 return atomic_long_read(&n->total_objects);
1546#else
1547 return 0;
1548#endif
1549}
1550
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001551static noinline void
1552slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1553{
1554 int node;
1555
1556 printk(KERN_WARNING
1557 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1558 nid, gfpflags);
1559 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1560 "default order: %d, min order: %d\n", s->name, s->objsize,
1561 s->size, oo_order(s->oo), oo_order(s->min));
1562
1563 for_each_online_node(node) {
1564 struct kmem_cache_node *n = get_node(s, node);
1565 unsigned long nr_slabs;
1566 unsigned long nr_objs;
1567 unsigned long nr_free;
1568
1569 if (!n)
1570 continue;
1571
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001572 nr_free = count_partial(n, count_free);
1573 nr_slabs = node_nr_slabs(n);
1574 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001575
1576 printk(KERN_WARNING
1577 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1578 node, nr_slabs, nr_objs, nr_free);
1579 }
1580}
1581
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001582/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001583 * Slow path. The lockless freelist is empty or we need to perform
1584 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001585 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001586 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001587 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001588 * Processing is still very fast if new objects have been freed to the
1589 * regular freelist. In that case we simply take over the regular freelist
1590 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001591 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001592 * If that is not working then we fall back to the partial lists. We take the
1593 * first element of the freelist as the object to allocate now and move the
1594 * rest of the freelist to the lockless freelist.
1595 *
1596 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001597 * we need to allocate a new slab. This is the slowest path since it involves
1598 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001599 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001600static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1601 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001602{
Christoph Lameter81819f02007-05-06 14:49:36 -07001603 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001604 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001605
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001606 /* We handle __GFP_ZERO in the caller */
1607 gfpflags &= ~__GFP_ZERO;
1608
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001609 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001610 goto new_slab;
1611
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001612 slab_lock(c->page);
1613 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001614 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001615
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001616 stat(c, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001617
Christoph Lameter894b8782007-05-10 03:15:16 -07001618load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001619 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001620 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001621 goto another_slab;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001622 if (unlikely(SLABDEBUG && PageSlubDebug(c->page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001623 goto debug;
1624
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001625 c->freelist = object[c->offset];
Christoph Lameter39b26462008-04-14 19:11:30 +03001626 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001627 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001628 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001629unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001630 slab_unlock(c->page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001631 stat(c, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001632 return object;
1633
1634another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001635 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001636
1637new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001638 new = get_partial(s, gfpflags, node);
1639 if (new) {
1640 c->page = new;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001641 stat(c, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001642 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001643 }
1644
Christoph Lameterb811c202007-10-16 23:25:51 -07001645 if (gfpflags & __GFP_WAIT)
1646 local_irq_enable();
1647
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001648 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001649
1650 if (gfpflags & __GFP_WAIT)
1651 local_irq_disable();
1652
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001653 if (new) {
1654 c = get_cpu_slab(s, smp_processor_id());
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001655 stat(c, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001656 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001657 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001658 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001659 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001660 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001661 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001662 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001663 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1664 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001665 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001666debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001667 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001668 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001669
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001670 c->page->inuse++;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001671 c->page->freelist = object[c->offset];
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001672 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001673 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001674}
1675
1676/*
1677 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1678 * have the fastpath folded into their functions. So no function call
1679 * overhead for requests that can be satisfied on the fastpath.
1680 *
1681 * The fastpath works by first checking if the lockless freelist can be used.
1682 * If not then __slab_alloc is called for slow processing.
1683 *
1684 * Otherwise we can simply pick the next object from the lockless free list.
1685 */
Pekka Enberg06428782008-01-07 23:20:27 -08001686static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001687 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001688{
Christoph Lameter894b8782007-05-10 03:15:16 -07001689 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001690 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001691 unsigned long flags;
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001692 unsigned int objsize;
Christoph Lameter1f842602008-01-07 23:20:30 -08001693
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001694 gfpflags &= gfp_allowed_mask;
Pekka Enberg7e85ee02009-06-12 14:03:06 +03001695
Nick Piggincf40bd12009-01-21 08:12:39 +01001696 lockdep_trace_alloc(gfpflags);
OGAWA Hirofumi89124d72008-11-19 21:23:59 +09001697 might_sleep_if(gfpflags & __GFP_WAIT);
Pekka Enberg3c506ef2008-12-29 11:47:05 +02001698
Akinobu Mita773ff602008-12-23 19:37:01 +09001699 if (should_failslab(s->objsize, gfpflags))
1700 return NULL;
1701
Christoph Lameter894b8782007-05-10 03:15:16 -07001702 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001703 c = get_cpu_slab(s, smp_processor_id());
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001704 objsize = c->objsize;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001705 if (unlikely(!c->freelist || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001706
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001707 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001708
1709 else {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001710 object = c->freelist;
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001711 c->freelist = object[c->offset];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001712 stat(c, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001713 }
1714 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001715
1716 if (unlikely((gfpflags & __GFP_ZERO) && object))
Dmitry Adamushkobdb21922008-07-10 22:21:58 +02001717 memset(object, 0, objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001718
Vegard Nossum5a896d92008-04-04 00:54:48 +02001719 kmemcheck_slab_alloc(s, gfpflags, object, c->objsize);
Catalin Marinas06f22f12009-06-11 13:23:18 +01001720 kmemleak_alloc_recursive(object, objsize, 1, s->flags, gfpflags);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001721
Christoph Lameter894b8782007-05-10 03:15:16 -07001722 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001723}
1724
1725void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1726{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001727 void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_);
1728
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001729 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001730
1731 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001732}
1733EXPORT_SYMBOL(kmem_cache_alloc);
1734
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001735#ifdef CONFIG_KMEMTRACE
1736void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1737{
1738 return slab_alloc(s, gfpflags, -1, _RET_IP_);
1739}
1740EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1741#endif
1742
Christoph Lameter81819f02007-05-06 14:49:36 -07001743#ifdef CONFIG_NUMA
1744void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1745{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001746 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1747
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001748 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1749 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001750
1751 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001752}
1753EXPORT_SYMBOL(kmem_cache_alloc_node);
1754#endif
1755
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001756#ifdef CONFIG_KMEMTRACE
1757void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1758 gfp_t gfpflags,
1759 int node)
1760{
1761 return slab_alloc(s, gfpflags, node, _RET_IP_);
1762}
1763EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1764#endif
1765
Christoph Lameter81819f02007-05-06 14:49:36 -07001766/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001767 * Slow patch handling. This may still be called frequently since objects
1768 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001769 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001770 * So we still attempt to reduce cache line usage. Just take the slab
1771 * lock and free the item. If there is no additional partial page
1772 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001773 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001774static void __slab_free(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001775 void *x, unsigned long addr, unsigned int offset)
Christoph Lameter81819f02007-05-06 14:49:36 -07001776{
1777 void *prior;
1778 void **object = (void *)x;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001779 struct kmem_cache_cpu *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07001780
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001781 c = get_cpu_slab(s, raw_smp_processor_id());
1782 stat(c, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001783 slab_lock(page);
1784
Andy Whitcroft8a380822008-07-23 21:27:18 -07001785 if (unlikely(SLABDEBUG && PageSlubDebug(page)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001786 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001787
Christoph Lameter81819f02007-05-06 14:49:36 -07001788checks_ok:
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001789 prior = object[offset] = page->freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001790 page->freelist = object;
1791 page->inuse--;
1792
Andy Whitcroft8a380822008-07-23 21:27:18 -07001793 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001794 stat(c, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001795 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001796 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001797
1798 if (unlikely(!page->inuse))
1799 goto slab_empty;
1800
1801 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001802 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001803 * then add it.
1804 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001805 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001806 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001807 stat(c, FREE_ADD_PARTIAL);
1808 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001809
1810out_unlock:
1811 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001812 return;
1813
1814slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001815 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001816 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001817 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001818 */
1819 remove_partial(s, page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001820 stat(c, FREE_REMOVE_PARTIAL);
1821 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001822 slab_unlock(page);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001823 stat(c, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001824 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001825 return;
1826
1827debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001828 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001829 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001830 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001831}
1832
Christoph Lameter894b8782007-05-10 03:15:16 -07001833/*
1834 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1835 * can perform fastpath freeing without additional function calls.
1836 *
1837 * The fastpath is only possible if we are freeing to the current cpu slab
1838 * of this processor. This typically the case if we have just allocated
1839 * the item before.
1840 *
1841 * If fastpath is not possible then fall back to __slab_free where we deal
1842 * with all sorts of special processing.
1843 */
Pekka Enberg06428782008-01-07 23:20:27 -08001844static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001845 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001846{
1847 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001848 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001849 unsigned long flags;
1850
Catalin Marinas06f22f12009-06-11 13:23:18 +01001851 kmemleak_free_recursive(x, s->flags);
Christoph Lameter894b8782007-05-10 03:15:16 -07001852 local_irq_save(flags);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001853 c = get_cpu_slab(s, smp_processor_id());
Vegard Nossum5a896d92008-04-04 00:54:48 +02001854 kmemcheck_slab_free(s, object, c->objsize);
Christoph Lameter27d9e4e2008-02-15 23:45:25 -08001855 debug_check_no_locks_freed(object, c->objsize);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001856 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Pekka Enberg6047a002009-01-14 12:22:25 +02001857 debug_check_no_obj_freed(object, c->objsize);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001858 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001859 object[c->offset] = c->freelist;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001860 c->freelist = object;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001861 stat(c, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001862 } else
Christoph Lameterb3fba8d2007-10-16 01:26:06 -07001863 __slab_free(s, page, x, addr, c->offset);
Christoph Lameter894b8782007-05-10 03:15:16 -07001864
1865 local_irq_restore(flags);
1866}
1867
Christoph Lameter81819f02007-05-06 14:49:36 -07001868void kmem_cache_free(struct kmem_cache *s, void *x)
1869{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001870 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001871
Christoph Lameterb49af682007-05-06 14:49:41 -07001872 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001873
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001874 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001875
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001876 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001877}
1878EXPORT_SYMBOL(kmem_cache_free);
1879
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001880/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001881static struct page *get_object_page(const void *x)
1882{
Christoph Lameterb49af682007-05-06 14:49:41 -07001883 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001884
1885 if (!PageSlab(page))
1886 return NULL;
1887
1888 return page;
1889}
1890
1891/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001892 * Object placement in a slab is made very easy because we always start at
1893 * offset 0. If we tune the size of the object to the alignment then we can
1894 * get the required alignment by putting one properly sized object after
1895 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001896 *
1897 * Notice that the allocation order determines the sizes of the per cpu
1898 * caches. Each processor has always one slab available for allocations.
1899 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001900 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001901 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001902 */
1903
1904/*
1905 * Mininum / Maximum order of slab pages. This influences locking overhead
1906 * and slab fragmentation. A higher order reduces the number of partial slabs
1907 * and increases the number of allocations possible without having to
1908 * take the list_lock.
1909 */
1910static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001911static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001912static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001913
1914/*
1915 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001916 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001917 */
1918static int slub_nomerge;
1919
1920/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001921 * Calculate the order of allocation given an slab object size.
1922 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001923 * The order of allocation has significant impact on performance and other
1924 * system components. Generally order 0 allocations should be preferred since
1925 * order 0 does not cause fragmentation in the page allocator. Larger objects
1926 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001927 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001928 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001929 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001930 * In order to reach satisfactory performance we must ensure that a minimum
1931 * number of objects is in one slab. Otherwise we may generate too much
1932 * activity on the partial lists which requires taking the list_lock. This is
1933 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001934 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001935 * slub_max_order specifies the order where we begin to stop considering the
1936 * number of objects in a slab as critical. If we reach slub_max_order then
1937 * we try to keep the page order as low as possible. So we accept more waste
1938 * of space in favor of a small page order.
1939 *
1940 * Higher order allocations also allow the placement of more objects in a
1941 * slab and thereby reduce object handling overhead. If the user has
1942 * requested a higher mininum order then we start with that one instead of
1943 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001944 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001945static inline int slab_order(int size, int min_objects,
1946 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001947{
1948 int order;
1949 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001950 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001951
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001952 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1953 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001954
Christoph Lameter6300ea72007-07-17 04:03:20 -07001955 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001956 fls(min_objects * size - 1) - PAGE_SHIFT);
1957 order <= max_order; order++) {
1958
Christoph Lameter81819f02007-05-06 14:49:36 -07001959 unsigned long slab_size = PAGE_SIZE << order;
1960
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001961 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001962 continue;
1963
Christoph Lameter81819f02007-05-06 14:49:36 -07001964 rem = slab_size % size;
1965
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001966 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001967 break;
1968
1969 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001970
Christoph Lameter81819f02007-05-06 14:49:36 -07001971 return order;
1972}
1973
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001974static inline int calculate_order(int size)
1975{
1976 int order;
1977 int min_objects;
1978 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001979 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001980
1981 /*
1982 * Attempt to find best configuration for a slab. This
1983 * works by first attempting to generate a layout with
1984 * the best configuration and backing off gradually.
1985 *
1986 * First we reduce the acceptable waste in a slab. Then
1987 * we reduce the minimum objects required in a slab.
1988 */
1989 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001990 if (!min_objects)
1991 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001992 max_objects = (PAGE_SIZE << slub_max_order)/size;
1993 min_objects = min(min_objects, max_objects);
1994
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001995 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001996 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001997 while (fraction >= 4) {
1998 order = slab_order(size, min_objects,
1999 slub_max_order, fraction);
2000 if (order <= slub_max_order)
2001 return order;
2002 fraction /= 2;
2003 }
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002004 min_objects --;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002005 }
2006
2007 /*
2008 * We were unable to place multiple objects in a slab. Now
2009 * lets see if we can place a single object there.
2010 */
2011 order = slab_order(size, 1, slub_max_order, 1);
2012 if (order <= slub_max_order)
2013 return order;
2014
2015 /*
2016 * Doh this slab cannot be placed using slub_max_order.
2017 */
2018 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002019 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002020 return order;
2021 return -ENOSYS;
2022}
2023
Christoph Lameter81819f02007-05-06 14:49:36 -07002024/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002025 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002026 */
2027static unsigned long calculate_alignment(unsigned long flags,
2028 unsigned long align, unsigned long size)
2029{
2030 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002031 * If the user wants hardware cache aligned objects then follow that
2032 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002033 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002034 * The hardware cache alignment cannot override the specified
2035 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002036 */
Nick Pigginb6210382008-03-05 14:05:56 -08002037 if (flags & SLAB_HWCACHE_ALIGN) {
2038 unsigned long ralign = cache_line_size();
2039 while (size <= ralign / 2)
2040 ralign /= 2;
2041 align = max(align, ralign);
2042 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002043
2044 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002045 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002046
2047 return ALIGN(align, sizeof(void *));
2048}
2049
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002050static void init_kmem_cache_cpu(struct kmem_cache *s,
2051 struct kmem_cache_cpu *c)
2052{
2053 c->page = NULL;
Christoph Lametera973e9d2008-03-01 13:40:44 -08002054 c->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002055 c->node = 0;
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07002056 c->offset = s->offset / sizeof(void *);
2057 c->objsize = s->objsize;
Pekka Enberg62f75532008-04-14 18:50:44 +03002058#ifdef CONFIG_SLUB_STATS
2059 memset(c->stat, 0, NR_SLUB_STAT_ITEMS * sizeof(unsigned));
2060#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002061}
2062
Pekka Enberg5595cff2008-08-05 09:28:47 +03002063static void
2064init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002065{
2066 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002067 spin_lock_init(&n->list_lock);
2068 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002069#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002070 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002071 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002072 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002073#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002074}
2075
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002076#ifdef CONFIG_SMP
2077/*
2078 * Per cpu array for per cpu structures.
2079 *
2080 * The per cpu array places all kmem_cache_cpu structures from one processor
2081 * close together meaning that it becomes possible that multiple per cpu
2082 * structures are contained in one cacheline. This may be particularly
2083 * beneficial for the kmalloc caches.
2084 *
2085 * A desktop system typically has around 60-80 slabs. With 100 here we are
2086 * likely able to get per cpu structures for all caches from the array defined
2087 * here. We must be able to cover all kmalloc caches during bootstrap.
2088 *
2089 * If the per cpu array is exhausted then fall back to kmalloc
2090 * of individual cachelines. No sharing is possible then.
2091 */
2092#define NR_KMEM_CACHE_CPU 100
2093
2094static DEFINE_PER_CPU(struct kmem_cache_cpu,
2095 kmem_cache_cpu)[NR_KMEM_CACHE_CPU];
2096
2097static DEFINE_PER_CPU(struct kmem_cache_cpu *, kmem_cache_cpu_free);
Rusty Russell174596a2009-01-01 10:12:29 +10302098static DECLARE_BITMAP(kmem_cach_cpu_free_init_once, CONFIG_NR_CPUS);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002099
2100static struct kmem_cache_cpu *alloc_kmem_cache_cpu(struct kmem_cache *s,
2101 int cpu, gfp_t flags)
2102{
2103 struct kmem_cache_cpu *c = per_cpu(kmem_cache_cpu_free, cpu);
2104
2105 if (c)
2106 per_cpu(kmem_cache_cpu_free, cpu) =
2107 (void *)c->freelist;
2108 else {
2109 /* Table overflow: So allocate ourselves */
2110 c = kmalloc_node(
2111 ALIGN(sizeof(struct kmem_cache_cpu), cache_line_size()),
2112 flags, cpu_to_node(cpu));
2113 if (!c)
2114 return NULL;
2115 }
2116
2117 init_kmem_cache_cpu(s, c);
2118 return c;
2119}
2120
2121static void free_kmem_cache_cpu(struct kmem_cache_cpu *c, int cpu)
2122{
2123 if (c < per_cpu(kmem_cache_cpu, cpu) ||
David Rientjes37189092009-01-27 18:59:46 -08002124 c >= per_cpu(kmem_cache_cpu, cpu) + NR_KMEM_CACHE_CPU) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002125 kfree(c);
2126 return;
2127 }
2128 c->freelist = (void *)per_cpu(kmem_cache_cpu_free, cpu);
2129 per_cpu(kmem_cache_cpu_free, cpu) = c;
2130}
2131
2132static void free_kmem_cache_cpus(struct kmem_cache *s)
2133{
2134 int cpu;
2135
2136 for_each_online_cpu(cpu) {
2137 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2138
2139 if (c) {
2140 s->cpu_slab[cpu] = NULL;
2141 free_kmem_cache_cpu(c, cpu);
2142 }
2143 }
2144}
2145
2146static int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2147{
2148 int cpu;
2149
2150 for_each_online_cpu(cpu) {
2151 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
2152
2153 if (c)
2154 continue;
2155
2156 c = alloc_kmem_cache_cpu(s, cpu, flags);
2157 if (!c) {
2158 free_kmem_cache_cpus(s);
2159 return 0;
2160 }
2161 s->cpu_slab[cpu] = c;
2162 }
2163 return 1;
2164}
2165
2166/*
2167 * Initialize the per cpu array.
2168 */
2169static void init_alloc_cpu_cpu(int cpu)
2170{
2171 int i;
2172
Rusty Russell174596a2009-01-01 10:12:29 +10302173 if (cpumask_test_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once)))
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002174 return;
2175
2176 for (i = NR_KMEM_CACHE_CPU - 1; i >= 0; i--)
2177 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu, cpu)[i], cpu);
2178
Rusty Russell174596a2009-01-01 10:12:29 +10302179 cpumask_set_cpu(cpu, to_cpumask(kmem_cach_cpu_free_init_once));
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002180}
2181
2182static void __init init_alloc_cpu(void)
2183{
2184 int cpu;
2185
2186 for_each_online_cpu(cpu)
2187 init_alloc_cpu_cpu(cpu);
2188 }
2189
2190#else
2191static inline void free_kmem_cache_cpus(struct kmem_cache *s) {}
2192static inline void init_alloc_cpu(void) {}
2193
2194static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2195{
2196 init_kmem_cache_cpu(s, &s->cpu_slab);
2197 return 1;
2198}
2199#endif
2200
Christoph Lameter81819f02007-05-06 14:49:36 -07002201#ifdef CONFIG_NUMA
2202/*
2203 * No kmalloc_node yet so do it by hand. We know that this is the first
2204 * slab on the node for this slabcache. There are no concurrent accesses
2205 * possible.
2206 *
2207 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002208 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2209 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002210 */
David Rientjes0094de92008-11-25 19:14:19 -08002211static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002212{
2213 struct page *page;
2214 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002215 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002216
2217 BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2218
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002219 page = new_slab(kmalloc_caches, gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002220
2221 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002222 if (page_to_nid(page) != node) {
2223 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2224 "node %d\n", node);
2225 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2226 "in order to be able to continue\n");
2227 }
2228
Christoph Lameter81819f02007-05-06 14:49:36 -07002229 n = page->freelist;
2230 BUG_ON(!n);
2231 page->freelist = get_freepointer(kmalloc_caches, n);
2232 page->inuse++;
2233 kmalloc_caches->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002234#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterd45f39c2007-07-17 04:03:21 -07002235 init_object(kmalloc_caches, n, 1);
2236 init_tracking(kmalloc_caches, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002237#endif
Pekka Enberg5595cff2008-08-05 09:28:47 +03002238 init_kmem_cache_node(n, kmalloc_caches);
Christoph Lameter205ab992008-04-14 19:11:40 +03002239 inc_slabs_node(kmalloc_caches, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002240
rootba84c732008-01-07 23:20:28 -08002241 /*
2242 * lockdep requires consistent irq usage for each lock
2243 * so even though there cannot be a race this early in
2244 * the boot sequence, we still disable irqs.
2245 */
2246 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002247 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002248 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002249}
2250
2251static void free_kmem_cache_nodes(struct kmem_cache *s)
2252{
2253 int node;
2254
Christoph Lameterf64dc582007-10-16 01:25:33 -07002255 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002256 struct kmem_cache_node *n = s->node[node];
2257 if (n && n != &s->local_node)
2258 kmem_cache_free(kmalloc_caches, n);
2259 s->node[node] = NULL;
2260 }
2261}
2262
2263static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2264{
2265 int node;
2266 int local_node;
2267
2268 if (slab_state >= UP)
2269 local_node = page_to_nid(virt_to_page(s));
2270 else
2271 local_node = 0;
2272
Christoph Lameterf64dc582007-10-16 01:25:33 -07002273 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002274 struct kmem_cache_node *n;
2275
2276 if (local_node == node)
2277 n = &s->local_node;
2278 else {
2279 if (slab_state == DOWN) {
David Rientjes0094de92008-11-25 19:14:19 -08002280 early_kmem_cache_node_alloc(gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002281 continue;
2282 }
2283 n = kmem_cache_alloc_node(kmalloc_caches,
2284 gfpflags, node);
2285
2286 if (!n) {
2287 free_kmem_cache_nodes(s);
2288 return 0;
2289 }
2290
2291 }
2292 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002293 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002294 }
2295 return 1;
2296}
2297#else
2298static void free_kmem_cache_nodes(struct kmem_cache *s)
2299{
2300}
2301
2302static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2303{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002304 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002305 return 1;
2306}
2307#endif
2308
David Rientjesc0bdb232009-02-25 09:16:35 +02002309static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002310{
2311 if (min < MIN_PARTIAL)
2312 min = MIN_PARTIAL;
2313 else if (min > MAX_PARTIAL)
2314 min = MAX_PARTIAL;
2315 s->min_partial = min;
2316}
2317
Christoph Lameter81819f02007-05-06 14:49:36 -07002318/*
2319 * calculate_sizes() determines the order and the distribution of data within
2320 * a slab object.
2321 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002322static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002323{
2324 unsigned long flags = s->flags;
2325 unsigned long size = s->objsize;
2326 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002327 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002328
2329 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002330 * Round up object size to the next word boundary. We can only
2331 * place the free pointer at word boundaries and this determines
2332 * the possible location of the free pointer.
2333 */
2334 size = ALIGN(size, sizeof(void *));
2335
2336#ifdef CONFIG_SLUB_DEBUG
2337 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002338 * Determine if we can poison the object itself. If the user of
2339 * the slab may touch the object after free or before allocation
2340 * then we should never poison the object itself.
2341 */
2342 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002343 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002344 s->flags |= __OBJECT_POISON;
2345 else
2346 s->flags &= ~__OBJECT_POISON;
2347
Christoph Lameter81819f02007-05-06 14:49:36 -07002348
2349 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002350 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002351 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002352 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002353 */
2354 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2355 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002356#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002357
2358 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002359 * With that we have determined the number of bytes in actual use
2360 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002361 */
2362 s->inuse = size;
2363
2364 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002365 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002366 /*
2367 * Relocate free pointer after the object if it is not
2368 * permitted to overwrite the first word of the object on
2369 * kmem_cache_free.
2370 *
2371 * This is the case if we do RCU, have a constructor or
2372 * destructor or are poisoning the objects.
2373 */
2374 s->offset = size;
2375 size += sizeof(void *);
2376 }
2377
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002378#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002379 if (flags & SLAB_STORE_USER)
2380 /*
2381 * Need to store information about allocs and frees after
2382 * the object.
2383 */
2384 size += 2 * sizeof(struct track);
2385
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002386 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002387 /*
2388 * Add some empty padding so that we can catch
2389 * overwrites from earlier objects rather than let
2390 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002391 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002392 * of the object.
2393 */
2394 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002395#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002396
Christoph Lameter81819f02007-05-06 14:49:36 -07002397 /*
2398 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002399 * user specified and the dynamic determination of cache line size
2400 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002401 */
2402 align = calculate_alignment(flags, align, s->objsize);
2403
2404 /*
2405 * SLUB stores one object immediately after another beginning from
2406 * offset 0. In order to align the objects we have to simply size
2407 * each object to conform to the alignment.
2408 */
2409 size = ALIGN(size, align);
2410 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002411 if (forced_order >= 0)
2412 order = forced_order;
2413 else
2414 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002415
Christoph Lameter834f3d12008-04-14 19:11:31 +03002416 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002417 return 0;
2418
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002419 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002420 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002421 s->allocflags |= __GFP_COMP;
2422
2423 if (s->flags & SLAB_CACHE_DMA)
2424 s->allocflags |= SLUB_DMA;
2425
2426 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2427 s->allocflags |= __GFP_RECLAIMABLE;
2428
Christoph Lameter81819f02007-05-06 14:49:36 -07002429 /*
2430 * Determine the number of objects per slab
2431 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002432 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002433 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002434 if (oo_objects(s->oo) > oo_objects(s->max))
2435 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002436
Christoph Lameter834f3d12008-04-14 19:11:31 +03002437 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002438
2439}
2440
Christoph Lameter81819f02007-05-06 14:49:36 -07002441static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2442 const char *name, size_t size,
2443 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002444 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002445{
2446 memset(s, 0, kmem_size);
2447 s->name = name;
2448 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002449 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002450 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002451 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002452
Christoph Lameter06b285d2008-04-14 19:11:41 +03002453 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002454 goto error;
2455
David Rientjes3b89d7d2009-02-22 17:40:07 -08002456 /*
2457 * The larger the object size is, the more pages we want on the partial
2458 * list to avoid pounding the page allocator excessively.
2459 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002460 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002461 s->refcount = 1;
2462#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002463 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002464#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002465 if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2466 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002467
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002468 if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
Christoph Lameter81819f02007-05-06 14:49:36 -07002469 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002470 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002471error:
2472 if (flags & SLAB_PANIC)
2473 panic("Cannot create slab %s size=%lu realsize=%u "
2474 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002475 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002476 s->offset, flags);
2477 return 0;
2478}
Christoph Lameter81819f02007-05-06 14:49:36 -07002479
2480/*
2481 * Check if a given pointer is valid
2482 */
2483int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2484{
Pekka Enberg06428782008-01-07 23:20:27 -08002485 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002486
2487 page = get_object_page(object);
2488
2489 if (!page || s != page->slab)
2490 /* No slab or wrong slab */
2491 return 0;
2492
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002493 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002494 return 0;
2495
2496 /*
2497 * We could also check if the object is on the slabs freelist.
2498 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002499 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002500 * to a certain slab.
2501 */
2502 return 1;
2503}
2504EXPORT_SYMBOL(kmem_ptr_validate);
2505
2506/*
2507 * Determine the size of a slab object
2508 */
2509unsigned int kmem_cache_size(struct kmem_cache *s)
2510{
2511 return s->objsize;
2512}
2513EXPORT_SYMBOL(kmem_cache_size);
2514
2515const char *kmem_cache_name(struct kmem_cache *s)
2516{
2517 return s->name;
2518}
2519EXPORT_SYMBOL(kmem_cache_name);
2520
Christoph Lameter33b12c32008-04-25 12:22:43 -07002521static void list_slab_objects(struct kmem_cache *s, struct page *page,
2522 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002523{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002524#ifdef CONFIG_SLUB_DEBUG
2525 void *addr = page_address(page);
2526 void *p;
2527 DECLARE_BITMAP(map, page->objects);
2528
2529 bitmap_zero(map, page->objects);
2530 slab_err(s, page, "%s", text);
2531 slab_lock(page);
2532 for_each_free_object(p, s, page->freelist)
2533 set_bit(slab_index(p, s, addr), map);
2534
2535 for_each_object(p, s, addr, page->objects) {
2536
2537 if (!test_bit(slab_index(p, s, addr), map)) {
2538 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2539 p, p - addr);
2540 print_tracking(s, p);
2541 }
2542 }
2543 slab_unlock(page);
2544#endif
2545}
2546
Christoph Lameter81819f02007-05-06 14:49:36 -07002547/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002548 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002549 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002550static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002551{
Christoph Lameter81819f02007-05-06 14:49:36 -07002552 unsigned long flags;
2553 struct page *page, *h;
2554
2555 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002556 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002557 if (!page->inuse) {
2558 list_del(&page->lru);
2559 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002560 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002561 } else {
2562 list_slab_objects(s, page,
2563 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002564 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002565 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002566 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002567}
2568
2569/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002570 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002571 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002572static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002573{
2574 int node;
2575
2576 flush_all(s);
2577
2578 /* Attempt to free all objects */
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002579 free_kmem_cache_cpus(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002580 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002581 struct kmem_cache_node *n = get_node(s, node);
2582
Christoph Lameter599870b2008-04-23 12:36:52 -07002583 free_partial(s, n);
2584 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002585 return 1;
2586 }
2587 free_kmem_cache_nodes(s);
2588 return 0;
2589}
2590
2591/*
2592 * Close a cache and release the kmem_cache structure
2593 * (must be used for caches created using kmem_cache_create)
2594 */
2595void kmem_cache_destroy(struct kmem_cache *s)
2596{
2597 down_write(&slub_lock);
2598 s->refcount--;
2599 if (!s->refcount) {
2600 list_del(&s->list);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002601 up_write(&slub_lock);
Pekka Enbergd629d812008-04-23 22:31:08 +03002602 if (kmem_cache_close(s)) {
2603 printk(KERN_ERR "SLUB %s: %s called for cache that "
2604 "still has objects.\n", s->name, __func__);
2605 dump_stack();
2606 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002607 if (s->flags & SLAB_DESTROY_BY_RCU)
2608 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002609 sysfs_slab_remove(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07002610 } else
2611 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002612}
2613EXPORT_SYMBOL(kmem_cache_destroy);
2614
2615/********************************************************************
2616 * Kmalloc subsystem
2617 *******************************************************************/
2618
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002619struct kmem_cache kmalloc_caches[SLUB_PAGE_SHIFT] __cacheline_aligned;
Christoph Lameter81819f02007-05-06 14:49:36 -07002620EXPORT_SYMBOL(kmalloc_caches);
2621
Christoph Lameter81819f02007-05-06 14:49:36 -07002622static int __init setup_slub_min_order(char *str)
2623{
Pekka Enberg06428782008-01-07 23:20:27 -08002624 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002625
2626 return 1;
2627}
2628
2629__setup("slub_min_order=", setup_slub_min_order);
2630
2631static int __init setup_slub_max_order(char *str)
2632{
Pekka Enberg06428782008-01-07 23:20:27 -08002633 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002634 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002635
2636 return 1;
2637}
2638
2639__setup("slub_max_order=", setup_slub_max_order);
2640
2641static int __init setup_slub_min_objects(char *str)
2642{
Pekka Enberg06428782008-01-07 23:20:27 -08002643 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002644
2645 return 1;
2646}
2647
2648__setup("slub_min_objects=", setup_slub_min_objects);
2649
2650static int __init setup_slub_nomerge(char *str)
2651{
2652 slub_nomerge = 1;
2653 return 1;
2654}
2655
2656__setup("slub_nomerge", setup_slub_nomerge);
2657
Christoph Lameter81819f02007-05-06 14:49:36 -07002658static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2659 const char *name, int size, gfp_t gfp_flags)
2660{
2661 unsigned int flags = 0;
2662
2663 if (gfp_flags & SLUB_DMA)
2664 flags = SLAB_CACHE_DMA;
2665
Pekka Enberg83b519e2009-06-10 19:40:04 +03002666 /*
2667 * This function is called with IRQs disabled during early-boot on
2668 * single CPU so there's no need to take slub_lock here.
2669 */
Christoph Lameter81819f02007-05-06 14:49:36 -07002670 if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002671 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002672 goto panic;
2673
2674 list_add(&s->list, &slab_caches);
Pekka Enberg83b519e2009-06-10 19:40:04 +03002675
Christoph Lameter81819f02007-05-06 14:49:36 -07002676 if (sysfs_slab_add(s))
2677 goto panic;
2678 return s;
2679
2680panic:
2681 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2682}
2683
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002684#ifdef CONFIG_ZONE_DMA
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002685static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
Christoph Lameter1ceef402007-08-07 15:11:48 -07002686
2687static void sysfs_add_func(struct work_struct *w)
2688{
2689 struct kmem_cache *s;
2690
2691 down_write(&slub_lock);
2692 list_for_each_entry(s, &slab_caches, list) {
2693 if (s->flags & __SYSFS_ADD_DEFERRED) {
2694 s->flags &= ~__SYSFS_ADD_DEFERRED;
2695 sysfs_slab_add(s);
2696 }
2697 }
2698 up_write(&slub_lock);
2699}
2700
2701static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2702
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002703static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2704{
2705 struct kmem_cache *s;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002706 char *text;
2707 size_t realsize;
Nick Piggin964cf352009-06-15 13:35:10 +03002708 unsigned long slabflags;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002709
2710 s = kmalloc_caches_dma[index];
2711 if (s)
2712 return s;
2713
2714 /* Dynamically create dma cache */
Christoph Lameter1ceef402007-08-07 15:11:48 -07002715 if (flags & __GFP_WAIT)
2716 down_write(&slub_lock);
2717 else {
2718 if (!down_write_trylock(&slub_lock))
2719 goto out;
2720 }
2721
2722 if (kmalloc_caches_dma[index])
2723 goto unlock_out;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002724
Christoph Lameter7b55f622007-07-17 04:03:27 -07002725 realsize = kmalloc_caches[index].objsize;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08002726 text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2727 (unsigned int)realsize);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002728 s = kmalloc(kmem_size, flags & ~SLUB_DMA);
2729
Nick Piggin964cf352009-06-15 13:35:10 +03002730 /*
2731 * Must defer sysfs creation to a workqueue because we don't know
2732 * what context we are called from. Before sysfs comes up, we don't
2733 * need to do anything because our sysfs initcall will start by
2734 * adding all existing slabs to sysfs.
2735 */
Pekka Enberg5caf5c72009-06-17 08:30:54 +03002736 slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK;
Nick Piggin964cf352009-06-15 13:35:10 +03002737 if (slab_state >= SYSFS)
2738 slabflags |= __SYSFS_ADD_DEFERRED;
2739
Christoph Lameter1ceef402007-08-07 15:11:48 -07002740 if (!s || !text || !kmem_cache_open(s, flags, text,
Nick Piggin964cf352009-06-15 13:35:10 +03002741 realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) {
Christoph Lameter1ceef402007-08-07 15:11:48 -07002742 kfree(s);
2743 kfree(text);
2744 goto unlock_out;
Christoph Lameterdfce8642007-07-17 04:03:25 -07002745 }
Christoph Lameter1ceef402007-08-07 15:11:48 -07002746
2747 list_add(&s->list, &slab_caches);
2748 kmalloc_caches_dma[index] = s;
2749
Nick Piggin964cf352009-06-15 13:35:10 +03002750 if (slab_state >= SYSFS)
2751 schedule_work(&sysfs_add_work);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002752
2753unlock_out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002754 up_write(&slub_lock);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002755out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002756 return kmalloc_caches_dma[index];
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002757}
2758#endif
2759
Christoph Lameterf1b26332007-07-17 04:03:26 -07002760/*
2761 * Conversion table for small slabs sizes / 8 to the index in the
2762 * kmalloc array. This is necessary for slabs < 192 since we have non power
2763 * of two cache sizes there. The size of larger slabs can be determined using
2764 * fls.
2765 */
2766static s8 size_index[24] = {
2767 3, /* 8 */
2768 4, /* 16 */
2769 5, /* 24 */
2770 5, /* 32 */
2771 6, /* 40 */
2772 6, /* 48 */
2773 6, /* 56 */
2774 6, /* 64 */
2775 1, /* 72 */
2776 1, /* 80 */
2777 1, /* 88 */
2778 1, /* 96 */
2779 7, /* 104 */
2780 7, /* 112 */
2781 7, /* 120 */
2782 7, /* 128 */
2783 2, /* 136 */
2784 2, /* 144 */
2785 2, /* 152 */
2786 2, /* 160 */
2787 2, /* 168 */
2788 2, /* 176 */
2789 2, /* 184 */
2790 2 /* 192 */
2791};
2792
Christoph Lameter81819f02007-05-06 14:49:36 -07002793static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2794{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002795 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002796
Christoph Lameterf1b26332007-07-17 04:03:26 -07002797 if (size <= 192) {
2798 if (!size)
2799 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002800
Christoph Lameterf1b26332007-07-17 04:03:26 -07002801 index = size_index[(size - 1) / 8];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002802 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002803 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002804
2805#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002806 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002807 return dma_kmalloc_cache(index, flags);
Christoph Lameterf1b26332007-07-17 04:03:26 -07002808
Christoph Lameter81819f02007-05-06 14:49:36 -07002809#endif
2810 return &kmalloc_caches[index];
2811}
2812
2813void *__kmalloc(size_t size, gfp_t flags)
2814{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002815 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002816 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002817
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002818 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002819 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002820
2821 s = get_slab(size, flags);
2822
2823 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002824 return s;
2825
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002826 ret = slab_alloc(s, flags, -1, _RET_IP_);
2827
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002828 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002829
2830 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002831}
2832EXPORT_SYMBOL(__kmalloc);
2833
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002834static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2835{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002836 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002837 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002838
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002839 flags |= __GFP_COMP | __GFP_NOTRACK;
2840 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002841 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002842 ptr = page_address(page);
2843
2844 kmemleak_alloc(ptr, size, 1, flags);
2845 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002846}
2847
Christoph Lameter81819f02007-05-06 14:49:36 -07002848#ifdef CONFIG_NUMA
2849void *__kmalloc_node(size_t size, gfp_t flags, int node)
2850{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002851 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002852 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002853
Ingo Molnar057685c2009-02-20 12:15:30 +01002854 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002855 ret = kmalloc_large_node(size, flags, node);
2856
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002857 trace_kmalloc_node(_RET_IP_, ret,
2858 size, PAGE_SIZE << get_order(size),
2859 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002860
2861 return ret;
2862 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002863
2864 s = get_slab(size, flags);
2865
2866 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002867 return s;
2868
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002869 ret = slab_alloc(s, flags, node, _RET_IP_);
2870
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002871 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002872
2873 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002874}
2875EXPORT_SYMBOL(__kmalloc_node);
2876#endif
2877
2878size_t ksize(const void *object)
2879{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002880 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002881 struct kmem_cache *s;
2882
Christoph Lameteref8b4522007-10-16 01:24:46 -07002883 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002884 return 0;
2885
Vegard Nossum294a80a2007-12-04 23:45:30 -08002886 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002887
Pekka Enberg76994412008-05-22 19:22:25 +03002888 if (unlikely(!PageSlab(page))) {
2889 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002890 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002891 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002892 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002893
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002894#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002895 /*
2896 * Debugging requires use of the padding between object
2897 * and whatever may come after it.
2898 */
2899 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2900 return s->objsize;
2901
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002902#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002903 /*
2904 * If we have the need to store the freelist pointer
2905 * back there or track user information then we can
2906 * only use the space before that information.
2907 */
2908 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2909 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002910 /*
2911 * Else we can use all the padding etc for the allocation
2912 */
2913 return s->size;
2914}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002915EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002916
2917void kfree(const void *x)
2918{
Christoph Lameter81819f02007-05-06 14:49:36 -07002919 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002920 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002921
Pekka Enberg2121db72009-03-25 11:05:57 +02002922 trace_kfree(_RET_IP_, x);
2923
Satyam Sharma2408c552007-10-16 01:24:44 -07002924 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002925 return;
2926
Christoph Lameterb49af682007-05-06 14:49:41 -07002927 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002928 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002929 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002930 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002931 put_page(page);
2932 return;
2933 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002934 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002935}
2936EXPORT_SYMBOL(kfree);
2937
Christoph Lameter2086d262007-05-06 14:49:46 -07002938/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002939 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2940 * the remaining slabs by the number of items in use. The slabs with the
2941 * most items in use come first. New allocations will then fill those up
2942 * and thus they can be removed from the partial lists.
2943 *
2944 * The slabs with the least items are placed last. This results in them
2945 * being allocated from last increasing the chance that the last objects
2946 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002947 */
2948int kmem_cache_shrink(struct kmem_cache *s)
2949{
2950 int node;
2951 int i;
2952 struct kmem_cache_node *n;
2953 struct page *page;
2954 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002955 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002956 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002957 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002958 unsigned long flags;
2959
2960 if (!slabs_by_inuse)
2961 return -ENOMEM;
2962
2963 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002964 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002965 n = get_node(s, node);
2966
2967 if (!n->nr_partial)
2968 continue;
2969
Christoph Lameter834f3d12008-04-14 19:11:31 +03002970 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002971 INIT_LIST_HEAD(slabs_by_inuse + i);
2972
2973 spin_lock_irqsave(&n->list_lock, flags);
2974
2975 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002976 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002977 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002978 * Note that concurrent frees may occur while we hold the
2979 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002980 */
2981 list_for_each_entry_safe(page, t, &n->partial, lru) {
2982 if (!page->inuse && slab_trylock(page)) {
2983 /*
2984 * Must hold slab lock here because slab_free
2985 * may have freed the last object and be
2986 * waiting to release the slab.
2987 */
2988 list_del(&page->lru);
2989 n->nr_partial--;
2990 slab_unlock(page);
2991 discard_slab(s, page);
2992 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002993 list_move(&page->lru,
2994 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002995 }
2996 }
2997
Christoph Lameter2086d262007-05-06 14:49:46 -07002998 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002999 * Rebuild the partial list with the slabs filled up most
3000 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003001 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03003002 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003003 list_splice(slabs_by_inuse + i, n->partial.prev);
3004
Christoph Lameter2086d262007-05-06 14:49:46 -07003005 spin_unlock_irqrestore(&n->list_lock, flags);
3006 }
3007
3008 kfree(slabs_by_inuse);
3009 return 0;
3010}
3011EXPORT_SYMBOL(kmem_cache_shrink);
3012
Yasunori Gotob9049e22007-10-21 16:41:37 -07003013#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
3014static int slab_mem_going_offline_callback(void *arg)
3015{
3016 struct kmem_cache *s;
3017
3018 down_read(&slub_lock);
3019 list_for_each_entry(s, &slab_caches, list)
3020 kmem_cache_shrink(s);
3021 up_read(&slub_lock);
3022
3023 return 0;
3024}
3025
3026static void slab_mem_offline_callback(void *arg)
3027{
3028 struct kmem_cache_node *n;
3029 struct kmem_cache *s;
3030 struct memory_notify *marg = arg;
3031 int offline_node;
3032
3033 offline_node = marg->status_change_nid;
3034
3035 /*
3036 * If the node still has available memory. we need kmem_cache_node
3037 * for it yet.
3038 */
3039 if (offline_node < 0)
3040 return;
3041
3042 down_read(&slub_lock);
3043 list_for_each_entry(s, &slab_caches, list) {
3044 n = get_node(s, offline_node);
3045 if (n) {
3046 /*
3047 * if n->nr_slabs > 0, slabs still exist on the node
3048 * that is going down. We were unable to free them,
3049 * and offline_pages() function shoudn't call this
3050 * callback. So, we must fail.
3051 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003052 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003053
3054 s->node[offline_node] = NULL;
3055 kmem_cache_free(kmalloc_caches, n);
3056 }
3057 }
3058 up_read(&slub_lock);
3059}
3060
3061static int slab_mem_going_online_callback(void *arg)
3062{
3063 struct kmem_cache_node *n;
3064 struct kmem_cache *s;
3065 struct memory_notify *marg = arg;
3066 int nid = marg->status_change_nid;
3067 int ret = 0;
3068
3069 /*
3070 * If the node's memory is already available, then kmem_cache_node is
3071 * already created. Nothing to do.
3072 */
3073 if (nid < 0)
3074 return 0;
3075
3076 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003077 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003078 * allocate a kmem_cache_node structure in order to bring the node
3079 * online.
3080 */
3081 down_read(&slub_lock);
3082 list_for_each_entry(s, &slab_caches, list) {
3083 /*
3084 * XXX: kmem_cache_alloc_node will fallback to other nodes
3085 * since memory is not yet available from the node that
3086 * is brought up.
3087 */
3088 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
3089 if (!n) {
3090 ret = -ENOMEM;
3091 goto out;
3092 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03003093 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003094 s->node[nid] = n;
3095 }
3096out:
3097 up_read(&slub_lock);
3098 return ret;
3099}
3100
3101static int slab_memory_callback(struct notifier_block *self,
3102 unsigned long action, void *arg)
3103{
3104 int ret = 0;
3105
3106 switch (action) {
3107 case MEM_GOING_ONLINE:
3108 ret = slab_mem_going_online_callback(arg);
3109 break;
3110 case MEM_GOING_OFFLINE:
3111 ret = slab_mem_going_offline_callback(arg);
3112 break;
3113 case MEM_OFFLINE:
3114 case MEM_CANCEL_ONLINE:
3115 slab_mem_offline_callback(arg);
3116 break;
3117 case MEM_ONLINE:
3118 case MEM_CANCEL_OFFLINE:
3119 break;
3120 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003121 if (ret)
3122 ret = notifier_from_errno(ret);
3123 else
3124 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003125 return ret;
3126}
3127
3128#endif /* CONFIG_MEMORY_HOTPLUG */
3129
Christoph Lameter81819f02007-05-06 14:49:36 -07003130/********************************************************************
3131 * Basic setup of slabs
3132 *******************************************************************/
3133
3134void __init kmem_cache_init(void)
3135{
3136 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003137 int caches = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003138
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003139 init_alloc_cpu();
3140
Christoph Lameter81819f02007-05-06 14:49:36 -07003141#ifdef CONFIG_NUMA
3142 /*
3143 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003144 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003145 * kmem_cache_open for slab_state == DOWN.
3146 */
3147 create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
Pekka Enberg83b519e2009-06-10 19:40:04 +03003148 sizeof(struct kmem_cache_node), GFP_NOWAIT);
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003149 kmalloc_caches[0].refcount = -1;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003150 caches++;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003151
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003152 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003153#endif
3154
3155 /* Able to allocate the per node structures */
3156 slab_state = PARTIAL;
3157
3158 /* Caches that are not of the two-to-the-power-of size */
Christoph Lameter4b356be2007-06-16 10:16:13 -07003159 if (KMALLOC_MIN_SIZE <= 64) {
3160 create_kmalloc_cache(&kmalloc_caches[1],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003161 "kmalloc-96", 96, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003162 caches++;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003163 create_kmalloc_cache(&kmalloc_caches[2],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003164 "kmalloc-192", 192, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003165 caches++;
3166 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003167
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003168 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003169 create_kmalloc_cache(&kmalloc_caches[i],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003170 "kmalloc", 1 << i, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003171 caches++;
3172 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003173
Christoph Lameterf1b26332007-07-17 04:03:26 -07003174
3175 /*
3176 * Patch up the size_index table if we have strange large alignment
3177 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003178 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003179 *
3180 * Largest permitted alignment is 256 bytes due to the way we
3181 * handle the index determination for the smaller caches.
3182 *
3183 * Make sure that nothing crazy happens if someone starts tinkering
3184 * around with ARCH_KMALLOC_MINALIGN
3185 */
3186 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3187 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3188
Christoph Lameter12ad6842007-07-17 04:03:28 -07003189 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8)
Christoph Lameterf1b26332007-07-17 04:03:26 -07003190 size_index[(i - 1) / 8] = KMALLOC_SHIFT_LOW;
3191
Christoph Lameter41d54d32008-07-03 09:14:26 -05003192 if (KMALLOC_MIN_SIZE == 128) {
3193 /*
3194 * The 192 byte sized cache is not used if the alignment
3195 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3196 * instead.
3197 */
3198 for (i = 128 + 8; i <= 192; i += 8)
3199 size_index[(i - 1) / 8] = 8;
3200 }
3201
Christoph Lameter81819f02007-05-06 14:49:36 -07003202 slab_state = UP;
3203
3204 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003205 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++)
Christoph Lameter81819f02007-05-06 14:49:36 -07003206 kmalloc_caches[i]. name =
Pekka Enberg83b519e2009-06-10 19:40:04 +03003207 kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
Christoph Lameter81819f02007-05-06 14:49:36 -07003208
3209#ifdef CONFIG_SMP
3210 register_cpu_notifier(&slab_notifier);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003211 kmem_size = offsetof(struct kmem_cache, cpu_slab) +
3212 nr_cpu_ids * sizeof(struct kmem_cache_cpu *);
3213#else
3214 kmem_size = sizeof(struct kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003215#endif
3216
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003217 printk(KERN_INFO
3218 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003219 " CPUs=%d, Nodes=%d\n",
3220 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003221 slub_min_order, slub_max_order, slub_min_objects,
3222 nr_cpu_ids, nr_node_ids);
3223}
3224
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003225void __init kmem_cache_init_late(void)
3226{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003227}
3228
Christoph Lameter81819f02007-05-06 14:49:36 -07003229/*
3230 * Find a mergeable slab cache
3231 */
3232static int slab_unmergeable(struct kmem_cache *s)
3233{
3234 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3235 return 1;
3236
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003237 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003238 return 1;
3239
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003240 /*
3241 * We may have set a slab to be unmergeable during bootstrap.
3242 */
3243 if (s->refcount < 0)
3244 return 1;
3245
Christoph Lameter81819f02007-05-06 14:49:36 -07003246 return 0;
3247}
3248
3249static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003250 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003251 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003252{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003253 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003254
3255 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3256 return NULL;
3257
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003258 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003259 return NULL;
3260
3261 size = ALIGN(size, sizeof(void *));
3262 align = calculate_alignment(flags, align, size);
3263 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003264 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003265
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003266 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003267 if (slab_unmergeable(s))
3268 continue;
3269
3270 if (size > s->size)
3271 continue;
3272
Christoph Lameterba0268a2007-09-11 15:24:11 -07003273 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003274 continue;
3275 /*
3276 * Check if alignment is compatible.
3277 * Courtesy of Adrian Drzewiecki
3278 */
Pekka Enberg06428782008-01-07 23:20:27 -08003279 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003280 continue;
3281
3282 if (s->size - size >= sizeof(void *))
3283 continue;
3284
3285 return s;
3286 }
3287 return NULL;
3288}
3289
3290struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003291 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003292{
3293 struct kmem_cache *s;
3294
3295 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003296 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003297 if (s) {
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003298 int cpu;
3299
Christoph Lameter81819f02007-05-06 14:49:36 -07003300 s->refcount++;
3301 /*
3302 * Adjust the object sizes so that we clear
3303 * the complete object on kzalloc.
3304 */
3305 s->objsize = max(s->objsize, (int)size);
Christoph Lameter42a9fdb2007-10-16 01:26:09 -07003306
3307 /*
3308 * And then we need to update the object size in the
3309 * per cpu structures
3310 */
3311 for_each_online_cpu(cpu)
3312 get_cpu_slab(s, cpu)->objsize = s->objsize;
Christoph Lameter6446faa2008-02-15 23:45:26 -08003313
Christoph Lameter81819f02007-05-06 14:49:36 -07003314 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003315 up_write(&slub_lock);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003316
David Rientjes7b8f3b62008-12-17 22:09:46 -08003317 if (sysfs_slab_alias(s, name)) {
3318 down_write(&slub_lock);
3319 s->refcount--;
3320 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003321 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003322 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003323 return s;
3324 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003325
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003326 s = kmalloc(kmem_size, GFP_KERNEL);
3327 if (s) {
3328 if (kmem_cache_open(s, GFP_KERNEL, name,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003329 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003330 list_add(&s->list, &slab_caches);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003331 up_write(&slub_lock);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003332 if (sysfs_slab_add(s)) {
3333 down_write(&slub_lock);
3334 list_del(&s->list);
3335 up_write(&slub_lock);
3336 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003337 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003338 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003339 return s;
3340 }
3341 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003342 }
3343 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003344
3345err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003346 if (flags & SLAB_PANIC)
3347 panic("Cannot create slabcache %s\n", name);
3348 else
3349 s = NULL;
3350 return s;
3351}
3352EXPORT_SYMBOL(kmem_cache_create);
3353
Christoph Lameter81819f02007-05-06 14:49:36 -07003354#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003355/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003356 * Use the cpu notifier to insure that the cpu slabs are flushed when
3357 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003358 */
3359static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3360 unsigned long action, void *hcpu)
3361{
3362 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003363 struct kmem_cache *s;
3364 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003365
3366 switch (action) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003367 case CPU_UP_PREPARE:
3368 case CPU_UP_PREPARE_FROZEN:
3369 init_alloc_cpu_cpu(cpu);
3370 down_read(&slub_lock);
3371 list_for_each_entry(s, &slab_caches, list)
3372 s->cpu_slab[cpu] = alloc_kmem_cache_cpu(s, cpu,
3373 GFP_KERNEL);
3374 up_read(&slub_lock);
3375 break;
3376
Christoph Lameter81819f02007-05-06 14:49:36 -07003377 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003378 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003379 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003380 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003381 down_read(&slub_lock);
3382 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003383 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
3384
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003385 local_irq_save(flags);
3386 __flush_cpu_slab(s, cpu);
3387 local_irq_restore(flags);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003388 free_kmem_cache_cpu(c, cpu);
3389 s->cpu_slab[cpu] = NULL;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003390 }
3391 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003392 break;
3393 default:
3394 break;
3395 }
3396 return NOTIFY_OK;
3397}
3398
Pekka Enberg06428782008-01-07 23:20:27 -08003399static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003400 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003401};
Christoph Lameter81819f02007-05-06 14:49:36 -07003402
3403#endif
3404
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003405void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003406{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003407 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003408 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003409
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003410 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003411 return kmalloc_large(size, gfpflags);
3412
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003413 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003414
Satyam Sharma2408c552007-10-16 01:24:44 -07003415 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003416 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003417
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003418 ret = slab_alloc(s, gfpflags, -1, caller);
3419
3420 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003421 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003422
3423 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003424}
3425
3426void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003427 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003428{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003429 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003430 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003431
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003432 if (unlikely(size > SLUB_MAX_SIZE))
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003433 return kmalloc_large_node(size, gfpflags, node);
Pekka Enbergeada35e2008-02-11 22:47:46 +02003434
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003435 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003436
Satyam Sharma2408c552007-10-16 01:24:44 -07003437 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003438 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003439
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003440 ret = slab_alloc(s, gfpflags, node, caller);
3441
3442 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003443 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003444
3445 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003446}
3447
Christoph Lameterf6acb632008-04-29 16:16:06 -07003448#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003449static int count_inuse(struct page *page)
3450{
3451 return page->inuse;
3452}
3453
3454static int count_total(struct page *page)
3455{
3456 return page->objects;
3457}
3458
Christoph Lameter434e2452007-07-17 04:03:30 -07003459static int validate_slab(struct kmem_cache *s, struct page *page,
3460 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003461{
3462 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003463 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003464
3465 if (!check_slab(s, page) ||
3466 !on_freelist(s, page, NULL))
3467 return 0;
3468
3469 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003470 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003471
Christoph Lameter7656c722007-05-09 02:32:40 -07003472 for_each_free_object(p, s, page->freelist) {
3473 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003474 if (!check_object(s, page, p, 0))
3475 return 0;
3476 }
3477
Christoph Lameter224a88b2008-04-14 19:11:31 +03003478 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003479 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003480 if (!check_object(s, page, p, 1))
3481 return 0;
3482 return 1;
3483}
3484
Christoph Lameter434e2452007-07-17 04:03:30 -07003485static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3486 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003487{
3488 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003489 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003490 slab_unlock(page);
3491 } else
3492 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3493 s->name, page);
3494
3495 if (s->flags & DEBUG_DEFAULT_FLAGS) {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003496 if (!PageSlubDebug(page))
3497 printk(KERN_ERR "SLUB %s: SlubDebug not set "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003498 "on slab 0x%p\n", s->name, page);
3499 } else {
Andy Whitcroft8a380822008-07-23 21:27:18 -07003500 if (PageSlubDebug(page))
3501 printk(KERN_ERR "SLUB %s: SlubDebug set on "
Christoph Lameter53e15af2007-05-06 14:49:43 -07003502 "slab 0x%p\n", s->name, page);
3503 }
3504}
3505
Christoph Lameter434e2452007-07-17 04:03:30 -07003506static int validate_slab_node(struct kmem_cache *s,
3507 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003508{
3509 unsigned long count = 0;
3510 struct page *page;
3511 unsigned long flags;
3512
3513 spin_lock_irqsave(&n->list_lock, flags);
3514
3515 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003516 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003517 count++;
3518 }
3519 if (count != n->nr_partial)
3520 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3521 "counter=%ld\n", s->name, count, n->nr_partial);
3522
3523 if (!(s->flags & SLAB_STORE_USER))
3524 goto out;
3525
3526 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003527 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003528 count++;
3529 }
3530 if (count != atomic_long_read(&n->nr_slabs))
3531 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3532 "counter=%ld\n", s->name, count,
3533 atomic_long_read(&n->nr_slabs));
3534
3535out:
3536 spin_unlock_irqrestore(&n->list_lock, flags);
3537 return count;
3538}
3539
Christoph Lameter434e2452007-07-17 04:03:30 -07003540static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003541{
3542 int node;
3543 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003544 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003545 sizeof(unsigned long), GFP_KERNEL);
3546
3547 if (!map)
3548 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003549
3550 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003551 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003552 struct kmem_cache_node *n = get_node(s, node);
3553
Christoph Lameter434e2452007-07-17 04:03:30 -07003554 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003555 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003556 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003557 return count;
3558}
3559
Christoph Lameterb3459702007-05-09 02:32:41 -07003560#ifdef SLUB_RESILIENCY_TEST
3561static void resiliency_test(void)
3562{
3563 u8 *p;
3564
3565 printk(KERN_ERR "SLUB resiliency testing\n");
3566 printk(KERN_ERR "-----------------------\n");
3567 printk(KERN_ERR "A. Corruption after allocation\n");
3568
3569 p = kzalloc(16, GFP_KERNEL);
3570 p[16] = 0x12;
3571 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3572 " 0x12->0x%p\n\n", p + 16);
3573
3574 validate_slab_cache(kmalloc_caches + 4);
3575
3576 /* Hmmm... The next two are dangerous */
3577 p = kzalloc(32, GFP_KERNEL);
3578 p[32 + sizeof(void *)] = 0x34;
3579 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003580 " 0x34 -> -0x%p\n", p);
3581 printk(KERN_ERR
3582 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003583
3584 validate_slab_cache(kmalloc_caches + 5);
3585 p = kzalloc(64, GFP_KERNEL);
3586 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3587 *p = 0x56;
3588 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3589 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003590 printk(KERN_ERR
3591 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003592 validate_slab_cache(kmalloc_caches + 6);
3593
3594 printk(KERN_ERR "\nB. Corruption after free\n");
3595 p = kzalloc(128, GFP_KERNEL);
3596 kfree(p);
3597 *p = 0x78;
3598 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3599 validate_slab_cache(kmalloc_caches + 7);
3600
3601 p = kzalloc(256, GFP_KERNEL);
3602 kfree(p);
3603 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003604 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3605 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003606 validate_slab_cache(kmalloc_caches + 8);
3607
3608 p = kzalloc(512, GFP_KERNEL);
3609 kfree(p);
3610 p[512] = 0xab;
3611 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3612 validate_slab_cache(kmalloc_caches + 9);
3613}
3614#else
3615static void resiliency_test(void) {};
3616#endif
3617
Christoph Lameter88a420e2007-05-06 14:49:45 -07003618/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003619 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003620 * and freed.
3621 */
3622
3623struct location {
3624 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003625 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003626 long long sum_time;
3627 long min_time;
3628 long max_time;
3629 long min_pid;
3630 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303631 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003632 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003633};
3634
3635struct loc_track {
3636 unsigned long max;
3637 unsigned long count;
3638 struct location *loc;
3639};
3640
3641static void free_loc_track(struct loc_track *t)
3642{
3643 if (t->max)
3644 free_pages((unsigned long)t->loc,
3645 get_order(sizeof(struct location) * t->max));
3646}
3647
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003648static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003649{
3650 struct location *l;
3651 int order;
3652
Christoph Lameter88a420e2007-05-06 14:49:45 -07003653 order = get_order(sizeof(struct location) * max);
3654
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003655 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003656 if (!l)
3657 return 0;
3658
3659 if (t->count) {
3660 memcpy(l, t->loc, sizeof(struct location) * t->count);
3661 free_loc_track(t);
3662 }
3663 t->max = max;
3664 t->loc = l;
3665 return 1;
3666}
3667
3668static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003669 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003670{
3671 long start, end, pos;
3672 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003673 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003674 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003675
3676 start = -1;
3677 end = t->count;
3678
3679 for ( ; ; ) {
3680 pos = start + (end - start + 1) / 2;
3681
3682 /*
3683 * There is nothing at "end". If we end up there
3684 * we need to add something to before end.
3685 */
3686 if (pos == end)
3687 break;
3688
3689 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003690 if (track->addr == caddr) {
3691
3692 l = &t->loc[pos];
3693 l->count++;
3694 if (track->when) {
3695 l->sum_time += age;
3696 if (age < l->min_time)
3697 l->min_time = age;
3698 if (age > l->max_time)
3699 l->max_time = age;
3700
3701 if (track->pid < l->min_pid)
3702 l->min_pid = track->pid;
3703 if (track->pid > l->max_pid)
3704 l->max_pid = track->pid;
3705
Rusty Russell174596a2009-01-01 10:12:29 +10303706 cpumask_set_cpu(track->cpu,
3707 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003708 }
3709 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003710 return 1;
3711 }
3712
Christoph Lameter45edfa52007-05-09 02:32:45 -07003713 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003714 end = pos;
3715 else
3716 start = pos;
3717 }
3718
3719 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003720 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003721 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003722 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003723 return 0;
3724
3725 l = t->loc + pos;
3726 if (pos < t->count)
3727 memmove(l + 1, l,
3728 (t->count - pos) * sizeof(struct location));
3729 t->count++;
3730 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003731 l->addr = track->addr;
3732 l->sum_time = age;
3733 l->min_time = age;
3734 l->max_time = age;
3735 l->min_pid = track->pid;
3736 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303737 cpumask_clear(to_cpumask(l->cpus));
3738 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003739 nodes_clear(l->nodes);
3740 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003741 return 1;
3742}
3743
3744static void process_slab(struct loc_track *t, struct kmem_cache *s,
3745 struct page *page, enum track_item alloc)
3746{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003747 void *addr = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +03003748 DECLARE_BITMAP(map, page->objects);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003749 void *p;
3750
Christoph Lameter39b26462008-04-14 19:11:30 +03003751 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003752 for_each_free_object(p, s, page->freelist)
3753 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003754
Christoph Lameter224a88b2008-04-14 19:11:31 +03003755 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003756 if (!test_bit(slab_index(p, s, addr), map))
3757 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003758}
3759
3760static int list_locations(struct kmem_cache *s, char *buf,
3761 enum track_item alloc)
3762{
Harvey Harrisone374d482008-01-31 15:20:50 -08003763 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003764 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003765 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003766 int node;
3767
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003768 if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
Andrew Mortonea3061d2007-10-16 01:26:09 -07003769 GFP_TEMPORARY))
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003770 return sprintf(buf, "Out of memory\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003771
3772 /* Push back cpu slabs */
3773 flush_all(s);
3774
Christoph Lameterf64dc582007-10-16 01:25:33 -07003775 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003776 struct kmem_cache_node *n = get_node(s, node);
3777 unsigned long flags;
3778 struct page *page;
3779
Christoph Lameter9e869432007-08-22 14:01:56 -07003780 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003781 continue;
3782
3783 spin_lock_irqsave(&n->list_lock, flags);
3784 list_for_each_entry(page, &n->partial, lru)
3785 process_slab(&t, s, page, alloc);
3786 list_for_each_entry(page, &n->full, lru)
3787 process_slab(&t, s, page, alloc);
3788 spin_unlock_irqrestore(&n->list_lock, flags);
3789 }
3790
3791 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003792 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003793
Hugh Dickins9c246242008-12-09 13:14:27 -08003794 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003795 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003796 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003797
3798 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003799 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003800 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003801 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003802
3803 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003804 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003805 l->min_time,
3806 (long)div_u64(l->sum_time, l->count),
3807 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003808 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003809 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003810 l->min_time);
3811
3812 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003813 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003814 l->min_pid, l->max_pid);
3815 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003816 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003817 l->min_pid);
3818
Rusty Russell174596a2009-01-01 10:12:29 +10303819 if (num_online_cpus() > 1 &&
3820 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003821 len < PAGE_SIZE - 60) {
3822 len += sprintf(buf + len, " cpus=");
3823 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303824 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003825 }
3826
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003827 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003828 len < PAGE_SIZE - 60) {
3829 len += sprintf(buf + len, " nodes=");
3830 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003831 l->nodes);
3832 }
3833
Harvey Harrisone374d482008-01-31 15:20:50 -08003834 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003835 }
3836
3837 free_loc_track(&t);
3838 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003839 len += sprintf(buf, "No data\n");
3840 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003841}
3842
Christoph Lameter81819f02007-05-06 14:49:36 -07003843enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003844 SL_ALL, /* All slabs */
3845 SL_PARTIAL, /* Only partially allocated slabs */
3846 SL_CPU, /* Only slabs used for cpu caches */
3847 SL_OBJECTS, /* Determine allocated objects not slabs */
3848 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003849};
3850
Christoph Lameter205ab992008-04-14 19:11:40 +03003851#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003852#define SO_PARTIAL (1 << SL_PARTIAL)
3853#define SO_CPU (1 << SL_CPU)
3854#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003855#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003856
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003857static ssize_t show_slab_objects(struct kmem_cache *s,
3858 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003859{
3860 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003861 int node;
3862 int x;
3863 unsigned long *nodes;
3864 unsigned long *per_cpu;
3865
3866 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003867 if (!nodes)
3868 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003869 per_cpu = nodes + nr_node_ids;
3870
Christoph Lameter205ab992008-04-14 19:11:40 +03003871 if (flags & SO_CPU) {
3872 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003873
Christoph Lameter205ab992008-04-14 19:11:40 +03003874 for_each_possible_cpu(cpu) {
3875 struct kmem_cache_cpu *c = get_cpu_slab(s, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003876
Christoph Lameter205ab992008-04-14 19:11:40 +03003877 if (!c || c->node < 0)
3878 continue;
3879
3880 if (c->page) {
3881 if (flags & SO_TOTAL)
3882 x = c->page->objects;
3883 else if (flags & SO_OBJECTS)
3884 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003885 else
3886 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003887
Christoph Lameter81819f02007-05-06 14:49:36 -07003888 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003889 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003890 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003891 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003892 }
3893 }
3894
Christoph Lameter205ab992008-04-14 19:11:40 +03003895 if (flags & SO_ALL) {
3896 for_each_node_state(node, N_NORMAL_MEMORY) {
3897 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003898
Christoph Lameter205ab992008-04-14 19:11:40 +03003899 if (flags & SO_TOTAL)
3900 x = atomic_long_read(&n->total_objects);
3901 else if (flags & SO_OBJECTS)
3902 x = atomic_long_read(&n->total_objects) -
3903 count_partial(n, count_free);
3904
3905 else
3906 x = atomic_long_read(&n->nr_slabs);
3907 total += x;
3908 nodes[node] += x;
3909 }
3910
3911 } else if (flags & SO_PARTIAL) {
3912 for_each_node_state(node, N_NORMAL_MEMORY) {
3913 struct kmem_cache_node *n = get_node(s, node);
3914
3915 if (flags & SO_TOTAL)
3916 x = count_partial(n, count_total);
3917 else if (flags & SO_OBJECTS)
3918 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003919 else
3920 x = n->nr_partial;
3921 total += x;
3922 nodes[node] += x;
3923 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003924 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003925 x = sprintf(buf, "%lu", total);
3926#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003927 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003928 if (nodes[node])
3929 x += sprintf(buf + x, " N%d=%lu",
3930 node, nodes[node]);
3931#endif
3932 kfree(nodes);
3933 return x + sprintf(buf + x, "\n");
3934}
3935
3936static int any_slab_objects(struct kmem_cache *s)
3937{
3938 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003939
3940 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003941 struct kmem_cache_node *n = get_node(s, node);
3942
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003943 if (!n)
3944 continue;
3945
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003946 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003947 return 1;
3948 }
3949 return 0;
3950}
3951
3952#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3953#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3954
3955struct slab_attribute {
3956 struct attribute attr;
3957 ssize_t (*show)(struct kmem_cache *s, char *buf);
3958 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3959};
3960
3961#define SLAB_ATTR_RO(_name) \
3962 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3963
3964#define SLAB_ATTR(_name) \
3965 static struct slab_attribute _name##_attr = \
3966 __ATTR(_name, 0644, _name##_show, _name##_store)
3967
Christoph Lameter81819f02007-05-06 14:49:36 -07003968static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3969{
3970 return sprintf(buf, "%d\n", s->size);
3971}
3972SLAB_ATTR_RO(slab_size);
3973
3974static ssize_t align_show(struct kmem_cache *s, char *buf)
3975{
3976 return sprintf(buf, "%d\n", s->align);
3977}
3978SLAB_ATTR_RO(align);
3979
3980static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3981{
3982 return sprintf(buf, "%d\n", s->objsize);
3983}
3984SLAB_ATTR_RO(object_size);
3985
3986static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3987{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003988 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003989}
3990SLAB_ATTR_RO(objs_per_slab);
3991
Christoph Lameter06b285d2008-04-14 19:11:41 +03003992static ssize_t order_store(struct kmem_cache *s,
3993 const char *buf, size_t length)
3994{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003995 unsigned long order;
3996 int err;
3997
3998 err = strict_strtoul(buf, 10, &order);
3999 if (err)
4000 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004001
4002 if (order > slub_max_order || order < slub_min_order)
4003 return -EINVAL;
4004
4005 calculate_sizes(s, order);
4006 return length;
4007}
4008
Christoph Lameter81819f02007-05-06 14:49:36 -07004009static ssize_t order_show(struct kmem_cache *s, char *buf)
4010{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004011 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004012}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004013SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004014
David Rientjes73d342b2009-02-22 17:40:09 -08004015static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4016{
4017 return sprintf(buf, "%lu\n", s->min_partial);
4018}
4019
4020static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4021 size_t length)
4022{
4023 unsigned long min;
4024 int err;
4025
4026 err = strict_strtoul(buf, 10, &min);
4027 if (err)
4028 return err;
4029
David Rientjesc0bdb232009-02-25 09:16:35 +02004030 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004031 return length;
4032}
4033SLAB_ATTR(min_partial);
4034
Christoph Lameter81819f02007-05-06 14:49:36 -07004035static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4036{
4037 if (s->ctor) {
4038 int n = sprint_symbol(buf, (unsigned long)s->ctor);
4039
4040 return n + sprintf(buf + n, "\n");
4041 }
4042 return 0;
4043}
4044SLAB_ATTR_RO(ctor);
4045
Christoph Lameter81819f02007-05-06 14:49:36 -07004046static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4047{
4048 return sprintf(buf, "%d\n", s->refcount - 1);
4049}
4050SLAB_ATTR_RO(aliases);
4051
4052static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4053{
Christoph Lameter205ab992008-04-14 19:11:40 +03004054 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004055}
4056SLAB_ATTR_RO(slabs);
4057
4058static ssize_t partial_show(struct kmem_cache *s, char *buf)
4059{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004060 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004061}
4062SLAB_ATTR_RO(partial);
4063
4064static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4065{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004066 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004067}
4068SLAB_ATTR_RO(cpu_slabs);
4069
4070static ssize_t objects_show(struct kmem_cache *s, char *buf)
4071{
Christoph Lameter205ab992008-04-14 19:11:40 +03004072 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004073}
4074SLAB_ATTR_RO(objects);
4075
Christoph Lameter205ab992008-04-14 19:11:40 +03004076static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4077{
4078 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4079}
4080SLAB_ATTR_RO(objects_partial);
4081
4082static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4083{
4084 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4085}
4086SLAB_ATTR_RO(total_objects);
4087
Christoph Lameter81819f02007-05-06 14:49:36 -07004088static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4089{
4090 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4091}
4092
4093static ssize_t sanity_checks_store(struct kmem_cache *s,
4094 const char *buf, size_t length)
4095{
4096 s->flags &= ~SLAB_DEBUG_FREE;
4097 if (buf[0] == '1')
4098 s->flags |= SLAB_DEBUG_FREE;
4099 return length;
4100}
4101SLAB_ATTR(sanity_checks);
4102
4103static ssize_t trace_show(struct kmem_cache *s, char *buf)
4104{
4105 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4106}
4107
4108static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4109 size_t length)
4110{
4111 s->flags &= ~SLAB_TRACE;
4112 if (buf[0] == '1')
4113 s->flags |= SLAB_TRACE;
4114 return length;
4115}
4116SLAB_ATTR(trace);
4117
4118static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4119{
4120 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4121}
4122
4123static ssize_t reclaim_account_store(struct kmem_cache *s,
4124 const char *buf, size_t length)
4125{
4126 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4127 if (buf[0] == '1')
4128 s->flags |= SLAB_RECLAIM_ACCOUNT;
4129 return length;
4130}
4131SLAB_ATTR(reclaim_account);
4132
4133static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4134{
Christoph Lameter5af60832007-05-06 14:49:56 -07004135 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004136}
4137SLAB_ATTR_RO(hwcache_align);
4138
4139#ifdef CONFIG_ZONE_DMA
4140static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4141{
4142 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4143}
4144SLAB_ATTR_RO(cache_dma);
4145#endif
4146
4147static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4148{
4149 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4150}
4151SLAB_ATTR_RO(destroy_by_rcu);
4152
4153static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4154{
4155 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4156}
4157
4158static ssize_t red_zone_store(struct kmem_cache *s,
4159 const char *buf, size_t length)
4160{
4161 if (any_slab_objects(s))
4162 return -EBUSY;
4163
4164 s->flags &= ~SLAB_RED_ZONE;
4165 if (buf[0] == '1')
4166 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004167 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004168 return length;
4169}
4170SLAB_ATTR(red_zone);
4171
4172static ssize_t poison_show(struct kmem_cache *s, char *buf)
4173{
4174 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4175}
4176
4177static ssize_t poison_store(struct kmem_cache *s,
4178 const char *buf, size_t length)
4179{
4180 if (any_slab_objects(s))
4181 return -EBUSY;
4182
4183 s->flags &= ~SLAB_POISON;
4184 if (buf[0] == '1')
4185 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004186 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004187 return length;
4188}
4189SLAB_ATTR(poison);
4190
4191static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4192{
4193 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4194}
4195
4196static ssize_t store_user_store(struct kmem_cache *s,
4197 const char *buf, size_t length)
4198{
4199 if (any_slab_objects(s))
4200 return -EBUSY;
4201
4202 s->flags &= ~SLAB_STORE_USER;
4203 if (buf[0] == '1')
4204 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004205 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004206 return length;
4207}
4208SLAB_ATTR(store_user);
4209
Christoph Lameter53e15af2007-05-06 14:49:43 -07004210static ssize_t validate_show(struct kmem_cache *s, char *buf)
4211{
4212 return 0;
4213}
4214
4215static ssize_t validate_store(struct kmem_cache *s,
4216 const char *buf, size_t length)
4217{
Christoph Lameter434e2452007-07-17 04:03:30 -07004218 int ret = -EINVAL;
4219
4220 if (buf[0] == '1') {
4221 ret = validate_slab_cache(s);
4222 if (ret >= 0)
4223 ret = length;
4224 }
4225 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004226}
4227SLAB_ATTR(validate);
4228
Christoph Lameter2086d262007-05-06 14:49:46 -07004229static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4230{
4231 return 0;
4232}
4233
4234static ssize_t shrink_store(struct kmem_cache *s,
4235 const char *buf, size_t length)
4236{
4237 if (buf[0] == '1') {
4238 int rc = kmem_cache_shrink(s);
4239
4240 if (rc)
4241 return rc;
4242 } else
4243 return -EINVAL;
4244 return length;
4245}
4246SLAB_ATTR(shrink);
4247
Christoph Lameter88a420e2007-05-06 14:49:45 -07004248static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4249{
4250 if (!(s->flags & SLAB_STORE_USER))
4251 return -ENOSYS;
4252 return list_locations(s, buf, TRACK_ALLOC);
4253}
4254SLAB_ATTR_RO(alloc_calls);
4255
4256static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4257{
4258 if (!(s->flags & SLAB_STORE_USER))
4259 return -ENOSYS;
4260 return list_locations(s, buf, TRACK_FREE);
4261}
4262SLAB_ATTR_RO(free_calls);
4263
Christoph Lameter81819f02007-05-06 14:49:36 -07004264#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004265static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004266{
Christoph Lameter98246012008-01-07 23:20:26 -08004267 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004268}
4269
Christoph Lameter98246012008-01-07 23:20:26 -08004270static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004271 const char *buf, size_t length)
4272{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004273 unsigned long ratio;
4274 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004275
Christoph Lameter0121c6192008-04-29 16:11:12 -07004276 err = strict_strtoul(buf, 10, &ratio);
4277 if (err)
4278 return err;
4279
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004280 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004281 s->remote_node_defrag_ratio = ratio * 10;
4282
Christoph Lameter81819f02007-05-06 14:49:36 -07004283 return length;
4284}
Christoph Lameter98246012008-01-07 23:20:26 -08004285SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004286#endif
4287
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004288#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004289static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4290{
4291 unsigned long sum = 0;
4292 int cpu;
4293 int len;
4294 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4295
4296 if (!data)
4297 return -ENOMEM;
4298
4299 for_each_online_cpu(cpu) {
4300 unsigned x = get_cpu_slab(s, cpu)->stat[si];
4301
4302 data[cpu] = x;
4303 sum += x;
4304 }
4305
4306 len = sprintf(buf, "%lu", sum);
4307
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004308#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004309 for_each_online_cpu(cpu) {
4310 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004311 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004312 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004313#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004314 kfree(data);
4315 return len + sprintf(buf + len, "\n");
4316}
4317
4318#define STAT_ATTR(si, text) \
4319static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4320{ \
4321 return show_stat(s, buf, si); \
4322} \
4323SLAB_ATTR_RO(text); \
4324
4325STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4326STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4327STAT_ATTR(FREE_FASTPATH, free_fastpath);
4328STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4329STAT_ATTR(FREE_FROZEN, free_frozen);
4330STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4331STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4332STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4333STAT_ATTR(ALLOC_SLAB, alloc_slab);
4334STAT_ATTR(ALLOC_REFILL, alloc_refill);
4335STAT_ATTR(FREE_SLAB, free_slab);
4336STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4337STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4338STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4339STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4340STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4341STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004342STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004343#endif
4344
Pekka Enberg06428782008-01-07 23:20:27 -08004345static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004346 &slab_size_attr.attr,
4347 &object_size_attr.attr,
4348 &objs_per_slab_attr.attr,
4349 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004350 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004351 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004352 &objects_partial_attr.attr,
4353 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004354 &slabs_attr.attr,
4355 &partial_attr.attr,
4356 &cpu_slabs_attr.attr,
4357 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004358 &aliases_attr.attr,
4359 &align_attr.attr,
4360 &sanity_checks_attr.attr,
4361 &trace_attr.attr,
4362 &hwcache_align_attr.attr,
4363 &reclaim_account_attr.attr,
4364 &destroy_by_rcu_attr.attr,
4365 &red_zone_attr.attr,
4366 &poison_attr.attr,
4367 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004368 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004369 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004370 &alloc_calls_attr.attr,
4371 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004372#ifdef CONFIG_ZONE_DMA
4373 &cache_dma_attr.attr,
4374#endif
4375#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004376 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004377#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004378#ifdef CONFIG_SLUB_STATS
4379 &alloc_fastpath_attr.attr,
4380 &alloc_slowpath_attr.attr,
4381 &free_fastpath_attr.attr,
4382 &free_slowpath_attr.attr,
4383 &free_frozen_attr.attr,
4384 &free_add_partial_attr.attr,
4385 &free_remove_partial_attr.attr,
4386 &alloc_from_partial_attr.attr,
4387 &alloc_slab_attr.attr,
4388 &alloc_refill_attr.attr,
4389 &free_slab_attr.attr,
4390 &cpuslab_flush_attr.attr,
4391 &deactivate_full_attr.attr,
4392 &deactivate_empty_attr.attr,
4393 &deactivate_to_head_attr.attr,
4394 &deactivate_to_tail_attr.attr,
4395 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004396 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004397#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004398 NULL
4399};
4400
4401static struct attribute_group slab_attr_group = {
4402 .attrs = slab_attrs,
4403};
4404
4405static ssize_t slab_attr_show(struct kobject *kobj,
4406 struct attribute *attr,
4407 char *buf)
4408{
4409 struct slab_attribute *attribute;
4410 struct kmem_cache *s;
4411 int err;
4412
4413 attribute = to_slab_attr(attr);
4414 s = to_slab(kobj);
4415
4416 if (!attribute->show)
4417 return -EIO;
4418
4419 err = attribute->show(s, buf);
4420
4421 return err;
4422}
4423
4424static ssize_t slab_attr_store(struct kobject *kobj,
4425 struct attribute *attr,
4426 const char *buf, size_t len)
4427{
4428 struct slab_attribute *attribute;
4429 struct kmem_cache *s;
4430 int err;
4431
4432 attribute = to_slab_attr(attr);
4433 s = to_slab(kobj);
4434
4435 if (!attribute->store)
4436 return -EIO;
4437
4438 err = attribute->store(s, buf, len);
4439
4440 return err;
4441}
4442
Christoph Lameter151c6022008-01-07 22:29:05 -08004443static void kmem_cache_release(struct kobject *kobj)
4444{
4445 struct kmem_cache *s = to_slab(kobj);
4446
4447 kfree(s);
4448}
4449
Christoph Lameter81819f02007-05-06 14:49:36 -07004450static struct sysfs_ops slab_sysfs_ops = {
4451 .show = slab_attr_show,
4452 .store = slab_attr_store,
4453};
4454
4455static struct kobj_type slab_ktype = {
4456 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004457 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004458};
4459
4460static int uevent_filter(struct kset *kset, struct kobject *kobj)
4461{
4462 struct kobj_type *ktype = get_ktype(kobj);
4463
4464 if (ktype == &slab_ktype)
4465 return 1;
4466 return 0;
4467}
4468
4469static struct kset_uevent_ops slab_uevent_ops = {
4470 .filter = uevent_filter,
4471};
4472
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004473static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004474
4475#define ID_STR_LENGTH 64
4476
4477/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004478 *
4479 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004480 */
4481static char *create_unique_id(struct kmem_cache *s)
4482{
4483 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4484 char *p = name;
4485
4486 BUG_ON(!name);
4487
4488 *p++ = ':';
4489 /*
4490 * First flags affecting slabcache operations. We will only
4491 * get here for aliasable slabs so we do not need to support
4492 * too many flags. The flags here must cover all flags that
4493 * are matched during merging to guarantee that the id is
4494 * unique.
4495 */
4496 if (s->flags & SLAB_CACHE_DMA)
4497 *p++ = 'd';
4498 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4499 *p++ = 'a';
4500 if (s->flags & SLAB_DEBUG_FREE)
4501 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004502 if (!(s->flags & SLAB_NOTRACK))
4503 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004504 if (p != name + 1)
4505 *p++ = '-';
4506 p += sprintf(p, "%07d", s->size);
4507 BUG_ON(p > name + ID_STR_LENGTH - 1);
4508 return name;
4509}
4510
4511static int sysfs_slab_add(struct kmem_cache *s)
4512{
4513 int err;
4514 const char *name;
4515 int unmergeable;
4516
4517 if (slab_state < SYSFS)
4518 /* Defer until later */
4519 return 0;
4520
4521 unmergeable = slab_unmergeable(s);
4522 if (unmergeable) {
4523 /*
4524 * Slabcache can never be merged so we can use the name proper.
4525 * This is typically the case for debug situations. In that
4526 * case we can catch duplicate names easily.
4527 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004528 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004529 name = s->name;
4530 } else {
4531 /*
4532 * Create a unique name for the slab as a target
4533 * for the symlinks.
4534 */
4535 name = create_unique_id(s);
4536 }
4537
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004538 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004539 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4540 if (err) {
4541 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004542 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004543 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004544
4545 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4546 if (err)
4547 return err;
4548 kobject_uevent(&s->kobj, KOBJ_ADD);
4549 if (!unmergeable) {
4550 /* Setup first alias */
4551 sysfs_slab_alias(s, s->name);
4552 kfree(name);
4553 }
4554 return 0;
4555}
4556
4557static void sysfs_slab_remove(struct kmem_cache *s)
4558{
4559 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4560 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004561 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004562}
4563
4564/*
4565 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004566 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004567 */
4568struct saved_alias {
4569 struct kmem_cache *s;
4570 const char *name;
4571 struct saved_alias *next;
4572};
4573
Adrian Bunk5af328a2007-07-17 04:03:27 -07004574static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004575
4576static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4577{
4578 struct saved_alias *al;
4579
4580 if (slab_state == SYSFS) {
4581 /*
4582 * If we have a leftover link then remove it.
4583 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004584 sysfs_remove_link(&slab_kset->kobj, name);
4585 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004586 }
4587
4588 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4589 if (!al)
4590 return -ENOMEM;
4591
4592 al->s = s;
4593 al->name = name;
4594 al->next = alias_list;
4595 alias_list = al;
4596 return 0;
4597}
4598
4599static int __init slab_sysfs_init(void)
4600{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004601 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004602 int err;
4603
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004604 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004605 if (!slab_kset) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004606 printk(KERN_ERR "Cannot register slab subsystem.\n");
4607 return -ENOSYS;
4608 }
4609
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004610 slab_state = SYSFS;
4611
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004612 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004613 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004614 if (err)
4615 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4616 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004617 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004618
4619 while (alias_list) {
4620 struct saved_alias *al = alias_list;
4621
4622 alias_list = alias_list->next;
4623 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004624 if (err)
4625 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4626 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004627 kfree(al);
4628 }
4629
4630 resiliency_test();
4631 return 0;
4632}
4633
4634__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004635#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004636
4637/*
4638 * The /proc/slabinfo ABI
4639 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004640#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004641static void print_slabinfo_header(struct seq_file *m)
4642{
4643 seq_puts(m, "slabinfo - version: 2.1\n");
4644 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4645 "<objperslab> <pagesperslab>");
4646 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4647 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4648 seq_putc(m, '\n');
4649}
4650
4651static void *s_start(struct seq_file *m, loff_t *pos)
4652{
4653 loff_t n = *pos;
4654
4655 down_read(&slub_lock);
4656 if (!n)
4657 print_slabinfo_header(m);
4658
4659 return seq_list_start(&slab_caches, *pos);
4660}
4661
4662static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4663{
4664 return seq_list_next(p, &slab_caches, pos);
4665}
4666
4667static void s_stop(struct seq_file *m, void *p)
4668{
4669 up_read(&slub_lock);
4670}
4671
4672static int s_show(struct seq_file *m, void *p)
4673{
4674 unsigned long nr_partials = 0;
4675 unsigned long nr_slabs = 0;
4676 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004677 unsigned long nr_objs = 0;
4678 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004679 struct kmem_cache *s;
4680 int node;
4681
4682 s = list_entry(p, struct kmem_cache, list);
4683
4684 for_each_online_node(node) {
4685 struct kmem_cache_node *n = get_node(s, node);
4686
4687 if (!n)
4688 continue;
4689
4690 nr_partials += n->nr_partial;
4691 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004692 nr_objs += atomic_long_read(&n->total_objects);
4693 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004694 }
4695
Christoph Lameter205ab992008-04-14 19:11:40 +03004696 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004697
4698 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004699 nr_objs, s->size, oo_objects(s->oo),
4700 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004701 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4702 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4703 0UL);
4704 seq_putc(m, '\n');
4705 return 0;
4706}
4707
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004708static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004709 .start = s_start,
4710 .next = s_next,
4711 .stop = s_stop,
4712 .show = s_show,
4713};
4714
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004715static int slabinfo_open(struct inode *inode, struct file *file)
4716{
4717 return seq_open(file, &slabinfo_op);
4718}
4719
4720static const struct file_operations proc_slabinfo_operations = {
4721 .open = slabinfo_open,
4722 .read = seq_read,
4723 .llseek = seq_lseek,
4724 .release = seq_release,
4725};
4726
4727static int __init slab_proc_init(void)
4728{
4729 proc_create("slabinfo",S_IWUSR|S_IRUGO,NULL,&proc_slabinfo_operations);
4730 return 0;
4731}
4732module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004733#endif /* CONFIG_SLABINFO */