blob: e15aa7f193c9734518f3c3508210c70f294e4ad3 [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>
Vegard Nossum5a896d92008-04-04 00:54:48 +020020#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070021#include <linux/cpu.h>
22#include <linux/cpuset.h>
23#include <linux/mempolicy.h>
24#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070025#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070026#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070027#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070028#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090029#include <linux/fault-inject.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070030
Richard Kennedy4a923792010-10-21 10:29:19 +010031#include <trace/events/kmem.h>
32
Christoph Lameter81819f02007-05-06 14:49:36 -070033/*
34 * Lock order:
35 * 1. slab_lock(page)
36 * 2. slab->list_lock
37 *
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
44 *
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
50 *
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
55 * the list lock.
56 *
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
69 *
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
74 *
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
77 *
Christoph Lameter672bba32007-05-09 02:32:39 -070078 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070080 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070081 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070083 *
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
87 *
88 * Overloading of page flags that are otherwise used for LRU management.
89 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070090 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
98 *
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700102 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700105 *
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700108 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700109 */
110
Christoph Lameteraf537b02010-07-09 14:07:14 -0500111#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
113
114static inline int kmem_cache_debug(struct kmem_cache *s)
115{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700116#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500117 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700118#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500119 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500121}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700122
Christoph Lameter81819f02007-05-06 14:49:36 -0700123/*
124 * Issues still to be resolved:
125 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
127 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700128 * - Variable sizing of the per node arrays
129 */
130
131/* Enable to test recovery from slab corruption on boot */
132#undef SLUB_RESILIENCY_TEST
133
Christoph Lameter81819f02007-05-06 14:49:36 -0700134/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700135 * Mininum number of partial slabs. These will be left on the partial
136 * lists even if they are empty. kmem_cache_shrink may reclaim them.
137 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800138#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700139
Christoph Lameter2086d262007-05-06 14:49:46 -0700140/*
141 * Maximum number of desirable partial slabs.
142 * The existence of more partial slabs makes kmem_cache_shrink
143 * sort the partial list by the number of objects in the.
144 */
145#define MAX_PARTIAL 10
146
Christoph Lameter81819f02007-05-06 14:49:36 -0700147#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
148 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700149
Christoph Lameter81819f02007-05-06 14:49:36 -0700150/*
David Rientjes3de47212009-07-27 18:30:35 -0700151 * Debugging flags that require metadata to be stored in the slab. These get
152 * disabled when slub_debug=O is used and a cache's min order increases with
153 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700154 */
David Rientjes3de47212009-07-27 18:30:35 -0700155#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700156
157/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700158 * Set of flags that will prevent slab merging
159 */
160#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300161 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
162 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700163
164#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200165 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400167#define OO_SHIFT 16
168#define OO_MASK ((1 << OO_SHIFT) - 1)
169#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
170
Christoph Lameter81819f02007-05-06 14:49:36 -0700171/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500172#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameter81819f02007-05-06 14:49:36 -0700173
174static int kmem_size = sizeof(struct kmem_cache);
175
176#ifdef CONFIG_SMP
177static struct notifier_block slab_notifier;
178#endif
179
180static enum {
181 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500182 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700183 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700184 SYSFS /* Sysfs up */
185} slab_state = DOWN;
186
187/* A list of all slab caches on the system */
188static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700189static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700190
Christoph Lameter02cbc872007-05-09 02:32:43 -0700191/*
192 * Tracking user of a slab.
193 */
194struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300195 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700196 int cpu; /* Was running on cpu */
197 int pid; /* Pid context */
198 unsigned long when; /* When did the operation occur */
199};
200
201enum track_item { TRACK_ALLOC, TRACK_FREE };
202
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500203#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700204static int sysfs_slab_add(struct kmem_cache *);
205static int sysfs_slab_alias(struct kmem_cache *, const char *);
206static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800207
Christoph Lameter81819f02007-05-06 14:49:36 -0700208#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700209static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
210static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
211 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800212static inline void sysfs_slab_remove(struct kmem_cache *s)
213{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300214 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800215 kfree(s);
216}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800217
Christoph Lameter81819f02007-05-06 14:49:36 -0700218#endif
219
Christoph Lameter84e554e2009-12-18 16:26:23 -0600220static inline void stat(struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800221{
222#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600223 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800224#endif
225}
226
Christoph Lameter81819f02007-05-06 14:49:36 -0700227/********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
230
231int slab_is_available(void)
232{
233 return slab_state >= UP;
234}
235
236static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
237{
Christoph Lameter81819f02007-05-06 14:49:36 -0700238 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700239}
240
Christoph Lameter6446faa2008-02-15 23:45:26 -0800241/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700242static inline int check_valid_pointer(struct kmem_cache *s,
243 struct page *page, const void *object)
244{
245 void *base;
246
Christoph Lametera973e9d2008-03-01 13:40:44 -0800247 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700248 return 1;
249
Christoph Lametera973e9d2008-03-01 13:40:44 -0800250 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300251 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700252 (object - base) % s->size) {
253 return 0;
254 }
255
256 return 1;
257}
258
Christoph Lameter7656c722007-05-09 02:32:40 -0700259static inline void *get_freepointer(struct kmem_cache *s, void *object)
260{
261 return *(void **)(object + s->offset);
262}
263
264static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
265{
266 *(void **)(object + s->offset) = fp;
267}
268
269/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300270#define for_each_object(__p, __s, __addr, __objects) \
271 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700272 __p += (__s)->size)
273
274/* Scan freelist */
275#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800276 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700277
278/* Determine object index from a given position */
279static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
280{
281 return (p - addr) / s->size;
282}
283
Christoph Lameter834f3d12008-04-14 19:11:31 +0300284static inline struct kmem_cache_order_objects oo_make(int order,
285 unsigned long size)
286{
287 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400288 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300289 };
290
291 return x;
292}
293
294static inline int oo_order(struct kmem_cache_order_objects x)
295{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400296 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300297}
298
299static inline int oo_objects(struct kmem_cache_order_objects x)
300{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400301 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300302}
303
Christoph Lameter41ecc552007-05-09 02:32:44 -0700304#ifdef CONFIG_SLUB_DEBUG
305/*
306 * Debug settings:
307 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700308#ifdef CONFIG_SLUB_DEBUG_ON
309static int slub_debug = DEBUG_DEFAULT_FLAGS;
310#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700311static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700312#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700313
314static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700315static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700316
Christoph Lameter7656c722007-05-09 02:32:40 -0700317/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700318 * Object debugging
319 */
320static void print_section(char *text, u8 *addr, unsigned int length)
321{
322 int i, offset;
323 int newline = 1;
324 char ascii[17];
325
326 ascii[16] = 0;
327
328 for (i = 0; i < length; i++) {
329 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700330 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700331 newline = 0;
332 }
Pekka Enberg06428782008-01-07 23:20:27 -0800333 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700334 offset = i % 16;
335 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
336 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800337 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700338 newline = 1;
339 }
340 }
341 if (!newline) {
342 i %= 16;
343 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800344 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700345 ascii[i] = ' ';
346 i++;
347 }
Pekka Enberg06428782008-01-07 23:20:27 -0800348 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700349 }
350}
351
Christoph Lameter81819f02007-05-06 14:49:36 -0700352static struct track *get_track(struct kmem_cache *s, void *object,
353 enum track_item alloc)
354{
355 struct track *p;
356
357 if (s->offset)
358 p = object + s->offset + sizeof(void *);
359 else
360 p = object + s->inuse;
361
362 return p + alloc;
363}
364
365static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300366 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700367{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900368 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700369
Christoph Lameter81819f02007-05-06 14:49:36 -0700370 if (addr) {
371 p->addr = addr;
372 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400373 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700374 p->when = jiffies;
375 } else
376 memset(p, 0, sizeof(struct track));
377}
378
Christoph Lameter81819f02007-05-06 14:49:36 -0700379static void init_tracking(struct kmem_cache *s, void *object)
380{
Christoph Lameter24922682007-07-17 04:03:18 -0700381 if (!(s->flags & SLAB_STORE_USER))
382 return;
383
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300384 set_track(s, object, TRACK_FREE, 0UL);
385 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700386}
387
388static void print_track(const char *s, struct track *t)
389{
390 if (!t->addr)
391 return;
392
Linus Torvalds7daf7052008-07-14 12:12:53 -0700393 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300394 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700395}
396
Christoph Lameter24922682007-07-17 04:03:18 -0700397static void print_tracking(struct kmem_cache *s, void *object)
398{
399 if (!(s->flags & SLAB_STORE_USER))
400 return;
401
402 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
403 print_track("Freed", get_track(s, object, TRACK_FREE));
404}
405
406static void print_page_info(struct page *page)
407{
Christoph Lameter39b26462008-04-14 19:11:30 +0300408 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
409 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700410
411}
412
413static void slab_bug(struct kmem_cache *s, char *fmt, ...)
414{
415 va_list args;
416 char buf[100];
417
418 va_start(args, fmt);
419 vsnprintf(buf, sizeof(buf), fmt, args);
420 va_end(args);
421 printk(KERN_ERR "========================================"
422 "=====================================\n");
423 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
424 printk(KERN_ERR "----------------------------------------"
425 "-------------------------------------\n\n");
426}
427
428static void slab_fix(struct kmem_cache *s, char *fmt, ...)
429{
430 va_list args;
431 char buf[100];
432
433 va_start(args, fmt);
434 vsnprintf(buf, sizeof(buf), fmt, args);
435 va_end(args);
436 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
437}
438
439static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700440{
441 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800442 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700443
444 print_tracking(s, p);
445
446 print_page_info(page);
447
448 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
449 p, p - addr, get_freepointer(s, p));
450
451 if (p > addr + 16)
452 print_section("Bytes b4", p - 16, 16);
453
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300454 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700455
456 if (s->flags & SLAB_RED_ZONE)
457 print_section("Redzone", p + s->objsize,
458 s->inuse - s->objsize);
459
Christoph Lameter81819f02007-05-06 14:49:36 -0700460 if (s->offset)
461 off = s->offset + sizeof(void *);
462 else
463 off = s->inuse;
464
Christoph Lameter24922682007-07-17 04:03:18 -0700465 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700466 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700467
468 if (off != s->size)
469 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700470 print_section("Padding", p + off, s->size - off);
471
472 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700473}
474
475static void object_err(struct kmem_cache *s, struct page *page,
476 u8 *object, char *reason)
477{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700478 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700479 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700480}
481
Christoph Lameter24922682007-07-17 04:03:18 -0700482static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700483{
484 va_list args;
485 char buf[100];
486
Christoph Lameter24922682007-07-17 04:03:18 -0700487 va_start(args, fmt);
488 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700489 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700490 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700491 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700492 dump_stack();
493}
494
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500495static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700496{
497 u8 *p = object;
498
499 if (s->flags & __OBJECT_POISON) {
500 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800501 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700502 }
503
504 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500505 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700506}
507
Christoph Lameter24922682007-07-17 04:03:18 -0700508static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700509{
510 while (bytes) {
511 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700512 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700513 start++;
514 bytes--;
515 }
Christoph Lameter24922682007-07-17 04:03:18 -0700516 return NULL;
517}
518
519static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
520 void *from, void *to)
521{
522 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
523 memset(from, data, to - from);
524}
525
526static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
527 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800528 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700529{
530 u8 *fault;
531 u8 *end;
532
533 fault = check_bytes(start, value, bytes);
534 if (!fault)
535 return 1;
536
537 end = start + bytes;
538 while (end > fault && end[-1] == value)
539 end--;
540
541 slab_bug(s, "%s overwritten", what);
542 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
543 fault, end - 1, fault[0], value);
544 print_trailer(s, page, object);
545
546 restore_bytes(s, what, value, fault, end);
547 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700548}
549
Christoph Lameter81819f02007-05-06 14:49:36 -0700550/*
551 * Object layout:
552 *
553 * object address
554 * Bytes of the object to be managed.
555 * If the freepointer may overlay the object then the free
556 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700557 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700558 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
559 * 0xa5 (POISON_END)
560 *
561 * object + s->objsize
562 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700563 * Padding is extended by another word if Redzoning is enabled and
564 * objsize == inuse.
565 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700566 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
567 * 0xcc (RED_ACTIVE) for objects in use.
568 *
569 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700570 * Meta data starts here.
571 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700572 * A. Free pointer (if we cannot overwrite object on free)
573 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700574 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800575 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700576 * before the word boundary.
577 *
578 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700579 *
580 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700581 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700582 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700583 * If slabcaches are merged then the objsize and inuse boundaries are mostly
584 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700585 * may be used with merged slabcaches.
586 */
587
Christoph Lameter81819f02007-05-06 14:49:36 -0700588static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
589{
590 unsigned long off = s->inuse; /* The end of info */
591
592 if (s->offset)
593 /* Freepointer is placed after the object. */
594 off += sizeof(void *);
595
596 if (s->flags & SLAB_STORE_USER)
597 /* We also have user information there */
598 off += 2 * sizeof(struct track);
599
600 if (s->size == off)
601 return 1;
602
Christoph Lameter24922682007-07-17 04:03:18 -0700603 return check_bytes_and_report(s, page, p, "Object padding",
604 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700605}
606
Christoph Lameter39b26462008-04-14 19:11:30 +0300607/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700608static int slab_pad_check(struct kmem_cache *s, struct page *page)
609{
Christoph Lameter24922682007-07-17 04:03:18 -0700610 u8 *start;
611 u8 *fault;
612 u8 *end;
613 int length;
614 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700615
616 if (!(s->flags & SLAB_POISON))
617 return 1;
618
Christoph Lametera973e9d2008-03-01 13:40:44 -0800619 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300620 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300621 end = start + length;
622 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700623 if (!remainder)
624 return 1;
625
Christoph Lameter39b26462008-04-14 19:11:30 +0300626 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700627 if (!fault)
628 return 1;
629 while (end > fault && end[-1] == POISON_INUSE)
630 end--;
631
632 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300633 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700634
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200635 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700636 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700637}
638
639static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500640 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700641{
642 u8 *p = object;
643 u8 *endobject = object + s->objsize;
644
645 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700646 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500647 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700648 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700649 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800650 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
651 check_bytes_and_report(s, page, p, "Alignment padding",
652 endobject, POISON_INUSE, s->inuse - s->objsize);
653 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700654 }
655
656 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500657 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700658 (!check_bytes_and_report(s, page, p, "Poison", p,
659 POISON_FREE, s->objsize - 1) ||
660 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800661 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700662 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700663 /*
664 * check_pad_bytes cleans up on its own.
665 */
666 check_pad_bytes(s, page, p);
667 }
668
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500669 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700670 /*
671 * Object and freepointer overlap. Cannot check
672 * freepointer while object is allocated.
673 */
674 return 1;
675
676 /* Check free pointer validity */
677 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
678 object_err(s, page, p, "Freepointer corrupt");
679 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100680 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700681 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700682 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700683 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800684 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700685 return 0;
686 }
687 return 1;
688}
689
690static int check_slab(struct kmem_cache *s, struct page *page)
691{
Christoph Lameter39b26462008-04-14 19:11:30 +0300692 int maxobj;
693
Christoph Lameter81819f02007-05-06 14:49:36 -0700694 VM_BUG_ON(!irqs_disabled());
695
696 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700697 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700698 return 0;
699 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300700
701 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
702 if (page->objects > maxobj) {
703 slab_err(s, page, "objects %u > max %u",
704 s->name, page->objects, maxobj);
705 return 0;
706 }
707 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700708 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300709 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700710 return 0;
711 }
712 /* Slab_pad_check fixes things up after itself */
713 slab_pad_check(s, page);
714 return 1;
715}
716
717/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700718 * Determine if a certain object on a page is on the freelist. Must hold the
719 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700720 */
721static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
722{
723 int nr = 0;
724 void *fp = page->freelist;
725 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300726 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700727
Christoph Lameter39b26462008-04-14 19:11:30 +0300728 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700729 if (fp == search)
730 return 1;
731 if (!check_valid_pointer(s, page, fp)) {
732 if (object) {
733 object_err(s, page, object,
734 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800735 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700736 break;
737 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700738 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800739 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300740 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700741 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700742 return 0;
743 }
744 break;
745 }
746 object = fp;
747 fp = get_freepointer(s, object);
748 nr++;
749 }
750
Christoph Lameter224a88b2008-04-14 19:11:31 +0300751 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400752 if (max_objects > MAX_OBJS_PER_PAGE)
753 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300754
755 if (page->objects != max_objects) {
756 slab_err(s, page, "Wrong number of objects. Found %d but "
757 "should be %d", page->objects, max_objects);
758 page->objects = max_objects;
759 slab_fix(s, "Number of objects adjusted.");
760 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300761 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700762 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300763 "counted were %d", page->inuse, page->objects - nr);
764 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700765 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700766 }
767 return search == NULL;
768}
769
Christoph Lameter0121c6192008-04-29 16:11:12 -0700770static void trace(struct kmem_cache *s, struct page *page, void *object,
771 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700772{
773 if (s->flags & SLAB_TRACE) {
774 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
775 s->name,
776 alloc ? "alloc" : "free",
777 object, page->inuse,
778 page->freelist);
779
780 if (!alloc)
781 print_section("Object", (void *)object, s->objsize);
782
783 dump_stack();
784 }
785}
786
Christoph Lameter643b1132007-05-06 14:49:42 -0700787/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500788 * Hooks for other subsystems that check memory allocations. In a typical
789 * production configuration these hooks all should produce no code at all.
790 */
791static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
792{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500793 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500794 lockdep_trace_alloc(flags);
795 might_sleep_if(flags & __GFP_WAIT);
796
797 return should_failslab(s->objsize, flags, s->flags);
798}
799
800static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
801{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500802 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500803 kmemcheck_slab_alloc(s, flags, object, s->objsize);
804 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
805}
806
807static inline void slab_free_hook(struct kmem_cache *s, void *x)
808{
809 kmemleak_free_recursive(x, s->flags);
810}
811
812static inline void slab_free_hook_irq(struct kmem_cache *s, void *object)
813{
814 kmemcheck_slab_free(s, object, s->objsize);
815 debug_check_no_locks_freed(object, s->objsize);
816 if (!(s->flags & SLAB_DEBUG_OBJECTS))
817 debug_check_no_obj_freed(object, s->objsize);
818}
819
820/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700821 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700822 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700823static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700824{
Christoph Lameter643b1132007-05-06 14:49:42 -0700825 spin_lock(&n->list_lock);
826 list_add(&page->lru, &n->full);
827 spin_unlock(&n->list_lock);
828}
829
830static void remove_full(struct kmem_cache *s, struct page *page)
831{
832 struct kmem_cache_node *n;
833
834 if (!(s->flags & SLAB_STORE_USER))
835 return;
836
837 n = get_node(s, page_to_nid(page));
838
839 spin_lock(&n->list_lock);
840 list_del(&page->lru);
841 spin_unlock(&n->list_lock);
842}
843
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300844/* Tracking of the number of slabs for debugging purposes */
845static inline unsigned long slabs_node(struct kmem_cache *s, int node)
846{
847 struct kmem_cache_node *n = get_node(s, node);
848
849 return atomic_long_read(&n->nr_slabs);
850}
851
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400852static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
853{
854 return atomic_long_read(&n->nr_slabs);
855}
856
Christoph Lameter205ab992008-04-14 19:11:40 +0300857static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300858{
859 struct kmem_cache_node *n = get_node(s, node);
860
861 /*
862 * May be called early in order to allocate a slab for the
863 * kmem_cache_node structure. Solve the chicken-egg
864 * dilemma by deferring the increment of the count during
865 * bootstrap (see early_kmem_cache_node_alloc).
866 */
Christoph Lameter7340cc82010-09-28 08:10:26 -0500867 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300868 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300869 atomic_long_add(objects, &n->total_objects);
870 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300871}
Christoph Lameter205ab992008-04-14 19:11:40 +0300872static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300873{
874 struct kmem_cache_node *n = get_node(s, node);
875
876 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300877 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300878}
879
880/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700881static void setup_object_debug(struct kmem_cache *s, struct page *page,
882 void *object)
883{
884 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
885 return;
886
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500887 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700888 init_tracking(s, object);
889}
890
Christoph Lameter15370662010-08-20 12:37:12 -0500891static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300892 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700893{
894 if (!check_slab(s, page))
895 goto bad;
896
Christoph Lameterd692ef62008-02-15 23:45:24 -0800897 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700898 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700899 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700900 }
901
902 if (!check_valid_pointer(s, page, object)) {
903 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700904 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700905 }
906
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500907 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -0700908 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700909
Christoph Lameter3ec09742007-05-16 22:11:00 -0700910 /* Success perform special debug activities for allocs */
911 if (s->flags & SLAB_STORE_USER)
912 set_track(s, object, TRACK_ALLOC, addr);
913 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500914 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700915 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700916
Christoph Lameter81819f02007-05-06 14:49:36 -0700917bad:
918 if (PageSlab(page)) {
919 /*
920 * If this is a slab page then lets do the best we can
921 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700922 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700923 */
Christoph Lameter24922682007-07-17 04:03:18 -0700924 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300925 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800926 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700927 }
928 return 0;
929}
930
Christoph Lameter15370662010-08-20 12:37:12 -0500931static noinline int free_debug_processing(struct kmem_cache *s,
932 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700933{
934 if (!check_slab(s, page))
935 goto fail;
936
937 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700938 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700939 goto fail;
940 }
941
942 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700943 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700944 goto fail;
945 }
946
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500947 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -0700948 return 0;
949
950 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800951 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700952 slab_err(s, page, "Attempt to free object(0x%p) "
953 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800954 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700955 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700956 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700957 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700958 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800959 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700960 object_err(s, page, object,
961 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700962 goto fail;
963 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700964
965 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700966 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700967 remove_full(s, page);
968 if (s->flags & SLAB_STORE_USER)
969 set_track(s, object, TRACK_FREE, addr);
970 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500971 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700972 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700973
Christoph Lameter81819f02007-05-06 14:49:36 -0700974fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700975 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700976 return 0;
977}
978
Christoph Lameter41ecc552007-05-09 02:32:44 -0700979static int __init setup_slub_debug(char *str)
980{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700981 slub_debug = DEBUG_DEFAULT_FLAGS;
982 if (*str++ != '=' || !*str)
983 /*
984 * No options specified. Switch on full debugging.
985 */
986 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700987
988 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700989 /*
990 * No options but restriction on slabs. This means full
991 * debugging for slabs matching a pattern.
992 */
993 goto check_slabs;
994
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700995 if (tolower(*str) == 'o') {
996 /*
997 * Avoid enabling debugging on caches if its minimum order
998 * would increase as a result.
999 */
1000 disable_higher_order_debug = 1;
1001 goto out;
1002 }
1003
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001004 slub_debug = 0;
1005 if (*str == '-')
1006 /*
1007 * Switch off all debugging measures.
1008 */
1009 goto out;
1010
1011 /*
1012 * Determine which debug features should be switched on
1013 */
Pekka Enberg06428782008-01-07 23:20:27 -08001014 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001015 switch (tolower(*str)) {
1016 case 'f':
1017 slub_debug |= SLAB_DEBUG_FREE;
1018 break;
1019 case 'z':
1020 slub_debug |= SLAB_RED_ZONE;
1021 break;
1022 case 'p':
1023 slub_debug |= SLAB_POISON;
1024 break;
1025 case 'u':
1026 slub_debug |= SLAB_STORE_USER;
1027 break;
1028 case 't':
1029 slub_debug |= SLAB_TRACE;
1030 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001031 case 'a':
1032 slub_debug |= SLAB_FAILSLAB;
1033 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001034 default:
1035 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001036 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001037 }
1038 }
1039
1040check_slabs:
1041 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001042 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001043out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001044 return 1;
1045}
1046
1047__setup("slub_debug", setup_slub_debug);
1048
Christoph Lameterba0268a2007-09-11 15:24:11 -07001049static 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 Lameter41ecc552007-05-09 02:32:44 -07001052{
1053 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001054 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001055 */
Christoph Lametere1533622008-02-15 23:45:24 -08001056 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001057 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1058 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001059
1060 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001061}
1062#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001063static inline void setup_object_debug(struct kmem_cache *s,
1064 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001065
Christoph Lameter3ec09742007-05-16 22:11:00 -07001066static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001067 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001068
Christoph Lameter3ec09742007-05-16 22:11:00 -07001069static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001070 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001071
Christoph Lameter41ecc552007-05-09 02:32:44 -07001072static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1073 { return 1; }
1074static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001075 void *object, u8 val) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001076static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001077static inline unsigned long kmem_cache_flags(unsigned long objsize,
1078 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001079 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001080{
1081 return flags;
1082}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001083#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001084
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001085#define disable_higher_order_debug 0
1086
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001087static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1088 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001089static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1090 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001091static inline void inc_slabs_node(struct kmem_cache *s, int node,
1092 int objects) {}
1093static inline void dec_slabs_node(struct kmem_cache *s, int node,
1094 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001095
1096static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1097 { return 0; }
1098
1099static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1100 void *object) {}
1101
1102static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1103
1104static inline void slab_free_hook_irq(struct kmem_cache *s,
1105 void *object) {}
1106
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001107#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001108
Christoph Lameter81819f02007-05-06 14:49:36 -07001109/*
1110 * Slab allocation and freeing
1111 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001112static inline struct page *alloc_slab_page(gfp_t flags, int node,
1113 struct kmem_cache_order_objects oo)
1114{
1115 int order = oo_order(oo);
1116
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001117 flags |= __GFP_NOTRACK;
1118
Christoph Lameter2154a332010-07-09 14:07:10 -05001119 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001120 return alloc_pages(flags, order);
1121 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001122 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001123}
1124
Christoph Lameter81819f02007-05-06 14:49:36 -07001125static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1126{
Pekka Enberg06428782008-01-07 23:20:27 -08001127 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001128 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001129 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001130
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001131 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001132
Pekka Enbergba522702009-06-24 21:59:51 +03001133 /*
1134 * Let the initial higher-order allocation fail under memory pressure
1135 * so we fall-back to the minimum order allocation.
1136 */
1137 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1138
1139 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001140 if (unlikely(!page)) {
1141 oo = s->min;
1142 /*
1143 * Allocation may have failed due to fragmentation.
1144 * Try a lower order alloc if possible
1145 */
1146 page = alloc_slab_page(flags, node, oo);
1147 if (!page)
1148 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001149
Christoph Lameter84e554e2009-12-18 16:26:23 -06001150 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001151 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001152
1153 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001154 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001155 int pages = 1 << oo_order(oo);
1156
1157 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1158
1159 /*
1160 * Objects from caches that have a constructor don't get
1161 * cleared when they're allocated, so we need to do it here.
1162 */
1163 if (s->ctor)
1164 kmemcheck_mark_uninitialized_pages(page, pages);
1165 else
1166 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001167 }
1168
Christoph Lameter834f3d12008-04-14 19:11:31 +03001169 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001170 mod_zone_page_state(page_zone(page),
1171 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1172 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001173 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001174
1175 return page;
1176}
1177
1178static void setup_object(struct kmem_cache *s, struct page *page,
1179 void *object)
1180{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001181 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001182 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001183 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001184}
1185
1186static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1187{
1188 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001189 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001190 void *last;
1191 void *p;
1192
Christoph Lameter6cb06222007-10-16 01:25:41 -07001193 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001194
Christoph Lameter6cb06222007-10-16 01:25:41 -07001195 page = allocate_slab(s,
1196 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001197 if (!page)
1198 goto out;
1199
Christoph Lameter205ab992008-04-14 19:11:40 +03001200 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001201 page->slab = s;
1202 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001203
1204 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001205
1206 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001207 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001208
1209 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001210 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001211 setup_object(s, page, last);
1212 set_freepointer(s, last, p);
1213 last = p;
1214 }
1215 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001216 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001217
1218 page->freelist = start;
1219 page->inuse = 0;
1220out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001221 return page;
1222}
1223
1224static void __free_slab(struct kmem_cache *s, struct page *page)
1225{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001226 int order = compound_order(page);
1227 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001228
Christoph Lameteraf537b02010-07-09 14:07:14 -05001229 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001230 void *p;
1231
1232 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001233 for_each_object(p, s, page_address(page),
1234 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001235 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001236 }
1237
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001238 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001239
Christoph Lameter81819f02007-05-06 14:49:36 -07001240 mod_zone_page_state(page_zone(page),
1241 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1242 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001243 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001244
Christoph Lameter49bd5222008-04-14 18:52:18 +03001245 __ClearPageSlab(page);
1246 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001247 if (current->reclaim_state)
1248 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001249 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001250}
1251
1252static void rcu_free_slab(struct rcu_head *h)
1253{
1254 struct page *page;
1255
1256 page = container_of((struct list_head *)h, struct page, lru);
1257 __free_slab(page->slab, page);
1258}
1259
1260static void free_slab(struct kmem_cache *s, struct page *page)
1261{
1262 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1263 /*
1264 * RCU free overloads the RCU head over the LRU
1265 */
1266 struct rcu_head *head = (void *)&page->lru;
1267
1268 call_rcu(head, rcu_free_slab);
1269 } else
1270 __free_slab(s, page);
1271}
1272
1273static void discard_slab(struct kmem_cache *s, struct page *page)
1274{
Christoph Lameter205ab992008-04-14 19:11:40 +03001275 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001276 free_slab(s, page);
1277}
1278
1279/*
1280 * Per slab locking using the pagelock
1281 */
1282static __always_inline void slab_lock(struct page *page)
1283{
1284 bit_spin_lock(PG_locked, &page->flags);
1285}
1286
1287static __always_inline void slab_unlock(struct page *page)
1288{
Nick Piggina76d3542008-01-07 23:20:27 -08001289 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001290}
1291
1292static __always_inline int slab_trylock(struct page *page)
1293{
1294 int rc = 1;
1295
1296 rc = bit_spin_trylock(PG_locked, &page->flags);
1297 return rc;
1298}
1299
1300/*
1301 * Management of partially allocated slabs
1302 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001303static void add_partial(struct kmem_cache_node *n,
1304 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001305{
Christoph Lametere95eed52007-05-06 14:49:44 -07001306 spin_lock(&n->list_lock);
1307 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001308 if (tail)
1309 list_add_tail(&page->lru, &n->partial);
1310 else
1311 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001312 spin_unlock(&n->list_lock);
1313}
1314
Christoph Lameter62e346a2010-09-28 08:10:28 -05001315static inline void __remove_partial(struct kmem_cache_node *n,
1316 struct page *page)
1317{
1318 list_del(&page->lru);
1319 n->nr_partial--;
1320}
1321
Christoph Lameter0121c6192008-04-29 16:11:12 -07001322static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001323{
1324 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1325
1326 spin_lock(&n->list_lock);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001327 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001328 spin_unlock(&n->list_lock);
1329}
1330
1331/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001332 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001333 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001334 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001335 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001336static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1337 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001338{
1339 if (slab_trylock(page)) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05001340 __remove_partial(n, page);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001341 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001342 return 1;
1343 }
1344 return 0;
1345}
1346
1347/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001348 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001349 */
1350static struct page *get_partial_node(struct kmem_cache_node *n)
1351{
1352 struct page *page;
1353
1354 /*
1355 * Racy check. If we mistakenly see no partial slabs then we
1356 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001357 * partial slab and there is none available then get_partials()
1358 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001359 */
1360 if (!n || !n->nr_partial)
1361 return NULL;
1362
1363 spin_lock(&n->list_lock);
1364 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001365 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001366 goto out;
1367 page = NULL;
1368out:
1369 spin_unlock(&n->list_lock);
1370 return page;
1371}
1372
1373/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001374 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001375 */
1376static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1377{
1378#ifdef CONFIG_NUMA
1379 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001380 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001381 struct zone *zone;
1382 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001383 struct page *page;
1384
1385 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001386 * The defrag ratio allows a configuration of the tradeoffs between
1387 * inter node defragmentation and node local allocations. A lower
1388 * defrag_ratio increases the tendency to do local allocations
1389 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001390 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001391 * If the defrag_ratio is set to 0 then kmalloc() always
1392 * returns node local objects. If the ratio is higher then kmalloc()
1393 * may return off node objects because partial slabs are obtained
1394 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001395 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001396 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001397 * defrag_ratio = 1000) then every (well almost) allocation will
1398 * first attempt to defrag slab caches on other nodes. This means
1399 * scanning over all nodes to look for partial slabs which may be
1400 * expensive if we do it every time we are trying to find a slab
1401 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001402 */
Christoph Lameter98246012008-01-07 23:20:26 -08001403 if (!s->remote_node_defrag_ratio ||
1404 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001405 return NULL;
1406
Miao Xiec0ff7452010-05-24 14:32:08 -07001407 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001408 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001409 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001410 struct kmem_cache_node *n;
1411
Mel Gorman54a6eb52008-04-28 02:12:16 -07001412 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001413
Mel Gorman54a6eb52008-04-28 02:12:16 -07001414 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001415 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001416 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001417 if (page) {
1418 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001419 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001420 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001421 }
1422 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001423 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001424#endif
1425 return NULL;
1426}
1427
1428/*
1429 * Get a partial page, lock it and return it.
1430 */
1431static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1432{
1433 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001434 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001435
1436 page = get_partial_node(get_node(s, searchnode));
Christoph Lameterbc6488e2010-07-26 10:41:14 -05001437 if (page || node != -1)
Christoph Lameter81819f02007-05-06 14:49:36 -07001438 return page;
1439
1440 return get_any_partial(s, flags);
1441}
1442
1443/*
1444 * Move a page back to the lists.
1445 *
1446 * Must be called with the slab lock held.
1447 *
1448 * On exit the slab lock will have been dropped.
1449 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001450static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Namhyung Kim34789732010-09-29 21:02:14 +09001451 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001452{
Christoph Lametere95eed52007-05-06 14:49:44 -07001453 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1454
Andy Whitcroft8a380822008-07-23 21:27:18 -07001455 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001456 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001457
Christoph Lametera973e9d2008-03-01 13:40:44 -08001458 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001459 add_partial(n, page, tail);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001460 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001461 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001462 stat(s, DEACTIVATE_FULL);
Christoph Lameteraf537b02010-07-09 14:07:14 -05001463 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001464 add_full(n, page);
1465 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001466 slab_unlock(page);
1467 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001468 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001469 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001470 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001471 * Adding an empty slab to the partial slabs in order
1472 * to avoid page allocator overhead. This slab needs
1473 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001474 * so that the others get filled first. That way the
1475 * size of the partial list stays small.
1476 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001477 * kmem_cache_shrink can reclaim any empty slabs from
1478 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001479 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001480 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001481 slab_unlock(page);
1482 } else {
1483 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001484 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001485 discard_slab(s, page);
1486 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001487 }
1488}
1489
1490/*
1491 * Remove the cpu slab
1492 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001493static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Namhyung Kim34789732010-09-29 21:02:14 +09001494 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001495{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001496 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001497 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001498
Christoph Lameterb773ad72008-03-04 11:10:17 -08001499 if (page->freelist)
Christoph Lameter84e554e2009-12-18 16:26:23 -06001500 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001501 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001502 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001503 * because both freelists are empty. So this is unlikely
1504 * to occur.
1505 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001506 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001507 void **object;
1508
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001509 tail = 0; /* Hot objects. Put the slab first */
1510
Christoph Lameter894b8782007-05-10 03:15:16 -07001511 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001512 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001513 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001514
1515 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001516 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001517 page->freelist = object;
1518 page->inuse--;
1519 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001520 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001521 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001522}
1523
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001524static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001525{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001526 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001527 slab_lock(c->page);
1528 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001529}
1530
1531/*
1532 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001533 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001534 * Called from IPI handler with interrupts disabled.
1535 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001536static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001537{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001538 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001539
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001540 if (likely(c && c->page))
1541 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001542}
1543
1544static void flush_cpu_slab(void *d)
1545{
1546 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001547
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001548 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001549}
1550
1551static void flush_all(struct kmem_cache *s)
1552{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001553 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001554}
1555
1556/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001557 * Check if the objects in a per cpu structure fit numa
1558 * locality expectations.
1559 */
1560static inline int node_match(struct kmem_cache_cpu *c, int node)
1561{
1562#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001563 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001564 return 0;
1565#endif
1566 return 1;
1567}
1568
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001569static int count_free(struct page *page)
1570{
1571 return page->objects - page->inuse;
1572}
1573
1574static unsigned long count_partial(struct kmem_cache_node *n,
1575 int (*get_count)(struct page *))
1576{
1577 unsigned long flags;
1578 unsigned long x = 0;
1579 struct page *page;
1580
1581 spin_lock_irqsave(&n->list_lock, flags);
1582 list_for_each_entry(page, &n->partial, lru)
1583 x += get_count(page);
1584 spin_unlock_irqrestore(&n->list_lock, flags);
1585 return x;
1586}
1587
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001588static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1589{
1590#ifdef CONFIG_SLUB_DEBUG
1591 return atomic_long_read(&n->total_objects);
1592#else
1593 return 0;
1594#endif
1595}
1596
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001597static noinline void
1598slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1599{
1600 int node;
1601
1602 printk(KERN_WARNING
1603 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1604 nid, gfpflags);
1605 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1606 "default order: %d, min order: %d\n", s->name, s->objsize,
1607 s->size, oo_order(s->oo), oo_order(s->min));
1608
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001609 if (oo_order(s->min) > get_order(s->objsize))
1610 printk(KERN_WARNING " %s debugging increased min order, use "
1611 "slub_debug=O to disable.\n", s->name);
1612
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001613 for_each_online_node(node) {
1614 struct kmem_cache_node *n = get_node(s, node);
1615 unsigned long nr_slabs;
1616 unsigned long nr_objs;
1617 unsigned long nr_free;
1618
1619 if (!n)
1620 continue;
1621
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001622 nr_free = count_partial(n, count_free);
1623 nr_slabs = node_nr_slabs(n);
1624 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001625
1626 printk(KERN_WARNING
1627 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1628 node, nr_slabs, nr_objs, nr_free);
1629 }
1630}
1631
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001632/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001633 * Slow path. The lockless freelist is empty or we need to perform
1634 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001635 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001636 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001637 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001638 * Processing is still very fast if new objects have been freed to the
1639 * regular freelist. In that case we simply take over the regular freelist
1640 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001641 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001642 * If that is not working then we fall back to the partial lists. We take the
1643 * first element of the freelist as the object to allocate now and move the
1644 * rest of the freelist to the lockless freelist.
1645 *
1646 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001647 * we need to allocate a new slab. This is the slowest path since it involves
1648 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001649 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001650static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1651 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001652{
Christoph Lameter81819f02007-05-06 14:49:36 -07001653 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001654 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001655
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001656 /* We handle __GFP_ZERO in the caller */
1657 gfpflags &= ~__GFP_ZERO;
1658
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001659 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001660 goto new_slab;
1661
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001662 slab_lock(c->page);
1663 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001664 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001665
Christoph Lameter84e554e2009-12-18 16:26:23 -06001666 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001667
Christoph Lameter894b8782007-05-10 03:15:16 -07001668load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001669 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001670 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001671 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001672 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001673 goto debug;
1674
Christoph Lameterff120592009-12-18 16:26:22 -06001675 c->freelist = get_freepointer(s, object);
Christoph Lameter39b26462008-04-14 19:11:30 +03001676 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001677 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001678 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001679unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001680 slab_unlock(c->page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001681 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001682 return object;
1683
1684another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001685 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001686
1687new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001688 new = get_partial(s, gfpflags, node);
1689 if (new) {
1690 c->page = new;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001691 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001692 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001693 }
1694
Christoph Lameterc1d50832010-08-20 12:37:17 -05001695 gfpflags &= gfp_allowed_mask;
Christoph Lameterb811c202007-10-16 23:25:51 -07001696 if (gfpflags & __GFP_WAIT)
1697 local_irq_enable();
1698
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001699 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001700
1701 if (gfpflags & __GFP_WAIT)
1702 local_irq_disable();
1703
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001704 if (new) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001705 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001706 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001707 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001708 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001709 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001710 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001711 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001712 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001713 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001714 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1715 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001716 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001717debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001718 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001719 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001720
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001721 c->page->inuse++;
Christoph Lameterff120592009-12-18 16:26:22 -06001722 c->page->freelist = get_freepointer(s, object);
Pekka Enberg15b7c512010-10-02 11:32:32 +03001723 c->node = NUMA_NO_NODE;
Christoph Lameter1f842602008-01-07 23:20:30 -08001724 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001725}
1726
1727/*
1728 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1729 * have the fastpath folded into their functions. So no function call
1730 * overhead for requests that can be satisfied on the fastpath.
1731 *
1732 * The fastpath works by first checking if the lockless freelist can be used.
1733 * If not then __slab_alloc is called for slow processing.
1734 *
1735 * Otherwise we can simply pick the next object from the lockless free list.
1736 */
Pekka Enberg06428782008-01-07 23:20:27 -08001737static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001738 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001739{
Christoph Lameter894b8782007-05-10 03:15:16 -07001740 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001741 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001742 unsigned long flags;
1743
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001744 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001745 return NULL;
1746
Christoph Lameter894b8782007-05-10 03:15:16 -07001747 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001748 c = __this_cpu_ptr(s->cpu_slab);
1749 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001750 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001751
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001752 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001753
1754 else {
Christoph Lameterff120592009-12-18 16:26:22 -06001755 c->freelist = get_freepointer(s, object);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001756 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001757 }
1758 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001759
Pekka Enberg74e21342009-11-25 20:14:48 +02001760 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001761 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001762
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001763 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001764
Christoph Lameter894b8782007-05-10 03:15:16 -07001765 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001766}
1767
1768void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1769{
Christoph Lameter2154a332010-07-09 14:07:10 -05001770 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001771
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001772 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001773
1774 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001775}
1776EXPORT_SYMBOL(kmem_cache_alloc);
1777
Li Zefan0f24f122009-12-11 15:45:30 +08001778#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01001779void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001780{
Richard Kennedy4a923792010-10-21 10:29:19 +01001781 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1782 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1783 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001784}
Richard Kennedy4a923792010-10-21 10:29:19 +01001785EXPORT_SYMBOL(kmem_cache_alloc_trace);
1786
1787void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1788{
1789 void *ret = kmalloc_order(size, flags, order);
1790 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1791 return ret;
1792}
1793EXPORT_SYMBOL(kmalloc_order_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001794#endif
1795
Christoph Lameter81819f02007-05-06 14:49:36 -07001796#ifdef CONFIG_NUMA
1797void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1798{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001799 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1800
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001801 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1802 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001803
1804 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001805}
1806EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001807
Li Zefan0f24f122009-12-11 15:45:30 +08001808#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01001809void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001810 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01001811 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001812{
Richard Kennedy4a923792010-10-21 10:29:19 +01001813 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1814
1815 trace_kmalloc_node(_RET_IP_, ret,
1816 size, s->size, gfpflags, node);
1817 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001818}
Richard Kennedy4a923792010-10-21 10:29:19 +01001819EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001820#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09001821#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001822
Christoph Lameter81819f02007-05-06 14:49:36 -07001823/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001824 * Slow patch handling. This may still be called frequently since objects
1825 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001826 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001827 * So we still attempt to reduce cache line usage. Just take the slab
1828 * lock and free the item. If there is no additional partial page
1829 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001830 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001831static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06001832 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001833{
1834 void *prior;
1835 void **object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07001836
Christoph Lameter84e554e2009-12-18 16:26:23 -06001837 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001838 slab_lock(page);
1839
Christoph Lameteraf537b02010-07-09 14:07:14 -05001840 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001841 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001842
Christoph Lameter81819f02007-05-06 14:49:36 -07001843checks_ok:
Christoph Lameterff120592009-12-18 16:26:22 -06001844 prior = page->freelist;
1845 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07001846 page->freelist = object;
1847 page->inuse--;
1848
Andy Whitcroft8a380822008-07-23 21:27:18 -07001849 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001850 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001851 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001852 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001853
1854 if (unlikely(!page->inuse))
1855 goto slab_empty;
1856
1857 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001858 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001859 * then add it.
1860 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001861 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001862 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001863 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001864 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001865
1866out_unlock:
1867 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001868 return;
1869
1870slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001871 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001872 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001873 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001874 */
1875 remove_partial(s, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001876 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001877 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001878 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001879 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001880 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001881 return;
1882
1883debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001884 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001885 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001886 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001887}
1888
Christoph Lameter894b8782007-05-10 03:15:16 -07001889/*
1890 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1891 * can perform fastpath freeing without additional function calls.
1892 *
1893 * The fastpath is only possible if we are freeing to the current cpu slab
1894 * of this processor. This typically the case if we have just allocated
1895 * the item before.
1896 *
1897 * If fastpath is not possible then fall back to __slab_free where we deal
1898 * with all sorts of special processing.
1899 */
Pekka Enberg06428782008-01-07 23:20:27 -08001900static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001901 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001902{
1903 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001904 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001905 unsigned long flags;
1906
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001907 slab_free_hook(s, x);
1908
Christoph Lameter894b8782007-05-10 03:15:16 -07001909 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001910 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001911
1912 slab_free_hook_irq(s, x);
1913
Pekka Enberg15b7c512010-10-02 11:32:32 +03001914 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
Christoph Lameterff120592009-12-18 16:26:22 -06001915 set_freepointer(s, object, c->freelist);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001916 c->freelist = object;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001917 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001918 } else
Christoph Lameterff120592009-12-18 16:26:22 -06001919 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07001920
1921 local_irq_restore(flags);
1922}
1923
Christoph Lameter81819f02007-05-06 14:49:36 -07001924void kmem_cache_free(struct kmem_cache *s, void *x)
1925{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001926 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001927
Christoph Lameterb49af682007-05-06 14:49:41 -07001928 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001929
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001930 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001931
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001932 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001933}
1934EXPORT_SYMBOL(kmem_cache_free);
1935
Christoph Lameter81819f02007-05-06 14:49:36 -07001936/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001937 * Object placement in a slab is made very easy because we always start at
1938 * offset 0. If we tune the size of the object to the alignment then we can
1939 * get the required alignment by putting one properly sized object after
1940 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001941 *
1942 * Notice that the allocation order determines the sizes of the per cpu
1943 * caches. Each processor has always one slab available for allocations.
1944 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001945 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001946 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001947 */
1948
1949/*
1950 * Mininum / Maximum order of slab pages. This influences locking overhead
1951 * and slab fragmentation. A higher order reduces the number of partial slabs
1952 * and increases the number of allocations possible without having to
1953 * take the list_lock.
1954 */
1955static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001956static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001957static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001958
1959/*
1960 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001961 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001962 */
1963static int slub_nomerge;
1964
1965/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001966 * Calculate the order of allocation given an slab object size.
1967 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001968 * The order of allocation has significant impact on performance and other
1969 * system components. Generally order 0 allocations should be preferred since
1970 * order 0 does not cause fragmentation in the page allocator. Larger objects
1971 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001972 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001973 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001974 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001975 * In order to reach satisfactory performance we must ensure that a minimum
1976 * number of objects is in one slab. Otherwise we may generate too much
1977 * activity on the partial lists which requires taking the list_lock. This is
1978 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001979 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001980 * slub_max_order specifies the order where we begin to stop considering the
1981 * number of objects in a slab as critical. If we reach slub_max_order then
1982 * we try to keep the page order as low as possible. So we accept more waste
1983 * of space in favor of a small page order.
1984 *
1985 * Higher order allocations also allow the placement of more objects in a
1986 * slab and thereby reduce object handling overhead. If the user has
1987 * requested a higher mininum order then we start with that one instead of
1988 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001989 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001990static inline int slab_order(int size, int min_objects,
1991 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001992{
1993 int order;
1994 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001995 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001996
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001997 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1998 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001999
Christoph Lameter6300ea72007-07-17 04:03:20 -07002000 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002001 fls(min_objects * size - 1) - PAGE_SHIFT);
2002 order <= max_order; order++) {
2003
Christoph Lameter81819f02007-05-06 14:49:36 -07002004 unsigned long slab_size = PAGE_SIZE << order;
2005
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002006 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07002007 continue;
2008
Christoph Lameter81819f02007-05-06 14:49:36 -07002009 rem = slab_size % size;
2010
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002011 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002012 break;
2013
2014 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002015
Christoph Lameter81819f02007-05-06 14:49:36 -07002016 return order;
2017}
2018
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002019static inline int calculate_order(int size)
2020{
2021 int order;
2022 int min_objects;
2023 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002024 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002025
2026 /*
2027 * Attempt to find best configuration for a slab. This
2028 * works by first attempting to generate a layout with
2029 * the best configuration and backing off gradually.
2030 *
2031 * First we reduce the acceptable waste in a slab. Then
2032 * we reduce the minimum objects required in a slab.
2033 */
2034 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002035 if (!min_objects)
2036 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002037 max_objects = (PAGE_SIZE << slub_max_order)/size;
2038 min_objects = min(min_objects, max_objects);
2039
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002040 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002041 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002042 while (fraction >= 4) {
2043 order = slab_order(size, min_objects,
2044 slub_max_order, fraction);
2045 if (order <= slub_max_order)
2046 return order;
2047 fraction /= 2;
2048 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002049 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002050 }
2051
2052 /*
2053 * We were unable to place multiple objects in a slab. Now
2054 * lets see if we can place a single object there.
2055 */
2056 order = slab_order(size, 1, slub_max_order, 1);
2057 if (order <= slub_max_order)
2058 return order;
2059
2060 /*
2061 * Doh this slab cannot be placed using slub_max_order.
2062 */
2063 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002064 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002065 return order;
2066 return -ENOSYS;
2067}
2068
Christoph Lameter81819f02007-05-06 14:49:36 -07002069/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002070 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002071 */
2072static unsigned long calculate_alignment(unsigned long flags,
2073 unsigned long align, unsigned long size)
2074{
2075 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002076 * If the user wants hardware cache aligned objects then follow that
2077 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002078 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002079 * The hardware cache alignment cannot override the specified
2080 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002081 */
Nick Pigginb6210382008-03-05 14:05:56 -08002082 if (flags & SLAB_HWCACHE_ALIGN) {
2083 unsigned long ralign = cache_line_size();
2084 while (size <= ralign / 2)
2085 ralign /= 2;
2086 align = max(align, ralign);
2087 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002088
2089 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002090 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002091
2092 return ALIGN(align, sizeof(void *));
2093}
2094
Pekka Enberg5595cff2008-08-05 09:28:47 +03002095static void
2096init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002097{
2098 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002099 spin_lock_init(&n->list_lock);
2100 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002101#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002102 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002103 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002104 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002105#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002106}
2107
Christoph Lameter55136592010-08-20 12:37:13 -05002108static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002109{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002110 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2111 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002112
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002113 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002114
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002115 return s->cpu_slab != NULL;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002116}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002117
Christoph Lameter51df1142010-08-20 12:37:15 -05002118static struct kmem_cache *kmem_cache_node;
2119
Christoph Lameter81819f02007-05-06 14:49:36 -07002120/*
2121 * No kmalloc_node yet so do it by hand. We know that this is the first
2122 * slab on the node for this slabcache. There are no concurrent accesses
2123 * possible.
2124 *
2125 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002126 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2127 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002128 */
Christoph Lameter55136592010-08-20 12:37:13 -05002129static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002130{
2131 struct page *page;
2132 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002133 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002134
Christoph Lameter51df1142010-08-20 12:37:15 -05002135 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002136
Christoph Lameter51df1142010-08-20 12:37:15 -05002137 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002138
2139 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002140 if (page_to_nid(page) != node) {
2141 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2142 "node %d\n", node);
2143 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2144 "in order to be able to continue\n");
2145 }
2146
Christoph Lameter81819f02007-05-06 14:49:36 -07002147 n = page->freelist;
2148 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002149 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002150 page->inuse++;
Christoph Lameter51df1142010-08-20 12:37:15 -05002151 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002152#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002153 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002154 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002155#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002156 init_kmem_cache_node(n, kmem_cache_node);
2157 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002158
rootba84c732008-01-07 23:20:28 -08002159 /*
2160 * lockdep requires consistent irq usage for each lock
2161 * so even though there cannot be a race this early in
2162 * the boot sequence, we still disable irqs.
2163 */
2164 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002165 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002166 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002167}
2168
2169static void free_kmem_cache_nodes(struct kmem_cache *s)
2170{
2171 int node;
2172
Christoph Lameterf64dc582007-10-16 01:25:33 -07002173 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002174 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002175
Alexander Duyck73367bd2010-05-21 14:41:35 -07002176 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002177 kmem_cache_free(kmem_cache_node, n);
2178
Christoph Lameter81819f02007-05-06 14:49:36 -07002179 s->node[node] = NULL;
2180 }
2181}
2182
Christoph Lameter55136592010-08-20 12:37:13 -05002183static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002184{
2185 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002186
Christoph Lameterf64dc582007-10-16 01:25:33 -07002187 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002188 struct kmem_cache_node *n;
2189
Alexander Duyck73367bd2010-05-21 14:41:35 -07002190 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002191 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002192 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002193 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002194 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002195 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002196
2197 if (!n) {
2198 free_kmem_cache_nodes(s);
2199 return 0;
2200 }
2201
Christoph Lameter81819f02007-05-06 14:49:36 -07002202 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002203 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002204 }
2205 return 1;
2206}
Christoph Lameter81819f02007-05-06 14:49:36 -07002207
David Rientjesc0bdb232009-02-25 09:16:35 +02002208static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002209{
2210 if (min < MIN_PARTIAL)
2211 min = MIN_PARTIAL;
2212 else if (min > MAX_PARTIAL)
2213 min = MAX_PARTIAL;
2214 s->min_partial = min;
2215}
2216
Christoph Lameter81819f02007-05-06 14:49:36 -07002217/*
2218 * calculate_sizes() determines the order and the distribution of data within
2219 * a slab object.
2220 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002221static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002222{
2223 unsigned long flags = s->flags;
2224 unsigned long size = s->objsize;
2225 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002226 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002227
2228 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002229 * Round up object size to the next word boundary. We can only
2230 * place the free pointer at word boundaries and this determines
2231 * the possible location of the free pointer.
2232 */
2233 size = ALIGN(size, sizeof(void *));
2234
2235#ifdef CONFIG_SLUB_DEBUG
2236 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002237 * Determine if we can poison the object itself. If the user of
2238 * the slab may touch the object after free or before allocation
2239 * then we should never poison the object itself.
2240 */
2241 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002242 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002243 s->flags |= __OBJECT_POISON;
2244 else
2245 s->flags &= ~__OBJECT_POISON;
2246
Christoph Lameter81819f02007-05-06 14:49:36 -07002247
2248 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002249 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002250 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002251 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002252 */
2253 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2254 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002255#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002256
2257 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002258 * With that we have determined the number of bytes in actual use
2259 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002260 */
2261 s->inuse = size;
2262
2263 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002264 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002265 /*
2266 * Relocate free pointer after the object if it is not
2267 * permitted to overwrite the first word of the object on
2268 * kmem_cache_free.
2269 *
2270 * This is the case if we do RCU, have a constructor or
2271 * destructor or are poisoning the objects.
2272 */
2273 s->offset = size;
2274 size += sizeof(void *);
2275 }
2276
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002277#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002278 if (flags & SLAB_STORE_USER)
2279 /*
2280 * Need to store information about allocs and frees after
2281 * the object.
2282 */
2283 size += 2 * sizeof(struct track);
2284
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002285 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002286 /*
2287 * Add some empty padding so that we can catch
2288 * overwrites from earlier objects rather than let
2289 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002290 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002291 * of the object.
2292 */
2293 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002294#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002295
Christoph Lameter81819f02007-05-06 14:49:36 -07002296 /*
2297 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002298 * user specified and the dynamic determination of cache line size
2299 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002300 */
2301 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002302 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002303
2304 /*
2305 * SLUB stores one object immediately after another beginning from
2306 * offset 0. In order to align the objects we have to simply size
2307 * each object to conform to the alignment.
2308 */
2309 size = ALIGN(size, align);
2310 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002311 if (forced_order >= 0)
2312 order = forced_order;
2313 else
2314 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002315
Christoph Lameter834f3d12008-04-14 19:11:31 +03002316 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002317 return 0;
2318
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002319 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002320 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002321 s->allocflags |= __GFP_COMP;
2322
2323 if (s->flags & SLAB_CACHE_DMA)
2324 s->allocflags |= SLUB_DMA;
2325
2326 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2327 s->allocflags |= __GFP_RECLAIMABLE;
2328
Christoph Lameter81819f02007-05-06 14:49:36 -07002329 /*
2330 * Determine the number of objects per slab
2331 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002332 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002333 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002334 if (oo_objects(s->oo) > oo_objects(s->max))
2335 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002336
Christoph Lameter834f3d12008-04-14 19:11:31 +03002337 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002338
2339}
2340
Christoph Lameter55136592010-08-20 12:37:13 -05002341static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002342 const char *name, size_t size,
2343 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002344 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002345{
2346 memset(s, 0, kmem_size);
2347 s->name = name;
2348 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002349 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002350 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002351 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002352
Christoph Lameter06b285d2008-04-14 19:11:41 +03002353 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002354 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002355 if (disable_higher_order_debug) {
2356 /*
2357 * Disable debugging flags that store metadata if the min slab
2358 * order increased.
2359 */
2360 if (get_order(s->size) > get_order(s->objsize)) {
2361 s->flags &= ~DEBUG_METADATA_FLAGS;
2362 s->offset = 0;
2363 if (!calculate_sizes(s, -1))
2364 goto error;
2365 }
2366 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002367
David Rientjes3b89d7d2009-02-22 17:40:07 -08002368 /*
2369 * The larger the object size is, the more pages we want on the partial
2370 * list to avoid pounding the page allocator excessively.
2371 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002372 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002373 s->refcount = 1;
2374#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002375 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002376#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002377 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002378 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002379
Christoph Lameter55136592010-08-20 12:37:13 -05002380 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002381 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002382
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002383 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002384error:
2385 if (flags & SLAB_PANIC)
2386 panic("Cannot create slab %s size=%lu realsize=%u "
2387 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002388 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002389 s->offset, flags);
2390 return 0;
2391}
Christoph Lameter81819f02007-05-06 14:49:36 -07002392
2393/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002394 * Determine the size of a slab object
2395 */
2396unsigned int kmem_cache_size(struct kmem_cache *s)
2397{
2398 return s->objsize;
2399}
2400EXPORT_SYMBOL(kmem_cache_size);
2401
2402const char *kmem_cache_name(struct kmem_cache *s)
2403{
2404 return s->name;
2405}
2406EXPORT_SYMBOL(kmem_cache_name);
2407
Christoph Lameter33b12c32008-04-25 12:22:43 -07002408static void list_slab_objects(struct kmem_cache *s, struct page *page,
2409 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002410{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002411#ifdef CONFIG_SLUB_DEBUG
2412 void *addr = page_address(page);
2413 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002414 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2415 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002416 if (!map)
2417 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002418 slab_err(s, page, "%s", text);
2419 slab_lock(page);
2420 for_each_free_object(p, s, page->freelist)
2421 set_bit(slab_index(p, s, addr), map);
2422
2423 for_each_object(p, s, addr, page->objects) {
2424
2425 if (!test_bit(slab_index(p, s, addr), map)) {
2426 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2427 p, p - addr);
2428 print_tracking(s, p);
2429 }
2430 }
2431 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002432 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002433#endif
2434}
2435
Christoph Lameter81819f02007-05-06 14:49:36 -07002436/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002437 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002438 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002439static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002440{
Christoph Lameter81819f02007-05-06 14:49:36 -07002441 unsigned long flags;
2442 struct page *page, *h;
2443
2444 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002445 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002446 if (!page->inuse) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05002447 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002448 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002449 } else {
2450 list_slab_objects(s, page,
2451 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002452 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002453 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002454 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002455}
2456
2457/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002458 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002459 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002460static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002461{
2462 int node;
2463
2464 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002465 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002466 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002467 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002468 struct kmem_cache_node *n = get_node(s, node);
2469
Christoph Lameter599870b2008-04-23 12:36:52 -07002470 free_partial(s, n);
2471 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002472 return 1;
2473 }
2474 free_kmem_cache_nodes(s);
2475 return 0;
2476}
2477
2478/*
2479 * Close a cache and release the kmem_cache structure
2480 * (must be used for caches created using kmem_cache_create)
2481 */
2482void kmem_cache_destroy(struct kmem_cache *s)
2483{
2484 down_write(&slub_lock);
2485 s->refcount--;
2486 if (!s->refcount) {
2487 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002488 if (kmem_cache_close(s)) {
2489 printk(KERN_ERR "SLUB %s: %s called for cache that "
2490 "still has objects.\n", s->name, __func__);
2491 dump_stack();
2492 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002493 if (s->flags & SLAB_DESTROY_BY_RCU)
2494 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002495 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002496 }
2497 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002498}
2499EXPORT_SYMBOL(kmem_cache_destroy);
2500
2501/********************************************************************
2502 * Kmalloc subsystem
2503 *******************************************************************/
2504
Christoph Lameter51df1142010-08-20 12:37:15 -05002505struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002506EXPORT_SYMBOL(kmalloc_caches);
2507
Christoph Lameter51df1142010-08-20 12:37:15 -05002508static struct kmem_cache *kmem_cache;
2509
Christoph Lameter55136592010-08-20 12:37:13 -05002510#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002511static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002512#endif
2513
Christoph Lameter81819f02007-05-06 14:49:36 -07002514static int __init setup_slub_min_order(char *str)
2515{
Pekka Enberg06428782008-01-07 23:20:27 -08002516 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002517
2518 return 1;
2519}
2520
2521__setup("slub_min_order=", setup_slub_min_order);
2522
2523static int __init setup_slub_max_order(char *str)
2524{
Pekka Enberg06428782008-01-07 23:20:27 -08002525 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002526 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002527
2528 return 1;
2529}
2530
2531__setup("slub_max_order=", setup_slub_max_order);
2532
2533static int __init setup_slub_min_objects(char *str)
2534{
Pekka Enberg06428782008-01-07 23:20:27 -08002535 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002536
2537 return 1;
2538}
2539
2540__setup("slub_min_objects=", setup_slub_min_objects);
2541
2542static int __init setup_slub_nomerge(char *str)
2543{
2544 slub_nomerge = 1;
2545 return 1;
2546}
2547
2548__setup("slub_nomerge", setup_slub_nomerge);
2549
Christoph Lameter51df1142010-08-20 12:37:15 -05002550static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2551 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002552{
Christoph Lameter51df1142010-08-20 12:37:15 -05002553 struct kmem_cache *s;
2554
2555 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2556
Pekka Enberg83b519e2009-06-10 19:40:04 +03002557 /*
2558 * This function is called with IRQs disabled during early-boot on
2559 * single CPU so there's no need to take slub_lock here.
2560 */
Christoph Lameter55136592010-08-20 12:37:13 -05002561 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002562 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002563 goto panic;
2564
2565 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002566 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002567
2568panic:
2569 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002570 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002571}
2572
Christoph Lameterf1b26332007-07-17 04:03:26 -07002573/*
2574 * Conversion table for small slabs sizes / 8 to the index in the
2575 * kmalloc array. This is necessary for slabs < 192 since we have non power
2576 * of two cache sizes there. The size of larger slabs can be determined using
2577 * fls.
2578 */
2579static s8 size_index[24] = {
2580 3, /* 8 */
2581 4, /* 16 */
2582 5, /* 24 */
2583 5, /* 32 */
2584 6, /* 40 */
2585 6, /* 48 */
2586 6, /* 56 */
2587 6, /* 64 */
2588 1, /* 72 */
2589 1, /* 80 */
2590 1, /* 88 */
2591 1, /* 96 */
2592 7, /* 104 */
2593 7, /* 112 */
2594 7, /* 120 */
2595 7, /* 128 */
2596 2, /* 136 */
2597 2, /* 144 */
2598 2, /* 152 */
2599 2, /* 160 */
2600 2, /* 168 */
2601 2, /* 176 */
2602 2, /* 184 */
2603 2 /* 192 */
2604};
2605
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002606static inline int size_index_elem(size_t bytes)
2607{
2608 return (bytes - 1) / 8;
2609}
2610
Christoph Lameter81819f02007-05-06 14:49:36 -07002611static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2612{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002613 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002614
Christoph Lameterf1b26332007-07-17 04:03:26 -07002615 if (size <= 192) {
2616 if (!size)
2617 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002618
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002619 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002620 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002621 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002622
2623#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002624 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002625 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002626
Christoph Lameter81819f02007-05-06 14:49:36 -07002627#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002628 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002629}
2630
2631void *__kmalloc(size_t size, gfp_t flags)
2632{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002633 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002634 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002635
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002636 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002637 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002638
2639 s = get_slab(size, flags);
2640
2641 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002642 return s;
2643
Christoph Lameter2154a332010-07-09 14:07:10 -05002644 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002645
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002646 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002647
2648 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002649}
2650EXPORT_SYMBOL(__kmalloc);
2651
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002652#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002653static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2654{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002655 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002656 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002657
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002658 flags |= __GFP_COMP | __GFP_NOTRACK;
2659 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002660 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002661 ptr = page_address(page);
2662
2663 kmemleak_alloc(ptr, size, 1, flags);
2664 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002665}
2666
Christoph Lameter81819f02007-05-06 14:49:36 -07002667void *__kmalloc_node(size_t size, gfp_t flags, int node)
2668{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002669 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002670 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002671
Ingo Molnar057685c2009-02-20 12:15:30 +01002672 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002673 ret = kmalloc_large_node(size, flags, node);
2674
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002675 trace_kmalloc_node(_RET_IP_, ret,
2676 size, PAGE_SIZE << get_order(size),
2677 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002678
2679 return ret;
2680 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002681
2682 s = get_slab(size, flags);
2683
2684 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002685 return s;
2686
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002687 ret = slab_alloc(s, flags, node, _RET_IP_);
2688
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002689 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002690
2691 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002692}
2693EXPORT_SYMBOL(__kmalloc_node);
2694#endif
2695
2696size_t ksize(const void *object)
2697{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002698 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002699 struct kmem_cache *s;
2700
Christoph Lameteref8b4522007-10-16 01:24:46 -07002701 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002702 return 0;
2703
Vegard Nossum294a80a2007-12-04 23:45:30 -08002704 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002705
Pekka Enberg76994412008-05-22 19:22:25 +03002706 if (unlikely(!PageSlab(page))) {
2707 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002708 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002709 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002710 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002711
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002712#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002713 /*
2714 * Debugging requires use of the padding between object
2715 * and whatever may come after it.
2716 */
2717 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2718 return s->objsize;
2719
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002720#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002721 /*
2722 * If we have the need to store the freelist pointer
2723 * back there or track user information then we can
2724 * only use the space before that information.
2725 */
2726 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2727 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002728 /*
2729 * Else we can use all the padding etc for the allocation
2730 */
2731 return s->size;
2732}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002733EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002734
2735void kfree(const void *x)
2736{
Christoph Lameter81819f02007-05-06 14:49:36 -07002737 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002738 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002739
Pekka Enberg2121db72009-03-25 11:05:57 +02002740 trace_kfree(_RET_IP_, x);
2741
Satyam Sharma2408c552007-10-16 01:24:44 -07002742 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002743 return;
2744
Christoph Lameterb49af682007-05-06 14:49:41 -07002745 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002746 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002747 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002748 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002749 put_page(page);
2750 return;
2751 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002752 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002753}
2754EXPORT_SYMBOL(kfree);
2755
Christoph Lameter2086d262007-05-06 14:49:46 -07002756/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002757 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2758 * the remaining slabs by the number of items in use. The slabs with the
2759 * most items in use come first. New allocations will then fill those up
2760 * and thus they can be removed from the partial lists.
2761 *
2762 * The slabs with the least items are placed last. This results in them
2763 * being allocated from last increasing the chance that the last objects
2764 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002765 */
2766int kmem_cache_shrink(struct kmem_cache *s)
2767{
2768 int node;
2769 int i;
2770 struct kmem_cache_node *n;
2771 struct page *page;
2772 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002773 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002774 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002775 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002776 unsigned long flags;
2777
2778 if (!slabs_by_inuse)
2779 return -ENOMEM;
2780
2781 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002782 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002783 n = get_node(s, node);
2784
2785 if (!n->nr_partial)
2786 continue;
2787
Christoph Lameter834f3d12008-04-14 19:11:31 +03002788 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002789 INIT_LIST_HEAD(slabs_by_inuse + i);
2790
2791 spin_lock_irqsave(&n->list_lock, flags);
2792
2793 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002794 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002795 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002796 * Note that concurrent frees may occur while we hold the
2797 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002798 */
2799 list_for_each_entry_safe(page, t, &n->partial, lru) {
2800 if (!page->inuse && slab_trylock(page)) {
2801 /*
2802 * Must hold slab lock here because slab_free
2803 * may have freed the last object and be
2804 * waiting to release the slab.
2805 */
Christoph Lameter62e346a2010-09-28 08:10:28 -05002806 __remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07002807 slab_unlock(page);
2808 discard_slab(s, page);
2809 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002810 list_move(&page->lru,
2811 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002812 }
2813 }
2814
Christoph Lameter2086d262007-05-06 14:49:46 -07002815 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002816 * Rebuild the partial list with the slabs filled up most
2817 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002818 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002819 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002820 list_splice(slabs_by_inuse + i, n->partial.prev);
2821
Christoph Lameter2086d262007-05-06 14:49:46 -07002822 spin_unlock_irqrestore(&n->list_lock, flags);
2823 }
2824
2825 kfree(slabs_by_inuse);
2826 return 0;
2827}
2828EXPORT_SYMBOL(kmem_cache_shrink);
2829
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03002830#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07002831static int slab_mem_going_offline_callback(void *arg)
2832{
2833 struct kmem_cache *s;
2834
2835 down_read(&slub_lock);
2836 list_for_each_entry(s, &slab_caches, list)
2837 kmem_cache_shrink(s);
2838 up_read(&slub_lock);
2839
2840 return 0;
2841}
2842
2843static void slab_mem_offline_callback(void *arg)
2844{
2845 struct kmem_cache_node *n;
2846 struct kmem_cache *s;
2847 struct memory_notify *marg = arg;
2848 int offline_node;
2849
2850 offline_node = marg->status_change_nid;
2851
2852 /*
2853 * If the node still has available memory. we need kmem_cache_node
2854 * for it yet.
2855 */
2856 if (offline_node < 0)
2857 return;
2858
2859 down_read(&slub_lock);
2860 list_for_each_entry(s, &slab_caches, list) {
2861 n = get_node(s, offline_node);
2862 if (n) {
2863 /*
2864 * if n->nr_slabs > 0, slabs still exist on the node
2865 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05002866 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07002867 * callback. So, we must fail.
2868 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002869 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002870
2871 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05002872 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002873 }
2874 }
2875 up_read(&slub_lock);
2876}
2877
2878static int slab_mem_going_online_callback(void *arg)
2879{
2880 struct kmem_cache_node *n;
2881 struct kmem_cache *s;
2882 struct memory_notify *marg = arg;
2883 int nid = marg->status_change_nid;
2884 int ret = 0;
2885
2886 /*
2887 * If the node's memory is already available, then kmem_cache_node is
2888 * already created. Nothing to do.
2889 */
2890 if (nid < 0)
2891 return 0;
2892
2893 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002894 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002895 * allocate a kmem_cache_node structure in order to bring the node
2896 * online.
2897 */
2898 down_read(&slub_lock);
2899 list_for_each_entry(s, &slab_caches, list) {
2900 /*
2901 * XXX: kmem_cache_alloc_node will fallback to other nodes
2902 * since memory is not yet available from the node that
2903 * is brought up.
2904 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05002905 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002906 if (!n) {
2907 ret = -ENOMEM;
2908 goto out;
2909 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03002910 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002911 s->node[nid] = n;
2912 }
2913out:
2914 up_read(&slub_lock);
2915 return ret;
2916}
2917
2918static int slab_memory_callback(struct notifier_block *self,
2919 unsigned long action, void *arg)
2920{
2921 int ret = 0;
2922
2923 switch (action) {
2924 case MEM_GOING_ONLINE:
2925 ret = slab_mem_going_online_callback(arg);
2926 break;
2927 case MEM_GOING_OFFLINE:
2928 ret = slab_mem_going_offline_callback(arg);
2929 break;
2930 case MEM_OFFLINE:
2931 case MEM_CANCEL_ONLINE:
2932 slab_mem_offline_callback(arg);
2933 break;
2934 case MEM_ONLINE:
2935 case MEM_CANCEL_OFFLINE:
2936 break;
2937 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08002938 if (ret)
2939 ret = notifier_from_errno(ret);
2940 else
2941 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07002942 return ret;
2943}
2944
2945#endif /* CONFIG_MEMORY_HOTPLUG */
2946
Christoph Lameter81819f02007-05-06 14:49:36 -07002947/********************************************************************
2948 * Basic setup of slabs
2949 *******************************************************************/
2950
Christoph Lameter51df1142010-08-20 12:37:15 -05002951/*
2952 * Used for early kmem_cache structures that were allocated using
2953 * the page allocator
2954 */
2955
2956static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
2957{
2958 int node;
2959
2960 list_add(&s->list, &slab_caches);
2961 s->refcount = -1;
2962
2963 for_each_node_state(node, N_NORMAL_MEMORY) {
2964 struct kmem_cache_node *n = get_node(s, node);
2965 struct page *p;
2966
2967 if (n) {
2968 list_for_each_entry(p, &n->partial, lru)
2969 p->slab = s;
2970
2971#ifdef CONFIG_SLAB_DEBUG
2972 list_for_each_entry(p, &n->full, lru)
2973 p->slab = s;
2974#endif
2975 }
2976 }
2977}
2978
Christoph Lameter81819f02007-05-06 14:49:36 -07002979void __init kmem_cache_init(void)
2980{
2981 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07002982 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002983 struct kmem_cache *temp_kmem_cache;
2984 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05002985 struct kmem_cache *temp_kmem_cache_node;
2986 unsigned long kmalloc_size;
2987
2988 kmem_size = offsetof(struct kmem_cache, node) +
2989 nr_node_ids * sizeof(struct kmem_cache_node *);
2990
2991 /* Allocate two kmem_caches from the page allocator */
2992 kmalloc_size = ALIGN(kmem_size, cache_line_size());
2993 order = get_order(2 * kmalloc_size);
2994 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
2995
Christoph Lameter81819f02007-05-06 14:49:36 -07002996 /*
2997 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07002998 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07002999 * kmem_cache_open for slab_state == DOWN.
3000 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003001 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3002
3003 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3004 sizeof(struct kmem_cache_node),
3005 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003006
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003007 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003008
3009 /* Able to allocate the per node structures */
3010 slab_state = PARTIAL;
3011
Christoph Lameter51df1142010-08-20 12:37:15 -05003012 temp_kmem_cache = kmem_cache;
3013 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3014 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3015 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3016 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003017
Christoph Lameter51df1142010-08-20 12:37:15 -05003018 /*
3019 * Allocate kmem_cache_node properly from the kmem_cache slab.
3020 * kmem_cache_node is separately allocated so no need to
3021 * update any list pointers.
3022 */
3023 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003024
Christoph Lameter51df1142010-08-20 12:37:15 -05003025 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3026 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3027
3028 kmem_cache_bootstrap_fixup(kmem_cache_node);
3029
3030 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003031 kmem_cache_bootstrap_fixup(kmem_cache);
3032 caches++;
3033 /* Free temporary boot structure */
3034 free_pages((unsigned long)temp_kmem_cache, order);
3035
3036 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003037
3038 /*
3039 * Patch up the size_index table if we have strange large alignment
3040 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003041 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003042 *
3043 * Largest permitted alignment is 256 bytes due to the way we
3044 * handle the index determination for the smaller caches.
3045 *
3046 * Make sure that nothing crazy happens if someone starts tinkering
3047 * around with ARCH_KMALLOC_MINALIGN
3048 */
3049 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3050 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3051
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003052 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3053 int elem = size_index_elem(i);
3054 if (elem >= ARRAY_SIZE(size_index))
3055 break;
3056 size_index[elem] = KMALLOC_SHIFT_LOW;
3057 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003058
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003059 if (KMALLOC_MIN_SIZE == 64) {
3060 /*
3061 * The 96 byte size cache is not used if the alignment
3062 * is 64 byte.
3063 */
3064 for (i = 64 + 8; i <= 96; i += 8)
3065 size_index[size_index_elem(i)] = 7;
3066 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003067 /*
3068 * The 192 byte sized cache is not used if the alignment
3069 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3070 * instead.
3071 */
3072 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003073 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003074 }
3075
Christoph Lameter51df1142010-08-20 12:37:15 -05003076 /* Caches that are not of the two-to-the-power-of size */
3077 if (KMALLOC_MIN_SIZE <= 32) {
3078 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3079 caches++;
3080 }
3081
3082 if (KMALLOC_MIN_SIZE <= 64) {
3083 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3084 caches++;
3085 }
3086
3087 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3088 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3089 caches++;
3090 }
3091
Christoph Lameter81819f02007-05-06 14:49:36 -07003092 slab_state = UP;
3093
3094 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003095 if (KMALLOC_MIN_SIZE <= 32) {
3096 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3097 BUG_ON(!kmalloc_caches[1]->name);
3098 }
3099
3100 if (KMALLOC_MIN_SIZE <= 64) {
3101 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3102 BUG_ON(!kmalloc_caches[2]->name);
3103 }
3104
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003105 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3106 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3107
3108 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003109 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003110 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003111
3112#ifdef CONFIG_SMP
3113 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003114#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003115
Christoph Lameter55136592010-08-20 12:37:13 -05003116#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003117 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3118 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003119
Christoph Lameter51df1142010-08-20 12:37:15 -05003120 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003121 char *name = kasprintf(GFP_NOWAIT,
3122 "dma-kmalloc-%d", s->objsize);
3123
3124 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003125 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3126 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003127 }
3128 }
3129#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003130 printk(KERN_INFO
3131 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003132 " CPUs=%d, Nodes=%d\n",
3133 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003134 slub_min_order, slub_max_order, slub_min_objects,
3135 nr_cpu_ids, nr_node_ids);
3136}
3137
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003138void __init kmem_cache_init_late(void)
3139{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003140}
3141
Christoph Lameter81819f02007-05-06 14:49:36 -07003142/*
3143 * Find a mergeable slab cache
3144 */
3145static int slab_unmergeable(struct kmem_cache *s)
3146{
3147 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3148 return 1;
3149
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003150 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003151 return 1;
3152
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003153 /*
3154 * We may have set a slab to be unmergeable during bootstrap.
3155 */
3156 if (s->refcount < 0)
3157 return 1;
3158
Christoph Lameter81819f02007-05-06 14:49:36 -07003159 return 0;
3160}
3161
3162static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003163 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003164 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003165{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003166 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003167
3168 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3169 return NULL;
3170
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003171 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003172 return NULL;
3173
3174 size = ALIGN(size, sizeof(void *));
3175 align = calculate_alignment(flags, align, size);
3176 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003177 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003178
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003179 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003180 if (slab_unmergeable(s))
3181 continue;
3182
3183 if (size > s->size)
3184 continue;
3185
Christoph Lameterba0268a2007-09-11 15:24:11 -07003186 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003187 continue;
3188 /*
3189 * Check if alignment is compatible.
3190 * Courtesy of Adrian Drzewiecki
3191 */
Pekka Enberg06428782008-01-07 23:20:27 -08003192 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003193 continue;
3194
3195 if (s->size - size >= sizeof(void *))
3196 continue;
3197
3198 return s;
3199 }
3200 return NULL;
3201}
3202
3203struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003204 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003205{
3206 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003207 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003208
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003209 if (WARN_ON(!name))
3210 return NULL;
3211
Christoph Lameter81819f02007-05-06 14:49:36 -07003212 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003213 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003214 if (s) {
3215 s->refcount++;
3216 /*
3217 * Adjust the object sizes so that we clear
3218 * the complete object on kzalloc.
3219 */
3220 s->objsize = max(s->objsize, (int)size);
3221 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003222
David Rientjes7b8f3b62008-12-17 22:09:46 -08003223 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003224 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003225 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003226 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003227 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003228 return s;
3229 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003230
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003231 n = kstrdup(name, GFP_KERNEL);
3232 if (!n)
3233 goto err;
3234
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003235 s = kmalloc(kmem_size, GFP_KERNEL);
3236 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003237 if (kmem_cache_open(s, n,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003238 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003239 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003240 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003241 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003242 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003243 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003244 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003245 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003246 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003247 return s;
3248 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003249 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003250 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003251 }
Pavel Emelyanov68cee4f2010-10-28 13:50:37 +04003252err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003253 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003254
Christoph Lameter81819f02007-05-06 14:49:36 -07003255 if (flags & SLAB_PANIC)
3256 panic("Cannot create slabcache %s\n", name);
3257 else
3258 s = NULL;
3259 return s;
3260}
3261EXPORT_SYMBOL(kmem_cache_create);
3262
Christoph Lameter81819f02007-05-06 14:49:36 -07003263#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003264/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003265 * Use the cpu notifier to insure that the cpu slabs are flushed when
3266 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003267 */
3268static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3269 unsigned long action, void *hcpu)
3270{
3271 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003272 struct kmem_cache *s;
3273 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003274
3275 switch (action) {
3276 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003277 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003278 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003279 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003280 down_read(&slub_lock);
3281 list_for_each_entry(s, &slab_caches, list) {
3282 local_irq_save(flags);
3283 __flush_cpu_slab(s, cpu);
3284 local_irq_restore(flags);
3285 }
3286 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003287 break;
3288 default:
3289 break;
3290 }
3291 return NOTIFY_OK;
3292}
3293
Pekka Enberg06428782008-01-07 23:20:27 -08003294static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003295 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003296};
Christoph Lameter81819f02007-05-06 14:49:36 -07003297
3298#endif
3299
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003300void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003301{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003302 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003303 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003304
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003305 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003306 return kmalloc_large(size, gfpflags);
3307
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003308 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003309
Satyam Sharma2408c552007-10-16 01:24:44 -07003310 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003311 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003312
Christoph Lameter2154a332010-07-09 14:07:10 -05003313 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003314
3315 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003316 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003317
3318 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003319}
3320
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003321#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003322void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003323 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003324{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003325 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003326 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003327
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003328 if (unlikely(size > SLUB_MAX_SIZE)) {
3329 ret = kmalloc_large_node(size, gfpflags, node);
3330
3331 trace_kmalloc_node(caller, ret,
3332 size, PAGE_SIZE << get_order(size),
3333 gfpflags, node);
3334
3335 return ret;
3336 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003337
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003338 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003339
Satyam Sharma2408c552007-10-16 01:24:44 -07003340 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003341 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003342
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003343 ret = slab_alloc(s, gfpflags, node, caller);
3344
3345 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003346 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003347
3348 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003349}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003350#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003351
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003352#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003353static int count_inuse(struct page *page)
3354{
3355 return page->inuse;
3356}
3357
3358static int count_total(struct page *page)
3359{
3360 return page->objects;
3361}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003362#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003363
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003364#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003365static int validate_slab(struct kmem_cache *s, struct page *page,
3366 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003367{
3368 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003369 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003370
3371 if (!check_slab(s, page) ||
3372 !on_freelist(s, page, NULL))
3373 return 0;
3374
3375 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003376 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003377
Christoph Lameter7656c722007-05-09 02:32:40 -07003378 for_each_free_object(p, s, page->freelist) {
3379 set_bit(slab_index(p, s, addr), map);
Tero Roponen37d57442010-12-01 20:04:20 +02003380 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003381 return 0;
3382 }
3383
Christoph Lameter224a88b2008-04-14 19:11:31 +03003384 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003385 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003386 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003387 return 0;
3388 return 1;
3389}
3390
Christoph Lameter434e2452007-07-17 04:03:30 -07003391static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3392 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003393{
3394 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003395 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003396 slab_unlock(page);
3397 } else
3398 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3399 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003400}
3401
Christoph Lameter434e2452007-07-17 04:03:30 -07003402static int validate_slab_node(struct kmem_cache *s,
3403 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003404{
3405 unsigned long count = 0;
3406 struct page *page;
3407 unsigned long flags;
3408
3409 spin_lock_irqsave(&n->list_lock, flags);
3410
3411 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003412 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003413 count++;
3414 }
3415 if (count != n->nr_partial)
3416 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3417 "counter=%ld\n", s->name, count, n->nr_partial);
3418
3419 if (!(s->flags & SLAB_STORE_USER))
3420 goto out;
3421
3422 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003423 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003424 count++;
3425 }
3426 if (count != atomic_long_read(&n->nr_slabs))
3427 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3428 "counter=%ld\n", s->name, count,
3429 atomic_long_read(&n->nr_slabs));
3430
3431out:
3432 spin_unlock_irqrestore(&n->list_lock, flags);
3433 return count;
3434}
3435
Christoph Lameter434e2452007-07-17 04:03:30 -07003436static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003437{
3438 int node;
3439 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003440 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003441 sizeof(unsigned long), GFP_KERNEL);
3442
3443 if (!map)
3444 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003445
3446 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003447 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003448 struct kmem_cache_node *n = get_node(s, node);
3449
Christoph Lameter434e2452007-07-17 04:03:30 -07003450 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003451 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003452 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003453 return count;
3454}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003455/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003456 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003457 * and freed.
3458 */
3459
3460struct location {
3461 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003462 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003463 long long sum_time;
3464 long min_time;
3465 long max_time;
3466 long min_pid;
3467 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303468 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003469 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003470};
3471
3472struct loc_track {
3473 unsigned long max;
3474 unsigned long count;
3475 struct location *loc;
3476};
3477
3478static void free_loc_track(struct loc_track *t)
3479{
3480 if (t->max)
3481 free_pages((unsigned long)t->loc,
3482 get_order(sizeof(struct location) * t->max));
3483}
3484
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003485static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003486{
3487 struct location *l;
3488 int order;
3489
Christoph Lameter88a420e2007-05-06 14:49:45 -07003490 order = get_order(sizeof(struct location) * max);
3491
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003492 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003493 if (!l)
3494 return 0;
3495
3496 if (t->count) {
3497 memcpy(l, t->loc, sizeof(struct location) * t->count);
3498 free_loc_track(t);
3499 }
3500 t->max = max;
3501 t->loc = l;
3502 return 1;
3503}
3504
3505static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003506 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003507{
3508 long start, end, pos;
3509 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003510 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003511 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003512
3513 start = -1;
3514 end = t->count;
3515
3516 for ( ; ; ) {
3517 pos = start + (end - start + 1) / 2;
3518
3519 /*
3520 * There is nothing at "end". If we end up there
3521 * we need to add something to before end.
3522 */
3523 if (pos == end)
3524 break;
3525
3526 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003527 if (track->addr == caddr) {
3528
3529 l = &t->loc[pos];
3530 l->count++;
3531 if (track->when) {
3532 l->sum_time += age;
3533 if (age < l->min_time)
3534 l->min_time = age;
3535 if (age > l->max_time)
3536 l->max_time = age;
3537
3538 if (track->pid < l->min_pid)
3539 l->min_pid = track->pid;
3540 if (track->pid > l->max_pid)
3541 l->max_pid = track->pid;
3542
Rusty Russell174596a2009-01-01 10:12:29 +10303543 cpumask_set_cpu(track->cpu,
3544 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003545 }
3546 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003547 return 1;
3548 }
3549
Christoph Lameter45edfa52007-05-09 02:32:45 -07003550 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003551 end = pos;
3552 else
3553 start = pos;
3554 }
3555
3556 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003557 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003558 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003559 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003560 return 0;
3561
3562 l = t->loc + pos;
3563 if (pos < t->count)
3564 memmove(l + 1, l,
3565 (t->count - pos) * sizeof(struct location));
3566 t->count++;
3567 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003568 l->addr = track->addr;
3569 l->sum_time = age;
3570 l->min_time = age;
3571 l->max_time = age;
3572 l->min_pid = track->pid;
3573 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303574 cpumask_clear(to_cpumask(l->cpus));
3575 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003576 nodes_clear(l->nodes);
3577 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003578 return 1;
3579}
3580
3581static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003582 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003583 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003584{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003585 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003586 void *p;
3587
Christoph Lameter39b26462008-04-14 19:11:30 +03003588 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003589 for_each_free_object(p, s, page->freelist)
3590 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003591
Christoph Lameter224a88b2008-04-14 19:11:31 +03003592 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003593 if (!test_bit(slab_index(p, s, addr), map))
3594 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003595}
3596
3597static int list_locations(struct kmem_cache *s, char *buf,
3598 enum track_item alloc)
3599{
Harvey Harrisone374d482008-01-31 15:20:50 -08003600 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003601 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003602 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003603 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003604 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3605 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003606
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003607 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3608 GFP_TEMPORARY)) {
3609 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003610 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003611 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003612 /* Push back cpu slabs */
3613 flush_all(s);
3614
Christoph Lameterf64dc582007-10-16 01:25:33 -07003615 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003616 struct kmem_cache_node *n = get_node(s, node);
3617 unsigned long flags;
3618 struct page *page;
3619
Christoph Lameter9e869432007-08-22 14:01:56 -07003620 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003621 continue;
3622
3623 spin_lock_irqsave(&n->list_lock, flags);
3624 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003625 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003626 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003627 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003628 spin_unlock_irqrestore(&n->list_lock, flags);
3629 }
3630
3631 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003632 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003633
Hugh Dickins9c246242008-12-09 13:14:27 -08003634 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003635 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003636 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003637
3638 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08003639 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003640 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003641 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003642
3643 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003644 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003645 l->min_time,
3646 (long)div_u64(l->sum_time, l->count),
3647 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003648 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003649 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003650 l->min_time);
3651
3652 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003653 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003654 l->min_pid, l->max_pid);
3655 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003656 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003657 l->min_pid);
3658
Rusty Russell174596a2009-01-01 10:12:29 +10303659 if (num_online_cpus() > 1 &&
3660 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003661 len < PAGE_SIZE - 60) {
3662 len += sprintf(buf + len, " cpus=");
3663 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303664 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003665 }
3666
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003667 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003668 len < PAGE_SIZE - 60) {
3669 len += sprintf(buf + len, " nodes=");
3670 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003671 l->nodes);
3672 }
3673
Harvey Harrisone374d482008-01-31 15:20:50 -08003674 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003675 }
3676
3677 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003678 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003679 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003680 len += sprintf(buf, "No data\n");
3681 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003682}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003683#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07003684
Christoph Lametera5a84752010-10-05 13:57:27 -05003685#ifdef SLUB_RESILIENCY_TEST
3686static void resiliency_test(void)
3687{
3688 u8 *p;
3689
3690 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3691
3692 printk(KERN_ERR "SLUB resiliency testing\n");
3693 printk(KERN_ERR "-----------------------\n");
3694 printk(KERN_ERR "A. Corruption after allocation\n");
3695
3696 p = kzalloc(16, GFP_KERNEL);
3697 p[16] = 0x12;
3698 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3699 " 0x12->0x%p\n\n", p + 16);
3700
3701 validate_slab_cache(kmalloc_caches[4]);
3702
3703 /* Hmmm... The next two are dangerous */
3704 p = kzalloc(32, GFP_KERNEL);
3705 p[32 + sizeof(void *)] = 0x34;
3706 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3707 " 0x34 -> -0x%p\n", p);
3708 printk(KERN_ERR
3709 "If allocated object is overwritten then not detectable\n\n");
3710
3711 validate_slab_cache(kmalloc_caches[5]);
3712 p = kzalloc(64, GFP_KERNEL);
3713 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3714 *p = 0x56;
3715 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3716 p);
3717 printk(KERN_ERR
3718 "If allocated object is overwritten then not detectable\n\n");
3719 validate_slab_cache(kmalloc_caches[6]);
3720
3721 printk(KERN_ERR "\nB. Corruption after free\n");
3722 p = kzalloc(128, GFP_KERNEL);
3723 kfree(p);
3724 *p = 0x78;
3725 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3726 validate_slab_cache(kmalloc_caches[7]);
3727
3728 p = kzalloc(256, GFP_KERNEL);
3729 kfree(p);
3730 p[50] = 0x9a;
3731 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3732 p);
3733 validate_slab_cache(kmalloc_caches[8]);
3734
3735 p = kzalloc(512, GFP_KERNEL);
3736 kfree(p);
3737 p[512] = 0xab;
3738 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3739 validate_slab_cache(kmalloc_caches[9]);
3740}
3741#else
3742#ifdef CONFIG_SYSFS
3743static void resiliency_test(void) {};
3744#endif
3745#endif
3746
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003747#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07003748enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003749 SL_ALL, /* All slabs */
3750 SL_PARTIAL, /* Only partially allocated slabs */
3751 SL_CPU, /* Only slabs used for cpu caches */
3752 SL_OBJECTS, /* Determine allocated objects not slabs */
3753 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003754};
3755
Christoph Lameter205ab992008-04-14 19:11:40 +03003756#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003757#define SO_PARTIAL (1 << SL_PARTIAL)
3758#define SO_CPU (1 << SL_CPU)
3759#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003760#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003761
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003762static ssize_t show_slab_objects(struct kmem_cache *s,
3763 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003764{
3765 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003766 int node;
3767 int x;
3768 unsigned long *nodes;
3769 unsigned long *per_cpu;
3770
3771 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003772 if (!nodes)
3773 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003774 per_cpu = nodes + nr_node_ids;
3775
Christoph Lameter205ab992008-04-14 19:11:40 +03003776 if (flags & SO_CPU) {
3777 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003778
Christoph Lameter205ab992008-04-14 19:11:40 +03003779 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003780 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003781
Christoph Lameter205ab992008-04-14 19:11:40 +03003782 if (!c || c->node < 0)
3783 continue;
3784
3785 if (c->page) {
3786 if (flags & SO_TOTAL)
3787 x = c->page->objects;
3788 else if (flags & SO_OBJECTS)
3789 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003790 else
3791 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003792
Christoph Lameter81819f02007-05-06 14:49:36 -07003793 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003794 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003795 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003796 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003797 }
3798 }
3799
Christoph Lameter04d94872011-01-10 10:15:15 -06003800 lock_memory_hotplug();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003801#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003802 if (flags & SO_ALL) {
3803 for_each_node_state(node, N_NORMAL_MEMORY) {
3804 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003805
Christoph Lameter205ab992008-04-14 19:11:40 +03003806 if (flags & SO_TOTAL)
3807 x = atomic_long_read(&n->total_objects);
3808 else if (flags & SO_OBJECTS)
3809 x = atomic_long_read(&n->total_objects) -
3810 count_partial(n, count_free);
3811
3812 else
3813 x = atomic_long_read(&n->nr_slabs);
3814 total += x;
3815 nodes[node] += x;
3816 }
3817
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003818 } else
3819#endif
3820 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03003821 for_each_node_state(node, N_NORMAL_MEMORY) {
3822 struct kmem_cache_node *n = get_node(s, node);
3823
3824 if (flags & SO_TOTAL)
3825 x = count_partial(n, count_total);
3826 else if (flags & SO_OBJECTS)
3827 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003828 else
3829 x = n->nr_partial;
3830 total += x;
3831 nodes[node] += x;
3832 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003833 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003834 x = sprintf(buf, "%lu", total);
3835#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003836 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003837 if (nodes[node])
3838 x += sprintf(buf + x, " N%d=%lu",
3839 node, nodes[node]);
3840#endif
Christoph Lameter04d94872011-01-10 10:15:15 -06003841 unlock_memory_hotplug();
Christoph Lameter81819f02007-05-06 14:49:36 -07003842 kfree(nodes);
3843 return x + sprintf(buf + x, "\n");
3844}
3845
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003846#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003847static int any_slab_objects(struct kmem_cache *s)
3848{
3849 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003850
3851 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003852 struct kmem_cache_node *n = get_node(s, node);
3853
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003854 if (!n)
3855 continue;
3856
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003857 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003858 return 1;
3859 }
3860 return 0;
3861}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003862#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003863
3864#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3865#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3866
3867struct slab_attribute {
3868 struct attribute attr;
3869 ssize_t (*show)(struct kmem_cache *s, char *buf);
3870 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3871};
3872
3873#define SLAB_ATTR_RO(_name) \
3874 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3875
3876#define SLAB_ATTR(_name) \
3877 static struct slab_attribute _name##_attr = \
3878 __ATTR(_name, 0644, _name##_show, _name##_store)
3879
Christoph Lameter81819f02007-05-06 14:49:36 -07003880static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3881{
3882 return sprintf(buf, "%d\n", s->size);
3883}
3884SLAB_ATTR_RO(slab_size);
3885
3886static ssize_t align_show(struct kmem_cache *s, char *buf)
3887{
3888 return sprintf(buf, "%d\n", s->align);
3889}
3890SLAB_ATTR_RO(align);
3891
3892static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3893{
3894 return sprintf(buf, "%d\n", s->objsize);
3895}
3896SLAB_ATTR_RO(object_size);
3897
3898static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3899{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003900 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003901}
3902SLAB_ATTR_RO(objs_per_slab);
3903
Christoph Lameter06b285d2008-04-14 19:11:41 +03003904static ssize_t order_store(struct kmem_cache *s,
3905 const char *buf, size_t length)
3906{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003907 unsigned long order;
3908 int err;
3909
3910 err = strict_strtoul(buf, 10, &order);
3911 if (err)
3912 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003913
3914 if (order > slub_max_order || order < slub_min_order)
3915 return -EINVAL;
3916
3917 calculate_sizes(s, order);
3918 return length;
3919}
3920
Christoph Lameter81819f02007-05-06 14:49:36 -07003921static ssize_t order_show(struct kmem_cache *s, char *buf)
3922{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003923 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003924}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003925SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003926
David Rientjes73d342b2009-02-22 17:40:09 -08003927static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3928{
3929 return sprintf(buf, "%lu\n", s->min_partial);
3930}
3931
3932static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3933 size_t length)
3934{
3935 unsigned long min;
3936 int err;
3937
3938 err = strict_strtoul(buf, 10, &min);
3939 if (err)
3940 return err;
3941
David Rientjesc0bdb232009-02-25 09:16:35 +02003942 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003943 return length;
3944}
3945SLAB_ATTR(min_partial);
3946
Christoph Lameter81819f02007-05-06 14:49:36 -07003947static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3948{
Joe Perches62c70bc2011-01-13 15:45:52 -08003949 if (!s->ctor)
3950 return 0;
3951 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003952}
3953SLAB_ATTR_RO(ctor);
3954
Christoph Lameter81819f02007-05-06 14:49:36 -07003955static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3956{
3957 return sprintf(buf, "%d\n", s->refcount - 1);
3958}
3959SLAB_ATTR_RO(aliases);
3960
Christoph Lameter81819f02007-05-06 14:49:36 -07003961static ssize_t partial_show(struct kmem_cache *s, char *buf)
3962{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003963 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003964}
3965SLAB_ATTR_RO(partial);
3966
3967static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3968{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003969 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003970}
3971SLAB_ATTR_RO(cpu_slabs);
3972
3973static ssize_t objects_show(struct kmem_cache *s, char *buf)
3974{
Christoph Lameter205ab992008-04-14 19:11:40 +03003975 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07003976}
3977SLAB_ATTR_RO(objects);
3978
Christoph Lameter205ab992008-04-14 19:11:40 +03003979static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
3980{
3981 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
3982}
3983SLAB_ATTR_RO(objects_partial);
3984
Christoph Lameter81819f02007-05-06 14:49:36 -07003985static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
3986{
3987 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
3988}
3989
3990static ssize_t reclaim_account_store(struct kmem_cache *s,
3991 const char *buf, size_t length)
3992{
3993 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
3994 if (buf[0] == '1')
3995 s->flags |= SLAB_RECLAIM_ACCOUNT;
3996 return length;
3997}
3998SLAB_ATTR(reclaim_account);
3999
4000static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4001{
Christoph Lameter5af60832007-05-06 14:49:56 -07004002 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004003}
4004SLAB_ATTR_RO(hwcache_align);
4005
4006#ifdef CONFIG_ZONE_DMA
4007static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4008{
4009 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4010}
4011SLAB_ATTR_RO(cache_dma);
4012#endif
4013
4014static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4015{
4016 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4017}
4018SLAB_ATTR_RO(destroy_by_rcu);
4019
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004020#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004021static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4022{
4023 return show_slab_objects(s, buf, SO_ALL);
4024}
4025SLAB_ATTR_RO(slabs);
4026
4027static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4028{
4029 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4030}
4031SLAB_ATTR_RO(total_objects);
4032
4033static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4034{
4035 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4036}
4037
4038static ssize_t sanity_checks_store(struct kmem_cache *s,
4039 const char *buf, size_t length)
4040{
4041 s->flags &= ~SLAB_DEBUG_FREE;
4042 if (buf[0] == '1')
4043 s->flags |= SLAB_DEBUG_FREE;
4044 return length;
4045}
4046SLAB_ATTR(sanity_checks);
4047
4048static ssize_t trace_show(struct kmem_cache *s, char *buf)
4049{
4050 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4051}
4052
4053static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4054 size_t length)
4055{
4056 s->flags &= ~SLAB_TRACE;
4057 if (buf[0] == '1')
4058 s->flags |= SLAB_TRACE;
4059 return length;
4060}
4061SLAB_ATTR(trace);
4062
Christoph Lameter81819f02007-05-06 14:49:36 -07004063static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4064{
4065 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4066}
4067
4068static ssize_t red_zone_store(struct kmem_cache *s,
4069 const char *buf, size_t length)
4070{
4071 if (any_slab_objects(s))
4072 return -EBUSY;
4073
4074 s->flags &= ~SLAB_RED_ZONE;
4075 if (buf[0] == '1')
4076 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004077 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004078 return length;
4079}
4080SLAB_ATTR(red_zone);
4081
4082static ssize_t poison_show(struct kmem_cache *s, char *buf)
4083{
4084 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4085}
4086
4087static ssize_t poison_store(struct kmem_cache *s,
4088 const char *buf, size_t length)
4089{
4090 if (any_slab_objects(s))
4091 return -EBUSY;
4092
4093 s->flags &= ~SLAB_POISON;
4094 if (buf[0] == '1')
4095 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004096 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004097 return length;
4098}
4099SLAB_ATTR(poison);
4100
4101static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4102{
4103 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4104}
4105
4106static ssize_t store_user_store(struct kmem_cache *s,
4107 const char *buf, size_t length)
4108{
4109 if (any_slab_objects(s))
4110 return -EBUSY;
4111
4112 s->flags &= ~SLAB_STORE_USER;
4113 if (buf[0] == '1')
4114 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004115 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004116 return length;
4117}
4118SLAB_ATTR(store_user);
4119
Christoph Lameter53e15af2007-05-06 14:49:43 -07004120static ssize_t validate_show(struct kmem_cache *s, char *buf)
4121{
4122 return 0;
4123}
4124
4125static ssize_t validate_store(struct kmem_cache *s,
4126 const char *buf, size_t length)
4127{
Christoph Lameter434e2452007-07-17 04:03:30 -07004128 int ret = -EINVAL;
4129
4130 if (buf[0] == '1') {
4131 ret = validate_slab_cache(s);
4132 if (ret >= 0)
4133 ret = length;
4134 }
4135 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004136}
4137SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004138
4139static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4140{
4141 if (!(s->flags & SLAB_STORE_USER))
4142 return -ENOSYS;
4143 return list_locations(s, buf, TRACK_ALLOC);
4144}
4145SLAB_ATTR_RO(alloc_calls);
4146
4147static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4148{
4149 if (!(s->flags & SLAB_STORE_USER))
4150 return -ENOSYS;
4151 return list_locations(s, buf, TRACK_FREE);
4152}
4153SLAB_ATTR_RO(free_calls);
4154#endif /* CONFIG_SLUB_DEBUG */
4155
4156#ifdef CONFIG_FAILSLAB
4157static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4158{
4159 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4160}
4161
4162static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4163 size_t length)
4164{
4165 s->flags &= ~SLAB_FAILSLAB;
4166 if (buf[0] == '1')
4167 s->flags |= SLAB_FAILSLAB;
4168 return length;
4169}
4170SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004171#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004172
Christoph Lameter2086d262007-05-06 14:49:46 -07004173static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4174{
4175 return 0;
4176}
4177
4178static ssize_t shrink_store(struct kmem_cache *s,
4179 const char *buf, size_t length)
4180{
4181 if (buf[0] == '1') {
4182 int rc = kmem_cache_shrink(s);
4183
4184 if (rc)
4185 return rc;
4186 } else
4187 return -EINVAL;
4188 return length;
4189}
4190SLAB_ATTR(shrink);
4191
Christoph Lameter81819f02007-05-06 14:49:36 -07004192#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004193static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004194{
Christoph Lameter98246012008-01-07 23:20:26 -08004195 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004196}
4197
Christoph Lameter98246012008-01-07 23:20:26 -08004198static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004199 const char *buf, size_t length)
4200{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004201 unsigned long ratio;
4202 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004203
Christoph Lameter0121c6192008-04-29 16:11:12 -07004204 err = strict_strtoul(buf, 10, &ratio);
4205 if (err)
4206 return err;
4207
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004208 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004209 s->remote_node_defrag_ratio = ratio * 10;
4210
Christoph Lameter81819f02007-05-06 14:49:36 -07004211 return length;
4212}
Christoph Lameter98246012008-01-07 23:20:26 -08004213SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004214#endif
4215
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004216#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004217static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4218{
4219 unsigned long sum = 0;
4220 int cpu;
4221 int len;
4222 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4223
4224 if (!data)
4225 return -ENOMEM;
4226
4227 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004228 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004229
4230 data[cpu] = x;
4231 sum += x;
4232 }
4233
4234 len = sprintf(buf, "%lu", sum);
4235
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004236#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004237 for_each_online_cpu(cpu) {
4238 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004239 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004240 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004241#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004242 kfree(data);
4243 return len + sprintf(buf + len, "\n");
4244}
4245
David Rientjes78eb00c2009-10-15 02:20:22 -07004246static void clear_stat(struct kmem_cache *s, enum stat_item si)
4247{
4248 int cpu;
4249
4250 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004251 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004252}
4253
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004254#define STAT_ATTR(si, text) \
4255static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4256{ \
4257 return show_stat(s, buf, si); \
4258} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004259static ssize_t text##_store(struct kmem_cache *s, \
4260 const char *buf, size_t length) \
4261{ \
4262 if (buf[0] != '0') \
4263 return -EINVAL; \
4264 clear_stat(s, si); \
4265 return length; \
4266} \
4267SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004268
4269STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4270STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4271STAT_ATTR(FREE_FASTPATH, free_fastpath);
4272STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4273STAT_ATTR(FREE_FROZEN, free_frozen);
4274STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4275STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4276STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4277STAT_ATTR(ALLOC_SLAB, alloc_slab);
4278STAT_ATTR(ALLOC_REFILL, alloc_refill);
4279STAT_ATTR(FREE_SLAB, free_slab);
4280STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4281STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4282STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4283STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4284STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4285STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004286STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004287#endif
4288
Pekka Enberg06428782008-01-07 23:20:27 -08004289static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004290 &slab_size_attr.attr,
4291 &object_size_attr.attr,
4292 &objs_per_slab_attr.attr,
4293 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004294 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004295 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004296 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004297 &partial_attr.attr,
4298 &cpu_slabs_attr.attr,
4299 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004300 &aliases_attr.attr,
4301 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004302 &hwcache_align_attr.attr,
4303 &reclaim_account_attr.attr,
4304 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004305 &shrink_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004306#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004307 &total_objects_attr.attr,
4308 &slabs_attr.attr,
4309 &sanity_checks_attr.attr,
4310 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004311 &red_zone_attr.attr,
4312 &poison_attr.attr,
4313 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004314 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004315 &alloc_calls_attr.attr,
4316 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004317#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004318#ifdef CONFIG_ZONE_DMA
4319 &cache_dma_attr.attr,
4320#endif
4321#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004322 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004323#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004324#ifdef CONFIG_SLUB_STATS
4325 &alloc_fastpath_attr.attr,
4326 &alloc_slowpath_attr.attr,
4327 &free_fastpath_attr.attr,
4328 &free_slowpath_attr.attr,
4329 &free_frozen_attr.attr,
4330 &free_add_partial_attr.attr,
4331 &free_remove_partial_attr.attr,
4332 &alloc_from_partial_attr.attr,
4333 &alloc_slab_attr.attr,
4334 &alloc_refill_attr.attr,
4335 &free_slab_attr.attr,
4336 &cpuslab_flush_attr.attr,
4337 &deactivate_full_attr.attr,
4338 &deactivate_empty_attr.attr,
4339 &deactivate_to_head_attr.attr,
4340 &deactivate_to_tail_attr.attr,
4341 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004342 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004343#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004344#ifdef CONFIG_FAILSLAB
4345 &failslab_attr.attr,
4346#endif
4347
Christoph Lameter81819f02007-05-06 14:49:36 -07004348 NULL
4349};
4350
4351static struct attribute_group slab_attr_group = {
4352 .attrs = slab_attrs,
4353};
4354
4355static ssize_t slab_attr_show(struct kobject *kobj,
4356 struct attribute *attr,
4357 char *buf)
4358{
4359 struct slab_attribute *attribute;
4360 struct kmem_cache *s;
4361 int err;
4362
4363 attribute = to_slab_attr(attr);
4364 s = to_slab(kobj);
4365
4366 if (!attribute->show)
4367 return -EIO;
4368
4369 err = attribute->show(s, buf);
4370
4371 return err;
4372}
4373
4374static ssize_t slab_attr_store(struct kobject *kobj,
4375 struct attribute *attr,
4376 const char *buf, size_t len)
4377{
4378 struct slab_attribute *attribute;
4379 struct kmem_cache *s;
4380 int err;
4381
4382 attribute = to_slab_attr(attr);
4383 s = to_slab(kobj);
4384
4385 if (!attribute->store)
4386 return -EIO;
4387
4388 err = attribute->store(s, buf, len);
4389
4390 return err;
4391}
4392
Christoph Lameter151c6022008-01-07 22:29:05 -08004393static void kmem_cache_release(struct kobject *kobj)
4394{
4395 struct kmem_cache *s = to_slab(kobj);
4396
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004397 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004398 kfree(s);
4399}
4400
Emese Revfy52cf25d2010-01-19 02:58:23 +01004401static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004402 .show = slab_attr_show,
4403 .store = slab_attr_store,
4404};
4405
4406static struct kobj_type slab_ktype = {
4407 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004408 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004409};
4410
4411static int uevent_filter(struct kset *kset, struct kobject *kobj)
4412{
4413 struct kobj_type *ktype = get_ktype(kobj);
4414
4415 if (ktype == &slab_ktype)
4416 return 1;
4417 return 0;
4418}
4419
Emese Revfy9cd43612009-12-31 14:52:51 +01004420static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004421 .filter = uevent_filter,
4422};
4423
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004424static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004425
4426#define ID_STR_LENGTH 64
4427
4428/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004429 *
4430 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004431 */
4432static char *create_unique_id(struct kmem_cache *s)
4433{
4434 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4435 char *p = name;
4436
4437 BUG_ON(!name);
4438
4439 *p++ = ':';
4440 /*
4441 * First flags affecting slabcache operations. We will only
4442 * get here for aliasable slabs so we do not need to support
4443 * too many flags. The flags here must cover all flags that
4444 * are matched during merging to guarantee that the id is
4445 * unique.
4446 */
4447 if (s->flags & SLAB_CACHE_DMA)
4448 *p++ = 'd';
4449 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4450 *p++ = 'a';
4451 if (s->flags & SLAB_DEBUG_FREE)
4452 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004453 if (!(s->flags & SLAB_NOTRACK))
4454 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004455 if (p != name + 1)
4456 *p++ = '-';
4457 p += sprintf(p, "%07d", s->size);
4458 BUG_ON(p > name + ID_STR_LENGTH - 1);
4459 return name;
4460}
4461
4462static int sysfs_slab_add(struct kmem_cache *s)
4463{
4464 int err;
4465 const char *name;
4466 int unmergeable;
4467
4468 if (slab_state < SYSFS)
4469 /* Defer until later */
4470 return 0;
4471
4472 unmergeable = slab_unmergeable(s);
4473 if (unmergeable) {
4474 /*
4475 * Slabcache can never be merged so we can use the name proper.
4476 * This is typically the case for debug situations. In that
4477 * case we can catch duplicate names easily.
4478 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004479 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004480 name = s->name;
4481 } else {
4482 /*
4483 * Create a unique name for the slab as a target
4484 * for the symlinks.
4485 */
4486 name = create_unique_id(s);
4487 }
4488
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004489 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004490 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4491 if (err) {
4492 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004493 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004494 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004495
4496 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004497 if (err) {
4498 kobject_del(&s->kobj);
4499 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004500 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004501 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004502 kobject_uevent(&s->kobj, KOBJ_ADD);
4503 if (!unmergeable) {
4504 /* Setup first alias */
4505 sysfs_slab_alias(s, s->name);
4506 kfree(name);
4507 }
4508 return 0;
4509}
4510
4511static void sysfs_slab_remove(struct kmem_cache *s)
4512{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004513 if (slab_state < SYSFS)
4514 /*
4515 * Sysfs has not been setup yet so no need to remove the
4516 * cache from sysfs.
4517 */
4518 return;
4519
Christoph Lameter81819f02007-05-06 14:49:36 -07004520 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4521 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004522 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004523}
4524
4525/*
4526 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004527 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004528 */
4529struct saved_alias {
4530 struct kmem_cache *s;
4531 const char *name;
4532 struct saved_alias *next;
4533};
4534
Adrian Bunk5af328a2007-07-17 04:03:27 -07004535static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004536
4537static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4538{
4539 struct saved_alias *al;
4540
4541 if (slab_state == SYSFS) {
4542 /*
4543 * If we have a leftover link then remove it.
4544 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004545 sysfs_remove_link(&slab_kset->kobj, name);
4546 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004547 }
4548
4549 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4550 if (!al)
4551 return -ENOMEM;
4552
4553 al->s = s;
4554 al->name = name;
4555 al->next = alias_list;
4556 alias_list = al;
4557 return 0;
4558}
4559
4560static int __init slab_sysfs_init(void)
4561{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004562 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004563 int err;
4564
Christoph Lameter2bce6482010-07-19 11:39:11 -05004565 down_write(&slub_lock);
4566
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004567 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004568 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004569 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004570 printk(KERN_ERR "Cannot register slab subsystem.\n");
4571 return -ENOSYS;
4572 }
4573
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004574 slab_state = SYSFS;
4575
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004576 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004577 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004578 if (err)
4579 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4580 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004581 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004582
4583 while (alias_list) {
4584 struct saved_alias *al = alias_list;
4585
4586 alias_list = alias_list->next;
4587 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004588 if (err)
4589 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4590 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004591 kfree(al);
4592 }
4593
Christoph Lameter2bce6482010-07-19 11:39:11 -05004594 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004595 resiliency_test();
4596 return 0;
4597}
4598
4599__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004600#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004601
4602/*
4603 * The /proc/slabinfo ABI
4604 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004605#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004606static void print_slabinfo_header(struct seq_file *m)
4607{
4608 seq_puts(m, "slabinfo - version: 2.1\n");
4609 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4610 "<objperslab> <pagesperslab>");
4611 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4612 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4613 seq_putc(m, '\n');
4614}
4615
4616static void *s_start(struct seq_file *m, loff_t *pos)
4617{
4618 loff_t n = *pos;
4619
4620 down_read(&slub_lock);
4621 if (!n)
4622 print_slabinfo_header(m);
4623
4624 return seq_list_start(&slab_caches, *pos);
4625}
4626
4627static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4628{
4629 return seq_list_next(p, &slab_caches, pos);
4630}
4631
4632static void s_stop(struct seq_file *m, void *p)
4633{
4634 up_read(&slub_lock);
4635}
4636
4637static int s_show(struct seq_file *m, void *p)
4638{
4639 unsigned long nr_partials = 0;
4640 unsigned long nr_slabs = 0;
4641 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004642 unsigned long nr_objs = 0;
4643 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004644 struct kmem_cache *s;
4645 int node;
4646
4647 s = list_entry(p, struct kmem_cache, list);
4648
4649 for_each_online_node(node) {
4650 struct kmem_cache_node *n = get_node(s, node);
4651
4652 if (!n)
4653 continue;
4654
4655 nr_partials += n->nr_partial;
4656 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004657 nr_objs += atomic_long_read(&n->total_objects);
4658 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004659 }
4660
Christoph Lameter205ab992008-04-14 19:11:40 +03004661 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004662
4663 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004664 nr_objs, s->size, oo_objects(s->oo),
4665 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004666 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4667 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4668 0UL);
4669 seq_putc(m, '\n');
4670 return 0;
4671}
4672
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004673static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004674 .start = s_start,
4675 .next = s_next,
4676 .stop = s_stop,
4677 .show = s_show,
4678};
4679
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004680static int slabinfo_open(struct inode *inode, struct file *file)
4681{
4682 return seq_open(file, &slabinfo_op);
4683}
4684
4685static const struct file_operations proc_slabinfo_operations = {
4686 .open = slabinfo_open,
4687 .read = seq_read,
4688 .llseek = seq_lseek,
4689 .release = seq_release,
4690};
4691
4692static int __init slab_proc_init(void)
4693{
WANG Congcf5d1132009-08-18 19:11:40 +03004694 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004695 return 0;
4696}
4697module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004698#endif /* CONFIG_SLABINFO */