blob: 8fd5401bb0717f3cfc73141bdad1c901e92764a7 [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
31/*
32 * Lock order:
33 * 1. slab_lock(page)
34 * 2. slab->list_lock
35 *
36 * The slab_lock protects operations on the object of a particular
37 * slab and its metadata in the page struct. If the slab lock
38 * has been taken then no allocations nor frees can be performed
39 * on the objects in the slab nor can the slab be added or removed
40 * from the partial or full lists since this would mean modifying
41 * the page_struct of the slab.
42 *
43 * The list_lock protects the partial and full list on each node and
44 * the partial slab counter. If taken then no new slabs may be added or
45 * removed from the lists nor make the number of partial slabs be modified.
46 * (Note that the total number of slabs is an atomic value that may be
47 * modified without taking the list lock).
48 *
49 * The list_lock is a centralized lock and thus we avoid taking it as
50 * much as possible. As long as SLUB does not have to handle partial
51 * slabs, operations can continue without any centralized lock. F.e.
52 * allocating a long series of objects that fill up slabs does not require
53 * the list lock.
54 *
55 * The lock order is sometimes inverted when we are trying to get a slab
56 * off a list. We take the list_lock and then look for a page on the list
57 * to use. While we do that objects in the slabs may be freed. We can
58 * only operate on the slab if we have also taken the slab_lock. So we use
59 * a slab_trylock() on the slab. If trylock was successful then no frees
60 * can occur anymore and we can use the slab for allocations etc. If the
61 * slab_trylock() does not succeed then frees are in progress in the slab and
62 * we must stay away from it for a while since we may cause a bouncing
63 * cacheline if we try to acquire the lock. So go onto the next slab.
64 * If all pages are busy then we may allocate a new slab instead of reusing
65 * a partial slab. A new slab has noone operating on it and thus there is
66 * no danger of cacheline contention.
67 *
68 * Interrupts are disabled during allocation and deallocation in order to
69 * make the slab allocator safe to use in the context of an irq. In addition
70 * interrupts are disabled to ensure that the processor does not change
71 * while handling per_cpu slabs, due to kernel preemption.
72 *
73 * SLUB assigns one slab for allocation to each processor.
74 * Allocations only occur from these slabs called cpu slabs.
75 *
Christoph Lameter672bba32007-05-09 02:32:39 -070076 * Slabs with free elements are kept on a partial list and during regular
77 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070078 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070079 * We track full slabs for debugging purposes though because otherwise we
80 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070081 *
82 * Slabs are freed when they become empty. Teardown and setup is
83 * minimal so we rely on the page allocators per cpu caches for
84 * fast frees and allocs.
85 *
86 * Overloading of page flags that are otherwise used for LRU management.
87 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070088 * PageActive The slab is frozen and exempt from list processing.
89 * This means that the slab is dedicated to a purpose
90 * such as satisfying allocations for a specific
91 * processor. Objects may be freed in the slab while
92 * it is frozen but slab_free will then skip the usual
93 * list operations. It is up to the processor holding
94 * the slab to integrate the slab into the slab lists
95 * when the slab is no longer needed.
96 *
97 * One use of this flag is to mark slabs that are
98 * used for allocations. Then such a slab becomes a cpu
99 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700100 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700101 * free objects in addition to the regular freelist
102 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700103 *
104 * PageError Slab requires special handling due to debug
105 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700106 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700107 */
108
Christoph Lameteraf537b02010-07-09 14:07:14 -0500109#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
110 SLAB_TRACE | SLAB_DEBUG_FREE)
111
112static inline int kmem_cache_debug(struct kmem_cache *s)
113{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700114#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500115 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700116#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500117 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700118#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500119}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120
Christoph Lameter81819f02007-05-06 14:49:36 -0700121/*
122 * Issues still to be resolved:
123 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700124 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
125 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700126 * - Variable sizing of the per node arrays
127 */
128
129/* Enable to test recovery from slab corruption on boot */
130#undef SLUB_RESILIENCY_TEST
131
Christoph Lameter81819f02007-05-06 14:49:36 -0700132/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700133 * Mininum number of partial slabs. These will be left on the partial
134 * lists even if they are empty. kmem_cache_shrink may reclaim them.
135 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800136#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700137
Christoph Lameter2086d262007-05-06 14:49:46 -0700138/*
139 * Maximum number of desirable partial slabs.
140 * The existence of more partial slabs makes kmem_cache_shrink
141 * sort the partial list by the number of objects in the.
142 */
143#define MAX_PARTIAL 10
144
Christoph Lameter81819f02007-05-06 14:49:36 -0700145#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
146 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700147
Christoph Lameter81819f02007-05-06 14:49:36 -0700148/*
David Rientjes3de47212009-07-27 18:30:35 -0700149 * Debugging flags that require metadata to be stored in the slab. These get
150 * disabled when slub_debug=O is used and a cache's min order increases with
151 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700152 */
David Rientjes3de47212009-07-27 18:30:35 -0700153#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700154
155/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700156 * Set of flags that will prevent slab merging
157 */
158#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300159 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
160 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700161
162#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200163 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700164
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400165#define OO_SHIFT 16
166#define OO_MASK ((1 << OO_SHIFT) - 1)
167#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
168
Christoph Lameter81819f02007-05-06 14:49:36 -0700169/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500170#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameter81819f02007-05-06 14:49:36 -0700171
172static int kmem_size = sizeof(struct kmem_cache);
173
174#ifdef CONFIG_SMP
175static struct notifier_block slab_notifier;
176#endif
177
178static enum {
179 DOWN, /* No slab functionality available */
Christoph Lameter51df1142010-08-20 12:37:15 -0500180 PARTIAL, /* Kmem_cache_node works */
Christoph Lameter672bba32007-05-09 02:32:39 -0700181 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700182 SYSFS /* Sysfs up */
183} slab_state = DOWN;
184
185/* A list of all slab caches on the system */
186static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700187static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700188
Christoph Lameter02cbc872007-05-09 02:32:43 -0700189/*
190 * Tracking user of a slab.
191 */
192struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300193 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700194 int cpu; /* Was running on cpu */
195 int pid; /* Pid context */
196 unsigned long when; /* When did the operation occur */
197};
198
199enum track_item { TRACK_ALLOC, TRACK_FREE };
200
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500201#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700202static int sysfs_slab_add(struct kmem_cache *);
203static int sysfs_slab_alias(struct kmem_cache *, const char *);
204static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800205
Christoph Lameter81819f02007-05-06 14:49:36 -0700206#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700207static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
208static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
209 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800210static inline void sysfs_slab_remove(struct kmem_cache *s)
211{
Pekka Enberg84c1cf62010-09-14 23:21:12 +0300212 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -0800213 kfree(s);
214}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800215
Christoph Lameter81819f02007-05-06 14:49:36 -0700216#endif
217
Christoph Lameter84e554e62009-12-18 16:26:23 -0600218static inline void stat(struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800219{
220#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e62009-12-18 16:26:23 -0600221 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800222#endif
223}
224
Christoph Lameter81819f02007-05-06 14:49:36 -0700225/********************************************************************
226 * Core slab cache functions
227 *******************************************************************/
228
229int slab_is_available(void)
230{
231 return slab_state >= UP;
232}
233
234static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
235{
Christoph Lameter81819f02007-05-06 14:49:36 -0700236 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700237}
238
Christoph Lameter6446faa2008-02-15 23:45:26 -0800239/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700240static inline int check_valid_pointer(struct kmem_cache *s,
241 struct page *page, const void *object)
242{
243 void *base;
244
Christoph Lametera973e9d2008-03-01 13:40:44 -0800245 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700246 return 1;
247
Christoph Lametera973e9d2008-03-01 13:40:44 -0800248 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300249 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700250 (object - base) % s->size) {
251 return 0;
252 }
253
254 return 1;
255}
256
Christoph Lameter7656c722007-05-09 02:32:40 -0700257static inline void *get_freepointer(struct kmem_cache *s, void *object)
258{
259 return *(void **)(object + s->offset);
260}
261
262static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
263{
264 *(void **)(object + s->offset) = fp;
265}
266
267/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300268#define for_each_object(__p, __s, __addr, __objects) \
269 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700270 __p += (__s)->size)
271
272/* Scan freelist */
273#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800274 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700275
276/* Determine object index from a given position */
277static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
278{
279 return (p - addr) / s->size;
280}
281
Christoph Lameter834f3d12008-04-14 19:11:31 +0300282static inline struct kmem_cache_order_objects oo_make(int order,
283 unsigned long size)
284{
285 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400286 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300287 };
288
289 return x;
290}
291
292static inline int oo_order(struct kmem_cache_order_objects x)
293{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400294 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300295}
296
297static inline int oo_objects(struct kmem_cache_order_objects x)
298{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400299 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300300}
301
Christoph Lameter41ecc552007-05-09 02:32:44 -0700302#ifdef CONFIG_SLUB_DEBUG
303/*
304 * Debug settings:
305 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700306#ifdef CONFIG_SLUB_DEBUG_ON
307static int slub_debug = DEBUG_DEFAULT_FLAGS;
308#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700309static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700310#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700311
312static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700313static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700314
Christoph Lameter7656c722007-05-09 02:32:40 -0700315/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700316 * Object debugging
317 */
318static void print_section(char *text, u8 *addr, unsigned int length)
319{
320 int i, offset;
321 int newline = 1;
322 char ascii[17];
323
324 ascii[16] = 0;
325
326 for (i = 0; i < length; i++) {
327 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700328 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700329 newline = 0;
330 }
Pekka Enberg06428782008-01-07 23:20:27 -0800331 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700332 offset = i % 16;
333 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
334 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800335 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700336 newline = 1;
337 }
338 }
339 if (!newline) {
340 i %= 16;
341 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800342 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700343 ascii[i] = ' ';
344 i++;
345 }
Pekka Enberg06428782008-01-07 23:20:27 -0800346 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700347 }
348}
349
Christoph Lameter81819f02007-05-06 14:49:36 -0700350static struct track *get_track(struct kmem_cache *s, void *object,
351 enum track_item alloc)
352{
353 struct track *p;
354
355 if (s->offset)
356 p = object + s->offset + sizeof(void *);
357 else
358 p = object + s->inuse;
359
360 return p + alloc;
361}
362
363static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300364 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700365{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900366 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700367
Christoph Lameter81819f02007-05-06 14:49:36 -0700368 if (addr) {
369 p->addr = addr;
370 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400371 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700372 p->when = jiffies;
373 } else
374 memset(p, 0, sizeof(struct track));
375}
376
Christoph Lameter81819f02007-05-06 14:49:36 -0700377static void init_tracking(struct kmem_cache *s, void *object)
378{
Christoph Lameter24922682007-07-17 04:03:18 -0700379 if (!(s->flags & SLAB_STORE_USER))
380 return;
381
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300382 set_track(s, object, TRACK_FREE, 0UL);
383 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700384}
385
386static void print_track(const char *s, struct track *t)
387{
388 if (!t->addr)
389 return;
390
Linus Torvalds7daf7052008-07-14 12:12:53 -0700391 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300392 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700393}
394
Christoph Lameter24922682007-07-17 04:03:18 -0700395static void print_tracking(struct kmem_cache *s, void *object)
396{
397 if (!(s->flags & SLAB_STORE_USER))
398 return;
399
400 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
401 print_track("Freed", get_track(s, object, TRACK_FREE));
402}
403
404static void print_page_info(struct page *page)
405{
Christoph Lameter39b26462008-04-14 19:11:30 +0300406 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
407 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700408
409}
410
411static void slab_bug(struct kmem_cache *s, char *fmt, ...)
412{
413 va_list args;
414 char buf[100];
415
416 va_start(args, fmt);
417 vsnprintf(buf, sizeof(buf), fmt, args);
418 va_end(args);
419 printk(KERN_ERR "========================================"
420 "=====================================\n");
421 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
422 printk(KERN_ERR "----------------------------------------"
423 "-------------------------------------\n\n");
424}
425
426static void slab_fix(struct kmem_cache *s, char *fmt, ...)
427{
428 va_list args;
429 char buf[100];
430
431 va_start(args, fmt);
432 vsnprintf(buf, sizeof(buf), fmt, args);
433 va_end(args);
434 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
435}
436
437static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700438{
439 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800440 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700441
442 print_tracking(s, p);
443
444 print_page_info(page);
445
446 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
447 p, p - addr, get_freepointer(s, p));
448
449 if (p > addr + 16)
450 print_section("Bytes b4", p - 16, 16);
451
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300452 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700453
454 if (s->flags & SLAB_RED_ZONE)
455 print_section("Redzone", p + s->objsize,
456 s->inuse - s->objsize);
457
Christoph Lameter81819f02007-05-06 14:49:36 -0700458 if (s->offset)
459 off = s->offset + sizeof(void *);
460 else
461 off = s->inuse;
462
Christoph Lameter24922682007-07-17 04:03:18 -0700463 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700464 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700465
466 if (off != s->size)
467 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700468 print_section("Padding", p + off, s->size - off);
469
470 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700471}
472
473static void object_err(struct kmem_cache *s, struct page *page,
474 u8 *object, char *reason)
475{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700476 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700477 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700478}
479
Christoph Lameter24922682007-07-17 04:03:18 -0700480static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700481{
482 va_list args;
483 char buf[100];
484
Christoph Lameter24922682007-07-17 04:03:18 -0700485 va_start(args, fmt);
486 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700487 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700488 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700489 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700490 dump_stack();
491}
492
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500493static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700494{
495 u8 *p = object;
496
497 if (s->flags & __OBJECT_POISON) {
498 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800499 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700500 }
501
502 if (s->flags & SLAB_RED_ZONE)
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500503 memset(p + s->objsize, val, s->inuse - s->objsize);
Christoph Lameter81819f02007-05-06 14:49:36 -0700504}
505
Christoph Lameter24922682007-07-17 04:03:18 -0700506static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700507{
508 while (bytes) {
509 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700510 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700511 start++;
512 bytes--;
513 }
Christoph Lameter24922682007-07-17 04:03:18 -0700514 return NULL;
515}
516
517static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
518 void *from, void *to)
519{
520 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
521 memset(from, data, to - from);
522}
523
524static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
525 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800526 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700527{
528 u8 *fault;
529 u8 *end;
530
531 fault = check_bytes(start, value, bytes);
532 if (!fault)
533 return 1;
534
535 end = start + bytes;
536 while (end > fault && end[-1] == value)
537 end--;
538
539 slab_bug(s, "%s overwritten", what);
540 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
541 fault, end - 1, fault[0], value);
542 print_trailer(s, page, object);
543
544 restore_bytes(s, what, value, fault, end);
545 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700546}
547
Christoph Lameter81819f02007-05-06 14:49:36 -0700548/*
549 * Object layout:
550 *
551 * object address
552 * Bytes of the object to be managed.
553 * If the freepointer may overlay the object then the free
554 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700555 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700556 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
557 * 0xa5 (POISON_END)
558 *
559 * object + s->objsize
560 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700561 * Padding is extended by another word if Redzoning is enabled and
562 * objsize == inuse.
563 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700564 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
565 * 0xcc (RED_ACTIVE) for objects in use.
566 *
567 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700568 * Meta data starts here.
569 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700570 * A. Free pointer (if we cannot overwrite object on free)
571 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700572 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800573 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700574 * before the word boundary.
575 *
576 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700577 *
578 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700579 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700580 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700581 * If slabcaches are merged then the objsize and inuse boundaries are mostly
582 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700583 * may be used with merged slabcaches.
584 */
585
Christoph Lameter81819f02007-05-06 14:49:36 -0700586static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
587{
588 unsigned long off = s->inuse; /* The end of info */
589
590 if (s->offset)
591 /* Freepointer is placed after the object. */
592 off += sizeof(void *);
593
594 if (s->flags & SLAB_STORE_USER)
595 /* We also have user information there */
596 off += 2 * sizeof(struct track);
597
598 if (s->size == off)
599 return 1;
600
Christoph Lameter24922682007-07-17 04:03:18 -0700601 return check_bytes_and_report(s, page, p, "Object padding",
602 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700603}
604
Christoph Lameter39b26462008-04-14 19:11:30 +0300605/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700606static int slab_pad_check(struct kmem_cache *s, struct page *page)
607{
Christoph Lameter24922682007-07-17 04:03:18 -0700608 u8 *start;
609 u8 *fault;
610 u8 *end;
611 int length;
612 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700613
614 if (!(s->flags & SLAB_POISON))
615 return 1;
616
Christoph Lametera973e9d2008-03-01 13:40:44 -0800617 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300618 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300619 end = start + length;
620 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700621 if (!remainder)
622 return 1;
623
Christoph Lameter39b26462008-04-14 19:11:30 +0300624 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700625 if (!fault)
626 return 1;
627 while (end > fault && end[-1] == POISON_INUSE)
628 end--;
629
630 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300631 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700632
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200633 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700634 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700635}
636
637static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500638 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700639{
640 u8 *p = object;
641 u8 *endobject = object + s->objsize;
642
643 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700644 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500645 endobject, val, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700646 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700647 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800648 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
649 check_bytes_and_report(s, page, p, "Alignment padding",
650 endobject, POISON_INUSE, s->inuse - s->objsize);
651 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700652 }
653
654 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500655 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700656 (!check_bytes_and_report(s, page, p, "Poison", p,
657 POISON_FREE, s->objsize - 1) ||
658 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800659 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700660 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700661 /*
662 * check_pad_bytes cleans up on its own.
663 */
664 check_pad_bytes(s, page, p);
665 }
666
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500667 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700668 /*
669 * Object and freepointer overlap. Cannot check
670 * freepointer while object is allocated.
671 */
672 return 1;
673
674 /* Check free pointer validity */
675 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
676 object_err(s, page, p, "Freepointer corrupt");
677 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100678 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700679 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700680 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700681 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800682 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700683 return 0;
684 }
685 return 1;
686}
687
688static int check_slab(struct kmem_cache *s, struct page *page)
689{
Christoph Lameter39b26462008-04-14 19:11:30 +0300690 int maxobj;
691
Christoph Lameter81819f02007-05-06 14:49:36 -0700692 VM_BUG_ON(!irqs_disabled());
693
694 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700695 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700696 return 0;
697 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300698
699 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
700 if (page->objects > maxobj) {
701 slab_err(s, page, "objects %u > max %u",
702 s->name, page->objects, maxobj);
703 return 0;
704 }
705 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700706 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300707 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700708 return 0;
709 }
710 /* Slab_pad_check fixes things up after itself */
711 slab_pad_check(s, page);
712 return 1;
713}
714
715/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700716 * Determine if a certain object on a page is on the freelist. Must hold the
717 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700718 */
719static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
720{
721 int nr = 0;
722 void *fp = page->freelist;
723 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300724 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700725
Christoph Lameter39b26462008-04-14 19:11:30 +0300726 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700727 if (fp == search)
728 return 1;
729 if (!check_valid_pointer(s, page, fp)) {
730 if (object) {
731 object_err(s, page, object,
732 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800733 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700734 break;
735 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700736 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800737 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300738 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700739 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700740 return 0;
741 }
742 break;
743 }
744 object = fp;
745 fp = get_freepointer(s, object);
746 nr++;
747 }
748
Christoph Lameter224a88b2008-04-14 19:11:31 +0300749 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400750 if (max_objects > MAX_OBJS_PER_PAGE)
751 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300752
753 if (page->objects != max_objects) {
754 slab_err(s, page, "Wrong number of objects. Found %d but "
755 "should be %d", page->objects, max_objects);
756 page->objects = max_objects;
757 slab_fix(s, "Number of objects adjusted.");
758 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300759 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700760 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300761 "counted were %d", page->inuse, page->objects - nr);
762 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700763 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700764 }
765 return search == NULL;
766}
767
Christoph Lameter0121c6192008-04-29 16:11:12 -0700768static void trace(struct kmem_cache *s, struct page *page, void *object,
769 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700770{
771 if (s->flags & SLAB_TRACE) {
772 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
773 s->name,
774 alloc ? "alloc" : "free",
775 object, page->inuse,
776 page->freelist);
777
778 if (!alloc)
779 print_section("Object", (void *)object, s->objsize);
780
781 dump_stack();
782 }
783}
784
Christoph Lameter643b1132007-05-06 14:49:42 -0700785/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500786 * Hooks for other subsystems that check memory allocations. In a typical
787 * production configuration these hooks all should produce no code at all.
788 */
789static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
790{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500791 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500792 lockdep_trace_alloc(flags);
793 might_sleep_if(flags & __GFP_WAIT);
794
795 return should_failslab(s->objsize, flags, s->flags);
796}
797
798static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
799{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500800 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500801 kmemcheck_slab_alloc(s, flags, object, s->objsize);
802 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
803}
804
805static inline void slab_free_hook(struct kmem_cache *s, void *x)
806{
807 kmemleak_free_recursive(x, s->flags);
808}
809
810static inline void slab_free_hook_irq(struct kmem_cache *s, void *object)
811{
812 kmemcheck_slab_free(s, object, s->objsize);
813 debug_check_no_locks_freed(object, s->objsize);
814 if (!(s->flags & SLAB_DEBUG_OBJECTS))
815 debug_check_no_obj_freed(object, s->objsize);
816}
817
818/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700819 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700820 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700821static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700822{
Christoph Lameter643b1132007-05-06 14:49:42 -0700823 spin_lock(&n->list_lock);
824 list_add(&page->lru, &n->full);
825 spin_unlock(&n->list_lock);
826}
827
828static void remove_full(struct kmem_cache *s, struct page *page)
829{
830 struct kmem_cache_node *n;
831
832 if (!(s->flags & SLAB_STORE_USER))
833 return;
834
835 n = get_node(s, page_to_nid(page));
836
837 spin_lock(&n->list_lock);
838 list_del(&page->lru);
839 spin_unlock(&n->list_lock);
840}
841
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300842/* Tracking of the number of slabs for debugging purposes */
843static inline unsigned long slabs_node(struct kmem_cache *s, int node)
844{
845 struct kmem_cache_node *n = get_node(s, node);
846
847 return atomic_long_read(&n->nr_slabs);
848}
849
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400850static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
851{
852 return atomic_long_read(&n->nr_slabs);
853}
854
Christoph Lameter205ab992008-04-14 19:11:40 +0300855static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300856{
857 struct kmem_cache_node *n = get_node(s, node);
858
859 /*
860 * May be called early in order to allocate a slab for the
861 * kmem_cache_node structure. Solve the chicken-egg
862 * dilemma by deferring the increment of the count during
863 * bootstrap (see early_kmem_cache_node_alloc).
864 */
Christoph Lameter7340cc82010-09-28 08:10:26 -0500865 if (n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300866 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300867 atomic_long_add(objects, &n->total_objects);
868 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300869}
Christoph Lameter205ab992008-04-14 19:11:40 +0300870static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300871{
872 struct kmem_cache_node *n = get_node(s, node);
873
874 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300875 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300876}
877
878/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700879static void setup_object_debug(struct kmem_cache *s, struct page *page,
880 void *object)
881{
882 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
883 return;
884
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500885 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700886 init_tracking(s, object);
887}
888
Christoph Lameter15370662010-08-20 12:37:12 -0500889static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300890 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700891{
892 if (!check_slab(s, page))
893 goto bad;
894
Christoph Lameterd692ef62008-02-15 23:45:24 -0800895 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700896 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700897 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700898 }
899
900 if (!check_valid_pointer(s, page, object)) {
901 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700902 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700903 }
904
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500905 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -0700906 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700907
Christoph Lameter3ec09742007-05-16 22:11:00 -0700908 /* Success perform special debug activities for allocs */
909 if (s->flags & SLAB_STORE_USER)
910 set_track(s, object, TRACK_ALLOC, addr);
911 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500912 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700913 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700914
Christoph Lameter81819f02007-05-06 14:49:36 -0700915bad:
916 if (PageSlab(page)) {
917 /*
918 * If this is a slab page then lets do the best we can
919 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700920 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700921 */
Christoph Lameter24922682007-07-17 04:03:18 -0700922 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300923 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800924 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700925 }
926 return 0;
927}
928
Christoph Lameter15370662010-08-20 12:37:12 -0500929static noinline int free_debug_processing(struct kmem_cache *s,
930 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700931{
932 if (!check_slab(s, page))
933 goto fail;
934
935 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700936 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700937 goto fail;
938 }
939
940 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700941 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700942 goto fail;
943 }
944
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500945 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -0700946 return 0;
947
948 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800949 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700950 slab_err(s, page, "Attempt to free object(0x%p) "
951 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800952 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700953 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700954 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700955 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700956 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800957 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700958 object_err(s, page, object,
959 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700960 goto fail;
961 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700962
963 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700964 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700965 remove_full(s, page);
966 if (s->flags & SLAB_STORE_USER)
967 set_track(s, object, TRACK_FREE, addr);
968 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500969 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700970 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700971
Christoph Lameter81819f02007-05-06 14:49:36 -0700972fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700973 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700974 return 0;
975}
976
Christoph Lameter41ecc552007-05-09 02:32:44 -0700977static int __init setup_slub_debug(char *str)
978{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700979 slub_debug = DEBUG_DEFAULT_FLAGS;
980 if (*str++ != '=' || !*str)
981 /*
982 * No options specified. Switch on full debugging.
983 */
984 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700985
986 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700987 /*
988 * No options but restriction on slabs. This means full
989 * debugging for slabs matching a pattern.
990 */
991 goto check_slabs;
992
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700993 if (tolower(*str) == 'o') {
994 /*
995 * Avoid enabling debugging on caches if its minimum order
996 * would increase as a result.
997 */
998 disable_higher_order_debug = 1;
999 goto out;
1000 }
1001
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001002 slub_debug = 0;
1003 if (*str == '-')
1004 /*
1005 * Switch off all debugging measures.
1006 */
1007 goto out;
1008
1009 /*
1010 * Determine which debug features should be switched on
1011 */
Pekka Enberg06428782008-01-07 23:20:27 -08001012 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001013 switch (tolower(*str)) {
1014 case 'f':
1015 slub_debug |= SLAB_DEBUG_FREE;
1016 break;
1017 case 'z':
1018 slub_debug |= SLAB_RED_ZONE;
1019 break;
1020 case 'p':
1021 slub_debug |= SLAB_POISON;
1022 break;
1023 case 'u':
1024 slub_debug |= SLAB_STORE_USER;
1025 break;
1026 case 't':
1027 slub_debug |= SLAB_TRACE;
1028 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001029 case 'a':
1030 slub_debug |= SLAB_FAILSLAB;
1031 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001032 default:
1033 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001034 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001035 }
1036 }
1037
1038check_slabs:
1039 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001040 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001041out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001042 return 1;
1043}
1044
1045__setup("slub_debug", setup_slub_debug);
1046
Christoph Lameterba0268a2007-09-11 15:24:11 -07001047static unsigned long kmem_cache_flags(unsigned long objsize,
1048 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001049 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001050{
1051 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001052 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001053 */
Christoph Lametere1533622008-02-15 23:45:24 -08001054 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001055 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1056 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001057
1058 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001059}
1060#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001061static inline void setup_object_debug(struct kmem_cache *s,
1062 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001063
Christoph Lameter3ec09742007-05-16 22:11:00 -07001064static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001065 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001066
Christoph Lameter3ec09742007-05-16 22:11:00 -07001067static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001068 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001069
Christoph Lameter41ecc552007-05-09 02:32:44 -07001070static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1071 { return 1; }
1072static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001073 void *object, u8 val) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001074static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001075static inline unsigned long kmem_cache_flags(unsigned long objsize,
1076 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001077 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001078{
1079 return flags;
1080}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001081#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001082
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001083#define disable_higher_order_debug 0
1084
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001085static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1086 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001087static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1088 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001089static inline void inc_slabs_node(struct kmem_cache *s, int node,
1090 int objects) {}
1091static inline void dec_slabs_node(struct kmem_cache *s, int node,
1092 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001093
1094static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1095 { return 0; }
1096
1097static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1098 void *object) {}
1099
1100static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1101
1102static inline void slab_free_hook_irq(struct kmem_cache *s,
1103 void *object) {}
1104
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001105#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001106
Christoph Lameter81819f02007-05-06 14:49:36 -07001107/*
1108 * Slab allocation and freeing
1109 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001110static inline struct page *alloc_slab_page(gfp_t flags, int node,
1111 struct kmem_cache_order_objects oo)
1112{
1113 int order = oo_order(oo);
1114
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001115 flags |= __GFP_NOTRACK;
1116
Christoph Lameter2154a332010-07-09 14:07:10 -05001117 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001118 return alloc_pages(flags, order);
1119 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001120 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001121}
1122
Christoph Lameter81819f02007-05-06 14:49:36 -07001123static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1124{
Pekka Enberg06428782008-01-07 23:20:27 -08001125 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001126 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001127 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001128
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001129 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001130
Pekka Enbergba522702009-06-24 21:59:51 +03001131 /*
1132 * Let the initial higher-order allocation fail under memory pressure
1133 * so we fall-back to the minimum order allocation.
1134 */
1135 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1136
1137 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001138 if (unlikely(!page)) {
1139 oo = s->min;
1140 /*
1141 * Allocation may have failed due to fragmentation.
1142 * Try a lower order alloc if possible
1143 */
1144 page = alloc_slab_page(flags, node, oo);
1145 if (!page)
1146 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001147
Christoph Lameter84e554e62009-12-18 16:26:23 -06001148 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001149 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001150
1151 if (kmemcheck_enabled
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001152 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001153 int pages = 1 << oo_order(oo);
1154
1155 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1156
1157 /*
1158 * Objects from caches that have a constructor don't get
1159 * cleared when they're allocated, so we need to do it here.
1160 */
1161 if (s->ctor)
1162 kmemcheck_mark_uninitialized_pages(page, pages);
1163 else
1164 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001165 }
1166
Christoph Lameter834f3d12008-04-14 19:11:31 +03001167 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001168 mod_zone_page_state(page_zone(page),
1169 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1170 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001171 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001172
1173 return page;
1174}
1175
1176static void setup_object(struct kmem_cache *s, struct page *page,
1177 void *object)
1178{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001179 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001180 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001181 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001182}
1183
1184static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1185{
1186 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001187 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001188 void *last;
1189 void *p;
1190
Christoph Lameter6cb06222007-10-16 01:25:41 -07001191 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001192
Christoph Lameter6cb06222007-10-16 01:25:41 -07001193 page = allocate_slab(s,
1194 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001195 if (!page)
1196 goto out;
1197
Christoph Lameter205ab992008-04-14 19:11:40 +03001198 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001199 page->slab = s;
1200 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001201
1202 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001203
1204 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001205 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001206
1207 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001208 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001209 setup_object(s, page, last);
1210 set_freepointer(s, last, p);
1211 last = p;
1212 }
1213 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001214 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001215
1216 page->freelist = start;
1217 page->inuse = 0;
1218out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001219 return page;
1220}
1221
1222static void __free_slab(struct kmem_cache *s, struct page *page)
1223{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001224 int order = compound_order(page);
1225 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001226
Christoph Lameteraf537b02010-07-09 14:07:14 -05001227 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001228 void *p;
1229
1230 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001231 for_each_object(p, s, page_address(page),
1232 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001233 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001234 }
1235
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001236 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001237
Christoph Lameter81819f02007-05-06 14:49:36 -07001238 mod_zone_page_state(page_zone(page),
1239 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1240 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001241 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001242
Christoph Lameter49bd5222008-04-14 18:52:18 +03001243 __ClearPageSlab(page);
1244 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001245 if (current->reclaim_state)
1246 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001247 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001248}
1249
1250static void rcu_free_slab(struct rcu_head *h)
1251{
1252 struct page *page;
1253
1254 page = container_of((struct list_head *)h, struct page, lru);
1255 __free_slab(page->slab, page);
1256}
1257
1258static void free_slab(struct kmem_cache *s, struct page *page)
1259{
1260 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1261 /*
1262 * RCU free overloads the RCU head over the LRU
1263 */
1264 struct rcu_head *head = (void *)&page->lru;
1265
1266 call_rcu(head, rcu_free_slab);
1267 } else
1268 __free_slab(s, page);
1269}
1270
1271static void discard_slab(struct kmem_cache *s, struct page *page)
1272{
Christoph Lameter205ab992008-04-14 19:11:40 +03001273 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001274 free_slab(s, page);
1275}
1276
1277/*
1278 * Per slab locking using the pagelock
1279 */
1280static __always_inline void slab_lock(struct page *page)
1281{
1282 bit_spin_lock(PG_locked, &page->flags);
1283}
1284
1285static __always_inline void slab_unlock(struct page *page)
1286{
Nick Piggina76d3542008-01-07 23:20:27 -08001287 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001288}
1289
1290static __always_inline int slab_trylock(struct page *page)
1291{
1292 int rc = 1;
1293
1294 rc = bit_spin_trylock(PG_locked, &page->flags);
1295 return rc;
1296}
1297
1298/*
1299 * Management of partially allocated slabs
1300 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001301static void add_partial(struct kmem_cache_node *n,
1302 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001303{
Christoph Lametere95eed52007-05-06 14:49:44 -07001304 spin_lock(&n->list_lock);
1305 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001306 if (tail)
1307 list_add_tail(&page->lru, &n->partial);
1308 else
1309 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001310 spin_unlock(&n->list_lock);
1311}
1312
Christoph Lameter62e346a2010-09-28 08:10:28 -05001313static inline void __remove_partial(struct kmem_cache_node *n,
1314 struct page *page)
1315{
1316 list_del(&page->lru);
1317 n->nr_partial--;
1318}
1319
Christoph Lameter0121c6192008-04-29 16:11:12 -07001320static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001321{
1322 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1323
1324 spin_lock(&n->list_lock);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001325 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001326 spin_unlock(&n->list_lock);
1327}
1328
1329/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001330 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001331 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001332 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001333 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001334static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1335 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001336{
1337 if (slab_trylock(page)) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05001338 __remove_partial(n, page);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001339 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001340 return 1;
1341 }
1342 return 0;
1343}
1344
1345/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001346 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001347 */
1348static struct page *get_partial_node(struct kmem_cache_node *n)
1349{
1350 struct page *page;
1351
1352 /*
1353 * Racy check. If we mistakenly see no partial slabs then we
1354 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001355 * partial slab and there is none available then get_partials()
1356 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001357 */
1358 if (!n || !n->nr_partial)
1359 return NULL;
1360
1361 spin_lock(&n->list_lock);
1362 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001363 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001364 goto out;
1365 page = NULL;
1366out:
1367 spin_unlock(&n->list_lock);
1368 return page;
1369}
1370
1371/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001372 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001373 */
1374static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1375{
1376#ifdef CONFIG_NUMA
1377 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001378 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001379 struct zone *zone;
1380 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001381 struct page *page;
1382
1383 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001384 * The defrag ratio allows a configuration of the tradeoffs between
1385 * inter node defragmentation and node local allocations. A lower
1386 * defrag_ratio increases the tendency to do local allocations
1387 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001388 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001389 * If the defrag_ratio is set to 0 then kmalloc() always
1390 * returns node local objects. If the ratio is higher then kmalloc()
1391 * may return off node objects because partial slabs are obtained
1392 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001393 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001394 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001395 * defrag_ratio = 1000) then every (well almost) allocation will
1396 * first attempt to defrag slab caches on other nodes. This means
1397 * scanning over all nodes to look for partial slabs which may be
1398 * expensive if we do it every time we are trying to find a slab
1399 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001400 */
Christoph Lameter98246012008-01-07 23:20:26 -08001401 if (!s->remote_node_defrag_ratio ||
1402 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001403 return NULL;
1404
Miao Xiec0ff7452010-05-24 14:32:08 -07001405 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001406 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001407 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001408 struct kmem_cache_node *n;
1409
Mel Gorman54a6eb52008-04-28 02:12:16 -07001410 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001411
Mel Gorman54a6eb52008-04-28 02:12:16 -07001412 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001413 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001414 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001415 if (page) {
1416 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001417 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001418 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001419 }
1420 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001421 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001422#endif
1423 return NULL;
1424}
1425
1426/*
1427 * Get a partial page, lock it and return it.
1428 */
1429static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1430{
1431 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001432 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001433
1434 page = get_partial_node(get_node(s, searchnode));
Christoph Lameterbc6488e2010-07-26 10:41:14 -05001435 if (page || node != -1)
Christoph Lameter81819f02007-05-06 14:49:36 -07001436 return page;
1437
1438 return get_any_partial(s, flags);
1439}
1440
1441/*
1442 * Move a page back to the lists.
1443 *
1444 * Must be called with the slab lock held.
1445 *
1446 * On exit the slab lock will have been dropped.
1447 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001448static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Namhyung Kim34789732010-09-29 21:02:14 +09001449 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001450{
Christoph Lametere95eed52007-05-06 14:49:44 -07001451 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1452
Andy Whitcroft8a380822008-07-23 21:27:18 -07001453 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001454 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001455
Christoph Lametera973e9d2008-03-01 13:40:44 -08001456 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001457 add_partial(n, page, tail);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001458 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001459 } else {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001460 stat(s, DEACTIVATE_FULL);
Christoph Lameteraf537b02010-07-09 14:07:14 -05001461 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001462 add_full(n, page);
1463 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001464 slab_unlock(page);
1465 } else {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001466 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001467 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001468 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001469 * Adding an empty slab to the partial slabs in order
1470 * to avoid page allocator overhead. This slab needs
1471 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001472 * so that the others get filled first. That way the
1473 * size of the partial list stays small.
1474 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001475 * kmem_cache_shrink can reclaim any empty slabs from
1476 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001477 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001478 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001479 slab_unlock(page);
1480 } else {
1481 slab_unlock(page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001482 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001483 discard_slab(s, page);
1484 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001485 }
1486}
1487
1488/*
1489 * Remove the cpu slab
1490 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001491static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Namhyung Kim34789732010-09-29 21:02:14 +09001492 __releases(bitlock)
Christoph Lameter81819f02007-05-06 14:49:36 -07001493{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001494 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001495 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001496
Christoph Lameterb773ad72008-03-04 11:10:17 -08001497 if (page->freelist)
Christoph Lameter84e554e62009-12-18 16:26:23 -06001498 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001499 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001500 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001501 * because both freelists are empty. So this is unlikely
1502 * to occur.
1503 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001504 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001505 void **object;
1506
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001507 tail = 0; /* Hot objects. Put the slab first */
1508
Christoph Lameter894b8782007-05-10 03:15:16 -07001509 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001510 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001511 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001512
1513 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001514 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001515 page->freelist = object;
1516 page->inuse--;
1517 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001518 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001519 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001520}
1521
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001522static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001523{
Christoph Lameter84e554e62009-12-18 16:26:23 -06001524 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001525 slab_lock(c->page);
1526 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001527}
1528
1529/*
1530 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001531 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001532 * Called from IPI handler with interrupts disabled.
1533 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001534static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001535{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001536 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001537
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001538 if (likely(c && c->page))
1539 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001540}
1541
1542static void flush_cpu_slab(void *d)
1543{
1544 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001545
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001546 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001547}
1548
1549static void flush_all(struct kmem_cache *s)
1550{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001551 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001552}
1553
1554/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001555 * Check if the objects in a per cpu structure fit numa
1556 * locality expectations.
1557 */
1558static inline int node_match(struct kmem_cache_cpu *c, int node)
1559{
1560#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001561 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001562 return 0;
1563#endif
1564 return 1;
1565}
1566
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001567static int count_free(struct page *page)
1568{
1569 return page->objects - page->inuse;
1570}
1571
1572static unsigned long count_partial(struct kmem_cache_node *n,
1573 int (*get_count)(struct page *))
1574{
1575 unsigned long flags;
1576 unsigned long x = 0;
1577 struct page *page;
1578
1579 spin_lock_irqsave(&n->list_lock, flags);
1580 list_for_each_entry(page, &n->partial, lru)
1581 x += get_count(page);
1582 spin_unlock_irqrestore(&n->list_lock, flags);
1583 return x;
1584}
1585
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001586static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1587{
1588#ifdef CONFIG_SLUB_DEBUG
1589 return atomic_long_read(&n->total_objects);
1590#else
1591 return 0;
1592#endif
1593}
1594
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001595static noinline void
1596slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1597{
1598 int node;
1599
1600 printk(KERN_WARNING
1601 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1602 nid, gfpflags);
1603 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1604 "default order: %d, min order: %d\n", s->name, s->objsize,
1605 s->size, oo_order(s->oo), oo_order(s->min));
1606
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001607 if (oo_order(s->min) > get_order(s->objsize))
1608 printk(KERN_WARNING " %s debugging increased min order, use "
1609 "slub_debug=O to disable.\n", s->name);
1610
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001611 for_each_online_node(node) {
1612 struct kmem_cache_node *n = get_node(s, node);
1613 unsigned long nr_slabs;
1614 unsigned long nr_objs;
1615 unsigned long nr_free;
1616
1617 if (!n)
1618 continue;
1619
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001620 nr_free = count_partial(n, count_free);
1621 nr_slabs = node_nr_slabs(n);
1622 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001623
1624 printk(KERN_WARNING
1625 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1626 node, nr_slabs, nr_objs, nr_free);
1627 }
1628}
1629
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001630/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001631 * Slow path. The lockless freelist is empty or we need to perform
1632 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001633 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001634 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001635 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001636 * Processing is still very fast if new objects have been freed to the
1637 * regular freelist. In that case we simply take over the regular freelist
1638 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001639 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001640 * If that is not working then we fall back to the partial lists. We take the
1641 * first element of the freelist as the object to allocate now and move the
1642 * rest of the freelist to the lockless freelist.
1643 *
1644 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001645 * we need to allocate a new slab. This is the slowest path since it involves
1646 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001647 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001648static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1649 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001650{
Christoph Lameter81819f02007-05-06 14:49:36 -07001651 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001652 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001653
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001654 /* We handle __GFP_ZERO in the caller */
1655 gfpflags &= ~__GFP_ZERO;
1656
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001657 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001658 goto new_slab;
1659
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001660 slab_lock(c->page);
1661 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001662 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001663
Christoph Lameter84e554e62009-12-18 16:26:23 -06001664 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001665
Christoph Lameter894b8782007-05-10 03:15:16 -07001666load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001667 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001668 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001669 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001670 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001671 goto debug;
1672
Christoph Lameterff120592009-12-18 16:26:22 -06001673 c->freelist = get_freepointer(s, object);
Christoph Lameter39b26462008-04-14 19:11:30 +03001674 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001675 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001676 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001677unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001678 slab_unlock(c->page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001679 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001680 return object;
1681
1682another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001683 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001684
1685new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001686 new = get_partial(s, gfpflags, node);
1687 if (new) {
1688 c->page = new;
Christoph Lameter84e554e62009-12-18 16:26:23 -06001689 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001690 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001691 }
1692
Christoph Lameterc1d50832010-08-20 12:37:17 -05001693 gfpflags &= gfp_allowed_mask;
Christoph Lameterb811c202007-10-16 23:25:51 -07001694 if (gfpflags & __GFP_WAIT)
1695 local_irq_enable();
1696
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001697 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001698
1699 if (gfpflags & __GFP_WAIT)
1700 local_irq_disable();
1701
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001702 if (new) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001703 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001704 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001705 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001706 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001707 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001708 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001709 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001710 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001711 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001712 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1713 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001714 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001715debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001716 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001717 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001718
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001719 c->page->inuse++;
Christoph Lameterff120592009-12-18 16:26:22 -06001720 c->page->freelist = get_freepointer(s, object);
Pekka Enberg15b7c512010-10-02 11:32:32 +03001721 c->node = NUMA_NO_NODE;
Christoph Lameter1f842602008-01-07 23:20:30 -08001722 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001723}
1724
1725/*
1726 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1727 * have the fastpath folded into their functions. So no function call
1728 * overhead for requests that can be satisfied on the fastpath.
1729 *
1730 * The fastpath works by first checking if the lockless freelist can be used.
1731 * If not then __slab_alloc is called for slow processing.
1732 *
1733 * Otherwise we can simply pick the next object from the lockless free list.
1734 */
Pekka Enberg06428782008-01-07 23:20:27 -08001735static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001736 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001737{
Christoph Lameter894b8782007-05-10 03:15:16 -07001738 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001739 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001740 unsigned long flags;
1741
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001742 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001743 return NULL;
1744
Christoph Lameter894b8782007-05-10 03:15:16 -07001745 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001746 c = __this_cpu_ptr(s->cpu_slab);
1747 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001748 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001749
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001750 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001751
1752 else {
Christoph Lameterff120592009-12-18 16:26:22 -06001753 c->freelist = get_freepointer(s, object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001754 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001755 }
1756 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001757
Pekka Enberg74e21342009-11-25 20:14:48 +02001758 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001759 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001760
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001761 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001762
Christoph Lameter894b8782007-05-10 03:15:16 -07001763 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001764}
1765
1766void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1767{
Christoph Lameter2154a332010-07-09 14:07:10 -05001768 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001769
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02001770 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001771
1772 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001773}
1774EXPORT_SYMBOL(kmem_cache_alloc);
1775
Li Zefan0f24f122009-12-11 15:45:30 +08001776#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001777void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1778{
Christoph Lameter2154a332010-07-09 14:07:10 -05001779 return slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001780}
1781EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1782#endif
1783
Christoph Lameter81819f02007-05-06 14:49:36 -07001784#ifdef CONFIG_NUMA
1785void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1786{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001787 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1788
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02001789 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1790 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001791
1792 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001793}
1794EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001795
Li Zefan0f24f122009-12-11 15:45:30 +08001796#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001797void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1798 gfp_t gfpflags,
1799 int node)
1800{
1801 return slab_alloc(s, gfpflags, node, _RET_IP_);
1802}
1803EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1804#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09001805#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001806
Christoph Lameter81819f02007-05-06 14:49:36 -07001807/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001808 * Slow patch handling. This may still be called frequently since objects
1809 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001810 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001811 * So we still attempt to reduce cache line usage. Just take the slab
1812 * lock and free the item. If there is no additional partial page
1813 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001814 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001815static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06001816 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001817{
1818 void *prior;
1819 void **object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07001820
Christoph Lameter84e554e62009-12-18 16:26:23 -06001821 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001822 slab_lock(page);
1823
Christoph Lameteraf537b02010-07-09 14:07:14 -05001824 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001825 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001826
Christoph Lameter81819f02007-05-06 14:49:36 -07001827checks_ok:
Christoph Lameterff120592009-12-18 16:26:22 -06001828 prior = page->freelist;
1829 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07001830 page->freelist = object;
1831 page->inuse--;
1832
Andy Whitcroft8a380822008-07-23 21:27:18 -07001833 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001834 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001835 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001836 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001837
1838 if (unlikely(!page->inuse))
1839 goto slab_empty;
1840
1841 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001842 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001843 * then add it.
1844 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001845 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001846 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001847 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001848 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001849
1850out_unlock:
1851 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001852 return;
1853
1854slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001855 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001856 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001857 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001858 */
1859 remove_partial(s, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001860 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001861 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001862 slab_unlock(page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06001863 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001864 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001865 return;
1866
1867debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001868 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001869 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001870 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001871}
1872
Christoph Lameter894b8782007-05-10 03:15:16 -07001873/*
1874 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1875 * can perform fastpath freeing without additional function calls.
1876 *
1877 * The fastpath is only possible if we are freeing to the current cpu slab
1878 * of this processor. This typically the case if we have just allocated
1879 * the item before.
1880 *
1881 * If fastpath is not possible then fall back to __slab_free where we deal
1882 * with all sorts of special processing.
1883 */
Pekka Enberg06428782008-01-07 23:20:27 -08001884static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001885 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001886{
1887 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001888 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001889 unsigned long flags;
1890
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001891 slab_free_hook(s, x);
1892
Christoph Lameter894b8782007-05-10 03:15:16 -07001893 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001894 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001895
1896 slab_free_hook_irq(s, x);
1897
Pekka Enberg15b7c512010-10-02 11:32:32 +03001898 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
Christoph Lameterff120592009-12-18 16:26:22 -06001899 set_freepointer(s, object, c->freelist);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001900 c->freelist = object;
Christoph Lameter84e554e62009-12-18 16:26:23 -06001901 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001902 } else
Christoph Lameterff120592009-12-18 16:26:22 -06001903 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07001904
1905 local_irq_restore(flags);
1906}
1907
Christoph Lameter81819f02007-05-06 14:49:36 -07001908void kmem_cache_free(struct kmem_cache *s, void *x)
1909{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001910 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001911
Christoph Lameterb49af682007-05-06 14:49:41 -07001912 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001913
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001914 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001915
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02001916 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001917}
1918EXPORT_SYMBOL(kmem_cache_free);
1919
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001920/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001921static struct page *get_object_page(const void *x)
1922{
Christoph Lameterb49af682007-05-06 14:49:41 -07001923 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001924
1925 if (!PageSlab(page))
1926 return NULL;
1927
1928 return page;
1929}
1930
1931/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001932 * Object placement in a slab is made very easy because we always start at
1933 * offset 0. If we tune the size of the object to the alignment then we can
1934 * get the required alignment by putting one properly sized object after
1935 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001936 *
1937 * Notice that the allocation order determines the sizes of the per cpu
1938 * caches. Each processor has always one slab available for allocations.
1939 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001940 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001941 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001942 */
1943
1944/*
1945 * Mininum / Maximum order of slab pages. This influences locking overhead
1946 * and slab fragmentation. A higher order reduces the number of partial slabs
1947 * and increases the number of allocations possible without having to
1948 * take the list_lock.
1949 */
1950static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001951static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001952static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001953
1954/*
1955 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001956 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001957 */
1958static int slub_nomerge;
1959
1960/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001961 * Calculate the order of allocation given an slab object size.
1962 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001963 * The order of allocation has significant impact on performance and other
1964 * system components. Generally order 0 allocations should be preferred since
1965 * order 0 does not cause fragmentation in the page allocator. Larger objects
1966 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001967 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001968 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001969 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001970 * In order to reach satisfactory performance we must ensure that a minimum
1971 * number of objects is in one slab. Otherwise we may generate too much
1972 * activity on the partial lists which requires taking the list_lock. This is
1973 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001974 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001975 * slub_max_order specifies the order where we begin to stop considering the
1976 * number of objects in a slab as critical. If we reach slub_max_order then
1977 * we try to keep the page order as low as possible. So we accept more waste
1978 * of space in favor of a small page order.
1979 *
1980 * Higher order allocations also allow the placement of more objects in a
1981 * slab and thereby reduce object handling overhead. If the user has
1982 * requested a higher mininum order then we start with that one instead of
1983 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001984 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001985static inline int slab_order(int size, int min_objects,
1986 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001987{
1988 int order;
1989 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001990 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001991
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001992 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1993 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001994
Christoph Lameter6300ea72007-07-17 04:03:20 -07001995 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001996 fls(min_objects * size - 1) - PAGE_SHIFT);
1997 order <= max_order; order++) {
1998
Christoph Lameter81819f02007-05-06 14:49:36 -07001999 unsigned long slab_size = PAGE_SIZE << order;
2000
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002001 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07002002 continue;
2003
Christoph Lameter81819f02007-05-06 14:49:36 -07002004 rem = slab_size % size;
2005
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002006 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002007 break;
2008
2009 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002010
Christoph Lameter81819f02007-05-06 14:49:36 -07002011 return order;
2012}
2013
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002014static inline int calculate_order(int size)
2015{
2016 int order;
2017 int min_objects;
2018 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002019 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002020
2021 /*
2022 * Attempt to find best configuration for a slab. This
2023 * works by first attempting to generate a layout with
2024 * the best configuration and backing off gradually.
2025 *
2026 * First we reduce the acceptable waste in a slab. Then
2027 * we reduce the minimum objects required in a slab.
2028 */
2029 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002030 if (!min_objects)
2031 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002032 max_objects = (PAGE_SIZE << slub_max_order)/size;
2033 min_objects = min(min_objects, max_objects);
2034
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002035 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002036 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002037 while (fraction >= 4) {
2038 order = slab_order(size, min_objects,
2039 slub_max_order, fraction);
2040 if (order <= slub_max_order)
2041 return order;
2042 fraction /= 2;
2043 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002044 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002045 }
2046
2047 /*
2048 * We were unable to place multiple objects in a slab. Now
2049 * lets see if we can place a single object there.
2050 */
2051 order = slab_order(size, 1, slub_max_order, 1);
2052 if (order <= slub_max_order)
2053 return order;
2054
2055 /*
2056 * Doh this slab cannot be placed using slub_max_order.
2057 */
2058 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002059 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002060 return order;
2061 return -ENOSYS;
2062}
2063
Christoph Lameter81819f02007-05-06 14:49:36 -07002064/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002065 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002066 */
2067static unsigned long calculate_alignment(unsigned long flags,
2068 unsigned long align, unsigned long size)
2069{
2070 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002071 * If the user wants hardware cache aligned objects then follow that
2072 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002073 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002074 * The hardware cache alignment cannot override the specified
2075 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002076 */
Nick Pigginb6210382008-03-05 14:05:56 -08002077 if (flags & SLAB_HWCACHE_ALIGN) {
2078 unsigned long ralign = cache_line_size();
2079 while (size <= ralign / 2)
2080 ralign /= 2;
2081 align = max(align, ralign);
2082 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002083
2084 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002085 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002086
2087 return ALIGN(align, sizeof(void *));
2088}
2089
Pekka Enberg5595cff2008-08-05 09:28:47 +03002090static void
2091init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002092{
2093 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002094 spin_lock_init(&n->list_lock);
2095 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002096#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002097 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002098 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002099 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002100#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002101}
2102
Christoph Lameter55136592010-08-20 12:37:13 -05002103static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002104{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002105 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2106 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002107
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002108 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002109
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002110 return s->cpu_slab != NULL;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002111}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002112
Christoph Lameter51df1142010-08-20 12:37:15 -05002113static struct kmem_cache *kmem_cache_node;
2114
Christoph Lameter81819f02007-05-06 14:49:36 -07002115/*
2116 * No kmalloc_node yet so do it by hand. We know that this is the first
2117 * slab on the node for this slabcache. There are no concurrent accesses
2118 * possible.
2119 *
2120 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002121 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2122 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002123 */
Christoph Lameter55136592010-08-20 12:37:13 -05002124static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002125{
2126 struct page *page;
2127 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002128 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002129
Christoph Lameter51df1142010-08-20 12:37:15 -05002130 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002131
Christoph Lameter51df1142010-08-20 12:37:15 -05002132 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002133
2134 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002135 if (page_to_nid(page) != node) {
2136 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2137 "node %d\n", node);
2138 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2139 "in order to be able to continue\n");
2140 }
2141
Christoph Lameter81819f02007-05-06 14:49:36 -07002142 n = page->freelist;
2143 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002144 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002145 page->inuse++;
Christoph Lameter51df1142010-08-20 12:37:15 -05002146 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002147#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002148 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002149 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002150#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002151 init_kmem_cache_node(n, kmem_cache_node);
2152 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002153
rootba84c732008-01-07 23:20:28 -08002154 /*
2155 * lockdep requires consistent irq usage for each lock
2156 * so even though there cannot be a race this early in
2157 * the boot sequence, we still disable irqs.
2158 */
2159 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002160 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002161 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002162}
2163
2164static void free_kmem_cache_nodes(struct kmem_cache *s)
2165{
2166 int node;
2167
Christoph Lameterf64dc582007-10-16 01:25:33 -07002168 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002169 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002170
Alexander Duyck73367bd2010-05-21 14:41:35 -07002171 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002172 kmem_cache_free(kmem_cache_node, n);
2173
Christoph Lameter81819f02007-05-06 14:49:36 -07002174 s->node[node] = NULL;
2175 }
2176}
2177
Christoph Lameter55136592010-08-20 12:37:13 -05002178static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002179{
2180 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002181
Christoph Lameterf64dc582007-10-16 01:25:33 -07002182 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002183 struct kmem_cache_node *n;
2184
Alexander Duyck73367bd2010-05-21 14:41:35 -07002185 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002186 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002187 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002188 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002189 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002190 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002191
2192 if (!n) {
2193 free_kmem_cache_nodes(s);
2194 return 0;
2195 }
2196
Christoph Lameter81819f02007-05-06 14:49:36 -07002197 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002198 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002199 }
2200 return 1;
2201}
Christoph Lameter81819f02007-05-06 14:49:36 -07002202
David Rientjesc0bdb232009-02-25 09:16:35 +02002203static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002204{
2205 if (min < MIN_PARTIAL)
2206 min = MIN_PARTIAL;
2207 else if (min > MAX_PARTIAL)
2208 min = MAX_PARTIAL;
2209 s->min_partial = min;
2210}
2211
Christoph Lameter81819f02007-05-06 14:49:36 -07002212/*
2213 * calculate_sizes() determines the order and the distribution of data within
2214 * a slab object.
2215 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002216static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002217{
2218 unsigned long flags = s->flags;
2219 unsigned long size = s->objsize;
2220 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002221 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002222
2223 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002224 * Round up object size to the next word boundary. We can only
2225 * place the free pointer at word boundaries and this determines
2226 * the possible location of the free pointer.
2227 */
2228 size = ALIGN(size, sizeof(void *));
2229
2230#ifdef CONFIG_SLUB_DEBUG
2231 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002232 * Determine if we can poison the object itself. If the user of
2233 * the slab may touch the object after free or before allocation
2234 * then we should never poison the object itself.
2235 */
2236 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07002237 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002238 s->flags |= __OBJECT_POISON;
2239 else
2240 s->flags &= ~__OBJECT_POISON;
2241
Christoph Lameter81819f02007-05-06 14:49:36 -07002242
2243 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002244 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002245 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002246 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002247 */
2248 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2249 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002250#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002251
2252 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002253 * With that we have determined the number of bytes in actual use
2254 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002255 */
2256 s->inuse = size;
2257
2258 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07002259 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002260 /*
2261 * Relocate free pointer after the object if it is not
2262 * permitted to overwrite the first word of the object on
2263 * kmem_cache_free.
2264 *
2265 * This is the case if we do RCU, have a constructor or
2266 * destructor or are poisoning the objects.
2267 */
2268 s->offset = size;
2269 size += sizeof(void *);
2270 }
2271
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002272#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002273 if (flags & SLAB_STORE_USER)
2274 /*
2275 * Need to store information about allocs and frees after
2276 * the object.
2277 */
2278 size += 2 * sizeof(struct track);
2279
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002280 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002281 /*
2282 * Add some empty padding so that we can catch
2283 * overwrites from earlier objects rather than let
2284 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002285 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002286 * of the object.
2287 */
2288 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002289#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002290
Christoph Lameter81819f02007-05-06 14:49:36 -07002291 /*
2292 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002293 * user specified and the dynamic determination of cache line size
2294 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002295 */
2296 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002297 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002298
2299 /*
2300 * SLUB stores one object immediately after another beginning from
2301 * offset 0. In order to align the objects we have to simply size
2302 * each object to conform to the alignment.
2303 */
2304 size = ALIGN(size, align);
2305 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002306 if (forced_order >= 0)
2307 order = forced_order;
2308 else
2309 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002310
Christoph Lameter834f3d12008-04-14 19:11:31 +03002311 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002312 return 0;
2313
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002314 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002315 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002316 s->allocflags |= __GFP_COMP;
2317
2318 if (s->flags & SLAB_CACHE_DMA)
2319 s->allocflags |= SLUB_DMA;
2320
2321 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2322 s->allocflags |= __GFP_RECLAIMABLE;
2323
Christoph Lameter81819f02007-05-06 14:49:36 -07002324 /*
2325 * Determine the number of objects per slab
2326 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002327 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002328 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002329 if (oo_objects(s->oo) > oo_objects(s->max))
2330 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002331
Christoph Lameter834f3d12008-04-14 19:11:31 +03002332 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002333
2334}
2335
Christoph Lameter55136592010-08-20 12:37:13 -05002336static int kmem_cache_open(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07002337 const char *name, size_t size,
2338 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002339 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002340{
2341 memset(s, 0, kmem_size);
2342 s->name = name;
2343 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002344 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002345 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002346 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002347
Christoph Lameter06b285d2008-04-14 19:11:41 +03002348 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002349 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002350 if (disable_higher_order_debug) {
2351 /*
2352 * Disable debugging flags that store metadata if the min slab
2353 * order increased.
2354 */
2355 if (get_order(s->size) > get_order(s->objsize)) {
2356 s->flags &= ~DEBUG_METADATA_FLAGS;
2357 s->offset = 0;
2358 if (!calculate_sizes(s, -1))
2359 goto error;
2360 }
2361 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002362
David Rientjes3b89d7d2009-02-22 17:40:07 -08002363 /*
2364 * The larger the object size is, the more pages we want on the partial
2365 * list to avoid pounding the page allocator excessively.
2366 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002367 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002368 s->refcount = 1;
2369#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002370 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002371#endif
Christoph Lameter55136592010-08-20 12:37:13 -05002372 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002373 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002374
Christoph Lameter55136592010-08-20 12:37:13 -05002375 if (alloc_kmem_cache_cpus(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07002376 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002377
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002378 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002379error:
2380 if (flags & SLAB_PANIC)
2381 panic("Cannot create slab %s size=%lu realsize=%u "
2382 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002383 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002384 s->offset, flags);
2385 return 0;
2386}
Christoph Lameter81819f02007-05-06 14:49:36 -07002387
2388/*
2389 * Check if a given pointer is valid
2390 */
2391int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2392{
Pekka Enberg06428782008-01-07 23:20:27 -08002393 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002394
Pekka Enbergd3e06e22010-04-07 19:23:41 +03002395 if (!kern_ptr_validate(object, s->size))
2396 return 0;
2397
Christoph Lameter81819f02007-05-06 14:49:36 -07002398 page = get_object_page(object);
2399
2400 if (!page || s != page->slab)
2401 /* No slab or wrong slab */
2402 return 0;
2403
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002404 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002405 return 0;
2406
2407 /*
2408 * We could also check if the object is on the slabs freelist.
2409 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002410 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002411 * to a certain slab.
2412 */
2413 return 1;
2414}
2415EXPORT_SYMBOL(kmem_ptr_validate);
2416
2417/*
2418 * Determine the size of a slab object
2419 */
2420unsigned int kmem_cache_size(struct kmem_cache *s)
2421{
2422 return s->objsize;
2423}
2424EXPORT_SYMBOL(kmem_cache_size);
2425
2426const char *kmem_cache_name(struct kmem_cache *s)
2427{
2428 return s->name;
2429}
2430EXPORT_SYMBOL(kmem_cache_name);
2431
Christoph Lameter33b12c32008-04-25 12:22:43 -07002432static void list_slab_objects(struct kmem_cache *s, struct page *page,
2433 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002434{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002435#ifdef CONFIG_SLUB_DEBUG
2436 void *addr = page_address(page);
2437 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09002438 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2439 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002440 if (!map)
2441 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002442 slab_err(s, page, "%s", text);
2443 slab_lock(page);
2444 for_each_free_object(p, s, page->freelist)
2445 set_bit(slab_index(p, s, addr), map);
2446
2447 for_each_object(p, s, addr, page->objects) {
2448
2449 if (!test_bit(slab_index(p, s, addr), map)) {
2450 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2451 p, p - addr);
2452 print_tracking(s, p);
2453 }
2454 }
2455 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002456 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002457#endif
2458}
2459
Christoph Lameter81819f02007-05-06 14:49:36 -07002460/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002461 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002462 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002463static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002464{
Christoph Lameter81819f02007-05-06 14:49:36 -07002465 unsigned long flags;
2466 struct page *page, *h;
2467
2468 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002469 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002470 if (!page->inuse) {
Christoph Lameter62e346a2010-09-28 08:10:28 -05002471 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002472 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002473 } else {
2474 list_slab_objects(s, page,
2475 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002476 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002477 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002478 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002479}
2480
2481/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002482 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002483 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002484static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002485{
2486 int node;
2487
2488 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002489 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002490 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002491 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002492 struct kmem_cache_node *n = get_node(s, node);
2493
Christoph Lameter599870b2008-04-23 12:36:52 -07002494 free_partial(s, n);
2495 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002496 return 1;
2497 }
2498 free_kmem_cache_nodes(s);
2499 return 0;
2500}
2501
2502/*
2503 * Close a cache and release the kmem_cache structure
2504 * (must be used for caches created using kmem_cache_create)
2505 */
2506void kmem_cache_destroy(struct kmem_cache *s)
2507{
2508 down_write(&slub_lock);
2509 s->refcount--;
2510 if (!s->refcount) {
2511 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002512 if (kmem_cache_close(s)) {
2513 printk(KERN_ERR "SLUB %s: %s called for cache that "
2514 "still has objects.\n", s->name, __func__);
2515 dump_stack();
2516 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002517 if (s->flags & SLAB_DESTROY_BY_RCU)
2518 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002519 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002520 }
2521 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002522}
2523EXPORT_SYMBOL(kmem_cache_destroy);
2524
2525/********************************************************************
2526 * Kmalloc subsystem
2527 *******************************************************************/
2528
Christoph Lameter51df1142010-08-20 12:37:15 -05002529struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002530EXPORT_SYMBOL(kmalloc_caches);
2531
Christoph Lameter51df1142010-08-20 12:37:15 -05002532static struct kmem_cache *kmem_cache;
2533
Christoph Lameter55136592010-08-20 12:37:13 -05002534#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002535static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002536#endif
2537
Christoph Lameter81819f02007-05-06 14:49:36 -07002538static int __init setup_slub_min_order(char *str)
2539{
Pekka Enberg06428782008-01-07 23:20:27 -08002540 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002541
2542 return 1;
2543}
2544
2545__setup("slub_min_order=", setup_slub_min_order);
2546
2547static int __init setup_slub_max_order(char *str)
2548{
Pekka Enberg06428782008-01-07 23:20:27 -08002549 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002550 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002551
2552 return 1;
2553}
2554
2555__setup("slub_max_order=", setup_slub_max_order);
2556
2557static int __init setup_slub_min_objects(char *str)
2558{
Pekka Enberg06428782008-01-07 23:20:27 -08002559 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002560
2561 return 1;
2562}
2563
2564__setup("slub_min_objects=", setup_slub_min_objects);
2565
2566static int __init setup_slub_nomerge(char *str)
2567{
2568 slub_nomerge = 1;
2569 return 1;
2570}
2571
2572__setup("slub_nomerge", setup_slub_nomerge);
2573
Christoph Lameter51df1142010-08-20 12:37:15 -05002574static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2575 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002576{
Christoph Lameter51df1142010-08-20 12:37:15 -05002577 struct kmem_cache *s;
2578
2579 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2580
Pekka Enberg83b519e2009-06-10 19:40:04 +03002581 /*
2582 * This function is called with IRQs disabled during early-boot on
2583 * single CPU so there's no need to take slub_lock here.
2584 */
Christoph Lameter55136592010-08-20 12:37:13 -05002585 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002586 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002587 goto panic;
2588
2589 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002590 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002591
2592panic:
2593 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002594 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002595}
2596
Christoph Lameterf1b26332007-07-17 04:03:26 -07002597/*
2598 * Conversion table for small slabs sizes / 8 to the index in the
2599 * kmalloc array. This is necessary for slabs < 192 since we have non power
2600 * of two cache sizes there. The size of larger slabs can be determined using
2601 * fls.
2602 */
2603static s8 size_index[24] = {
2604 3, /* 8 */
2605 4, /* 16 */
2606 5, /* 24 */
2607 5, /* 32 */
2608 6, /* 40 */
2609 6, /* 48 */
2610 6, /* 56 */
2611 6, /* 64 */
2612 1, /* 72 */
2613 1, /* 80 */
2614 1, /* 88 */
2615 1, /* 96 */
2616 7, /* 104 */
2617 7, /* 112 */
2618 7, /* 120 */
2619 7, /* 128 */
2620 2, /* 136 */
2621 2, /* 144 */
2622 2, /* 152 */
2623 2, /* 160 */
2624 2, /* 168 */
2625 2, /* 176 */
2626 2, /* 184 */
2627 2 /* 192 */
2628};
2629
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002630static inline int size_index_elem(size_t bytes)
2631{
2632 return (bytes - 1) / 8;
2633}
2634
Christoph Lameter81819f02007-05-06 14:49:36 -07002635static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2636{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002637 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002638
Christoph Lameterf1b26332007-07-17 04:03:26 -07002639 if (size <= 192) {
2640 if (!size)
2641 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002642
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002643 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002644 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002645 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002646
2647#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002648 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002649 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002650
Christoph Lameter81819f02007-05-06 14:49:36 -07002651#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002652 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002653}
2654
2655void *__kmalloc(size_t size, gfp_t flags)
2656{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002657 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002658 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002659
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002660 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002661 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002662
2663 s = get_slab(size, flags);
2664
2665 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002666 return s;
2667
Christoph Lameter2154a332010-07-09 14:07:10 -05002668 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002669
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002670 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002671
2672 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002673}
2674EXPORT_SYMBOL(__kmalloc);
2675
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002676#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002677static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2678{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002679 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002680 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002681
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002682 flags |= __GFP_COMP | __GFP_NOTRACK;
2683 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002684 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002685 ptr = page_address(page);
2686
2687 kmemleak_alloc(ptr, size, 1, flags);
2688 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002689}
2690
Christoph Lameter81819f02007-05-06 14:49:36 -07002691void *__kmalloc_node(size_t size, gfp_t flags, int node)
2692{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002693 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002694 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002695
Ingo Molnar057685c2009-02-20 12:15:30 +01002696 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002697 ret = kmalloc_large_node(size, flags, node);
2698
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002699 trace_kmalloc_node(_RET_IP_, ret,
2700 size, PAGE_SIZE << get_order(size),
2701 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002702
2703 return ret;
2704 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002705
2706 s = get_slab(size, flags);
2707
2708 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002709 return s;
2710
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002711 ret = slab_alloc(s, flags, node, _RET_IP_);
2712
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002713 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002714
2715 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002716}
2717EXPORT_SYMBOL(__kmalloc_node);
2718#endif
2719
2720size_t ksize(const void *object)
2721{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002722 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002723 struct kmem_cache *s;
2724
Christoph Lameteref8b4522007-10-16 01:24:46 -07002725 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002726 return 0;
2727
Vegard Nossum294a80a2007-12-04 23:45:30 -08002728 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002729
Pekka Enberg76994412008-05-22 19:22:25 +03002730 if (unlikely(!PageSlab(page))) {
2731 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002732 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002733 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002734 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002735
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002736#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002737 /*
2738 * Debugging requires use of the padding between object
2739 * and whatever may come after it.
2740 */
2741 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2742 return s->objsize;
2743
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002744#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002745 /*
2746 * If we have the need to store the freelist pointer
2747 * back there or track user information then we can
2748 * only use the space before that information.
2749 */
2750 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2751 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002752 /*
2753 * Else we can use all the padding etc for the allocation
2754 */
2755 return s->size;
2756}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002757EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002758
2759void kfree(const void *x)
2760{
Christoph Lameter81819f02007-05-06 14:49:36 -07002761 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002762 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002763
Pekka Enberg2121db72009-03-25 11:05:57 +02002764 trace_kfree(_RET_IP_, x);
2765
Satyam Sharma2408c552007-10-16 01:24:44 -07002766 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002767 return;
2768
Christoph Lameterb49af682007-05-06 14:49:41 -07002769 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002770 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002771 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002772 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002773 put_page(page);
2774 return;
2775 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002776 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002777}
2778EXPORT_SYMBOL(kfree);
2779
Christoph Lameter2086d262007-05-06 14:49:46 -07002780/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002781 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2782 * the remaining slabs by the number of items in use. The slabs with the
2783 * most items in use come first. New allocations will then fill those up
2784 * and thus they can be removed from the partial lists.
2785 *
2786 * The slabs with the least items are placed last. This results in them
2787 * being allocated from last increasing the chance that the last objects
2788 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002789 */
2790int kmem_cache_shrink(struct kmem_cache *s)
2791{
2792 int node;
2793 int i;
2794 struct kmem_cache_node *n;
2795 struct page *page;
2796 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002797 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002798 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002799 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002800 unsigned long flags;
2801
2802 if (!slabs_by_inuse)
2803 return -ENOMEM;
2804
2805 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002806 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002807 n = get_node(s, node);
2808
2809 if (!n->nr_partial)
2810 continue;
2811
Christoph Lameter834f3d12008-04-14 19:11:31 +03002812 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002813 INIT_LIST_HEAD(slabs_by_inuse + i);
2814
2815 spin_lock_irqsave(&n->list_lock, flags);
2816
2817 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002818 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002819 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002820 * Note that concurrent frees may occur while we hold the
2821 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002822 */
2823 list_for_each_entry_safe(page, t, &n->partial, lru) {
2824 if (!page->inuse && slab_trylock(page)) {
2825 /*
2826 * Must hold slab lock here because slab_free
2827 * may have freed the last object and be
2828 * waiting to release the slab.
2829 */
Christoph Lameter62e346a2010-09-28 08:10:28 -05002830 __remove_partial(n, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07002831 slab_unlock(page);
2832 discard_slab(s, page);
2833 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002834 list_move(&page->lru,
2835 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002836 }
2837 }
2838
Christoph Lameter2086d262007-05-06 14:49:46 -07002839 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002840 * Rebuild the partial list with the slabs filled up most
2841 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002842 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002843 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002844 list_splice(slabs_by_inuse + i, n->partial.prev);
2845
Christoph Lameter2086d262007-05-06 14:49:46 -07002846 spin_unlock_irqrestore(&n->list_lock, flags);
2847 }
2848
2849 kfree(slabs_by_inuse);
2850 return 0;
2851}
2852EXPORT_SYMBOL(kmem_cache_shrink);
2853
Pekka Enberg92a5bbc2010-10-06 16:58:16 +03002854#if defined(CONFIG_MEMORY_HOTPLUG)
Yasunori Gotob9049e22007-10-21 16:41:37 -07002855static int slab_mem_going_offline_callback(void *arg)
2856{
2857 struct kmem_cache *s;
2858
2859 down_read(&slub_lock);
2860 list_for_each_entry(s, &slab_caches, list)
2861 kmem_cache_shrink(s);
2862 up_read(&slub_lock);
2863
2864 return 0;
2865}
2866
2867static void slab_mem_offline_callback(void *arg)
2868{
2869 struct kmem_cache_node *n;
2870 struct kmem_cache *s;
2871 struct memory_notify *marg = arg;
2872 int offline_node;
2873
2874 offline_node = marg->status_change_nid;
2875
2876 /*
2877 * If the node still has available memory. we need kmem_cache_node
2878 * for it yet.
2879 */
2880 if (offline_node < 0)
2881 return;
2882
2883 down_read(&slub_lock);
2884 list_for_each_entry(s, &slab_caches, list) {
2885 n = get_node(s, offline_node);
2886 if (n) {
2887 /*
2888 * if n->nr_slabs > 0, slabs still exist on the node
2889 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05002890 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07002891 * callback. So, we must fail.
2892 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002893 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002894
2895 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05002896 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002897 }
2898 }
2899 up_read(&slub_lock);
2900}
2901
2902static int slab_mem_going_online_callback(void *arg)
2903{
2904 struct kmem_cache_node *n;
2905 struct kmem_cache *s;
2906 struct memory_notify *marg = arg;
2907 int nid = marg->status_change_nid;
2908 int ret = 0;
2909
2910 /*
2911 * If the node's memory is already available, then kmem_cache_node is
2912 * already created. Nothing to do.
2913 */
2914 if (nid < 0)
2915 return 0;
2916
2917 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002918 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002919 * allocate a kmem_cache_node structure in order to bring the node
2920 * online.
2921 */
2922 down_read(&slub_lock);
2923 list_for_each_entry(s, &slab_caches, list) {
2924 /*
2925 * XXX: kmem_cache_alloc_node will fallback to other nodes
2926 * since memory is not yet available from the node that
2927 * is brought up.
2928 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05002929 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002930 if (!n) {
2931 ret = -ENOMEM;
2932 goto out;
2933 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03002934 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002935 s->node[nid] = n;
2936 }
2937out:
2938 up_read(&slub_lock);
2939 return ret;
2940}
2941
2942static int slab_memory_callback(struct notifier_block *self,
2943 unsigned long action, void *arg)
2944{
2945 int ret = 0;
2946
2947 switch (action) {
2948 case MEM_GOING_ONLINE:
2949 ret = slab_mem_going_online_callback(arg);
2950 break;
2951 case MEM_GOING_OFFLINE:
2952 ret = slab_mem_going_offline_callback(arg);
2953 break;
2954 case MEM_OFFLINE:
2955 case MEM_CANCEL_ONLINE:
2956 slab_mem_offline_callback(arg);
2957 break;
2958 case MEM_ONLINE:
2959 case MEM_CANCEL_OFFLINE:
2960 break;
2961 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08002962 if (ret)
2963 ret = notifier_from_errno(ret);
2964 else
2965 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07002966 return ret;
2967}
2968
2969#endif /* CONFIG_MEMORY_HOTPLUG */
2970
Christoph Lameter81819f02007-05-06 14:49:36 -07002971/********************************************************************
2972 * Basic setup of slabs
2973 *******************************************************************/
2974
Christoph Lameter51df1142010-08-20 12:37:15 -05002975/*
2976 * Used for early kmem_cache structures that were allocated using
2977 * the page allocator
2978 */
2979
2980static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
2981{
2982 int node;
2983
2984 list_add(&s->list, &slab_caches);
2985 s->refcount = -1;
2986
2987 for_each_node_state(node, N_NORMAL_MEMORY) {
2988 struct kmem_cache_node *n = get_node(s, node);
2989 struct page *p;
2990
2991 if (n) {
2992 list_for_each_entry(p, &n->partial, lru)
2993 p->slab = s;
2994
2995#ifdef CONFIG_SLAB_DEBUG
2996 list_for_each_entry(p, &n->full, lru)
2997 p->slab = s;
2998#endif
2999 }
3000 }
3001}
3002
Christoph Lameter81819f02007-05-06 14:49:36 -07003003void __init kmem_cache_init(void)
3004{
3005 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003006 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003007 struct kmem_cache *temp_kmem_cache;
3008 int order;
Christoph Lameter51df1142010-08-20 12:37:15 -05003009 struct kmem_cache *temp_kmem_cache_node;
3010 unsigned long kmalloc_size;
3011
3012 kmem_size = offsetof(struct kmem_cache, node) +
3013 nr_node_ids * sizeof(struct kmem_cache_node *);
3014
3015 /* Allocate two kmem_caches from the page allocator */
3016 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3017 order = get_order(2 * kmalloc_size);
3018 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3019
Christoph Lameter81819f02007-05-06 14:49:36 -07003020 /*
3021 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003022 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003023 * kmem_cache_open for slab_state == DOWN.
3024 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003025 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3026
3027 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3028 sizeof(struct kmem_cache_node),
3029 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003030
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003031 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003032
3033 /* Able to allocate the per node structures */
3034 slab_state = PARTIAL;
3035
Christoph Lameter51df1142010-08-20 12:37:15 -05003036 temp_kmem_cache = kmem_cache;
3037 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3038 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3039 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3040 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003041
Christoph Lameter51df1142010-08-20 12:37:15 -05003042 /*
3043 * Allocate kmem_cache_node properly from the kmem_cache slab.
3044 * kmem_cache_node is separately allocated so no need to
3045 * update any list pointers.
3046 */
3047 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003048
Christoph Lameter51df1142010-08-20 12:37:15 -05003049 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3050 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3051
3052 kmem_cache_bootstrap_fixup(kmem_cache_node);
3053
3054 caches++;
Christoph Lameter51df1142010-08-20 12:37:15 -05003055 kmem_cache_bootstrap_fixup(kmem_cache);
3056 caches++;
3057 /* Free temporary boot structure */
3058 free_pages((unsigned long)temp_kmem_cache, order);
3059
3060 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003061
3062 /*
3063 * Patch up the size_index table if we have strange large alignment
3064 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003065 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003066 *
3067 * Largest permitted alignment is 256 bytes due to the way we
3068 * handle the index determination for the smaller caches.
3069 *
3070 * Make sure that nothing crazy happens if someone starts tinkering
3071 * around with ARCH_KMALLOC_MINALIGN
3072 */
3073 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3074 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3075
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003076 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3077 int elem = size_index_elem(i);
3078 if (elem >= ARRAY_SIZE(size_index))
3079 break;
3080 size_index[elem] = KMALLOC_SHIFT_LOW;
3081 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003082
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003083 if (KMALLOC_MIN_SIZE == 64) {
3084 /*
3085 * The 96 byte size cache is not used if the alignment
3086 * is 64 byte.
3087 */
3088 for (i = 64 + 8; i <= 96; i += 8)
3089 size_index[size_index_elem(i)] = 7;
3090 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003091 /*
3092 * The 192 byte sized cache is not used if the alignment
3093 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3094 * instead.
3095 */
3096 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003097 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003098 }
3099
Christoph Lameter51df1142010-08-20 12:37:15 -05003100 /* Caches that are not of the two-to-the-power-of size */
3101 if (KMALLOC_MIN_SIZE <= 32) {
3102 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3103 caches++;
3104 }
3105
3106 if (KMALLOC_MIN_SIZE <= 64) {
3107 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3108 caches++;
3109 }
3110
3111 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3112 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3113 caches++;
3114 }
3115
Christoph Lameter81819f02007-05-06 14:49:36 -07003116 slab_state = UP;
3117
3118 /* Provide the correct kmalloc names now that the caches are up */
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003119 if (KMALLOC_MIN_SIZE <= 32) {
3120 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3121 BUG_ON(!kmalloc_caches[1]->name);
3122 }
3123
3124 if (KMALLOC_MIN_SIZE <= 64) {
3125 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3126 BUG_ON(!kmalloc_caches[2]->name);
3127 }
3128
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003129 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3130 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3131
3132 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003133 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003134 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003135
3136#ifdef CONFIG_SMP
3137 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003138#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003139
Christoph Lameter55136592010-08-20 12:37:13 -05003140#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003141 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3142 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003143
Christoph Lameter51df1142010-08-20 12:37:15 -05003144 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003145 char *name = kasprintf(GFP_NOWAIT,
3146 "dma-kmalloc-%d", s->objsize);
3147
3148 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003149 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3150 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003151 }
3152 }
3153#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003154 printk(KERN_INFO
3155 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003156 " CPUs=%d, Nodes=%d\n",
3157 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003158 slub_min_order, slub_max_order, slub_min_objects,
3159 nr_cpu_ids, nr_node_ids);
3160}
3161
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003162void __init kmem_cache_init_late(void)
3163{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003164}
3165
Christoph Lameter81819f02007-05-06 14:49:36 -07003166/*
3167 * Find a mergeable slab cache
3168 */
3169static int slab_unmergeable(struct kmem_cache *s)
3170{
3171 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3172 return 1;
3173
Christoph Lameterc59def92007-05-16 22:10:50 -07003174 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003175 return 1;
3176
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003177 /*
3178 * We may have set a slab to be unmergeable during bootstrap.
3179 */
3180 if (s->refcount < 0)
3181 return 1;
3182
Christoph Lameter81819f02007-05-06 14:49:36 -07003183 return 0;
3184}
3185
3186static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003187 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003188 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003189{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003190 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003191
3192 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3193 return NULL;
3194
Christoph Lameterc59def92007-05-16 22:10:50 -07003195 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003196 return NULL;
3197
3198 size = ALIGN(size, sizeof(void *));
3199 align = calculate_alignment(flags, align, size);
3200 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003201 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003202
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003203 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003204 if (slab_unmergeable(s))
3205 continue;
3206
3207 if (size > s->size)
3208 continue;
3209
Christoph Lameterba0268a2007-09-11 15:24:11 -07003210 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003211 continue;
3212 /*
3213 * Check if alignment is compatible.
3214 * Courtesy of Adrian Drzewiecki
3215 */
Pekka Enberg06428782008-01-07 23:20:27 -08003216 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003217 continue;
3218
3219 if (s->size - size >= sizeof(void *))
3220 continue;
3221
3222 return s;
3223 }
3224 return NULL;
3225}
3226
3227struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003228 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003229{
3230 struct kmem_cache *s;
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003231 char *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003232
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003233 if (WARN_ON(!name))
3234 return NULL;
3235
Christoph Lameter81819f02007-05-06 14:49:36 -07003236 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003237 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003238 if (s) {
3239 s->refcount++;
3240 /*
3241 * Adjust the object sizes so that we clear
3242 * the complete object on kzalloc.
3243 */
3244 s->objsize = max(s->objsize, (int)size);
3245 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003246
David Rientjes7b8f3b62008-12-17 22:09:46 -08003247 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003248 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003249 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003250 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003251 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003252 return s;
3253 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003254
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003255 n = kstrdup(name, GFP_KERNEL);
3256 if (!n)
3257 goto err;
3258
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003259 s = kmalloc(kmem_size, GFP_KERNEL);
3260 if (s) {
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003261 if (kmem_cache_open(s, n,
Christoph Lameterc59def92007-05-16 22:10:50 -07003262 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003263 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003264 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003265 list_del(&s->list);
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003266 kfree(n);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003267 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003268 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003269 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003270 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003271 return s;
3272 }
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003273 kfree(n);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003274 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003275 }
3276 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003277
3278err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003279 if (flags & SLAB_PANIC)
3280 panic("Cannot create slabcache %s\n", name);
3281 else
3282 s = NULL;
3283 return s;
3284}
3285EXPORT_SYMBOL(kmem_cache_create);
3286
Christoph Lameter81819f02007-05-06 14:49:36 -07003287#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003288/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003289 * Use the cpu notifier to insure that the cpu slabs are flushed when
3290 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003291 */
3292static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3293 unsigned long action, void *hcpu)
3294{
3295 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003296 struct kmem_cache *s;
3297 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003298
3299 switch (action) {
3300 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003301 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003302 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003303 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003304 down_read(&slub_lock);
3305 list_for_each_entry(s, &slab_caches, list) {
3306 local_irq_save(flags);
3307 __flush_cpu_slab(s, cpu);
3308 local_irq_restore(flags);
3309 }
3310 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003311 break;
3312 default:
3313 break;
3314 }
3315 return NOTIFY_OK;
3316}
3317
Pekka Enberg06428782008-01-07 23:20:27 -08003318static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003319 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003320};
Christoph Lameter81819f02007-05-06 14:49:36 -07003321
3322#endif
3323
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003324void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003325{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003326 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003327 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003328
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003329 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003330 return kmalloc_large(size, gfpflags);
3331
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003332 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003333
Satyam Sharma2408c552007-10-16 01:24:44 -07003334 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003335 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003336
Christoph Lameter2154a332010-07-09 14:07:10 -05003337 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003338
3339 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003340 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003341
3342 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003343}
3344
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003345#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003346void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003347 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003348{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003349 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003350 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003351
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003352 if (unlikely(size > SLUB_MAX_SIZE)) {
3353 ret = kmalloc_large_node(size, gfpflags, node);
3354
3355 trace_kmalloc_node(caller, ret,
3356 size, PAGE_SIZE << get_order(size),
3357 gfpflags, node);
3358
3359 return ret;
3360 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003361
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003362 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003363
Satyam Sharma2408c552007-10-16 01:24:44 -07003364 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003365 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003366
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003367 ret = slab_alloc(s, gfpflags, node, caller);
3368
3369 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003370 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003371
3372 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003373}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003374#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003375
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003376#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003377static int count_inuse(struct page *page)
3378{
3379 return page->inuse;
3380}
3381
3382static int count_total(struct page *page)
3383{
3384 return page->objects;
3385}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003386#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003387
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003388#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003389static int validate_slab(struct kmem_cache *s, struct page *page,
3390 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003391{
3392 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003393 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003394
3395 if (!check_slab(s, page) ||
3396 !on_freelist(s, page, NULL))
3397 return 0;
3398
3399 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003400 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003401
Christoph Lameter7656c722007-05-09 02:32:40 -07003402 for_each_free_object(p, s, page->freelist) {
3403 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003404 if (!check_object(s, page, p, 0))
3405 return 0;
3406 }
3407
Christoph Lameter224a88b2008-04-14 19:11:31 +03003408 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003409 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003410 if (!check_object(s, page, p, 1))
3411 return 0;
3412 return 1;
3413}
3414
Christoph Lameter434e2452007-07-17 04:03:30 -07003415static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3416 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003417{
3418 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003419 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003420 slab_unlock(page);
3421 } else
3422 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3423 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003424}
3425
Christoph Lameter434e2452007-07-17 04:03:30 -07003426static int validate_slab_node(struct kmem_cache *s,
3427 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003428{
3429 unsigned long count = 0;
3430 struct page *page;
3431 unsigned long flags;
3432
3433 spin_lock_irqsave(&n->list_lock, flags);
3434
3435 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003436 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003437 count++;
3438 }
3439 if (count != n->nr_partial)
3440 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3441 "counter=%ld\n", s->name, count, n->nr_partial);
3442
3443 if (!(s->flags & SLAB_STORE_USER))
3444 goto out;
3445
3446 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003447 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003448 count++;
3449 }
3450 if (count != atomic_long_read(&n->nr_slabs))
3451 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3452 "counter=%ld\n", s->name, count,
3453 atomic_long_read(&n->nr_slabs));
3454
3455out:
3456 spin_unlock_irqrestore(&n->list_lock, flags);
3457 return count;
3458}
3459
Christoph Lameter434e2452007-07-17 04:03:30 -07003460static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003461{
3462 int node;
3463 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003464 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003465 sizeof(unsigned long), GFP_KERNEL);
3466
3467 if (!map)
3468 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003469
3470 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003471 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003472 struct kmem_cache_node *n = get_node(s, node);
3473
Christoph Lameter434e2452007-07-17 04:03:30 -07003474 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003475 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003476 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003477 return count;
3478}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003479/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003480 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003481 * and freed.
3482 */
3483
3484struct location {
3485 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003486 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003487 long long sum_time;
3488 long min_time;
3489 long max_time;
3490 long min_pid;
3491 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303492 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003493 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003494};
3495
3496struct loc_track {
3497 unsigned long max;
3498 unsigned long count;
3499 struct location *loc;
3500};
3501
3502static void free_loc_track(struct loc_track *t)
3503{
3504 if (t->max)
3505 free_pages((unsigned long)t->loc,
3506 get_order(sizeof(struct location) * t->max));
3507}
3508
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003509static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003510{
3511 struct location *l;
3512 int order;
3513
Christoph Lameter88a420e2007-05-06 14:49:45 -07003514 order = get_order(sizeof(struct location) * max);
3515
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003516 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003517 if (!l)
3518 return 0;
3519
3520 if (t->count) {
3521 memcpy(l, t->loc, sizeof(struct location) * t->count);
3522 free_loc_track(t);
3523 }
3524 t->max = max;
3525 t->loc = l;
3526 return 1;
3527}
3528
3529static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003530 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003531{
3532 long start, end, pos;
3533 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003534 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003535 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003536
3537 start = -1;
3538 end = t->count;
3539
3540 for ( ; ; ) {
3541 pos = start + (end - start + 1) / 2;
3542
3543 /*
3544 * There is nothing at "end". If we end up there
3545 * we need to add something to before end.
3546 */
3547 if (pos == end)
3548 break;
3549
3550 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003551 if (track->addr == caddr) {
3552
3553 l = &t->loc[pos];
3554 l->count++;
3555 if (track->when) {
3556 l->sum_time += age;
3557 if (age < l->min_time)
3558 l->min_time = age;
3559 if (age > l->max_time)
3560 l->max_time = age;
3561
3562 if (track->pid < l->min_pid)
3563 l->min_pid = track->pid;
3564 if (track->pid > l->max_pid)
3565 l->max_pid = track->pid;
3566
Rusty Russell174596a2009-01-01 10:12:29 +10303567 cpumask_set_cpu(track->cpu,
3568 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003569 }
3570 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003571 return 1;
3572 }
3573
Christoph Lameter45edfa52007-05-09 02:32:45 -07003574 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003575 end = pos;
3576 else
3577 start = pos;
3578 }
3579
3580 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003581 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003582 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003583 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003584 return 0;
3585
3586 l = t->loc + pos;
3587 if (pos < t->count)
3588 memmove(l + 1, l,
3589 (t->count - pos) * sizeof(struct location));
3590 t->count++;
3591 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003592 l->addr = track->addr;
3593 l->sum_time = age;
3594 l->min_time = age;
3595 l->max_time = age;
3596 l->min_pid = track->pid;
3597 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303598 cpumask_clear(to_cpumask(l->cpus));
3599 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003600 nodes_clear(l->nodes);
3601 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003602 return 1;
3603}
3604
3605static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003606 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003607 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003608{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003609 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003610 void *p;
3611
Christoph Lameter39b26462008-04-14 19:11:30 +03003612 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003613 for_each_free_object(p, s, page->freelist)
3614 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003615
Christoph Lameter224a88b2008-04-14 19:11:31 +03003616 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003617 if (!test_bit(slab_index(p, s, addr), map))
3618 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003619}
3620
3621static int list_locations(struct kmem_cache *s, char *buf,
3622 enum track_item alloc)
3623{
Harvey Harrisone374d482008-01-31 15:20:50 -08003624 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003625 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003626 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003627 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003628 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3629 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003630
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003631 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3632 GFP_TEMPORARY)) {
3633 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003634 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003635 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003636 /* Push back cpu slabs */
3637 flush_all(s);
3638
Christoph Lameterf64dc582007-10-16 01:25:33 -07003639 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003640 struct kmem_cache_node *n = get_node(s, node);
3641 unsigned long flags;
3642 struct page *page;
3643
Christoph Lameter9e869432007-08-22 14:01:56 -07003644 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003645 continue;
3646
3647 spin_lock_irqsave(&n->list_lock, flags);
3648 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003649 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003650 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003651 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003652 spin_unlock_irqrestore(&n->list_lock, flags);
3653 }
3654
3655 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003656 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003657
Hugh Dickins9c246242008-12-09 13:14:27 -08003658 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003659 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003660 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003661
3662 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003663 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003664 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003665 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003666
3667 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003668 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003669 l->min_time,
3670 (long)div_u64(l->sum_time, l->count),
3671 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003672 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003673 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003674 l->min_time);
3675
3676 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003677 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003678 l->min_pid, l->max_pid);
3679 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003680 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003681 l->min_pid);
3682
Rusty Russell174596a2009-01-01 10:12:29 +10303683 if (num_online_cpus() > 1 &&
3684 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003685 len < PAGE_SIZE - 60) {
3686 len += sprintf(buf + len, " cpus=");
3687 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303688 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003689 }
3690
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003691 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003692 len < PAGE_SIZE - 60) {
3693 len += sprintf(buf + len, " nodes=");
3694 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003695 l->nodes);
3696 }
3697
Harvey Harrisone374d482008-01-31 15:20:50 -08003698 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003699 }
3700
3701 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003702 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003703 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003704 len += sprintf(buf, "No data\n");
3705 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003706}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003707#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07003708
Christoph Lametera5a84752010-10-05 13:57:27 -05003709#ifdef SLUB_RESILIENCY_TEST
3710static void resiliency_test(void)
3711{
3712 u8 *p;
3713
3714 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3715
3716 printk(KERN_ERR "SLUB resiliency testing\n");
3717 printk(KERN_ERR "-----------------------\n");
3718 printk(KERN_ERR "A. Corruption after allocation\n");
3719
3720 p = kzalloc(16, GFP_KERNEL);
3721 p[16] = 0x12;
3722 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3723 " 0x12->0x%p\n\n", p + 16);
3724
3725 validate_slab_cache(kmalloc_caches[4]);
3726
3727 /* Hmmm... The next two are dangerous */
3728 p = kzalloc(32, GFP_KERNEL);
3729 p[32 + sizeof(void *)] = 0x34;
3730 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3731 " 0x34 -> -0x%p\n", p);
3732 printk(KERN_ERR
3733 "If allocated object is overwritten then not detectable\n\n");
3734
3735 validate_slab_cache(kmalloc_caches[5]);
3736 p = kzalloc(64, GFP_KERNEL);
3737 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3738 *p = 0x56;
3739 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3740 p);
3741 printk(KERN_ERR
3742 "If allocated object is overwritten then not detectable\n\n");
3743 validate_slab_cache(kmalloc_caches[6]);
3744
3745 printk(KERN_ERR "\nB. Corruption after free\n");
3746 p = kzalloc(128, GFP_KERNEL);
3747 kfree(p);
3748 *p = 0x78;
3749 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3750 validate_slab_cache(kmalloc_caches[7]);
3751
3752 p = kzalloc(256, GFP_KERNEL);
3753 kfree(p);
3754 p[50] = 0x9a;
3755 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3756 p);
3757 validate_slab_cache(kmalloc_caches[8]);
3758
3759 p = kzalloc(512, GFP_KERNEL);
3760 kfree(p);
3761 p[512] = 0xab;
3762 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3763 validate_slab_cache(kmalloc_caches[9]);
3764}
3765#else
3766#ifdef CONFIG_SYSFS
3767static void resiliency_test(void) {};
3768#endif
3769#endif
3770
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003771#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07003772enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003773 SL_ALL, /* All slabs */
3774 SL_PARTIAL, /* Only partially allocated slabs */
3775 SL_CPU, /* Only slabs used for cpu caches */
3776 SL_OBJECTS, /* Determine allocated objects not slabs */
3777 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003778};
3779
Christoph Lameter205ab992008-04-14 19:11:40 +03003780#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003781#define SO_PARTIAL (1 << SL_PARTIAL)
3782#define SO_CPU (1 << SL_CPU)
3783#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003784#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003785
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003786static ssize_t show_slab_objects(struct kmem_cache *s,
3787 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003788{
3789 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003790 int node;
3791 int x;
3792 unsigned long *nodes;
3793 unsigned long *per_cpu;
3794
3795 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003796 if (!nodes)
3797 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003798 per_cpu = nodes + nr_node_ids;
3799
Christoph Lameter205ab992008-04-14 19:11:40 +03003800 if (flags & SO_CPU) {
3801 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003802
Christoph Lameter205ab992008-04-14 19:11:40 +03003803 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003804 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003805
Christoph Lameter205ab992008-04-14 19:11:40 +03003806 if (!c || c->node < 0)
3807 continue;
3808
3809 if (c->page) {
3810 if (flags & SO_TOTAL)
3811 x = c->page->objects;
3812 else if (flags & SO_OBJECTS)
3813 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003814 else
3815 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003816
Christoph Lameter81819f02007-05-06 14:49:36 -07003817 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003818 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003819 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003820 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003821 }
3822 }
3823
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003824 down_read(&slub_lock);
3825#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003826 if (flags & SO_ALL) {
3827 for_each_node_state(node, N_NORMAL_MEMORY) {
3828 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003829
Christoph Lameter205ab992008-04-14 19:11:40 +03003830 if (flags & SO_TOTAL)
3831 x = atomic_long_read(&n->total_objects);
3832 else if (flags & SO_OBJECTS)
3833 x = atomic_long_read(&n->total_objects) -
3834 count_partial(n, count_free);
3835
3836 else
3837 x = atomic_long_read(&n->nr_slabs);
3838 total += x;
3839 nodes[node] += x;
3840 }
3841
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003842 } else
3843#endif
3844 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03003845 for_each_node_state(node, N_NORMAL_MEMORY) {
3846 struct kmem_cache_node *n = get_node(s, node);
3847
3848 if (flags & SO_TOTAL)
3849 x = count_partial(n, count_total);
3850 else if (flags & SO_OBJECTS)
3851 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003852 else
3853 x = n->nr_partial;
3854 total += x;
3855 nodes[node] += x;
3856 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003857 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003858 x = sprintf(buf, "%lu", total);
3859#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003860 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003861 if (nodes[node])
3862 x += sprintf(buf + x, " N%d=%lu",
3863 node, nodes[node]);
3864#endif
3865 kfree(nodes);
3866 return x + sprintf(buf + x, "\n");
3867}
3868
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003869#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003870static int any_slab_objects(struct kmem_cache *s)
3871{
3872 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003873
3874 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003875 struct kmem_cache_node *n = get_node(s, node);
3876
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003877 if (!n)
3878 continue;
3879
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003880 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003881 return 1;
3882 }
3883 return 0;
3884}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003885#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003886
3887#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3888#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3889
3890struct slab_attribute {
3891 struct attribute attr;
3892 ssize_t (*show)(struct kmem_cache *s, char *buf);
3893 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3894};
3895
3896#define SLAB_ATTR_RO(_name) \
3897 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3898
3899#define SLAB_ATTR(_name) \
3900 static struct slab_attribute _name##_attr = \
3901 __ATTR(_name, 0644, _name##_show, _name##_store)
3902
Christoph Lameter81819f02007-05-06 14:49:36 -07003903static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3904{
3905 return sprintf(buf, "%d\n", s->size);
3906}
3907SLAB_ATTR_RO(slab_size);
3908
3909static ssize_t align_show(struct kmem_cache *s, char *buf)
3910{
3911 return sprintf(buf, "%d\n", s->align);
3912}
3913SLAB_ATTR_RO(align);
3914
3915static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3916{
3917 return sprintf(buf, "%d\n", s->objsize);
3918}
3919SLAB_ATTR_RO(object_size);
3920
3921static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3922{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003923 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003924}
3925SLAB_ATTR_RO(objs_per_slab);
3926
Christoph Lameter06b285d2008-04-14 19:11:41 +03003927static ssize_t order_store(struct kmem_cache *s,
3928 const char *buf, size_t length)
3929{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003930 unsigned long order;
3931 int err;
3932
3933 err = strict_strtoul(buf, 10, &order);
3934 if (err)
3935 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003936
3937 if (order > slub_max_order || order < slub_min_order)
3938 return -EINVAL;
3939
3940 calculate_sizes(s, order);
3941 return length;
3942}
3943
Christoph Lameter81819f02007-05-06 14:49:36 -07003944static ssize_t order_show(struct kmem_cache *s, char *buf)
3945{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003946 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003947}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003948SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003949
David Rientjes73d342b2009-02-22 17:40:09 -08003950static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3951{
3952 return sprintf(buf, "%lu\n", s->min_partial);
3953}
3954
3955static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3956 size_t length)
3957{
3958 unsigned long min;
3959 int err;
3960
3961 err = strict_strtoul(buf, 10, &min);
3962 if (err)
3963 return err;
3964
David Rientjesc0bdb232009-02-25 09:16:35 +02003965 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003966 return length;
3967}
3968SLAB_ATTR(min_partial);
3969
Christoph Lameter81819f02007-05-06 14:49:36 -07003970static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3971{
3972 if (s->ctor) {
3973 int n = sprint_symbol(buf, (unsigned long)s->ctor);
3974
3975 return n + sprintf(buf + n, "\n");
3976 }
3977 return 0;
3978}
3979SLAB_ATTR_RO(ctor);
3980
Christoph Lameter81819f02007-05-06 14:49:36 -07003981static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3982{
3983 return sprintf(buf, "%d\n", s->refcount - 1);
3984}
3985SLAB_ATTR_RO(aliases);
3986
Christoph Lameter81819f02007-05-06 14:49:36 -07003987static ssize_t partial_show(struct kmem_cache *s, char *buf)
3988{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003989 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003990}
3991SLAB_ATTR_RO(partial);
3992
3993static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3994{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003995 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003996}
3997SLAB_ATTR_RO(cpu_slabs);
3998
3999static ssize_t objects_show(struct kmem_cache *s, char *buf)
4000{
Christoph Lameter205ab992008-04-14 19:11:40 +03004001 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004002}
4003SLAB_ATTR_RO(objects);
4004
Christoph Lameter205ab992008-04-14 19:11:40 +03004005static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4006{
4007 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4008}
4009SLAB_ATTR_RO(objects_partial);
4010
Christoph Lameter81819f02007-05-06 14:49:36 -07004011static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4012{
4013 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4014}
4015
4016static ssize_t reclaim_account_store(struct kmem_cache *s,
4017 const char *buf, size_t length)
4018{
4019 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4020 if (buf[0] == '1')
4021 s->flags |= SLAB_RECLAIM_ACCOUNT;
4022 return length;
4023}
4024SLAB_ATTR(reclaim_account);
4025
4026static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4027{
Christoph Lameter5af60832007-05-06 14:49:56 -07004028 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004029}
4030SLAB_ATTR_RO(hwcache_align);
4031
4032#ifdef CONFIG_ZONE_DMA
4033static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4034{
4035 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4036}
4037SLAB_ATTR_RO(cache_dma);
4038#endif
4039
4040static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4041{
4042 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4043}
4044SLAB_ATTR_RO(destroy_by_rcu);
4045
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004046#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004047static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4048{
4049 return show_slab_objects(s, buf, SO_ALL);
4050}
4051SLAB_ATTR_RO(slabs);
4052
4053static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4054{
4055 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4056}
4057SLAB_ATTR_RO(total_objects);
4058
4059static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4060{
4061 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4062}
4063
4064static ssize_t sanity_checks_store(struct kmem_cache *s,
4065 const char *buf, size_t length)
4066{
4067 s->flags &= ~SLAB_DEBUG_FREE;
4068 if (buf[0] == '1')
4069 s->flags |= SLAB_DEBUG_FREE;
4070 return length;
4071}
4072SLAB_ATTR(sanity_checks);
4073
4074static ssize_t trace_show(struct kmem_cache *s, char *buf)
4075{
4076 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4077}
4078
4079static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4080 size_t length)
4081{
4082 s->flags &= ~SLAB_TRACE;
4083 if (buf[0] == '1')
4084 s->flags |= SLAB_TRACE;
4085 return length;
4086}
4087SLAB_ATTR(trace);
4088
Christoph Lameter81819f02007-05-06 14:49:36 -07004089static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4090{
4091 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4092}
4093
4094static ssize_t red_zone_store(struct kmem_cache *s,
4095 const char *buf, size_t length)
4096{
4097 if (any_slab_objects(s))
4098 return -EBUSY;
4099
4100 s->flags &= ~SLAB_RED_ZONE;
4101 if (buf[0] == '1')
4102 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004103 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004104 return length;
4105}
4106SLAB_ATTR(red_zone);
4107
4108static ssize_t poison_show(struct kmem_cache *s, char *buf)
4109{
4110 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4111}
4112
4113static ssize_t poison_store(struct kmem_cache *s,
4114 const char *buf, size_t length)
4115{
4116 if (any_slab_objects(s))
4117 return -EBUSY;
4118
4119 s->flags &= ~SLAB_POISON;
4120 if (buf[0] == '1')
4121 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004122 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004123 return length;
4124}
4125SLAB_ATTR(poison);
4126
4127static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4128{
4129 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4130}
4131
4132static ssize_t store_user_store(struct kmem_cache *s,
4133 const char *buf, size_t length)
4134{
4135 if (any_slab_objects(s))
4136 return -EBUSY;
4137
4138 s->flags &= ~SLAB_STORE_USER;
4139 if (buf[0] == '1')
4140 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004141 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004142 return length;
4143}
4144SLAB_ATTR(store_user);
4145
Christoph Lameter53e15af2007-05-06 14:49:43 -07004146static ssize_t validate_show(struct kmem_cache *s, char *buf)
4147{
4148 return 0;
4149}
4150
4151static ssize_t validate_store(struct kmem_cache *s,
4152 const char *buf, size_t length)
4153{
Christoph Lameter434e2452007-07-17 04:03:30 -07004154 int ret = -EINVAL;
4155
4156 if (buf[0] == '1') {
4157 ret = validate_slab_cache(s);
4158 if (ret >= 0)
4159 ret = length;
4160 }
4161 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004162}
4163SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004164
4165static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4166{
4167 if (!(s->flags & SLAB_STORE_USER))
4168 return -ENOSYS;
4169 return list_locations(s, buf, TRACK_ALLOC);
4170}
4171SLAB_ATTR_RO(alloc_calls);
4172
4173static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4174{
4175 if (!(s->flags & SLAB_STORE_USER))
4176 return -ENOSYS;
4177 return list_locations(s, buf, TRACK_FREE);
4178}
4179SLAB_ATTR_RO(free_calls);
4180#endif /* CONFIG_SLUB_DEBUG */
4181
4182#ifdef CONFIG_FAILSLAB
4183static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4184{
4185 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4186}
4187
4188static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4189 size_t length)
4190{
4191 s->flags &= ~SLAB_FAILSLAB;
4192 if (buf[0] == '1')
4193 s->flags |= SLAB_FAILSLAB;
4194 return length;
4195}
4196SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004197#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004198
Christoph Lameter2086d262007-05-06 14:49:46 -07004199static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4200{
4201 return 0;
4202}
4203
4204static ssize_t shrink_store(struct kmem_cache *s,
4205 const char *buf, size_t length)
4206{
4207 if (buf[0] == '1') {
4208 int rc = kmem_cache_shrink(s);
4209
4210 if (rc)
4211 return rc;
4212 } else
4213 return -EINVAL;
4214 return length;
4215}
4216SLAB_ATTR(shrink);
4217
Christoph Lameter81819f02007-05-06 14:49:36 -07004218#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004219static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004220{
Christoph Lameter98246012008-01-07 23:20:26 -08004221 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004222}
4223
Christoph Lameter98246012008-01-07 23:20:26 -08004224static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004225 const char *buf, size_t length)
4226{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004227 unsigned long ratio;
4228 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004229
Christoph Lameter0121c6192008-04-29 16:11:12 -07004230 err = strict_strtoul(buf, 10, &ratio);
4231 if (err)
4232 return err;
4233
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004234 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004235 s->remote_node_defrag_ratio = ratio * 10;
4236
Christoph Lameter81819f02007-05-06 14:49:36 -07004237 return length;
4238}
Christoph Lameter98246012008-01-07 23:20:26 -08004239SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004240#endif
4241
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004242#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004243static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4244{
4245 unsigned long sum = 0;
4246 int cpu;
4247 int len;
4248 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4249
4250 if (!data)
4251 return -ENOMEM;
4252
4253 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004254 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004255
4256 data[cpu] = x;
4257 sum += x;
4258 }
4259
4260 len = sprintf(buf, "%lu", sum);
4261
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004262#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004263 for_each_online_cpu(cpu) {
4264 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004265 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004266 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004267#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004268 kfree(data);
4269 return len + sprintf(buf + len, "\n");
4270}
4271
David Rientjes78eb00c2009-10-15 02:20:22 -07004272static void clear_stat(struct kmem_cache *s, enum stat_item si)
4273{
4274 int cpu;
4275
4276 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004277 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004278}
4279
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004280#define STAT_ATTR(si, text) \
4281static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4282{ \
4283 return show_stat(s, buf, si); \
4284} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004285static ssize_t text##_store(struct kmem_cache *s, \
4286 const char *buf, size_t length) \
4287{ \
4288 if (buf[0] != '0') \
4289 return -EINVAL; \
4290 clear_stat(s, si); \
4291 return length; \
4292} \
4293SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004294
4295STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4296STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4297STAT_ATTR(FREE_FASTPATH, free_fastpath);
4298STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4299STAT_ATTR(FREE_FROZEN, free_frozen);
4300STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4301STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4302STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4303STAT_ATTR(ALLOC_SLAB, alloc_slab);
4304STAT_ATTR(ALLOC_REFILL, alloc_refill);
4305STAT_ATTR(FREE_SLAB, free_slab);
4306STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4307STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4308STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4309STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4310STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4311STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004312STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004313#endif
4314
Pekka Enberg06428782008-01-07 23:20:27 -08004315static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004316 &slab_size_attr.attr,
4317 &object_size_attr.attr,
4318 &objs_per_slab_attr.attr,
4319 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004320 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004321 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004322 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004323 &partial_attr.attr,
4324 &cpu_slabs_attr.attr,
4325 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004326 &aliases_attr.attr,
4327 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004328 &hwcache_align_attr.attr,
4329 &reclaim_account_attr.attr,
4330 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004331 &shrink_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004332#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004333 &total_objects_attr.attr,
4334 &slabs_attr.attr,
4335 &sanity_checks_attr.attr,
4336 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004337 &red_zone_attr.attr,
4338 &poison_attr.attr,
4339 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004340 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004341 &alloc_calls_attr.attr,
4342 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004343#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004344#ifdef CONFIG_ZONE_DMA
4345 &cache_dma_attr.attr,
4346#endif
4347#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004348 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004349#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004350#ifdef CONFIG_SLUB_STATS
4351 &alloc_fastpath_attr.attr,
4352 &alloc_slowpath_attr.attr,
4353 &free_fastpath_attr.attr,
4354 &free_slowpath_attr.attr,
4355 &free_frozen_attr.attr,
4356 &free_add_partial_attr.attr,
4357 &free_remove_partial_attr.attr,
4358 &alloc_from_partial_attr.attr,
4359 &alloc_slab_attr.attr,
4360 &alloc_refill_attr.attr,
4361 &free_slab_attr.attr,
4362 &cpuslab_flush_attr.attr,
4363 &deactivate_full_attr.attr,
4364 &deactivate_empty_attr.attr,
4365 &deactivate_to_head_attr.attr,
4366 &deactivate_to_tail_attr.attr,
4367 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004368 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004369#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004370#ifdef CONFIG_FAILSLAB
4371 &failslab_attr.attr,
4372#endif
4373
Christoph Lameter81819f02007-05-06 14:49:36 -07004374 NULL
4375};
4376
4377static struct attribute_group slab_attr_group = {
4378 .attrs = slab_attrs,
4379};
4380
4381static ssize_t slab_attr_show(struct kobject *kobj,
4382 struct attribute *attr,
4383 char *buf)
4384{
4385 struct slab_attribute *attribute;
4386 struct kmem_cache *s;
4387 int err;
4388
4389 attribute = to_slab_attr(attr);
4390 s = to_slab(kobj);
4391
4392 if (!attribute->show)
4393 return -EIO;
4394
4395 err = attribute->show(s, buf);
4396
4397 return err;
4398}
4399
4400static ssize_t slab_attr_store(struct kobject *kobj,
4401 struct attribute *attr,
4402 const char *buf, size_t len)
4403{
4404 struct slab_attribute *attribute;
4405 struct kmem_cache *s;
4406 int err;
4407
4408 attribute = to_slab_attr(attr);
4409 s = to_slab(kobj);
4410
4411 if (!attribute->store)
4412 return -EIO;
4413
4414 err = attribute->store(s, buf, len);
4415
4416 return err;
4417}
4418
Christoph Lameter151c6022008-01-07 22:29:05 -08004419static void kmem_cache_release(struct kobject *kobj)
4420{
4421 struct kmem_cache *s = to_slab(kobj);
4422
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004423 kfree(s->name);
Christoph Lameter151c6022008-01-07 22:29:05 -08004424 kfree(s);
4425}
4426
Emese Revfy52cf25d2010-01-19 02:58:23 +01004427static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004428 .show = slab_attr_show,
4429 .store = slab_attr_store,
4430};
4431
4432static struct kobj_type slab_ktype = {
4433 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004434 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004435};
4436
4437static int uevent_filter(struct kset *kset, struct kobject *kobj)
4438{
4439 struct kobj_type *ktype = get_ktype(kobj);
4440
4441 if (ktype == &slab_ktype)
4442 return 1;
4443 return 0;
4444}
4445
Emese Revfy9cd43612009-12-31 14:52:51 +01004446static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004447 .filter = uevent_filter,
4448};
4449
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004450static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004451
4452#define ID_STR_LENGTH 64
4453
4454/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004455 *
4456 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004457 */
4458static char *create_unique_id(struct kmem_cache *s)
4459{
4460 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4461 char *p = name;
4462
4463 BUG_ON(!name);
4464
4465 *p++ = ':';
4466 /*
4467 * First flags affecting slabcache operations. We will only
4468 * get here for aliasable slabs so we do not need to support
4469 * too many flags. The flags here must cover all flags that
4470 * are matched during merging to guarantee that the id is
4471 * unique.
4472 */
4473 if (s->flags & SLAB_CACHE_DMA)
4474 *p++ = 'd';
4475 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4476 *p++ = 'a';
4477 if (s->flags & SLAB_DEBUG_FREE)
4478 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004479 if (!(s->flags & SLAB_NOTRACK))
4480 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004481 if (p != name + 1)
4482 *p++ = '-';
4483 p += sprintf(p, "%07d", s->size);
4484 BUG_ON(p > name + ID_STR_LENGTH - 1);
4485 return name;
4486}
4487
4488static int sysfs_slab_add(struct kmem_cache *s)
4489{
4490 int err;
4491 const char *name;
4492 int unmergeable;
4493
4494 if (slab_state < SYSFS)
4495 /* Defer until later */
4496 return 0;
4497
4498 unmergeable = slab_unmergeable(s);
4499 if (unmergeable) {
4500 /*
4501 * Slabcache can never be merged so we can use the name proper.
4502 * This is typically the case for debug situations. In that
4503 * case we can catch duplicate names easily.
4504 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004505 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004506 name = s->name;
4507 } else {
4508 /*
4509 * Create a unique name for the slab as a target
4510 * for the symlinks.
4511 */
4512 name = create_unique_id(s);
4513 }
4514
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004515 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004516 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4517 if (err) {
4518 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004519 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004520 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004521
4522 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004523 if (err) {
4524 kobject_del(&s->kobj);
4525 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004526 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004527 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004528 kobject_uevent(&s->kobj, KOBJ_ADD);
4529 if (!unmergeable) {
4530 /* Setup first alias */
4531 sysfs_slab_alias(s, s->name);
4532 kfree(name);
4533 }
4534 return 0;
4535}
4536
4537static void sysfs_slab_remove(struct kmem_cache *s)
4538{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004539 if (slab_state < SYSFS)
4540 /*
4541 * Sysfs has not been setup yet so no need to remove the
4542 * cache from sysfs.
4543 */
4544 return;
4545
Christoph Lameter81819f02007-05-06 14:49:36 -07004546 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4547 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004548 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004549}
4550
4551/*
4552 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004553 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004554 */
4555struct saved_alias {
4556 struct kmem_cache *s;
4557 const char *name;
4558 struct saved_alias *next;
4559};
4560
Adrian Bunk5af328a2007-07-17 04:03:27 -07004561static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004562
4563static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4564{
4565 struct saved_alias *al;
4566
4567 if (slab_state == SYSFS) {
4568 /*
4569 * If we have a leftover link then remove it.
4570 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004571 sysfs_remove_link(&slab_kset->kobj, name);
4572 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004573 }
4574
4575 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4576 if (!al)
4577 return -ENOMEM;
4578
4579 al->s = s;
4580 al->name = name;
4581 al->next = alias_list;
4582 alias_list = al;
4583 return 0;
4584}
4585
4586static int __init slab_sysfs_init(void)
4587{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004588 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004589 int err;
4590
Christoph Lameter2bce6482010-07-19 11:39:11 -05004591 down_write(&slub_lock);
4592
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004593 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004594 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004595 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004596 printk(KERN_ERR "Cannot register slab subsystem.\n");
4597 return -ENOSYS;
4598 }
4599
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004600 slab_state = SYSFS;
4601
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07004602 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004603 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004604 if (err)
4605 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4606 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004607 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004608
4609 while (alias_list) {
4610 struct saved_alias *al = alias_list;
4611
4612 alias_list = alias_list->next;
4613 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004614 if (err)
4615 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4616 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004617 kfree(al);
4618 }
4619
Christoph Lameter2bce6482010-07-19 11:39:11 -05004620 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004621 resiliency_test();
4622 return 0;
4623}
4624
4625__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004626#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004627
4628/*
4629 * The /proc/slabinfo ABI
4630 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004631#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004632static void print_slabinfo_header(struct seq_file *m)
4633{
4634 seq_puts(m, "slabinfo - version: 2.1\n");
4635 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4636 "<objperslab> <pagesperslab>");
4637 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4638 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4639 seq_putc(m, '\n');
4640}
4641
4642static void *s_start(struct seq_file *m, loff_t *pos)
4643{
4644 loff_t n = *pos;
4645
4646 down_read(&slub_lock);
4647 if (!n)
4648 print_slabinfo_header(m);
4649
4650 return seq_list_start(&slab_caches, *pos);
4651}
4652
4653static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4654{
4655 return seq_list_next(p, &slab_caches, pos);
4656}
4657
4658static void s_stop(struct seq_file *m, void *p)
4659{
4660 up_read(&slub_lock);
4661}
4662
4663static int s_show(struct seq_file *m, void *p)
4664{
4665 unsigned long nr_partials = 0;
4666 unsigned long nr_slabs = 0;
4667 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004668 unsigned long nr_objs = 0;
4669 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004670 struct kmem_cache *s;
4671 int node;
4672
4673 s = list_entry(p, struct kmem_cache, list);
4674
4675 for_each_online_node(node) {
4676 struct kmem_cache_node *n = get_node(s, node);
4677
4678 if (!n)
4679 continue;
4680
4681 nr_partials += n->nr_partial;
4682 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004683 nr_objs += atomic_long_read(&n->total_objects);
4684 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004685 }
4686
Christoph Lameter205ab992008-04-14 19:11:40 +03004687 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004688
4689 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004690 nr_objs, s->size, oo_objects(s->oo),
4691 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004692 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4693 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4694 0UL);
4695 seq_putc(m, '\n');
4696 return 0;
4697}
4698
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004699static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004700 .start = s_start,
4701 .next = s_next,
4702 .stop = s_stop,
4703 .show = s_show,
4704};
4705
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004706static int slabinfo_open(struct inode *inode, struct file *file)
4707{
4708 return seq_open(file, &slabinfo_op);
4709}
4710
4711static const struct file_operations proc_slabinfo_operations = {
4712 .open = slabinfo_open,
4713 .read = seq_read,
4714 .llseek = seq_lseek,
4715 .release = seq_release,
4716};
4717
4718static int __init slab_proc_init(void)
4719{
WANG Congcf5d1132009-08-18 19:11:40 +03004720 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004721 return 0;
4722}
4723module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004724#endif /* CONFIG_SLABINFO */