blob: ca49d02b5ff80ada8c4f4480682352738ad659c6 [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 Lameterf6acb632008-04-29 16:16:06 -0700201#ifdef CONFIG_SLUB_DEBUG
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{
212 kfree(s);
213}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800214
Christoph Lameter81819f02007-05-06 14:49:36 -0700215#endif
216
Christoph Lameter84e554e2009-12-18 16:26:23 -0600217static inline void stat(struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800218{
219#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600220 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800221#endif
222}
223
Christoph Lameter81819f02007-05-06 14:49:36 -0700224/********************************************************************
225 * Core slab cache functions
226 *******************************************************************/
227
228int slab_is_available(void)
229{
230 return slab_state >= UP;
231}
232
233static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
234{
235#ifdef CONFIG_NUMA
236 return s->node[node];
237#else
238 return &s->local_node;
239#endif
240}
241
Christoph Lameter6446faa2008-02-15 23:45:26 -0800242/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700243static inline int check_valid_pointer(struct kmem_cache *s,
244 struct page *page, const void *object)
245{
246 void *base;
247
Christoph Lametera973e9d2008-03-01 13:40:44 -0800248 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700249 return 1;
250
Christoph Lametera973e9d2008-03-01 13:40:44 -0800251 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300252 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700253 (object - base) % s->size) {
254 return 0;
255 }
256
257 return 1;
258}
259
Christoph Lameter7656c722007-05-09 02:32:40 -0700260static inline void *get_freepointer(struct kmem_cache *s, void *object)
261{
262 return *(void **)(object + s->offset);
263}
264
265static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
266{
267 *(void **)(object + s->offset) = fp;
268}
269
270/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300271#define for_each_object(__p, __s, __addr, __objects) \
272 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700273 __p += (__s)->size)
274
275/* Scan freelist */
276#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800277 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700278
279/* Determine object index from a given position */
280static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
281{
282 return (p - addr) / s->size;
283}
284
Christoph Lameter834f3d12008-04-14 19:11:31 +0300285static inline struct kmem_cache_order_objects oo_make(int order,
286 unsigned long size)
287{
288 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400289 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300290 };
291
292 return x;
293}
294
295static inline int oo_order(struct kmem_cache_order_objects x)
296{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400297 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300298}
299
300static inline int oo_objects(struct kmem_cache_order_objects x)
301{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400302 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300303}
304
Christoph Lameter41ecc552007-05-09 02:32:44 -0700305#ifdef CONFIG_SLUB_DEBUG
306/*
307 * Debug settings:
308 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700309#ifdef CONFIG_SLUB_DEBUG_ON
310static int slub_debug = DEBUG_DEFAULT_FLAGS;
311#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700312static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700313#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700314
315static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700316static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700317
Christoph Lameter7656c722007-05-09 02:32:40 -0700318/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700319 * Object debugging
320 */
321static void print_section(char *text, u8 *addr, unsigned int length)
322{
323 int i, offset;
324 int newline = 1;
325 char ascii[17];
326
327 ascii[16] = 0;
328
329 for (i = 0; i < length; i++) {
330 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700331 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700332 newline = 0;
333 }
Pekka Enberg06428782008-01-07 23:20:27 -0800334 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700335 offset = i % 16;
336 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
337 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800338 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700339 newline = 1;
340 }
341 }
342 if (!newline) {
343 i %= 16;
344 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800345 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700346 ascii[i] = ' ';
347 i++;
348 }
Pekka Enberg06428782008-01-07 23:20:27 -0800349 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700350 }
351}
352
Christoph Lameter81819f02007-05-06 14:49:36 -0700353static struct track *get_track(struct kmem_cache *s, void *object,
354 enum track_item alloc)
355{
356 struct track *p;
357
358 if (s->offset)
359 p = object + s->offset + sizeof(void *);
360 else
361 p = object + s->inuse;
362
363 return p + alloc;
364}
365
366static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300367 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700368{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900369 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700370
Christoph Lameter81819f02007-05-06 14:49:36 -0700371 if (addr) {
372 p->addr = addr;
373 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400374 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700375 p->when = jiffies;
376 } else
377 memset(p, 0, sizeof(struct track));
378}
379
Christoph Lameter81819f02007-05-06 14:49:36 -0700380static void init_tracking(struct kmem_cache *s, void *object)
381{
Christoph Lameter24922682007-07-17 04:03:18 -0700382 if (!(s->flags & SLAB_STORE_USER))
383 return;
384
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300385 set_track(s, object, TRACK_FREE, 0UL);
386 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700387}
388
389static void print_track(const char *s, struct track *t)
390{
391 if (!t->addr)
392 return;
393
Linus Torvalds7daf7052008-07-14 12:12:53 -0700394 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300395 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700396}
397
Christoph Lameter24922682007-07-17 04:03:18 -0700398static void print_tracking(struct kmem_cache *s, void *object)
399{
400 if (!(s->flags & SLAB_STORE_USER))
401 return;
402
403 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
404 print_track("Freed", get_track(s, object, TRACK_FREE));
405}
406
407static void print_page_info(struct page *page)
408{
Christoph Lameter39b26462008-04-14 19:11:30 +0300409 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
410 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700411
412}
413
414static void slab_bug(struct kmem_cache *s, char *fmt, ...)
415{
416 va_list args;
417 char buf[100];
418
419 va_start(args, fmt);
420 vsnprintf(buf, sizeof(buf), fmt, args);
421 va_end(args);
422 printk(KERN_ERR "========================================"
423 "=====================================\n");
424 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
425 printk(KERN_ERR "----------------------------------------"
426 "-------------------------------------\n\n");
427}
428
429static void slab_fix(struct kmem_cache *s, char *fmt, ...)
430{
431 va_list args;
432 char buf[100];
433
434 va_start(args, fmt);
435 vsnprintf(buf, sizeof(buf), fmt, args);
436 va_end(args);
437 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
438}
439
440static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700441{
442 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800443 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700444
445 print_tracking(s, p);
446
447 print_page_info(page);
448
449 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
450 p, p - addr, get_freepointer(s, p));
451
452 if (p > addr + 16)
453 print_section("Bytes b4", p - 16, 16);
454
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300455 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700456
457 if (s->flags & SLAB_RED_ZONE)
458 print_section("Redzone", p + s->objsize,
459 s->inuse - s->objsize);
460
Christoph Lameter81819f02007-05-06 14:49:36 -0700461 if (s->offset)
462 off = s->offset + sizeof(void *);
463 else
464 off = s->inuse;
465
Christoph Lameter24922682007-07-17 04:03:18 -0700466 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700467 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700468
469 if (off != s->size)
470 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700471 print_section("Padding", p + off, s->size - off);
472
473 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700474}
475
476static void object_err(struct kmem_cache *s, struct page *page,
477 u8 *object, char *reason)
478{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700479 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700480 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700481}
482
Christoph Lameter24922682007-07-17 04:03:18 -0700483static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700484{
485 va_list args;
486 char buf[100];
487
Christoph Lameter24922682007-07-17 04:03:18 -0700488 va_start(args, fmt);
489 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700490 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700491 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700492 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700493 dump_stack();
494}
495
496static void init_object(struct kmem_cache *s, void *object, int active)
497{
498 u8 *p = object;
499
500 if (s->flags & __OBJECT_POISON) {
501 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800502 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700503 }
504
505 if (s->flags & SLAB_RED_ZONE)
506 memset(p + s->objsize,
507 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
508 s->inuse - s->objsize);
509}
510
Christoph Lameter24922682007-07-17 04:03:18 -0700511static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700512{
513 while (bytes) {
514 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700515 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700516 start++;
517 bytes--;
518 }
Christoph Lameter24922682007-07-17 04:03:18 -0700519 return NULL;
520}
521
522static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
523 void *from, void *to)
524{
525 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
526 memset(from, data, to - from);
527}
528
529static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
530 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800531 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700532{
533 u8 *fault;
534 u8 *end;
535
536 fault = check_bytes(start, value, bytes);
537 if (!fault)
538 return 1;
539
540 end = start + bytes;
541 while (end > fault && end[-1] == value)
542 end--;
543
544 slab_bug(s, "%s overwritten", what);
545 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
546 fault, end - 1, fault[0], value);
547 print_trailer(s, page, object);
548
549 restore_bytes(s, what, value, fault, end);
550 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700551}
552
Christoph Lameter81819f02007-05-06 14:49:36 -0700553/*
554 * Object layout:
555 *
556 * object address
557 * Bytes of the object to be managed.
558 * If the freepointer may overlay the object then the free
559 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700560 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700561 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
562 * 0xa5 (POISON_END)
563 *
564 * object + s->objsize
565 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700566 * Padding is extended by another word if Redzoning is enabled and
567 * objsize == inuse.
568 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700569 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
570 * 0xcc (RED_ACTIVE) for objects in use.
571 *
572 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700573 * Meta data starts here.
574 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700575 * A. Free pointer (if we cannot overwrite object on free)
576 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700577 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800578 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700579 * before the word boundary.
580 *
581 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700582 *
583 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700584 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700585 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700586 * If slabcaches are merged then the objsize and inuse boundaries are mostly
587 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700588 * may be used with merged slabcaches.
589 */
590
Christoph Lameter81819f02007-05-06 14:49:36 -0700591static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
592{
593 unsigned long off = s->inuse; /* The end of info */
594
595 if (s->offset)
596 /* Freepointer is placed after the object. */
597 off += sizeof(void *);
598
599 if (s->flags & SLAB_STORE_USER)
600 /* We also have user information there */
601 off += 2 * sizeof(struct track);
602
603 if (s->size == off)
604 return 1;
605
Christoph Lameter24922682007-07-17 04:03:18 -0700606 return check_bytes_and_report(s, page, p, "Object padding",
607 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700608}
609
Christoph Lameter39b26462008-04-14 19:11:30 +0300610/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700611static int slab_pad_check(struct kmem_cache *s, struct page *page)
612{
Christoph Lameter24922682007-07-17 04:03:18 -0700613 u8 *start;
614 u8 *fault;
615 u8 *end;
616 int length;
617 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700618
619 if (!(s->flags & SLAB_POISON))
620 return 1;
621
Christoph Lametera973e9d2008-03-01 13:40:44 -0800622 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300623 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300624 end = start + length;
625 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700626 if (!remainder)
627 return 1;
628
Christoph Lameter39b26462008-04-14 19:11:30 +0300629 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700630 if (!fault)
631 return 1;
632 while (end > fault && end[-1] == POISON_INUSE)
633 end--;
634
635 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300636 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700637
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200638 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700639 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700640}
641
642static int check_object(struct kmem_cache *s, struct page *page,
643 void *object, int active)
644{
645 u8 *p = object;
646 u8 *endobject = object + s->objsize;
647
648 if (s->flags & SLAB_RED_ZONE) {
649 unsigned int red =
650 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
651
Christoph Lameter24922682007-07-17 04:03:18 -0700652 if (!check_bytes_and_report(s, page, object, "Redzone",
653 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700654 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700655 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800656 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
657 check_bytes_and_report(s, page, p, "Alignment padding",
658 endobject, POISON_INUSE, s->inuse - s->objsize);
659 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700660 }
661
662 if (s->flags & SLAB_POISON) {
663 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700664 (!check_bytes_and_report(s, page, p, "Poison", p,
665 POISON_FREE, s->objsize - 1) ||
666 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800667 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700668 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700669 /*
670 * check_pad_bytes cleans up on its own.
671 */
672 check_pad_bytes(s, page, p);
673 }
674
675 if (!s->offset && active)
676 /*
677 * Object and freepointer overlap. Cannot check
678 * freepointer while object is allocated.
679 */
680 return 1;
681
682 /* Check free pointer validity */
683 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
684 object_err(s, page, p, "Freepointer corrupt");
685 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100686 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700687 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700688 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700689 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800690 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700691 return 0;
692 }
693 return 1;
694}
695
696static int check_slab(struct kmem_cache *s, struct page *page)
697{
Christoph Lameter39b26462008-04-14 19:11:30 +0300698 int maxobj;
699
Christoph Lameter81819f02007-05-06 14:49:36 -0700700 VM_BUG_ON(!irqs_disabled());
701
702 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700703 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700704 return 0;
705 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300706
707 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
708 if (page->objects > maxobj) {
709 slab_err(s, page, "objects %u > max %u",
710 s->name, page->objects, maxobj);
711 return 0;
712 }
713 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700714 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300715 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700716 return 0;
717 }
718 /* Slab_pad_check fixes things up after itself */
719 slab_pad_check(s, page);
720 return 1;
721}
722
723/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700724 * Determine if a certain object on a page is on the freelist. Must hold the
725 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700726 */
727static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
728{
729 int nr = 0;
730 void *fp = page->freelist;
731 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300732 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700733
Christoph Lameter39b26462008-04-14 19:11:30 +0300734 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700735 if (fp == search)
736 return 1;
737 if (!check_valid_pointer(s, page, fp)) {
738 if (object) {
739 object_err(s, page, object,
740 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800741 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700742 break;
743 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700744 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800745 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300746 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700747 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700748 return 0;
749 }
750 break;
751 }
752 object = fp;
753 fp = get_freepointer(s, object);
754 nr++;
755 }
756
Christoph Lameter224a88b2008-04-14 19:11:31 +0300757 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400758 if (max_objects > MAX_OBJS_PER_PAGE)
759 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300760
761 if (page->objects != max_objects) {
762 slab_err(s, page, "Wrong number of objects. Found %d but "
763 "should be %d", page->objects, max_objects);
764 page->objects = max_objects;
765 slab_fix(s, "Number of objects adjusted.");
766 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300767 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700768 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300769 "counted were %d", page->inuse, page->objects - nr);
770 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700771 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700772 }
773 return search == NULL;
774}
775
Christoph Lameter0121c6192008-04-29 16:11:12 -0700776static void trace(struct kmem_cache *s, struct page *page, void *object,
777 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700778{
779 if (s->flags & SLAB_TRACE) {
780 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
781 s->name,
782 alloc ? "alloc" : "free",
783 object, page->inuse,
784 page->freelist);
785
786 if (!alloc)
787 print_section("Object", (void *)object, s->objsize);
788
789 dump_stack();
790 }
791}
792
Christoph Lameter643b1132007-05-06 14:49:42 -0700793/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500794 * Hooks for other subsystems that check memory allocations. In a typical
795 * production configuration these hooks all should produce no code at all.
796 */
797static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
798{
799 lockdep_trace_alloc(flags);
800 might_sleep_if(flags & __GFP_WAIT);
801
802 return should_failslab(s->objsize, flags, s->flags);
803}
804
805static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
806{
807 kmemcheck_slab_alloc(s, flags, object, s->objsize);
808 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
809}
810
811static inline void slab_free_hook(struct kmem_cache *s, void *x)
812{
813 kmemleak_free_recursive(x, s->flags);
814}
815
816static inline void slab_free_hook_irq(struct kmem_cache *s, void *object)
817{
818 kmemcheck_slab_free(s, object, s->objsize);
819 debug_check_no_locks_freed(object, s->objsize);
820 if (!(s->flags & SLAB_DEBUG_OBJECTS))
821 debug_check_no_obj_freed(object, s->objsize);
822}
823
824/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700825 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700826 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700827static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700828{
Christoph Lameter643b1132007-05-06 14:49:42 -0700829 spin_lock(&n->list_lock);
830 list_add(&page->lru, &n->full);
831 spin_unlock(&n->list_lock);
832}
833
834static void remove_full(struct kmem_cache *s, struct page *page)
835{
836 struct kmem_cache_node *n;
837
838 if (!(s->flags & SLAB_STORE_USER))
839 return;
840
841 n = get_node(s, page_to_nid(page));
842
843 spin_lock(&n->list_lock);
844 list_del(&page->lru);
845 spin_unlock(&n->list_lock);
846}
847
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300848/* Tracking of the number of slabs for debugging purposes */
849static inline unsigned long slabs_node(struct kmem_cache *s, int node)
850{
851 struct kmem_cache_node *n = get_node(s, node);
852
853 return atomic_long_read(&n->nr_slabs);
854}
855
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400856static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
857{
858 return atomic_long_read(&n->nr_slabs);
859}
860
Christoph Lameter205ab992008-04-14 19:11:40 +0300861static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300862{
863 struct kmem_cache_node *n = get_node(s, node);
864
865 /*
866 * May be called early in order to allocate a slab for the
867 * kmem_cache_node structure. Solve the chicken-egg
868 * dilemma by deferring the increment of the count during
869 * bootstrap (see early_kmem_cache_node_alloc).
870 */
Christoph Lameter205ab992008-04-14 19:11:40 +0300871 if (!NUMA_BUILD || n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300872 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300873 atomic_long_add(objects, &n->total_objects);
874 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300875}
Christoph Lameter205ab992008-04-14 19:11:40 +0300876static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300877{
878 struct kmem_cache_node *n = get_node(s, node);
879
880 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300881 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300882}
883
884/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700885static void setup_object_debug(struct kmem_cache *s, struct page *page,
886 void *object)
887{
888 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
889 return;
890
891 init_object(s, object, 0);
892 init_tracking(s, object);
893}
894
Christoph Lameter15370662010-08-20 12:37:12 -0500895static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300896 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700897{
898 if (!check_slab(s, page))
899 goto bad;
900
Christoph Lameterd692ef62008-02-15 23:45:24 -0800901 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700902 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700903 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700904 }
905
906 if (!check_valid_pointer(s, page, object)) {
907 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700908 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700909 }
910
Christoph Lameterd692ef62008-02-15 23:45:24 -0800911 if (!check_object(s, page, object, 0))
Christoph Lameter81819f02007-05-06 14:49:36 -0700912 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700913
Christoph Lameter3ec09742007-05-16 22:11:00 -0700914 /* Success perform special debug activities for allocs */
915 if (s->flags & SLAB_STORE_USER)
916 set_track(s, object, TRACK_ALLOC, addr);
917 trace(s, page, object, 1);
918 init_object(s, object, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700919 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700920
Christoph Lameter81819f02007-05-06 14:49:36 -0700921bad:
922 if (PageSlab(page)) {
923 /*
924 * If this is a slab page then lets do the best we can
925 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700926 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700927 */
Christoph Lameter24922682007-07-17 04:03:18 -0700928 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300929 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800930 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700931 }
932 return 0;
933}
934
Christoph Lameter15370662010-08-20 12:37:12 -0500935static noinline int free_debug_processing(struct kmem_cache *s,
936 struct page *page, void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700937{
938 if (!check_slab(s, page))
939 goto fail;
940
941 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700942 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700943 goto fail;
944 }
945
946 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700947 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700948 goto fail;
949 }
950
951 if (!check_object(s, page, object, 1))
952 return 0;
953
954 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800955 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700956 slab_err(s, page, "Attempt to free object(0x%p) "
957 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800958 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700959 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700960 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700961 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700962 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800963 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700964 object_err(s, page, object,
965 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700966 goto fail;
967 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700968
969 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700970 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700971 remove_full(s, page);
972 if (s->flags & SLAB_STORE_USER)
973 set_track(s, object, TRACK_FREE, addr);
974 trace(s, page, object, 0);
975 init_object(s, object, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -0700976 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700977
Christoph Lameter81819f02007-05-06 14:49:36 -0700978fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700979 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700980 return 0;
981}
982
Christoph Lameter41ecc552007-05-09 02:32:44 -0700983static int __init setup_slub_debug(char *str)
984{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700985 slub_debug = DEBUG_DEFAULT_FLAGS;
986 if (*str++ != '=' || !*str)
987 /*
988 * No options specified. Switch on full debugging.
989 */
990 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700991
992 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700993 /*
994 * No options but restriction on slabs. This means full
995 * debugging for slabs matching a pattern.
996 */
997 goto check_slabs;
998
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700999 if (tolower(*str) == 'o') {
1000 /*
1001 * Avoid enabling debugging on caches if its minimum order
1002 * would increase as a result.
1003 */
1004 disable_higher_order_debug = 1;
1005 goto out;
1006 }
1007
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001008 slub_debug = 0;
1009 if (*str == '-')
1010 /*
1011 * Switch off all debugging measures.
1012 */
1013 goto out;
1014
1015 /*
1016 * Determine which debug features should be switched on
1017 */
Pekka Enberg06428782008-01-07 23:20:27 -08001018 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001019 switch (tolower(*str)) {
1020 case 'f':
1021 slub_debug |= SLAB_DEBUG_FREE;
1022 break;
1023 case 'z':
1024 slub_debug |= SLAB_RED_ZONE;
1025 break;
1026 case 'p':
1027 slub_debug |= SLAB_POISON;
1028 break;
1029 case 'u':
1030 slub_debug |= SLAB_STORE_USER;
1031 break;
1032 case 't':
1033 slub_debug |= SLAB_TRACE;
1034 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001035 case 'a':
1036 slub_debug |= SLAB_FAILSLAB;
1037 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001038 default:
1039 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001040 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001041 }
1042 }
1043
1044check_slabs:
1045 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001046 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001047out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001048 return 1;
1049}
1050
1051__setup("slub_debug", setup_slub_debug);
1052
Christoph Lameterba0268a2007-09-11 15:24:11 -07001053static unsigned long kmem_cache_flags(unsigned long objsize,
1054 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001055 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001056{
1057 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001058 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001059 */
Christoph Lametere1533622008-02-15 23:45:24 -08001060 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001061 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1062 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001063
1064 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001065}
1066#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001067static inline void setup_object_debug(struct kmem_cache *s,
1068 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001069
Christoph Lameter3ec09742007-05-16 22:11:00 -07001070static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001071 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001072
Christoph Lameter3ec09742007-05-16 22:11:00 -07001073static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001074 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001075
Christoph Lameter41ecc552007-05-09 02:32:44 -07001076static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1077 { return 1; }
1078static inline int check_object(struct kmem_cache *s, struct page *page,
1079 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001080static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001081static inline unsigned long kmem_cache_flags(unsigned long objsize,
1082 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001083 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001084{
1085 return flags;
1086}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001087#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001088
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001089#define disable_higher_order_debug 0
1090
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001091static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1092 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001093static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1094 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001095static inline void inc_slabs_node(struct kmem_cache *s, int node,
1096 int objects) {}
1097static inline void dec_slabs_node(struct kmem_cache *s, int node,
1098 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001099#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001100
Christoph Lameter81819f02007-05-06 14:49:36 -07001101/*
1102 * Slab allocation and freeing
1103 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001104static inline struct page *alloc_slab_page(gfp_t flags, int node,
1105 struct kmem_cache_order_objects oo)
1106{
1107 int order = oo_order(oo);
1108
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001109 flags |= __GFP_NOTRACK;
1110
Christoph Lameter2154a332010-07-09 14:07:10 -05001111 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001112 return alloc_pages(flags, order);
1113 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001114 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001115}
1116
Christoph Lameter81819f02007-05-06 14:49:36 -07001117static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1118{
Pekka Enberg06428782008-01-07 23:20:27 -08001119 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001120 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001121 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001122
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001123 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001124
Pekka Enbergba522702009-06-24 21:59:51 +03001125 /*
1126 * Let the initial higher-order allocation fail under memory pressure
1127 * so we fall-back to the minimum order allocation.
1128 */
1129 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1130
1131 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001132 if (unlikely(!page)) {
1133 oo = s->min;
1134 /*
1135 * Allocation may have failed due to fragmentation.
1136 * Try a lower order alloc if possible
1137 */
1138 page = alloc_slab_page(flags, node, oo);
1139 if (!page)
1140 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001141
Christoph Lameter84e554e2009-12-18 16:26:23 -06001142 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001143 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001144
1145 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001146 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001147 int pages = 1 << oo_order(oo);
1148
1149 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1150
1151 /*
1152 * Objects from caches that have a constructor don't get
1153 * cleared when they're allocated, so we need to do it here.
1154 */
1155 if (s->ctor)
1156 kmemcheck_mark_uninitialized_pages(page, pages);
1157 else
1158 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001159 }
1160
Christoph Lameter834f3d12008-04-14 19:11:31 +03001161 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001162 mod_zone_page_state(page_zone(page),
1163 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1164 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001165 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001166
1167 return page;
1168}
1169
1170static void setup_object(struct kmem_cache *s, struct page *page,
1171 void *object)
1172{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001173 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001174 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001175 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001176}
1177
1178static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1179{
1180 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001181 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001182 void *last;
1183 void *p;
1184
Christoph Lameter6cb06222007-10-16 01:25:41 -07001185 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001186
Christoph Lameter6cb06222007-10-16 01:25:41 -07001187 page = allocate_slab(s,
1188 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001189 if (!page)
1190 goto out;
1191
Christoph Lameter205ab992008-04-14 19:11:40 +03001192 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001193 page->slab = s;
1194 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001195
1196 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001197
1198 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001199 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001200
1201 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001202 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001203 setup_object(s, page, last);
1204 set_freepointer(s, last, p);
1205 last = p;
1206 }
1207 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001208 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001209
1210 page->freelist = start;
1211 page->inuse = 0;
1212out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001213 return page;
1214}
1215
1216static void __free_slab(struct kmem_cache *s, struct page *page)
1217{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001218 int order = compound_order(page);
1219 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001220
Christoph Lameteraf537b02010-07-09 14:07:14 -05001221 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001222 void *p;
1223
1224 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001225 for_each_object(p, s, page_address(page),
1226 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001227 check_object(s, page, p, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -07001228 }
1229
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001230 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001231
Christoph Lameter81819f02007-05-06 14:49:36 -07001232 mod_zone_page_state(page_zone(page),
1233 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1234 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001235 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001236
Christoph Lameter49bd5222008-04-14 18:52:18 +03001237 __ClearPageSlab(page);
1238 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001239 if (current->reclaim_state)
1240 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001241 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001242}
1243
1244static void rcu_free_slab(struct rcu_head *h)
1245{
1246 struct page *page;
1247
1248 page = container_of((struct list_head *)h, struct page, lru);
1249 __free_slab(page->slab, page);
1250}
1251
1252static void free_slab(struct kmem_cache *s, struct page *page)
1253{
1254 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1255 /*
1256 * RCU free overloads the RCU head over the LRU
1257 */
1258 struct rcu_head *head = (void *)&page->lru;
1259
1260 call_rcu(head, rcu_free_slab);
1261 } else
1262 __free_slab(s, page);
1263}
1264
1265static void discard_slab(struct kmem_cache *s, struct page *page)
1266{
Christoph Lameter205ab992008-04-14 19:11:40 +03001267 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001268 free_slab(s, page);
1269}
1270
1271/*
1272 * Per slab locking using the pagelock
1273 */
1274static __always_inline void slab_lock(struct page *page)
1275{
1276 bit_spin_lock(PG_locked, &page->flags);
1277}
1278
1279static __always_inline void slab_unlock(struct page *page)
1280{
Nick Piggina76d3542008-01-07 23:20:27 -08001281 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001282}
1283
1284static __always_inline int slab_trylock(struct page *page)
1285{
1286 int rc = 1;
1287
1288 rc = bit_spin_trylock(PG_locked, &page->flags);
1289 return rc;
1290}
1291
1292/*
1293 * Management of partially allocated slabs
1294 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001295static void add_partial(struct kmem_cache_node *n,
1296 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001297{
Christoph Lametere95eed52007-05-06 14:49:44 -07001298 spin_lock(&n->list_lock);
1299 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001300 if (tail)
1301 list_add_tail(&page->lru, &n->partial);
1302 else
1303 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001304 spin_unlock(&n->list_lock);
1305}
1306
Christoph Lameter0121c6192008-04-29 16:11:12 -07001307static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001308{
1309 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1310
1311 spin_lock(&n->list_lock);
1312 list_del(&page->lru);
1313 n->nr_partial--;
1314 spin_unlock(&n->list_lock);
1315}
1316
1317/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001318 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001319 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001320 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001321 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001322static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1323 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001324{
1325 if (slab_trylock(page)) {
1326 list_del(&page->lru);
1327 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001328 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001329 return 1;
1330 }
1331 return 0;
1332}
1333
1334/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001335 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001336 */
1337static struct page *get_partial_node(struct kmem_cache_node *n)
1338{
1339 struct page *page;
1340
1341 /*
1342 * Racy check. If we mistakenly see no partial slabs then we
1343 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001344 * partial slab and there is none available then get_partials()
1345 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001346 */
1347 if (!n || !n->nr_partial)
1348 return NULL;
1349
1350 spin_lock(&n->list_lock);
1351 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001352 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001353 goto out;
1354 page = NULL;
1355out:
1356 spin_unlock(&n->list_lock);
1357 return page;
1358}
1359
1360/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001361 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001362 */
1363static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1364{
1365#ifdef CONFIG_NUMA
1366 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001367 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001368 struct zone *zone;
1369 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001370 struct page *page;
1371
1372 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001373 * The defrag ratio allows a configuration of the tradeoffs between
1374 * inter node defragmentation and node local allocations. A lower
1375 * defrag_ratio increases the tendency to do local allocations
1376 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001377 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001378 * If the defrag_ratio is set to 0 then kmalloc() always
1379 * returns node local objects. If the ratio is higher then kmalloc()
1380 * may return off node objects because partial slabs are obtained
1381 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001382 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001383 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001384 * defrag_ratio = 1000) then every (well almost) allocation will
1385 * first attempt to defrag slab caches on other nodes. This means
1386 * scanning over all nodes to look for partial slabs which may be
1387 * expensive if we do it every time we are trying to find a slab
1388 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001389 */
Christoph Lameter98246012008-01-07 23:20:26 -08001390 if (!s->remote_node_defrag_ratio ||
1391 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001392 return NULL;
1393
Miao Xiec0ff7452010-05-24 14:32:08 -07001394 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001395 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001396 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001397 struct kmem_cache_node *n;
1398
Mel Gorman54a6eb52008-04-28 02:12:16 -07001399 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001400
Mel Gorman54a6eb52008-04-28 02:12:16 -07001401 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001402 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001403 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001404 if (page) {
1405 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001406 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001407 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001408 }
1409 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001410 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001411#endif
1412 return NULL;
1413}
1414
1415/*
1416 * Get a partial page, lock it and return it.
1417 */
1418static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1419{
1420 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001421 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001422
1423 page = get_partial_node(get_node(s, searchnode));
Christoph Lameterbc6488e2010-07-26 10:41:14 -05001424 if (page || node != -1)
Christoph Lameter81819f02007-05-06 14:49:36 -07001425 return page;
1426
1427 return get_any_partial(s, flags);
1428}
1429
1430/*
1431 * Move a page back to the lists.
1432 *
1433 * Must be called with the slab lock held.
1434 *
1435 * On exit the slab lock will have been dropped.
1436 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001437static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001438{
Christoph Lametere95eed52007-05-06 14:49:44 -07001439 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1440
Andy Whitcroft8a380822008-07-23 21:27:18 -07001441 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001442 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001443
Christoph Lametera973e9d2008-03-01 13:40:44 -08001444 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001445 add_partial(n, page, tail);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001446 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001447 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001448 stat(s, DEACTIVATE_FULL);
Christoph Lameteraf537b02010-07-09 14:07:14 -05001449 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001450 add_full(n, page);
1451 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001452 slab_unlock(page);
1453 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001454 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001455 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001456 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001457 * Adding an empty slab to the partial slabs in order
1458 * to avoid page allocator overhead. This slab needs
1459 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001460 * so that the others get filled first. That way the
1461 * size of the partial list stays small.
1462 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001463 * kmem_cache_shrink can reclaim any empty slabs from
1464 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001465 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001466 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001467 slab_unlock(page);
1468 } else {
1469 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001470 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001471 discard_slab(s, page);
1472 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001473 }
1474}
1475
1476/*
1477 * Remove the cpu slab
1478 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001479static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001480{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001481 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001482 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001483
Christoph Lameterb773ad72008-03-04 11:10:17 -08001484 if (page->freelist)
Christoph Lameter84e554e2009-12-18 16:26:23 -06001485 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001486 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001487 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001488 * because both freelists are empty. So this is unlikely
1489 * to occur.
1490 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001491 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001492 void **object;
1493
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001494 tail = 0; /* Hot objects. Put the slab first */
1495
Christoph Lameter894b8782007-05-10 03:15:16 -07001496 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001497 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001498 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001499
1500 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001501 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001502 page->freelist = object;
1503 page->inuse--;
1504 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001505 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001506 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001507}
1508
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001509static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001510{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001511 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001512 slab_lock(c->page);
1513 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001514}
1515
1516/*
1517 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001518 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001519 * Called from IPI handler with interrupts disabled.
1520 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001521static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001522{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001523 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001524
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001525 if (likely(c && c->page))
1526 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001527}
1528
1529static void flush_cpu_slab(void *d)
1530{
1531 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001532
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001533 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001534}
1535
1536static void flush_all(struct kmem_cache *s)
1537{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001538 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001539}
1540
1541/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001542 * Check if the objects in a per cpu structure fit numa
1543 * locality expectations.
1544 */
1545static inline int node_match(struct kmem_cache_cpu *c, int node)
1546{
1547#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001548 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001549 return 0;
1550#endif
1551 return 1;
1552}
1553
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001554static int count_free(struct page *page)
1555{
1556 return page->objects - page->inuse;
1557}
1558
1559static unsigned long count_partial(struct kmem_cache_node *n,
1560 int (*get_count)(struct page *))
1561{
1562 unsigned long flags;
1563 unsigned long x = 0;
1564 struct page *page;
1565
1566 spin_lock_irqsave(&n->list_lock, flags);
1567 list_for_each_entry(page, &n->partial, lru)
1568 x += get_count(page);
1569 spin_unlock_irqrestore(&n->list_lock, flags);
1570 return x;
1571}
1572
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001573static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1574{
1575#ifdef CONFIG_SLUB_DEBUG
1576 return atomic_long_read(&n->total_objects);
1577#else
1578 return 0;
1579#endif
1580}
1581
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001582static noinline void
1583slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1584{
1585 int node;
1586
1587 printk(KERN_WARNING
1588 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1589 nid, gfpflags);
1590 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1591 "default order: %d, min order: %d\n", s->name, s->objsize,
1592 s->size, oo_order(s->oo), oo_order(s->min));
1593
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001594 if (oo_order(s->min) > get_order(s->objsize))
1595 printk(KERN_WARNING " %s debugging increased min order, use "
1596 "slub_debug=O to disable.\n", s->name);
1597
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001598 for_each_online_node(node) {
1599 struct kmem_cache_node *n = get_node(s, node);
1600 unsigned long nr_slabs;
1601 unsigned long nr_objs;
1602 unsigned long nr_free;
1603
1604 if (!n)
1605 continue;
1606
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001607 nr_free = count_partial(n, count_free);
1608 nr_slabs = node_nr_slabs(n);
1609 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001610
1611 printk(KERN_WARNING
1612 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1613 node, nr_slabs, nr_objs, nr_free);
1614 }
1615}
1616
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001617/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001618 * Slow path. The lockless freelist is empty or we need to perform
1619 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001620 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001621 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001622 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001623 * Processing is still very fast if new objects have been freed to the
1624 * regular freelist. In that case we simply take over the regular freelist
1625 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001626 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001627 * If that is not working then we fall back to the partial lists. We take the
1628 * first element of the freelist as the object to allocate now and move the
1629 * rest of the freelist to the lockless freelist.
1630 *
1631 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001632 * we need to allocate a new slab. This is the slowest path since it involves
1633 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001634 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001635static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1636 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001637{
Christoph Lameter81819f02007-05-06 14:49:36 -07001638 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001639 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001640
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001641 /* We handle __GFP_ZERO in the caller */
1642 gfpflags &= ~__GFP_ZERO;
1643
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001644 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001645 goto new_slab;
1646
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001647 slab_lock(c->page);
1648 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001649 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001650
Christoph Lameter84e554e2009-12-18 16:26:23 -06001651 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001652
Christoph Lameter894b8782007-05-10 03:15:16 -07001653load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001654 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001655 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001656 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001657 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001658 goto debug;
1659
Christoph Lameterff120592009-12-18 16:26:22 -06001660 c->freelist = get_freepointer(s, object);
Christoph Lameter39b26462008-04-14 19:11:30 +03001661 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001662 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001663 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001664unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001665 slab_unlock(c->page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001666 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001667 return object;
1668
1669another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001670 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001671
1672new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001673 new = get_partial(s, gfpflags, node);
1674 if (new) {
1675 c->page = new;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001676 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001677 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001678 }
1679
Christoph Lameterb811c202007-10-16 23:25:51 -07001680 if (gfpflags & __GFP_WAIT)
1681 local_irq_enable();
1682
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001683 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001684
1685 if (gfpflags & __GFP_WAIT)
1686 local_irq_disable();
1687
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001688 if (new) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001689 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001690 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001691 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001692 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001693 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001694 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001695 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001696 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001697 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001698 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1699 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001700 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001701debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001702 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001703 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001704
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001705 c->page->inuse++;
Christoph Lameterff120592009-12-18 16:26:22 -06001706 c->page->freelist = get_freepointer(s, object);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001707 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001708 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001709}
1710
1711/*
1712 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1713 * have the fastpath folded into their functions. So no function call
1714 * overhead for requests that can be satisfied on the fastpath.
1715 *
1716 * The fastpath works by first checking if the lockless freelist can be used.
1717 * If not then __slab_alloc is called for slow processing.
1718 *
1719 * Otherwise we can simply pick the next object from the lockless free list.
1720 */
Pekka Enberg06428782008-01-07 23:20:27 -08001721static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001722 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001723{
Christoph Lameter894b8782007-05-10 03:15:16 -07001724 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001725 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001726 unsigned long flags;
1727
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001728 gfpflags &= gfp_allowed_mask;
Pekka Enberg7e85ee02009-06-12 14:03:06 +03001729
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001730 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001731 return NULL;
1732
Christoph Lameter894b8782007-05-10 03:15:16 -07001733 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001734 c = __this_cpu_ptr(s->cpu_slab);
1735 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001736 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001737
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001738 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001739
1740 else {
Christoph Lameterff120592009-12-18 16:26:22 -06001741 c->freelist = get_freepointer(s, object);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001742 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001743 }
1744 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001745
Pekka Enberg74e21342009-11-25 20:14:48 +02001746 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001747 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001748
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001749 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001750
Christoph Lameter894b8782007-05-10 03:15:16 -07001751 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001752}
1753
1754void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1755{
Christoph Lameter2154a332010-07-09 14:07:10 -05001756 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001757
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001758 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001759
1760 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001761}
1762EXPORT_SYMBOL(kmem_cache_alloc);
1763
Li Zefan0f24f122009-12-11 15:45:30 +08001764#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001765void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1766{
Christoph Lameter2154a332010-07-09 14:07:10 -05001767 return slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001768}
1769EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1770#endif
1771
Christoph Lameter81819f02007-05-06 14:49:36 -07001772#ifdef CONFIG_NUMA
1773void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1774{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001775 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1776
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001777 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1778 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001779
1780 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001781}
1782EXPORT_SYMBOL(kmem_cache_alloc_node);
1783#endif
1784
Li Zefan0f24f122009-12-11 15:45:30 +08001785#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001786void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1787 gfp_t gfpflags,
1788 int node)
1789{
1790 return slab_alloc(s, gfpflags, node, _RET_IP_);
1791}
1792EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1793#endif
1794
Christoph Lameter81819f02007-05-06 14:49:36 -07001795/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001796 * Slow patch handling. This may still be called frequently since objects
1797 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001798 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001799 * So we still attempt to reduce cache line usage. Just take the slab
1800 * lock and free the item. If there is no additional partial page
1801 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001802 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001803static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06001804 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001805{
1806 void *prior;
1807 void **object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07001808
Christoph Lameter84e554e2009-12-18 16:26:23 -06001809 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001810 slab_lock(page);
1811
Christoph Lameteraf537b02010-07-09 14:07:14 -05001812 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001813 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001814
Christoph Lameter81819f02007-05-06 14:49:36 -07001815checks_ok:
Christoph Lameterff120592009-12-18 16:26:22 -06001816 prior = page->freelist;
1817 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07001818 page->freelist = object;
1819 page->inuse--;
1820
Andy Whitcroft8a380822008-07-23 21:27:18 -07001821 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001822 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001823 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001824 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001825
1826 if (unlikely(!page->inuse))
1827 goto slab_empty;
1828
1829 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001830 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001831 * then add it.
1832 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001833 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001834 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001835 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001836 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001837
1838out_unlock:
1839 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001840 return;
1841
1842slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001843 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001844 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001845 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001846 */
1847 remove_partial(s, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001848 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001849 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001850 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001851 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001852 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001853 return;
1854
1855debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001856 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001857 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001858 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001859}
1860
Christoph Lameter894b8782007-05-10 03:15:16 -07001861/*
1862 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1863 * can perform fastpath freeing without additional function calls.
1864 *
1865 * The fastpath is only possible if we are freeing to the current cpu slab
1866 * of this processor. This typically the case if we have just allocated
1867 * the item before.
1868 *
1869 * If fastpath is not possible then fall back to __slab_free where we deal
1870 * with all sorts of special processing.
1871 */
Pekka Enberg06428782008-01-07 23:20:27 -08001872static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001873 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001874{
1875 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001876 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001877 unsigned long flags;
1878
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001879 slab_free_hook(s, x);
1880
Christoph Lameter894b8782007-05-10 03:15:16 -07001881 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001882 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05001883
1884 slab_free_hook_irq(s, x);
1885
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001886 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterff120592009-12-18 16:26:22 -06001887 set_freepointer(s, object, c->freelist);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001888 c->freelist = object;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001889 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001890 } else
Christoph Lameterff120592009-12-18 16:26:22 -06001891 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07001892
1893 local_irq_restore(flags);
1894}
1895
Christoph Lameter81819f02007-05-06 14:49:36 -07001896void kmem_cache_free(struct kmem_cache *s, void *x)
1897{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001898 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001899
Christoph Lameterb49af682007-05-06 14:49:41 -07001900 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001901
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001902 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001903
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001904 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001905}
1906EXPORT_SYMBOL(kmem_cache_free);
1907
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001908/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001909static struct page *get_object_page(const void *x)
1910{
Christoph Lameterb49af682007-05-06 14:49:41 -07001911 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001912
1913 if (!PageSlab(page))
1914 return NULL;
1915
1916 return page;
1917}
1918
1919/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001920 * Object placement in a slab is made very easy because we always start at
1921 * offset 0. If we tune the size of the object to the alignment then we can
1922 * get the required alignment by putting one properly sized object after
1923 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001924 *
1925 * Notice that the allocation order determines the sizes of the per cpu
1926 * caches. Each processor has always one slab available for allocations.
1927 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001928 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001929 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001930 */
1931
1932/*
1933 * Mininum / Maximum order of slab pages. This influences locking overhead
1934 * and slab fragmentation. A higher order reduces the number of partial slabs
1935 * and increases the number of allocations possible without having to
1936 * take the list_lock.
1937 */
1938static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001939static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001940static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001941
1942/*
1943 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001944 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001945 */
1946static int slub_nomerge;
1947
1948/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001949 * Calculate the order of allocation given an slab object size.
1950 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001951 * The order of allocation has significant impact on performance and other
1952 * system components. Generally order 0 allocations should be preferred since
1953 * order 0 does not cause fragmentation in the page allocator. Larger objects
1954 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001955 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001956 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001957 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001958 * In order to reach satisfactory performance we must ensure that a minimum
1959 * number of objects is in one slab. Otherwise we may generate too much
1960 * activity on the partial lists which requires taking the list_lock. This is
1961 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001962 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001963 * slub_max_order specifies the order where we begin to stop considering the
1964 * number of objects in a slab as critical. If we reach slub_max_order then
1965 * we try to keep the page order as low as possible. So we accept more waste
1966 * of space in favor of a small page order.
1967 *
1968 * Higher order allocations also allow the placement of more objects in a
1969 * slab and thereby reduce object handling overhead. If the user has
1970 * requested a higher mininum order then we start with that one instead of
1971 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001972 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001973static inline int slab_order(int size, int min_objects,
1974 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001975{
1976 int order;
1977 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001978 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001979
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001980 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1981 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001982
Christoph Lameter6300ea72007-07-17 04:03:20 -07001983 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001984 fls(min_objects * size - 1) - PAGE_SHIFT);
1985 order <= max_order; order++) {
1986
Christoph Lameter81819f02007-05-06 14:49:36 -07001987 unsigned long slab_size = PAGE_SIZE << order;
1988
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001989 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001990 continue;
1991
Christoph Lameter81819f02007-05-06 14:49:36 -07001992 rem = slab_size % size;
1993
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001994 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001995 break;
1996
1997 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001998
Christoph Lameter81819f02007-05-06 14:49:36 -07001999 return order;
2000}
2001
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002002static inline int calculate_order(int size)
2003{
2004 int order;
2005 int min_objects;
2006 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002007 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002008
2009 /*
2010 * Attempt to find best configuration for a slab. This
2011 * works by first attempting to generate a layout with
2012 * the best configuration and backing off gradually.
2013 *
2014 * First we reduce the acceptable waste in a slab. Then
2015 * we reduce the minimum objects required in a slab.
2016 */
2017 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002018 if (!min_objects)
2019 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002020 max_objects = (PAGE_SIZE << slub_max_order)/size;
2021 min_objects = min(min_objects, max_objects);
2022
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002023 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002024 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002025 while (fraction >= 4) {
2026 order = slab_order(size, min_objects,
2027 slub_max_order, fraction);
2028 if (order <= slub_max_order)
2029 return order;
2030 fraction /= 2;
2031 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002032 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002033 }
2034
2035 /*
2036 * We were unable to place multiple objects in a slab. Now
2037 * lets see if we can place a single object there.
2038 */
2039 order = slab_order(size, 1, slub_max_order, 1);
2040 if (order <= slub_max_order)
2041 return order;
2042
2043 /*
2044 * Doh this slab cannot be placed using slub_max_order.
2045 */
2046 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002047 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002048 return order;
2049 return -ENOSYS;
2050}
2051
Christoph Lameter81819f02007-05-06 14:49:36 -07002052/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002053 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002054 */
2055static unsigned long calculate_alignment(unsigned long flags,
2056 unsigned long align, unsigned long size)
2057{
2058 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002059 * If the user wants hardware cache aligned objects then follow that
2060 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002061 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002062 * The hardware cache alignment cannot override the specified
2063 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002064 */
Nick Pigginb6210382008-03-05 14:05:56 -08002065 if (flags & SLAB_HWCACHE_ALIGN) {
2066 unsigned long ralign = cache_line_size();
2067 while (size <= ralign / 2)
2068 ralign /= 2;
2069 align = max(align, ralign);
2070 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002071
2072 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002073 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002074
2075 return ALIGN(align, sizeof(void *));
2076}
2077
Pekka Enberg5595cff2008-08-05 09:28:47 +03002078static void
2079init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002080{
2081 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002082 spin_lock_init(&n->list_lock);
2083 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002084#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002085 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002086 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002087 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002088#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002089}
2090
Christoph Lameter55136592010-08-20 12:37:13 -05002091static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002092{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002093 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2094 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002095
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002096 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002097
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002098 return s->cpu_slab != NULL;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002099}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002100
Christoph Lameter81819f02007-05-06 14:49:36 -07002101#ifdef CONFIG_NUMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002102static struct kmem_cache *kmem_cache_node;
2103
Christoph Lameter81819f02007-05-06 14:49:36 -07002104/*
2105 * No kmalloc_node yet so do it by hand. We know that this is the first
2106 * slab on the node for this slabcache. There are no concurrent accesses
2107 * possible.
2108 *
2109 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002110 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2111 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002112 */
Christoph Lameter55136592010-08-20 12:37:13 -05002113static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002114{
2115 struct page *page;
2116 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002117 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002118
Christoph Lameter51df1142010-08-20 12:37:15 -05002119 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002120
Christoph Lameter51df1142010-08-20 12:37:15 -05002121 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002122
2123 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002124 if (page_to_nid(page) != node) {
2125 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2126 "node %d\n", node);
2127 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2128 "in order to be able to continue\n");
2129 }
2130
Christoph Lameter81819f02007-05-06 14:49:36 -07002131 n = page->freelist;
2132 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002133 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002134 page->inuse++;
Christoph Lameter51df1142010-08-20 12:37:15 -05002135 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002136#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05002137 init_object(kmem_cache_node, n, 1);
2138 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002139#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002140 init_kmem_cache_node(n, kmem_cache_node);
2141 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002142
rootba84c732008-01-07 23:20:28 -08002143 /*
2144 * lockdep requires consistent irq usage for each lock
2145 * so even though there cannot be a race this early in
2146 * the boot sequence, we still disable irqs.
2147 */
2148 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002149 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002150 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002151}
2152
2153static void free_kmem_cache_nodes(struct kmem_cache *s)
2154{
2155 int node;
2156
Christoph Lameterf64dc582007-10-16 01:25:33 -07002157 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002158 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002159
Alexander Duyck73367bd2010-05-21 14:41:35 -07002160 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002161 kmem_cache_free(kmem_cache_node, n);
2162
Christoph Lameter81819f02007-05-06 14:49:36 -07002163 s->node[node] = NULL;
2164 }
2165}
2166
Christoph Lameter55136592010-08-20 12:37:13 -05002167static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002168{
2169 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002170
Christoph Lameterf64dc582007-10-16 01:25:33 -07002171 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002172 struct kmem_cache_node *n;
2173
Alexander Duyck73367bd2010-05-21 14:41:35 -07002174 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002175 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002176 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002177 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002178 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002179 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002180
2181 if (!n) {
2182 free_kmem_cache_nodes(s);
2183 return 0;
2184 }
2185
Christoph Lameter81819f02007-05-06 14:49:36 -07002186 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002187 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002188 }
2189 return 1;
2190}
2191#else
2192static void free_kmem_cache_nodes(struct kmem_cache *s)
2193{
2194}
2195
Christoph Lameter55136592010-08-20 12:37:13 -05002196static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002197{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002198 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002199 return 1;
2200}
2201#endif
2202
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 Lameterc59def9f2007-05-16 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 Lameterc59def9f2007-05-16 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;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002438 long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long),
2439 GFP_ATOMIC);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002440
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002441 if (!map)
2442 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002443 slab_err(s, page, "%s", text);
2444 slab_lock(page);
2445 for_each_free_object(p, s, page->freelist)
2446 set_bit(slab_index(p, s, addr), map);
2447
2448 for_each_object(p, s, addr, page->objects) {
2449
2450 if (!test_bit(slab_index(p, s, addr), map)) {
2451 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2452 p, p - addr);
2453 print_tracking(s, p);
2454 }
2455 }
2456 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002457 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002458#endif
2459}
2460
Christoph Lameter81819f02007-05-06 14:49:36 -07002461/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002462 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002463 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002464static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002465{
Christoph Lameter81819f02007-05-06 14:49:36 -07002466 unsigned long flags;
2467 struct page *page, *h;
2468
2469 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002470 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002471 if (!page->inuse) {
2472 list_del(&page->lru);
2473 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002474 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002475 } else {
2476 list_slab_objects(s, page,
2477 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002478 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002479 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002480 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002481}
2482
2483/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002484 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002485 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002486static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002487{
2488 int node;
2489
2490 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002491 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002492 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002493 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002494 struct kmem_cache_node *n = get_node(s, node);
2495
Christoph Lameter599870b2008-04-23 12:36:52 -07002496 free_partial(s, n);
2497 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002498 return 1;
2499 }
2500 free_kmem_cache_nodes(s);
2501 return 0;
2502}
2503
2504/*
2505 * Close a cache and release the kmem_cache structure
2506 * (must be used for caches created using kmem_cache_create)
2507 */
2508void kmem_cache_destroy(struct kmem_cache *s)
2509{
2510 down_write(&slub_lock);
2511 s->refcount--;
2512 if (!s->refcount) {
2513 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002514 if (kmem_cache_close(s)) {
2515 printk(KERN_ERR "SLUB %s: %s called for cache that "
2516 "still has objects.\n", s->name, __func__);
2517 dump_stack();
2518 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002519 if (s->flags & SLAB_DESTROY_BY_RCU)
2520 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002521 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002522 }
2523 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002524}
2525EXPORT_SYMBOL(kmem_cache_destroy);
2526
2527/********************************************************************
2528 * Kmalloc subsystem
2529 *******************************************************************/
2530
Christoph Lameter51df1142010-08-20 12:37:15 -05002531struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -07002532EXPORT_SYMBOL(kmalloc_caches);
2533
Christoph Lameter51df1142010-08-20 12:37:15 -05002534static struct kmem_cache *kmem_cache;
2535
Christoph Lameter55136592010-08-20 12:37:13 -05002536#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05002537static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
Christoph Lameter55136592010-08-20 12:37:13 -05002538#endif
2539
Christoph Lameter81819f02007-05-06 14:49:36 -07002540static int __init setup_slub_min_order(char *str)
2541{
Pekka Enberg06428782008-01-07 23:20:27 -08002542 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002543
2544 return 1;
2545}
2546
2547__setup("slub_min_order=", setup_slub_min_order);
2548
2549static int __init setup_slub_max_order(char *str)
2550{
Pekka Enberg06428782008-01-07 23:20:27 -08002551 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002552 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002553
2554 return 1;
2555}
2556
2557__setup("slub_max_order=", setup_slub_max_order);
2558
2559static int __init setup_slub_min_objects(char *str)
2560{
Pekka Enberg06428782008-01-07 23:20:27 -08002561 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002562
2563 return 1;
2564}
2565
2566__setup("slub_min_objects=", setup_slub_min_objects);
2567
2568static int __init setup_slub_nomerge(char *str)
2569{
2570 slub_nomerge = 1;
2571 return 1;
2572}
2573
2574__setup("slub_nomerge", setup_slub_nomerge);
2575
Christoph Lameter51df1142010-08-20 12:37:15 -05002576static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2577 int size, unsigned int flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002578{
Christoph Lameter51df1142010-08-20 12:37:15 -05002579 struct kmem_cache *s;
2580
2581 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2582
Pekka Enberg83b519e2009-06-10 19:40:04 +03002583 /*
2584 * This function is called with IRQs disabled during early-boot on
2585 * single CPU so there's no need to take slub_lock here.
2586 */
Christoph Lameter55136592010-08-20 12:37:13 -05002587 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002588 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002589 goto panic;
2590
2591 list_add(&s->list, &slab_caches);
Christoph Lameter51df1142010-08-20 12:37:15 -05002592 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07002593
2594panic:
2595 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
Christoph Lameter51df1142010-08-20 12:37:15 -05002596 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002597}
2598
Christoph Lameterf1b26332007-07-17 04:03:26 -07002599/*
2600 * Conversion table for small slabs sizes / 8 to the index in the
2601 * kmalloc array. This is necessary for slabs < 192 since we have non power
2602 * of two cache sizes there. The size of larger slabs can be determined using
2603 * fls.
2604 */
2605static s8 size_index[24] = {
2606 3, /* 8 */
2607 4, /* 16 */
2608 5, /* 24 */
2609 5, /* 32 */
2610 6, /* 40 */
2611 6, /* 48 */
2612 6, /* 56 */
2613 6, /* 64 */
2614 1, /* 72 */
2615 1, /* 80 */
2616 1, /* 88 */
2617 1, /* 96 */
2618 7, /* 104 */
2619 7, /* 112 */
2620 7, /* 120 */
2621 7, /* 128 */
2622 2, /* 136 */
2623 2, /* 144 */
2624 2, /* 152 */
2625 2, /* 160 */
2626 2, /* 168 */
2627 2, /* 176 */
2628 2, /* 184 */
2629 2 /* 192 */
2630};
2631
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002632static inline int size_index_elem(size_t bytes)
2633{
2634 return (bytes - 1) / 8;
2635}
2636
Christoph Lameter81819f02007-05-06 14:49:36 -07002637static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2638{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002639 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002640
Christoph Lameterf1b26332007-07-17 04:03:26 -07002641 if (size <= 192) {
2642 if (!size)
2643 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002644
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002645 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002646 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002647 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002648
2649#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002650 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter51df1142010-08-20 12:37:15 -05002651 return kmalloc_dma_caches[index];
Christoph Lameterf1b26332007-07-17 04:03:26 -07002652
Christoph Lameter81819f02007-05-06 14:49:36 -07002653#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05002654 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -07002655}
2656
2657void *__kmalloc(size_t size, gfp_t flags)
2658{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002659 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002660 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002661
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002662 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002663 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002664
2665 s = get_slab(size, flags);
2666
2667 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002668 return s;
2669
Christoph Lameter2154a332010-07-09 14:07:10 -05002670 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002671
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002672 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002673
2674 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002675}
2676EXPORT_SYMBOL(__kmalloc);
2677
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002678static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2679{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002680 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002681 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002682
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002683 flags |= __GFP_COMP | __GFP_NOTRACK;
2684 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002685 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002686 ptr = page_address(page);
2687
2688 kmemleak_alloc(ptr, size, 1, flags);
2689 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002690}
2691
Christoph Lameter81819f02007-05-06 14:49:36 -07002692#ifdef CONFIG_NUMA
2693void *__kmalloc_node(size_t size, gfp_t flags, int node)
2694{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002695 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002696 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002697
Ingo Molnar057685c2009-02-20 12:15:30 +01002698 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002699 ret = kmalloc_large_node(size, flags, node);
2700
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002701 trace_kmalloc_node(_RET_IP_, ret,
2702 size, PAGE_SIZE << get_order(size),
2703 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002704
2705 return ret;
2706 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002707
2708 s = get_slab(size, flags);
2709
2710 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002711 return s;
2712
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002713 ret = slab_alloc(s, flags, node, _RET_IP_);
2714
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002715 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002716
2717 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002718}
2719EXPORT_SYMBOL(__kmalloc_node);
2720#endif
2721
2722size_t ksize(const void *object)
2723{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002724 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002725 struct kmem_cache *s;
2726
Christoph Lameteref8b4522007-10-16 01:24:46 -07002727 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002728 return 0;
2729
Vegard Nossum294a80a2007-12-04 23:45:30 -08002730 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002731
Pekka Enberg76994412008-05-22 19:22:25 +03002732 if (unlikely(!PageSlab(page))) {
2733 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002734 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002735 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002736 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002737
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002738#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002739 /*
2740 * Debugging requires use of the padding between object
2741 * and whatever may come after it.
2742 */
2743 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2744 return s->objsize;
2745
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002746#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002747 /*
2748 * If we have the need to store the freelist pointer
2749 * back there or track user information then we can
2750 * only use the space before that information.
2751 */
2752 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2753 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002754 /*
2755 * Else we can use all the padding etc for the allocation
2756 */
2757 return s->size;
2758}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002759EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002760
2761void kfree(const void *x)
2762{
Christoph Lameter81819f02007-05-06 14:49:36 -07002763 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002764 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002765
Pekka Enberg2121db72009-03-25 11:05:57 +02002766 trace_kfree(_RET_IP_, x);
2767
Satyam Sharma2408c552007-10-16 01:24:44 -07002768 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002769 return;
2770
Christoph Lameterb49af682007-05-06 14:49:41 -07002771 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002772 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002773 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002774 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002775 put_page(page);
2776 return;
2777 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002778 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002779}
2780EXPORT_SYMBOL(kfree);
2781
Christoph Lameter2086d262007-05-06 14:49:46 -07002782/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002783 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2784 * the remaining slabs by the number of items in use. The slabs with the
2785 * most items in use come first. New allocations will then fill those up
2786 * and thus they can be removed from the partial lists.
2787 *
2788 * The slabs with the least items are placed last. This results in them
2789 * being allocated from last increasing the chance that the last objects
2790 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002791 */
2792int kmem_cache_shrink(struct kmem_cache *s)
2793{
2794 int node;
2795 int i;
2796 struct kmem_cache_node *n;
2797 struct page *page;
2798 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002799 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002800 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002801 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002802 unsigned long flags;
2803
2804 if (!slabs_by_inuse)
2805 return -ENOMEM;
2806
2807 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002808 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002809 n = get_node(s, node);
2810
2811 if (!n->nr_partial)
2812 continue;
2813
Christoph Lameter834f3d12008-04-14 19:11:31 +03002814 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002815 INIT_LIST_HEAD(slabs_by_inuse + i);
2816
2817 spin_lock_irqsave(&n->list_lock, flags);
2818
2819 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002820 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002821 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002822 * Note that concurrent frees may occur while we hold the
2823 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002824 */
2825 list_for_each_entry_safe(page, t, &n->partial, lru) {
2826 if (!page->inuse && slab_trylock(page)) {
2827 /*
2828 * Must hold slab lock here because slab_free
2829 * may have freed the last object and be
2830 * waiting to release the slab.
2831 */
2832 list_del(&page->lru);
2833 n->nr_partial--;
2834 slab_unlock(page);
2835 discard_slab(s, page);
2836 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002837 list_move(&page->lru,
2838 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002839 }
2840 }
2841
Christoph Lameter2086d262007-05-06 14:49:46 -07002842 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002843 * Rebuild the partial list with the slabs filled up most
2844 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002845 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002846 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002847 list_splice(slabs_by_inuse + i, n->partial.prev);
2848
Christoph Lameter2086d262007-05-06 14:49:46 -07002849 spin_unlock_irqrestore(&n->list_lock, flags);
2850 }
2851
2852 kfree(slabs_by_inuse);
2853 return 0;
2854}
2855EXPORT_SYMBOL(kmem_cache_shrink);
2856
Yasunori Gotob9049e22007-10-21 16:41:37 -07002857#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2858static int slab_mem_going_offline_callback(void *arg)
2859{
2860 struct kmem_cache *s;
2861
2862 down_read(&slub_lock);
2863 list_for_each_entry(s, &slab_caches, list)
2864 kmem_cache_shrink(s);
2865 up_read(&slub_lock);
2866
2867 return 0;
2868}
2869
2870static void slab_mem_offline_callback(void *arg)
2871{
2872 struct kmem_cache_node *n;
2873 struct kmem_cache *s;
2874 struct memory_notify *marg = arg;
2875 int offline_node;
2876
2877 offline_node = marg->status_change_nid;
2878
2879 /*
2880 * If the node still has available memory. we need kmem_cache_node
2881 * for it yet.
2882 */
2883 if (offline_node < 0)
2884 return;
2885
2886 down_read(&slub_lock);
2887 list_for_each_entry(s, &slab_caches, list) {
2888 n = get_node(s, offline_node);
2889 if (n) {
2890 /*
2891 * if n->nr_slabs > 0, slabs still exist on the node
2892 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05002893 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07002894 * callback. So, we must fail.
2895 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002896 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002897
2898 s->node[offline_node] = NULL;
2899 kmem_cache_free(kmalloc_caches, n);
2900 }
2901 }
2902 up_read(&slub_lock);
2903}
2904
2905static int slab_mem_going_online_callback(void *arg)
2906{
2907 struct kmem_cache_node *n;
2908 struct kmem_cache *s;
2909 struct memory_notify *marg = arg;
2910 int nid = marg->status_change_nid;
2911 int ret = 0;
2912
2913 /*
2914 * If the node's memory is already available, then kmem_cache_node is
2915 * already created. Nothing to do.
2916 */
2917 if (nid < 0)
2918 return 0;
2919
2920 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002921 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002922 * allocate a kmem_cache_node structure in order to bring the node
2923 * online.
2924 */
2925 down_read(&slub_lock);
2926 list_for_each_entry(s, &slab_caches, list) {
2927 /*
2928 * XXX: kmem_cache_alloc_node will fallback to other nodes
2929 * since memory is not yet available from the node that
2930 * is brought up.
2931 */
2932 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
2933 if (!n) {
2934 ret = -ENOMEM;
2935 goto out;
2936 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03002937 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002938 s->node[nid] = n;
2939 }
2940out:
2941 up_read(&slub_lock);
2942 return ret;
2943}
2944
2945static int slab_memory_callback(struct notifier_block *self,
2946 unsigned long action, void *arg)
2947{
2948 int ret = 0;
2949
2950 switch (action) {
2951 case MEM_GOING_ONLINE:
2952 ret = slab_mem_going_online_callback(arg);
2953 break;
2954 case MEM_GOING_OFFLINE:
2955 ret = slab_mem_going_offline_callback(arg);
2956 break;
2957 case MEM_OFFLINE:
2958 case MEM_CANCEL_ONLINE:
2959 slab_mem_offline_callback(arg);
2960 break;
2961 case MEM_ONLINE:
2962 case MEM_CANCEL_OFFLINE:
2963 break;
2964 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08002965 if (ret)
2966 ret = notifier_from_errno(ret);
2967 else
2968 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07002969 return ret;
2970}
2971
2972#endif /* CONFIG_MEMORY_HOTPLUG */
2973
Christoph Lameter81819f02007-05-06 14:49:36 -07002974/********************************************************************
2975 * Basic setup of slabs
2976 *******************************************************************/
2977
Christoph Lameter51df1142010-08-20 12:37:15 -05002978/*
2979 * Used for early kmem_cache structures that were allocated using
2980 * the page allocator
2981 */
2982
2983static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
2984{
2985 int node;
2986
2987 list_add(&s->list, &slab_caches);
2988 s->refcount = -1;
2989
2990 for_each_node_state(node, N_NORMAL_MEMORY) {
2991 struct kmem_cache_node *n = get_node(s, node);
2992 struct page *p;
2993
2994 if (n) {
2995 list_for_each_entry(p, &n->partial, lru)
2996 p->slab = s;
2997
2998#ifdef CONFIG_SLAB_DEBUG
2999 list_for_each_entry(p, &n->full, lru)
3000 p->slab = s;
3001#endif
3002 }
3003 }
3004}
3005
Christoph Lameter81819f02007-05-06 14:49:36 -07003006void __init kmem_cache_init(void)
3007{
3008 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003009 int caches = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003010 struct kmem_cache *temp_kmem_cache;
3011 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003012
3013#ifdef CONFIG_NUMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003014 struct kmem_cache *temp_kmem_cache_node;
3015 unsigned long kmalloc_size;
3016
3017 kmem_size = offsetof(struct kmem_cache, node) +
3018 nr_node_ids * sizeof(struct kmem_cache_node *);
3019
3020 /* Allocate two kmem_caches from the page allocator */
3021 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3022 order = get_order(2 * kmalloc_size);
3023 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3024
Christoph Lameter81819f02007-05-06 14:49:36 -07003025 /*
3026 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003027 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003028 * kmem_cache_open for slab_state == DOWN.
3029 */
Christoph Lameter51df1142010-08-20 12:37:15 -05003030 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3031
3032 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3033 sizeof(struct kmem_cache_node),
3034 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003035
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003036 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter51df1142010-08-20 12:37:15 -05003037#else
3038 /* Allocate a single kmem_cache from the page allocator */
3039 kmem_size = sizeof(struct kmem_cache);
3040 order = get_order(kmem_size);
3041 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003042#endif
3043
3044 /* Able to allocate the per node structures */
3045 slab_state = PARTIAL;
3046
Christoph Lameter51df1142010-08-20 12:37:15 -05003047 temp_kmem_cache = kmem_cache;
3048 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3049 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3050 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3051 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003052
Christoph Lameter51df1142010-08-20 12:37:15 -05003053#ifdef CONFIG_NUMA
3054 /*
3055 * Allocate kmem_cache_node properly from the kmem_cache slab.
3056 * kmem_cache_node is separately allocated so no need to
3057 * update any list pointers.
3058 */
3059 temp_kmem_cache_node = kmem_cache_node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003060
Christoph Lameter51df1142010-08-20 12:37:15 -05003061 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3062 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3063
3064 kmem_cache_bootstrap_fixup(kmem_cache_node);
3065
3066 caches++;
3067#else
3068 /*
3069 * kmem_cache has kmem_cache_node embedded and we moved it!
3070 * Update the list heads
3071 */
3072 INIT_LIST_HEAD(&kmem_cache->local_node.partial);
3073 list_splice(&temp_kmem_cache->local_node.partial, &kmem_cache->local_node.partial);
3074#ifdef CONFIG_SLUB_DEBUG
3075 INIT_LIST_HEAD(&kmem_cache->local_node.full);
3076 list_splice(&temp_kmem_cache->local_node.full, &kmem_cache->local_node.full);
3077#endif
3078#endif
3079 kmem_cache_bootstrap_fixup(kmem_cache);
3080 caches++;
3081 /* Free temporary boot structure */
3082 free_pages((unsigned long)temp_kmem_cache, order);
3083
3084 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf1b26332007-07-17 04:03:26 -07003085
3086 /*
3087 * Patch up the size_index table if we have strange large alignment
3088 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003089 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003090 *
3091 * Largest permitted alignment is 256 bytes due to the way we
3092 * handle the index determination for the smaller caches.
3093 *
3094 * Make sure that nothing crazy happens if someone starts tinkering
3095 * around with ARCH_KMALLOC_MINALIGN
3096 */
3097 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3098 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3099
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003100 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3101 int elem = size_index_elem(i);
3102 if (elem >= ARRAY_SIZE(size_index))
3103 break;
3104 size_index[elem] = KMALLOC_SHIFT_LOW;
3105 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003106
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003107 if (KMALLOC_MIN_SIZE == 64) {
3108 /*
3109 * The 96 byte size cache is not used if the alignment
3110 * is 64 byte.
3111 */
3112 for (i = 64 + 8; i <= 96; i += 8)
3113 size_index[size_index_elem(i)] = 7;
3114 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003115 /*
3116 * The 192 byte sized cache is not used if the alignment
3117 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3118 * instead.
3119 */
3120 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003121 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003122 }
3123
Christoph Lameter51df1142010-08-20 12:37:15 -05003124 /* Caches that are not of the two-to-the-power-of size */
3125 if (KMALLOC_MIN_SIZE <= 32) {
3126 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3127 caches++;
3128 }
3129
3130 if (KMALLOC_MIN_SIZE <= 64) {
3131 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3132 caches++;
3133 }
3134
3135 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3136 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3137 caches++;
3138 }
3139
Christoph Lameter81819f02007-05-06 14:49:36 -07003140 slab_state = UP;
3141
3142 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003143 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3144 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3145
3146 BUG_ON(!s);
Christoph Lameter51df1142010-08-20 12:37:15 -05003147 kmalloc_caches[i]->name = s;
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003148 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003149
3150#ifdef CONFIG_SMP
3151 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003152#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003153
Christoph Lameter55136592010-08-20 12:37:13 -05003154#ifdef CONFIG_ZONE_DMA
Christoph Lameter51df1142010-08-20 12:37:15 -05003155 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3156 struct kmem_cache *s = kmalloc_caches[i];
Christoph Lameter55136592010-08-20 12:37:13 -05003157
Christoph Lameter51df1142010-08-20 12:37:15 -05003158 if (s && s->size) {
Christoph Lameter55136592010-08-20 12:37:13 -05003159 char *name = kasprintf(GFP_NOWAIT,
3160 "dma-kmalloc-%d", s->objsize);
3161
3162 BUG_ON(!name);
Christoph Lameter51df1142010-08-20 12:37:15 -05003163 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3164 s->objsize, SLAB_CACHE_DMA);
Christoph Lameter55136592010-08-20 12:37:13 -05003165 }
3166 }
3167#endif
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003168 printk(KERN_INFO
3169 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003170 " CPUs=%d, Nodes=%d\n",
3171 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003172 slub_min_order, slub_max_order, slub_min_objects,
3173 nr_cpu_ids, nr_node_ids);
3174}
3175
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003176void __init kmem_cache_init_late(void)
3177{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003178}
3179
Christoph Lameter81819f02007-05-06 14:49:36 -07003180/*
3181 * Find a mergeable slab cache
3182 */
3183static int slab_unmergeable(struct kmem_cache *s)
3184{
3185 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3186 return 1;
3187
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003188 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003189 return 1;
3190
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003191 /*
3192 * We may have set a slab to be unmergeable during bootstrap.
3193 */
3194 if (s->refcount < 0)
3195 return 1;
3196
Christoph Lameter81819f02007-05-06 14:49:36 -07003197 return 0;
3198}
3199
3200static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003201 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003202 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003203{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003204 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003205
3206 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3207 return NULL;
3208
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003209 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003210 return NULL;
3211
3212 size = ALIGN(size, sizeof(void *));
3213 align = calculate_alignment(flags, align, size);
3214 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003215 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003216
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003217 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003218 if (slab_unmergeable(s))
3219 continue;
3220
3221 if (size > s->size)
3222 continue;
3223
Christoph Lameterba0268a2007-09-11 15:24:11 -07003224 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003225 continue;
3226 /*
3227 * Check if alignment is compatible.
3228 * Courtesy of Adrian Drzewiecki
3229 */
Pekka Enberg06428782008-01-07 23:20:27 -08003230 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003231 continue;
3232
3233 if (s->size - size >= sizeof(void *))
3234 continue;
3235
3236 return s;
3237 }
3238 return NULL;
3239}
3240
3241struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003242 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003243{
3244 struct kmem_cache *s;
3245
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003246 if (WARN_ON(!name))
3247 return NULL;
3248
Christoph Lameter81819f02007-05-06 14:49:36 -07003249 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003250 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003251 if (s) {
3252 s->refcount++;
3253 /*
3254 * Adjust the object sizes so that we clear
3255 * the complete object on kzalloc.
3256 */
3257 s->objsize = max(s->objsize, (int)size);
3258 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003259
David Rientjes7b8f3b62008-12-17 22:09:46 -08003260 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003261 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003262 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003263 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003264 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003265 return s;
3266 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003267
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003268 s = kmalloc(kmem_size, GFP_KERNEL);
3269 if (s) {
Christoph Lameter55136592010-08-20 12:37:13 -05003270 if (kmem_cache_open(s, name,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003271 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003272 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003273 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003274 list_del(&s->list);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003275 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003276 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003277 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003278 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003279 return s;
3280 }
3281 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003282 }
3283 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003284
3285err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003286 if (flags & SLAB_PANIC)
3287 panic("Cannot create slabcache %s\n", name);
3288 else
3289 s = NULL;
3290 return s;
3291}
3292EXPORT_SYMBOL(kmem_cache_create);
3293
Christoph Lameter81819f02007-05-06 14:49:36 -07003294#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003295/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003296 * Use the cpu notifier to insure that the cpu slabs are flushed when
3297 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003298 */
3299static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3300 unsigned long action, void *hcpu)
3301{
3302 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003303 struct kmem_cache *s;
3304 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003305
3306 switch (action) {
3307 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003308 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003309 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003310 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003311 down_read(&slub_lock);
3312 list_for_each_entry(s, &slab_caches, list) {
3313 local_irq_save(flags);
3314 __flush_cpu_slab(s, cpu);
3315 local_irq_restore(flags);
3316 }
3317 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003318 break;
3319 default:
3320 break;
3321 }
3322 return NOTIFY_OK;
3323}
3324
Pekka Enberg06428782008-01-07 23:20:27 -08003325static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003326 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003327};
Christoph Lameter81819f02007-05-06 14:49:36 -07003328
3329#endif
3330
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003331void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003332{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003333 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003334 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003335
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003336 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003337 return kmalloc_large(size, gfpflags);
3338
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003339 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003340
Satyam Sharma2408c552007-10-16 01:24:44 -07003341 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003342 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003343
Christoph Lameter2154a332010-07-09 14:07:10 -05003344 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003345
3346 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003347 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003348
3349 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003350}
3351
3352void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003353 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003354{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003355 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003356 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003357
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003358 if (unlikely(size > SLUB_MAX_SIZE)) {
3359 ret = kmalloc_large_node(size, gfpflags, node);
3360
3361 trace_kmalloc_node(caller, ret,
3362 size, PAGE_SIZE << get_order(size),
3363 gfpflags, node);
3364
3365 return ret;
3366 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003367
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003368 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003369
Satyam Sharma2408c552007-10-16 01:24:44 -07003370 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003371 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003372
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003373 ret = slab_alloc(s, gfpflags, node, caller);
3374
3375 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003376 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003377
3378 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003379}
3380
Christoph Lameterf6acb632008-04-29 16:16:06 -07003381#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003382static int count_inuse(struct page *page)
3383{
3384 return page->inuse;
3385}
3386
3387static int count_total(struct page *page)
3388{
3389 return page->objects;
3390}
3391
Christoph Lameter434e2452007-07-17 04:03:30 -07003392static int validate_slab(struct kmem_cache *s, struct page *page,
3393 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003394{
3395 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003396 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003397
3398 if (!check_slab(s, page) ||
3399 !on_freelist(s, page, NULL))
3400 return 0;
3401
3402 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003403 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003404
Christoph Lameter7656c722007-05-09 02:32:40 -07003405 for_each_free_object(p, s, page->freelist) {
3406 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003407 if (!check_object(s, page, p, 0))
3408 return 0;
3409 }
3410
Christoph Lameter224a88b2008-04-14 19:11:31 +03003411 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003412 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003413 if (!check_object(s, page, p, 1))
3414 return 0;
3415 return 1;
3416}
3417
Christoph Lameter434e2452007-07-17 04:03:30 -07003418static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3419 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003420{
3421 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003422 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003423 slab_unlock(page);
3424 } else
3425 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3426 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003427}
3428
Christoph Lameter434e2452007-07-17 04:03:30 -07003429static int validate_slab_node(struct kmem_cache *s,
3430 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003431{
3432 unsigned long count = 0;
3433 struct page *page;
3434 unsigned long flags;
3435
3436 spin_lock_irqsave(&n->list_lock, flags);
3437
3438 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003439 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003440 count++;
3441 }
3442 if (count != n->nr_partial)
3443 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3444 "counter=%ld\n", s->name, count, n->nr_partial);
3445
3446 if (!(s->flags & SLAB_STORE_USER))
3447 goto out;
3448
3449 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003450 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003451 count++;
3452 }
3453 if (count != atomic_long_read(&n->nr_slabs))
3454 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3455 "counter=%ld\n", s->name, count,
3456 atomic_long_read(&n->nr_slabs));
3457
3458out:
3459 spin_unlock_irqrestore(&n->list_lock, flags);
3460 return count;
3461}
3462
Christoph Lameter434e2452007-07-17 04:03:30 -07003463static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003464{
3465 int node;
3466 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003467 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003468 sizeof(unsigned long), GFP_KERNEL);
3469
3470 if (!map)
3471 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003472
3473 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003474 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003475 struct kmem_cache_node *n = get_node(s, node);
3476
Christoph Lameter434e2452007-07-17 04:03:30 -07003477 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003478 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003479 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003480 return count;
3481}
3482
Christoph Lameterb3459702007-05-09 02:32:41 -07003483#ifdef SLUB_RESILIENCY_TEST
3484static void resiliency_test(void)
3485{
3486 u8 *p;
3487
3488 printk(KERN_ERR "SLUB resiliency testing\n");
3489 printk(KERN_ERR "-----------------------\n");
3490 printk(KERN_ERR "A. Corruption after allocation\n");
3491
3492 p = kzalloc(16, GFP_KERNEL);
3493 p[16] = 0x12;
3494 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3495 " 0x12->0x%p\n\n", p + 16);
3496
3497 validate_slab_cache(kmalloc_caches + 4);
3498
3499 /* Hmmm... The next two are dangerous */
3500 p = kzalloc(32, GFP_KERNEL);
3501 p[32 + sizeof(void *)] = 0x34;
3502 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003503 " 0x34 -> -0x%p\n", p);
3504 printk(KERN_ERR
3505 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003506
3507 validate_slab_cache(kmalloc_caches + 5);
3508 p = kzalloc(64, GFP_KERNEL);
3509 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3510 *p = 0x56;
3511 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3512 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003513 printk(KERN_ERR
3514 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003515 validate_slab_cache(kmalloc_caches + 6);
3516
3517 printk(KERN_ERR "\nB. Corruption after free\n");
3518 p = kzalloc(128, GFP_KERNEL);
3519 kfree(p);
3520 *p = 0x78;
3521 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3522 validate_slab_cache(kmalloc_caches + 7);
3523
3524 p = kzalloc(256, GFP_KERNEL);
3525 kfree(p);
3526 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003527 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3528 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003529 validate_slab_cache(kmalloc_caches + 8);
3530
3531 p = kzalloc(512, GFP_KERNEL);
3532 kfree(p);
3533 p[512] = 0xab;
3534 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3535 validate_slab_cache(kmalloc_caches + 9);
3536}
3537#else
3538static void resiliency_test(void) {};
3539#endif
3540
Christoph Lameter88a420e2007-05-06 14:49:45 -07003541/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003542 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003543 * and freed.
3544 */
3545
3546struct location {
3547 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003548 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003549 long long sum_time;
3550 long min_time;
3551 long max_time;
3552 long min_pid;
3553 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303554 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003555 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003556};
3557
3558struct loc_track {
3559 unsigned long max;
3560 unsigned long count;
3561 struct location *loc;
3562};
3563
3564static void free_loc_track(struct loc_track *t)
3565{
3566 if (t->max)
3567 free_pages((unsigned long)t->loc,
3568 get_order(sizeof(struct location) * t->max));
3569}
3570
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003571static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003572{
3573 struct location *l;
3574 int order;
3575
Christoph Lameter88a420e2007-05-06 14:49:45 -07003576 order = get_order(sizeof(struct location) * max);
3577
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003578 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003579 if (!l)
3580 return 0;
3581
3582 if (t->count) {
3583 memcpy(l, t->loc, sizeof(struct location) * t->count);
3584 free_loc_track(t);
3585 }
3586 t->max = max;
3587 t->loc = l;
3588 return 1;
3589}
3590
3591static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003592 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003593{
3594 long start, end, pos;
3595 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003596 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003597 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003598
3599 start = -1;
3600 end = t->count;
3601
3602 for ( ; ; ) {
3603 pos = start + (end - start + 1) / 2;
3604
3605 /*
3606 * There is nothing at "end". If we end up there
3607 * we need to add something to before end.
3608 */
3609 if (pos == end)
3610 break;
3611
3612 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003613 if (track->addr == caddr) {
3614
3615 l = &t->loc[pos];
3616 l->count++;
3617 if (track->when) {
3618 l->sum_time += age;
3619 if (age < l->min_time)
3620 l->min_time = age;
3621 if (age > l->max_time)
3622 l->max_time = age;
3623
3624 if (track->pid < l->min_pid)
3625 l->min_pid = track->pid;
3626 if (track->pid > l->max_pid)
3627 l->max_pid = track->pid;
3628
Rusty Russell174596a2009-01-01 10:12:29 +10303629 cpumask_set_cpu(track->cpu,
3630 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003631 }
3632 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003633 return 1;
3634 }
3635
Christoph Lameter45edfa52007-05-09 02:32:45 -07003636 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003637 end = pos;
3638 else
3639 start = pos;
3640 }
3641
3642 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003643 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003644 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003645 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003646 return 0;
3647
3648 l = t->loc + pos;
3649 if (pos < t->count)
3650 memmove(l + 1, l,
3651 (t->count - pos) * sizeof(struct location));
3652 t->count++;
3653 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003654 l->addr = track->addr;
3655 l->sum_time = age;
3656 l->min_time = age;
3657 l->max_time = age;
3658 l->min_pid = track->pid;
3659 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303660 cpumask_clear(to_cpumask(l->cpus));
3661 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003662 nodes_clear(l->nodes);
3663 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003664 return 1;
3665}
3666
3667static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003668 struct page *page, enum track_item alloc,
3669 long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003670{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003671 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003672 void *p;
3673
Christoph Lameter39b26462008-04-14 19:11:30 +03003674 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003675 for_each_free_object(p, s, page->freelist)
3676 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003677
Christoph Lameter224a88b2008-04-14 19:11:31 +03003678 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003679 if (!test_bit(slab_index(p, s, addr), map))
3680 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003681}
3682
3683static int list_locations(struct kmem_cache *s, char *buf,
3684 enum track_item alloc)
3685{
Harvey Harrisone374d482008-01-31 15:20:50 -08003686 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003687 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003688 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003689 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003690 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3691 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003692
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003693 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3694 GFP_TEMPORARY)) {
3695 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003696 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003697 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003698 /* Push back cpu slabs */
3699 flush_all(s);
3700
Christoph Lameterf64dc582007-10-16 01:25:33 -07003701 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003702 struct kmem_cache_node *n = get_node(s, node);
3703 unsigned long flags;
3704 struct page *page;
3705
Christoph Lameter9e869432007-08-22 14:01:56 -07003706 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003707 continue;
3708
3709 spin_lock_irqsave(&n->list_lock, flags);
3710 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003711 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003712 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003713 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003714 spin_unlock_irqrestore(&n->list_lock, flags);
3715 }
3716
3717 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003718 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003719
Hugh Dickins9c246242008-12-09 13:14:27 -08003720 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003721 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003722 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003723
3724 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003725 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003726 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003727 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003728
3729 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003730 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003731 l->min_time,
3732 (long)div_u64(l->sum_time, l->count),
3733 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003734 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003735 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003736 l->min_time);
3737
3738 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003739 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003740 l->min_pid, l->max_pid);
3741 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003742 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003743 l->min_pid);
3744
Rusty Russell174596a2009-01-01 10:12:29 +10303745 if (num_online_cpus() > 1 &&
3746 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003747 len < PAGE_SIZE - 60) {
3748 len += sprintf(buf + len, " cpus=");
3749 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303750 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003751 }
3752
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003753 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003754 len < PAGE_SIZE - 60) {
3755 len += sprintf(buf + len, " nodes=");
3756 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003757 l->nodes);
3758 }
3759
Harvey Harrisone374d482008-01-31 15:20:50 -08003760 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003761 }
3762
3763 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003764 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003765 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003766 len += sprintf(buf, "No data\n");
3767 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003768}
3769
Christoph Lameter81819f02007-05-06 14:49:36 -07003770enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003771 SL_ALL, /* All slabs */
3772 SL_PARTIAL, /* Only partially allocated slabs */
3773 SL_CPU, /* Only slabs used for cpu caches */
3774 SL_OBJECTS, /* Determine allocated objects not slabs */
3775 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003776};
3777
Christoph Lameter205ab992008-04-14 19:11:40 +03003778#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003779#define SO_PARTIAL (1 << SL_PARTIAL)
3780#define SO_CPU (1 << SL_CPU)
3781#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003782#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003783
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003784static ssize_t show_slab_objects(struct kmem_cache *s,
3785 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003786{
3787 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003788 int node;
3789 int x;
3790 unsigned long *nodes;
3791 unsigned long *per_cpu;
3792
3793 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003794 if (!nodes)
3795 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003796 per_cpu = nodes + nr_node_ids;
3797
Christoph Lameter205ab992008-04-14 19:11:40 +03003798 if (flags & SO_CPU) {
3799 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003800
Christoph Lameter205ab992008-04-14 19:11:40 +03003801 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003802 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003803
Christoph Lameter205ab992008-04-14 19:11:40 +03003804 if (!c || c->node < 0)
3805 continue;
3806
3807 if (c->page) {
3808 if (flags & SO_TOTAL)
3809 x = c->page->objects;
3810 else if (flags & SO_OBJECTS)
3811 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003812 else
3813 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003814
Christoph Lameter81819f02007-05-06 14:49:36 -07003815 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003816 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003817 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003818 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003819 }
3820 }
3821
Christoph Lameter205ab992008-04-14 19:11:40 +03003822 if (flags & SO_ALL) {
3823 for_each_node_state(node, N_NORMAL_MEMORY) {
3824 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003825
Christoph Lameter205ab992008-04-14 19:11:40 +03003826 if (flags & SO_TOTAL)
3827 x = atomic_long_read(&n->total_objects);
3828 else if (flags & SO_OBJECTS)
3829 x = atomic_long_read(&n->total_objects) -
3830 count_partial(n, count_free);
3831
3832 else
3833 x = atomic_long_read(&n->nr_slabs);
3834 total += x;
3835 nodes[node] += x;
3836 }
3837
3838 } else if (flags & SO_PARTIAL) {
3839 for_each_node_state(node, N_NORMAL_MEMORY) {
3840 struct kmem_cache_node *n = get_node(s, node);
3841
3842 if (flags & SO_TOTAL)
3843 x = count_partial(n, count_total);
3844 else if (flags & SO_OBJECTS)
3845 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003846 else
3847 x = n->nr_partial;
3848 total += x;
3849 nodes[node] += x;
3850 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003851 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003852 x = sprintf(buf, "%lu", total);
3853#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003854 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003855 if (nodes[node])
3856 x += sprintf(buf + x, " N%d=%lu",
3857 node, nodes[node]);
3858#endif
3859 kfree(nodes);
3860 return x + sprintf(buf + x, "\n");
3861}
3862
3863static int any_slab_objects(struct kmem_cache *s)
3864{
3865 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003866
3867 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003868 struct kmem_cache_node *n = get_node(s, node);
3869
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003870 if (!n)
3871 continue;
3872
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003873 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003874 return 1;
3875 }
3876 return 0;
3877}
3878
3879#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3880#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3881
3882struct slab_attribute {
3883 struct attribute attr;
3884 ssize_t (*show)(struct kmem_cache *s, char *buf);
3885 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3886};
3887
3888#define SLAB_ATTR_RO(_name) \
3889 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3890
3891#define SLAB_ATTR(_name) \
3892 static struct slab_attribute _name##_attr = \
3893 __ATTR(_name, 0644, _name##_show, _name##_store)
3894
Christoph Lameter81819f02007-05-06 14:49:36 -07003895static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3896{
3897 return sprintf(buf, "%d\n", s->size);
3898}
3899SLAB_ATTR_RO(slab_size);
3900
3901static ssize_t align_show(struct kmem_cache *s, char *buf)
3902{
3903 return sprintf(buf, "%d\n", s->align);
3904}
3905SLAB_ATTR_RO(align);
3906
3907static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3908{
3909 return sprintf(buf, "%d\n", s->objsize);
3910}
3911SLAB_ATTR_RO(object_size);
3912
3913static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3914{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003915 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003916}
3917SLAB_ATTR_RO(objs_per_slab);
3918
Christoph Lameter06b285d2008-04-14 19:11:41 +03003919static ssize_t order_store(struct kmem_cache *s,
3920 const char *buf, size_t length)
3921{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003922 unsigned long order;
3923 int err;
3924
3925 err = strict_strtoul(buf, 10, &order);
3926 if (err)
3927 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003928
3929 if (order > slub_max_order || order < slub_min_order)
3930 return -EINVAL;
3931
3932 calculate_sizes(s, order);
3933 return length;
3934}
3935
Christoph Lameter81819f02007-05-06 14:49:36 -07003936static ssize_t order_show(struct kmem_cache *s, char *buf)
3937{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003938 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003939}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003940SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003941
David Rientjes73d342b2009-02-22 17:40:09 -08003942static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3943{
3944 return sprintf(buf, "%lu\n", s->min_partial);
3945}
3946
3947static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3948 size_t length)
3949{
3950 unsigned long min;
3951 int err;
3952
3953 err = strict_strtoul(buf, 10, &min);
3954 if (err)
3955 return err;
3956
David Rientjesc0bdb232009-02-25 09:16:35 +02003957 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003958 return length;
3959}
3960SLAB_ATTR(min_partial);
3961
Christoph Lameter81819f02007-05-06 14:49:36 -07003962static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3963{
3964 if (s->ctor) {
3965 int n = sprint_symbol(buf, (unsigned long)s->ctor);
3966
3967 return n + sprintf(buf + n, "\n");
3968 }
3969 return 0;
3970}
3971SLAB_ATTR_RO(ctor);
3972
Christoph Lameter81819f02007-05-06 14:49:36 -07003973static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3974{
3975 return sprintf(buf, "%d\n", s->refcount - 1);
3976}
3977SLAB_ATTR_RO(aliases);
3978
3979static ssize_t slabs_show(struct kmem_cache *s, char *buf)
3980{
Christoph Lameter205ab992008-04-14 19:11:40 +03003981 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003982}
3983SLAB_ATTR_RO(slabs);
3984
3985static ssize_t partial_show(struct kmem_cache *s, char *buf)
3986{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003987 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003988}
3989SLAB_ATTR_RO(partial);
3990
3991static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3992{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003993 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003994}
3995SLAB_ATTR_RO(cpu_slabs);
3996
3997static ssize_t objects_show(struct kmem_cache *s, char *buf)
3998{
Christoph Lameter205ab992008-04-14 19:11:40 +03003999 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004000}
4001SLAB_ATTR_RO(objects);
4002
Christoph Lameter205ab992008-04-14 19:11:40 +03004003static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4004{
4005 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4006}
4007SLAB_ATTR_RO(objects_partial);
4008
4009static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4010{
4011 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4012}
4013SLAB_ATTR_RO(total_objects);
4014
Christoph Lameter81819f02007-05-06 14:49:36 -07004015static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4016{
4017 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4018}
4019
4020static ssize_t sanity_checks_store(struct kmem_cache *s,
4021 const char *buf, size_t length)
4022{
4023 s->flags &= ~SLAB_DEBUG_FREE;
4024 if (buf[0] == '1')
4025 s->flags |= SLAB_DEBUG_FREE;
4026 return length;
4027}
4028SLAB_ATTR(sanity_checks);
4029
4030static ssize_t trace_show(struct kmem_cache *s, char *buf)
4031{
4032 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4033}
4034
4035static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4036 size_t length)
4037{
4038 s->flags &= ~SLAB_TRACE;
4039 if (buf[0] == '1')
4040 s->flags |= SLAB_TRACE;
4041 return length;
4042}
4043SLAB_ATTR(trace);
4044
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004045#ifdef CONFIG_FAILSLAB
4046static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4047{
4048 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4049}
4050
4051static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4052 size_t length)
4053{
4054 s->flags &= ~SLAB_FAILSLAB;
4055 if (buf[0] == '1')
4056 s->flags |= SLAB_FAILSLAB;
4057 return length;
4058}
4059SLAB_ATTR(failslab);
4060#endif
4061
Christoph Lameter81819f02007-05-06 14:49:36 -07004062static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4063{
4064 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4065}
4066
4067static ssize_t reclaim_account_store(struct kmem_cache *s,
4068 const char *buf, size_t length)
4069{
4070 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4071 if (buf[0] == '1')
4072 s->flags |= SLAB_RECLAIM_ACCOUNT;
4073 return length;
4074}
4075SLAB_ATTR(reclaim_account);
4076
4077static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4078{
Christoph Lameter5af60832007-05-06 14:49:56 -07004079 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004080}
4081SLAB_ATTR_RO(hwcache_align);
4082
4083#ifdef CONFIG_ZONE_DMA
4084static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4085{
4086 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4087}
4088SLAB_ATTR_RO(cache_dma);
4089#endif
4090
4091static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4092{
4093 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4094}
4095SLAB_ATTR_RO(destroy_by_rcu);
4096
4097static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4098{
4099 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4100}
4101
4102static ssize_t red_zone_store(struct kmem_cache *s,
4103 const char *buf, size_t length)
4104{
4105 if (any_slab_objects(s))
4106 return -EBUSY;
4107
4108 s->flags &= ~SLAB_RED_ZONE;
4109 if (buf[0] == '1')
4110 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004111 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004112 return length;
4113}
4114SLAB_ATTR(red_zone);
4115
4116static ssize_t poison_show(struct kmem_cache *s, char *buf)
4117{
4118 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4119}
4120
4121static ssize_t poison_store(struct kmem_cache *s,
4122 const char *buf, size_t length)
4123{
4124 if (any_slab_objects(s))
4125 return -EBUSY;
4126
4127 s->flags &= ~SLAB_POISON;
4128 if (buf[0] == '1')
4129 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004130 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004131 return length;
4132}
4133SLAB_ATTR(poison);
4134
4135static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4136{
4137 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4138}
4139
4140static ssize_t store_user_store(struct kmem_cache *s,
4141 const char *buf, size_t length)
4142{
4143 if (any_slab_objects(s))
4144 return -EBUSY;
4145
4146 s->flags &= ~SLAB_STORE_USER;
4147 if (buf[0] == '1')
4148 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004149 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004150 return length;
4151}
4152SLAB_ATTR(store_user);
4153
Christoph Lameter53e15af2007-05-06 14:49:43 -07004154static ssize_t validate_show(struct kmem_cache *s, char *buf)
4155{
4156 return 0;
4157}
4158
4159static ssize_t validate_store(struct kmem_cache *s,
4160 const char *buf, size_t length)
4161{
Christoph Lameter434e2452007-07-17 04:03:30 -07004162 int ret = -EINVAL;
4163
4164 if (buf[0] == '1') {
4165 ret = validate_slab_cache(s);
4166 if (ret >= 0)
4167 ret = length;
4168 }
4169 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004170}
4171SLAB_ATTR(validate);
4172
Christoph Lameter2086d262007-05-06 14:49:46 -07004173static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4174{
4175 return 0;
4176}
4177
4178static ssize_t shrink_store(struct kmem_cache *s,
4179 const char *buf, size_t length)
4180{
4181 if (buf[0] == '1') {
4182 int rc = kmem_cache_shrink(s);
4183
4184 if (rc)
4185 return rc;
4186 } else
4187 return -EINVAL;
4188 return length;
4189}
4190SLAB_ATTR(shrink);
4191
Christoph Lameter88a420e2007-05-06 14:49:45 -07004192static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4193{
4194 if (!(s->flags & SLAB_STORE_USER))
4195 return -ENOSYS;
4196 return list_locations(s, buf, TRACK_ALLOC);
4197}
4198SLAB_ATTR_RO(alloc_calls);
4199
4200static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4201{
4202 if (!(s->flags & SLAB_STORE_USER))
4203 return -ENOSYS;
4204 return list_locations(s, buf, TRACK_FREE);
4205}
4206SLAB_ATTR_RO(free_calls);
4207
Christoph Lameter81819f02007-05-06 14:49:36 -07004208#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004209static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004210{
Christoph Lameter98246012008-01-07 23:20:26 -08004211 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004212}
4213
Christoph Lameter98246012008-01-07 23:20:26 -08004214static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004215 const char *buf, size_t length)
4216{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004217 unsigned long ratio;
4218 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004219
Christoph Lameter0121c6192008-04-29 16:11:12 -07004220 err = strict_strtoul(buf, 10, &ratio);
4221 if (err)
4222 return err;
4223
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004224 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004225 s->remote_node_defrag_ratio = ratio * 10;
4226
Christoph Lameter81819f02007-05-06 14:49:36 -07004227 return length;
4228}
Christoph Lameter98246012008-01-07 23:20:26 -08004229SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004230#endif
4231
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004232#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004233static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4234{
4235 unsigned long sum = 0;
4236 int cpu;
4237 int len;
4238 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4239
4240 if (!data)
4241 return -ENOMEM;
4242
4243 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004244 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004245
4246 data[cpu] = x;
4247 sum += x;
4248 }
4249
4250 len = sprintf(buf, "%lu", sum);
4251
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004252#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004253 for_each_online_cpu(cpu) {
4254 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004255 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004256 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004257#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004258 kfree(data);
4259 return len + sprintf(buf + len, "\n");
4260}
4261
David Rientjes78eb00c2009-10-15 02:20:22 -07004262static void clear_stat(struct kmem_cache *s, enum stat_item si)
4263{
4264 int cpu;
4265
4266 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004267 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004268}
4269
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004270#define STAT_ATTR(si, text) \
4271static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4272{ \
4273 return show_stat(s, buf, si); \
4274} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004275static ssize_t text##_store(struct kmem_cache *s, \
4276 const char *buf, size_t length) \
4277{ \
4278 if (buf[0] != '0') \
4279 return -EINVAL; \
4280 clear_stat(s, si); \
4281 return length; \
4282} \
4283SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004284
4285STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4286STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4287STAT_ATTR(FREE_FASTPATH, free_fastpath);
4288STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4289STAT_ATTR(FREE_FROZEN, free_frozen);
4290STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4291STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4292STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4293STAT_ATTR(ALLOC_SLAB, alloc_slab);
4294STAT_ATTR(ALLOC_REFILL, alloc_refill);
4295STAT_ATTR(FREE_SLAB, free_slab);
4296STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4297STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4298STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4299STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4300STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4301STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004302STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004303#endif
4304
Pekka Enberg06428782008-01-07 23:20:27 -08004305static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004306 &slab_size_attr.attr,
4307 &object_size_attr.attr,
4308 &objs_per_slab_attr.attr,
4309 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004310 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004311 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004312 &objects_partial_attr.attr,
4313 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004314 &slabs_attr.attr,
4315 &partial_attr.attr,
4316 &cpu_slabs_attr.attr,
4317 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004318 &aliases_attr.attr,
4319 &align_attr.attr,
4320 &sanity_checks_attr.attr,
4321 &trace_attr.attr,
4322 &hwcache_align_attr.attr,
4323 &reclaim_account_attr.attr,
4324 &destroy_by_rcu_attr.attr,
4325 &red_zone_attr.attr,
4326 &poison_attr.attr,
4327 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004328 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004329 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004330 &alloc_calls_attr.attr,
4331 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004332#ifdef CONFIG_ZONE_DMA
4333 &cache_dma_attr.attr,
4334#endif
4335#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004336 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004337#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004338#ifdef CONFIG_SLUB_STATS
4339 &alloc_fastpath_attr.attr,
4340 &alloc_slowpath_attr.attr,
4341 &free_fastpath_attr.attr,
4342 &free_slowpath_attr.attr,
4343 &free_frozen_attr.attr,
4344 &free_add_partial_attr.attr,
4345 &free_remove_partial_attr.attr,
4346 &alloc_from_partial_attr.attr,
4347 &alloc_slab_attr.attr,
4348 &alloc_refill_attr.attr,
4349 &free_slab_attr.attr,
4350 &cpuslab_flush_attr.attr,
4351 &deactivate_full_attr.attr,
4352 &deactivate_empty_attr.attr,
4353 &deactivate_to_head_attr.attr,
4354 &deactivate_to_tail_attr.attr,
4355 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004356 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004357#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004358#ifdef CONFIG_FAILSLAB
4359 &failslab_attr.attr,
4360#endif
4361
Christoph Lameter81819f02007-05-06 14:49:36 -07004362 NULL
4363};
4364
4365static struct attribute_group slab_attr_group = {
4366 .attrs = slab_attrs,
4367};
4368
4369static ssize_t slab_attr_show(struct kobject *kobj,
4370 struct attribute *attr,
4371 char *buf)
4372{
4373 struct slab_attribute *attribute;
4374 struct kmem_cache *s;
4375 int err;
4376
4377 attribute = to_slab_attr(attr);
4378 s = to_slab(kobj);
4379
4380 if (!attribute->show)
4381 return -EIO;
4382
4383 err = attribute->show(s, buf);
4384
4385 return err;
4386}
4387
4388static ssize_t slab_attr_store(struct kobject *kobj,
4389 struct attribute *attr,
4390 const char *buf, size_t len)
4391{
4392 struct slab_attribute *attribute;
4393 struct kmem_cache *s;
4394 int err;
4395
4396 attribute = to_slab_attr(attr);
4397 s = to_slab(kobj);
4398
4399 if (!attribute->store)
4400 return -EIO;
4401
4402 err = attribute->store(s, buf, len);
4403
4404 return err;
4405}
4406
Christoph Lameter151c6022008-01-07 22:29:05 -08004407static void kmem_cache_release(struct kobject *kobj)
4408{
4409 struct kmem_cache *s = to_slab(kobj);
4410
4411 kfree(s);
4412}
4413
Emese Revfy52cf25d2010-01-19 02:58:23 +01004414static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004415 .show = slab_attr_show,
4416 .store = slab_attr_store,
4417};
4418
4419static struct kobj_type slab_ktype = {
4420 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004421 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004422};
4423
4424static int uevent_filter(struct kset *kset, struct kobject *kobj)
4425{
4426 struct kobj_type *ktype = get_ktype(kobj);
4427
4428 if (ktype == &slab_ktype)
4429 return 1;
4430 return 0;
4431}
4432
Emese Revfy9cd43612009-12-31 14:52:51 +01004433static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004434 .filter = uevent_filter,
4435};
4436
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004437static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004438
4439#define ID_STR_LENGTH 64
4440
4441/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004442 *
4443 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004444 */
4445static char *create_unique_id(struct kmem_cache *s)
4446{
4447 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4448 char *p = name;
4449
4450 BUG_ON(!name);
4451
4452 *p++ = ':';
4453 /*
4454 * First flags affecting slabcache operations. We will only
4455 * get here for aliasable slabs so we do not need to support
4456 * too many flags. The flags here must cover all flags that
4457 * are matched during merging to guarantee that the id is
4458 * unique.
4459 */
4460 if (s->flags & SLAB_CACHE_DMA)
4461 *p++ = 'd';
4462 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4463 *p++ = 'a';
4464 if (s->flags & SLAB_DEBUG_FREE)
4465 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004466 if (!(s->flags & SLAB_NOTRACK))
4467 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004468 if (p != name + 1)
4469 *p++ = '-';
4470 p += sprintf(p, "%07d", s->size);
4471 BUG_ON(p > name + ID_STR_LENGTH - 1);
4472 return name;
4473}
4474
4475static int sysfs_slab_add(struct kmem_cache *s)
4476{
4477 int err;
4478 const char *name;
4479 int unmergeable;
4480
4481 if (slab_state < SYSFS)
4482 /* Defer until later */
4483 return 0;
4484
4485 unmergeable = slab_unmergeable(s);
4486 if (unmergeable) {
4487 /*
4488 * Slabcache can never be merged so we can use the name proper.
4489 * This is typically the case for debug situations. In that
4490 * case we can catch duplicate names easily.
4491 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004492 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004493 name = s->name;
4494 } else {
4495 /*
4496 * Create a unique name for the slab as a target
4497 * for the symlinks.
4498 */
4499 name = create_unique_id(s);
4500 }
4501
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004502 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004503 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4504 if (err) {
4505 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004506 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004507 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004508
4509 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004510 if (err) {
4511 kobject_del(&s->kobj);
4512 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004513 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004514 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004515 kobject_uevent(&s->kobj, KOBJ_ADD);
4516 if (!unmergeable) {
4517 /* Setup first alias */
4518 sysfs_slab_alias(s, s->name);
4519 kfree(name);
4520 }
4521 return 0;
4522}
4523
4524static void sysfs_slab_remove(struct kmem_cache *s)
4525{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004526 if (slab_state < SYSFS)
4527 /*
4528 * Sysfs has not been setup yet so no need to remove the
4529 * cache from sysfs.
4530 */
4531 return;
4532
Christoph Lameter81819f02007-05-06 14:49:36 -07004533 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4534 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004535 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004536}
4537
4538/*
4539 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004540 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004541 */
4542struct saved_alias {
4543 struct kmem_cache *s;
4544 const char *name;
4545 struct saved_alias *next;
4546};
4547
Adrian Bunk5af328a2007-07-17 04:03:27 -07004548static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004549
4550static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4551{
4552 struct saved_alias *al;
4553
4554 if (slab_state == SYSFS) {
4555 /*
4556 * If we have a leftover link then remove it.
4557 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004558 sysfs_remove_link(&slab_kset->kobj, name);
4559 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004560 }
4561
4562 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4563 if (!al)
4564 return -ENOMEM;
4565
4566 al->s = s;
4567 al->name = name;
4568 al->next = alias_list;
4569 alias_list = al;
4570 return 0;
4571}
4572
4573static int __init slab_sysfs_init(void)
4574{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004575 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004576 int err;
4577
Christoph Lameter2bce6482010-07-19 11:39:11 -05004578 down_write(&slub_lock);
4579
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004580 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004581 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004582 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004583 printk(KERN_ERR "Cannot register slab subsystem.\n");
4584 return -ENOSYS;
4585 }
4586
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004587 slab_state = SYSFS;
4588
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004589 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004590 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004591 if (err)
4592 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4593 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004594 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004595
4596 while (alias_list) {
4597 struct saved_alias *al = alias_list;
4598
4599 alias_list = alias_list->next;
4600 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004601 if (err)
4602 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4603 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004604 kfree(al);
4605 }
4606
Christoph Lameter2bce6482010-07-19 11:39:11 -05004607 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004608 resiliency_test();
4609 return 0;
4610}
4611
4612__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004613#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004614
4615/*
4616 * The /proc/slabinfo ABI
4617 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004618#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004619static void print_slabinfo_header(struct seq_file *m)
4620{
4621 seq_puts(m, "slabinfo - version: 2.1\n");
4622 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4623 "<objperslab> <pagesperslab>");
4624 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4625 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4626 seq_putc(m, '\n');
4627}
4628
4629static void *s_start(struct seq_file *m, loff_t *pos)
4630{
4631 loff_t n = *pos;
4632
4633 down_read(&slub_lock);
4634 if (!n)
4635 print_slabinfo_header(m);
4636
4637 return seq_list_start(&slab_caches, *pos);
4638}
4639
4640static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4641{
4642 return seq_list_next(p, &slab_caches, pos);
4643}
4644
4645static void s_stop(struct seq_file *m, void *p)
4646{
4647 up_read(&slub_lock);
4648}
4649
4650static int s_show(struct seq_file *m, void *p)
4651{
4652 unsigned long nr_partials = 0;
4653 unsigned long nr_slabs = 0;
4654 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004655 unsigned long nr_objs = 0;
4656 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004657 struct kmem_cache *s;
4658 int node;
4659
4660 s = list_entry(p, struct kmem_cache, list);
4661
4662 for_each_online_node(node) {
4663 struct kmem_cache_node *n = get_node(s, node);
4664
4665 if (!n)
4666 continue;
4667
4668 nr_partials += n->nr_partial;
4669 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004670 nr_objs += atomic_long_read(&n->total_objects);
4671 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004672 }
4673
Christoph Lameter205ab992008-04-14 19:11:40 +03004674 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004675
4676 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004677 nr_objs, s->size, oo_objects(s->oo),
4678 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004679 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4680 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4681 0UL);
4682 seq_putc(m, '\n');
4683 return 0;
4684}
4685
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004686static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004687 .start = s_start,
4688 .next = s_next,
4689 .stop = s_stop,
4690 .show = s_show,
4691};
4692
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004693static int slabinfo_open(struct inode *inode, struct file *file)
4694{
4695 return seq_open(file, &slabinfo_op);
4696}
4697
4698static const struct file_operations proc_slabinfo_operations = {
4699 .open = slabinfo_open,
4700 .read = seq_read,
4701 .llseek = seq_lseek,
4702 .release = seq_release,
4703};
4704
4705static int __init slab_proc_init(void)
4706{
WANG Congcf5d1132009-08-18 19:11:40 +03004707 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004708 return 0;
4709}
4710module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004711#endif /* CONFIG_SLABINFO */