blob: 13fffe1f0f3dc5992471ffd590326b9517b0cc3a [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 */
171#define __SYSFS_ADD_DEFERRED 0x40000000UL /* Not yet visible via sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700172
173static int kmem_size = sizeof(struct kmem_cache);
174
175#ifdef CONFIG_SMP
176static struct notifier_block slab_notifier;
177#endif
178
179static enum {
180 DOWN, /* No slab functionality available */
181 PARTIAL, /* kmem_cache_open() works but kmalloc does not */
Christoph Lameter672bba32007-05-09 02:32:39 -0700182 UP, /* Everything works but does not show up in sysfs */
Christoph Lameter81819f02007-05-06 14:49:36 -0700183 SYSFS /* Sysfs up */
184} slab_state = DOWN;
185
186/* A list of all slab caches on the system */
187static DECLARE_RWSEM(slub_lock);
Adrian Bunk5af328a2007-07-17 04:03:27 -0700188static LIST_HEAD(slab_caches);
Christoph Lameter81819f02007-05-06 14:49:36 -0700189
Christoph Lameter02cbc872007-05-09 02:32:43 -0700190/*
191 * Tracking user of a slab.
192 */
193struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300194 unsigned long addr; /* Called from address */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700195 int cpu; /* Was running on cpu */
196 int pid; /* Pid context */
197 unsigned long when; /* When did the operation occur */
198};
199
200enum track_item { TRACK_ALLOC, TRACK_FREE };
201
Christoph Lameterf6acb632008-04-29 16:16:06 -0700202#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -0700203static int sysfs_slab_add(struct kmem_cache *);
204static int sysfs_slab_alias(struct kmem_cache *, const char *);
205static void sysfs_slab_remove(struct kmem_cache *);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800206
Christoph Lameter81819f02007-05-06 14:49:36 -0700207#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700208static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
209static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
210 { return 0; }
Christoph Lameter151c6022008-01-07 22:29:05 -0800211static inline void sysfs_slab_remove(struct kmem_cache *s)
212{
213 kfree(s);
214}
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800215
Christoph Lameter81819f02007-05-06 14:49:36 -0700216#endif
217
Christoph Lameter84e554e2009-12-18 16:26:23 -0600218static inline void stat(struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800219{
220#ifdef CONFIG_SLUB_STATS
Christoph Lameter84e554e2009-12-18 16:26:23 -0600221 __this_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800222#endif
223}
224
Christoph Lameter81819f02007-05-06 14:49:36 -0700225/********************************************************************
226 * Core slab cache functions
227 *******************************************************************/
228
229int slab_is_available(void)
230{
231 return slab_state >= UP;
232}
233
234static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
235{
236#ifdef CONFIG_NUMA
237 return s->node[node];
238#else
239 return &s->local_node;
240#endif
241}
242
Christoph Lameter6446faa2008-02-15 23:45:26 -0800243/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700244static inline int check_valid_pointer(struct kmem_cache *s,
245 struct page *page, const void *object)
246{
247 void *base;
248
Christoph Lametera973e9d2008-03-01 13:40:44 -0800249 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700250 return 1;
251
Christoph Lametera973e9d2008-03-01 13:40:44 -0800252 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300253 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700254 (object - base) % s->size) {
255 return 0;
256 }
257
258 return 1;
259}
260
Christoph Lameter7656c722007-05-09 02:32:40 -0700261static inline void *get_freepointer(struct kmem_cache *s, void *object)
262{
263 return *(void **)(object + s->offset);
264}
265
266static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
267{
268 *(void **)(object + s->offset) = fp;
269}
270
271/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300272#define for_each_object(__p, __s, __addr, __objects) \
273 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700274 __p += (__s)->size)
275
276/* Scan freelist */
277#define for_each_free_object(__p, __s, __free) \
Christoph Lametera973e9d2008-03-01 13:40:44 -0800278 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
Christoph Lameter7656c722007-05-09 02:32:40 -0700279
280/* Determine object index from a given position */
281static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
282{
283 return (p - addr) / s->size;
284}
285
Christoph Lameter834f3d12008-04-14 19:11:31 +0300286static inline struct kmem_cache_order_objects oo_make(int order,
287 unsigned long size)
288{
289 struct kmem_cache_order_objects x = {
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400290 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
Christoph Lameter834f3d12008-04-14 19:11:31 +0300291 };
292
293 return x;
294}
295
296static inline int oo_order(struct kmem_cache_order_objects x)
297{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400298 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300299}
300
301static inline int oo_objects(struct kmem_cache_order_objects x)
302{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400303 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300304}
305
Christoph Lameter41ecc552007-05-09 02:32:44 -0700306#ifdef CONFIG_SLUB_DEBUG
307/*
308 * Debug settings:
309 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700310#ifdef CONFIG_SLUB_DEBUG_ON
311static int slub_debug = DEBUG_DEFAULT_FLAGS;
312#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700313static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700314#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700315
316static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700317static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700318
Christoph Lameter7656c722007-05-09 02:32:40 -0700319/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700320 * Object debugging
321 */
322static void print_section(char *text, u8 *addr, unsigned int length)
323{
324 int i, offset;
325 int newline = 1;
326 char ascii[17];
327
328 ascii[16] = 0;
329
330 for (i = 0; i < length; i++) {
331 if (newline) {
Christoph Lameter24922682007-07-17 04:03:18 -0700332 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
Christoph Lameter81819f02007-05-06 14:49:36 -0700333 newline = 0;
334 }
Pekka Enberg06428782008-01-07 23:20:27 -0800335 printk(KERN_CONT " %02x", addr[i]);
Christoph Lameter81819f02007-05-06 14:49:36 -0700336 offset = i % 16;
337 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
338 if (offset == 15) {
Pekka Enberg06428782008-01-07 23:20:27 -0800339 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700340 newline = 1;
341 }
342 }
343 if (!newline) {
344 i %= 16;
345 while (i < 16) {
Pekka Enberg06428782008-01-07 23:20:27 -0800346 printk(KERN_CONT " ");
Christoph Lameter81819f02007-05-06 14:49:36 -0700347 ascii[i] = ' ';
348 i++;
349 }
Pekka Enberg06428782008-01-07 23:20:27 -0800350 printk(KERN_CONT " %s\n", ascii);
Christoph Lameter81819f02007-05-06 14:49:36 -0700351 }
352}
353
Christoph Lameter81819f02007-05-06 14:49:36 -0700354static struct track *get_track(struct kmem_cache *s, void *object,
355 enum track_item alloc)
356{
357 struct track *p;
358
359 if (s->offset)
360 p = object + s->offset + sizeof(void *);
361 else
362 p = object + s->inuse;
363
364 return p + alloc;
365}
366
367static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300368 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700369{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900370 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700371
Christoph Lameter81819f02007-05-06 14:49:36 -0700372 if (addr) {
373 p->addr = addr;
374 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400375 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700376 p->when = jiffies;
377 } else
378 memset(p, 0, sizeof(struct track));
379}
380
Christoph Lameter81819f02007-05-06 14:49:36 -0700381static void init_tracking(struct kmem_cache *s, void *object)
382{
Christoph Lameter24922682007-07-17 04:03:18 -0700383 if (!(s->flags & SLAB_STORE_USER))
384 return;
385
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300386 set_track(s, object, TRACK_FREE, 0UL);
387 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700388}
389
390static void print_track(const char *s, struct track *t)
391{
392 if (!t->addr)
393 return;
394
Linus Torvalds7daf7052008-07-14 12:12:53 -0700395 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300396 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Christoph Lameter81819f02007-05-06 14:49:36 -0700397}
398
Christoph Lameter24922682007-07-17 04:03:18 -0700399static void print_tracking(struct kmem_cache *s, void *object)
400{
401 if (!(s->flags & SLAB_STORE_USER))
402 return;
403
404 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
405 print_track("Freed", get_track(s, object, TRACK_FREE));
406}
407
408static void print_page_info(struct page *page)
409{
Christoph Lameter39b26462008-04-14 19:11:30 +0300410 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
411 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700412
413}
414
415static void slab_bug(struct kmem_cache *s, char *fmt, ...)
416{
417 va_list args;
418 char buf[100];
419
420 va_start(args, fmt);
421 vsnprintf(buf, sizeof(buf), fmt, args);
422 va_end(args);
423 printk(KERN_ERR "========================================"
424 "=====================================\n");
425 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
426 printk(KERN_ERR "----------------------------------------"
427 "-------------------------------------\n\n");
428}
429
430static void slab_fix(struct kmem_cache *s, char *fmt, ...)
431{
432 va_list args;
433 char buf[100];
434
435 va_start(args, fmt);
436 vsnprintf(buf, sizeof(buf), fmt, args);
437 va_end(args);
438 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
439}
440
441static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700442{
443 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800444 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700445
446 print_tracking(s, p);
447
448 print_page_info(page);
449
450 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
451 p, p - addr, get_freepointer(s, p));
452
453 if (p > addr + 16)
454 print_section("Bytes b4", p - 16, 16);
455
Pekka Enberg0ebd6522008-07-19 14:17:22 +0300456 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700457
458 if (s->flags & SLAB_RED_ZONE)
459 print_section("Redzone", p + s->objsize,
460 s->inuse - s->objsize);
461
Christoph Lameter81819f02007-05-06 14:49:36 -0700462 if (s->offset)
463 off = s->offset + sizeof(void *);
464 else
465 off = s->inuse;
466
Christoph Lameter24922682007-07-17 04:03:18 -0700467 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700468 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700469
470 if (off != s->size)
471 /* Beginning of the filler is the free pointer */
Christoph Lameter24922682007-07-17 04:03:18 -0700472 print_section("Padding", p + off, s->size - off);
473
474 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700475}
476
477static void object_err(struct kmem_cache *s, struct page *page,
478 u8 *object, char *reason)
479{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700480 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700481 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700482}
483
Christoph Lameter24922682007-07-17 04:03:18 -0700484static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700485{
486 va_list args;
487 char buf[100];
488
Christoph Lameter24922682007-07-17 04:03:18 -0700489 va_start(args, fmt);
490 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700491 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700492 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700493 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700494 dump_stack();
495}
496
497static void init_object(struct kmem_cache *s, void *object, int active)
498{
499 u8 *p = object;
500
501 if (s->flags & __OBJECT_POISON) {
502 memset(p, POISON_FREE, s->objsize - 1);
Pekka Enberg06428782008-01-07 23:20:27 -0800503 p[s->objsize - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700504 }
505
506 if (s->flags & SLAB_RED_ZONE)
507 memset(p + s->objsize,
508 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
509 s->inuse - s->objsize);
510}
511
Christoph Lameter24922682007-07-17 04:03:18 -0700512static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter81819f02007-05-06 14:49:36 -0700513{
514 while (bytes) {
515 if (*start != (u8)value)
Christoph Lameter24922682007-07-17 04:03:18 -0700516 return start;
Christoph Lameter81819f02007-05-06 14:49:36 -0700517 start++;
518 bytes--;
519 }
Christoph Lameter24922682007-07-17 04:03:18 -0700520 return NULL;
521}
522
523static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
524 void *from, void *to)
525{
526 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
527 memset(from, data, to - from);
528}
529
530static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
531 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800532 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700533{
534 u8 *fault;
535 u8 *end;
536
537 fault = check_bytes(start, value, bytes);
538 if (!fault)
539 return 1;
540
541 end = start + bytes;
542 while (end > fault && end[-1] == value)
543 end--;
544
545 slab_bug(s, "%s overwritten", what);
546 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
547 fault, end - 1, fault[0], value);
548 print_trailer(s, page, object);
549
550 restore_bytes(s, what, value, fault, end);
551 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700552}
553
Christoph Lameter81819f02007-05-06 14:49:36 -0700554/*
555 * Object layout:
556 *
557 * object address
558 * Bytes of the object to be managed.
559 * If the freepointer may overlay the object then the free
560 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700561 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700562 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
563 * 0xa5 (POISON_END)
564 *
565 * object + s->objsize
566 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700567 * Padding is extended by another word if Redzoning is enabled and
568 * objsize == inuse.
569 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700570 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
571 * 0xcc (RED_ACTIVE) for objects in use.
572 *
573 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700574 * Meta data starts here.
575 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700576 * A. Free pointer (if we cannot overwrite object on free)
577 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700578 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800579 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700580 * before the word boundary.
581 *
582 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700583 *
584 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700585 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700586 *
Christoph Lameter672bba32007-05-09 02:32:39 -0700587 * If slabcaches are merged then the objsize and inuse boundaries are mostly
588 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700589 * may be used with merged slabcaches.
590 */
591
Christoph Lameter81819f02007-05-06 14:49:36 -0700592static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
593{
594 unsigned long off = s->inuse; /* The end of info */
595
596 if (s->offset)
597 /* Freepointer is placed after the object. */
598 off += sizeof(void *);
599
600 if (s->flags & SLAB_STORE_USER)
601 /* We also have user information there */
602 off += 2 * sizeof(struct track);
603
604 if (s->size == off)
605 return 1;
606
Christoph Lameter24922682007-07-17 04:03:18 -0700607 return check_bytes_and_report(s, page, p, "Object padding",
608 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700609}
610
Christoph Lameter39b26462008-04-14 19:11:30 +0300611/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700612static int slab_pad_check(struct kmem_cache *s, struct page *page)
613{
Christoph Lameter24922682007-07-17 04:03:18 -0700614 u8 *start;
615 u8 *fault;
616 u8 *end;
617 int length;
618 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700619
620 if (!(s->flags & SLAB_POISON))
621 return 1;
622
Christoph Lametera973e9d2008-03-01 13:40:44 -0800623 start = page_address(page);
Christoph Lameter834f3d12008-04-14 19:11:31 +0300624 length = (PAGE_SIZE << compound_order(page));
Christoph Lameter39b26462008-04-14 19:11:30 +0300625 end = start + length;
626 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700627 if (!remainder)
628 return 1;
629
Christoph Lameter39b26462008-04-14 19:11:30 +0300630 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700631 if (!fault)
632 return 1;
633 while (end > fault && end[-1] == POISON_INUSE)
634 end--;
635
636 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Christoph Lameter39b26462008-04-14 19:11:30 +0300637 print_section("Padding", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700638
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200639 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700640 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700641}
642
643static int check_object(struct kmem_cache *s, struct page *page,
644 void *object, int active)
645{
646 u8 *p = object;
647 u8 *endobject = object + s->objsize;
648
649 if (s->flags & SLAB_RED_ZONE) {
650 unsigned int red =
651 active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
652
Christoph Lameter24922682007-07-17 04:03:18 -0700653 if (!check_bytes_and_report(s, page, object, "Redzone",
654 endobject, red, s->inuse - s->objsize))
Christoph Lameter81819f02007-05-06 14:49:36 -0700655 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700656 } else {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800657 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
658 check_bytes_and_report(s, page, p, "Alignment padding",
659 endobject, POISON_INUSE, s->inuse - s->objsize);
660 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700661 }
662
663 if (s->flags & SLAB_POISON) {
664 if (!active && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700665 (!check_bytes_and_report(s, page, p, "Poison", p,
666 POISON_FREE, s->objsize - 1) ||
667 !check_bytes_and_report(s, page, p, "Poison",
Pekka Enberg06428782008-01-07 23:20:27 -0800668 p + s->objsize - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700669 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700670 /*
671 * check_pad_bytes cleans up on its own.
672 */
673 check_pad_bytes(s, page, p);
674 }
675
676 if (!s->offset && active)
677 /*
678 * Object and freepointer overlap. Cannot check
679 * freepointer while object is allocated.
680 */
681 return 1;
682
683 /* Check free pointer validity */
684 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
685 object_err(s, page, p, "Freepointer corrupt");
686 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100687 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700688 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700689 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700690 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800691 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700692 return 0;
693 }
694 return 1;
695}
696
697static int check_slab(struct kmem_cache *s, struct page *page)
698{
Christoph Lameter39b26462008-04-14 19:11:30 +0300699 int maxobj;
700
Christoph Lameter81819f02007-05-06 14:49:36 -0700701 VM_BUG_ON(!irqs_disabled());
702
703 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700704 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700705 return 0;
706 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300707
708 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
709 if (page->objects > maxobj) {
710 slab_err(s, page, "objects %u > max %u",
711 s->name, page->objects, maxobj);
712 return 0;
713 }
714 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700715 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300716 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700717 return 0;
718 }
719 /* Slab_pad_check fixes things up after itself */
720 slab_pad_check(s, page);
721 return 1;
722}
723
724/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700725 * Determine if a certain object on a page is on the freelist. Must hold the
726 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700727 */
728static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
729{
730 int nr = 0;
731 void *fp = page->freelist;
732 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300733 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700734
Christoph Lameter39b26462008-04-14 19:11:30 +0300735 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700736 if (fp == search)
737 return 1;
738 if (!check_valid_pointer(s, page, fp)) {
739 if (object) {
740 object_err(s, page, object,
741 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800742 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700743 break;
744 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700745 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800746 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300747 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700748 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700749 return 0;
750 }
751 break;
752 }
753 object = fp;
754 fp = get_freepointer(s, object);
755 nr++;
756 }
757
Christoph Lameter224a88b2008-04-14 19:11:31 +0300758 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400759 if (max_objects > MAX_OBJS_PER_PAGE)
760 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300761
762 if (page->objects != max_objects) {
763 slab_err(s, page, "Wrong number of objects. Found %d but "
764 "should be %d", page->objects, max_objects);
765 page->objects = max_objects;
766 slab_fix(s, "Number of objects adjusted.");
767 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300768 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700769 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300770 "counted were %d", page->inuse, page->objects - nr);
771 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700772 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700773 }
774 return search == NULL;
775}
776
Christoph Lameter0121c6192008-04-29 16:11:12 -0700777static void trace(struct kmem_cache *s, struct page *page, void *object,
778 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700779{
780 if (s->flags & SLAB_TRACE) {
781 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
782 s->name,
783 alloc ? "alloc" : "free",
784 object, page->inuse,
785 page->freelist);
786
787 if (!alloc)
788 print_section("Object", (void *)object, s->objsize);
789
790 dump_stack();
791 }
792}
793
Christoph Lameter643b1132007-05-06 14:49:42 -0700794/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700795 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700796 */
Christoph Lametere95eed52007-05-06 14:49:44 -0700797static void add_full(struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700798{
Christoph Lameter643b1132007-05-06 14:49:42 -0700799 spin_lock(&n->list_lock);
800 list_add(&page->lru, &n->full);
801 spin_unlock(&n->list_lock);
802}
803
804static void remove_full(struct kmem_cache *s, struct page *page)
805{
806 struct kmem_cache_node *n;
807
808 if (!(s->flags & SLAB_STORE_USER))
809 return;
810
811 n = get_node(s, page_to_nid(page));
812
813 spin_lock(&n->list_lock);
814 list_del(&page->lru);
815 spin_unlock(&n->list_lock);
816}
817
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300818/* Tracking of the number of slabs for debugging purposes */
819static inline unsigned long slabs_node(struct kmem_cache *s, int node)
820{
821 struct kmem_cache_node *n = get_node(s, node);
822
823 return atomic_long_read(&n->nr_slabs);
824}
825
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400826static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
827{
828 return atomic_long_read(&n->nr_slabs);
829}
830
Christoph Lameter205ab992008-04-14 19:11:40 +0300831static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300832{
833 struct kmem_cache_node *n = get_node(s, node);
834
835 /*
836 * May be called early in order to allocate a slab for the
837 * kmem_cache_node structure. Solve the chicken-egg
838 * dilemma by deferring the increment of the count during
839 * bootstrap (see early_kmem_cache_node_alloc).
840 */
Christoph Lameter205ab992008-04-14 19:11:40 +0300841 if (!NUMA_BUILD || n) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300842 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300843 atomic_long_add(objects, &n->total_objects);
844 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300845}
Christoph Lameter205ab992008-04-14 19:11:40 +0300846static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300847{
848 struct kmem_cache_node *n = get_node(s, node);
849
850 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +0300851 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300852}
853
854/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -0700855static void setup_object_debug(struct kmem_cache *s, struct page *page,
856 void *object)
857{
858 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
859 return;
860
861 init_object(s, object, 0);
862 init_tracking(s, object);
863}
864
865static int alloc_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300866 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700867{
868 if (!check_slab(s, page))
869 goto bad;
870
Christoph Lameterd692ef62008-02-15 23:45:24 -0800871 if (!on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700872 object_err(s, page, object, "Object already allocated");
Christoph Lameter70d71222007-05-06 14:49:47 -0700873 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700874 }
875
876 if (!check_valid_pointer(s, page, object)) {
877 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -0700878 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700879 }
880
Christoph Lameterd692ef62008-02-15 23:45:24 -0800881 if (!check_object(s, page, object, 0))
Christoph Lameter81819f02007-05-06 14:49:36 -0700882 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -0700883
Christoph Lameter3ec09742007-05-16 22:11:00 -0700884 /* Success perform special debug activities for allocs */
885 if (s->flags & SLAB_STORE_USER)
886 set_track(s, object, TRACK_ALLOC, addr);
887 trace(s, page, object, 1);
888 init_object(s, object, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700889 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700890
Christoph Lameter81819f02007-05-06 14:49:36 -0700891bad:
892 if (PageSlab(page)) {
893 /*
894 * If this is a slab page then lets do the best we can
895 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -0700896 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -0700897 */
Christoph Lameter24922682007-07-17 04:03:18 -0700898 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +0300899 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -0800900 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -0700901 }
902 return 0;
903}
904
Christoph Lameter3ec09742007-05-16 22:11:00 -0700905static int free_debug_processing(struct kmem_cache *s, struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300906 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700907{
908 if (!check_slab(s, page))
909 goto fail;
910
911 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700912 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700913 goto fail;
914 }
915
916 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700917 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -0700918 goto fail;
919 }
920
921 if (!check_object(s, page, object, 1))
922 return 0;
923
924 if (unlikely(s != page->slab)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800925 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700926 slab_err(s, page, "Attempt to free object(0x%p) "
927 "outside of slab", object);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800928 } else if (!page->slab) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700929 printk(KERN_ERR
Christoph Lameter70d71222007-05-06 14:49:47 -0700930 "SLUB <none>: no slab for object 0x%p.\n",
Christoph Lameter81819f02007-05-06 14:49:36 -0700931 object);
Christoph Lameter70d71222007-05-06 14:49:47 -0700932 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -0800933 } else
Christoph Lameter24922682007-07-17 04:03:18 -0700934 object_err(s, page, object,
935 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700936 goto fail;
937 }
Christoph Lameter3ec09742007-05-16 22:11:00 -0700938
939 /* Special debug activities for freeing objects */
Andy Whitcroft8a380822008-07-23 21:27:18 -0700940 if (!PageSlubFrozen(page) && !page->freelist)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700941 remove_full(s, page);
942 if (s->flags & SLAB_STORE_USER)
943 set_track(s, object, TRACK_FREE, addr);
944 trace(s, page, object, 0);
945 init_object(s, object, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -0700946 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -0700947
Christoph Lameter81819f02007-05-06 14:49:36 -0700948fail:
Christoph Lameter24922682007-07-17 04:03:18 -0700949 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700950 return 0;
951}
952
Christoph Lameter41ecc552007-05-09 02:32:44 -0700953static int __init setup_slub_debug(char *str)
954{
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700955 slub_debug = DEBUG_DEFAULT_FLAGS;
956 if (*str++ != '=' || !*str)
957 /*
958 * No options specified. Switch on full debugging.
959 */
960 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700961
962 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700963 /*
964 * No options but restriction on slabs. This means full
965 * debugging for slabs matching a pattern.
966 */
967 goto check_slabs;
968
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700969 if (tolower(*str) == 'o') {
970 /*
971 * Avoid enabling debugging on caches if its minimum order
972 * would increase as a result.
973 */
974 disable_higher_order_debug = 1;
975 goto out;
976 }
977
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700978 slub_debug = 0;
979 if (*str == '-')
980 /*
981 * Switch off all debugging measures.
982 */
983 goto out;
984
985 /*
986 * Determine which debug features should be switched on
987 */
Pekka Enberg06428782008-01-07 23:20:27 -0800988 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700989 switch (tolower(*str)) {
990 case 'f':
991 slub_debug |= SLAB_DEBUG_FREE;
992 break;
993 case 'z':
994 slub_debug |= SLAB_RED_ZONE;
995 break;
996 case 'p':
997 slub_debug |= SLAB_POISON;
998 break;
999 case 'u':
1000 slub_debug |= SLAB_STORE_USER;
1001 break;
1002 case 't':
1003 slub_debug |= SLAB_TRACE;
1004 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001005 case 'a':
1006 slub_debug |= SLAB_FAILSLAB;
1007 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001008 default:
1009 printk(KERN_ERR "slub_debug option '%c' "
Pekka Enberg06428782008-01-07 23:20:27 -08001010 "unknown. skipped\n", *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001011 }
1012 }
1013
1014check_slabs:
1015 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001016 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001017out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001018 return 1;
1019}
1020
1021__setup("slub_debug", setup_slub_debug);
1022
Christoph Lameterba0268a2007-09-11 15:24:11 -07001023static unsigned long kmem_cache_flags(unsigned long objsize,
1024 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001025 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001026{
1027 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001028 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001029 */
Christoph Lametere1533622008-02-15 23:45:24 -08001030 if (slub_debug && (!slub_debug_slabs ||
David Rientjes3de47212009-07-27 18:30:35 -07001031 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1032 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001033
1034 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001035}
1036#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001037static inline void setup_object_debug(struct kmem_cache *s,
1038 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001039
Christoph Lameter3ec09742007-05-16 22:11:00 -07001040static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001041 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001042
Christoph Lameter3ec09742007-05-16 22:11:00 -07001043static inline int free_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001044 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001045
Christoph Lameter41ecc552007-05-09 02:32:44 -07001046static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1047 { return 1; }
1048static inline int check_object(struct kmem_cache *s, struct page *page,
1049 void *object, int active) { return 1; }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001050static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
Christoph Lameterba0268a2007-09-11 15:24:11 -07001051static inline unsigned long kmem_cache_flags(unsigned long objsize,
1052 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001053 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001054{
1055 return flags;
1056}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001057#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001058
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001059#define disable_higher_order_debug 0
1060
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001061static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1062 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001063static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1064 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001065static inline void inc_slabs_node(struct kmem_cache *s, int node,
1066 int objects) {}
1067static inline void dec_slabs_node(struct kmem_cache *s, int node,
1068 int objects) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001069#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03001070
Christoph Lameter81819f02007-05-06 14:49:36 -07001071/*
1072 * Slab allocation and freeing
1073 */
Christoph Lameter65c33762008-04-14 19:11:40 +03001074static inline struct page *alloc_slab_page(gfp_t flags, int node,
1075 struct kmem_cache_order_objects oo)
1076{
1077 int order = oo_order(oo);
1078
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001079 flags |= __GFP_NOTRACK;
1080
Christoph Lameter2154a332010-07-09 14:07:10 -05001081 if (node == NUMA_NO_NODE)
Christoph Lameter65c33762008-04-14 19:11:40 +03001082 return alloc_pages(flags, order);
1083 else
Minchan Kim6b65aaf2010-04-14 23:58:36 +09001084 return alloc_pages_exact_node(node, flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001085}
1086
Christoph Lameter81819f02007-05-06 14:49:36 -07001087static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1088{
Pekka Enberg06428782008-01-07 23:20:27 -08001089 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001090 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001091 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001092
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001093 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001094
Pekka Enbergba522702009-06-24 21:59:51 +03001095 /*
1096 * Let the initial higher-order allocation fail under memory pressure
1097 * so we fall-back to the minimum order allocation.
1098 */
1099 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1100
1101 page = alloc_slab_page(alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001102 if (unlikely(!page)) {
1103 oo = s->min;
1104 /*
1105 * Allocation may have failed due to fragmentation.
1106 * Try a lower order alloc if possible
1107 */
1108 page = alloc_slab_page(flags, node, oo);
1109 if (!page)
1110 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001111
Christoph Lameter84e554e2009-12-18 16:26:23 -06001112 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001113 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001114
1115 if (kmemcheck_enabled
Amerigo Wang5086c382009-08-19 21:44:13 +03001116 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001117 int pages = 1 << oo_order(oo);
1118
1119 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1120
1121 /*
1122 * Objects from caches that have a constructor don't get
1123 * cleared when they're allocated, so we need to do it here.
1124 */
1125 if (s->ctor)
1126 kmemcheck_mark_uninitialized_pages(page, pages);
1127 else
1128 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001129 }
1130
Christoph Lameter834f3d12008-04-14 19:11:31 +03001131 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001132 mod_zone_page_state(page_zone(page),
1133 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1134 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001135 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001136
1137 return page;
1138}
1139
1140static void setup_object(struct kmem_cache *s, struct page *page,
1141 void *object)
1142{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001143 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001144 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001145 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001146}
1147
1148static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1149{
1150 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001151 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001152 void *last;
1153 void *p;
1154
Christoph Lameter6cb06222007-10-16 01:25:41 -07001155 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001156
Christoph Lameter6cb06222007-10-16 01:25:41 -07001157 page = allocate_slab(s,
1158 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001159 if (!page)
1160 goto out;
1161
Christoph Lameter205ab992008-04-14 19:11:40 +03001162 inc_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001163 page->slab = s;
1164 page->flags |= 1 << PG_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07001165
1166 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001167
1168 if (unlikely(s->flags & SLAB_POISON))
Christoph Lameter834f3d12008-04-14 19:11:31 +03001169 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
Christoph Lameter81819f02007-05-06 14:49:36 -07001170
1171 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001172 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001173 setup_object(s, page, last);
1174 set_freepointer(s, last, p);
1175 last = p;
1176 }
1177 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001178 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001179
1180 page->freelist = start;
1181 page->inuse = 0;
1182out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001183 return page;
1184}
1185
1186static void __free_slab(struct kmem_cache *s, struct page *page)
1187{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001188 int order = compound_order(page);
1189 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001190
Christoph Lameteraf537b02010-07-09 14:07:14 -05001191 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001192 void *p;
1193
1194 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001195 for_each_object(p, s, page_address(page),
1196 page->objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001197 check_object(s, page, p, 0);
Christoph Lameter81819f02007-05-06 14:49:36 -07001198 }
1199
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001200 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001201
Christoph Lameter81819f02007-05-06 14:49:36 -07001202 mod_zone_page_state(page_zone(page),
1203 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1204 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001205 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001206
Christoph Lameter49bd5222008-04-14 18:52:18 +03001207 __ClearPageSlab(page);
1208 reset_page_mapcount(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001209 if (current->reclaim_state)
1210 current->reclaim_state->reclaimed_slab += pages;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001211 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001212}
1213
1214static void rcu_free_slab(struct rcu_head *h)
1215{
1216 struct page *page;
1217
1218 page = container_of((struct list_head *)h, struct page, lru);
1219 __free_slab(page->slab, page);
1220}
1221
1222static void free_slab(struct kmem_cache *s, struct page *page)
1223{
1224 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1225 /*
1226 * RCU free overloads the RCU head over the LRU
1227 */
1228 struct rcu_head *head = (void *)&page->lru;
1229
1230 call_rcu(head, rcu_free_slab);
1231 } else
1232 __free_slab(s, page);
1233}
1234
1235static void discard_slab(struct kmem_cache *s, struct page *page)
1236{
Christoph Lameter205ab992008-04-14 19:11:40 +03001237 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001238 free_slab(s, page);
1239}
1240
1241/*
1242 * Per slab locking using the pagelock
1243 */
1244static __always_inline void slab_lock(struct page *page)
1245{
1246 bit_spin_lock(PG_locked, &page->flags);
1247}
1248
1249static __always_inline void slab_unlock(struct page *page)
1250{
Nick Piggina76d3542008-01-07 23:20:27 -08001251 __bit_spin_unlock(PG_locked, &page->flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001252}
1253
1254static __always_inline int slab_trylock(struct page *page)
1255{
1256 int rc = 1;
1257
1258 rc = bit_spin_trylock(PG_locked, &page->flags);
1259 return rc;
1260}
1261
1262/*
1263 * Management of partially allocated slabs
1264 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001265static void add_partial(struct kmem_cache_node *n,
1266 struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001267{
Christoph Lametere95eed52007-05-06 14:49:44 -07001268 spin_lock(&n->list_lock);
1269 n->nr_partial++;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001270 if (tail)
1271 list_add_tail(&page->lru, &n->partial);
1272 else
1273 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001274 spin_unlock(&n->list_lock);
1275}
1276
Christoph Lameter0121c6192008-04-29 16:11:12 -07001277static void remove_partial(struct kmem_cache *s, struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001278{
1279 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1280
1281 spin_lock(&n->list_lock);
1282 list_del(&page->lru);
1283 n->nr_partial--;
1284 spin_unlock(&n->list_lock);
1285}
1286
1287/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001288 * Lock slab and remove from the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001289 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001290 * Must hold list_lock.
Christoph Lameter81819f02007-05-06 14:49:36 -07001291 */
Christoph Lameter0121c6192008-04-29 16:11:12 -07001292static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1293 struct page *page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001294{
1295 if (slab_trylock(page)) {
1296 list_del(&page->lru);
1297 n->nr_partial--;
Andy Whitcroft8a380822008-07-23 21:27:18 -07001298 __SetPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001299 return 1;
1300 }
1301 return 0;
1302}
1303
1304/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001305 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001306 */
1307static struct page *get_partial_node(struct kmem_cache_node *n)
1308{
1309 struct page *page;
1310
1311 /*
1312 * Racy check. If we mistakenly see no partial slabs then we
1313 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001314 * partial slab and there is none available then get_partials()
1315 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001316 */
1317 if (!n || !n->nr_partial)
1318 return NULL;
1319
1320 spin_lock(&n->list_lock);
1321 list_for_each_entry(page, &n->partial, lru)
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001322 if (lock_and_freeze_slab(n, page))
Christoph Lameter81819f02007-05-06 14:49:36 -07001323 goto out;
1324 page = NULL;
1325out:
1326 spin_unlock(&n->list_lock);
1327 return page;
1328}
1329
1330/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001331 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001332 */
1333static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1334{
1335#ifdef CONFIG_NUMA
1336 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001337 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001338 struct zone *zone;
1339 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07001340 struct page *page;
1341
1342 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001343 * The defrag ratio allows a configuration of the tradeoffs between
1344 * inter node defragmentation and node local allocations. A lower
1345 * defrag_ratio increases the tendency to do local allocations
1346 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001347 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001348 * If the defrag_ratio is set to 0 then kmalloc() always
1349 * returns node local objects. If the ratio is higher then kmalloc()
1350 * may return off node objects because partial slabs are obtained
1351 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001352 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001353 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001354 * defrag_ratio = 1000) then every (well almost) allocation will
1355 * first attempt to defrag slab caches on other nodes. This means
1356 * scanning over all nodes to look for partial slabs which may be
1357 * expensive if we do it every time we are trying to find a slab
1358 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001359 */
Christoph Lameter98246012008-01-07 23:20:26 -08001360 if (!s->remote_node_defrag_ratio ||
1361 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001362 return NULL;
1363
Miao Xiec0ff7452010-05-24 14:32:08 -07001364 get_mems_allowed();
Mel Gorman0e884602008-04-28 02:12:14 -07001365 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
Mel Gorman54a6eb52008-04-28 02:12:16 -07001366 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001367 struct kmem_cache_node *n;
1368
Mel Gorman54a6eb52008-04-28 02:12:16 -07001369 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001370
Mel Gorman54a6eb52008-04-28 02:12:16 -07001371 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
David Rientjes3b89d7d2009-02-22 17:40:07 -08001372 n->nr_partial > s->min_partial) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001373 page = get_partial_node(n);
Miao Xiec0ff7452010-05-24 14:32:08 -07001374 if (page) {
1375 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001376 return page;
Miao Xiec0ff7452010-05-24 14:32:08 -07001377 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001378 }
1379 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001380 put_mems_allowed();
Christoph Lameter81819f02007-05-06 14:49:36 -07001381#endif
1382 return NULL;
1383}
1384
1385/*
1386 * Get a partial page, lock it and return it.
1387 */
1388static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1389{
1390 struct page *page;
Christoph Lameter2154a332010-07-09 14:07:10 -05001391 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001392
1393 page = get_partial_node(get_node(s, searchnode));
Christoph Lameterbc6488e2010-07-26 10:41:14 -05001394 if (page || node != -1)
Christoph Lameter81819f02007-05-06 14:49:36 -07001395 return page;
1396
1397 return get_any_partial(s, flags);
1398}
1399
1400/*
1401 * Move a page back to the lists.
1402 *
1403 * Must be called with the slab lock held.
1404 *
1405 * On exit the slab lock will have been dropped.
1406 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001407static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001408{
Christoph Lametere95eed52007-05-06 14:49:44 -07001409 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1410
Andy Whitcroft8a380822008-07-23 21:27:18 -07001411 __ClearPageSlubFrozen(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001412 if (page->inuse) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001413
Christoph Lametera973e9d2008-03-01 13:40:44 -08001414 if (page->freelist) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001415 add_partial(n, page, tail);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001416 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001417 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001418 stat(s, DEACTIVATE_FULL);
Christoph Lameteraf537b02010-07-09 14:07:14 -05001419 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001420 add_full(n, page);
1421 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001422 slab_unlock(page);
1423 } else {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001424 stat(s, DEACTIVATE_EMPTY);
David Rientjes3b89d7d2009-02-22 17:40:07 -08001425 if (n->nr_partial < s->min_partial) {
Christoph Lametere95eed52007-05-06 14:49:44 -07001426 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001427 * Adding an empty slab to the partial slabs in order
1428 * to avoid page allocator overhead. This slab needs
1429 * to come after the other slabs with objects in
Christoph Lameter6446faa2008-02-15 23:45:26 -08001430 * so that the others get filled first. That way the
1431 * size of the partial list stays small.
1432 *
Christoph Lameter0121c6192008-04-29 16:11:12 -07001433 * kmem_cache_shrink can reclaim any empty slabs from
1434 * the partial list.
Christoph Lametere95eed52007-05-06 14:49:44 -07001435 */
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001436 add_partial(n, page, 1);
Christoph Lametere95eed52007-05-06 14:49:44 -07001437 slab_unlock(page);
1438 } else {
1439 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001440 stat(s, FREE_SLAB);
Christoph Lametere95eed52007-05-06 14:49:44 -07001441 discard_slab(s, page);
1442 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001443 }
1444}
1445
1446/*
1447 * Remove the cpu slab
1448 */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001449static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001450{
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001451 struct page *page = c->page;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001452 int tail = 1;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001453
Christoph Lameterb773ad72008-03-04 11:10:17 -08001454 if (page->freelist)
Christoph Lameter84e554e2009-12-18 16:26:23 -06001455 stat(s, DEACTIVATE_REMOTE_FREES);
Christoph Lameter894b8782007-05-10 03:15:16 -07001456 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001457 * Merge cpu freelist into slab freelist. Typically we get here
Christoph Lameter894b8782007-05-10 03:15:16 -07001458 * because both freelists are empty. So this is unlikely
1459 * to occur.
1460 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001461 while (unlikely(c->freelist)) {
Christoph Lameter894b8782007-05-10 03:15:16 -07001462 void **object;
1463
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001464 tail = 0; /* Hot objects. Put the slab first */
1465
Christoph Lameter894b8782007-05-10 03:15:16 -07001466 /* Retrieve object from cpu_freelist */
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001467 object = c->freelist;
Christoph Lameterff120592009-12-18 16:26:22 -06001468 c->freelist = get_freepointer(s, c->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001469
1470 /* And put onto the regular freelist */
Christoph Lameterff120592009-12-18 16:26:22 -06001471 set_freepointer(s, object, page->freelist);
Christoph Lameter894b8782007-05-10 03:15:16 -07001472 page->freelist = object;
1473 page->inuse--;
1474 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001475 c->page = NULL;
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001476 unfreeze_slab(s, page, tail);
Christoph Lameter81819f02007-05-06 14:49:36 -07001477}
1478
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001479static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001480{
Christoph Lameter84e554e2009-12-18 16:26:23 -06001481 stat(s, CPUSLAB_FLUSH);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001482 slab_lock(c->page);
1483 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001484}
1485
1486/*
1487 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08001488 *
Christoph Lameter81819f02007-05-06 14:49:36 -07001489 * Called from IPI handler with interrupts disabled.
1490 */
Christoph Lameter0c710012007-07-17 04:03:24 -07001491static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07001492{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001493 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07001494
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001495 if (likely(c && c->page))
1496 flush_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001497}
1498
1499static void flush_cpu_slab(void *d)
1500{
1501 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07001502
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001503 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07001504}
1505
1506static void flush_all(struct kmem_cache *s)
1507{
Jens Axboe15c8b6c2008-05-09 09:39:44 +02001508 on_each_cpu(flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07001509}
1510
1511/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001512 * Check if the objects in a per cpu structure fit numa
1513 * locality expectations.
1514 */
1515static inline int node_match(struct kmem_cache_cpu *c, int node)
1516{
1517#ifdef CONFIG_NUMA
Christoph Lameter2154a332010-07-09 14:07:10 -05001518 if (node != NUMA_NO_NODE && c->node != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001519 return 0;
1520#endif
1521 return 1;
1522}
1523
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001524static int count_free(struct page *page)
1525{
1526 return page->objects - page->inuse;
1527}
1528
1529static unsigned long count_partial(struct kmem_cache_node *n,
1530 int (*get_count)(struct page *))
1531{
1532 unsigned long flags;
1533 unsigned long x = 0;
1534 struct page *page;
1535
1536 spin_lock_irqsave(&n->list_lock, flags);
1537 list_for_each_entry(page, &n->partial, lru)
1538 x += get_count(page);
1539 spin_unlock_irqrestore(&n->list_lock, flags);
1540 return x;
1541}
1542
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001543static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1544{
1545#ifdef CONFIG_SLUB_DEBUG
1546 return atomic_long_read(&n->total_objects);
1547#else
1548 return 0;
1549#endif
1550}
1551
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001552static noinline void
1553slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1554{
1555 int node;
1556
1557 printk(KERN_WARNING
1558 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1559 nid, gfpflags);
1560 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1561 "default order: %d, min order: %d\n", s->name, s->objsize,
1562 s->size, oo_order(s->oo), oo_order(s->min));
1563
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001564 if (oo_order(s->min) > get_order(s->objsize))
1565 printk(KERN_WARNING " %s debugging increased min order, use "
1566 "slub_debug=O to disable.\n", s->name);
1567
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001568 for_each_online_node(node) {
1569 struct kmem_cache_node *n = get_node(s, node);
1570 unsigned long nr_slabs;
1571 unsigned long nr_objs;
1572 unsigned long nr_free;
1573
1574 if (!n)
1575 continue;
1576
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001577 nr_free = count_partial(n, count_free);
1578 nr_slabs = node_nr_slabs(n);
1579 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03001580
1581 printk(KERN_WARNING
1582 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1583 node, nr_slabs, nr_objs, nr_free);
1584 }
1585}
1586
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001587/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001588 * Slow path. The lockless freelist is empty or we need to perform
1589 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07001590 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001591 * Interrupts are disabled.
Christoph Lameter81819f02007-05-06 14:49:36 -07001592 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001593 * Processing is still very fast if new objects have been freed to the
1594 * regular freelist. In that case we simply take over the regular freelist
1595 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001596 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001597 * If that is not working then we fall back to the partial lists. We take the
1598 * first element of the freelist as the object to allocate now and move the
1599 * rest of the freelist to the lockless freelist.
1600 *
1601 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08001602 * we need to allocate a new slab. This is the slowest path since it involves
1603 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07001604 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001605static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1606 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001607{
Christoph Lameter81819f02007-05-06 14:49:36 -07001608 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001609 struct page *new;
Christoph Lameter81819f02007-05-06 14:49:36 -07001610
Linus Torvaldse72e9c22008-03-27 20:56:33 -07001611 /* We handle __GFP_ZERO in the caller */
1612 gfpflags &= ~__GFP_ZERO;
1613
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001614 if (!c->page)
Christoph Lameter81819f02007-05-06 14:49:36 -07001615 goto new_slab;
1616
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001617 slab_lock(c->page);
1618 if (unlikely(!node_match(c, node)))
Christoph Lameter81819f02007-05-06 14:49:36 -07001619 goto another_slab;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001620
Christoph Lameter84e554e2009-12-18 16:26:23 -06001621 stat(s, ALLOC_REFILL);
Christoph Lameter6446faa2008-02-15 23:45:26 -08001622
Christoph Lameter894b8782007-05-10 03:15:16 -07001623load_freelist:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001624 object = c->page->freelist;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001625 if (unlikely(!object))
Christoph Lameter81819f02007-05-06 14:49:36 -07001626 goto another_slab;
Christoph Lameteraf537b02010-07-09 14:07:14 -05001627 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001628 goto debug;
1629
Christoph Lameterff120592009-12-18 16:26:22 -06001630 c->freelist = get_freepointer(s, object);
Christoph Lameter39b26462008-04-14 19:11:30 +03001631 c->page->inuse = c->page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001632 c->page->freelist = NULL;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001633 c->node = page_to_nid(c->page);
Christoph Lameter1f842602008-01-07 23:20:30 -08001634unlock_out:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001635 slab_unlock(c->page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001636 stat(s, ALLOC_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001637 return object;
1638
1639another_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001640 deactivate_slab(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001641
1642new_slab:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001643 new = get_partial(s, gfpflags, node);
1644 if (new) {
1645 c->page = new;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001646 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter894b8782007-05-10 03:15:16 -07001647 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001648 }
1649
Christoph Lameterb811c202007-10-16 23:25:51 -07001650 if (gfpflags & __GFP_WAIT)
1651 local_irq_enable();
1652
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001653 new = new_slab(s, gfpflags, node);
Christoph Lameterb811c202007-10-16 23:25:51 -07001654
1655 if (gfpflags & __GFP_WAIT)
1656 local_irq_disable();
1657
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001658 if (new) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001659 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001660 stat(s, ALLOC_SLAB);
Christoph Lameter05aa3452007-11-05 11:31:58 -08001661 if (c->page)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001662 flush_slab(s, c);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001663 slab_lock(new);
Andy Whitcroft8a380822008-07-23 21:27:18 -07001664 __SetPageSlubFrozen(new);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001665 c->page = new;
Christoph Lameter4b6f0752007-05-16 22:10:53 -07001666 goto load_freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001667 }
Pekka Enberg95f85982009-06-11 16:18:09 +03001668 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1669 slab_out_of_memory(s, gfpflags, node);
Christoph Lameter71c7a062008-02-14 14:28:01 -08001670 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001671debug:
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001672 if (!alloc_debug_processing(s, c->page, object, addr))
Christoph Lameter81819f02007-05-06 14:49:36 -07001673 goto another_slab;
Christoph Lameter894b8782007-05-10 03:15:16 -07001674
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001675 c->page->inuse++;
Christoph Lameterff120592009-12-18 16:26:22 -06001676 c->page->freelist = get_freepointer(s, object);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001677 c->node = -1;
Christoph Lameter1f842602008-01-07 23:20:30 -08001678 goto unlock_out;
Christoph Lameter894b8782007-05-10 03:15:16 -07001679}
1680
1681/*
1682 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1683 * have the fastpath folded into their functions. So no function call
1684 * overhead for requests that can be satisfied on the fastpath.
1685 *
1686 * The fastpath works by first checking if the lockless freelist can be used.
1687 * If not then __slab_alloc is called for slow processing.
1688 *
1689 * Otherwise we can simply pick the next object from the lockless free list.
1690 */
Pekka Enberg06428782008-01-07 23:20:27 -08001691static __always_inline void *slab_alloc(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001692 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001693{
Christoph Lameter894b8782007-05-10 03:15:16 -07001694 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001695 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001696 unsigned long flags;
1697
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10001698 gfpflags &= gfp_allowed_mask;
Pekka Enberg7e85ee02009-06-12 14:03:06 +03001699
Nick Piggincf40bd12009-01-21 08:12:39 +01001700 lockdep_trace_alloc(gfpflags);
OGAWA Hirofumi89124d72008-11-19 21:23:59 +09001701 might_sleep_if(gfpflags & __GFP_WAIT);
Pekka Enberg3c506ef2008-12-29 11:47:05 +02001702
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001703 if (should_failslab(s->objsize, gfpflags, s->flags))
Akinobu Mita773ff602008-12-23 19:37:01 +09001704 return NULL;
1705
Christoph Lameter894b8782007-05-10 03:15:16 -07001706 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001707 c = __this_cpu_ptr(s->cpu_slab);
1708 object = c->freelist;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001709 if (unlikely(!object || !node_match(c, node)))
Christoph Lameter894b8782007-05-10 03:15:16 -07001710
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001711 object = __slab_alloc(s, gfpflags, node, addr, c);
Christoph Lameter894b8782007-05-10 03:15:16 -07001712
1713 else {
Christoph Lameterff120592009-12-18 16:26:22 -06001714 c->freelist = get_freepointer(s, object);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001715 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001716 }
1717 local_irq_restore(flags);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001718
Pekka Enberg74e21342009-11-25 20:14:48 +02001719 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameterff120592009-12-18 16:26:22 -06001720 memset(object, 0, s->objsize);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07001721
Christoph Lameterff120592009-12-18 16:26:22 -06001722 kmemcheck_slab_alloc(s, gfpflags, object, s->objsize);
1723 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, gfpflags);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001724
Christoph Lameter894b8782007-05-10 03:15:16 -07001725 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001726}
1727
1728void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1729{
Christoph Lameter2154a332010-07-09 14:07:10 -05001730 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001731
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001732 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001733
1734 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001735}
1736EXPORT_SYMBOL(kmem_cache_alloc);
1737
Li Zefan0f24f122009-12-11 15:45:30 +08001738#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001739void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1740{
Christoph Lameter2154a332010-07-09 14:07:10 -05001741 return slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001742}
1743EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1744#endif
1745
Christoph Lameter81819f02007-05-06 14:49:36 -07001746#ifdef CONFIG_NUMA
1747void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1748{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001749 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1750
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001751 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1752 s->objsize, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001753
1754 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001755}
1756EXPORT_SYMBOL(kmem_cache_alloc_node);
1757#endif
1758
Li Zefan0f24f122009-12-11 15:45:30 +08001759#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001760void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1761 gfp_t gfpflags,
1762 int node)
1763{
1764 return slab_alloc(s, gfpflags, node, _RET_IP_);
1765}
1766EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1767#endif
1768
Christoph Lameter81819f02007-05-06 14:49:36 -07001769/*
Christoph Lameter894b8782007-05-10 03:15:16 -07001770 * Slow patch handling. This may still be called frequently since objects
1771 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07001772 *
Christoph Lameter894b8782007-05-10 03:15:16 -07001773 * So we still attempt to reduce cache line usage. Just take the slab
1774 * lock and free the item. If there is no additional partial page
1775 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07001776 */
Christoph Lameter894b8782007-05-10 03:15:16 -07001777static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06001778 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001779{
1780 void *prior;
1781 void **object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07001782
Christoph Lameter84e554e2009-12-18 16:26:23 -06001783 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07001784 slab_lock(page);
1785
Christoph Lameteraf537b02010-07-09 14:07:14 -05001786 if (kmem_cache_debug(s))
Christoph Lameter81819f02007-05-06 14:49:36 -07001787 goto debug;
Christoph Lameter6446faa2008-02-15 23:45:26 -08001788
Christoph Lameter81819f02007-05-06 14:49:36 -07001789checks_ok:
Christoph Lameterff120592009-12-18 16:26:22 -06001790 prior = page->freelist;
1791 set_freepointer(s, object, prior);
Christoph Lameter81819f02007-05-06 14:49:36 -07001792 page->freelist = object;
1793 page->inuse--;
1794
Andy Whitcroft8a380822008-07-23 21:27:18 -07001795 if (unlikely(PageSlubFrozen(page))) {
Christoph Lameter84e554e2009-12-18 16:26:23 -06001796 stat(s, FREE_FROZEN);
Christoph Lameter81819f02007-05-06 14:49:36 -07001797 goto out_unlock;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001798 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001799
1800 if (unlikely(!page->inuse))
1801 goto slab_empty;
1802
1803 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08001804 * Objects left in the slab. If it was not on the partial list before
Christoph Lameter81819f02007-05-06 14:49:36 -07001805 * then add it.
1806 */
Christoph Lametera973e9d2008-03-01 13:40:44 -08001807 if (unlikely(!prior)) {
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001808 add_partial(get_node(s, page_to_nid(page)), page, 1);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001809 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001810 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001811
1812out_unlock:
1813 slab_unlock(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001814 return;
1815
1816slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08001817 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001818 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001819 * Slab still on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07001820 */
1821 remove_partial(s, page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001822 stat(s, FREE_REMOVE_PARTIAL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001823 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001824 slab_unlock(page);
Christoph Lameter84e554e2009-12-18 16:26:23 -06001825 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07001826 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001827 return;
1828
1829debug:
Christoph Lameter3ec09742007-05-16 22:11:00 -07001830 if (!free_debug_processing(s, page, x, addr))
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001831 goto out_unlock;
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001832 goto checks_ok;
Christoph Lameter81819f02007-05-06 14:49:36 -07001833}
1834
Christoph Lameter894b8782007-05-10 03:15:16 -07001835/*
1836 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1837 * can perform fastpath freeing without additional function calls.
1838 *
1839 * The fastpath is only possible if we are freeing to the current cpu slab
1840 * of this processor. This typically the case if we have just allocated
1841 * the item before.
1842 *
1843 * If fastpath is not possible then fall back to __slab_free where we deal
1844 * with all sorts of special processing.
1845 */
Pekka Enberg06428782008-01-07 23:20:27 -08001846static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001847 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07001848{
1849 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001850 struct kmem_cache_cpu *c;
Christoph Lameter1f842602008-01-07 23:20:30 -08001851 unsigned long flags;
1852
Catalin Marinas06f22f12009-06-11 13:23:18 +01001853 kmemleak_free_recursive(x, s->flags);
Christoph Lameter894b8782007-05-10 03:15:16 -07001854 local_irq_save(flags);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06001855 c = __this_cpu_ptr(s->cpu_slab);
Christoph Lameterff120592009-12-18 16:26:22 -06001856 kmemcheck_slab_free(s, object, s->objsize);
1857 debug_check_no_locks_freed(object, s->objsize);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07001858 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameterff120592009-12-18 16:26:22 -06001859 debug_check_no_obj_freed(object, s->objsize);
Christoph Lameteree3c72a2007-10-16 01:26:07 -07001860 if (likely(page == c->page && c->node >= 0)) {
Christoph Lameterff120592009-12-18 16:26:22 -06001861 set_freepointer(s, object, c->freelist);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07001862 c->freelist = object;
Christoph Lameter84e554e2009-12-18 16:26:23 -06001863 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07001864 } else
Christoph Lameterff120592009-12-18 16:26:22 -06001865 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07001866
1867 local_irq_restore(flags);
1868}
1869
Christoph Lameter81819f02007-05-06 14:49:36 -07001870void kmem_cache_free(struct kmem_cache *s, void *x)
1871{
Christoph Lameter77c5e2d2007-05-06 14:49:42 -07001872 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001873
Christoph Lameterb49af682007-05-06 14:49:41 -07001874 page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001875
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001876 slab_free(s, page, x, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03001877
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02001878 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001879}
1880EXPORT_SYMBOL(kmem_cache_free);
1881
Cyrill Gorcunove9beef12008-10-28 22:02:26 +03001882/* Figure out on which slab page the object resides */
Christoph Lameter81819f02007-05-06 14:49:36 -07001883static struct page *get_object_page(const void *x)
1884{
Christoph Lameterb49af682007-05-06 14:49:41 -07001885 struct page *page = virt_to_head_page(x);
Christoph Lameter81819f02007-05-06 14:49:36 -07001886
1887 if (!PageSlab(page))
1888 return NULL;
1889
1890 return page;
1891}
1892
1893/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001894 * Object placement in a slab is made very easy because we always start at
1895 * offset 0. If we tune the size of the object to the alignment then we can
1896 * get the required alignment by putting one properly sized object after
1897 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07001898 *
1899 * Notice that the allocation order determines the sizes of the per cpu
1900 * caches. Each processor has always one slab available for allocations.
1901 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07001902 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07001903 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07001904 */
1905
1906/*
1907 * Mininum / Maximum order of slab pages. This influences locking overhead
1908 * and slab fragmentation. A higher order reduces the number of partial slabs
1909 * and increases the number of allocations possible without having to
1910 * take the list_lock.
1911 */
1912static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03001913static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001914static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001915
1916/*
1917 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07001918 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07001919 */
1920static int slub_nomerge;
1921
1922/*
Christoph Lameter81819f02007-05-06 14:49:36 -07001923 * Calculate the order of allocation given an slab object size.
1924 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001925 * The order of allocation has significant impact on performance and other
1926 * system components. Generally order 0 allocations should be preferred since
1927 * order 0 does not cause fragmentation in the page allocator. Larger objects
1928 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001929 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001930 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07001931 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001932 * In order to reach satisfactory performance we must ensure that a minimum
1933 * number of objects is in one slab. Otherwise we may generate too much
1934 * activity on the partial lists which requires taking the list_lock. This is
1935 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07001936 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001937 * slub_max_order specifies the order where we begin to stop considering the
1938 * number of objects in a slab as critical. If we reach slub_max_order then
1939 * we try to keep the page order as low as possible. So we accept more waste
1940 * of space in favor of a small page order.
1941 *
1942 * Higher order allocations also allow the placement of more objects in a
1943 * slab and thereby reduce object handling overhead. If the user has
1944 * requested a higher mininum order then we start with that one instead of
1945 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07001946 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001947static inline int slab_order(int size, int min_objects,
1948 int max_order, int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001949{
1950 int order;
1951 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07001952 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001953
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001954 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1955 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03001956
Christoph Lameter6300ea72007-07-17 04:03:20 -07001957 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001958 fls(min_objects * size - 1) - PAGE_SHIFT);
1959 order <= max_order; order++) {
1960
Christoph Lameter81819f02007-05-06 14:49:36 -07001961 unsigned long slab_size = PAGE_SIZE << order;
1962
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001963 if (slab_size < min_objects * size)
Christoph Lameter81819f02007-05-06 14:49:36 -07001964 continue;
1965
Christoph Lameter81819f02007-05-06 14:49:36 -07001966 rem = slab_size % size;
1967
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001968 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07001969 break;
1970
1971 }
Christoph Lameter672bba32007-05-09 02:32:39 -07001972
Christoph Lameter81819f02007-05-06 14:49:36 -07001973 return order;
1974}
1975
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001976static inline int calculate_order(int size)
1977{
1978 int order;
1979 int min_objects;
1980 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001981 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001982
1983 /*
1984 * Attempt to find best configuration for a slab. This
1985 * works by first attempting to generate a layout with
1986 * the best configuration and backing off gradually.
1987 *
1988 * First we reduce the acceptable waste in a slab. Then
1989 * we reduce the minimum objects required in a slab.
1990 */
1991 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03001992 if (!min_objects)
1993 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02001994 max_objects = (PAGE_SIZE << slub_max_order)/size;
1995 min_objects = min(min_objects, max_objects);
1996
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001997 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03001998 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07001999 while (fraction >= 4) {
2000 order = slab_order(size, min_objects,
2001 slub_max_order, fraction);
2002 if (order <= slub_max_order)
2003 return order;
2004 fraction /= 2;
2005 }
Amerigo Wang5086c382009-08-19 21:44:13 +03002006 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002007 }
2008
2009 /*
2010 * We were unable to place multiple objects in a slab. Now
2011 * lets see if we can place a single object there.
2012 */
2013 order = slab_order(size, 1, slub_max_order, 1);
2014 if (order <= slub_max_order)
2015 return order;
2016
2017 /*
2018 * Doh this slab cannot be placed using slub_max_order.
2019 */
2020 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03002021 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002022 return order;
2023 return -ENOSYS;
2024}
2025
Christoph Lameter81819f02007-05-06 14:49:36 -07002026/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002027 * Figure out what the alignment of the objects will be.
Christoph Lameter81819f02007-05-06 14:49:36 -07002028 */
2029static unsigned long calculate_alignment(unsigned long flags,
2030 unsigned long align, unsigned long size)
2031{
2032 /*
Christoph Lameter6446faa2008-02-15 23:45:26 -08002033 * If the user wants hardware cache aligned objects then follow that
2034 * suggestion if the object is sufficiently large.
Christoph Lameter81819f02007-05-06 14:49:36 -07002035 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08002036 * The hardware cache alignment cannot override the specified
2037 * alignment though. If that is greater then use it.
Christoph Lameter81819f02007-05-06 14:49:36 -07002038 */
Nick Pigginb6210382008-03-05 14:05:56 -08002039 if (flags & SLAB_HWCACHE_ALIGN) {
2040 unsigned long ralign = cache_line_size();
2041 while (size <= ralign / 2)
2042 ralign /= 2;
2043 align = max(align, ralign);
2044 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002045
2046 if (align < ARCH_SLAB_MINALIGN)
Nick Pigginb6210382008-03-05 14:05:56 -08002047 align = ARCH_SLAB_MINALIGN;
Christoph Lameter81819f02007-05-06 14:49:36 -07002048
2049 return ALIGN(align, sizeof(void *));
2050}
2051
Pekka Enberg5595cff2008-08-05 09:28:47 +03002052static void
2053init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002054{
2055 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002056 spin_lock_init(&n->list_lock);
2057 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002058#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002059 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002060 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002061 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002062#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002063}
2064
Christoph Lameter91efd772010-01-21 17:43:35 -06002065static DEFINE_PER_CPU(struct kmem_cache_cpu, kmalloc_percpu[KMALLOC_CACHES]);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002066
2067static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2068{
Christoph Lameter756dee72009-12-18 16:26:21 -06002069 if (s < kmalloc_caches + KMALLOC_CACHES && s >= kmalloc_caches)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002070 /*
2071 * Boot time creation of the kmalloc array. Use static per cpu data
2072 * since the per cpu allocator is not available yet.
2073 */
Stephen Rothwell1154fab2010-03-01 16:04:45 +11002074 s->cpu_slab = kmalloc_percpu + (s - kmalloc_caches);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002075 else
2076 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2077
2078 if (!s->cpu_slab)
2079 return 0;
2080
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002081 return 1;
2082}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002083
Christoph Lameter81819f02007-05-06 14:49:36 -07002084#ifdef CONFIG_NUMA
2085/*
2086 * No kmalloc_node yet so do it by hand. We know that this is the first
2087 * slab on the node for this slabcache. There are no concurrent accesses
2088 * possible.
2089 *
2090 * Note that this function only works on the kmalloc_node_cache
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002091 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2092 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002093 */
David Rientjes0094de92008-11-25 19:14:19 -08002094static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002095{
2096 struct page *page;
2097 struct kmem_cache_node *n;
rootba84c732008-01-07 23:20:28 -08002098 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07002099
2100 BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2101
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002102 page = new_slab(kmalloc_caches, gfpflags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002103
2104 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002105 if (page_to_nid(page) != node) {
2106 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2107 "node %d\n", node);
2108 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2109 "in order to be able to continue\n");
2110 }
2111
Christoph Lameter81819f02007-05-06 14:49:36 -07002112 n = page->freelist;
2113 BUG_ON(!n);
2114 page->freelist = get_freepointer(kmalloc_caches, n);
2115 page->inuse++;
2116 kmalloc_caches->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002117#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterd45f39c2007-07-17 04:03:21 -07002118 init_object(kmalloc_caches, n, 1);
2119 init_tracking(kmalloc_caches, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002120#endif
Pekka Enberg5595cff2008-08-05 09:28:47 +03002121 init_kmem_cache_node(n, kmalloc_caches);
Christoph Lameter205ab992008-04-14 19:11:40 +03002122 inc_slabs_node(kmalloc_caches, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002123
rootba84c732008-01-07 23:20:28 -08002124 /*
2125 * lockdep requires consistent irq usage for each lock
2126 * so even though there cannot be a race this early in
2127 * the boot sequence, we still disable irqs.
2128 */
2129 local_irq_save(flags);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08002130 add_partial(n, page, 0);
rootba84c732008-01-07 23:20:28 -08002131 local_irq_restore(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002132}
2133
2134static void free_kmem_cache_nodes(struct kmem_cache *s)
2135{
2136 int node;
2137
Christoph Lameterf64dc582007-10-16 01:25:33 -07002138 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002139 struct kmem_cache_node *n = s->node[node];
Alexander Duyck73367bd2010-05-21 14:41:35 -07002140 if (n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002141 kmem_cache_free(kmalloc_caches, n);
2142 s->node[node] = NULL;
2143 }
2144}
2145
2146static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2147{
2148 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002149
Christoph Lameterf64dc582007-10-16 01:25:33 -07002150 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002151 struct kmem_cache_node *n;
2152
Alexander Duyck73367bd2010-05-21 14:41:35 -07002153 if (slab_state == DOWN) {
2154 early_kmem_cache_node_alloc(gfpflags, node);
2155 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002156 }
Alexander Duyck73367bd2010-05-21 14:41:35 -07002157 n = kmem_cache_alloc_node(kmalloc_caches,
2158 gfpflags, node);
2159
2160 if (!n) {
2161 free_kmem_cache_nodes(s);
2162 return 0;
2163 }
2164
Christoph Lameter81819f02007-05-06 14:49:36 -07002165 s->node[node] = n;
Pekka Enberg5595cff2008-08-05 09:28:47 +03002166 init_kmem_cache_node(n, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002167 }
2168 return 1;
2169}
2170#else
2171static void free_kmem_cache_nodes(struct kmem_cache *s)
2172{
2173}
2174
2175static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2176{
Pekka Enberg5595cff2008-08-05 09:28:47 +03002177 init_kmem_cache_node(&s->local_node, s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002178 return 1;
2179}
2180#endif
2181
David Rientjesc0bdb232009-02-25 09:16:35 +02002182static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002183{
2184 if (min < MIN_PARTIAL)
2185 min = MIN_PARTIAL;
2186 else if (min > MAX_PARTIAL)
2187 min = MAX_PARTIAL;
2188 s->min_partial = min;
2189}
2190
Christoph Lameter81819f02007-05-06 14:49:36 -07002191/*
2192 * calculate_sizes() determines the order and the distribution of data within
2193 * a slab object.
2194 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002195static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002196{
2197 unsigned long flags = s->flags;
2198 unsigned long size = s->objsize;
2199 unsigned long align = s->align;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002200 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002201
2202 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002203 * Round up object size to the next word boundary. We can only
2204 * place the free pointer at word boundaries and this determines
2205 * the possible location of the free pointer.
2206 */
2207 size = ALIGN(size, sizeof(void *));
2208
2209#ifdef CONFIG_SLUB_DEBUG
2210 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002211 * Determine if we can poison the object itself. If the user of
2212 * the slab may touch the object after free or before allocation
2213 * then we should never poison the object itself.
2214 */
2215 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002216 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07002217 s->flags |= __OBJECT_POISON;
2218 else
2219 s->flags &= ~__OBJECT_POISON;
2220
Christoph Lameter81819f02007-05-06 14:49:36 -07002221
2222 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002223 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07002224 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07002225 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07002226 */
2227 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2228 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002229#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002230
2231 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002232 * With that we have determined the number of bytes in actual use
2233 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07002234 */
2235 s->inuse = size;
2236
2237 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def9f2007-05-16 22:10:50 -07002238 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002239 /*
2240 * Relocate free pointer after the object if it is not
2241 * permitted to overwrite the first word of the object on
2242 * kmem_cache_free.
2243 *
2244 * This is the case if we do RCU, have a constructor or
2245 * destructor or are poisoning the objects.
2246 */
2247 s->offset = size;
2248 size += sizeof(void *);
2249 }
2250
Christoph Lameterc12b3c62007-05-23 13:57:31 -07002251#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002252 if (flags & SLAB_STORE_USER)
2253 /*
2254 * Need to store information about allocs and frees after
2255 * the object.
2256 */
2257 size += 2 * sizeof(struct track);
2258
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07002259 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07002260 /*
2261 * Add some empty padding so that we can catch
2262 * overwrites from earlier objects rather than let
2263 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01002264 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07002265 * of the object.
2266 */
2267 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07002268#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07002269
Christoph Lameter81819f02007-05-06 14:49:36 -07002270 /*
2271 * Determine the alignment based on various parameters that the
Christoph Lameter65c02d42007-05-09 02:32:35 -07002272 * user specified and the dynamic determination of cache line size
2273 * on bootup.
Christoph Lameter81819f02007-05-06 14:49:36 -07002274 */
2275 align = calculate_alignment(flags, align, s->objsize);
Zhang, Yanmindcb0ce12009-07-30 11:28:11 +08002276 s->align = align;
Christoph Lameter81819f02007-05-06 14:49:36 -07002277
2278 /*
2279 * SLUB stores one object immediately after another beginning from
2280 * offset 0. In order to align the objects we have to simply size
2281 * each object to conform to the alignment.
2282 */
2283 size = ALIGN(size, align);
2284 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03002285 if (forced_order >= 0)
2286 order = forced_order;
2287 else
2288 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07002289
Christoph Lameter834f3d12008-04-14 19:11:31 +03002290 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07002291 return 0;
2292
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002293 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002294 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08002295 s->allocflags |= __GFP_COMP;
2296
2297 if (s->flags & SLAB_CACHE_DMA)
2298 s->allocflags |= SLUB_DMA;
2299
2300 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2301 s->allocflags |= __GFP_RECLAIMABLE;
2302
Christoph Lameter81819f02007-05-06 14:49:36 -07002303 /*
2304 * Determine the number of objects per slab
2305 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002306 s->oo = oo_make(order, size);
Christoph Lameter65c33762008-04-14 19:11:40 +03002307 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03002308 if (oo_objects(s->oo) > oo_objects(s->max))
2309 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002310
Christoph Lameter834f3d12008-04-14 19:11:31 +03002311 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07002312
2313}
2314
Christoph Lameter81819f02007-05-06 14:49:36 -07002315static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2316 const char *name, size_t size,
2317 size_t align, unsigned long flags,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002318 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07002319{
2320 memset(s, 0, kmem_size);
2321 s->name = name;
2322 s->ctor = ctor;
Christoph Lameter81819f02007-05-06 14:49:36 -07002323 s->objsize = size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002324 s->align = align;
Christoph Lameterba0268a2007-09-11 15:24:11 -07002325 s->flags = kmem_cache_flags(size, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07002326
Christoph Lameter06b285d2008-04-14 19:11:41 +03002327 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07002328 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07002329 if (disable_higher_order_debug) {
2330 /*
2331 * Disable debugging flags that store metadata if the min slab
2332 * order increased.
2333 */
2334 if (get_order(s->size) > get_order(s->objsize)) {
2335 s->flags &= ~DEBUG_METADATA_FLAGS;
2336 s->offset = 0;
2337 if (!calculate_sizes(s, -1))
2338 goto error;
2339 }
2340 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002341
David Rientjes3b89d7d2009-02-22 17:40:07 -08002342 /*
2343 * The larger the object size is, the more pages we want on the partial
2344 * list to avoid pounding the page allocator excessively.
2345 */
David Rientjesc0bdb232009-02-25 09:16:35 +02002346 set_min_partial(s, ilog2(s->size));
Christoph Lameter81819f02007-05-06 14:49:36 -07002347 s->refcount = 1;
2348#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05002349 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07002350#endif
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002351 if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2352 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07002353
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002354 if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
Christoph Lameter81819f02007-05-06 14:49:36 -07002355 return 1;
Christoph Lameterff120592009-12-18 16:26:22 -06002356
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002357 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07002358error:
2359 if (flags & SLAB_PANIC)
2360 panic("Cannot create slab %s size=%lu realsize=%u "
2361 "order=%u offset=%u flags=%lx\n",
Christoph Lameter834f3d12008-04-14 19:11:31 +03002362 s->name, (unsigned long)size, s->size, oo_order(s->oo),
Christoph Lameter81819f02007-05-06 14:49:36 -07002363 s->offset, flags);
2364 return 0;
2365}
Christoph Lameter81819f02007-05-06 14:49:36 -07002366
2367/*
2368 * Check if a given pointer is valid
2369 */
2370int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2371{
Pekka Enberg06428782008-01-07 23:20:27 -08002372 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002373
Pekka Enbergd3e06e22010-04-07 19:23:41 +03002374 if (!kern_ptr_validate(object, s->size))
2375 return 0;
2376
Christoph Lameter81819f02007-05-06 14:49:36 -07002377 page = get_object_page(object);
2378
2379 if (!page || s != page->slab)
2380 /* No slab or wrong slab */
2381 return 0;
2382
Christoph Lameterabcd08a2007-05-09 02:32:37 -07002383 if (!check_valid_pointer(s, page, object))
Christoph Lameter81819f02007-05-06 14:49:36 -07002384 return 0;
2385
2386 /*
2387 * We could also check if the object is on the slabs freelist.
2388 * But this would be too expensive and it seems that the main
Christoph Lameter6446faa2008-02-15 23:45:26 -08002389 * purpose of kmem_ptr_valid() is to check if the object belongs
Christoph Lameter81819f02007-05-06 14:49:36 -07002390 * to a certain slab.
2391 */
2392 return 1;
2393}
2394EXPORT_SYMBOL(kmem_ptr_validate);
2395
2396/*
2397 * Determine the size of a slab object
2398 */
2399unsigned int kmem_cache_size(struct kmem_cache *s)
2400{
2401 return s->objsize;
2402}
2403EXPORT_SYMBOL(kmem_cache_size);
2404
2405const char *kmem_cache_name(struct kmem_cache *s)
2406{
2407 return s->name;
2408}
2409EXPORT_SYMBOL(kmem_cache_name);
2410
Christoph Lameter33b12c32008-04-25 12:22:43 -07002411static void list_slab_objects(struct kmem_cache *s, struct page *page,
2412 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07002413{
Christoph Lameter33b12c32008-04-25 12:22:43 -07002414#ifdef CONFIG_SLUB_DEBUG
2415 void *addr = page_address(page);
2416 void *p;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002417 long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long),
2418 GFP_ATOMIC);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002419
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002420 if (!map)
2421 return;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002422 slab_err(s, page, "%s", text);
2423 slab_lock(page);
2424 for_each_free_object(p, s, page->freelist)
2425 set_bit(slab_index(p, s, addr), map);
2426
2427 for_each_object(p, s, addr, page->objects) {
2428
2429 if (!test_bit(slab_index(p, s, addr), map)) {
2430 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2431 p, p - addr);
2432 print_tracking(s, p);
2433 }
2434 }
2435 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01002436 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002437#endif
2438}
2439
Christoph Lameter81819f02007-05-06 14:49:36 -07002440/*
Christoph Lameter599870b2008-04-23 12:36:52 -07002441 * Attempt to free all partial slabs on a node.
Christoph Lameter81819f02007-05-06 14:49:36 -07002442 */
Christoph Lameter599870b2008-04-23 12:36:52 -07002443static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002444{
Christoph Lameter81819f02007-05-06 14:49:36 -07002445 unsigned long flags;
2446 struct page *page, *h;
2447
2448 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter33b12c32008-04-25 12:22:43 -07002449 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002450 if (!page->inuse) {
2451 list_del(&page->lru);
2452 discard_slab(s, page);
Christoph Lameter599870b2008-04-23 12:36:52 -07002453 n->nr_partial--;
Christoph Lameter33b12c32008-04-25 12:22:43 -07002454 } else {
2455 list_slab_objects(s, page,
2456 "Objects remaining on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07002457 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07002458 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002459 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002460}
2461
2462/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002463 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07002464 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002465static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002466{
2467 int node;
2468
2469 flush_all(s);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002470 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07002471 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07002472 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002473 struct kmem_cache_node *n = get_node(s, node);
2474
Christoph Lameter599870b2008-04-23 12:36:52 -07002475 free_partial(s, n);
2476 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07002477 return 1;
2478 }
2479 free_kmem_cache_nodes(s);
2480 return 0;
2481}
2482
2483/*
2484 * Close a cache and release the kmem_cache structure
2485 * (must be used for caches created using kmem_cache_create)
2486 */
2487void kmem_cache_destroy(struct kmem_cache *s)
2488{
2489 down_write(&slub_lock);
2490 s->refcount--;
2491 if (!s->refcount) {
2492 list_del(&s->list);
Pekka Enbergd629d812008-04-23 22:31:08 +03002493 if (kmem_cache_close(s)) {
2494 printk(KERN_ERR "SLUB %s: %s called for cache that "
2495 "still has objects.\n", s->name, __func__);
2496 dump_stack();
2497 }
Eric Dumazetd76b1592009-09-03 22:38:59 +03002498 if (s->flags & SLAB_DESTROY_BY_RCU)
2499 rcu_barrier();
Christoph Lameter81819f02007-05-06 14:49:36 -07002500 sysfs_slab_remove(s);
Christoph Lameter2bce6482010-07-19 11:39:11 -05002501 }
2502 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07002503}
2504EXPORT_SYMBOL(kmem_cache_destroy);
2505
2506/********************************************************************
2507 * Kmalloc subsystem
2508 *******************************************************************/
2509
Christoph Lameter756dee72009-12-18 16:26:21 -06002510struct kmem_cache kmalloc_caches[KMALLOC_CACHES] __cacheline_aligned;
Christoph Lameter81819f02007-05-06 14:49:36 -07002511EXPORT_SYMBOL(kmalloc_caches);
2512
Christoph Lameter81819f02007-05-06 14:49:36 -07002513static int __init setup_slub_min_order(char *str)
2514{
Pekka Enberg06428782008-01-07 23:20:27 -08002515 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07002516
2517 return 1;
2518}
2519
2520__setup("slub_min_order=", setup_slub_min_order);
2521
2522static int __init setup_slub_max_order(char *str)
2523{
Pekka Enberg06428782008-01-07 23:20:27 -08002524 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03002525 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002526
2527 return 1;
2528}
2529
2530__setup("slub_max_order=", setup_slub_max_order);
2531
2532static int __init setup_slub_min_objects(char *str)
2533{
Pekka Enberg06428782008-01-07 23:20:27 -08002534 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07002535
2536 return 1;
2537}
2538
2539__setup("slub_min_objects=", setup_slub_min_objects);
2540
2541static int __init setup_slub_nomerge(char *str)
2542{
2543 slub_nomerge = 1;
2544 return 1;
2545}
2546
2547__setup("slub_nomerge", setup_slub_nomerge);
2548
Christoph Lameter81819f02007-05-06 14:49:36 -07002549static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2550 const char *name, int size, gfp_t gfp_flags)
2551{
2552 unsigned int flags = 0;
2553
2554 if (gfp_flags & SLUB_DMA)
2555 flags = SLAB_CACHE_DMA;
2556
Pekka Enberg83b519e2009-06-10 19:40:04 +03002557 /*
2558 * This function is called with IRQs disabled during early-boot on
2559 * single CPU so there's no need to take slub_lock here.
2560 */
Christoph Lameter81819f02007-05-06 14:49:36 -07002561 if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
Christoph Lameter319d1e22008-04-14 19:11:41 +03002562 flags, NULL))
Christoph Lameter81819f02007-05-06 14:49:36 -07002563 goto panic;
2564
2565 list_add(&s->list, &slab_caches);
Pekka Enberg83b519e2009-06-10 19:40:04 +03002566
Christoph Lameter81819f02007-05-06 14:49:36 -07002567 if (sysfs_slab_add(s))
2568 goto panic;
2569 return s;
2570
2571panic:
2572 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2573}
2574
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002575#ifdef CONFIG_ZONE_DMA
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002576static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
Christoph Lameter1ceef402007-08-07 15:11:48 -07002577
2578static void sysfs_add_func(struct work_struct *w)
2579{
2580 struct kmem_cache *s;
2581
2582 down_write(&slub_lock);
2583 list_for_each_entry(s, &slab_caches, list) {
2584 if (s->flags & __SYSFS_ADD_DEFERRED) {
2585 s->flags &= ~__SYSFS_ADD_DEFERRED;
2586 sysfs_slab_add(s);
2587 }
2588 }
2589 up_write(&slub_lock);
2590}
2591
2592static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2593
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002594static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2595{
2596 struct kmem_cache *s;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002597 char *text;
2598 size_t realsize;
Nick Piggin964cf352009-06-15 13:35:10 +03002599 unsigned long slabflags;
Christoph Lameter756dee72009-12-18 16:26:21 -06002600 int i;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002601
2602 s = kmalloc_caches_dma[index];
2603 if (s)
2604 return s;
2605
2606 /* Dynamically create dma cache */
Christoph Lameter1ceef402007-08-07 15:11:48 -07002607 if (flags & __GFP_WAIT)
2608 down_write(&slub_lock);
2609 else {
2610 if (!down_write_trylock(&slub_lock))
2611 goto out;
2612 }
2613
2614 if (kmalloc_caches_dma[index])
2615 goto unlock_out;
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002616
Christoph Lameter7b55f622007-07-17 04:03:27 -07002617 realsize = kmalloc_caches[index].objsize;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08002618 text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2619 (unsigned int)realsize);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002620
Christoph Lameter756dee72009-12-18 16:26:21 -06002621 s = NULL;
2622 for (i = 0; i < KMALLOC_CACHES; i++)
2623 if (!kmalloc_caches[i].size)
2624 break;
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002625
Christoph Lameter756dee72009-12-18 16:26:21 -06002626 BUG_ON(i >= KMALLOC_CACHES);
2627 s = kmalloc_caches + i;
Christoph Lameter1ceef402007-08-07 15:11:48 -07002628
Nick Piggin964cf352009-06-15 13:35:10 +03002629 /*
2630 * Must defer sysfs creation to a workqueue because we don't know
2631 * what context we are called from. Before sysfs comes up, we don't
2632 * need to do anything because our sysfs initcall will start by
2633 * adding all existing slabs to sysfs.
2634 */
Pekka Enberg5caf5c72009-06-17 08:30:54 +03002635 slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK;
Nick Piggin964cf352009-06-15 13:35:10 +03002636 if (slab_state >= SYSFS)
2637 slabflags |= __SYSFS_ADD_DEFERRED;
2638
David Rientjes7738dd92010-01-15 12:49:56 -08002639 if (!text || !kmem_cache_open(s, flags, text,
Nick Piggin964cf352009-06-15 13:35:10 +03002640 realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) {
Christoph Lameter756dee72009-12-18 16:26:21 -06002641 s->size = 0;
Christoph Lameter1ceef402007-08-07 15:11:48 -07002642 kfree(text);
2643 goto unlock_out;
Christoph Lameterdfce8642007-07-17 04:03:25 -07002644 }
Christoph Lameter1ceef402007-08-07 15:11:48 -07002645
2646 list_add(&s->list, &slab_caches);
2647 kmalloc_caches_dma[index] = s;
2648
Nick Piggin964cf352009-06-15 13:35:10 +03002649 if (slab_state >= SYSFS)
2650 schedule_work(&sysfs_add_work);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002651
2652unlock_out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002653 up_write(&slub_lock);
Christoph Lameter1ceef402007-08-07 15:11:48 -07002654out:
Christoph Lameterdfce8642007-07-17 04:03:25 -07002655 return kmalloc_caches_dma[index];
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002656}
2657#endif
2658
Christoph Lameterf1b26332007-07-17 04:03:26 -07002659/*
2660 * Conversion table for small slabs sizes / 8 to the index in the
2661 * kmalloc array. This is necessary for slabs < 192 since we have non power
2662 * of two cache sizes there. The size of larger slabs can be determined using
2663 * fls.
2664 */
2665static s8 size_index[24] = {
2666 3, /* 8 */
2667 4, /* 16 */
2668 5, /* 24 */
2669 5, /* 32 */
2670 6, /* 40 */
2671 6, /* 48 */
2672 6, /* 56 */
2673 6, /* 64 */
2674 1, /* 72 */
2675 1, /* 80 */
2676 1, /* 88 */
2677 1, /* 96 */
2678 7, /* 104 */
2679 7, /* 112 */
2680 7, /* 120 */
2681 7, /* 128 */
2682 2, /* 136 */
2683 2, /* 144 */
2684 2, /* 152 */
2685 2, /* 160 */
2686 2, /* 168 */
2687 2, /* 176 */
2688 2, /* 184 */
2689 2 /* 192 */
2690};
2691
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002692static inline int size_index_elem(size_t bytes)
2693{
2694 return (bytes - 1) / 8;
2695}
2696
Christoph Lameter81819f02007-05-06 14:49:36 -07002697static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2698{
Christoph Lameterf1b26332007-07-17 04:03:26 -07002699 int index;
Christoph Lameter81819f02007-05-06 14:49:36 -07002700
Christoph Lameterf1b26332007-07-17 04:03:26 -07002701 if (size <= 192) {
2702 if (!size)
2703 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -07002704
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03002705 index = size_index[size_index_elem(size)];
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002706 } else
Christoph Lameterf1b26332007-07-17 04:03:26 -07002707 index = fls(size - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002708
2709#ifdef CONFIG_ZONE_DMA
Christoph Lameterf1b26332007-07-17 04:03:26 -07002710 if (unlikely((flags & SLUB_DMA)))
Christoph Lameter2e443fd2007-07-17 04:03:24 -07002711 return dma_kmalloc_cache(index, flags);
Christoph Lameterf1b26332007-07-17 04:03:26 -07002712
Christoph Lameter81819f02007-05-06 14:49:36 -07002713#endif
2714 return &kmalloc_caches[index];
2715}
2716
2717void *__kmalloc(size_t size, gfp_t flags)
2718{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002719 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002720 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002721
Christoph Lameterffadd4d2009-02-17 12:05:07 -05002722 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02002723 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002724
2725 s = get_slab(size, flags);
2726
2727 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002728 return s;
2729
Christoph Lameter2154a332010-07-09 14:07:10 -05002730 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002731
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002732 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002733
2734 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002735}
2736EXPORT_SYMBOL(__kmalloc);
2737
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002738static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2739{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002740 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002741 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002742
Vegard Nossumb1eeab62008-11-25 16:55:53 +01002743 flags |= __GFP_COMP | __GFP_NOTRACK;
2744 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002745 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002746 ptr = page_address(page);
2747
2748 kmemleak_alloc(ptr, size, 1, flags);
2749 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08002750}
2751
Christoph Lameter81819f02007-05-06 14:49:36 -07002752#ifdef CONFIG_NUMA
2753void *__kmalloc_node(size_t size, gfp_t flags, int node)
2754{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002755 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002756 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002757
Ingo Molnar057685c2009-02-20 12:15:30 +01002758 if (unlikely(size > SLUB_MAX_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002759 ret = kmalloc_large_node(size, flags, node);
2760
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002761 trace_kmalloc_node(_RET_IP_, ret,
2762 size, PAGE_SIZE << get_order(size),
2763 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002764
2765 return ret;
2766 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002767
2768 s = get_slab(size, flags);
2769
2770 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07002771 return s;
2772
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002773 ret = slab_alloc(s, flags, node, _RET_IP_);
2774
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02002775 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002776
2777 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002778}
2779EXPORT_SYMBOL(__kmalloc_node);
2780#endif
2781
2782size_t ksize(const void *object)
2783{
Christoph Lameter272c1d22007-06-08 13:46:49 -07002784 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002785 struct kmem_cache *s;
2786
Christoph Lameteref8b4522007-10-16 01:24:46 -07002787 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07002788 return 0;
2789
Vegard Nossum294a80a2007-12-04 23:45:30 -08002790 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08002791
Pekka Enberg76994412008-05-22 19:22:25 +03002792 if (unlikely(!PageSlab(page))) {
2793 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08002794 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03002795 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002796 s = page->slab;
Christoph Lameter81819f02007-05-06 14:49:36 -07002797
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002798#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07002799 /*
2800 * Debugging requires use of the padding between object
2801 * and whatever may come after it.
2802 */
2803 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2804 return s->objsize;
2805
Christoph Lameterae20bfd2008-02-15 23:45:25 -08002806#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002807 /*
2808 * If we have the need to store the freelist pointer
2809 * back there or track user information then we can
2810 * only use the space before that information.
2811 */
2812 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2813 return s->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07002814 /*
2815 * Else we can use all the padding etc for the allocation
2816 */
2817 return s->size;
2818}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02002819EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07002820
2821void kfree(const void *x)
2822{
Christoph Lameter81819f02007-05-06 14:49:36 -07002823 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08002824 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07002825
Pekka Enberg2121db72009-03-25 11:05:57 +02002826 trace_kfree(_RET_IP_, x);
2827
Satyam Sharma2408c552007-10-16 01:24:44 -07002828 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002829 return;
2830
Christoph Lameterb49af682007-05-06 14:49:41 -07002831 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002832 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07002833 BUG_ON(!PageCompound(page));
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01002834 kmemleak_free(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07002835 put_page(page);
2836 return;
2837 }
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002838 slab_free(page->slab, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07002839}
2840EXPORT_SYMBOL(kfree);
2841
Christoph Lameter2086d262007-05-06 14:49:46 -07002842/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002843 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2844 * the remaining slabs by the number of items in use. The slabs with the
2845 * most items in use come first. New allocations will then fill those up
2846 * and thus they can be removed from the partial lists.
2847 *
2848 * The slabs with the least items are placed last. This results in them
2849 * being allocated from last increasing the chance that the last objects
2850 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07002851 */
2852int kmem_cache_shrink(struct kmem_cache *s)
2853{
2854 int node;
2855 int i;
2856 struct kmem_cache_node *n;
2857 struct page *page;
2858 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03002859 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07002860 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03002861 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07002862 unsigned long flags;
2863
2864 if (!slabs_by_inuse)
2865 return -ENOMEM;
2866
2867 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07002868 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07002869 n = get_node(s, node);
2870
2871 if (!n->nr_partial)
2872 continue;
2873
Christoph Lameter834f3d12008-04-14 19:11:31 +03002874 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07002875 INIT_LIST_HEAD(slabs_by_inuse + i);
2876
2877 spin_lock_irqsave(&n->list_lock, flags);
2878
2879 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002880 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07002881 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002882 * Note that concurrent frees may occur while we hold the
2883 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07002884 */
2885 list_for_each_entry_safe(page, t, &n->partial, lru) {
2886 if (!page->inuse && slab_trylock(page)) {
2887 /*
2888 * Must hold slab lock here because slab_free
2889 * may have freed the last object and be
2890 * waiting to release the slab.
2891 */
2892 list_del(&page->lru);
2893 n->nr_partial--;
2894 slab_unlock(page);
2895 discard_slab(s, page);
2896 } else {
Christoph Lameterfcda3d82007-07-30 13:06:46 -07002897 list_move(&page->lru,
2898 slabs_by_inuse + page->inuse);
Christoph Lameter2086d262007-05-06 14:49:46 -07002899 }
2900 }
2901
Christoph Lameter2086d262007-05-06 14:49:46 -07002902 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002903 * Rebuild the partial list with the slabs filled up most
2904 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07002905 */
Christoph Lameter834f3d12008-04-14 19:11:31 +03002906 for (i = objects - 1; i >= 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07002907 list_splice(slabs_by_inuse + i, n->partial.prev);
2908
Christoph Lameter2086d262007-05-06 14:49:46 -07002909 spin_unlock_irqrestore(&n->list_lock, flags);
2910 }
2911
2912 kfree(slabs_by_inuse);
2913 return 0;
2914}
2915EXPORT_SYMBOL(kmem_cache_shrink);
2916
Yasunori Gotob9049e22007-10-21 16:41:37 -07002917#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2918static int slab_mem_going_offline_callback(void *arg)
2919{
2920 struct kmem_cache *s;
2921
2922 down_read(&slub_lock);
2923 list_for_each_entry(s, &slab_caches, list)
2924 kmem_cache_shrink(s);
2925 up_read(&slub_lock);
2926
2927 return 0;
2928}
2929
2930static void slab_mem_offline_callback(void *arg)
2931{
2932 struct kmem_cache_node *n;
2933 struct kmem_cache *s;
2934 struct memory_notify *marg = arg;
2935 int offline_node;
2936
2937 offline_node = marg->status_change_nid;
2938
2939 /*
2940 * If the node still has available memory. we need kmem_cache_node
2941 * for it yet.
2942 */
2943 if (offline_node < 0)
2944 return;
2945
2946 down_read(&slub_lock);
2947 list_for_each_entry(s, &slab_caches, list) {
2948 n = get_node(s, offline_node);
2949 if (n) {
2950 /*
2951 * if n->nr_slabs > 0, slabs still exist on the node
2952 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05002953 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07002954 * callback. So, we must fail.
2955 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002956 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07002957
2958 s->node[offline_node] = NULL;
2959 kmem_cache_free(kmalloc_caches, n);
2960 }
2961 }
2962 up_read(&slub_lock);
2963}
2964
2965static int slab_mem_going_online_callback(void *arg)
2966{
2967 struct kmem_cache_node *n;
2968 struct kmem_cache *s;
2969 struct memory_notify *marg = arg;
2970 int nid = marg->status_change_nid;
2971 int ret = 0;
2972
2973 /*
2974 * If the node's memory is already available, then kmem_cache_node is
2975 * already created. Nothing to do.
2976 */
2977 if (nid < 0)
2978 return 0;
2979
2980 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07002981 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07002982 * allocate a kmem_cache_node structure in order to bring the node
2983 * online.
2984 */
2985 down_read(&slub_lock);
2986 list_for_each_entry(s, &slab_caches, list) {
2987 /*
2988 * XXX: kmem_cache_alloc_node will fallback to other nodes
2989 * since memory is not yet available from the node that
2990 * is brought up.
2991 */
2992 n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
2993 if (!n) {
2994 ret = -ENOMEM;
2995 goto out;
2996 }
Pekka Enberg5595cff2008-08-05 09:28:47 +03002997 init_kmem_cache_node(n, s);
Yasunori Gotob9049e22007-10-21 16:41:37 -07002998 s->node[nid] = n;
2999 }
3000out:
3001 up_read(&slub_lock);
3002 return ret;
3003}
3004
3005static int slab_memory_callback(struct notifier_block *self,
3006 unsigned long action, void *arg)
3007{
3008 int ret = 0;
3009
3010 switch (action) {
3011 case MEM_GOING_ONLINE:
3012 ret = slab_mem_going_online_callback(arg);
3013 break;
3014 case MEM_GOING_OFFLINE:
3015 ret = slab_mem_going_offline_callback(arg);
3016 break;
3017 case MEM_OFFLINE:
3018 case MEM_CANCEL_ONLINE:
3019 slab_mem_offline_callback(arg);
3020 break;
3021 case MEM_ONLINE:
3022 case MEM_CANCEL_OFFLINE:
3023 break;
3024 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003025 if (ret)
3026 ret = notifier_from_errno(ret);
3027 else
3028 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003029 return ret;
3030}
3031
3032#endif /* CONFIG_MEMORY_HOTPLUG */
3033
Christoph Lameter81819f02007-05-06 14:49:36 -07003034/********************************************************************
3035 * Basic setup of slabs
3036 *******************************************************************/
3037
3038void __init kmem_cache_init(void)
3039{
3040 int i;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003041 int caches = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003042
3043#ifdef CONFIG_NUMA
3044 /*
3045 * Must first have the slab cache available for the allocations of the
Christoph Lameter672bba32007-05-09 02:32:39 -07003046 * struct kmem_cache_node's. There is special bootstrap code in
Christoph Lameter81819f02007-05-06 14:49:36 -07003047 * kmem_cache_open for slab_state == DOWN.
3048 */
3049 create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
Pekka Enberg83b519e2009-06-10 19:40:04 +03003050 sizeof(struct kmem_cache_node), GFP_NOWAIT);
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003051 kmalloc_caches[0].refcount = -1;
Christoph Lameter4b356be2007-06-16 10:16:13 -07003052 caches++;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003053
Nadia Derbey0c40ba42008-04-29 01:00:41 -07003054 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
Christoph Lameter81819f02007-05-06 14:49:36 -07003055#endif
3056
3057 /* Able to allocate the per node structures */
3058 slab_state = PARTIAL;
3059
3060 /* Caches that are not of the two-to-the-power-of size */
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003061 if (KMALLOC_MIN_SIZE <= 32) {
Christoph Lameter4b356be2007-06-16 10:16:13 -07003062 create_kmalloc_cache(&kmalloc_caches[1],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003063 "kmalloc-96", 96, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003064 caches++;
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003065 }
3066 if (KMALLOC_MIN_SIZE <= 64) {
Christoph Lameter4b356be2007-06-16 10:16:13 -07003067 create_kmalloc_cache(&kmalloc_caches[2],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003068 "kmalloc-192", 192, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003069 caches++;
3070 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003071
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003072 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003073 create_kmalloc_cache(&kmalloc_caches[i],
Pekka Enberg83b519e2009-06-10 19:40:04 +03003074 "kmalloc", 1 << i, GFP_NOWAIT);
Christoph Lameter4b356be2007-06-16 10:16:13 -07003075 caches++;
3076 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003077
Christoph Lameterf1b26332007-07-17 04:03:26 -07003078
3079 /*
3080 * Patch up the size_index table if we have strange large alignment
3081 * requirements for the kmalloc array. This is only the case for
Christoph Lameter6446faa2008-02-15 23:45:26 -08003082 * MIPS it seems. The standard arches will not generate any code here.
Christoph Lameterf1b26332007-07-17 04:03:26 -07003083 *
3084 * Largest permitted alignment is 256 bytes due to the way we
3085 * handle the index determination for the smaller caches.
3086 *
3087 * Make sure that nothing crazy happens if someone starts tinkering
3088 * around with ARCH_KMALLOC_MINALIGN
3089 */
3090 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3091 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3092
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003093 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3094 int elem = size_index_elem(i);
3095 if (elem >= ARRAY_SIZE(size_index))
3096 break;
3097 size_index[elem] = KMALLOC_SHIFT_LOW;
3098 }
Christoph Lameterf1b26332007-07-17 04:03:26 -07003099
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003100 if (KMALLOC_MIN_SIZE == 64) {
3101 /*
3102 * The 96 byte size cache is not used if the alignment
3103 * is 64 byte.
3104 */
3105 for (i = 64 + 8; i <= 96; i += 8)
3106 size_index[size_index_elem(i)] = 7;
3107 } else if (KMALLOC_MIN_SIZE == 128) {
Christoph Lameter41d54d32008-07-03 09:14:26 -05003108 /*
3109 * The 192 byte sized cache is not used if the alignment
3110 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3111 * instead.
3112 */
3113 for (i = 128 + 8; i <= 192; i += 8)
Aaro Koskinenacdfcd02009-08-28 14:28:54 +03003114 size_index[size_index_elem(i)] = 8;
Christoph Lameter41d54d32008-07-03 09:14:26 -05003115 }
3116
Christoph Lameter81819f02007-05-06 14:49:36 -07003117 slab_state = UP;
3118
3119 /* Provide the correct kmalloc names now that the caches are up */
Christoph Lameterd7278bd2010-07-09 14:07:12 -05003120 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3121 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3122
3123 BUG_ON(!s);
3124 kmalloc_caches[i].name = s;
3125 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003126
3127#ifdef CONFIG_SMP
3128 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003129#endif
3130#ifdef CONFIG_NUMA
3131 kmem_size = offsetof(struct kmem_cache, node) +
3132 nr_node_ids * sizeof(struct kmem_cache_node *);
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003133#else
3134 kmem_size = sizeof(struct kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003135#endif
3136
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003137 printk(KERN_INFO
3138 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
Christoph Lameter4b356be2007-06-16 10:16:13 -07003139 " CPUs=%d, Nodes=%d\n",
3140 caches, cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003141 slub_min_order, slub_max_order, slub_min_objects,
3142 nr_cpu_ids, nr_node_ids);
3143}
3144
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003145void __init kmem_cache_init_late(void)
3146{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003147}
3148
Christoph Lameter81819f02007-05-06 14:49:36 -07003149/*
3150 * Find a mergeable slab cache
3151 */
3152static int slab_unmergeable(struct kmem_cache *s)
3153{
3154 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3155 return 1;
3156
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003157 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003158 return 1;
3159
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003160 /*
3161 * We may have set a slab to be unmergeable during bootstrap.
3162 */
3163 if (s->refcount < 0)
3164 return 1;
3165
Christoph Lameter81819f02007-05-06 14:49:36 -07003166 return 0;
3167}
3168
3169static struct kmem_cache *find_mergeable(size_t size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07003170 size_t align, unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003171 void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003172{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003173 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003174
3175 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3176 return NULL;
3177
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003178 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003179 return NULL;
3180
3181 size = ALIGN(size, sizeof(void *));
3182 align = calculate_alignment(flags, align, size);
3183 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003184 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003185
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003186 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003187 if (slab_unmergeable(s))
3188 continue;
3189
3190 if (size > s->size)
3191 continue;
3192
Christoph Lameterba0268a2007-09-11 15:24:11 -07003193 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Christoph Lameter81819f02007-05-06 14:49:36 -07003194 continue;
3195 /*
3196 * Check if alignment is compatible.
3197 * Courtesy of Adrian Drzewiecki
3198 */
Pekka Enberg06428782008-01-07 23:20:27 -08003199 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003200 continue;
3201
3202 if (s->size - size >= sizeof(void *))
3203 continue;
3204
3205 return s;
3206 }
3207 return NULL;
3208}
3209
3210struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003211 size_t align, unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003212{
3213 struct kmem_cache *s;
3214
Benjamin Herrenschmidtfe1ff492009-09-21 17:02:30 -07003215 if (WARN_ON(!name))
3216 return NULL;
3217
Christoph Lameter81819f02007-05-06 14:49:36 -07003218 down_write(&slub_lock);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003219 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003220 if (s) {
3221 s->refcount++;
3222 /*
3223 * Adjust the object sizes so that we clear
3224 * the complete object on kzalloc.
3225 */
3226 s->objsize = max(s->objsize, (int)size);
3227 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003228
David Rientjes7b8f3b62008-12-17 22:09:46 -08003229 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003230 s->refcount--;
Christoph Lameter81819f02007-05-06 14:49:36 -07003231 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003232 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003233 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003234 return s;
3235 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003236
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003237 s = kmalloc(kmem_size, GFP_KERNEL);
3238 if (s) {
3239 if (kmem_cache_open(s, GFP_KERNEL, name,
Christoph Lameterc59def9f2007-05-16 22:10:50 -07003240 size, align, flags, ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003241 list_add(&s->list, &slab_caches);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003242 if (sysfs_slab_add(s)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003243 list_del(&s->list);
David Rientjes7b8f3b62008-12-17 22:09:46 -08003244 kfree(s);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003245 goto err;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003246 }
Christoph Lameter2bce6482010-07-19 11:39:11 -05003247 up_write(&slub_lock);
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003248 return s;
3249 }
3250 kfree(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003251 }
3252 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003253
3254err:
Christoph Lameter81819f02007-05-06 14:49:36 -07003255 if (flags & SLAB_PANIC)
3256 panic("Cannot create slabcache %s\n", name);
3257 else
3258 s = NULL;
3259 return s;
3260}
3261EXPORT_SYMBOL(kmem_cache_create);
3262
Christoph Lameter81819f02007-05-06 14:49:36 -07003263#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003264/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003265 * Use the cpu notifier to insure that the cpu slabs are flushed when
3266 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003267 */
3268static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3269 unsigned long action, void *hcpu)
3270{
3271 long cpu = (long)hcpu;
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003272 struct kmem_cache *s;
3273 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003274
3275 switch (action) {
3276 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003277 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003278 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003279 case CPU_DEAD_FROZEN:
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07003280 down_read(&slub_lock);
3281 list_for_each_entry(s, &slab_caches, list) {
3282 local_irq_save(flags);
3283 __flush_cpu_slab(s, cpu);
3284 local_irq_restore(flags);
3285 }
3286 up_read(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07003287 break;
3288 default:
3289 break;
3290 }
3291 return NOTIFY_OK;
3292}
3293
Pekka Enberg06428782008-01-07 23:20:27 -08003294static struct notifier_block __cpuinitdata slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003295 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003296};
Christoph Lameter81819f02007-05-06 14:49:36 -07003297
3298#endif
3299
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003300void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003301{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003302 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003303 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003304
Christoph Lameterffadd4d2009-02-17 12:05:07 -05003305 if (unlikely(size > SLUB_MAX_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003306 return kmalloc_large(size, gfpflags);
3307
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003308 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003309
Satyam Sharma2408c552007-10-16 01:24:44 -07003310 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003311 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003312
Christoph Lameter2154a332010-07-09 14:07:10 -05003313 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003314
3315 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003316 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003317
3318 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003319}
3320
3321void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003322 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003323{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003324 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003325 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003326
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003327 if (unlikely(size > SLUB_MAX_SIZE)) {
3328 ret = kmalloc_large_node(size, gfpflags, node);
3329
3330 trace_kmalloc_node(caller, ret,
3331 size, PAGE_SIZE << get_order(size),
3332 gfpflags, node);
3333
3334 return ret;
3335 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003336
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003337 s = get_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003338
Satyam Sharma2408c552007-10-16 01:24:44 -07003339 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003340 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003341
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003342 ret = slab_alloc(s, gfpflags, node, caller);
3343
3344 /* Honor the call site pointer we recieved. */
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +02003345 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003346
3347 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003348}
3349
Christoph Lameterf6acb632008-04-29 16:16:06 -07003350#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03003351static int count_inuse(struct page *page)
3352{
3353 return page->inuse;
3354}
3355
3356static int count_total(struct page *page)
3357{
3358 return page->objects;
3359}
3360
Christoph Lameter434e2452007-07-17 04:03:30 -07003361static int validate_slab(struct kmem_cache *s, struct page *page,
3362 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003363{
3364 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003365 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003366
3367 if (!check_slab(s, page) ||
3368 !on_freelist(s, page, NULL))
3369 return 0;
3370
3371 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003372 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003373
Christoph Lameter7656c722007-05-09 02:32:40 -07003374 for_each_free_object(p, s, page->freelist) {
3375 set_bit(slab_index(p, s, addr), map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003376 if (!check_object(s, page, p, 0))
3377 return 0;
3378 }
3379
Christoph Lameter224a88b2008-04-14 19:11:31 +03003380 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003381 if (!test_bit(slab_index(p, s, addr), map))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003382 if (!check_object(s, page, p, 1))
3383 return 0;
3384 return 1;
3385}
3386
Christoph Lameter434e2452007-07-17 04:03:30 -07003387static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3388 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003389{
3390 if (slab_trylock(page)) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003391 validate_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003392 slab_unlock(page);
3393 } else
3394 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3395 s->name, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003396}
3397
Christoph Lameter434e2452007-07-17 04:03:30 -07003398static int validate_slab_node(struct kmem_cache *s,
3399 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003400{
3401 unsigned long count = 0;
3402 struct page *page;
3403 unsigned long flags;
3404
3405 spin_lock_irqsave(&n->list_lock, flags);
3406
3407 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003408 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003409 count++;
3410 }
3411 if (count != n->nr_partial)
3412 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3413 "counter=%ld\n", s->name, count, n->nr_partial);
3414
3415 if (!(s->flags & SLAB_STORE_USER))
3416 goto out;
3417
3418 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003419 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003420 count++;
3421 }
3422 if (count != atomic_long_read(&n->nr_slabs))
3423 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3424 "counter=%ld\n", s->name, count,
3425 atomic_long_read(&n->nr_slabs));
3426
3427out:
3428 spin_unlock_irqrestore(&n->list_lock, flags);
3429 return count;
3430}
3431
Christoph Lameter434e2452007-07-17 04:03:30 -07003432static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003433{
3434 int node;
3435 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003436 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003437 sizeof(unsigned long), GFP_KERNEL);
3438
3439 if (!map)
3440 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003441
3442 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003443 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003444 struct kmem_cache_node *n = get_node(s, node);
3445
Christoph Lameter434e2452007-07-17 04:03:30 -07003446 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003447 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003448 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003449 return count;
3450}
3451
Christoph Lameterb3459702007-05-09 02:32:41 -07003452#ifdef SLUB_RESILIENCY_TEST
3453static void resiliency_test(void)
3454{
3455 u8 *p;
3456
3457 printk(KERN_ERR "SLUB resiliency testing\n");
3458 printk(KERN_ERR "-----------------------\n");
3459 printk(KERN_ERR "A. Corruption after allocation\n");
3460
3461 p = kzalloc(16, GFP_KERNEL);
3462 p[16] = 0x12;
3463 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3464 " 0x12->0x%p\n\n", p + 16);
3465
3466 validate_slab_cache(kmalloc_caches + 4);
3467
3468 /* Hmmm... The next two are dangerous */
3469 p = kzalloc(32, GFP_KERNEL);
3470 p[32 + sizeof(void *)] = 0x34;
3471 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003472 " 0x34 -> -0x%p\n", p);
3473 printk(KERN_ERR
3474 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003475
3476 validate_slab_cache(kmalloc_caches + 5);
3477 p = kzalloc(64, GFP_KERNEL);
3478 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3479 *p = 0x56;
3480 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3481 p);
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003482 printk(KERN_ERR
3483 "If allocated object is overwritten then not detectable\n\n");
Christoph Lameterb3459702007-05-09 02:32:41 -07003484 validate_slab_cache(kmalloc_caches + 6);
3485
3486 printk(KERN_ERR "\nB. Corruption after free\n");
3487 p = kzalloc(128, GFP_KERNEL);
3488 kfree(p);
3489 *p = 0x78;
3490 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3491 validate_slab_cache(kmalloc_caches + 7);
3492
3493 p = kzalloc(256, GFP_KERNEL);
3494 kfree(p);
3495 p[50] = 0x9a;
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003496 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3497 p);
Christoph Lameterb3459702007-05-09 02:32:41 -07003498 validate_slab_cache(kmalloc_caches + 8);
3499
3500 p = kzalloc(512, GFP_KERNEL);
3501 kfree(p);
3502 p[512] = 0xab;
3503 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3504 validate_slab_cache(kmalloc_caches + 9);
3505}
3506#else
3507static void resiliency_test(void) {};
3508#endif
3509
Christoph Lameter88a420e2007-05-06 14:49:45 -07003510/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003511 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003512 * and freed.
3513 */
3514
3515struct location {
3516 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003517 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003518 long long sum_time;
3519 long min_time;
3520 long max_time;
3521 long min_pid;
3522 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303523 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003524 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003525};
3526
3527struct loc_track {
3528 unsigned long max;
3529 unsigned long count;
3530 struct location *loc;
3531};
3532
3533static void free_loc_track(struct loc_track *t)
3534{
3535 if (t->max)
3536 free_pages((unsigned long)t->loc,
3537 get_order(sizeof(struct location) * t->max));
3538}
3539
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003540static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003541{
3542 struct location *l;
3543 int order;
3544
Christoph Lameter88a420e2007-05-06 14:49:45 -07003545 order = get_order(sizeof(struct location) * max);
3546
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003547 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003548 if (!l)
3549 return 0;
3550
3551 if (t->count) {
3552 memcpy(l, t->loc, sizeof(struct location) * t->count);
3553 free_loc_track(t);
3554 }
3555 t->max = max;
3556 t->loc = l;
3557 return 1;
3558}
3559
3560static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003561 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003562{
3563 long start, end, pos;
3564 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003565 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003566 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003567
3568 start = -1;
3569 end = t->count;
3570
3571 for ( ; ; ) {
3572 pos = start + (end - start + 1) / 2;
3573
3574 /*
3575 * There is nothing at "end". If we end up there
3576 * we need to add something to before end.
3577 */
3578 if (pos == end)
3579 break;
3580
3581 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003582 if (track->addr == caddr) {
3583
3584 l = &t->loc[pos];
3585 l->count++;
3586 if (track->when) {
3587 l->sum_time += age;
3588 if (age < l->min_time)
3589 l->min_time = age;
3590 if (age > l->max_time)
3591 l->max_time = age;
3592
3593 if (track->pid < l->min_pid)
3594 l->min_pid = track->pid;
3595 if (track->pid > l->max_pid)
3596 l->max_pid = track->pid;
3597
Rusty Russell174596a2009-01-01 10:12:29 +10303598 cpumask_set_cpu(track->cpu,
3599 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003600 }
3601 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003602 return 1;
3603 }
3604
Christoph Lameter45edfa52007-05-09 02:32:45 -07003605 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003606 end = pos;
3607 else
3608 start = pos;
3609 }
3610
3611 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003612 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07003613 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003614 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003615 return 0;
3616
3617 l = t->loc + pos;
3618 if (pos < t->count)
3619 memmove(l + 1, l,
3620 (t->count - pos) * sizeof(struct location));
3621 t->count++;
3622 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003623 l->addr = track->addr;
3624 l->sum_time = age;
3625 l->min_time = age;
3626 l->max_time = age;
3627 l->min_pid = track->pid;
3628 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303629 cpumask_clear(to_cpumask(l->cpus));
3630 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003631 nodes_clear(l->nodes);
3632 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003633 return 1;
3634}
3635
3636static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003637 struct page *page, enum track_item alloc,
3638 long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003639{
Christoph Lametera973e9d2008-03-01 13:40:44 -08003640 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003641 void *p;
3642
Christoph Lameter39b26462008-04-14 19:11:30 +03003643 bitmap_zero(map, page->objects);
Christoph Lameter7656c722007-05-09 02:32:40 -07003644 for_each_free_object(p, s, page->freelist)
3645 set_bit(slab_index(p, s, addr), map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003646
Christoph Lameter224a88b2008-04-14 19:11:31 +03003647 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07003648 if (!test_bit(slab_index(p, s, addr), map))
3649 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07003650}
3651
3652static int list_locations(struct kmem_cache *s, char *buf,
3653 enum track_item alloc)
3654{
Harvey Harrisone374d482008-01-31 15:20:50 -08003655 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003656 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003657 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07003658 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003659 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3660 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003661
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003662 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3663 GFP_TEMPORARY)) {
3664 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003665 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003666 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07003667 /* Push back cpu slabs */
3668 flush_all(s);
3669
Christoph Lameterf64dc582007-10-16 01:25:33 -07003670 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07003671 struct kmem_cache_node *n = get_node(s, node);
3672 unsigned long flags;
3673 struct page *page;
3674
Christoph Lameter9e869432007-08-22 14:01:56 -07003675 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07003676 continue;
3677
3678 spin_lock_irqsave(&n->list_lock, flags);
3679 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003680 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003681 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003682 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003683 spin_unlock_irqrestore(&n->list_lock, flags);
3684 }
3685
3686 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07003687 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07003688
Hugh Dickins9c246242008-12-09 13:14:27 -08003689 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003690 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08003691 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003692
3693 if (l->addr)
Harvey Harrisone374d482008-01-31 15:20:50 -08003694 len += sprint_symbol(buf + len, (unsigned long)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003695 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003696 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07003697
3698 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08003699 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07003700 l->min_time,
3701 (long)div_u64(l->sum_time, l->count),
3702 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003703 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08003704 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003705 l->min_time);
3706
3707 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08003708 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003709 l->min_pid, l->max_pid);
3710 else
Harvey Harrisone374d482008-01-31 15:20:50 -08003711 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07003712 l->min_pid);
3713
Rusty Russell174596a2009-01-01 10:12:29 +10303714 if (num_online_cpus() > 1 &&
3715 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003716 len < PAGE_SIZE - 60) {
3717 len += sprintf(buf + len, " cpus=");
3718 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10303719 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07003720 }
3721
Christoph Lameter62bc62a2009-06-16 15:32:15 -07003722 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08003723 len < PAGE_SIZE - 60) {
3724 len += sprintf(buf + len, " nodes=");
3725 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
Christoph Lameter45edfa52007-05-09 02:32:45 -07003726 l->nodes);
3727 }
3728
Harvey Harrisone374d482008-01-31 15:20:50 -08003729 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07003730 }
3731
3732 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003733 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003734 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08003735 len += sprintf(buf, "No data\n");
3736 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003737}
3738
Christoph Lameter81819f02007-05-06 14:49:36 -07003739enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03003740 SL_ALL, /* All slabs */
3741 SL_PARTIAL, /* Only partially allocated slabs */
3742 SL_CPU, /* Only slabs used for cpu caches */
3743 SL_OBJECTS, /* Determine allocated objects not slabs */
3744 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07003745};
3746
Christoph Lameter205ab992008-04-14 19:11:40 +03003747#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003748#define SO_PARTIAL (1 << SL_PARTIAL)
3749#define SO_CPU (1 << SL_CPU)
3750#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03003751#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07003752
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003753static ssize_t show_slab_objects(struct kmem_cache *s,
3754 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003755{
3756 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003757 int node;
3758 int x;
3759 unsigned long *nodes;
3760 unsigned long *per_cpu;
3761
3762 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03003763 if (!nodes)
3764 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07003765 per_cpu = nodes + nr_node_ids;
3766
Christoph Lameter205ab992008-04-14 19:11:40 +03003767 if (flags & SO_CPU) {
3768 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07003769
Christoph Lameter205ab992008-04-14 19:11:40 +03003770 for_each_possible_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003771 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003772
Christoph Lameter205ab992008-04-14 19:11:40 +03003773 if (!c || c->node < 0)
3774 continue;
3775
3776 if (c->page) {
3777 if (flags & SO_TOTAL)
3778 x = c->page->objects;
3779 else if (flags & SO_OBJECTS)
3780 x = c->page->inuse;
Christoph Lameter81819f02007-05-06 14:49:36 -07003781 else
3782 x = 1;
Christoph Lameter205ab992008-04-14 19:11:40 +03003783
Christoph Lameter81819f02007-05-06 14:49:36 -07003784 total += x;
Christoph Lameter205ab992008-04-14 19:11:40 +03003785 nodes[c->node] += x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003786 }
Christoph Lameter205ab992008-04-14 19:11:40 +03003787 per_cpu[c->node]++;
Christoph Lameter81819f02007-05-06 14:49:36 -07003788 }
3789 }
3790
Christoph Lameter205ab992008-04-14 19:11:40 +03003791 if (flags & SO_ALL) {
3792 for_each_node_state(node, N_NORMAL_MEMORY) {
3793 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003794
Christoph Lameter205ab992008-04-14 19:11:40 +03003795 if (flags & SO_TOTAL)
3796 x = atomic_long_read(&n->total_objects);
3797 else if (flags & SO_OBJECTS)
3798 x = atomic_long_read(&n->total_objects) -
3799 count_partial(n, count_free);
3800
3801 else
3802 x = atomic_long_read(&n->nr_slabs);
3803 total += x;
3804 nodes[node] += x;
3805 }
3806
3807 } else if (flags & SO_PARTIAL) {
3808 for_each_node_state(node, N_NORMAL_MEMORY) {
3809 struct kmem_cache_node *n = get_node(s, node);
3810
3811 if (flags & SO_TOTAL)
3812 x = count_partial(n, count_total);
3813 else if (flags & SO_OBJECTS)
3814 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07003815 else
3816 x = n->nr_partial;
3817 total += x;
3818 nodes[node] += x;
3819 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003820 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003821 x = sprintf(buf, "%lu", total);
3822#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07003823 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07003824 if (nodes[node])
3825 x += sprintf(buf + x, " N%d=%lu",
3826 node, nodes[node]);
3827#endif
3828 kfree(nodes);
3829 return x + sprintf(buf + x, "\n");
3830}
3831
3832static int any_slab_objects(struct kmem_cache *s)
3833{
3834 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003835
3836 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003837 struct kmem_cache_node *n = get_node(s, node);
3838
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003839 if (!n)
3840 continue;
3841
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07003842 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07003843 return 1;
3844 }
3845 return 0;
3846}
3847
3848#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3849#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3850
3851struct slab_attribute {
3852 struct attribute attr;
3853 ssize_t (*show)(struct kmem_cache *s, char *buf);
3854 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3855};
3856
3857#define SLAB_ATTR_RO(_name) \
3858 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3859
3860#define SLAB_ATTR(_name) \
3861 static struct slab_attribute _name##_attr = \
3862 __ATTR(_name, 0644, _name##_show, _name##_store)
3863
Christoph Lameter81819f02007-05-06 14:49:36 -07003864static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3865{
3866 return sprintf(buf, "%d\n", s->size);
3867}
3868SLAB_ATTR_RO(slab_size);
3869
3870static ssize_t align_show(struct kmem_cache *s, char *buf)
3871{
3872 return sprintf(buf, "%d\n", s->align);
3873}
3874SLAB_ATTR_RO(align);
3875
3876static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3877{
3878 return sprintf(buf, "%d\n", s->objsize);
3879}
3880SLAB_ATTR_RO(object_size);
3881
3882static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3883{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003884 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003885}
3886SLAB_ATTR_RO(objs_per_slab);
3887
Christoph Lameter06b285d2008-04-14 19:11:41 +03003888static ssize_t order_store(struct kmem_cache *s,
3889 const char *buf, size_t length)
3890{
Christoph Lameter0121c6192008-04-29 16:11:12 -07003891 unsigned long order;
3892 int err;
3893
3894 err = strict_strtoul(buf, 10, &order);
3895 if (err)
3896 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003897
3898 if (order > slub_max_order || order < slub_min_order)
3899 return -EINVAL;
3900
3901 calculate_sizes(s, order);
3902 return length;
3903}
3904
Christoph Lameter81819f02007-05-06 14:49:36 -07003905static ssize_t order_show(struct kmem_cache *s, char *buf)
3906{
Christoph Lameter834f3d12008-04-14 19:11:31 +03003907 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07003908}
Christoph Lameter06b285d2008-04-14 19:11:41 +03003909SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003910
David Rientjes73d342b2009-02-22 17:40:09 -08003911static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3912{
3913 return sprintf(buf, "%lu\n", s->min_partial);
3914}
3915
3916static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3917 size_t length)
3918{
3919 unsigned long min;
3920 int err;
3921
3922 err = strict_strtoul(buf, 10, &min);
3923 if (err)
3924 return err;
3925
David Rientjesc0bdb232009-02-25 09:16:35 +02003926 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08003927 return length;
3928}
3929SLAB_ATTR(min_partial);
3930
Christoph Lameter81819f02007-05-06 14:49:36 -07003931static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3932{
3933 if (s->ctor) {
3934 int n = sprint_symbol(buf, (unsigned long)s->ctor);
3935
3936 return n + sprintf(buf + n, "\n");
3937 }
3938 return 0;
3939}
3940SLAB_ATTR_RO(ctor);
3941
Christoph Lameter81819f02007-05-06 14:49:36 -07003942static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3943{
3944 return sprintf(buf, "%d\n", s->refcount - 1);
3945}
3946SLAB_ATTR_RO(aliases);
3947
3948static ssize_t slabs_show(struct kmem_cache *s, char *buf)
3949{
Christoph Lameter205ab992008-04-14 19:11:40 +03003950 return show_slab_objects(s, buf, SO_ALL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003951}
3952SLAB_ATTR_RO(slabs);
3953
3954static ssize_t partial_show(struct kmem_cache *s, char *buf)
3955{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003956 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003957}
3958SLAB_ATTR_RO(partial);
3959
3960static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3961{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08003962 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07003963}
3964SLAB_ATTR_RO(cpu_slabs);
3965
3966static ssize_t objects_show(struct kmem_cache *s, char *buf)
3967{
Christoph Lameter205ab992008-04-14 19:11:40 +03003968 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07003969}
3970SLAB_ATTR_RO(objects);
3971
Christoph Lameter205ab992008-04-14 19:11:40 +03003972static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
3973{
3974 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
3975}
3976SLAB_ATTR_RO(objects_partial);
3977
3978static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
3979{
3980 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
3981}
3982SLAB_ATTR_RO(total_objects);
3983
Christoph Lameter81819f02007-05-06 14:49:36 -07003984static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
3985{
3986 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
3987}
3988
3989static ssize_t sanity_checks_store(struct kmem_cache *s,
3990 const char *buf, size_t length)
3991{
3992 s->flags &= ~SLAB_DEBUG_FREE;
3993 if (buf[0] == '1')
3994 s->flags |= SLAB_DEBUG_FREE;
3995 return length;
3996}
3997SLAB_ATTR(sanity_checks);
3998
3999static ssize_t trace_show(struct kmem_cache *s, char *buf)
4000{
4001 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4002}
4003
4004static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4005 size_t length)
4006{
4007 s->flags &= ~SLAB_TRACE;
4008 if (buf[0] == '1')
4009 s->flags |= SLAB_TRACE;
4010 return length;
4011}
4012SLAB_ATTR(trace);
4013
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004014#ifdef CONFIG_FAILSLAB
4015static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4016{
4017 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4018}
4019
4020static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4021 size_t length)
4022{
4023 s->flags &= ~SLAB_FAILSLAB;
4024 if (buf[0] == '1')
4025 s->flags |= SLAB_FAILSLAB;
4026 return length;
4027}
4028SLAB_ATTR(failslab);
4029#endif
4030
Christoph Lameter81819f02007-05-06 14:49:36 -07004031static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4032{
4033 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4034}
4035
4036static ssize_t reclaim_account_store(struct kmem_cache *s,
4037 const char *buf, size_t length)
4038{
4039 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4040 if (buf[0] == '1')
4041 s->flags |= SLAB_RECLAIM_ACCOUNT;
4042 return length;
4043}
4044SLAB_ATTR(reclaim_account);
4045
4046static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4047{
Christoph Lameter5af60832007-05-06 14:49:56 -07004048 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004049}
4050SLAB_ATTR_RO(hwcache_align);
4051
4052#ifdef CONFIG_ZONE_DMA
4053static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4054{
4055 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4056}
4057SLAB_ATTR_RO(cache_dma);
4058#endif
4059
4060static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4061{
4062 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4063}
4064SLAB_ATTR_RO(destroy_by_rcu);
4065
4066static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4067{
4068 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4069}
4070
4071static ssize_t red_zone_store(struct kmem_cache *s,
4072 const char *buf, size_t length)
4073{
4074 if (any_slab_objects(s))
4075 return -EBUSY;
4076
4077 s->flags &= ~SLAB_RED_ZONE;
4078 if (buf[0] == '1')
4079 s->flags |= SLAB_RED_ZONE;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004080 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004081 return length;
4082}
4083SLAB_ATTR(red_zone);
4084
4085static ssize_t poison_show(struct kmem_cache *s, char *buf)
4086{
4087 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4088}
4089
4090static ssize_t poison_store(struct kmem_cache *s,
4091 const char *buf, size_t length)
4092{
4093 if (any_slab_objects(s))
4094 return -EBUSY;
4095
4096 s->flags &= ~SLAB_POISON;
4097 if (buf[0] == '1')
4098 s->flags |= SLAB_POISON;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004099 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004100 return length;
4101}
4102SLAB_ATTR(poison);
4103
4104static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4105{
4106 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4107}
4108
4109static ssize_t store_user_store(struct kmem_cache *s,
4110 const char *buf, size_t length)
4111{
4112 if (any_slab_objects(s))
4113 return -EBUSY;
4114
4115 s->flags &= ~SLAB_STORE_USER;
4116 if (buf[0] == '1')
4117 s->flags |= SLAB_STORE_USER;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004118 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004119 return length;
4120}
4121SLAB_ATTR(store_user);
4122
Christoph Lameter53e15af2007-05-06 14:49:43 -07004123static ssize_t validate_show(struct kmem_cache *s, char *buf)
4124{
4125 return 0;
4126}
4127
4128static ssize_t validate_store(struct kmem_cache *s,
4129 const char *buf, size_t length)
4130{
Christoph Lameter434e2452007-07-17 04:03:30 -07004131 int ret = -EINVAL;
4132
4133 if (buf[0] == '1') {
4134 ret = validate_slab_cache(s);
4135 if (ret >= 0)
4136 ret = length;
4137 }
4138 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004139}
4140SLAB_ATTR(validate);
4141
Christoph Lameter2086d262007-05-06 14:49:46 -07004142static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4143{
4144 return 0;
4145}
4146
4147static ssize_t shrink_store(struct kmem_cache *s,
4148 const char *buf, size_t length)
4149{
4150 if (buf[0] == '1') {
4151 int rc = kmem_cache_shrink(s);
4152
4153 if (rc)
4154 return rc;
4155 } else
4156 return -EINVAL;
4157 return length;
4158}
4159SLAB_ATTR(shrink);
4160
Christoph Lameter88a420e2007-05-06 14:49:45 -07004161static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4162{
4163 if (!(s->flags & SLAB_STORE_USER))
4164 return -ENOSYS;
4165 return list_locations(s, buf, TRACK_ALLOC);
4166}
4167SLAB_ATTR_RO(alloc_calls);
4168
4169static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4170{
4171 if (!(s->flags & SLAB_STORE_USER))
4172 return -ENOSYS;
4173 return list_locations(s, buf, TRACK_FREE);
4174}
4175SLAB_ATTR_RO(free_calls);
4176
Christoph Lameter81819f02007-05-06 14:49:36 -07004177#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004178static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004179{
Christoph Lameter98246012008-01-07 23:20:26 -08004180 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004181}
4182
Christoph Lameter98246012008-01-07 23:20:26 -08004183static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004184 const char *buf, size_t length)
4185{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004186 unsigned long ratio;
4187 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004188
Christoph Lameter0121c6192008-04-29 16:11:12 -07004189 err = strict_strtoul(buf, 10, &ratio);
4190 if (err)
4191 return err;
4192
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004193 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004194 s->remote_node_defrag_ratio = ratio * 10;
4195
Christoph Lameter81819f02007-05-06 14:49:36 -07004196 return length;
4197}
Christoph Lameter98246012008-01-07 23:20:26 -08004198SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004199#endif
4200
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004201#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004202static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4203{
4204 unsigned long sum = 0;
4205 int cpu;
4206 int len;
4207 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4208
4209 if (!data)
4210 return -ENOMEM;
4211
4212 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004213 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004214
4215 data[cpu] = x;
4216 sum += x;
4217 }
4218
4219 len = sprintf(buf, "%lu", sum);
4220
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004221#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004222 for_each_online_cpu(cpu) {
4223 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004224 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004225 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004226#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004227 kfree(data);
4228 return len + sprintf(buf + len, "\n");
4229}
4230
David Rientjes78eb00c2009-10-15 02:20:22 -07004231static void clear_stat(struct kmem_cache *s, enum stat_item si)
4232{
4233 int cpu;
4234
4235 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004236 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004237}
4238
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004239#define STAT_ATTR(si, text) \
4240static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4241{ \
4242 return show_stat(s, buf, si); \
4243} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004244static ssize_t text##_store(struct kmem_cache *s, \
4245 const char *buf, size_t length) \
4246{ \
4247 if (buf[0] != '0') \
4248 return -EINVAL; \
4249 clear_stat(s, si); \
4250 return length; \
4251} \
4252SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004253
4254STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4255STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4256STAT_ATTR(FREE_FASTPATH, free_fastpath);
4257STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4258STAT_ATTR(FREE_FROZEN, free_frozen);
4259STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4260STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4261STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4262STAT_ATTR(ALLOC_SLAB, alloc_slab);
4263STAT_ATTR(ALLOC_REFILL, alloc_refill);
4264STAT_ATTR(FREE_SLAB, free_slab);
4265STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4266STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4267STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4268STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4269STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4270STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter65c33762008-04-14 19:11:40 +03004271STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004272#endif
4273
Pekka Enberg06428782008-01-07 23:20:27 -08004274static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004275 &slab_size_attr.attr,
4276 &object_size_attr.attr,
4277 &objs_per_slab_attr.attr,
4278 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004279 &min_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004280 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004281 &objects_partial_attr.attr,
4282 &total_objects_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004283 &slabs_attr.attr,
4284 &partial_attr.attr,
4285 &cpu_slabs_attr.attr,
4286 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004287 &aliases_attr.attr,
4288 &align_attr.attr,
4289 &sanity_checks_attr.attr,
4290 &trace_attr.attr,
4291 &hwcache_align_attr.attr,
4292 &reclaim_account_attr.attr,
4293 &destroy_by_rcu_attr.attr,
4294 &red_zone_attr.attr,
4295 &poison_attr.attr,
4296 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004297 &validate_attr.attr,
Christoph Lameter2086d262007-05-06 14:49:46 -07004298 &shrink_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004299 &alloc_calls_attr.attr,
4300 &free_calls_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004301#ifdef CONFIG_ZONE_DMA
4302 &cache_dma_attr.attr,
4303#endif
4304#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004305 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004306#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004307#ifdef CONFIG_SLUB_STATS
4308 &alloc_fastpath_attr.attr,
4309 &alloc_slowpath_attr.attr,
4310 &free_fastpath_attr.attr,
4311 &free_slowpath_attr.attr,
4312 &free_frozen_attr.attr,
4313 &free_add_partial_attr.attr,
4314 &free_remove_partial_attr.attr,
4315 &alloc_from_partial_attr.attr,
4316 &alloc_slab_attr.attr,
4317 &alloc_refill_attr.attr,
4318 &free_slab_attr.attr,
4319 &cpuslab_flush_attr.attr,
4320 &deactivate_full_attr.attr,
4321 &deactivate_empty_attr.attr,
4322 &deactivate_to_head_attr.attr,
4323 &deactivate_to_tail_attr.attr,
4324 &deactivate_remote_frees_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004325 &order_fallback_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004326#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004327#ifdef CONFIG_FAILSLAB
4328 &failslab_attr.attr,
4329#endif
4330
Christoph Lameter81819f02007-05-06 14:49:36 -07004331 NULL
4332};
4333
4334static struct attribute_group slab_attr_group = {
4335 .attrs = slab_attrs,
4336};
4337
4338static ssize_t slab_attr_show(struct kobject *kobj,
4339 struct attribute *attr,
4340 char *buf)
4341{
4342 struct slab_attribute *attribute;
4343 struct kmem_cache *s;
4344 int err;
4345
4346 attribute = to_slab_attr(attr);
4347 s = to_slab(kobj);
4348
4349 if (!attribute->show)
4350 return -EIO;
4351
4352 err = attribute->show(s, buf);
4353
4354 return err;
4355}
4356
4357static ssize_t slab_attr_store(struct kobject *kobj,
4358 struct attribute *attr,
4359 const char *buf, size_t len)
4360{
4361 struct slab_attribute *attribute;
4362 struct kmem_cache *s;
4363 int err;
4364
4365 attribute = to_slab_attr(attr);
4366 s = to_slab(kobj);
4367
4368 if (!attribute->store)
4369 return -EIO;
4370
4371 err = attribute->store(s, buf, len);
4372
4373 return err;
4374}
4375
Christoph Lameter151c6022008-01-07 22:29:05 -08004376static void kmem_cache_release(struct kobject *kobj)
4377{
4378 struct kmem_cache *s = to_slab(kobj);
4379
4380 kfree(s);
4381}
4382
Emese Revfy52cf25d2010-01-19 02:58:23 +01004383static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004384 .show = slab_attr_show,
4385 .store = slab_attr_store,
4386};
4387
4388static struct kobj_type slab_ktype = {
4389 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter151c6022008-01-07 22:29:05 -08004390 .release = kmem_cache_release
Christoph Lameter81819f02007-05-06 14:49:36 -07004391};
4392
4393static int uevent_filter(struct kset *kset, struct kobject *kobj)
4394{
4395 struct kobj_type *ktype = get_ktype(kobj);
4396
4397 if (ktype == &slab_ktype)
4398 return 1;
4399 return 0;
4400}
4401
Emese Revfy9cd43612009-12-31 14:52:51 +01004402static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004403 .filter = uevent_filter,
4404};
4405
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004406static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07004407
4408#define ID_STR_LENGTH 64
4409
4410/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08004411 *
4412 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07004413 */
4414static char *create_unique_id(struct kmem_cache *s)
4415{
4416 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4417 char *p = name;
4418
4419 BUG_ON(!name);
4420
4421 *p++ = ':';
4422 /*
4423 * First flags affecting slabcache operations. We will only
4424 * get here for aliasable slabs so we do not need to support
4425 * too many flags. The flags here must cover all flags that
4426 * are matched during merging to guarantee that the id is
4427 * unique.
4428 */
4429 if (s->flags & SLAB_CACHE_DMA)
4430 *p++ = 'd';
4431 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4432 *p++ = 'a';
4433 if (s->flags & SLAB_DEBUG_FREE)
4434 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02004435 if (!(s->flags & SLAB_NOTRACK))
4436 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07004437 if (p != name + 1)
4438 *p++ = '-';
4439 p += sprintf(p, "%07d", s->size);
4440 BUG_ON(p > name + ID_STR_LENGTH - 1);
4441 return name;
4442}
4443
4444static int sysfs_slab_add(struct kmem_cache *s)
4445{
4446 int err;
4447 const char *name;
4448 int unmergeable;
4449
4450 if (slab_state < SYSFS)
4451 /* Defer until later */
4452 return 0;
4453
4454 unmergeable = slab_unmergeable(s);
4455 if (unmergeable) {
4456 /*
4457 * Slabcache can never be merged so we can use the name proper.
4458 * This is typically the case for debug situations. In that
4459 * case we can catch duplicate names easily.
4460 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004461 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004462 name = s->name;
4463 } else {
4464 /*
4465 * Create a unique name for the slab as a target
4466 * for the symlinks.
4467 */
4468 name = create_unique_id(s);
4469 }
4470
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004471 s->kobj.kset = slab_kset;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004472 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4473 if (err) {
4474 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004475 return err;
Greg Kroah-Hartman1eada112007-12-17 23:05:35 -07004476 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004477
4478 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004479 if (err) {
4480 kobject_del(&s->kobj);
4481 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004482 return err;
Xiaotian Feng5788d8a2009-07-22 11:28:53 +08004483 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004484 kobject_uevent(&s->kobj, KOBJ_ADD);
4485 if (!unmergeable) {
4486 /* Setup first alias */
4487 sysfs_slab_alias(s, s->name);
4488 kfree(name);
4489 }
4490 return 0;
4491}
4492
4493static void sysfs_slab_remove(struct kmem_cache *s)
4494{
Christoph Lameter2bce6482010-07-19 11:39:11 -05004495 if (slab_state < SYSFS)
4496 /*
4497 * Sysfs has not been setup yet so no need to remove the
4498 * cache from sysfs.
4499 */
4500 return;
4501
Christoph Lameter81819f02007-05-06 14:49:36 -07004502 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4503 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08004504 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07004505}
4506
4507/*
4508 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11004509 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07004510 */
4511struct saved_alias {
4512 struct kmem_cache *s;
4513 const char *name;
4514 struct saved_alias *next;
4515};
4516
Adrian Bunk5af328a2007-07-17 04:03:27 -07004517static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07004518
4519static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4520{
4521 struct saved_alias *al;
4522
4523 if (slab_state == SYSFS) {
4524 /*
4525 * If we have a leftover link then remove it.
4526 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004527 sysfs_remove_link(&slab_kset->kobj, name);
4528 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004529 }
4530
4531 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4532 if (!al)
4533 return -ENOMEM;
4534
4535 al->s = s;
4536 al->name = name;
4537 al->next = alias_list;
4538 alias_list = al;
4539 return 0;
4540}
4541
4542static int __init slab_sysfs_init(void)
4543{
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004544 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004545 int err;
4546
Christoph Lameter2bce6482010-07-19 11:39:11 -05004547 down_write(&slub_lock);
4548
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08004549 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06004550 if (!slab_kset) {
Christoph Lameter2bce6482010-07-19 11:39:11 -05004551 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004552 printk(KERN_ERR "Cannot register slab subsystem.\n");
4553 return -ENOSYS;
4554 }
4555
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004556 slab_state = SYSFS;
4557
Christoph Lameter5b95a4a2007-07-17 04:03:19 -07004558 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004559 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004560 if (err)
4561 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4562 " to sysfs\n", s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07004563 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004564
4565 while (alias_list) {
4566 struct saved_alias *al = alias_list;
4567
4568 alias_list = alias_list->next;
4569 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07004570 if (err)
4571 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4572 " %s to sysfs\n", s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07004573 kfree(al);
4574 }
4575
Christoph Lameter2bce6482010-07-19 11:39:11 -05004576 up_write(&slub_lock);
Christoph Lameter81819f02007-05-06 14:49:36 -07004577 resiliency_test();
4578 return 0;
4579}
4580
4581__initcall(slab_sysfs_init);
Christoph Lameter81819f02007-05-06 14:49:36 -07004582#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004583
4584/*
4585 * The /proc/slabinfo ABI
4586 */
Linus Torvalds158a9622008-01-02 13:04:48 -08004587#ifdef CONFIG_SLABINFO
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004588static void print_slabinfo_header(struct seq_file *m)
4589{
4590 seq_puts(m, "slabinfo - version: 2.1\n");
4591 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4592 "<objperslab> <pagesperslab>");
4593 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4594 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4595 seq_putc(m, '\n');
4596}
4597
4598static void *s_start(struct seq_file *m, loff_t *pos)
4599{
4600 loff_t n = *pos;
4601
4602 down_read(&slub_lock);
4603 if (!n)
4604 print_slabinfo_header(m);
4605
4606 return seq_list_start(&slab_caches, *pos);
4607}
4608
4609static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4610{
4611 return seq_list_next(p, &slab_caches, pos);
4612}
4613
4614static void s_stop(struct seq_file *m, void *p)
4615{
4616 up_read(&slub_lock);
4617}
4618
4619static int s_show(struct seq_file *m, void *p)
4620{
4621 unsigned long nr_partials = 0;
4622 unsigned long nr_slabs = 0;
4623 unsigned long nr_inuse = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004624 unsigned long nr_objs = 0;
4625 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004626 struct kmem_cache *s;
4627 int node;
4628
4629 s = list_entry(p, struct kmem_cache, list);
4630
4631 for_each_online_node(node) {
4632 struct kmem_cache_node *n = get_node(s, node);
4633
4634 if (!n)
4635 continue;
4636
4637 nr_partials += n->nr_partial;
4638 nr_slabs += atomic_long_read(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03004639 nr_objs += atomic_long_read(&n->total_objects);
4640 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004641 }
4642
Christoph Lameter205ab992008-04-14 19:11:40 +03004643 nr_inuse = nr_objs - nr_free;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004644
4645 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
Christoph Lameter834f3d12008-04-14 19:11:31 +03004646 nr_objs, s->size, oo_objects(s->oo),
4647 (1 << oo_order(s->oo)));
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004648 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4649 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4650 0UL);
4651 seq_putc(m, '\n');
4652 return 0;
4653}
4654
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004655static const struct seq_operations slabinfo_op = {
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01004656 .start = s_start,
4657 .next = s_next,
4658 .stop = s_stop,
4659 .show = s_show,
4660};
4661
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004662static int slabinfo_open(struct inode *inode, struct file *file)
4663{
4664 return seq_open(file, &slabinfo_op);
4665}
4666
4667static const struct file_operations proc_slabinfo_operations = {
4668 .open = slabinfo_open,
4669 .read = seq_read,
4670 .llseek = seq_lseek,
4671 .release = seq_release,
4672};
4673
4674static int __init slab_proc_init(void)
4675{
WANG Congcf5d1132009-08-18 19:11:40 +03004676 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04004677 return 0;
4678}
4679module_init(slab_proc_init);
Linus Torvalds158a9622008-01-02 13:04:48 -08004680#endif /* CONFIG_SLABINFO */