2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter <clameter@sgi.com>
12 #include <linux/module.h>
13 #include <linux/bit_spinlock.h>
14 #include <linux/interrupt.h>
15 #include <linux/bitops.h>
16 #include <linux/slab.h>
17 #include <linux/seq_file.h>
18 #include <linux/cpu.h>
19 #include <linux/cpuset.h>
20 #include <linux/mempolicy.h>
21 #include <linux/ctype.h>
22 #include <linux/kallsyms.h>
29 * The slab_lock protects operations on the object of a particular
30 * slab and its metadata in the page struct. If the slab lock
31 * has been taken then no allocations nor frees can be performed
32 * on the objects in the slab nor can the slab be added or removed
33 * from the partial or full lists since this would mean modifying
34 * the page_struct of the slab.
36 * The list_lock protects the partial and full list on each node and
37 * the partial slab counter. If taken then no new slabs may be added or
38 * removed from the lists nor make the number of partial slabs be modified.
39 * (Note that the total number of slabs is an atomic value that may be
40 * modified without taking the list lock).
42 * The list_lock is a centralized lock and thus we avoid taking it as
43 * much as possible. As long as SLUB does not have to handle partial
44 * slabs, operations can continue without any centralized lock. F.e.
45 * allocating a long series of objects that fill up slabs does not require
48 * The lock order is sometimes inverted when we are trying to get a slab
49 * off a list. We take the list_lock and then look for a page on the list
50 * to use. While we do that objects in the slabs may be freed. We can
51 * only operate on the slab if we have also taken the slab_lock. So we use
52 * a slab_trylock() on the slab. If trylock was successful then no frees
53 * can occur anymore and we can use the slab for allocations etc. If the
54 * slab_trylock() does not succeed then frees are in progress in the slab and
55 * we must stay away from it for a while since we may cause a bouncing
56 * cacheline if we try to acquire the lock. So go onto the next slab.
57 * If all pages are busy then we may allocate a new slab instead of reusing
58 * a partial slab. A new slab has noone operating on it and thus there is
59 * no danger of cacheline contention.
61 * Interrupts are disabled during allocation and deallocation in order to
62 * make the slab allocator safe to use in the context of an irq. In addition
63 * interrupts are disabled to ensure that the processor does not change
64 * while handling per_cpu slabs, due to kernel preemption.
66 * SLUB assigns one slab for allocation to each processor.
67 * Allocations only occur from these slabs called cpu slabs.
69 * Slabs with free elements are kept on a partial list and during regular
70 * operations no list for full slabs is used. If an object in a full slab is
71 * freed then the slab will show up again on the partial lists.
72 * We track full slabs for debugging purposes though because otherwise we
73 * cannot scan all objects.
75 * Slabs are freed when they become empty. Teardown and setup is
76 * minimal so we rely on the page allocators per cpu caches for
77 * fast frees and allocs.
79 * Overloading of page flags that are otherwise used for LRU management.
81 * PageActive The slab is frozen and exempt from list processing.
82 * This means that the slab is dedicated to a purpose
83 * such as satisfying allocations for a specific
84 * processor. Objects may be freed in the slab while
85 * it is frozen but slab_free will then skip the usual
86 * list operations. It is up to the processor holding
87 * the slab to integrate the slab into the slab lists
88 * when the slab is no longer needed.
90 * One use of this flag is to mark slabs that are
91 * used for allocations. Then such a slab becomes a cpu
92 * slab. The cpu slab may be equipped with an additional
93 * lockless_freelist that allows lockless access to
94 * free objects in addition to the regular freelist
95 * that requires the slab lock.
97 * PageError Slab requires special handling due to debug
98 * options set. This moves slab handling out of
99 * the fast path and disables lockless freelists.
102 #define FROZEN (1 << PG_active)
104 #ifdef CONFIG_SLUB_DEBUG
105 #define SLABDEBUG (1 << PG_error)
110 static inline int SlabFrozen(struct page
*page
)
112 return page
->flags
& FROZEN
;
115 static inline void SetSlabFrozen(struct page
*page
)
117 page
->flags
|= FROZEN
;
120 static inline void ClearSlabFrozen(struct page
*page
)
122 page
->flags
&= ~FROZEN
;
125 static inline int SlabDebug(struct page
*page
)
127 return page
->flags
& SLABDEBUG
;
130 static inline void SetSlabDebug(struct page
*page
)
132 page
->flags
|= SLABDEBUG
;
135 static inline void ClearSlabDebug(struct page
*page
)
137 page
->flags
&= ~SLABDEBUG
;
141 * Issues still to be resolved:
143 * - The per cpu array is updated for each new slab and and is a remote
144 * cacheline for most nodes. This could become a bouncing cacheline given
145 * enough frequent updates. There are 16 pointers in a cacheline, so at
146 * max 16 cpus could compete for the cacheline which may be okay.
148 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
150 * - Variable sizing of the per node arrays
153 /* Enable to test recovery from slab corruption on boot */
154 #undef SLUB_RESILIENCY_TEST
159 * Small page size. Make sure that we do not fragment memory
161 #define DEFAULT_MAX_ORDER 1
162 #define DEFAULT_MIN_OBJECTS 4
167 * Large page machines are customarily able to handle larger
170 #define DEFAULT_MAX_ORDER 2
171 #define DEFAULT_MIN_OBJECTS 8
176 * Mininum number of partial slabs. These will be left on the partial
177 * lists even if they are empty. kmem_cache_shrink may reclaim them.
179 #define MIN_PARTIAL 2
182 * Maximum number of desirable partial slabs.
183 * The existence of more partial slabs makes kmem_cache_shrink
184 * sort the partial list by the number of objects in the.
186 #define MAX_PARTIAL 10
188 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
189 SLAB_POISON | SLAB_STORE_USER)
192 * Set of flags that will prevent slab merging
194 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
195 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
197 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
200 #ifndef ARCH_KMALLOC_MINALIGN
201 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
204 #ifndef ARCH_SLAB_MINALIGN
205 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
208 /* Internal SLUB flags */
209 #define __OBJECT_POISON 0x80000000 /* Poison object */
211 /* Not all arches define cache_line_size */
212 #ifndef cache_line_size
213 #define cache_line_size() L1_CACHE_BYTES
216 static int kmem_size
= sizeof(struct kmem_cache
);
219 static struct notifier_block slab_notifier
;
223 DOWN
, /* No slab functionality available */
224 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
225 UP
, /* Everything works but does not show up in sysfs */
229 /* A list of all slab caches on the system */
230 static DECLARE_RWSEM(slub_lock
);
231 LIST_HEAD(slab_caches
);
234 * Tracking user of a slab.
237 void *addr
; /* Called from address */
238 int cpu
; /* Was running on cpu */
239 int pid
; /* Pid context */
240 unsigned long when
; /* When did the operation occur */
243 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
245 #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)
246 static int sysfs_slab_add(struct kmem_cache
*);
247 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
248 static void sysfs_slab_remove(struct kmem_cache
*);
250 static int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
251 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
) { return 0; }
252 static void sysfs_slab_remove(struct kmem_cache
*s
) {}
255 /********************************************************************
256 * Core slab cache functions
257 *******************************************************************/
259 int slab_is_available(void)
261 return slab_state
>= UP
;
264 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
267 return s
->node
[node
];
269 return &s
->local_node
;
273 static inline int check_valid_pointer(struct kmem_cache
*s
,
274 struct page
*page
, const void *object
)
281 base
= page_address(page
);
282 if (object
< base
|| object
>= base
+ s
->objects
* s
->size
||
283 (object
- base
) % s
->size
) {
291 * Slow version of get and set free pointer.
293 * This version requires touching the cache lines of kmem_cache which
294 * we avoid to do in the fast alloc free paths. There we obtain the offset
295 * from the page struct.
297 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
299 return *(void **)(object
+ s
->offset
);
302 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
304 *(void **)(object
+ s
->offset
) = fp
;
307 /* Loop over all objects in a slab */
308 #define for_each_object(__p, __s, __addr) \
309 for (__p = (__addr); __p < (__addr) + (__s)->objects * (__s)->size;\
313 #define for_each_free_object(__p, __s, __free) \
314 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
316 /* Determine object index from a given position */
317 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
319 return (p
- addr
) / s
->size
;
322 #ifdef CONFIG_SLUB_DEBUG
326 static int slub_debug
;
328 static char *slub_debug_slabs
;
333 static void print_section(char *text
, u8
*addr
, unsigned int length
)
341 for (i
= 0; i
< length
; i
++) {
343 printk(KERN_ERR
"%10s 0x%p: ", text
, addr
+ i
);
346 printk(" %02x", addr
[i
]);
348 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
350 printk(" %s\n",ascii
);
361 printk(" %s\n", ascii
);
365 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
366 enum track_item alloc
)
371 p
= object
+ s
->offset
+ sizeof(void *);
373 p
= object
+ s
->inuse
;
378 static void set_track(struct kmem_cache
*s
, void *object
,
379 enum track_item alloc
, void *addr
)
384 p
= object
+ s
->offset
+ sizeof(void *);
386 p
= object
+ s
->inuse
;
391 p
->cpu
= smp_processor_id();
392 p
->pid
= current
? current
->pid
: -1;
395 memset(p
, 0, sizeof(struct track
));
398 static void init_tracking(struct kmem_cache
*s
, void *object
)
400 if (s
->flags
& SLAB_STORE_USER
) {
401 set_track(s
, object
, TRACK_FREE
, NULL
);
402 set_track(s
, object
, TRACK_ALLOC
, NULL
);
406 static void print_track(const char *s
, struct track
*t
)
411 printk(KERN_ERR
"%s: ", s
);
412 __print_symbol("%s", (unsigned long)t
->addr
);
413 printk(" jiffies_ago=%lu cpu=%u pid=%d\n", jiffies
- t
->when
, t
->cpu
, t
->pid
);
416 static void print_trailer(struct kmem_cache
*s
, u8
*p
)
418 unsigned int off
; /* Offset of last byte */
420 if (s
->flags
& SLAB_RED_ZONE
)
421 print_section("Redzone", p
+ s
->objsize
,
422 s
->inuse
- s
->objsize
);
424 printk(KERN_ERR
"FreePointer 0x%p -> 0x%p\n",
426 get_freepointer(s
, p
));
429 off
= s
->offset
+ sizeof(void *);
433 if (s
->flags
& SLAB_STORE_USER
) {
434 print_track("Last alloc", get_track(s
, p
, TRACK_ALLOC
));
435 print_track("Last free ", get_track(s
, p
, TRACK_FREE
));
436 off
+= 2 * sizeof(struct track
);
440 /* Beginning of the filler is the free pointer */
441 print_section("Filler", p
+ off
, s
->size
- off
);
444 static void object_err(struct kmem_cache
*s
, struct page
*page
,
445 u8
*object
, char *reason
)
447 u8
*addr
= page_address(page
);
449 printk(KERN_ERR
"*** SLUB %s: %s@0x%p slab 0x%p\n",
450 s
->name
, reason
, object
, page
);
451 printk(KERN_ERR
" offset=%tu flags=0x%04lx inuse=%u freelist=0x%p\n",
452 object
- addr
, page
->flags
, page
->inuse
, page
->freelist
);
453 if (object
> addr
+ 16)
454 print_section("Bytes b4", object
- 16, 16);
455 print_section("Object", object
, min(s
->objsize
, 128));
456 print_trailer(s
, object
);
460 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *reason
, ...)
465 va_start(args
, reason
);
466 vsnprintf(buf
, sizeof(buf
), reason
, args
);
468 printk(KERN_ERR
"*** SLUB %s: %s in slab @0x%p\n", s
->name
, buf
,
473 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
477 if (s
->flags
& __OBJECT_POISON
) {
478 memset(p
, POISON_FREE
, s
->objsize
- 1);
479 p
[s
->objsize
-1] = POISON_END
;
482 if (s
->flags
& SLAB_RED_ZONE
)
483 memset(p
+ s
->objsize
,
484 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
485 s
->inuse
- s
->objsize
);
488 static int check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
491 if (*start
!= (u8
)value
)
503 * Bytes of the object to be managed.
504 * If the freepointer may overlay the object then the free
505 * pointer is the first word of the object.
507 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
510 * object + s->objsize
511 * Padding to reach word boundary. This is also used for Redzoning.
512 * Padding is extended by another word if Redzoning is enabled and
515 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
516 * 0xcc (RED_ACTIVE) for objects in use.
519 * Meta data starts here.
521 * A. Free pointer (if we cannot overwrite object on free)
522 * B. Tracking data for SLAB_STORE_USER
523 * C. Padding to reach required alignment boundary or at mininum
524 * one word if debuggin is on to be able to detect writes
525 * before the word boundary.
527 * Padding is done using 0x5a (POISON_INUSE)
530 * Nothing is used beyond s->size.
532 * If slabcaches are merged then the objsize and inuse boundaries are mostly
533 * ignored. And therefore no slab options that rely on these boundaries
534 * may be used with merged slabcaches.
537 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
538 void *from
, void *to
)
540 printk(KERN_ERR
"@@@ SLUB %s: Restoring %s (0x%x) from 0x%p-0x%p\n",
541 s
->name
, message
, data
, from
, to
- 1);
542 memset(from
, data
, to
- from
);
545 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
547 unsigned long off
= s
->inuse
; /* The end of info */
550 /* Freepointer is placed after the object. */
551 off
+= sizeof(void *);
553 if (s
->flags
& SLAB_STORE_USER
)
554 /* We also have user information there */
555 off
+= 2 * sizeof(struct track
);
560 if (check_bytes(p
+ off
, POISON_INUSE
, s
->size
- off
))
563 object_err(s
, page
, p
, "Object padding check fails");
568 restore_bytes(s
, "object padding", POISON_INUSE
, p
+ off
, p
+ s
->size
);
572 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
575 int length
, remainder
;
577 if (!(s
->flags
& SLAB_POISON
))
580 p
= page_address(page
);
581 length
= s
->objects
* s
->size
;
582 remainder
= (PAGE_SIZE
<< s
->order
) - length
;
586 if (!check_bytes(p
+ length
, POISON_INUSE
, remainder
)) {
587 slab_err(s
, page
, "Padding check failed");
588 restore_bytes(s
, "slab padding", POISON_INUSE
, p
+ length
,
589 p
+ length
+ remainder
);
595 static int check_object(struct kmem_cache
*s
, struct page
*page
,
596 void *object
, int active
)
599 u8
*endobject
= object
+ s
->objsize
;
601 if (s
->flags
& SLAB_RED_ZONE
) {
603 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
605 if (!check_bytes(endobject
, red
, s
->inuse
- s
->objsize
)) {
606 object_err(s
, page
, object
,
607 active
? "Redzone Active" : "Redzone Inactive");
608 restore_bytes(s
, "redzone", red
,
609 endobject
, object
+ s
->inuse
);
613 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
&&
614 !check_bytes(endobject
, POISON_INUSE
,
615 s
->inuse
- s
->objsize
)) {
616 object_err(s
, page
, p
, "Alignment padding check fails");
618 * Fix it so that there will not be another report.
620 * Hmmm... We may be corrupting an object that now expects
621 * to be longer than allowed.
623 restore_bytes(s
, "alignment padding", POISON_INUSE
,
624 endobject
, object
+ s
->inuse
);
628 if (s
->flags
& SLAB_POISON
) {
629 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
630 (!check_bytes(p
, POISON_FREE
, s
->objsize
- 1) ||
631 p
[s
->objsize
- 1] != POISON_END
)) {
633 object_err(s
, page
, p
, "Poison check failed");
634 restore_bytes(s
, "Poison", POISON_FREE
,
635 p
, p
+ s
->objsize
-1);
636 restore_bytes(s
, "Poison", POISON_END
,
637 p
+ s
->objsize
- 1, p
+ s
->objsize
);
641 * check_pad_bytes cleans up on its own.
643 check_pad_bytes(s
, page
, p
);
646 if (!s
->offset
&& active
)
648 * Object and freepointer overlap. Cannot check
649 * freepointer while object is allocated.
653 /* Check free pointer validity */
654 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
655 object_err(s
, page
, p
, "Freepointer corrupt");
657 * No choice but to zap it and thus loose the remainder
658 * of the free objects in this slab. May cause
659 * another error because the object count is now wrong.
661 set_freepointer(s
, p
, NULL
);
667 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
669 VM_BUG_ON(!irqs_disabled());
671 if (!PageSlab(page
)) {
672 slab_err(s
, page
, "Not a valid slab page flags=%lx "
673 "mapping=0x%p count=%d", page
->flags
, page
->mapping
,
677 if (page
->offset
* sizeof(void *) != s
->offset
) {
678 slab_err(s
, page
, "Corrupted offset %lu flags=0x%lx "
679 "mapping=0x%p count=%d",
680 (unsigned long)(page
->offset
* sizeof(void *)),
686 if (page
->inuse
> s
->objects
) {
687 slab_err(s
, page
, "inuse %u > max %u @0x%p flags=%lx "
688 "mapping=0x%p count=%d",
689 s
->name
, page
->inuse
, s
->objects
, page
->flags
,
690 page
->mapping
, page_count(page
));
693 /* Slab_pad_check fixes things up after itself */
694 slab_pad_check(s
, page
);
699 * Determine if a certain object on a page is on the freelist. Must hold the
700 * slab lock to guarantee that the chains are in a consistent state.
702 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
705 void *fp
= page
->freelist
;
708 while (fp
&& nr
<= s
->objects
) {
711 if (!check_valid_pointer(s
, page
, fp
)) {
713 object_err(s
, page
, object
,
714 "Freechain corrupt");
715 set_freepointer(s
, object
, NULL
);
718 slab_err(s
, page
, "Freepointer 0x%p corrupt",
720 page
->freelist
= NULL
;
721 page
->inuse
= s
->objects
;
722 printk(KERN_ERR
"@@@ SLUB %s: Freelist "
723 "cleared. Slab 0x%p\n",
730 fp
= get_freepointer(s
, object
);
734 if (page
->inuse
!= s
->objects
- nr
) {
735 slab_err(s
, page
, "Wrong object count. Counter is %d but "
736 "counted were %d", s
, page
, page
->inuse
,
738 page
->inuse
= s
->objects
- nr
;
739 printk(KERN_ERR
"@@@ SLUB %s: Object count adjusted. "
740 "Slab @0x%p\n", s
->name
, page
);
742 return search
== NULL
;
745 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
, int alloc
)
747 if (s
->flags
& SLAB_TRACE
) {
748 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
750 alloc
? "alloc" : "free",
755 print_section("Object", (void *)object
, s
->objsize
);
762 * Tracking of fully allocated slabs for debugging purposes.
764 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
766 spin_lock(&n
->list_lock
);
767 list_add(&page
->lru
, &n
->full
);
768 spin_unlock(&n
->list_lock
);
771 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
773 struct kmem_cache_node
*n
;
775 if (!(s
->flags
& SLAB_STORE_USER
))
778 n
= get_node(s
, page_to_nid(page
));
780 spin_lock(&n
->list_lock
);
781 list_del(&page
->lru
);
782 spin_unlock(&n
->list_lock
);
785 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
788 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
791 init_object(s
, object
, 0);
792 init_tracking(s
, object
);
795 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
796 void *object
, void *addr
)
798 if (!check_slab(s
, page
))
801 if (object
&& !on_freelist(s
, page
, object
)) {
802 slab_err(s
, page
, "Object 0x%p already allocated", object
);
806 if (!check_valid_pointer(s
, page
, object
)) {
807 object_err(s
, page
, object
, "Freelist Pointer check fails");
811 if (object
&& !check_object(s
, page
, object
, 0))
814 /* Success perform special debug activities for allocs */
815 if (s
->flags
& SLAB_STORE_USER
)
816 set_track(s
, object
, TRACK_ALLOC
, addr
);
817 trace(s
, page
, object
, 1);
818 init_object(s
, object
, 1);
822 if (PageSlab(page
)) {
824 * If this is a slab page then lets do the best we can
825 * to avoid issues in the future. Marking all objects
826 * as used avoids touching the remaining objects.
828 printk(KERN_ERR
"@@@ SLUB: %s slab 0x%p. Marking all objects used.\n",
830 page
->inuse
= s
->objects
;
831 page
->freelist
= NULL
;
832 /* Fix up fields that may be corrupted */
833 page
->offset
= s
->offset
/ sizeof(void *);
838 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
839 void *object
, void *addr
)
841 if (!check_slab(s
, page
))
844 if (!check_valid_pointer(s
, page
, object
)) {
845 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
849 if (on_freelist(s
, page
, object
)) {
850 slab_err(s
, page
, "Object 0x%p already free", object
);
854 if (!check_object(s
, page
, object
, 1))
857 if (unlikely(s
!= page
->slab
)) {
859 slab_err(s
, page
, "Attempt to free object(0x%p) "
860 "outside of slab", object
);
864 "SLUB <none>: no slab for object 0x%p.\n",
869 slab_err(s
, page
, "object at 0x%p belongs "
870 "to slab %s", object
, page
->slab
->name
);
874 /* Special debug activities for freeing objects */
875 if (!SlabFrozen(page
) && !page
->freelist
)
876 remove_full(s
, page
);
877 if (s
->flags
& SLAB_STORE_USER
)
878 set_track(s
, object
, TRACK_FREE
, addr
);
879 trace(s
, page
, object
, 0);
880 init_object(s
, object
, 0);
884 printk(KERN_ERR
"@@@ SLUB: %s slab 0x%p object at 0x%p not freed.\n",
885 s
->name
, page
, object
);
889 static int __init
setup_slub_debug(char *str
)
891 if (!str
|| *str
!= '=')
892 slub_debug
= DEBUG_DEFAULT_FLAGS
;
895 if (*str
== 0 || *str
== ',')
896 slub_debug
= DEBUG_DEFAULT_FLAGS
;
898 for( ;*str
&& *str
!= ','; str
++)
900 case 'f' : case 'F' :
901 slub_debug
|= SLAB_DEBUG_FREE
;
903 case 'z' : case 'Z' :
904 slub_debug
|= SLAB_RED_ZONE
;
906 case 'p' : case 'P' :
907 slub_debug
|= SLAB_POISON
;
909 case 'u' : case 'U' :
910 slub_debug
|= SLAB_STORE_USER
;
912 case 't' : case 'T' :
913 slub_debug
|= SLAB_TRACE
;
916 printk(KERN_ERR
"slub_debug option '%c' "
917 "unknown. skipped\n",*str
);
922 slub_debug_slabs
= str
+ 1;
926 __setup("slub_debug", setup_slub_debug
);
928 static void kmem_cache_open_debug_check(struct kmem_cache
*s
)
931 * The page->offset field is only 16 bit wide. This is an offset
932 * in units of words from the beginning of an object. If the slab
933 * size is bigger then we cannot move the free pointer behind the
936 * On 32 bit platforms the limit is 256k. On 64bit platforms
939 * Debugging or ctor may create a need to move the free
940 * pointer. Fail if this happens.
942 if (s
->objsize
>= 65535 * sizeof(void *)) {
943 BUG_ON(s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
|
944 SLAB_STORE_USER
| SLAB_DESTROY_BY_RCU
));
949 * Enable debugging if selected on the kernel commandline.
951 if (slub_debug
&& (!slub_debug_slabs
||
952 strncmp(slub_debug_slabs
, s
->name
,
953 strlen(slub_debug_slabs
)) == 0))
954 s
->flags
|= slub_debug
;
957 static inline void setup_object_debug(struct kmem_cache
*s
,
958 struct page
*page
, void *object
) {}
960 static inline int alloc_debug_processing(struct kmem_cache
*s
,
961 struct page
*page
, void *object
, void *addr
) { return 0; }
963 static inline int free_debug_processing(struct kmem_cache
*s
,
964 struct page
*page
, void *object
, void *addr
) { return 0; }
966 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
968 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
969 void *object
, int active
) { return 1; }
970 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
971 static inline void kmem_cache_open_debug_check(struct kmem_cache
*s
) {}
975 * Slab allocation and freeing
977 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
980 int pages
= 1 << s
->order
;
985 if (s
->flags
& SLAB_CACHE_DMA
)
989 page
= alloc_pages(flags
, s
->order
);
991 page
= alloc_pages_node(node
, flags
, s
->order
);
996 mod_zone_page_state(page_zone(page
),
997 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
998 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1004 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1007 setup_object_debug(s
, page
, object
);
1008 if (unlikely(s
->ctor
))
1009 s
->ctor(object
, s
, 0);
1012 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1015 struct kmem_cache_node
*n
;
1021 BUG_ON(flags
& ~(GFP_DMA
| GFP_LEVEL_MASK
));
1023 if (flags
& __GFP_WAIT
)
1026 page
= allocate_slab(s
, flags
& GFP_LEVEL_MASK
, node
);
1030 n
= get_node(s
, page_to_nid(page
));
1032 atomic_long_inc(&n
->nr_slabs
);
1033 page
->offset
= s
->offset
/ sizeof(void *);
1035 page
->flags
|= 1 << PG_slab
;
1036 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1037 SLAB_STORE_USER
| SLAB_TRACE
))
1040 start
= page_address(page
);
1041 end
= start
+ s
->objects
* s
->size
;
1043 if (unlikely(s
->flags
& SLAB_POISON
))
1044 memset(start
, POISON_INUSE
, PAGE_SIZE
<< s
->order
);
1047 for_each_object(p
, s
, start
) {
1048 setup_object(s
, page
, last
);
1049 set_freepointer(s
, last
, p
);
1052 setup_object(s
, page
, last
);
1053 set_freepointer(s
, last
, NULL
);
1055 page
->freelist
= start
;
1056 page
->lockless_freelist
= NULL
;
1059 if (flags
& __GFP_WAIT
)
1060 local_irq_disable();
1064 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1066 int pages
= 1 << s
->order
;
1068 if (unlikely(SlabDebug(page
))) {
1071 slab_pad_check(s
, page
);
1072 for_each_object(p
, s
, page_address(page
))
1073 check_object(s
, page
, p
, 0);
1076 mod_zone_page_state(page_zone(page
),
1077 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1078 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1081 page
->mapping
= NULL
;
1082 __free_pages(page
, s
->order
);
1085 static void rcu_free_slab(struct rcu_head
*h
)
1089 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1090 __free_slab(page
->slab
, page
);
1093 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1095 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1097 * RCU free overloads the RCU head over the LRU
1099 struct rcu_head
*head
= (void *)&page
->lru
;
1101 call_rcu(head
, rcu_free_slab
);
1103 __free_slab(s
, page
);
1106 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1108 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1110 atomic_long_dec(&n
->nr_slabs
);
1111 reset_page_mapcount(page
);
1112 ClearSlabDebug(page
);
1113 __ClearPageSlab(page
);
1118 * Per slab locking using the pagelock
1120 static __always_inline
void slab_lock(struct page
*page
)
1122 bit_spin_lock(PG_locked
, &page
->flags
);
1125 static __always_inline
void slab_unlock(struct page
*page
)
1127 bit_spin_unlock(PG_locked
, &page
->flags
);
1130 static __always_inline
int slab_trylock(struct page
*page
)
1134 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1139 * Management of partially allocated slabs
1141 static void add_partial_tail(struct kmem_cache_node
*n
, struct page
*page
)
1143 spin_lock(&n
->list_lock
);
1145 list_add_tail(&page
->lru
, &n
->partial
);
1146 spin_unlock(&n
->list_lock
);
1149 static void add_partial(struct kmem_cache_node
*n
, struct page
*page
)
1151 spin_lock(&n
->list_lock
);
1153 list_add(&page
->lru
, &n
->partial
);
1154 spin_unlock(&n
->list_lock
);
1157 static void remove_partial(struct kmem_cache
*s
,
1160 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1162 spin_lock(&n
->list_lock
);
1163 list_del(&page
->lru
);
1165 spin_unlock(&n
->list_lock
);
1169 * Lock slab and remove from the partial list.
1171 * Must hold list_lock.
1173 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
, struct page
*page
)
1175 if (slab_trylock(page
)) {
1176 list_del(&page
->lru
);
1178 SetSlabFrozen(page
);
1185 * Try to allocate a partial slab from a specific node.
1187 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1192 * Racy check. If we mistakenly see no partial slabs then we
1193 * just allocate an empty slab. If we mistakenly try to get a
1194 * partial slab and there is none available then get_partials()
1197 if (!n
|| !n
->nr_partial
)
1200 spin_lock(&n
->list_lock
);
1201 list_for_each_entry(page
, &n
->partial
, lru
)
1202 if (lock_and_freeze_slab(n
, page
))
1206 spin_unlock(&n
->list_lock
);
1211 * Get a page from somewhere. Search in increasing NUMA distances.
1213 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1216 struct zonelist
*zonelist
;
1221 * The defrag ratio allows a configuration of the tradeoffs between
1222 * inter node defragmentation and node local allocations. A lower
1223 * defrag_ratio increases the tendency to do local allocations
1224 * instead of attempting to obtain partial slabs from other nodes.
1226 * If the defrag_ratio is set to 0 then kmalloc() always
1227 * returns node local objects. If the ratio is higher then kmalloc()
1228 * may return off node objects because partial slabs are obtained
1229 * from other nodes and filled up.
1231 * If /sys/slab/xx/defrag_ratio is set to 100 (which makes
1232 * defrag_ratio = 1000) then every (well almost) allocation will
1233 * first attempt to defrag slab caches on other nodes. This means
1234 * scanning over all nodes to look for partial slabs which may be
1235 * expensive if we do it every time we are trying to find a slab
1236 * with available objects.
1238 if (!s
->defrag_ratio
|| get_cycles() % 1024 > s
->defrag_ratio
)
1241 zonelist
= &NODE_DATA(slab_node(current
->mempolicy
))
1242 ->node_zonelists
[gfp_zone(flags
)];
1243 for (z
= zonelist
->zones
; *z
; z
++) {
1244 struct kmem_cache_node
*n
;
1246 n
= get_node(s
, zone_to_nid(*z
));
1248 if (n
&& cpuset_zone_allowed_hardwall(*z
, flags
) &&
1249 n
->nr_partial
> MIN_PARTIAL
) {
1250 page
= get_partial_node(n
);
1260 * Get a partial page, lock it and return it.
1262 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1265 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1267 page
= get_partial_node(get_node(s
, searchnode
));
1268 if (page
|| (flags
& __GFP_THISNODE
))
1271 return get_any_partial(s
, flags
);
1275 * Move a page back to the lists.
1277 * Must be called with the slab lock held.
1279 * On exit the slab lock will have been dropped.
1281 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
)
1283 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1285 ClearSlabFrozen(page
);
1289 add_partial(n
, page
);
1290 else if (SlabDebug(page
) && (s
->flags
& SLAB_STORE_USER
))
1295 if (n
->nr_partial
< MIN_PARTIAL
) {
1297 * Adding an empty slab to the partial slabs in order
1298 * to avoid page allocator overhead. This slab needs
1299 * to come after the other slabs with objects in
1300 * order to fill them up. That way the size of the
1301 * partial list stays small. kmem_cache_shrink can
1302 * reclaim empty slabs from the partial list.
1304 add_partial_tail(n
, page
);
1308 discard_slab(s
, page
);
1314 * Remove the cpu slab
1316 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
, int cpu
)
1319 * Merge cpu freelist into freelist. Typically we get here
1320 * because both freelists are empty. So this is unlikely
1323 while (unlikely(page
->lockless_freelist
)) {
1326 /* Retrieve object from cpu_freelist */
1327 object
= page
->lockless_freelist
;
1328 page
->lockless_freelist
= page
->lockless_freelist
[page
->offset
];
1330 /* And put onto the regular freelist */
1331 object
[page
->offset
] = page
->freelist
;
1332 page
->freelist
= object
;
1335 s
->cpu_slab
[cpu
] = NULL
;
1336 unfreeze_slab(s
, page
);
1339 static void flush_slab(struct kmem_cache
*s
, struct page
*page
, int cpu
)
1342 deactivate_slab(s
, page
, cpu
);
1347 * Called from IPI handler with interrupts disabled.
1349 static void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1351 struct page
*page
= s
->cpu_slab
[cpu
];
1354 flush_slab(s
, page
, cpu
);
1357 static void flush_cpu_slab(void *d
)
1359 struct kmem_cache
*s
= d
;
1360 int cpu
= smp_processor_id();
1362 __flush_cpu_slab(s
, cpu
);
1365 static void flush_all(struct kmem_cache
*s
)
1368 on_each_cpu(flush_cpu_slab
, s
, 1, 1);
1370 unsigned long flags
;
1372 local_irq_save(flags
);
1374 local_irq_restore(flags
);
1379 * Slow path. The lockless freelist is empty or we need to perform
1382 * Interrupts are disabled.
1384 * Processing is still very fast if new objects have been freed to the
1385 * regular freelist. In that case we simply take over the regular freelist
1386 * as the lockless freelist and zap the regular freelist.
1388 * If that is not working then we fall back to the partial lists. We take the
1389 * first element of the freelist as the object to allocate now and move the
1390 * rest of the freelist to the lockless freelist.
1392 * And if we were unable to get a new slab from the partial slab lists then
1393 * we need to allocate a new slab. This is slowest path since we may sleep.
1395 static void *__slab_alloc(struct kmem_cache
*s
,
1396 gfp_t gfpflags
, int node
, void *addr
, struct page
*page
)
1399 int cpu
= smp_processor_id();
1405 if (unlikely(node
!= -1 && page_to_nid(page
) != node
))
1408 object
= page
->freelist
;
1409 if (unlikely(!object
))
1411 if (unlikely(SlabDebug(page
)))
1414 object
= page
->freelist
;
1415 page
->lockless_freelist
= object
[page
->offset
];
1416 page
->inuse
= s
->objects
;
1417 page
->freelist
= NULL
;
1422 deactivate_slab(s
, page
, cpu
);
1425 page
= get_partial(s
, gfpflags
, node
);
1427 s
->cpu_slab
[cpu
] = page
;
1431 page
= new_slab(s
, gfpflags
, node
);
1433 cpu
= smp_processor_id();
1434 if (s
->cpu_slab
[cpu
])
1435 flush_slab(s
, s
->cpu_slab
[cpu
], cpu
);
1437 SetSlabFrozen(page
);
1438 s
->cpu_slab
[cpu
] = page
;
1443 object
= page
->freelist
;
1444 if (!alloc_debug_processing(s
, page
, object
, addr
))
1448 page
->freelist
= object
[page
->offset
];
1454 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1455 * have the fastpath folded into their functions. So no function call
1456 * overhead for requests that can be satisfied on the fastpath.
1458 * The fastpath works by first checking if the lockless freelist can be used.
1459 * If not then __slab_alloc is called for slow processing.
1461 * Otherwise we can simply pick the next object from the lockless free list.
1463 static void __always_inline
*slab_alloc(struct kmem_cache
*s
,
1464 gfp_t gfpflags
, int node
, void *addr
)
1468 unsigned long flags
;
1470 local_irq_save(flags
);
1471 page
= s
->cpu_slab
[smp_processor_id()];
1472 if (unlikely(!page
|| !page
->lockless_freelist
||
1473 (node
!= -1 && page_to_nid(page
) != node
)))
1475 object
= __slab_alloc(s
, gfpflags
, node
, addr
, page
);
1478 object
= page
->lockless_freelist
;
1479 page
->lockless_freelist
= object
[page
->offset
];
1481 local_irq_restore(flags
);
1485 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1487 return slab_alloc(s
, gfpflags
, -1, __builtin_return_address(0));
1489 EXPORT_SYMBOL(kmem_cache_alloc
);
1492 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1494 return slab_alloc(s
, gfpflags
, node
, __builtin_return_address(0));
1496 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1500 * Slow patch handling. This may still be called frequently since objects
1501 * have a longer lifetime than the cpu slabs in most processing loads.
1503 * So we still attempt to reduce cache line usage. Just take the slab
1504 * lock and free the item. If there is no additional partial page
1505 * handling required then we can return immediately.
1507 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1508 void *x
, void *addr
)
1511 void **object
= (void *)x
;
1515 if (unlikely(SlabDebug(page
)))
1518 prior
= object
[page
->offset
] = page
->freelist
;
1519 page
->freelist
= object
;
1522 if (unlikely(SlabFrozen(page
)))
1525 if (unlikely(!page
->inuse
))
1529 * Objects left in the slab. If it
1530 * was not on the partial list before
1533 if (unlikely(!prior
))
1534 add_partial(get_node(s
, page_to_nid(page
)), page
);
1543 * Slab still on the partial list.
1545 remove_partial(s
, page
);
1548 discard_slab(s
, page
);
1552 if (!free_debug_processing(s
, page
, x
, addr
))
1558 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1559 * can perform fastpath freeing without additional function calls.
1561 * The fastpath is only possible if we are freeing to the current cpu slab
1562 * of this processor. This typically the case if we have just allocated
1565 * If fastpath is not possible then fall back to __slab_free where we deal
1566 * with all sorts of special processing.
1568 static void __always_inline
slab_free(struct kmem_cache
*s
,
1569 struct page
*page
, void *x
, void *addr
)
1571 void **object
= (void *)x
;
1572 unsigned long flags
;
1574 local_irq_save(flags
);
1575 if (likely(page
== s
->cpu_slab
[smp_processor_id()] &&
1576 !SlabDebug(page
))) {
1577 object
[page
->offset
] = page
->lockless_freelist
;
1578 page
->lockless_freelist
= object
;
1580 __slab_free(s
, page
, x
, addr
);
1582 local_irq_restore(flags
);
1585 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1589 page
= virt_to_head_page(x
);
1591 slab_free(s
, page
, x
, __builtin_return_address(0));
1593 EXPORT_SYMBOL(kmem_cache_free
);
1595 /* Figure out on which slab object the object resides */
1596 static struct page
*get_object_page(const void *x
)
1598 struct page
*page
= virt_to_head_page(x
);
1600 if (!PageSlab(page
))
1607 * Object placement in a slab is made very easy because we always start at
1608 * offset 0. If we tune the size of the object to the alignment then we can
1609 * get the required alignment by putting one properly sized object after
1612 * Notice that the allocation order determines the sizes of the per cpu
1613 * caches. Each processor has always one slab available for allocations.
1614 * Increasing the allocation order reduces the number of times that slabs
1615 * must be moved on and off the partial lists and is therefore a factor in
1620 * Mininum / Maximum order of slab pages. This influences locking overhead
1621 * and slab fragmentation. A higher order reduces the number of partial slabs
1622 * and increases the number of allocations possible without having to
1623 * take the list_lock.
1625 static int slub_min_order
;
1626 static int slub_max_order
= DEFAULT_MAX_ORDER
;
1627 static int slub_min_objects
= DEFAULT_MIN_OBJECTS
;
1630 * Merge control. If this is set then no merging of slab caches will occur.
1631 * (Could be removed. This was introduced to pacify the merge skeptics.)
1633 static int slub_nomerge
;
1636 * Calculate the order of allocation given an slab object size.
1638 * The order of allocation has significant impact on performance and other
1639 * system components. Generally order 0 allocations should be preferred since
1640 * order 0 does not cause fragmentation in the page allocator. Larger objects
1641 * be problematic to put into order 0 slabs because there may be too much
1642 * unused space left. We go to a higher order if more than 1/8th of the slab
1645 * In order to reach satisfactory performance we must ensure that a minimum
1646 * number of objects is in one slab. Otherwise we may generate too much
1647 * activity on the partial lists which requires taking the list_lock. This is
1648 * less a concern for large slabs though which are rarely used.
1650 * slub_max_order specifies the order where we begin to stop considering the
1651 * number of objects in a slab as critical. If we reach slub_max_order then
1652 * we try to keep the page order as low as possible. So we accept more waste
1653 * of space in favor of a small page order.
1655 * Higher order allocations also allow the placement of more objects in a
1656 * slab and thereby reduce object handling overhead. If the user has
1657 * requested a higher mininum order then we start with that one instead of
1658 * the smallest order which will fit the object.
1660 static inline int slab_order(int size
, int min_objects
,
1661 int max_order
, int fract_leftover
)
1666 for (order
= max(slub_min_order
,
1667 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1668 order
<= max_order
; order
++) {
1670 unsigned long slab_size
= PAGE_SIZE
<< order
;
1672 if (slab_size
< min_objects
* size
)
1675 rem
= slab_size
% size
;
1677 if (rem
<= slab_size
/ fract_leftover
)
1685 static inline int calculate_order(int size
)
1692 * Attempt to find best configuration for a slab. This
1693 * works by first attempting to generate a layout with
1694 * the best configuration and backing off gradually.
1696 * First we reduce the acceptable waste in a slab. Then
1697 * we reduce the minimum objects required in a slab.
1699 min_objects
= slub_min_objects
;
1700 while (min_objects
> 1) {
1702 while (fraction
>= 4) {
1703 order
= slab_order(size
, min_objects
,
1704 slub_max_order
, fraction
);
1705 if (order
<= slub_max_order
)
1713 * We were unable to place multiple objects in a slab. Now
1714 * lets see if we can place a single object there.
1716 order
= slab_order(size
, 1, slub_max_order
, 1);
1717 if (order
<= slub_max_order
)
1721 * Doh this slab cannot be placed using slub_max_order.
1723 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1724 if (order
<= MAX_ORDER
)
1730 * Figure out what the alignment of the objects will be.
1732 static unsigned long calculate_alignment(unsigned long flags
,
1733 unsigned long align
, unsigned long size
)
1736 * If the user wants hardware cache aligned objects then
1737 * follow that suggestion if the object is sufficiently
1740 * The hardware cache alignment cannot override the
1741 * specified alignment though. If that is greater
1744 if ((flags
& SLAB_HWCACHE_ALIGN
) &&
1745 size
> cache_line_size() / 2)
1746 return max_t(unsigned long, align
, cache_line_size());
1748 if (align
< ARCH_SLAB_MINALIGN
)
1749 return ARCH_SLAB_MINALIGN
;
1751 return ALIGN(align
, sizeof(void *));
1754 static void init_kmem_cache_node(struct kmem_cache_node
*n
)
1757 atomic_long_set(&n
->nr_slabs
, 0);
1758 spin_lock_init(&n
->list_lock
);
1759 INIT_LIST_HEAD(&n
->partial
);
1760 INIT_LIST_HEAD(&n
->full
);
1765 * No kmalloc_node yet so do it by hand. We know that this is the first
1766 * slab on the node for this slabcache. There are no concurrent accesses
1769 * Note that this function only works on the kmalloc_node_cache
1770 * when allocating for the kmalloc_node_cache.
1772 static struct kmem_cache_node
* __init
early_kmem_cache_node_alloc(gfp_t gfpflags
,
1776 struct kmem_cache_node
*n
;
1778 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
1780 page
= new_slab(kmalloc_caches
, gfpflags
| GFP_THISNODE
, node
);
1785 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
1787 kmalloc_caches
->node
[node
] = n
;
1788 setup_object_debug(kmalloc_caches
, page
, n
);
1789 init_kmem_cache_node(n
);
1790 atomic_long_inc(&n
->nr_slabs
);
1791 add_partial(n
, page
);
1794 * new_slab() disables interupts. If we do not reenable interrupts here
1795 * then bootup would continue with interrupts disabled.
1801 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
1805 for_each_online_node(node
) {
1806 struct kmem_cache_node
*n
= s
->node
[node
];
1807 if (n
&& n
!= &s
->local_node
)
1808 kmem_cache_free(kmalloc_caches
, n
);
1809 s
->node
[node
] = NULL
;
1813 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
1818 if (slab_state
>= UP
)
1819 local_node
= page_to_nid(virt_to_page(s
));
1823 for_each_online_node(node
) {
1824 struct kmem_cache_node
*n
;
1826 if (local_node
== node
)
1829 if (slab_state
== DOWN
) {
1830 n
= early_kmem_cache_node_alloc(gfpflags
,
1834 n
= kmem_cache_alloc_node(kmalloc_caches
,
1838 free_kmem_cache_nodes(s
);
1844 init_kmem_cache_node(n
);
1849 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
1853 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
1855 init_kmem_cache_node(&s
->local_node
);
1861 * calculate_sizes() determines the order and the distribution of data within
1864 static int calculate_sizes(struct kmem_cache
*s
)
1866 unsigned long flags
= s
->flags
;
1867 unsigned long size
= s
->objsize
;
1868 unsigned long align
= s
->align
;
1871 * Determine if we can poison the object itself. If the user of
1872 * the slab may touch the object after free or before allocation
1873 * then we should never poison the object itself.
1875 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
1877 s
->flags
|= __OBJECT_POISON
;
1879 s
->flags
&= ~__OBJECT_POISON
;
1882 * Round up object size to the next word boundary. We can only
1883 * place the free pointer at word boundaries and this determines
1884 * the possible location of the free pointer.
1886 size
= ALIGN(size
, sizeof(void *));
1888 #ifdef CONFIG_SLUB_DEBUG
1890 * If we are Redzoning then check if there is some space between the
1891 * end of the object and the free pointer. If not then add an
1892 * additional word to have some bytes to store Redzone information.
1894 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
1895 size
+= sizeof(void *);
1899 * With that we have determined the number of bytes in actual use
1900 * by the object. This is the potential offset to the free pointer.
1904 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
1907 * Relocate free pointer after the object if it is not
1908 * permitted to overwrite the first word of the object on
1911 * This is the case if we do RCU, have a constructor or
1912 * destructor or are poisoning the objects.
1915 size
+= sizeof(void *);
1918 #ifdef CONFIG_SLUB_DEBUG
1919 if (flags
& SLAB_STORE_USER
)
1921 * Need to store information about allocs and frees after
1924 size
+= 2 * sizeof(struct track
);
1926 if (flags
& SLAB_RED_ZONE
)
1928 * Add some empty padding so that we can catch
1929 * overwrites from earlier objects rather than let
1930 * tracking information or the free pointer be
1931 * corrupted if an user writes before the start
1934 size
+= sizeof(void *);
1938 * Determine the alignment based on various parameters that the
1939 * user specified and the dynamic determination of cache line size
1942 align
= calculate_alignment(flags
, align
, s
->objsize
);
1945 * SLUB stores one object immediately after another beginning from
1946 * offset 0. In order to align the objects we have to simply size
1947 * each object to conform to the alignment.
1949 size
= ALIGN(size
, align
);
1952 s
->order
= calculate_order(size
);
1957 * Determine the number of objects per slab
1959 s
->objects
= (PAGE_SIZE
<< s
->order
) / size
;
1962 * Verify that the number of objects is within permitted limits.
1963 * The page->inuse field is only 16 bit wide! So we cannot have
1964 * more than 64k objects per slab.
1966 if (!s
->objects
|| s
->objects
> 65535)
1972 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
1973 const char *name
, size_t size
,
1974 size_t align
, unsigned long flags
,
1975 void (*ctor
)(void *, struct kmem_cache
*, unsigned long))
1977 memset(s
, 0, kmem_size
);
1983 kmem_cache_open_debug_check(s
);
1985 if (!calculate_sizes(s
))
1990 s
->defrag_ratio
= 100;
1993 if (init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
1996 if (flags
& SLAB_PANIC
)
1997 panic("Cannot create slab %s size=%lu realsize=%u "
1998 "order=%u offset=%u flags=%lx\n",
1999 s
->name
, (unsigned long)size
, s
->size
, s
->order
,
2005 * Check if a given pointer is valid
2007 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2011 page
= get_object_page(object
);
2013 if (!page
|| s
!= page
->slab
)
2014 /* No slab or wrong slab */
2017 if (!check_valid_pointer(s
, page
, object
))
2021 * We could also check if the object is on the slabs freelist.
2022 * But this would be too expensive and it seems that the main
2023 * purpose of kmem_ptr_valid is to check if the object belongs
2024 * to a certain slab.
2028 EXPORT_SYMBOL(kmem_ptr_validate
);
2031 * Determine the size of a slab object
2033 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2037 EXPORT_SYMBOL(kmem_cache_size
);
2039 const char *kmem_cache_name(struct kmem_cache
*s
)
2043 EXPORT_SYMBOL(kmem_cache_name
);
2046 * Attempt to free all slabs on a node. Return the number of slabs we
2047 * were unable to free.
2049 static int free_list(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
2050 struct list_head
*list
)
2052 int slabs_inuse
= 0;
2053 unsigned long flags
;
2054 struct page
*page
, *h
;
2056 spin_lock_irqsave(&n
->list_lock
, flags
);
2057 list_for_each_entry_safe(page
, h
, list
, lru
)
2059 list_del(&page
->lru
);
2060 discard_slab(s
, page
);
2063 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2068 * Release all resources used by a slab cache.
2070 static int kmem_cache_close(struct kmem_cache
*s
)
2076 /* Attempt to free all objects */
2077 for_each_online_node(node
) {
2078 struct kmem_cache_node
*n
= get_node(s
, node
);
2080 n
->nr_partial
-= free_list(s
, n
, &n
->partial
);
2081 if (atomic_long_read(&n
->nr_slabs
))
2084 free_kmem_cache_nodes(s
);
2089 * Close a cache and release the kmem_cache structure
2090 * (must be used for caches created using kmem_cache_create)
2092 void kmem_cache_destroy(struct kmem_cache
*s
)
2094 down_write(&slub_lock
);
2098 if (kmem_cache_close(s
))
2100 sysfs_slab_remove(s
);
2103 up_write(&slub_lock
);
2105 EXPORT_SYMBOL(kmem_cache_destroy
);
2107 /********************************************************************
2109 *******************************************************************/
2111 struct kmem_cache kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1] __cacheline_aligned
;
2112 EXPORT_SYMBOL(kmalloc_caches
);
2114 #ifdef CONFIG_ZONE_DMA
2115 static struct kmem_cache
*kmalloc_caches_dma
[KMALLOC_SHIFT_HIGH
+ 1];
2118 static int __init
setup_slub_min_order(char *str
)
2120 get_option (&str
, &slub_min_order
);
2125 __setup("slub_min_order=", setup_slub_min_order
);
2127 static int __init
setup_slub_max_order(char *str
)
2129 get_option (&str
, &slub_max_order
);
2134 __setup("slub_max_order=", setup_slub_max_order
);
2136 static int __init
setup_slub_min_objects(char *str
)
2138 get_option (&str
, &slub_min_objects
);
2143 __setup("slub_min_objects=", setup_slub_min_objects
);
2145 static int __init
setup_slub_nomerge(char *str
)
2151 __setup("slub_nomerge", setup_slub_nomerge
);
2153 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2154 const char *name
, int size
, gfp_t gfp_flags
)
2156 unsigned int flags
= 0;
2158 if (gfp_flags
& SLUB_DMA
)
2159 flags
= SLAB_CACHE_DMA
;
2161 down_write(&slub_lock
);
2162 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2166 list_add(&s
->list
, &slab_caches
);
2167 up_write(&slub_lock
);
2168 if (sysfs_slab_add(s
))
2173 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2176 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2178 int index
= kmalloc_index(size
);
2183 /* Allocation too large? */
2186 #ifdef CONFIG_ZONE_DMA
2187 if ((flags
& SLUB_DMA
)) {
2188 struct kmem_cache
*s
;
2189 struct kmem_cache
*x
;
2193 s
= kmalloc_caches_dma
[index
];
2197 /* Dynamically create dma cache */
2198 x
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2200 panic("Unable to allocate memory for dma cache\n");
2202 if (index
<= KMALLOC_SHIFT_HIGH
)
2203 realsize
= 1 << index
;
2211 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2212 (unsigned int)realsize
);
2213 s
= create_kmalloc_cache(x
, text
, realsize
, flags
);
2214 kmalloc_caches_dma
[index
] = s
;
2218 return &kmalloc_caches
[index
];
2221 void *__kmalloc(size_t size
, gfp_t flags
)
2223 struct kmem_cache
*s
= get_slab(size
, flags
);
2226 return slab_alloc(s
, flags
, -1, __builtin_return_address(0));
2227 return ZERO_SIZE_PTR
;
2229 EXPORT_SYMBOL(__kmalloc
);
2232 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2234 struct kmem_cache
*s
= get_slab(size
, flags
);
2237 return slab_alloc(s
, flags
, node
, __builtin_return_address(0));
2238 return ZERO_SIZE_PTR
;
2240 EXPORT_SYMBOL(__kmalloc_node
);
2243 size_t ksize(const void *object
)
2246 struct kmem_cache
*s
;
2248 if (object
== ZERO_SIZE_PTR
)
2251 page
= get_object_page(object
);
2257 * Debugging requires use of the padding between object
2258 * and whatever may come after it.
2260 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2264 * If we have the need to store the freelist pointer
2265 * back there or track user information then we can
2266 * only use the space before that information.
2268 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2272 * Else we can use all the padding etc for the allocation
2276 EXPORT_SYMBOL(ksize
);
2278 void kfree(const void *x
)
2280 struct kmem_cache
*s
;
2284 * This has to be an unsigned comparison. According to Linus
2285 * some gcc version treat a pointer as a signed entity. Then
2286 * this comparison would be true for all "negative" pointers
2287 * (which would cover the whole upper half of the address space).
2289 if ((unsigned long)x
<= (unsigned long)ZERO_SIZE_PTR
)
2292 page
= virt_to_head_page(x
);
2295 slab_free(s
, page
, (void *)x
, __builtin_return_address(0));
2297 EXPORT_SYMBOL(kfree
);
2300 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2301 * the remaining slabs by the number of items in use. The slabs with the
2302 * most items in use come first. New allocations will then fill those up
2303 * and thus they can be removed from the partial lists.
2305 * The slabs with the least items are placed last. This results in them
2306 * being allocated from last increasing the chance that the last objects
2307 * are freed in them.
2309 int kmem_cache_shrink(struct kmem_cache
*s
)
2313 struct kmem_cache_node
*n
;
2316 struct list_head
*slabs_by_inuse
=
2317 kmalloc(sizeof(struct list_head
) * s
->objects
, GFP_KERNEL
);
2318 unsigned long flags
;
2320 if (!slabs_by_inuse
)
2324 for_each_online_node(node
) {
2325 n
= get_node(s
, node
);
2330 for (i
= 0; i
< s
->objects
; i
++)
2331 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2333 spin_lock_irqsave(&n
->list_lock
, flags
);
2336 * Build lists indexed by the items in use in each slab.
2338 * Note that concurrent frees may occur while we hold the
2339 * list_lock. page->inuse here is the upper limit.
2341 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2342 if (!page
->inuse
&& slab_trylock(page
)) {
2344 * Must hold slab lock here because slab_free
2345 * may have freed the last object and be
2346 * waiting to release the slab.
2348 list_del(&page
->lru
);
2351 discard_slab(s
, page
);
2353 if (n
->nr_partial
> MAX_PARTIAL
)
2354 list_move(&page
->lru
,
2355 slabs_by_inuse
+ page
->inuse
);
2359 if (n
->nr_partial
<= MAX_PARTIAL
)
2363 * Rebuild the partial list with the slabs filled up most
2364 * first and the least used slabs at the end.
2366 for (i
= s
->objects
- 1; i
>= 0; i
--)
2367 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2370 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2373 kfree(slabs_by_inuse
);
2376 EXPORT_SYMBOL(kmem_cache_shrink
);
2379 * krealloc - reallocate memory. The contents will remain unchanged.
2380 * @p: object to reallocate memory for.
2381 * @new_size: how many bytes of memory are required.
2382 * @flags: the type of memory to allocate.
2384 * The contents of the object pointed to are preserved up to the
2385 * lesser of the new and old sizes. If @p is %NULL, krealloc()
2386 * behaves exactly like kmalloc(). If @size is 0 and @p is not a
2387 * %NULL pointer, the object pointed to is freed.
2389 void *krealloc(const void *p
, size_t new_size
, gfp_t flags
)
2394 if (unlikely(!p
|| p
== ZERO_SIZE_PTR
))
2395 return kmalloc(new_size
, flags
);
2397 if (unlikely(!new_size
)) {
2399 return ZERO_SIZE_PTR
;
2406 ret
= kmalloc(new_size
, flags
);
2408 memcpy(ret
, p
, min(new_size
, ks
));
2413 EXPORT_SYMBOL(krealloc
);
2415 /********************************************************************
2416 * Basic setup of slabs
2417 *******************************************************************/
2419 void __init
kmem_cache_init(void)
2426 * Must first have the slab cache available for the allocations of the
2427 * struct kmem_cache_node's. There is special bootstrap code in
2428 * kmem_cache_open for slab_state == DOWN.
2430 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
2431 sizeof(struct kmem_cache_node
), GFP_KERNEL
);
2432 kmalloc_caches
[0].refcount
= -1;
2436 /* Able to allocate the per node structures */
2437 slab_state
= PARTIAL
;
2439 /* Caches that are not of the two-to-the-power-of size */
2440 if (KMALLOC_MIN_SIZE
<= 64) {
2441 create_kmalloc_cache(&kmalloc_caches
[1],
2442 "kmalloc-96", 96, GFP_KERNEL
);
2445 if (KMALLOC_MIN_SIZE
<= 128) {
2446 create_kmalloc_cache(&kmalloc_caches
[2],
2447 "kmalloc-192", 192, GFP_KERNEL
);
2451 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++) {
2452 create_kmalloc_cache(&kmalloc_caches
[i
],
2453 "kmalloc", 1 << i
, GFP_KERNEL
);
2459 /* Provide the correct kmalloc names now that the caches are up */
2460 for (i
= KMALLOC_SHIFT_LOW
; i
<= KMALLOC_SHIFT_HIGH
; i
++)
2461 kmalloc_caches
[i
]. name
=
2462 kasprintf(GFP_KERNEL
, "kmalloc-%d", 1 << i
);
2465 register_cpu_notifier(&slab_notifier
);
2468 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
2469 nr_cpu_ids
* sizeof(struct page
*);
2471 printk(KERN_INFO
"SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
2472 " CPUs=%d, Nodes=%d\n",
2473 caches
, cache_line_size(),
2474 slub_min_order
, slub_max_order
, slub_min_objects
,
2475 nr_cpu_ids
, nr_node_ids
);
2479 * Find a mergeable slab cache
2481 static int slab_unmergeable(struct kmem_cache
*s
)
2483 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
2490 * We may have set a slab to be unmergeable during bootstrap.
2492 if (s
->refcount
< 0)
2498 static struct kmem_cache
*find_mergeable(size_t size
,
2499 size_t align
, unsigned long flags
,
2500 void (*ctor
)(void *, struct kmem_cache
*, unsigned long))
2502 struct list_head
*h
;
2504 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
2510 size
= ALIGN(size
, sizeof(void *));
2511 align
= calculate_alignment(flags
, align
, size
);
2512 size
= ALIGN(size
, align
);
2514 list_for_each(h
, &slab_caches
) {
2515 struct kmem_cache
*s
=
2516 container_of(h
, struct kmem_cache
, list
);
2518 if (slab_unmergeable(s
))
2524 if (((flags
| slub_debug
) & SLUB_MERGE_SAME
) !=
2525 (s
->flags
& SLUB_MERGE_SAME
))
2528 * Check if alignment is compatible.
2529 * Courtesy of Adrian Drzewiecki
2531 if ((s
->size
& ~(align
-1)) != s
->size
)
2534 if (s
->size
- size
>= sizeof(void *))
2542 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
2543 size_t align
, unsigned long flags
,
2544 void (*ctor
)(void *, struct kmem_cache
*, unsigned long),
2545 void (*dtor
)(void *, struct kmem_cache
*, unsigned long))
2547 struct kmem_cache
*s
;
2550 down_write(&slub_lock
);
2551 s
= find_mergeable(size
, align
, flags
, ctor
);
2555 * Adjust the object sizes so that we clear
2556 * the complete object on kzalloc.
2558 s
->objsize
= max(s
->objsize
, (int)size
);
2559 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
2560 if (sysfs_slab_alias(s
, name
))
2563 s
= kmalloc(kmem_size
, GFP_KERNEL
);
2564 if (s
&& kmem_cache_open(s
, GFP_KERNEL
, name
,
2565 size
, align
, flags
, ctor
)) {
2566 if (sysfs_slab_add(s
)) {
2570 list_add(&s
->list
, &slab_caches
);
2574 up_write(&slub_lock
);
2578 up_write(&slub_lock
);
2579 if (flags
& SLAB_PANIC
)
2580 panic("Cannot create slabcache %s\n", name
);
2585 EXPORT_SYMBOL(kmem_cache_create
);
2587 void *kmem_cache_zalloc(struct kmem_cache
*s
, gfp_t flags
)
2591 x
= slab_alloc(s
, flags
, -1, __builtin_return_address(0));
2593 memset(x
, 0, s
->objsize
);
2596 EXPORT_SYMBOL(kmem_cache_zalloc
);
2599 static void for_all_slabs(void (*func
)(struct kmem_cache
*, int), int cpu
)
2601 struct list_head
*h
;
2603 down_read(&slub_lock
);
2604 list_for_each(h
, &slab_caches
) {
2605 struct kmem_cache
*s
=
2606 container_of(h
, struct kmem_cache
, list
);
2610 up_read(&slub_lock
);
2614 * Version of __flush_cpu_slab for the case that interrupts
2617 static void cpu_slab_flush(struct kmem_cache
*s
, int cpu
)
2619 unsigned long flags
;
2621 local_irq_save(flags
);
2622 __flush_cpu_slab(s
, cpu
);
2623 local_irq_restore(flags
);
2627 * Use the cpu notifier to insure that the cpu slabs are flushed when
2630 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
2631 unsigned long action
, void *hcpu
)
2633 long cpu
= (long)hcpu
;
2636 case CPU_UP_CANCELED
:
2637 case CPU_UP_CANCELED_FROZEN
:
2639 case CPU_DEAD_FROZEN
:
2640 for_all_slabs(cpu_slab_flush
, cpu
);
2648 static struct notifier_block __cpuinitdata slab_notifier
=
2649 { &slab_cpuup_callback
, NULL
, 0 };
2653 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, void *caller
)
2655 struct kmem_cache
*s
= get_slab(size
, gfpflags
);
2658 return ZERO_SIZE_PTR
;
2660 return slab_alloc(s
, gfpflags
, -1, caller
);
2663 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
2664 int node
, void *caller
)
2666 struct kmem_cache
*s
= get_slab(size
, gfpflags
);
2669 return ZERO_SIZE_PTR
;
2671 return slab_alloc(s
, gfpflags
, node
, caller
);
2674 #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)
2675 static int validate_slab(struct kmem_cache
*s
, struct page
*page
)
2678 void *addr
= page_address(page
);
2679 DECLARE_BITMAP(map
, s
->objects
);
2681 if (!check_slab(s
, page
) ||
2682 !on_freelist(s
, page
, NULL
))
2685 /* Now we know that a valid freelist exists */
2686 bitmap_zero(map
, s
->objects
);
2688 for_each_free_object(p
, s
, page
->freelist
) {
2689 set_bit(slab_index(p
, s
, addr
), map
);
2690 if (!check_object(s
, page
, p
, 0))
2694 for_each_object(p
, s
, addr
)
2695 if (!test_bit(slab_index(p
, s
, addr
), map
))
2696 if (!check_object(s
, page
, p
, 1))
2701 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
)
2703 if (slab_trylock(page
)) {
2704 validate_slab(s
, page
);
2707 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
2710 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
2711 if (!SlabDebug(page
))
2712 printk(KERN_ERR
"SLUB %s: SlabDebug not set "
2713 "on slab 0x%p\n", s
->name
, page
);
2715 if (SlabDebug(page
))
2716 printk(KERN_ERR
"SLUB %s: SlabDebug set on "
2717 "slab 0x%p\n", s
->name
, page
);
2721 static int validate_slab_node(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2723 unsigned long count
= 0;
2725 unsigned long flags
;
2727 spin_lock_irqsave(&n
->list_lock
, flags
);
2729 list_for_each_entry(page
, &n
->partial
, lru
) {
2730 validate_slab_slab(s
, page
);
2733 if (count
!= n
->nr_partial
)
2734 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
2735 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
2737 if (!(s
->flags
& SLAB_STORE_USER
))
2740 list_for_each_entry(page
, &n
->full
, lru
) {
2741 validate_slab_slab(s
, page
);
2744 if (count
!= atomic_long_read(&n
->nr_slabs
))
2745 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
2746 "counter=%ld\n", s
->name
, count
,
2747 atomic_long_read(&n
->nr_slabs
));
2750 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2754 static unsigned long validate_slab_cache(struct kmem_cache
*s
)
2757 unsigned long count
= 0;
2760 for_each_online_node(node
) {
2761 struct kmem_cache_node
*n
= get_node(s
, node
);
2763 count
+= validate_slab_node(s
, n
);
2768 #ifdef SLUB_RESILIENCY_TEST
2769 static void resiliency_test(void)
2773 printk(KERN_ERR
"SLUB resiliency testing\n");
2774 printk(KERN_ERR
"-----------------------\n");
2775 printk(KERN_ERR
"A. Corruption after allocation\n");
2777 p
= kzalloc(16, GFP_KERNEL
);
2779 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
2780 " 0x12->0x%p\n\n", p
+ 16);
2782 validate_slab_cache(kmalloc_caches
+ 4);
2784 /* Hmmm... The next two are dangerous */
2785 p
= kzalloc(32, GFP_KERNEL
);
2786 p
[32 + sizeof(void *)] = 0x34;
2787 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
2788 " 0x34 -> -0x%p\n", p
);
2789 printk(KERN_ERR
"If allocated object is overwritten then not detectable\n\n");
2791 validate_slab_cache(kmalloc_caches
+ 5);
2792 p
= kzalloc(64, GFP_KERNEL
);
2793 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
2795 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
2797 printk(KERN_ERR
"If allocated object is overwritten then not detectable\n\n");
2798 validate_slab_cache(kmalloc_caches
+ 6);
2800 printk(KERN_ERR
"\nB. Corruption after free\n");
2801 p
= kzalloc(128, GFP_KERNEL
);
2804 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
2805 validate_slab_cache(kmalloc_caches
+ 7);
2807 p
= kzalloc(256, GFP_KERNEL
);
2810 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p
);
2811 validate_slab_cache(kmalloc_caches
+ 8);
2813 p
= kzalloc(512, GFP_KERNEL
);
2816 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
2817 validate_slab_cache(kmalloc_caches
+ 9);
2820 static void resiliency_test(void) {};
2824 * Generate lists of code addresses where slabcache objects are allocated
2829 unsigned long count
;
2842 unsigned long count
;
2843 struct location
*loc
;
2846 static void free_loc_track(struct loc_track
*t
)
2849 free_pages((unsigned long)t
->loc
,
2850 get_order(sizeof(struct location
) * t
->max
));
2853 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
)
2859 max
= PAGE_SIZE
/ sizeof(struct location
);
2861 order
= get_order(sizeof(struct location
) * max
);
2863 l
= (void *)__get_free_pages(GFP_ATOMIC
, order
);
2869 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
2877 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
2878 const struct track
*track
)
2880 long start
, end
, pos
;
2883 unsigned long age
= jiffies
- track
->when
;
2889 pos
= start
+ (end
- start
+ 1) / 2;
2892 * There is nothing at "end". If we end up there
2893 * we need to add something to before end.
2898 caddr
= t
->loc
[pos
].addr
;
2899 if (track
->addr
== caddr
) {
2905 if (age
< l
->min_time
)
2907 if (age
> l
->max_time
)
2910 if (track
->pid
< l
->min_pid
)
2911 l
->min_pid
= track
->pid
;
2912 if (track
->pid
> l
->max_pid
)
2913 l
->max_pid
= track
->pid
;
2915 cpu_set(track
->cpu
, l
->cpus
);
2917 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
2921 if (track
->addr
< caddr
)
2928 * Not found. Insert new tracking element.
2930 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
))
2936 (t
->count
- pos
) * sizeof(struct location
));
2939 l
->addr
= track
->addr
;
2943 l
->min_pid
= track
->pid
;
2944 l
->max_pid
= track
->pid
;
2945 cpus_clear(l
->cpus
);
2946 cpu_set(track
->cpu
, l
->cpus
);
2947 nodes_clear(l
->nodes
);
2948 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
2952 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
2953 struct page
*page
, enum track_item alloc
)
2955 void *addr
= page_address(page
);
2956 DECLARE_BITMAP(map
, s
->objects
);
2959 bitmap_zero(map
, s
->objects
);
2960 for_each_free_object(p
, s
, page
->freelist
)
2961 set_bit(slab_index(p
, s
, addr
), map
);
2963 for_each_object(p
, s
, addr
)
2964 if (!test_bit(slab_index(p
, s
, addr
), map
))
2965 add_location(t
, s
, get_track(s
, p
, alloc
));
2968 static int list_locations(struct kmem_cache
*s
, char *buf
,
2969 enum track_item alloc
)
2979 /* Push back cpu slabs */
2982 for_each_online_node(node
) {
2983 struct kmem_cache_node
*n
= get_node(s
, node
);
2984 unsigned long flags
;
2987 if (!atomic_read(&n
->nr_slabs
))
2990 spin_lock_irqsave(&n
->list_lock
, flags
);
2991 list_for_each_entry(page
, &n
->partial
, lru
)
2992 process_slab(&t
, s
, page
, alloc
);
2993 list_for_each_entry(page
, &n
->full
, lru
)
2994 process_slab(&t
, s
, page
, alloc
);
2995 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2998 for (i
= 0; i
< t
.count
; i
++) {
2999 struct location
*l
= &t
.loc
[i
];
3001 if (n
> PAGE_SIZE
- 100)
3003 n
+= sprintf(buf
+ n
, "%7ld ", l
->count
);
3006 n
+= sprint_symbol(buf
+ n
, (unsigned long)l
->addr
);
3008 n
+= sprintf(buf
+ n
, "<not-available>");
3010 if (l
->sum_time
!= l
->min_time
) {
3011 unsigned long remainder
;
3013 n
+= sprintf(buf
+ n
, " age=%ld/%ld/%ld",
3015 div_long_long_rem(l
->sum_time
, l
->count
, &remainder
),
3018 n
+= sprintf(buf
+ n
, " age=%ld",
3021 if (l
->min_pid
!= l
->max_pid
)
3022 n
+= sprintf(buf
+ n
, " pid=%ld-%ld",
3023 l
->min_pid
, l
->max_pid
);
3025 n
+= sprintf(buf
+ n
, " pid=%ld",
3028 if (num_online_cpus() > 1 && !cpus_empty(l
->cpus
) &&
3029 n
< PAGE_SIZE
- 60) {
3030 n
+= sprintf(buf
+ n
, " cpus=");
3031 n
+= cpulist_scnprintf(buf
+ n
, PAGE_SIZE
- n
- 50,
3035 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3036 n
< PAGE_SIZE
- 60) {
3037 n
+= sprintf(buf
+ n
, " nodes=");
3038 n
+= nodelist_scnprintf(buf
+ n
, PAGE_SIZE
- n
- 50,
3042 n
+= sprintf(buf
+ n
, "\n");
3047 n
+= sprintf(buf
, "No data\n");
3051 static unsigned long count_partial(struct kmem_cache_node
*n
)
3053 unsigned long flags
;
3054 unsigned long x
= 0;
3057 spin_lock_irqsave(&n
->list_lock
, flags
);
3058 list_for_each_entry(page
, &n
->partial
, lru
)
3060 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3064 enum slab_stat_type
{
3071 #define SO_FULL (1 << SL_FULL)
3072 #define SO_PARTIAL (1 << SL_PARTIAL)
3073 #define SO_CPU (1 << SL_CPU)
3074 #define SO_OBJECTS (1 << SL_OBJECTS)
3076 static unsigned long slab_objects(struct kmem_cache
*s
,
3077 char *buf
, unsigned long flags
)
3079 unsigned long total
= 0;
3083 unsigned long *nodes
;
3084 unsigned long *per_cpu
;
3086 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3087 per_cpu
= nodes
+ nr_node_ids
;
3089 for_each_possible_cpu(cpu
) {
3090 struct page
*page
= s
->cpu_slab
[cpu
];
3094 node
= page_to_nid(page
);
3095 if (flags
& SO_CPU
) {
3098 if (flags
& SO_OBJECTS
)
3109 for_each_online_node(node
) {
3110 struct kmem_cache_node
*n
= get_node(s
, node
);
3112 if (flags
& SO_PARTIAL
) {
3113 if (flags
& SO_OBJECTS
)
3114 x
= count_partial(n
);
3121 if (flags
& SO_FULL
) {
3122 int full_slabs
= atomic_read(&n
->nr_slabs
)
3126 if (flags
& SO_OBJECTS
)
3127 x
= full_slabs
* s
->objects
;
3135 x
= sprintf(buf
, "%lu", total
);
3137 for_each_online_node(node
)
3139 x
+= sprintf(buf
+ x
, " N%d=%lu",
3143 return x
+ sprintf(buf
+ x
, "\n");
3146 static int any_slab_objects(struct kmem_cache
*s
)
3151 for_each_possible_cpu(cpu
)
3152 if (s
->cpu_slab
[cpu
])
3155 for_each_node(node
) {
3156 struct kmem_cache_node
*n
= get_node(s
, node
);
3158 if (n
->nr_partial
|| atomic_read(&n
->nr_slabs
))
3164 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3165 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3167 struct slab_attribute
{
3168 struct attribute attr
;
3169 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3170 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3173 #define SLAB_ATTR_RO(_name) \
3174 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3176 #define SLAB_ATTR(_name) \
3177 static struct slab_attribute _name##_attr = \
3178 __ATTR(_name, 0644, _name##_show, _name##_store)
3180 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3182 return sprintf(buf
, "%d\n", s
->size
);
3184 SLAB_ATTR_RO(slab_size
);
3186 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3188 return sprintf(buf
, "%d\n", s
->align
);
3190 SLAB_ATTR_RO(align
);
3192 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3194 return sprintf(buf
, "%d\n", s
->objsize
);
3196 SLAB_ATTR_RO(object_size
);
3198 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3200 return sprintf(buf
, "%d\n", s
->objects
);
3202 SLAB_ATTR_RO(objs_per_slab
);
3204 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3206 return sprintf(buf
, "%d\n", s
->order
);
3208 SLAB_ATTR_RO(order
);
3210 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3213 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3215 return n
+ sprintf(buf
+ n
, "\n");
3221 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3223 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3225 SLAB_ATTR_RO(aliases
);
3227 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3229 return slab_objects(s
, buf
, SO_FULL
|SO_PARTIAL
|SO_CPU
);
3231 SLAB_ATTR_RO(slabs
);
3233 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3235 return slab_objects(s
, buf
, SO_PARTIAL
);
3237 SLAB_ATTR_RO(partial
);
3239 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3241 return slab_objects(s
, buf
, SO_CPU
);
3243 SLAB_ATTR_RO(cpu_slabs
);
3245 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3247 return slab_objects(s
, buf
, SO_FULL
|SO_PARTIAL
|SO_CPU
|SO_OBJECTS
);
3249 SLAB_ATTR_RO(objects
);
3251 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3253 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3256 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3257 const char *buf
, size_t length
)
3259 s
->flags
&= ~SLAB_DEBUG_FREE
;
3261 s
->flags
|= SLAB_DEBUG_FREE
;
3264 SLAB_ATTR(sanity_checks
);
3266 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
3268 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
3271 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
3274 s
->flags
&= ~SLAB_TRACE
;
3276 s
->flags
|= SLAB_TRACE
;
3281 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
3283 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
3286 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
3287 const char *buf
, size_t length
)
3289 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
3291 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
3294 SLAB_ATTR(reclaim_account
);
3296 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
3298 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
3300 SLAB_ATTR_RO(hwcache_align
);
3302 #ifdef CONFIG_ZONE_DMA
3303 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
3305 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
3307 SLAB_ATTR_RO(cache_dma
);
3310 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
3312 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
3314 SLAB_ATTR_RO(destroy_by_rcu
);
3316 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
3318 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
3321 static ssize_t
red_zone_store(struct kmem_cache
*s
,
3322 const char *buf
, size_t length
)
3324 if (any_slab_objects(s
))
3327 s
->flags
&= ~SLAB_RED_ZONE
;
3329 s
->flags
|= SLAB_RED_ZONE
;
3333 SLAB_ATTR(red_zone
);
3335 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
3337 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
3340 static ssize_t
poison_store(struct kmem_cache
*s
,
3341 const char *buf
, size_t length
)
3343 if (any_slab_objects(s
))
3346 s
->flags
&= ~SLAB_POISON
;
3348 s
->flags
|= SLAB_POISON
;
3354 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
3356 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
3359 static ssize_t
store_user_store(struct kmem_cache
*s
,
3360 const char *buf
, size_t length
)
3362 if (any_slab_objects(s
))
3365 s
->flags
&= ~SLAB_STORE_USER
;
3367 s
->flags
|= SLAB_STORE_USER
;
3371 SLAB_ATTR(store_user
);
3373 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
3378 static ssize_t
validate_store(struct kmem_cache
*s
,
3379 const char *buf
, size_t length
)
3382 validate_slab_cache(s
);
3387 SLAB_ATTR(validate
);
3389 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
3394 static ssize_t
shrink_store(struct kmem_cache
*s
,
3395 const char *buf
, size_t length
)
3397 if (buf
[0] == '1') {
3398 int rc
= kmem_cache_shrink(s
);
3408 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
3410 if (!(s
->flags
& SLAB_STORE_USER
))
3412 return list_locations(s
, buf
, TRACK_ALLOC
);
3414 SLAB_ATTR_RO(alloc_calls
);
3416 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
3418 if (!(s
->flags
& SLAB_STORE_USER
))
3420 return list_locations(s
, buf
, TRACK_FREE
);
3422 SLAB_ATTR_RO(free_calls
);
3425 static ssize_t
defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
3427 return sprintf(buf
, "%d\n", s
->defrag_ratio
/ 10);
3430 static ssize_t
defrag_ratio_store(struct kmem_cache
*s
,
3431 const char *buf
, size_t length
)
3433 int n
= simple_strtoul(buf
, NULL
, 10);
3436 s
->defrag_ratio
= n
* 10;
3439 SLAB_ATTR(defrag_ratio
);
3442 static struct attribute
* slab_attrs
[] = {
3443 &slab_size_attr
.attr
,
3444 &object_size_attr
.attr
,
3445 &objs_per_slab_attr
.attr
,
3450 &cpu_slabs_attr
.attr
,
3454 &sanity_checks_attr
.attr
,
3456 &hwcache_align_attr
.attr
,
3457 &reclaim_account_attr
.attr
,
3458 &destroy_by_rcu_attr
.attr
,
3459 &red_zone_attr
.attr
,
3461 &store_user_attr
.attr
,
3462 &validate_attr
.attr
,
3464 &alloc_calls_attr
.attr
,
3465 &free_calls_attr
.attr
,
3466 #ifdef CONFIG_ZONE_DMA
3467 &cache_dma_attr
.attr
,
3470 &defrag_ratio_attr
.attr
,
3475 static struct attribute_group slab_attr_group
= {
3476 .attrs
= slab_attrs
,
3479 static ssize_t
slab_attr_show(struct kobject
*kobj
,
3480 struct attribute
*attr
,
3483 struct slab_attribute
*attribute
;
3484 struct kmem_cache
*s
;
3487 attribute
= to_slab_attr(attr
);
3490 if (!attribute
->show
)
3493 err
= attribute
->show(s
, buf
);
3498 static ssize_t
slab_attr_store(struct kobject
*kobj
,
3499 struct attribute
*attr
,
3500 const char *buf
, size_t len
)
3502 struct slab_attribute
*attribute
;
3503 struct kmem_cache
*s
;
3506 attribute
= to_slab_attr(attr
);
3509 if (!attribute
->store
)
3512 err
= attribute
->store(s
, buf
, len
);
3517 static struct sysfs_ops slab_sysfs_ops
= {
3518 .show
= slab_attr_show
,
3519 .store
= slab_attr_store
,
3522 static struct kobj_type slab_ktype
= {
3523 .sysfs_ops
= &slab_sysfs_ops
,
3526 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
3528 struct kobj_type
*ktype
= get_ktype(kobj
);
3530 if (ktype
== &slab_ktype
)
3535 static struct kset_uevent_ops slab_uevent_ops
= {
3536 .filter
= uevent_filter
,
3539 decl_subsys(slab
, &slab_ktype
, &slab_uevent_ops
);
3541 #define ID_STR_LENGTH 64
3543 /* Create a unique string id for a slab cache:
3545 * :[flags-]size:[memory address of kmemcache]
3547 static char *create_unique_id(struct kmem_cache
*s
)
3549 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
3556 * First flags affecting slabcache operations. We will only
3557 * get here for aliasable slabs so we do not need to support
3558 * too many flags. The flags here must cover all flags that
3559 * are matched during merging to guarantee that the id is
3562 if (s
->flags
& SLAB_CACHE_DMA
)
3564 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3566 if (s
->flags
& SLAB_DEBUG_FREE
)
3570 p
+= sprintf(p
, "%07d", s
->size
);
3571 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
3575 static int sysfs_slab_add(struct kmem_cache
*s
)
3581 if (slab_state
< SYSFS
)
3582 /* Defer until later */
3585 unmergeable
= slab_unmergeable(s
);
3588 * Slabcache can never be merged so we can use the name proper.
3589 * This is typically the case for debug situations. In that
3590 * case we can catch duplicate names easily.
3592 sysfs_remove_link(&slab_subsys
.kobj
, s
->name
);
3596 * Create a unique name for the slab as a target
3599 name
= create_unique_id(s
);
3602 kobj_set_kset_s(s
, slab_subsys
);
3603 kobject_set_name(&s
->kobj
, name
);
3604 kobject_init(&s
->kobj
);
3605 err
= kobject_add(&s
->kobj
);
3609 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
3612 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
3614 /* Setup first alias */
3615 sysfs_slab_alias(s
, s
->name
);
3621 static void sysfs_slab_remove(struct kmem_cache
*s
)
3623 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
3624 kobject_del(&s
->kobj
);
3628 * Need to buffer aliases during bootup until sysfs becomes
3629 * available lest we loose that information.
3631 struct saved_alias
{
3632 struct kmem_cache
*s
;
3634 struct saved_alias
*next
;
3637 struct saved_alias
*alias_list
;
3639 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
3641 struct saved_alias
*al
;
3643 if (slab_state
== SYSFS
) {
3645 * If we have a leftover link then remove it.
3647 sysfs_remove_link(&slab_subsys
.kobj
, name
);
3648 return sysfs_create_link(&slab_subsys
.kobj
,
3652 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
3658 al
->next
= alias_list
;
3663 static int __init
slab_sysfs_init(void)
3665 struct list_head
*h
;
3668 err
= subsystem_register(&slab_subsys
);
3670 printk(KERN_ERR
"Cannot register slab subsystem.\n");
3676 list_for_each(h
, &slab_caches
) {
3677 struct kmem_cache
*s
=
3678 container_of(h
, struct kmem_cache
, list
);
3680 err
= sysfs_slab_add(s
);
3684 while (alias_list
) {
3685 struct saved_alias
*al
= alias_list
;
3687 alias_list
= alias_list
->next
;
3688 err
= sysfs_slab_alias(al
->s
, al
->name
);
3697 __initcall(slab_sysfs_init
);