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>
23 #include <linux/memory.h>
30 * The slab_lock protects operations on the object of a particular
31 * slab and its metadata in the page struct. If the slab lock
32 * has been taken then no allocations nor frees can be performed
33 * on the objects in the slab nor can the slab be added or removed
34 * from the partial or full lists since this would mean modifying
35 * the page_struct of the slab.
37 * The list_lock protects the partial and full list on each node and
38 * the partial slab counter. If taken then no new slabs may be added or
39 * removed from the lists nor make the number of partial slabs be modified.
40 * (Note that the total number of slabs is an atomic value that may be
41 * modified without taking the list lock).
43 * The list_lock is a centralized lock and thus we avoid taking it as
44 * much as possible. As long as SLUB does not have to handle partial
45 * slabs, operations can continue without any centralized lock. F.e.
46 * allocating a long series of objects that fill up slabs does not require
49 * The lock order is sometimes inverted when we are trying to get a slab
50 * off a list. We take the list_lock and then look for a page on the list
51 * to use. While we do that objects in the slabs may be freed. We can
52 * only operate on the slab if we have also taken the slab_lock. So we use
53 * a slab_trylock() on the slab. If trylock was successful then no frees
54 * can occur anymore and we can use the slab for allocations etc. If the
55 * slab_trylock() does not succeed then frees are in progress in the slab and
56 * we must stay away from it for a while since we may cause a bouncing
57 * cacheline if we try to acquire the lock. So go onto the next slab.
58 * If all pages are busy then we may allocate a new slab instead of reusing
59 * a partial slab. A new slab has noone operating on it and thus there is
60 * no danger of cacheline contention.
62 * Interrupts are disabled during allocation and deallocation in order to
63 * make the slab allocator safe to use in the context of an irq. In addition
64 * interrupts are disabled to ensure that the processor does not change
65 * while handling per_cpu slabs, due to kernel preemption.
67 * SLUB assigns one slab for allocation to each processor.
68 * Allocations only occur from these slabs called cpu slabs.
70 * Slabs with free elements are kept on a partial list and during regular
71 * operations no list for full slabs is used. If an object in a full slab is
72 * freed then the slab will show up again on the partial lists.
73 * We track full slabs for debugging purposes though because otherwise we
74 * cannot scan all objects.
76 * Slabs are freed when they become empty. Teardown and setup is
77 * minimal so we rely on the page allocators per cpu caches for
78 * fast frees and allocs.
80 * Overloading of page flags that are otherwise used for LRU management.
82 * PageActive The slab is frozen and exempt from list processing.
83 * This means that the slab is dedicated to a purpose
84 * such as satisfying allocations for a specific
85 * processor. Objects may be freed in the slab while
86 * it is frozen but slab_free will then skip the usual
87 * list operations. It is up to the processor holding
88 * the slab to integrate the slab into the slab lists
89 * when the slab is no longer needed.
91 * One use of this flag is to mark slabs that are
92 * used for allocations. Then such a slab becomes a cpu
93 * slab. The cpu slab may be equipped with an additional
94 * freelist that allows lockless access to
95 * free objects in addition to the regular freelist
96 * that requires the slab lock.
98 * PageError Slab requires special handling due to debug
99 * options set. This moves slab handling out of
100 * the fast path and disables lockless freelists.
103 #define FROZEN (1 << PG_active)
105 #ifdef CONFIG_SLUB_DEBUG
106 #define SLABDEBUG (1 << PG_error)
111 static inline int SlabFrozen(struct page
*page
)
113 return page
->flags
& FROZEN
;
116 static inline void SetSlabFrozen(struct page
*page
)
118 page
->flags
|= FROZEN
;
121 static inline void ClearSlabFrozen(struct page
*page
)
123 page
->flags
&= ~FROZEN
;
126 static inline int SlabDebug(struct page
*page
)
128 return page
->flags
& SLABDEBUG
;
131 static inline void SetSlabDebug(struct page
*page
)
133 page
->flags
|= SLABDEBUG
;
136 static inline void ClearSlabDebug(struct page
*page
)
138 page
->flags
&= ~SLABDEBUG
;
142 * Issues still to be resolved:
144 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
146 * - Variable sizing of the per node arrays
149 /* Enable to test recovery from slab corruption on boot */
150 #undef SLUB_RESILIENCY_TEST
155 * Small page size. Make sure that we do not fragment memory
157 #define DEFAULT_MAX_ORDER 1
158 #define DEFAULT_MIN_OBJECTS 4
163 * Large page machines are customarily able to handle larger
166 #define DEFAULT_MAX_ORDER 2
167 #define DEFAULT_MIN_OBJECTS 8
172 * Mininum number of partial slabs. These will be left on the partial
173 * lists even if they are empty. kmem_cache_shrink may reclaim them.
175 #define MIN_PARTIAL 5
178 * Maximum number of desirable partial slabs.
179 * The existence of more partial slabs makes kmem_cache_shrink
180 * sort the partial list by the number of objects in the.
182 #define MAX_PARTIAL 10
184 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
185 SLAB_POISON | SLAB_STORE_USER)
188 * Set of flags that will prevent slab merging
190 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
191 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
193 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
196 #ifndef ARCH_KMALLOC_MINALIGN
197 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
200 #ifndef ARCH_SLAB_MINALIGN
201 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
204 /* Internal SLUB flags */
205 #define __OBJECT_POISON 0x80000000 /* Poison object */
206 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
207 #define __KMALLOC_CACHE 0x20000000 /* objects freed using kfree */
208 #define __PAGE_ALLOC_FALLBACK 0x10000000 /* Allow fallback to page alloc */
210 /* Not all arches define cache_line_size */
211 #ifndef cache_line_size
212 #define cache_line_size() L1_CACHE_BYTES
215 static int kmem_size
= sizeof(struct kmem_cache
);
218 static struct notifier_block slab_notifier
;
222 DOWN
, /* No slab functionality available */
223 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
224 UP
, /* Everything works but does not show up in sysfs */
228 /* A list of all slab caches on the system */
229 static DECLARE_RWSEM(slub_lock
);
230 static LIST_HEAD(slab_caches
);
233 * Tracking user of a slab.
236 void *addr
; /* Called from address */
237 int cpu
; /* Was running on cpu */
238 int pid
; /* Pid context */
239 unsigned long when
; /* When did the operation occur */
242 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
244 #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)
245 static int sysfs_slab_add(struct kmem_cache
*);
246 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
247 static void sysfs_slab_remove(struct kmem_cache
*);
250 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
251 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
253 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
260 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
262 #ifdef CONFIG_SLUB_STATS
267 /********************************************************************
268 * Core slab cache functions
269 *******************************************************************/
271 int slab_is_available(void)
273 return slab_state
>= UP
;
276 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
279 return s
->node
[node
];
281 return &s
->local_node
;
285 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
288 return s
->cpu_slab
[cpu
];
294 /* Verify that a pointer has an address that is valid within a slab page */
295 static inline int check_valid_pointer(struct kmem_cache
*s
,
296 struct page
*page
, const void *object
)
303 base
= page_address(page
);
304 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
305 (object
- base
) % s
->size
) {
313 * Slow version of get and set free pointer.
315 * This version requires touching the cache lines of kmem_cache which
316 * we avoid to do in the fast alloc free paths. There we obtain the offset
317 * from the page struct.
319 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
321 return *(void **)(object
+ s
->offset
);
324 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
326 *(void **)(object
+ s
->offset
) = fp
;
329 /* Loop over all objects in a slab */
330 #define for_each_object(__p, __s, __addr, __objects) \
331 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
335 #define for_each_free_object(__p, __s, __free) \
336 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
338 /* Determine object index from a given position */
339 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
341 return (p
- addr
) / s
->size
;
344 static inline struct kmem_cache_order_objects
oo_make(int order
,
347 struct kmem_cache_order_objects x
= {
348 (order
<< 16) + (PAGE_SIZE
<< order
) / size
354 static inline int oo_order(struct kmem_cache_order_objects x
)
359 static inline int oo_objects(struct kmem_cache_order_objects x
)
361 return x
.x
& ((1 << 16) - 1);
364 #ifdef CONFIG_SLUB_DEBUG
368 #ifdef CONFIG_SLUB_DEBUG_ON
369 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
371 static int slub_debug
;
374 static char *slub_debug_slabs
;
379 static void print_section(char *text
, u8
*addr
, unsigned int length
)
387 for (i
= 0; i
< length
; i
++) {
389 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
392 printk(KERN_CONT
" %02x", addr
[i
]);
394 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
396 printk(KERN_CONT
" %s\n", ascii
);
403 printk(KERN_CONT
" ");
407 printk(KERN_CONT
" %s\n", ascii
);
411 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
412 enum track_item alloc
)
417 p
= object
+ s
->offset
+ sizeof(void *);
419 p
= object
+ s
->inuse
;
424 static void set_track(struct kmem_cache
*s
, void *object
,
425 enum track_item alloc
, void *addr
)
430 p
= object
+ s
->offset
+ sizeof(void *);
432 p
= object
+ s
->inuse
;
437 p
->cpu
= smp_processor_id();
438 p
->pid
= current
? current
->pid
: -1;
441 memset(p
, 0, sizeof(struct track
));
444 static void init_tracking(struct kmem_cache
*s
, void *object
)
446 if (!(s
->flags
& SLAB_STORE_USER
))
449 set_track(s
, object
, TRACK_FREE
, NULL
);
450 set_track(s
, object
, TRACK_ALLOC
, NULL
);
453 static void print_track(const char *s
, struct track
*t
)
458 printk(KERN_ERR
"INFO: %s in ", s
);
459 __print_symbol("%s", (unsigned long)t
->addr
);
460 printk(" age=%lu cpu=%u pid=%d\n", jiffies
- t
->when
, t
->cpu
, t
->pid
);
463 static void print_tracking(struct kmem_cache
*s
, void *object
)
465 if (!(s
->flags
& SLAB_STORE_USER
))
468 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
469 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
472 static void print_page_info(struct page
*page
)
474 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
475 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
479 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
485 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
487 printk(KERN_ERR
"========================================"
488 "=====================================\n");
489 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
490 printk(KERN_ERR
"----------------------------------------"
491 "-------------------------------------\n\n");
494 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
500 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
502 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
505 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
507 unsigned int off
; /* Offset of last byte */
508 u8
*addr
= page_address(page
);
510 print_tracking(s
, p
);
512 print_page_info(page
);
514 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
515 p
, p
- addr
, get_freepointer(s
, p
));
518 print_section("Bytes b4", p
- 16, 16);
520 print_section("Object", p
, min(s
->objsize
, 128));
522 if (s
->flags
& SLAB_RED_ZONE
)
523 print_section("Redzone", p
+ s
->objsize
,
524 s
->inuse
- s
->objsize
);
527 off
= s
->offset
+ sizeof(void *);
531 if (s
->flags
& SLAB_STORE_USER
)
532 off
+= 2 * sizeof(struct track
);
535 /* Beginning of the filler is the free pointer */
536 print_section("Padding", p
+ off
, s
->size
- off
);
541 static void object_err(struct kmem_cache
*s
, struct page
*page
,
542 u8
*object
, char *reason
)
544 slab_bug(s
, "%s", reason
);
545 print_trailer(s
, page
, object
);
548 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
554 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
556 slab_bug(s
, "%s", buf
);
557 print_page_info(page
);
561 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
565 if (s
->flags
& __OBJECT_POISON
) {
566 memset(p
, POISON_FREE
, s
->objsize
- 1);
567 p
[s
->objsize
- 1] = POISON_END
;
570 if (s
->flags
& SLAB_RED_ZONE
)
571 memset(p
+ s
->objsize
,
572 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
573 s
->inuse
- s
->objsize
);
576 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
579 if (*start
!= (u8
)value
)
587 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
588 void *from
, void *to
)
590 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
591 memset(from
, data
, to
- from
);
594 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
595 u8
*object
, char *what
,
596 u8
*start
, unsigned int value
, unsigned int bytes
)
601 fault
= check_bytes(start
, value
, bytes
);
606 while (end
> fault
&& end
[-1] == value
)
609 slab_bug(s
, "%s overwritten", what
);
610 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
611 fault
, end
- 1, fault
[0], value
);
612 print_trailer(s
, page
, object
);
614 restore_bytes(s
, what
, value
, fault
, end
);
622 * Bytes of the object to be managed.
623 * If the freepointer may overlay the object then the free
624 * pointer is the first word of the object.
626 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
629 * object + s->objsize
630 * Padding to reach word boundary. This is also used for Redzoning.
631 * Padding is extended by another word if Redzoning is enabled and
634 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
635 * 0xcc (RED_ACTIVE) for objects in use.
638 * Meta data starts here.
640 * A. Free pointer (if we cannot overwrite object on free)
641 * B. Tracking data for SLAB_STORE_USER
642 * C. Padding to reach required alignment boundary or at mininum
643 * one word if debugging is on to be able to detect writes
644 * before the word boundary.
646 * Padding is done using 0x5a (POISON_INUSE)
649 * Nothing is used beyond s->size.
651 * If slabcaches are merged then the objsize and inuse boundaries are mostly
652 * ignored. And therefore no slab options that rely on these boundaries
653 * may be used with merged slabcaches.
656 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
658 unsigned long off
= s
->inuse
; /* The end of info */
661 /* Freepointer is placed after the object. */
662 off
+= sizeof(void *);
664 if (s
->flags
& SLAB_STORE_USER
)
665 /* We also have user information there */
666 off
+= 2 * sizeof(struct track
);
671 return check_bytes_and_report(s
, page
, p
, "Object padding",
672 p
+ off
, POISON_INUSE
, s
->size
- off
);
675 /* Check the pad bytes at the end of a slab page */
676 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
684 if (!(s
->flags
& SLAB_POISON
))
687 start
= page_address(page
);
688 length
= (PAGE_SIZE
<< compound_order(page
));
689 end
= start
+ length
;
690 remainder
= length
% s
->size
;
694 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
697 while (end
> fault
&& end
[-1] == POISON_INUSE
)
700 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
701 print_section("Padding", end
- remainder
, remainder
);
703 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
707 static int check_object(struct kmem_cache
*s
, struct page
*page
,
708 void *object
, int active
)
711 u8
*endobject
= object
+ s
->objsize
;
713 if (s
->flags
& SLAB_RED_ZONE
) {
715 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
717 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
718 endobject
, red
, s
->inuse
- s
->objsize
))
721 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
722 check_bytes_and_report(s
, page
, p
, "Alignment padding",
723 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
727 if (s
->flags
& SLAB_POISON
) {
728 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
729 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
730 POISON_FREE
, s
->objsize
- 1) ||
731 !check_bytes_and_report(s
, page
, p
, "Poison",
732 p
+ s
->objsize
- 1, POISON_END
, 1)))
735 * check_pad_bytes cleans up on its own.
737 check_pad_bytes(s
, page
, p
);
740 if (!s
->offset
&& active
)
742 * Object and freepointer overlap. Cannot check
743 * freepointer while object is allocated.
747 /* Check free pointer validity */
748 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
749 object_err(s
, page
, p
, "Freepointer corrupt");
751 * No choice but to zap it and thus loose the remainder
752 * of the free objects in this slab. May cause
753 * another error because the object count is now wrong.
755 set_freepointer(s
, p
, NULL
);
761 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
765 VM_BUG_ON(!irqs_disabled());
767 if (!PageSlab(page
)) {
768 slab_err(s
, page
, "Not a valid slab page");
772 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
773 if (page
->objects
> maxobj
) {
774 slab_err(s
, page
, "objects %u > max %u",
775 s
->name
, page
->objects
, maxobj
);
778 if (page
->inuse
> page
->objects
) {
779 slab_err(s
, page
, "inuse %u > max %u",
780 s
->name
, page
->inuse
, page
->objects
);
783 /* Slab_pad_check fixes things up after itself */
784 slab_pad_check(s
, page
);
789 * Determine if a certain object on a page is on the freelist. Must hold the
790 * slab lock to guarantee that the chains are in a consistent state.
792 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
795 void *fp
= page
->freelist
;
797 unsigned long max_objects
;
799 while (fp
&& nr
<= page
->objects
) {
802 if (!check_valid_pointer(s
, page
, fp
)) {
804 object_err(s
, page
, object
,
805 "Freechain corrupt");
806 set_freepointer(s
, object
, NULL
);
809 slab_err(s
, page
, "Freepointer corrupt");
810 page
->freelist
= NULL
;
811 page
->inuse
= page
->objects
;
812 slab_fix(s
, "Freelist cleared");
818 fp
= get_freepointer(s
, object
);
822 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
823 if (max_objects
> 65535)
826 if (page
->objects
!= max_objects
) {
827 slab_err(s
, page
, "Wrong number of objects. Found %d but "
828 "should be %d", page
->objects
, max_objects
);
829 page
->objects
= max_objects
;
830 slab_fix(s
, "Number of objects adjusted.");
832 if (page
->inuse
!= page
->objects
- nr
) {
833 slab_err(s
, page
, "Wrong object count. Counter is %d but "
834 "counted were %d", page
->inuse
, page
->objects
- nr
);
835 page
->inuse
= page
->objects
- nr
;
836 slab_fix(s
, "Object count adjusted.");
838 return search
== NULL
;
841 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
, int alloc
)
843 if (s
->flags
& SLAB_TRACE
) {
844 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
846 alloc
? "alloc" : "free",
851 print_section("Object", (void *)object
, s
->objsize
);
858 * Tracking of fully allocated slabs for debugging purposes.
860 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
862 spin_lock(&n
->list_lock
);
863 list_add(&page
->lru
, &n
->full
);
864 spin_unlock(&n
->list_lock
);
867 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
869 struct kmem_cache_node
*n
;
871 if (!(s
->flags
& SLAB_STORE_USER
))
874 n
= get_node(s
, page_to_nid(page
));
876 spin_lock(&n
->list_lock
);
877 list_del(&page
->lru
);
878 spin_unlock(&n
->list_lock
);
881 /* Tracking of the number of slabs for debugging purposes */
882 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
884 struct kmem_cache_node
*n
= get_node(s
, node
);
886 return atomic_long_read(&n
->nr_slabs
);
889 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
891 struct kmem_cache_node
*n
= get_node(s
, node
);
894 * May be called early in order to allocate a slab for the
895 * kmem_cache_node structure. Solve the chicken-egg
896 * dilemma by deferring the increment of the count during
897 * bootstrap (see early_kmem_cache_node_alloc).
899 if (!NUMA_BUILD
|| n
) {
900 atomic_long_inc(&n
->nr_slabs
);
901 atomic_long_add(objects
, &n
->total_objects
);
904 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
906 struct kmem_cache_node
*n
= get_node(s
, node
);
908 atomic_long_dec(&n
->nr_slabs
);
909 atomic_long_sub(objects
, &n
->total_objects
);
912 /* Object debug checks for alloc/free paths */
913 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
916 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
919 init_object(s
, object
, 0);
920 init_tracking(s
, object
);
923 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
924 void *object
, void *addr
)
926 if (!check_slab(s
, page
))
929 if (!on_freelist(s
, page
, object
)) {
930 object_err(s
, page
, object
, "Object already allocated");
934 if (!check_valid_pointer(s
, page
, object
)) {
935 object_err(s
, page
, object
, "Freelist Pointer check fails");
939 if (!check_object(s
, page
, object
, 0))
942 /* Success perform special debug activities for allocs */
943 if (s
->flags
& SLAB_STORE_USER
)
944 set_track(s
, object
, TRACK_ALLOC
, addr
);
945 trace(s
, page
, object
, 1);
946 init_object(s
, object
, 1);
950 if (PageSlab(page
)) {
952 * If this is a slab page then lets do the best we can
953 * to avoid issues in the future. Marking all objects
954 * as used avoids touching the remaining objects.
956 slab_fix(s
, "Marking all objects used");
957 page
->inuse
= page
->objects
;
958 page
->freelist
= NULL
;
963 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
964 void *object
, void *addr
)
966 if (!check_slab(s
, page
))
969 if (!check_valid_pointer(s
, page
, object
)) {
970 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
974 if (on_freelist(s
, page
, object
)) {
975 object_err(s
, page
, object
, "Object already free");
979 if (!check_object(s
, page
, object
, 1))
982 if (unlikely(s
!= page
->slab
)) {
983 if (!PageSlab(page
)) {
984 slab_err(s
, page
, "Attempt to free object(0x%p) "
985 "outside of slab", object
);
986 } else if (!page
->slab
) {
988 "SLUB <none>: no slab for object 0x%p.\n",
992 object_err(s
, page
, object
,
993 "page slab pointer corrupt.");
997 /* Special debug activities for freeing objects */
998 if (!SlabFrozen(page
) && !page
->freelist
)
999 remove_full(s
, page
);
1000 if (s
->flags
& SLAB_STORE_USER
)
1001 set_track(s
, object
, TRACK_FREE
, addr
);
1002 trace(s
, page
, object
, 0);
1003 init_object(s
, object
, 0);
1007 slab_fix(s
, "Object at 0x%p not freed", object
);
1011 static int __init
setup_slub_debug(char *str
)
1013 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1014 if (*str
++ != '=' || !*str
)
1016 * No options specified. Switch on full debugging.
1022 * No options but restriction on slabs. This means full
1023 * debugging for slabs matching a pattern.
1030 * Switch off all debugging measures.
1035 * Determine which debug features should be switched on
1037 for (; *str
&& *str
!= ','; str
++) {
1038 switch (tolower(*str
)) {
1040 slub_debug
|= SLAB_DEBUG_FREE
;
1043 slub_debug
|= SLAB_RED_ZONE
;
1046 slub_debug
|= SLAB_POISON
;
1049 slub_debug
|= SLAB_STORE_USER
;
1052 slub_debug
|= SLAB_TRACE
;
1055 printk(KERN_ERR
"slub_debug option '%c' "
1056 "unknown. skipped\n", *str
);
1062 slub_debug_slabs
= str
+ 1;
1067 __setup("slub_debug", setup_slub_debug
);
1069 static unsigned long kmem_cache_flags(unsigned long objsize
,
1070 unsigned long flags
, const char *name
,
1071 void (*ctor
)(struct kmem_cache
*, void *))
1074 * Enable debugging if selected on the kernel commandline.
1076 if (slub_debug
&& (!slub_debug_slabs
||
1077 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1078 flags
|= slub_debug
;
1083 static inline void setup_object_debug(struct kmem_cache
*s
,
1084 struct page
*page
, void *object
) {}
1086 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1087 struct page
*page
, void *object
, void *addr
) { return 0; }
1089 static inline int free_debug_processing(struct kmem_cache
*s
,
1090 struct page
*page
, void *object
, void *addr
) { return 0; }
1092 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1094 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1095 void *object
, int active
) { return 1; }
1096 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1097 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1098 unsigned long flags
, const char *name
,
1099 void (*ctor
)(struct kmem_cache
*, void *))
1103 #define slub_debug 0
1105 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1107 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1109 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1114 * Slab allocation and freeing
1116 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1117 struct kmem_cache_order_objects oo
)
1119 int order
= oo_order(oo
);
1122 return alloc_pages(flags
, order
);
1124 return alloc_pages_node(node
, flags
, order
);
1127 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1130 struct kmem_cache_order_objects oo
= s
->oo
;
1132 flags
|= s
->allocflags
;
1134 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1136 if (unlikely(!page
)) {
1139 * Allocation may have failed due to fragmentation.
1140 * Try a lower order alloc if possible
1142 page
= alloc_slab_page(flags
, node
, oo
);
1146 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1148 page
->objects
= oo_objects(oo
);
1149 mod_zone_page_state(page_zone(page
),
1150 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1151 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1157 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1160 setup_object_debug(s
, page
, object
);
1161 if (unlikely(s
->ctor
))
1165 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1172 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1174 page
= allocate_slab(s
,
1175 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1179 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1181 page
->flags
|= 1 << PG_slab
;
1182 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1183 SLAB_STORE_USER
| SLAB_TRACE
))
1186 start
= page_address(page
);
1188 if (unlikely(s
->flags
& SLAB_POISON
))
1189 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1192 for_each_object(p
, s
, start
, page
->objects
) {
1193 setup_object(s
, page
, last
);
1194 set_freepointer(s
, last
, p
);
1197 setup_object(s
, page
, last
);
1198 set_freepointer(s
, last
, NULL
);
1200 page
->freelist
= start
;
1206 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1208 int order
= compound_order(page
);
1209 int pages
= 1 << order
;
1211 if (unlikely(SlabDebug(page
))) {
1214 slab_pad_check(s
, page
);
1215 for_each_object(p
, s
, page_address(page
),
1217 check_object(s
, page
, p
, 0);
1218 ClearSlabDebug(page
);
1221 mod_zone_page_state(page_zone(page
),
1222 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1223 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1226 __ClearPageSlab(page
);
1227 reset_page_mapcount(page
);
1228 __free_pages(page
, order
);
1231 static void rcu_free_slab(struct rcu_head
*h
)
1235 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1236 __free_slab(page
->slab
, page
);
1239 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1241 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1243 * RCU free overloads the RCU head over the LRU
1245 struct rcu_head
*head
= (void *)&page
->lru
;
1247 call_rcu(head
, rcu_free_slab
);
1249 __free_slab(s
, page
);
1252 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1254 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1259 * Per slab locking using the pagelock
1261 static __always_inline
void slab_lock(struct page
*page
)
1263 bit_spin_lock(PG_locked
, &page
->flags
);
1266 static __always_inline
void slab_unlock(struct page
*page
)
1268 __bit_spin_unlock(PG_locked
, &page
->flags
);
1271 static __always_inline
int slab_trylock(struct page
*page
)
1275 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1280 * Management of partially allocated slabs
1282 static void add_partial(struct kmem_cache_node
*n
,
1283 struct page
*page
, int tail
)
1285 spin_lock(&n
->list_lock
);
1288 list_add_tail(&page
->lru
, &n
->partial
);
1290 list_add(&page
->lru
, &n
->partial
);
1291 spin_unlock(&n
->list_lock
);
1294 static void remove_partial(struct kmem_cache
*s
,
1297 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1299 spin_lock(&n
->list_lock
);
1300 list_del(&page
->lru
);
1302 spin_unlock(&n
->list_lock
);
1306 * Lock slab and remove from the partial list.
1308 * Must hold list_lock.
1310 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
, struct page
*page
)
1312 if (slab_trylock(page
)) {
1313 list_del(&page
->lru
);
1315 SetSlabFrozen(page
);
1322 * Try to allocate a partial slab from a specific node.
1324 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1329 * Racy check. If we mistakenly see no partial slabs then we
1330 * just allocate an empty slab. If we mistakenly try to get a
1331 * partial slab and there is none available then get_partials()
1334 if (!n
|| !n
->nr_partial
)
1337 spin_lock(&n
->list_lock
);
1338 list_for_each_entry(page
, &n
->partial
, lru
)
1339 if (lock_and_freeze_slab(n
, page
))
1343 spin_unlock(&n
->list_lock
);
1348 * Get a page from somewhere. Search in increasing NUMA distances.
1350 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1353 struct zonelist
*zonelist
;
1358 * The defrag ratio allows a configuration of the tradeoffs between
1359 * inter node defragmentation and node local allocations. A lower
1360 * defrag_ratio increases the tendency to do local allocations
1361 * instead of attempting to obtain partial slabs from other nodes.
1363 * If the defrag_ratio is set to 0 then kmalloc() always
1364 * returns node local objects. If the ratio is higher then kmalloc()
1365 * may return off node objects because partial slabs are obtained
1366 * from other nodes and filled up.
1368 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1369 * defrag_ratio = 1000) then every (well almost) allocation will
1370 * first attempt to defrag slab caches on other nodes. This means
1371 * scanning over all nodes to look for partial slabs which may be
1372 * expensive if we do it every time we are trying to find a slab
1373 * with available objects.
1375 if (!s
->remote_node_defrag_ratio
||
1376 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1379 zonelist
= &NODE_DATA(
1380 slab_node(current
->mempolicy
))->node_zonelists
[gfp_zone(flags
)];
1381 for (z
= zonelist
->zones
; *z
; z
++) {
1382 struct kmem_cache_node
*n
;
1384 n
= get_node(s
, zone_to_nid(*z
));
1386 if (n
&& cpuset_zone_allowed_hardwall(*z
, flags
) &&
1387 n
->nr_partial
> MIN_PARTIAL
) {
1388 page
= get_partial_node(n
);
1398 * Get a partial page, lock it and return it.
1400 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1403 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1405 page
= get_partial_node(get_node(s
, searchnode
));
1406 if (page
|| (flags
& __GFP_THISNODE
))
1409 return get_any_partial(s
, flags
);
1413 * Move a page back to the lists.
1415 * Must be called with the slab lock held.
1417 * On exit the slab lock will have been dropped.
1419 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1421 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1422 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1424 ClearSlabFrozen(page
);
1427 if (page
->freelist
) {
1428 add_partial(n
, page
, tail
);
1429 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1431 stat(c
, DEACTIVATE_FULL
);
1432 if (SlabDebug(page
) && (s
->flags
& SLAB_STORE_USER
))
1437 stat(c
, DEACTIVATE_EMPTY
);
1438 if (n
->nr_partial
< MIN_PARTIAL
) {
1440 * Adding an empty slab to the partial slabs in order
1441 * to avoid page allocator overhead. This slab needs
1442 * to come after the other slabs with objects in
1443 * so that the others get filled first. That way the
1444 * size of the partial list stays small.
1446 * kmem_cache_shrink can reclaim any empty slabs from the
1449 add_partial(n
, page
, 1);
1453 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1454 discard_slab(s
, page
);
1460 * Remove the cpu slab
1462 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1464 struct page
*page
= c
->page
;
1468 stat(c
, DEACTIVATE_REMOTE_FREES
);
1470 * Merge cpu freelist into slab freelist. Typically we get here
1471 * because both freelists are empty. So this is unlikely
1474 while (unlikely(c
->freelist
)) {
1477 tail
= 0; /* Hot objects. Put the slab first */
1479 /* Retrieve object from cpu_freelist */
1480 object
= c
->freelist
;
1481 c
->freelist
= c
->freelist
[c
->offset
];
1483 /* And put onto the regular freelist */
1484 object
[c
->offset
] = page
->freelist
;
1485 page
->freelist
= object
;
1489 unfreeze_slab(s
, page
, tail
);
1492 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1494 stat(c
, CPUSLAB_FLUSH
);
1496 deactivate_slab(s
, c
);
1502 * Called from IPI handler with interrupts disabled.
1504 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1506 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1508 if (likely(c
&& c
->page
))
1512 static void flush_cpu_slab(void *d
)
1514 struct kmem_cache
*s
= d
;
1516 __flush_cpu_slab(s
, smp_processor_id());
1519 static void flush_all(struct kmem_cache
*s
)
1522 on_each_cpu(flush_cpu_slab
, s
, 1, 1);
1524 unsigned long flags
;
1526 local_irq_save(flags
);
1528 local_irq_restore(flags
);
1533 * Check if the objects in a per cpu structure fit numa
1534 * locality expectations.
1536 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1539 if (node
!= -1 && c
->node
!= node
)
1546 * Slow path. The lockless freelist is empty or we need to perform
1549 * Interrupts are disabled.
1551 * Processing is still very fast if new objects have been freed to the
1552 * regular freelist. In that case we simply take over the regular freelist
1553 * as the lockless freelist and zap the regular freelist.
1555 * If that is not working then we fall back to the partial lists. We take the
1556 * first element of the freelist as the object to allocate now and move the
1557 * rest of the freelist to the lockless freelist.
1559 * And if we were unable to get a new slab from the partial slab lists then
1560 * we need to allocate a new slab. This is the slowest path since it involves
1561 * a call to the page allocator and the setup of a new slab.
1563 static void *__slab_alloc(struct kmem_cache
*s
,
1564 gfp_t gfpflags
, int node
, void *addr
, struct kmem_cache_cpu
*c
)
1569 /* We handle __GFP_ZERO in the caller */
1570 gfpflags
&= ~__GFP_ZERO
;
1576 if (unlikely(!node_match(c
, node
)))
1579 stat(c
, ALLOC_REFILL
);
1582 object
= c
->page
->freelist
;
1583 if (unlikely(!object
))
1585 if (unlikely(SlabDebug(c
->page
)))
1588 c
->freelist
= object
[c
->offset
];
1589 c
->page
->inuse
= c
->page
->objects
;
1590 c
->page
->freelist
= NULL
;
1591 c
->node
= page_to_nid(c
->page
);
1593 slab_unlock(c
->page
);
1594 stat(c
, ALLOC_SLOWPATH
);
1598 deactivate_slab(s
, c
);
1601 new = get_partial(s
, gfpflags
, node
);
1604 stat(c
, ALLOC_FROM_PARTIAL
);
1608 if (gfpflags
& __GFP_WAIT
)
1611 new = new_slab(s
, gfpflags
, node
);
1613 if (gfpflags
& __GFP_WAIT
)
1614 local_irq_disable();
1617 c
= get_cpu_slab(s
, smp_processor_id());
1618 stat(c
, ALLOC_SLAB
);
1628 * No memory available.
1630 * If the slab uses higher order allocs but the object is
1631 * smaller than a page size then we can fallback in emergencies
1632 * to the page allocator via kmalloc_large. The page allocator may
1633 * have failed to obtain a higher order page and we can try to
1634 * allocate a single page if the object fits into a single page.
1635 * That is only possible if certain conditions are met that are being
1636 * checked when a slab is created.
1638 if (!(gfpflags
& __GFP_NORETRY
) &&
1639 (s
->flags
& __PAGE_ALLOC_FALLBACK
)) {
1640 if (gfpflags
& __GFP_WAIT
)
1642 object
= kmalloc_large(s
->objsize
, gfpflags
);
1643 if (gfpflags
& __GFP_WAIT
)
1644 local_irq_disable();
1649 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1653 c
->page
->freelist
= object
[c
->offset
];
1659 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1660 * have the fastpath folded into their functions. So no function call
1661 * overhead for requests that can be satisfied on the fastpath.
1663 * The fastpath works by first checking if the lockless freelist can be used.
1664 * If not then __slab_alloc is called for slow processing.
1666 * Otherwise we can simply pick the next object from the lockless free list.
1668 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1669 gfp_t gfpflags
, int node
, void *addr
)
1672 struct kmem_cache_cpu
*c
;
1673 unsigned long flags
;
1675 local_irq_save(flags
);
1676 c
= get_cpu_slab(s
, smp_processor_id());
1677 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1679 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1682 object
= c
->freelist
;
1683 c
->freelist
= object
[c
->offset
];
1684 stat(c
, ALLOC_FASTPATH
);
1686 local_irq_restore(flags
);
1688 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1689 memset(object
, 0, c
->objsize
);
1694 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1696 return slab_alloc(s
, gfpflags
, -1, __builtin_return_address(0));
1698 EXPORT_SYMBOL(kmem_cache_alloc
);
1701 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1703 return slab_alloc(s
, gfpflags
, node
, __builtin_return_address(0));
1705 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1709 * Slow patch handling. This may still be called frequently since objects
1710 * have a longer lifetime than the cpu slabs in most processing loads.
1712 * So we still attempt to reduce cache line usage. Just take the slab
1713 * lock and free the item. If there is no additional partial page
1714 * handling required then we can return immediately.
1716 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1717 void *x
, void *addr
, unsigned int offset
)
1720 void **object
= (void *)x
;
1721 struct kmem_cache_cpu
*c
;
1723 c
= get_cpu_slab(s
, raw_smp_processor_id());
1724 stat(c
, FREE_SLOWPATH
);
1727 if (unlikely(SlabDebug(page
)))
1731 prior
= object
[offset
] = page
->freelist
;
1732 page
->freelist
= object
;
1735 if (unlikely(SlabFrozen(page
))) {
1736 stat(c
, FREE_FROZEN
);
1740 if (unlikely(!page
->inuse
))
1744 * Objects left in the slab. If it was not on the partial list before
1747 if (unlikely(!prior
)) {
1748 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1749 stat(c
, FREE_ADD_PARTIAL
);
1759 * Slab still on the partial list.
1761 remove_partial(s
, page
);
1762 stat(c
, FREE_REMOVE_PARTIAL
);
1766 discard_slab(s
, page
);
1770 if (!free_debug_processing(s
, page
, x
, addr
))
1776 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1777 * can perform fastpath freeing without additional function calls.
1779 * The fastpath is only possible if we are freeing to the current cpu slab
1780 * of this processor. This typically the case if we have just allocated
1783 * If fastpath is not possible then fall back to __slab_free where we deal
1784 * with all sorts of special processing.
1786 static __always_inline
void slab_free(struct kmem_cache
*s
,
1787 struct page
*page
, void *x
, void *addr
)
1789 void **object
= (void *)x
;
1790 struct kmem_cache_cpu
*c
;
1791 unsigned long flags
;
1793 local_irq_save(flags
);
1794 c
= get_cpu_slab(s
, smp_processor_id());
1795 debug_check_no_locks_freed(object
, c
->objsize
);
1796 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1797 object
[c
->offset
] = c
->freelist
;
1798 c
->freelist
= object
;
1799 stat(c
, FREE_FASTPATH
);
1801 __slab_free(s
, page
, x
, addr
, c
->offset
);
1803 local_irq_restore(flags
);
1806 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1810 page
= virt_to_head_page(x
);
1812 slab_free(s
, page
, x
, __builtin_return_address(0));
1814 EXPORT_SYMBOL(kmem_cache_free
);
1816 /* Figure out on which slab object the object resides */
1817 static struct page
*get_object_page(const void *x
)
1819 struct page
*page
= virt_to_head_page(x
);
1821 if (!PageSlab(page
))
1828 * Object placement in a slab is made very easy because we always start at
1829 * offset 0. If we tune the size of the object to the alignment then we can
1830 * get the required alignment by putting one properly sized object after
1833 * Notice that the allocation order determines the sizes of the per cpu
1834 * caches. Each processor has always one slab available for allocations.
1835 * Increasing the allocation order reduces the number of times that slabs
1836 * must be moved on and off the partial lists and is therefore a factor in
1841 * Mininum / Maximum order of slab pages. This influences locking overhead
1842 * and slab fragmentation. A higher order reduces the number of partial slabs
1843 * and increases the number of allocations possible without having to
1844 * take the list_lock.
1846 static int slub_min_order
;
1847 static int slub_max_order
= DEFAULT_MAX_ORDER
;
1848 static int slub_min_objects
= DEFAULT_MIN_OBJECTS
;
1851 * Merge control. If this is set then no merging of slab caches will occur.
1852 * (Could be removed. This was introduced to pacify the merge skeptics.)
1854 static int slub_nomerge
;
1857 * Calculate the order of allocation given an slab object size.
1859 * The order of allocation has significant impact on performance and other
1860 * system components. Generally order 0 allocations should be preferred since
1861 * order 0 does not cause fragmentation in the page allocator. Larger objects
1862 * be problematic to put into order 0 slabs because there may be too much
1863 * unused space left. We go to a higher order if more than 1/8th of the slab
1866 * In order to reach satisfactory performance we must ensure that a minimum
1867 * number of objects is in one slab. Otherwise we may generate too much
1868 * activity on the partial lists which requires taking the list_lock. This is
1869 * less a concern for large slabs though which are rarely used.
1871 * slub_max_order specifies the order where we begin to stop considering the
1872 * number of objects in a slab as critical. If we reach slub_max_order then
1873 * we try to keep the page order as low as possible. So we accept more waste
1874 * of space in favor of a small page order.
1876 * Higher order allocations also allow the placement of more objects in a
1877 * slab and thereby reduce object handling overhead. If the user has
1878 * requested a higher mininum order then we start with that one instead of
1879 * the smallest order which will fit the object.
1881 static inline int slab_order(int size
, int min_objects
,
1882 int max_order
, int fract_leftover
)
1886 int min_order
= slub_min_order
;
1888 if ((PAGE_SIZE
<< min_order
) / size
> 65535)
1889 return get_order(size
* 65535) - 1;
1891 for (order
= max(min_order
,
1892 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1893 order
<= max_order
; order
++) {
1895 unsigned long slab_size
= PAGE_SIZE
<< order
;
1897 if (slab_size
< min_objects
* size
)
1900 rem
= slab_size
% size
;
1902 if (rem
<= slab_size
/ fract_leftover
)
1910 static inline int calculate_order(int size
)
1917 * Attempt to find best configuration for a slab. This
1918 * works by first attempting to generate a layout with
1919 * the best configuration and backing off gradually.
1921 * First we reduce the acceptable waste in a slab. Then
1922 * we reduce the minimum objects required in a slab.
1924 min_objects
= slub_min_objects
;
1925 while (min_objects
> 1) {
1927 while (fraction
>= 4) {
1928 order
= slab_order(size
, min_objects
,
1929 slub_max_order
, fraction
);
1930 if (order
<= slub_max_order
)
1938 * We were unable to place multiple objects in a slab. Now
1939 * lets see if we can place a single object there.
1941 order
= slab_order(size
, 1, slub_max_order
, 1);
1942 if (order
<= slub_max_order
)
1946 * Doh this slab cannot be placed using slub_max_order.
1948 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1949 if (order
<= MAX_ORDER
)
1955 * Figure out what the alignment of the objects will be.
1957 static unsigned long calculate_alignment(unsigned long flags
,
1958 unsigned long align
, unsigned long size
)
1961 * If the user wants hardware cache aligned objects then follow that
1962 * suggestion if the object is sufficiently large.
1964 * The hardware cache alignment cannot override the specified
1965 * alignment though. If that is greater then use it.
1967 if (flags
& SLAB_HWCACHE_ALIGN
) {
1968 unsigned long ralign
= cache_line_size();
1969 while (size
<= ralign
/ 2)
1971 align
= max(align
, ralign
);
1974 if (align
< ARCH_SLAB_MINALIGN
)
1975 align
= ARCH_SLAB_MINALIGN
;
1977 return ALIGN(align
, sizeof(void *));
1980 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1981 struct kmem_cache_cpu
*c
)
1986 c
->offset
= s
->offset
/ sizeof(void *);
1987 c
->objsize
= s
->objsize
;
1988 #ifdef CONFIG_SLUB_STATS
1989 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1993 static void init_kmem_cache_node(struct kmem_cache_node
*n
)
1996 spin_lock_init(&n
->list_lock
);
1997 INIT_LIST_HEAD(&n
->partial
);
1998 #ifdef CONFIG_SLUB_DEBUG
1999 atomic_long_set(&n
->nr_slabs
, 0);
2000 INIT_LIST_HEAD(&n
->full
);
2006 * Per cpu array for per cpu structures.
2008 * The per cpu array places all kmem_cache_cpu structures from one processor
2009 * close together meaning that it becomes possible that multiple per cpu
2010 * structures are contained in one cacheline. This may be particularly
2011 * beneficial for the kmalloc caches.
2013 * A desktop system typically has around 60-80 slabs. With 100 here we are
2014 * likely able to get per cpu structures for all caches from the array defined
2015 * here. We must be able to cover all kmalloc caches during bootstrap.
2017 * If the per cpu array is exhausted then fall back to kmalloc
2018 * of individual cachelines. No sharing is possible then.
2020 #define NR_KMEM_CACHE_CPU 100
2022 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
2023 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
2025 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
2026 static cpumask_t kmem_cach_cpu_free_init_once
= CPU_MASK_NONE
;
2028 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
2029 int cpu
, gfp_t flags
)
2031 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
2034 per_cpu(kmem_cache_cpu_free
, cpu
) =
2035 (void *)c
->freelist
;
2037 /* Table overflow: So allocate ourselves */
2039 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2040 flags
, cpu_to_node(cpu
));
2045 init_kmem_cache_cpu(s
, c
);
2049 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2051 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2052 c
> per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2056 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2057 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2060 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2064 for_each_online_cpu(cpu
) {
2065 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2068 s
->cpu_slab
[cpu
] = NULL
;
2069 free_kmem_cache_cpu(c
, cpu
);
2074 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2078 for_each_online_cpu(cpu
) {
2079 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2084 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2086 free_kmem_cache_cpus(s
);
2089 s
->cpu_slab
[cpu
] = c
;
2095 * Initialize the per cpu array.
2097 static void init_alloc_cpu_cpu(int cpu
)
2101 if (cpu_isset(cpu
, kmem_cach_cpu_free_init_once
))
2104 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2105 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2107 cpu_set(cpu
, kmem_cach_cpu_free_init_once
);
2110 static void __init
init_alloc_cpu(void)
2114 for_each_online_cpu(cpu
)
2115 init_alloc_cpu_cpu(cpu
);
2119 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2120 static inline void init_alloc_cpu(void) {}
2122 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2124 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2131 * No kmalloc_node yet so do it by hand. We know that this is the first
2132 * slab on the node for this slabcache. There are no concurrent accesses
2135 * Note that this function only works on the kmalloc_node_cache
2136 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2137 * memory on a fresh node that has no slab structures yet.
2139 static struct kmem_cache_node
*early_kmem_cache_node_alloc(gfp_t gfpflags
,
2143 struct kmem_cache_node
*n
;
2144 unsigned long flags
;
2146 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2148 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2151 if (page_to_nid(page
) != node
) {
2152 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2154 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2155 "in order to be able to continue\n");
2160 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2162 kmalloc_caches
->node
[node
] = n
;
2163 #ifdef CONFIG_SLUB_DEBUG
2164 init_object(kmalloc_caches
, n
, 1);
2165 init_tracking(kmalloc_caches
, n
);
2167 init_kmem_cache_node(n
);
2168 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2171 * lockdep requires consistent irq usage for each lock
2172 * so even though there cannot be a race this early in
2173 * the boot sequence, we still disable irqs.
2175 local_irq_save(flags
);
2176 add_partial(n
, page
, 0);
2177 local_irq_restore(flags
);
2181 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2185 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2186 struct kmem_cache_node
*n
= s
->node
[node
];
2187 if (n
&& n
!= &s
->local_node
)
2188 kmem_cache_free(kmalloc_caches
, n
);
2189 s
->node
[node
] = NULL
;
2193 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2198 if (slab_state
>= UP
)
2199 local_node
= page_to_nid(virt_to_page(s
));
2203 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2204 struct kmem_cache_node
*n
;
2206 if (local_node
== node
)
2209 if (slab_state
== DOWN
) {
2210 n
= early_kmem_cache_node_alloc(gfpflags
,
2214 n
= kmem_cache_alloc_node(kmalloc_caches
,
2218 free_kmem_cache_nodes(s
);
2224 init_kmem_cache_node(n
);
2229 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2233 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2235 init_kmem_cache_node(&s
->local_node
);
2241 * calculate_sizes() determines the order and the distribution of data within
2244 static int calculate_sizes(struct kmem_cache
*s
)
2246 unsigned long flags
= s
->flags
;
2247 unsigned long size
= s
->objsize
;
2248 unsigned long align
= s
->align
;
2252 * Round up object size to the next word boundary. We can only
2253 * place the free pointer at word boundaries and this determines
2254 * the possible location of the free pointer.
2256 size
= ALIGN(size
, sizeof(void *));
2258 #ifdef CONFIG_SLUB_DEBUG
2260 * Determine if we can poison the object itself. If the user of
2261 * the slab may touch the object after free or before allocation
2262 * then we should never poison the object itself.
2264 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2266 s
->flags
|= __OBJECT_POISON
;
2268 s
->flags
&= ~__OBJECT_POISON
;
2272 * If we are Redzoning then check if there is some space between the
2273 * end of the object and the free pointer. If not then add an
2274 * additional word to have some bytes to store Redzone information.
2276 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2277 size
+= sizeof(void *);
2281 * With that we have determined the number of bytes in actual use
2282 * by the object. This is the potential offset to the free pointer.
2286 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2289 * Relocate free pointer after the object if it is not
2290 * permitted to overwrite the first word of the object on
2293 * This is the case if we do RCU, have a constructor or
2294 * destructor or are poisoning the objects.
2297 size
+= sizeof(void *);
2300 #ifdef CONFIG_SLUB_DEBUG
2301 if (flags
& SLAB_STORE_USER
)
2303 * Need to store information about allocs and frees after
2306 size
+= 2 * sizeof(struct track
);
2308 if (flags
& SLAB_RED_ZONE
)
2310 * Add some empty padding so that we can catch
2311 * overwrites from earlier objects rather than let
2312 * tracking information or the free pointer be
2313 * corrupted if an user writes before the start
2316 size
+= sizeof(void *);
2320 * Determine the alignment based on various parameters that the
2321 * user specified and the dynamic determination of cache line size
2324 align
= calculate_alignment(flags
, align
, s
->objsize
);
2327 * SLUB stores one object immediately after another beginning from
2328 * offset 0. In order to align the objects we have to simply size
2329 * each object to conform to the alignment.
2331 size
= ALIGN(size
, align
);
2334 if ((flags
& __KMALLOC_CACHE
) &&
2335 PAGE_SIZE
/ size
< slub_min_objects
) {
2337 * Kmalloc cache that would not have enough objects in
2338 * an order 0 page. Kmalloc slabs can fallback to
2339 * page allocator order 0 allocs so take a reasonably large
2340 * order that will allows us a good number of objects.
2342 order
= max(slub_max_order
, PAGE_ALLOC_COSTLY_ORDER
);
2343 s
->flags
|= __PAGE_ALLOC_FALLBACK
;
2344 s
->allocflags
|= __GFP_NOWARN
;
2346 order
= calculate_order(size
);
2353 s
->allocflags
|= __GFP_COMP
;
2355 if (s
->flags
& SLAB_CACHE_DMA
)
2356 s
->allocflags
|= SLUB_DMA
;
2358 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2359 s
->allocflags
|= __GFP_RECLAIMABLE
;
2362 * Determine the number of objects per slab
2364 s
->oo
= oo_make(order
, size
);
2365 s
->min
= oo_make(get_order(size
), size
);
2366 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2369 return !!oo_objects(s
->oo
);
2373 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2374 const char *name
, size_t size
,
2375 size_t align
, unsigned long flags
,
2376 void (*ctor
)(struct kmem_cache
*, void *))
2378 memset(s
, 0, kmem_size
);
2383 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2385 if (!calculate_sizes(s
))
2390 s
->remote_node_defrag_ratio
= 100;
2392 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2395 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2397 free_kmem_cache_nodes(s
);
2399 if (flags
& SLAB_PANIC
)
2400 panic("Cannot create slab %s size=%lu realsize=%u "
2401 "order=%u offset=%u flags=%lx\n",
2402 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2408 * Check if a given pointer is valid
2410 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2414 page
= get_object_page(object
);
2416 if (!page
|| s
!= page
->slab
)
2417 /* No slab or wrong slab */
2420 if (!check_valid_pointer(s
, page
, object
))
2424 * We could also check if the object is on the slabs freelist.
2425 * But this would be too expensive and it seems that the main
2426 * purpose of kmem_ptr_valid() is to check if the object belongs
2427 * to a certain slab.
2431 EXPORT_SYMBOL(kmem_ptr_validate
);
2434 * Determine the size of a slab object
2436 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2440 EXPORT_SYMBOL(kmem_cache_size
);
2442 const char *kmem_cache_name(struct kmem_cache
*s
)
2446 EXPORT_SYMBOL(kmem_cache_name
);
2448 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2451 #ifdef CONFIG_SLUB_DEBUG
2452 void *addr
= page_address(page
);
2454 DECLARE_BITMAP(map
, page
->objects
);
2456 bitmap_zero(map
, page
->objects
);
2457 slab_err(s
, page
, "%s", text
);
2459 for_each_free_object(p
, s
, page
->freelist
)
2460 set_bit(slab_index(p
, s
, addr
), map
);
2462 for_each_object(p
, s
, addr
, page
->objects
) {
2464 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2465 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2467 print_tracking(s
, p
);
2475 * Attempt to free all partial slabs on a node.
2477 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2479 unsigned long flags
;
2480 struct page
*page
, *h
;
2482 spin_lock_irqsave(&n
->list_lock
, flags
);
2483 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2485 list_del(&page
->lru
);
2486 discard_slab(s
, page
);
2489 list_slab_objects(s
, page
,
2490 "Objects remaining on kmem_cache_close()");
2493 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2497 * Release all resources used by a slab cache.
2499 static inline int kmem_cache_close(struct kmem_cache
*s
)
2505 /* Attempt to free all objects */
2506 free_kmem_cache_cpus(s
);
2507 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2508 struct kmem_cache_node
*n
= get_node(s
, node
);
2511 if (n
->nr_partial
|| slabs_node(s
, node
))
2514 free_kmem_cache_nodes(s
);
2519 * Close a cache and release the kmem_cache structure
2520 * (must be used for caches created using kmem_cache_create)
2522 void kmem_cache_destroy(struct kmem_cache
*s
)
2524 down_write(&slub_lock
);
2528 up_write(&slub_lock
);
2529 if (kmem_cache_close(s
)) {
2530 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2531 "still has objects.\n", s
->name
, __func__
);
2534 sysfs_slab_remove(s
);
2536 up_write(&slub_lock
);
2538 EXPORT_SYMBOL(kmem_cache_destroy
);
2540 /********************************************************************
2542 *******************************************************************/
2544 struct kmem_cache kmalloc_caches
[PAGE_SHIFT
+ 1] __cacheline_aligned
;
2545 EXPORT_SYMBOL(kmalloc_caches
);
2547 static int __init
setup_slub_min_order(char *str
)
2549 get_option(&str
, &slub_min_order
);
2554 __setup("slub_min_order=", setup_slub_min_order
);
2556 static int __init
setup_slub_max_order(char *str
)
2558 get_option(&str
, &slub_max_order
);
2563 __setup("slub_max_order=", setup_slub_max_order
);
2565 static int __init
setup_slub_min_objects(char *str
)
2567 get_option(&str
, &slub_min_objects
);
2572 __setup("slub_min_objects=", setup_slub_min_objects
);
2574 static int __init
setup_slub_nomerge(char *str
)
2580 __setup("slub_nomerge", setup_slub_nomerge
);
2582 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2583 const char *name
, int size
, gfp_t gfp_flags
)
2585 unsigned int flags
= 0;
2587 if (gfp_flags
& SLUB_DMA
)
2588 flags
= SLAB_CACHE_DMA
;
2590 down_write(&slub_lock
);
2591 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2592 flags
| __KMALLOC_CACHE
, NULL
))
2595 list_add(&s
->list
, &slab_caches
);
2596 up_write(&slub_lock
);
2597 if (sysfs_slab_add(s
))
2602 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2605 #ifdef CONFIG_ZONE_DMA
2606 static struct kmem_cache
*kmalloc_caches_dma
[PAGE_SHIFT
+ 1];
2608 static void sysfs_add_func(struct work_struct
*w
)
2610 struct kmem_cache
*s
;
2612 down_write(&slub_lock
);
2613 list_for_each_entry(s
, &slab_caches
, list
) {
2614 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2615 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2619 up_write(&slub_lock
);
2622 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2624 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2626 struct kmem_cache
*s
;
2630 s
= kmalloc_caches_dma
[index
];
2634 /* Dynamically create dma cache */
2635 if (flags
& __GFP_WAIT
)
2636 down_write(&slub_lock
);
2638 if (!down_write_trylock(&slub_lock
))
2642 if (kmalloc_caches_dma
[index
])
2645 realsize
= kmalloc_caches
[index
].objsize
;
2646 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2647 (unsigned int)realsize
);
2648 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2650 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2651 realsize
, ARCH_KMALLOC_MINALIGN
,
2652 SLAB_CACHE_DMA
|__SYSFS_ADD_DEFERRED
, NULL
)) {
2658 list_add(&s
->list
, &slab_caches
);
2659 kmalloc_caches_dma
[index
] = s
;
2661 schedule_work(&sysfs_add_work
);
2664 up_write(&slub_lock
);
2666 return kmalloc_caches_dma
[index
];
2671 * Conversion table for small slabs sizes / 8 to the index in the
2672 * kmalloc array. This is necessary for slabs < 192 since we have non power
2673 * of two cache sizes there. The size of larger slabs can be determined using
2676 static s8 size_index
[24] = {
2703 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2709 return ZERO_SIZE_PTR
;
2711 index
= size_index
[(size
- 1) / 8];
2713 index
= fls(size
- 1);
2715 #ifdef CONFIG_ZONE_DMA
2716 if (unlikely((flags
& SLUB_DMA
)))
2717 return dma_kmalloc_cache(index
, flags
);
2720 return &kmalloc_caches
[index
];
2723 void *__kmalloc(size_t size
, gfp_t flags
)
2725 struct kmem_cache
*s
;
2727 if (unlikely(size
> PAGE_SIZE
))
2728 return kmalloc_large(size
, flags
);
2730 s
= get_slab(size
, flags
);
2732 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2735 return slab_alloc(s
, flags
, -1, __builtin_return_address(0));
2737 EXPORT_SYMBOL(__kmalloc
);
2739 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2741 struct page
*page
= alloc_pages_node(node
, flags
| __GFP_COMP
,
2745 return page_address(page
);
2751 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2753 struct kmem_cache
*s
;
2755 if (unlikely(size
> PAGE_SIZE
))
2756 return kmalloc_large_node(size
, flags
, node
);
2758 s
= get_slab(size
, flags
);
2760 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2763 return slab_alloc(s
, flags
, node
, __builtin_return_address(0));
2765 EXPORT_SYMBOL(__kmalloc_node
);
2768 size_t ksize(const void *object
)
2771 struct kmem_cache
*s
;
2773 if (unlikely(object
== ZERO_SIZE_PTR
))
2776 page
= virt_to_head_page(object
);
2778 if (unlikely(!PageSlab(page
)))
2779 return PAGE_SIZE
<< compound_order(page
);
2783 #ifdef CONFIG_SLUB_DEBUG
2785 * Debugging requires use of the padding between object
2786 * and whatever may come after it.
2788 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2793 * If we have the need to store the freelist pointer
2794 * back there or track user information then we can
2795 * only use the space before that information.
2797 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2800 * Else we can use all the padding etc for the allocation
2804 EXPORT_SYMBOL(ksize
);
2806 void kfree(const void *x
)
2809 void *object
= (void *)x
;
2811 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2814 page
= virt_to_head_page(x
);
2815 if (unlikely(!PageSlab(page
))) {
2819 slab_free(page
->slab
, page
, object
, __builtin_return_address(0));
2821 EXPORT_SYMBOL(kfree
);
2824 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2825 * the remaining slabs by the number of items in use. The slabs with the
2826 * most items in use come first. New allocations will then fill those up
2827 * and thus they can be removed from the partial lists.
2829 * The slabs with the least items are placed last. This results in them
2830 * being allocated from last increasing the chance that the last objects
2831 * are freed in them.
2833 int kmem_cache_shrink(struct kmem_cache
*s
)
2837 struct kmem_cache_node
*n
;
2840 int objects
= oo_objects(s
->max
);
2841 struct list_head
*slabs_by_inuse
=
2842 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2843 unsigned long flags
;
2845 if (!slabs_by_inuse
)
2849 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2850 n
= get_node(s
, node
);
2855 for (i
= 0; i
< objects
; i
++)
2856 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2858 spin_lock_irqsave(&n
->list_lock
, flags
);
2861 * Build lists indexed by the items in use in each slab.
2863 * Note that concurrent frees may occur while we hold the
2864 * list_lock. page->inuse here is the upper limit.
2866 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2867 if (!page
->inuse
&& slab_trylock(page
)) {
2869 * Must hold slab lock here because slab_free
2870 * may have freed the last object and be
2871 * waiting to release the slab.
2873 list_del(&page
->lru
);
2876 discard_slab(s
, page
);
2878 list_move(&page
->lru
,
2879 slabs_by_inuse
+ page
->inuse
);
2884 * Rebuild the partial list with the slabs filled up most
2885 * first and the least used slabs at the end.
2887 for (i
= objects
- 1; i
>= 0; i
--)
2888 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2890 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2893 kfree(slabs_by_inuse
);
2896 EXPORT_SYMBOL(kmem_cache_shrink
);
2898 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2899 static int slab_mem_going_offline_callback(void *arg
)
2901 struct kmem_cache
*s
;
2903 down_read(&slub_lock
);
2904 list_for_each_entry(s
, &slab_caches
, list
)
2905 kmem_cache_shrink(s
);
2906 up_read(&slub_lock
);
2911 static void slab_mem_offline_callback(void *arg
)
2913 struct kmem_cache_node
*n
;
2914 struct kmem_cache
*s
;
2915 struct memory_notify
*marg
= arg
;
2918 offline_node
= marg
->status_change_nid
;
2921 * If the node still has available memory. we need kmem_cache_node
2924 if (offline_node
< 0)
2927 down_read(&slub_lock
);
2928 list_for_each_entry(s
, &slab_caches
, list
) {
2929 n
= get_node(s
, offline_node
);
2932 * if n->nr_slabs > 0, slabs still exist on the node
2933 * that is going down. We were unable to free them,
2934 * and offline_pages() function shoudn't call this
2935 * callback. So, we must fail.
2937 BUG_ON(slabs_node(s
, offline_node
));
2939 s
->node
[offline_node
] = NULL
;
2940 kmem_cache_free(kmalloc_caches
, n
);
2943 up_read(&slub_lock
);
2946 static int slab_mem_going_online_callback(void *arg
)
2948 struct kmem_cache_node
*n
;
2949 struct kmem_cache
*s
;
2950 struct memory_notify
*marg
= arg
;
2951 int nid
= marg
->status_change_nid
;
2955 * If the node's memory is already available, then kmem_cache_node is
2956 * already created. Nothing to do.
2962 * We are bringing a node online. No memory is availabe yet. We must
2963 * allocate a kmem_cache_node structure in order to bring the node
2966 down_read(&slub_lock
);
2967 list_for_each_entry(s
, &slab_caches
, list
) {
2969 * XXX: kmem_cache_alloc_node will fallback to other nodes
2970 * since memory is not yet available from the node that
2973 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2978 init_kmem_cache_node(n
);
2982 up_read(&slub_lock
);
2986 static int slab_memory_callback(struct notifier_block
*self
,
2987 unsigned long action
, void *arg
)
2992 case MEM_GOING_ONLINE
:
2993 ret
= slab_mem_going_online_callback(arg
);
2995 case MEM_GOING_OFFLINE
:
2996 ret
= slab_mem_going_offline_callback(arg
);
2999 case MEM_CANCEL_ONLINE
:
3000 slab_mem_offline_callback(arg
);
3003 case MEM_CANCEL_OFFLINE
:
3007 ret
= notifier_from_errno(ret
);
3011 #endif /* CONFIG_MEMORY_HOTPLUG */
3013 /********************************************************************
3014 * Basic setup of slabs
3015 *******************************************************************/
3017 void __init
kmem_cache_init(void)
3026 * Must first have the slab cache available for the allocations of the
3027 * struct kmem_cache_node's. There is special bootstrap code in
3028 * kmem_cache_open for slab_state == DOWN.
3030 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3031 sizeof(struct kmem_cache_node
), GFP_KERNEL
);
3032 kmalloc_caches
[0].refcount
= -1;
3035 hotplug_memory_notifier(slab_memory_callback
, 1);
3038 /* Able to allocate the per node structures */
3039 slab_state
= PARTIAL
;
3041 /* Caches that are not of the two-to-the-power-of size */
3042 if (KMALLOC_MIN_SIZE
<= 64) {
3043 create_kmalloc_cache(&kmalloc_caches
[1],
3044 "kmalloc-96", 96, GFP_KERNEL
);
3047 if (KMALLOC_MIN_SIZE
<= 128) {
3048 create_kmalloc_cache(&kmalloc_caches
[2],
3049 "kmalloc-192", 192, GFP_KERNEL
);
3053 for (i
= KMALLOC_SHIFT_LOW
; i
<= PAGE_SHIFT
; i
++) {
3054 create_kmalloc_cache(&kmalloc_caches
[i
],
3055 "kmalloc", 1 << i
, GFP_KERNEL
);
3061 * Patch up the size_index table if we have strange large alignment
3062 * requirements for the kmalloc array. This is only the case for
3063 * MIPS it seems. The standard arches will not generate any code here.
3065 * Largest permitted alignment is 256 bytes due to the way we
3066 * handle the index determination for the smaller caches.
3068 * Make sure that nothing crazy happens if someone starts tinkering
3069 * around with ARCH_KMALLOC_MINALIGN
3071 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3072 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3074 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3075 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3079 /* Provide the correct kmalloc names now that the caches are up */
3080 for (i
= KMALLOC_SHIFT_LOW
; i
<= PAGE_SHIFT
; i
++)
3081 kmalloc_caches
[i
]. name
=
3082 kasprintf(GFP_KERNEL
, "kmalloc-%d", 1 << i
);
3085 register_cpu_notifier(&slab_notifier
);
3086 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3087 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3089 kmem_size
= sizeof(struct kmem_cache
);
3093 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3094 " CPUs=%d, Nodes=%d\n",
3095 caches
, cache_line_size(),
3096 slub_min_order
, slub_max_order
, slub_min_objects
,
3097 nr_cpu_ids
, nr_node_ids
);
3101 * Find a mergeable slab cache
3103 static int slab_unmergeable(struct kmem_cache
*s
)
3105 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3108 if ((s
->flags
& __PAGE_ALLOC_FALLBACK
))
3115 * We may have set a slab to be unmergeable during bootstrap.
3117 if (s
->refcount
< 0)
3123 static struct kmem_cache
*find_mergeable(size_t size
,
3124 size_t align
, unsigned long flags
, const char *name
,
3125 void (*ctor
)(struct kmem_cache
*, void *))
3127 struct kmem_cache
*s
;
3129 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3135 size
= ALIGN(size
, sizeof(void *));
3136 align
= calculate_alignment(flags
, align
, size
);
3137 size
= ALIGN(size
, align
);
3138 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3140 list_for_each_entry(s
, &slab_caches
, list
) {
3141 if (slab_unmergeable(s
))
3147 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3150 * Check if alignment is compatible.
3151 * Courtesy of Adrian Drzewiecki
3153 if ((s
->size
& ~(align
- 1)) != s
->size
)
3156 if (s
->size
- size
>= sizeof(void *))
3164 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3165 size_t align
, unsigned long flags
,
3166 void (*ctor
)(struct kmem_cache
*, void *))
3168 struct kmem_cache
*s
;
3170 down_write(&slub_lock
);
3171 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3177 * Adjust the object sizes so that we clear
3178 * the complete object on kzalloc.
3180 s
->objsize
= max(s
->objsize
, (int)size
);
3183 * And then we need to update the object size in the
3184 * per cpu structures
3186 for_each_online_cpu(cpu
)
3187 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3189 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3190 up_write(&slub_lock
);
3192 if (sysfs_slab_alias(s
, name
))
3197 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3199 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3200 size
, align
, flags
, ctor
)) {
3201 list_add(&s
->list
, &slab_caches
);
3202 up_write(&slub_lock
);
3203 if (sysfs_slab_add(s
))
3209 up_write(&slub_lock
);
3212 if (flags
& SLAB_PANIC
)
3213 panic("Cannot create slabcache %s\n", name
);
3218 EXPORT_SYMBOL(kmem_cache_create
);
3222 * Use the cpu notifier to insure that the cpu slabs are flushed when
3225 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3226 unsigned long action
, void *hcpu
)
3228 long cpu
= (long)hcpu
;
3229 struct kmem_cache
*s
;
3230 unsigned long flags
;
3233 case CPU_UP_PREPARE
:
3234 case CPU_UP_PREPARE_FROZEN
:
3235 init_alloc_cpu_cpu(cpu
);
3236 down_read(&slub_lock
);
3237 list_for_each_entry(s
, &slab_caches
, list
)
3238 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3240 up_read(&slub_lock
);
3243 case CPU_UP_CANCELED
:
3244 case CPU_UP_CANCELED_FROZEN
:
3246 case CPU_DEAD_FROZEN
:
3247 down_read(&slub_lock
);
3248 list_for_each_entry(s
, &slab_caches
, list
) {
3249 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3251 local_irq_save(flags
);
3252 __flush_cpu_slab(s
, cpu
);
3253 local_irq_restore(flags
);
3254 free_kmem_cache_cpu(c
, cpu
);
3255 s
->cpu_slab
[cpu
] = NULL
;
3257 up_read(&slub_lock
);
3265 static struct notifier_block __cpuinitdata slab_notifier
= {
3266 .notifier_call
= slab_cpuup_callback
3271 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, void *caller
)
3273 struct kmem_cache
*s
;
3275 if (unlikely(size
> PAGE_SIZE
))
3276 return kmalloc_large(size
, gfpflags
);
3278 s
= get_slab(size
, gfpflags
);
3280 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3283 return slab_alloc(s
, gfpflags
, -1, caller
);
3286 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3287 int node
, void *caller
)
3289 struct kmem_cache
*s
;
3291 if (unlikely(size
> PAGE_SIZE
))
3292 return kmalloc_large_node(size
, gfpflags
, node
);
3294 s
= get_slab(size
, gfpflags
);
3296 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3299 return slab_alloc(s
, gfpflags
, node
, caller
);
3302 #if (defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)) || defined(CONFIG_SLABINFO)
3303 static unsigned long count_partial(struct kmem_cache_node
*n
,
3304 int (*get_count
)(struct page
*))
3306 unsigned long flags
;
3307 unsigned long x
= 0;
3310 spin_lock_irqsave(&n
->list_lock
, flags
);
3311 list_for_each_entry(page
, &n
->partial
, lru
)
3312 x
+= get_count(page
);
3313 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3317 static int count_inuse(struct page
*page
)
3322 static int count_total(struct page
*page
)
3324 return page
->objects
;
3327 static int count_free(struct page
*page
)
3329 return page
->objects
- page
->inuse
;
3333 #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)
3334 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3338 void *addr
= page_address(page
);
3340 if (!check_slab(s
, page
) ||
3341 !on_freelist(s
, page
, NULL
))
3344 /* Now we know that a valid freelist exists */
3345 bitmap_zero(map
, page
->objects
);
3347 for_each_free_object(p
, s
, page
->freelist
) {
3348 set_bit(slab_index(p
, s
, addr
), map
);
3349 if (!check_object(s
, page
, p
, 0))
3353 for_each_object(p
, s
, addr
, page
->objects
)
3354 if (!test_bit(slab_index(p
, s
, addr
), map
))
3355 if (!check_object(s
, page
, p
, 1))
3360 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3363 if (slab_trylock(page
)) {
3364 validate_slab(s
, page
, map
);
3367 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3370 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3371 if (!SlabDebug(page
))
3372 printk(KERN_ERR
"SLUB %s: SlabDebug not set "
3373 "on slab 0x%p\n", s
->name
, page
);
3375 if (SlabDebug(page
))
3376 printk(KERN_ERR
"SLUB %s: SlabDebug set on "
3377 "slab 0x%p\n", s
->name
, page
);
3381 static int validate_slab_node(struct kmem_cache
*s
,
3382 struct kmem_cache_node
*n
, unsigned long *map
)
3384 unsigned long count
= 0;
3386 unsigned long flags
;
3388 spin_lock_irqsave(&n
->list_lock
, flags
);
3390 list_for_each_entry(page
, &n
->partial
, lru
) {
3391 validate_slab_slab(s
, page
, map
);
3394 if (count
!= n
->nr_partial
)
3395 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3396 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3398 if (!(s
->flags
& SLAB_STORE_USER
))
3401 list_for_each_entry(page
, &n
->full
, lru
) {
3402 validate_slab_slab(s
, page
, map
);
3405 if (count
!= atomic_long_read(&n
->nr_slabs
))
3406 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3407 "counter=%ld\n", s
->name
, count
,
3408 atomic_long_read(&n
->nr_slabs
));
3411 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3415 static long validate_slab_cache(struct kmem_cache
*s
)
3418 unsigned long count
= 0;
3419 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3420 sizeof(unsigned long), GFP_KERNEL
);
3426 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3427 struct kmem_cache_node
*n
= get_node(s
, node
);
3429 count
+= validate_slab_node(s
, n
, map
);
3435 #ifdef SLUB_RESILIENCY_TEST
3436 static void resiliency_test(void)
3440 printk(KERN_ERR
"SLUB resiliency testing\n");
3441 printk(KERN_ERR
"-----------------------\n");
3442 printk(KERN_ERR
"A. Corruption after allocation\n");
3444 p
= kzalloc(16, GFP_KERNEL
);
3446 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3447 " 0x12->0x%p\n\n", p
+ 16);
3449 validate_slab_cache(kmalloc_caches
+ 4);
3451 /* Hmmm... The next two are dangerous */
3452 p
= kzalloc(32, GFP_KERNEL
);
3453 p
[32 + sizeof(void *)] = 0x34;
3454 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3455 " 0x34 -> -0x%p\n", p
);
3457 "If allocated object is overwritten then not detectable\n\n");
3459 validate_slab_cache(kmalloc_caches
+ 5);
3460 p
= kzalloc(64, GFP_KERNEL
);
3461 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3463 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3466 "If allocated object is overwritten then not detectable\n\n");
3467 validate_slab_cache(kmalloc_caches
+ 6);
3469 printk(KERN_ERR
"\nB. Corruption after free\n");
3470 p
= kzalloc(128, GFP_KERNEL
);
3473 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3474 validate_slab_cache(kmalloc_caches
+ 7);
3476 p
= kzalloc(256, GFP_KERNEL
);
3479 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3481 validate_slab_cache(kmalloc_caches
+ 8);
3483 p
= kzalloc(512, GFP_KERNEL
);
3486 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3487 validate_slab_cache(kmalloc_caches
+ 9);
3490 static void resiliency_test(void) {};
3494 * Generate lists of code addresses where slabcache objects are allocated
3499 unsigned long count
;
3512 unsigned long count
;
3513 struct location
*loc
;
3516 static void free_loc_track(struct loc_track
*t
)
3519 free_pages((unsigned long)t
->loc
,
3520 get_order(sizeof(struct location
) * t
->max
));
3523 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3528 order
= get_order(sizeof(struct location
) * max
);
3530 l
= (void *)__get_free_pages(flags
, order
);
3535 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3543 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3544 const struct track
*track
)
3546 long start
, end
, pos
;
3549 unsigned long age
= jiffies
- track
->when
;
3555 pos
= start
+ (end
- start
+ 1) / 2;
3558 * There is nothing at "end". If we end up there
3559 * we need to add something to before end.
3564 caddr
= t
->loc
[pos
].addr
;
3565 if (track
->addr
== caddr
) {
3571 if (age
< l
->min_time
)
3573 if (age
> l
->max_time
)
3576 if (track
->pid
< l
->min_pid
)
3577 l
->min_pid
= track
->pid
;
3578 if (track
->pid
> l
->max_pid
)
3579 l
->max_pid
= track
->pid
;
3581 cpu_set(track
->cpu
, l
->cpus
);
3583 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3587 if (track
->addr
< caddr
)
3594 * Not found. Insert new tracking element.
3596 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3602 (t
->count
- pos
) * sizeof(struct location
));
3605 l
->addr
= track
->addr
;
3609 l
->min_pid
= track
->pid
;
3610 l
->max_pid
= track
->pid
;
3611 cpus_clear(l
->cpus
);
3612 cpu_set(track
->cpu
, l
->cpus
);
3613 nodes_clear(l
->nodes
);
3614 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3618 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3619 struct page
*page
, enum track_item alloc
)
3621 void *addr
= page_address(page
);
3622 DECLARE_BITMAP(map
, page
->objects
);
3625 bitmap_zero(map
, page
->objects
);
3626 for_each_free_object(p
, s
, page
->freelist
)
3627 set_bit(slab_index(p
, s
, addr
), map
);
3629 for_each_object(p
, s
, addr
, page
->objects
)
3630 if (!test_bit(slab_index(p
, s
, addr
), map
))
3631 add_location(t
, s
, get_track(s
, p
, alloc
));
3634 static int list_locations(struct kmem_cache
*s
, char *buf
,
3635 enum track_item alloc
)
3639 struct loc_track t
= { 0, 0, NULL
};
3642 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3644 return sprintf(buf
, "Out of memory\n");
3646 /* Push back cpu slabs */
3649 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3650 struct kmem_cache_node
*n
= get_node(s
, node
);
3651 unsigned long flags
;
3654 if (!atomic_long_read(&n
->nr_slabs
))
3657 spin_lock_irqsave(&n
->list_lock
, flags
);
3658 list_for_each_entry(page
, &n
->partial
, lru
)
3659 process_slab(&t
, s
, page
, alloc
);
3660 list_for_each_entry(page
, &n
->full
, lru
)
3661 process_slab(&t
, s
, page
, alloc
);
3662 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3665 for (i
= 0; i
< t
.count
; i
++) {
3666 struct location
*l
= &t
.loc
[i
];
3668 if (len
> PAGE_SIZE
- 100)
3670 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3673 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3675 len
+= sprintf(buf
+ len
, "<not-available>");
3677 if (l
->sum_time
!= l
->min_time
) {
3678 unsigned long remainder
;
3680 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3682 div_long_long_rem(l
->sum_time
, l
->count
, &remainder
),
3685 len
+= sprintf(buf
+ len
, " age=%ld",
3688 if (l
->min_pid
!= l
->max_pid
)
3689 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3690 l
->min_pid
, l
->max_pid
);
3692 len
+= sprintf(buf
+ len
, " pid=%ld",
3695 if (num_online_cpus() > 1 && !cpus_empty(l
->cpus
) &&
3696 len
< PAGE_SIZE
- 60) {
3697 len
+= sprintf(buf
+ len
, " cpus=");
3698 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3702 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3703 len
< PAGE_SIZE
- 60) {
3704 len
+= sprintf(buf
+ len
, " nodes=");
3705 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3709 len
+= sprintf(buf
+ len
, "\n");
3714 len
+= sprintf(buf
, "No data\n");
3718 enum slab_stat_type
{
3719 SL_ALL
, /* All slabs */
3720 SL_PARTIAL
, /* Only partially allocated slabs */
3721 SL_CPU
, /* Only slabs used for cpu caches */
3722 SL_OBJECTS
, /* Determine allocated objects not slabs */
3723 SL_TOTAL
/* Determine object capacity not slabs */
3726 #define SO_ALL (1 << SL_ALL)
3727 #define SO_PARTIAL (1 << SL_PARTIAL)
3728 #define SO_CPU (1 << SL_CPU)
3729 #define SO_OBJECTS (1 << SL_OBJECTS)
3730 #define SO_TOTAL (1 << SL_TOTAL)
3732 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3733 char *buf
, unsigned long flags
)
3735 unsigned long total
= 0;
3738 unsigned long *nodes
;
3739 unsigned long *per_cpu
;
3741 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3744 per_cpu
= nodes
+ nr_node_ids
;
3746 if (flags
& SO_CPU
) {
3749 for_each_possible_cpu(cpu
) {
3750 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3752 if (!c
|| c
->node
< 0)
3756 if (flags
& SO_TOTAL
)
3757 x
= c
->page
->objects
;
3758 else if (flags
& SO_OBJECTS
)
3764 nodes
[c
->node
] += x
;
3770 if (flags
& SO_ALL
) {
3771 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3772 struct kmem_cache_node
*n
= get_node(s
, node
);
3774 if (flags
& SO_TOTAL
)
3775 x
= atomic_long_read(&n
->total_objects
);
3776 else if (flags
& SO_OBJECTS
)
3777 x
= atomic_long_read(&n
->total_objects
) -
3778 count_partial(n
, count_free
);
3781 x
= atomic_long_read(&n
->nr_slabs
);
3786 } else if (flags
& SO_PARTIAL
) {
3787 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3788 struct kmem_cache_node
*n
= get_node(s
, node
);
3790 if (flags
& SO_TOTAL
)
3791 x
= count_partial(n
, count_total
);
3792 else if (flags
& SO_OBJECTS
)
3793 x
= count_partial(n
, count_inuse
);
3800 x
= sprintf(buf
, "%lu", total
);
3802 for_each_node_state(node
, N_NORMAL_MEMORY
)
3804 x
+= sprintf(buf
+ x
, " N%d=%lu",
3808 return x
+ sprintf(buf
+ x
, "\n");
3811 static int any_slab_objects(struct kmem_cache
*s
)
3816 for_each_possible_cpu(cpu
) {
3817 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3823 for_each_online_node(node
) {
3824 struct kmem_cache_node
*n
= get_node(s
, node
);
3829 if (n
->nr_partial
|| atomic_long_read(&n
->nr_slabs
))
3835 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3836 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3838 struct slab_attribute
{
3839 struct attribute attr
;
3840 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3841 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3844 #define SLAB_ATTR_RO(_name) \
3845 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3847 #define SLAB_ATTR(_name) \
3848 static struct slab_attribute _name##_attr = \
3849 __ATTR(_name, 0644, _name##_show, _name##_store)
3851 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3853 return sprintf(buf
, "%d\n", s
->size
);
3855 SLAB_ATTR_RO(slab_size
);
3857 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3859 return sprintf(buf
, "%d\n", s
->align
);
3861 SLAB_ATTR_RO(align
);
3863 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3865 return sprintf(buf
, "%d\n", s
->objsize
);
3867 SLAB_ATTR_RO(object_size
);
3869 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3871 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3873 SLAB_ATTR_RO(objs_per_slab
);
3875 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3877 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3879 SLAB_ATTR_RO(order
);
3881 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3884 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3886 return n
+ sprintf(buf
+ n
, "\n");
3892 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3894 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3896 SLAB_ATTR_RO(aliases
);
3898 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3900 return show_slab_objects(s
, buf
, SO_ALL
);
3902 SLAB_ATTR_RO(slabs
);
3904 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3906 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3908 SLAB_ATTR_RO(partial
);
3910 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3912 return show_slab_objects(s
, buf
, SO_CPU
);
3914 SLAB_ATTR_RO(cpu_slabs
);
3916 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3918 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3920 SLAB_ATTR_RO(objects
);
3922 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3924 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3926 SLAB_ATTR_RO(objects_partial
);
3928 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3930 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3932 SLAB_ATTR_RO(total_objects
);
3934 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3936 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3939 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3940 const char *buf
, size_t length
)
3942 s
->flags
&= ~SLAB_DEBUG_FREE
;
3944 s
->flags
|= SLAB_DEBUG_FREE
;
3947 SLAB_ATTR(sanity_checks
);
3949 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
3951 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
3954 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
3957 s
->flags
&= ~SLAB_TRACE
;
3959 s
->flags
|= SLAB_TRACE
;
3964 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
3966 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
3969 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
3970 const char *buf
, size_t length
)
3972 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
3974 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
3977 SLAB_ATTR(reclaim_account
);
3979 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
3981 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
3983 SLAB_ATTR_RO(hwcache_align
);
3985 #ifdef CONFIG_ZONE_DMA
3986 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
3988 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
3990 SLAB_ATTR_RO(cache_dma
);
3993 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
3995 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
3997 SLAB_ATTR_RO(destroy_by_rcu
);
3999 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4001 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4004 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4005 const char *buf
, size_t length
)
4007 if (any_slab_objects(s
))
4010 s
->flags
&= ~SLAB_RED_ZONE
;
4012 s
->flags
|= SLAB_RED_ZONE
;
4016 SLAB_ATTR(red_zone
);
4018 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4020 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4023 static ssize_t
poison_store(struct kmem_cache
*s
,
4024 const char *buf
, size_t length
)
4026 if (any_slab_objects(s
))
4029 s
->flags
&= ~SLAB_POISON
;
4031 s
->flags
|= SLAB_POISON
;
4037 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4039 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4042 static ssize_t
store_user_store(struct kmem_cache
*s
,
4043 const char *buf
, size_t length
)
4045 if (any_slab_objects(s
))
4048 s
->flags
&= ~SLAB_STORE_USER
;
4050 s
->flags
|= SLAB_STORE_USER
;
4054 SLAB_ATTR(store_user
);
4056 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4061 static ssize_t
validate_store(struct kmem_cache
*s
,
4062 const char *buf
, size_t length
)
4066 if (buf
[0] == '1') {
4067 ret
= validate_slab_cache(s
);
4073 SLAB_ATTR(validate
);
4075 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4080 static ssize_t
shrink_store(struct kmem_cache
*s
,
4081 const char *buf
, size_t length
)
4083 if (buf
[0] == '1') {
4084 int rc
= kmem_cache_shrink(s
);
4094 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4096 if (!(s
->flags
& SLAB_STORE_USER
))
4098 return list_locations(s
, buf
, TRACK_ALLOC
);
4100 SLAB_ATTR_RO(alloc_calls
);
4102 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4104 if (!(s
->flags
& SLAB_STORE_USER
))
4106 return list_locations(s
, buf
, TRACK_FREE
);
4108 SLAB_ATTR_RO(free_calls
);
4111 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4113 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4116 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4117 const char *buf
, size_t length
)
4119 int n
= simple_strtoul(buf
, NULL
, 10);
4122 s
->remote_node_defrag_ratio
= n
* 10;
4125 SLAB_ATTR(remote_node_defrag_ratio
);
4128 #ifdef CONFIG_SLUB_STATS
4129 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4131 unsigned long sum
= 0;
4134 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4139 for_each_online_cpu(cpu
) {
4140 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4146 len
= sprintf(buf
, "%lu", sum
);
4149 for_each_online_cpu(cpu
) {
4150 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4151 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4155 return len
+ sprintf(buf
+ len
, "\n");
4158 #define STAT_ATTR(si, text) \
4159 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4161 return show_stat(s, buf, si); \
4163 SLAB_ATTR_RO(text); \
4165 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4166 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4167 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4168 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4169 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4170 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4171 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4172 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4173 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4174 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4175 STAT_ATTR(FREE_SLAB
, free_slab
);
4176 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4177 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4178 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4179 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4180 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4181 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4182 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4185 static struct attribute
*slab_attrs
[] = {
4186 &slab_size_attr
.attr
,
4187 &object_size_attr
.attr
,
4188 &objs_per_slab_attr
.attr
,
4191 &objects_partial_attr
.attr
,
4192 &total_objects_attr
.attr
,
4195 &cpu_slabs_attr
.attr
,
4199 &sanity_checks_attr
.attr
,
4201 &hwcache_align_attr
.attr
,
4202 &reclaim_account_attr
.attr
,
4203 &destroy_by_rcu_attr
.attr
,
4204 &red_zone_attr
.attr
,
4206 &store_user_attr
.attr
,
4207 &validate_attr
.attr
,
4209 &alloc_calls_attr
.attr
,
4210 &free_calls_attr
.attr
,
4211 #ifdef CONFIG_ZONE_DMA
4212 &cache_dma_attr
.attr
,
4215 &remote_node_defrag_ratio_attr
.attr
,
4217 #ifdef CONFIG_SLUB_STATS
4218 &alloc_fastpath_attr
.attr
,
4219 &alloc_slowpath_attr
.attr
,
4220 &free_fastpath_attr
.attr
,
4221 &free_slowpath_attr
.attr
,
4222 &free_frozen_attr
.attr
,
4223 &free_add_partial_attr
.attr
,
4224 &free_remove_partial_attr
.attr
,
4225 &alloc_from_partial_attr
.attr
,
4226 &alloc_slab_attr
.attr
,
4227 &alloc_refill_attr
.attr
,
4228 &free_slab_attr
.attr
,
4229 &cpuslab_flush_attr
.attr
,
4230 &deactivate_full_attr
.attr
,
4231 &deactivate_empty_attr
.attr
,
4232 &deactivate_to_head_attr
.attr
,
4233 &deactivate_to_tail_attr
.attr
,
4234 &deactivate_remote_frees_attr
.attr
,
4235 &order_fallback_attr
.attr
,
4240 static struct attribute_group slab_attr_group
= {
4241 .attrs
= slab_attrs
,
4244 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4245 struct attribute
*attr
,
4248 struct slab_attribute
*attribute
;
4249 struct kmem_cache
*s
;
4252 attribute
= to_slab_attr(attr
);
4255 if (!attribute
->show
)
4258 err
= attribute
->show(s
, buf
);
4263 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4264 struct attribute
*attr
,
4265 const char *buf
, size_t len
)
4267 struct slab_attribute
*attribute
;
4268 struct kmem_cache
*s
;
4271 attribute
= to_slab_attr(attr
);
4274 if (!attribute
->store
)
4277 err
= attribute
->store(s
, buf
, len
);
4282 static void kmem_cache_release(struct kobject
*kobj
)
4284 struct kmem_cache
*s
= to_slab(kobj
);
4289 static struct sysfs_ops slab_sysfs_ops
= {
4290 .show
= slab_attr_show
,
4291 .store
= slab_attr_store
,
4294 static struct kobj_type slab_ktype
= {
4295 .sysfs_ops
= &slab_sysfs_ops
,
4296 .release
= kmem_cache_release
4299 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4301 struct kobj_type
*ktype
= get_ktype(kobj
);
4303 if (ktype
== &slab_ktype
)
4308 static struct kset_uevent_ops slab_uevent_ops
= {
4309 .filter
= uevent_filter
,
4312 static struct kset
*slab_kset
;
4314 #define ID_STR_LENGTH 64
4316 /* Create a unique string id for a slab cache:
4318 * Format :[flags-]size
4320 static char *create_unique_id(struct kmem_cache
*s
)
4322 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4329 * First flags affecting slabcache operations. We will only
4330 * get here for aliasable slabs so we do not need to support
4331 * too many flags. The flags here must cover all flags that
4332 * are matched during merging to guarantee that the id is
4335 if (s
->flags
& SLAB_CACHE_DMA
)
4337 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4339 if (s
->flags
& SLAB_DEBUG_FREE
)
4343 p
+= sprintf(p
, "%07d", s
->size
);
4344 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4348 static int sysfs_slab_add(struct kmem_cache
*s
)
4354 if (slab_state
< SYSFS
)
4355 /* Defer until later */
4358 unmergeable
= slab_unmergeable(s
);
4361 * Slabcache can never be merged so we can use the name proper.
4362 * This is typically the case for debug situations. In that
4363 * case we can catch duplicate names easily.
4365 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4369 * Create a unique name for the slab as a target
4372 name
= create_unique_id(s
);
4375 s
->kobj
.kset
= slab_kset
;
4376 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4378 kobject_put(&s
->kobj
);
4382 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4385 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4387 /* Setup first alias */
4388 sysfs_slab_alias(s
, s
->name
);
4394 static void sysfs_slab_remove(struct kmem_cache
*s
)
4396 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4397 kobject_del(&s
->kobj
);
4398 kobject_put(&s
->kobj
);
4402 * Need to buffer aliases during bootup until sysfs becomes
4403 * available lest we loose that information.
4405 struct saved_alias
{
4406 struct kmem_cache
*s
;
4408 struct saved_alias
*next
;
4411 static struct saved_alias
*alias_list
;
4413 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4415 struct saved_alias
*al
;
4417 if (slab_state
== SYSFS
) {
4419 * If we have a leftover link then remove it.
4421 sysfs_remove_link(&slab_kset
->kobj
, name
);
4422 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4425 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4431 al
->next
= alias_list
;
4436 static int __init
slab_sysfs_init(void)
4438 struct kmem_cache
*s
;
4441 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4443 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4449 list_for_each_entry(s
, &slab_caches
, list
) {
4450 err
= sysfs_slab_add(s
);
4452 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4453 " to sysfs\n", s
->name
);
4456 while (alias_list
) {
4457 struct saved_alias
*al
= alias_list
;
4459 alias_list
= alias_list
->next
;
4460 err
= sysfs_slab_alias(al
->s
, al
->name
);
4462 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4463 " %s to sysfs\n", s
->name
);
4471 __initcall(slab_sysfs_init
);
4475 * The /proc/slabinfo ABI
4477 #ifdef CONFIG_SLABINFO
4479 ssize_t
slabinfo_write(struct file
*file
, const char __user
* buffer
,
4480 size_t count
, loff_t
*ppos
)
4486 static void print_slabinfo_header(struct seq_file
*m
)
4488 seq_puts(m
, "slabinfo - version: 2.1\n");
4489 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4490 "<objperslab> <pagesperslab>");
4491 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4492 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4496 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4500 down_read(&slub_lock
);
4502 print_slabinfo_header(m
);
4504 return seq_list_start(&slab_caches
, *pos
);
4507 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4509 return seq_list_next(p
, &slab_caches
, pos
);
4512 static void s_stop(struct seq_file
*m
, void *p
)
4514 up_read(&slub_lock
);
4517 static int s_show(struct seq_file
*m
, void *p
)
4519 unsigned long nr_partials
= 0;
4520 unsigned long nr_slabs
= 0;
4521 unsigned long nr_inuse
= 0;
4522 unsigned long nr_objs
= 0;
4523 unsigned long nr_free
= 0;
4524 struct kmem_cache
*s
;
4527 s
= list_entry(p
, struct kmem_cache
, list
);
4529 for_each_online_node(node
) {
4530 struct kmem_cache_node
*n
= get_node(s
, node
);
4535 nr_partials
+= n
->nr_partial
;
4536 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4537 nr_objs
+= atomic_long_read(&n
->total_objects
);
4538 nr_free
+= count_partial(n
, count_free
);
4541 nr_inuse
= nr_objs
- nr_free
;
4543 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4544 nr_objs
, s
->size
, oo_objects(s
->oo
),
4545 (1 << oo_order(s
->oo
)));
4546 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4547 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4553 const struct seq_operations slabinfo_op
= {
4560 #endif /* CONFIG_SLABINFO */