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
153 * Mininum number of partial slabs. These will be left on the partial
154 * lists even if they are empty. kmem_cache_shrink may reclaim them.
156 #define MIN_PARTIAL 5
159 * Maximum number of desirable partial slabs.
160 * The existence of more partial slabs makes kmem_cache_shrink
161 * sort the partial list by the number of objects in the.
163 #define MAX_PARTIAL 10
165 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
166 SLAB_POISON | SLAB_STORE_USER)
169 * Set of flags that will prevent slab merging
171 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
172 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
174 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
177 #ifndef ARCH_KMALLOC_MINALIGN
178 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
181 #ifndef ARCH_SLAB_MINALIGN
182 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
185 /* Internal SLUB flags */
186 #define __OBJECT_POISON 0x80000000 /* Poison object */
187 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
189 /* Not all arches define cache_line_size */
190 #ifndef cache_line_size
191 #define cache_line_size() L1_CACHE_BYTES
194 static int kmem_size
= sizeof(struct kmem_cache
);
197 static struct notifier_block slab_notifier
;
201 DOWN
, /* No slab functionality available */
202 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
203 UP
, /* Everything works but does not show up in sysfs */
207 /* A list of all slab caches on the system */
208 static DECLARE_RWSEM(slub_lock
);
209 static LIST_HEAD(slab_caches
);
212 * Tracking user of a slab.
215 void *addr
; /* Called from address */
216 int cpu
; /* Was running on cpu */
217 int pid
; /* Pid context */
218 unsigned long when
; /* When did the operation occur */
221 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
223 #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)
224 static int sysfs_slab_add(struct kmem_cache
*);
225 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
226 static void sysfs_slab_remove(struct kmem_cache
*);
229 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
230 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
232 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
239 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
241 #ifdef CONFIG_SLUB_STATS
246 /********************************************************************
247 * Core slab cache functions
248 *******************************************************************/
250 int slab_is_available(void)
252 return slab_state
>= UP
;
255 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
258 return s
->node
[node
];
260 return &s
->local_node
;
264 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
267 return s
->cpu_slab
[cpu
];
273 /* Verify that a pointer has an address that is valid within a slab page */
274 static inline int check_valid_pointer(struct kmem_cache
*s
,
275 struct page
*page
, const void *object
)
282 base
= page_address(page
);
283 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
284 (object
- base
) % s
->size
) {
292 * Slow version of get and set free pointer.
294 * This version requires touching the cache lines of kmem_cache which
295 * we avoid to do in the fast alloc free paths. There we obtain the offset
296 * from the page struct.
298 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
300 return *(void **)(object
+ s
->offset
);
303 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
305 *(void **)(object
+ s
->offset
) = fp
;
308 /* Loop over all objects in a slab */
309 #define for_each_object(__p, __s, __addr, __objects) \
310 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
314 #define for_each_free_object(__p, __s, __free) \
315 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
317 /* Determine object index from a given position */
318 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
320 return (p
- addr
) / s
->size
;
323 static inline struct kmem_cache_order_objects
oo_make(int order
,
326 struct kmem_cache_order_objects x
= {
327 (order
<< 16) + (PAGE_SIZE
<< order
) / size
333 static inline int oo_order(struct kmem_cache_order_objects x
)
338 static inline int oo_objects(struct kmem_cache_order_objects x
)
340 return x
.x
& ((1 << 16) - 1);
343 #ifdef CONFIG_SLUB_DEBUG
347 #ifdef CONFIG_SLUB_DEBUG_ON
348 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
350 static int slub_debug
;
353 static char *slub_debug_slabs
;
358 static void print_section(char *text
, u8
*addr
, unsigned int length
)
366 for (i
= 0; i
< length
; i
++) {
368 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
371 printk(KERN_CONT
" %02x", addr
[i
]);
373 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
375 printk(KERN_CONT
" %s\n", ascii
);
382 printk(KERN_CONT
" ");
386 printk(KERN_CONT
" %s\n", ascii
);
390 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
391 enum track_item alloc
)
396 p
= object
+ s
->offset
+ sizeof(void *);
398 p
= object
+ s
->inuse
;
403 static void set_track(struct kmem_cache
*s
, void *object
,
404 enum track_item alloc
, void *addr
)
409 p
= object
+ s
->offset
+ sizeof(void *);
411 p
= object
+ s
->inuse
;
416 p
->cpu
= smp_processor_id();
417 p
->pid
= current
? current
->pid
: -1;
420 memset(p
, 0, sizeof(struct track
));
423 static void init_tracking(struct kmem_cache
*s
, void *object
)
425 if (!(s
->flags
& SLAB_STORE_USER
))
428 set_track(s
, object
, TRACK_FREE
, NULL
);
429 set_track(s
, object
, TRACK_ALLOC
, NULL
);
432 static void print_track(const char *s
, struct track
*t
)
437 printk(KERN_ERR
"INFO: %s in ", s
);
438 __print_symbol("%s", (unsigned long)t
->addr
);
439 printk(" age=%lu cpu=%u pid=%d\n", jiffies
- t
->when
, t
->cpu
, t
->pid
);
442 static void print_tracking(struct kmem_cache
*s
, void *object
)
444 if (!(s
->flags
& SLAB_STORE_USER
))
447 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
448 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
451 static void print_page_info(struct page
*page
)
453 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
454 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
458 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
464 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
466 printk(KERN_ERR
"========================================"
467 "=====================================\n");
468 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
469 printk(KERN_ERR
"----------------------------------------"
470 "-------------------------------------\n\n");
473 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
479 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
481 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
484 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
486 unsigned int off
; /* Offset of last byte */
487 u8
*addr
= page_address(page
);
489 print_tracking(s
, p
);
491 print_page_info(page
);
493 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
494 p
, p
- addr
, get_freepointer(s
, p
));
497 print_section("Bytes b4", p
- 16, 16);
499 print_section("Object", p
, min(s
->objsize
, 128));
501 if (s
->flags
& SLAB_RED_ZONE
)
502 print_section("Redzone", p
+ s
->objsize
,
503 s
->inuse
- s
->objsize
);
506 off
= s
->offset
+ sizeof(void *);
510 if (s
->flags
& SLAB_STORE_USER
)
511 off
+= 2 * sizeof(struct track
);
514 /* Beginning of the filler is the free pointer */
515 print_section("Padding", p
+ off
, s
->size
- off
);
520 static void object_err(struct kmem_cache
*s
, struct page
*page
,
521 u8
*object
, char *reason
)
523 slab_bug(s
, "%s", reason
);
524 print_trailer(s
, page
, object
);
527 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
533 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
535 slab_bug(s
, "%s", buf
);
536 print_page_info(page
);
540 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
544 if (s
->flags
& __OBJECT_POISON
) {
545 memset(p
, POISON_FREE
, s
->objsize
- 1);
546 p
[s
->objsize
- 1] = POISON_END
;
549 if (s
->flags
& SLAB_RED_ZONE
)
550 memset(p
+ s
->objsize
,
551 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
552 s
->inuse
- s
->objsize
);
555 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
558 if (*start
!= (u8
)value
)
566 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
567 void *from
, void *to
)
569 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
570 memset(from
, data
, to
- from
);
573 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
574 u8
*object
, char *what
,
575 u8
*start
, unsigned int value
, unsigned int bytes
)
580 fault
= check_bytes(start
, value
, bytes
);
585 while (end
> fault
&& end
[-1] == value
)
588 slab_bug(s
, "%s overwritten", what
);
589 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
590 fault
, end
- 1, fault
[0], value
);
591 print_trailer(s
, page
, object
);
593 restore_bytes(s
, what
, value
, fault
, end
);
601 * Bytes of the object to be managed.
602 * If the freepointer may overlay the object then the free
603 * pointer is the first word of the object.
605 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
608 * object + s->objsize
609 * Padding to reach word boundary. This is also used for Redzoning.
610 * Padding is extended by another word if Redzoning is enabled and
613 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
614 * 0xcc (RED_ACTIVE) for objects in use.
617 * Meta data starts here.
619 * A. Free pointer (if we cannot overwrite object on free)
620 * B. Tracking data for SLAB_STORE_USER
621 * C. Padding to reach required alignment boundary or at mininum
622 * one word if debugging is on to be able to detect writes
623 * before the word boundary.
625 * Padding is done using 0x5a (POISON_INUSE)
628 * Nothing is used beyond s->size.
630 * If slabcaches are merged then the objsize and inuse boundaries are mostly
631 * ignored. And therefore no slab options that rely on these boundaries
632 * may be used with merged slabcaches.
635 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
637 unsigned long off
= s
->inuse
; /* The end of info */
640 /* Freepointer is placed after the object. */
641 off
+= sizeof(void *);
643 if (s
->flags
& SLAB_STORE_USER
)
644 /* We also have user information there */
645 off
+= 2 * sizeof(struct track
);
650 return check_bytes_and_report(s
, page
, p
, "Object padding",
651 p
+ off
, POISON_INUSE
, s
->size
- off
);
654 /* Check the pad bytes at the end of a slab page */
655 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
663 if (!(s
->flags
& SLAB_POISON
))
666 start
= page_address(page
);
667 length
= (PAGE_SIZE
<< compound_order(page
));
668 end
= start
+ length
;
669 remainder
= length
% s
->size
;
673 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
676 while (end
> fault
&& end
[-1] == POISON_INUSE
)
679 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
680 print_section("Padding", end
- remainder
, remainder
);
682 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
686 static int check_object(struct kmem_cache
*s
, struct page
*page
,
687 void *object
, int active
)
690 u8
*endobject
= object
+ s
->objsize
;
692 if (s
->flags
& SLAB_RED_ZONE
) {
694 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
696 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
697 endobject
, red
, s
->inuse
- s
->objsize
))
700 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
701 check_bytes_and_report(s
, page
, p
, "Alignment padding",
702 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
706 if (s
->flags
& SLAB_POISON
) {
707 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
708 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
709 POISON_FREE
, s
->objsize
- 1) ||
710 !check_bytes_and_report(s
, page
, p
, "Poison",
711 p
+ s
->objsize
- 1, POISON_END
, 1)))
714 * check_pad_bytes cleans up on its own.
716 check_pad_bytes(s
, page
, p
);
719 if (!s
->offset
&& active
)
721 * Object and freepointer overlap. Cannot check
722 * freepointer while object is allocated.
726 /* Check free pointer validity */
727 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
728 object_err(s
, page
, p
, "Freepointer corrupt");
730 * No choice but to zap it and thus loose the remainder
731 * of the free objects in this slab. May cause
732 * another error because the object count is now wrong.
734 set_freepointer(s
, p
, NULL
);
740 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
744 VM_BUG_ON(!irqs_disabled());
746 if (!PageSlab(page
)) {
747 slab_err(s
, page
, "Not a valid slab page");
751 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
752 if (page
->objects
> maxobj
) {
753 slab_err(s
, page
, "objects %u > max %u",
754 s
->name
, page
->objects
, maxobj
);
757 if (page
->inuse
> page
->objects
) {
758 slab_err(s
, page
, "inuse %u > max %u",
759 s
->name
, page
->inuse
, page
->objects
);
762 /* Slab_pad_check fixes things up after itself */
763 slab_pad_check(s
, page
);
768 * Determine if a certain object on a page is on the freelist. Must hold the
769 * slab lock to guarantee that the chains are in a consistent state.
771 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
774 void *fp
= page
->freelist
;
776 unsigned long max_objects
;
778 while (fp
&& nr
<= page
->objects
) {
781 if (!check_valid_pointer(s
, page
, fp
)) {
783 object_err(s
, page
, object
,
784 "Freechain corrupt");
785 set_freepointer(s
, object
, NULL
);
788 slab_err(s
, page
, "Freepointer corrupt");
789 page
->freelist
= NULL
;
790 page
->inuse
= page
->objects
;
791 slab_fix(s
, "Freelist cleared");
797 fp
= get_freepointer(s
, object
);
801 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
802 if (max_objects
> 65535)
805 if (page
->objects
!= max_objects
) {
806 slab_err(s
, page
, "Wrong number of objects. Found %d but "
807 "should be %d", page
->objects
, max_objects
);
808 page
->objects
= max_objects
;
809 slab_fix(s
, "Number of objects adjusted.");
811 if (page
->inuse
!= page
->objects
- nr
) {
812 slab_err(s
, page
, "Wrong object count. Counter is %d but "
813 "counted were %d", page
->inuse
, page
->objects
- nr
);
814 page
->inuse
= page
->objects
- nr
;
815 slab_fix(s
, "Object count adjusted.");
817 return search
== NULL
;
820 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
, int alloc
)
822 if (s
->flags
& SLAB_TRACE
) {
823 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
825 alloc
? "alloc" : "free",
830 print_section("Object", (void *)object
, s
->objsize
);
837 * Tracking of fully allocated slabs for debugging purposes.
839 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
841 spin_lock(&n
->list_lock
);
842 list_add(&page
->lru
, &n
->full
);
843 spin_unlock(&n
->list_lock
);
846 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
848 struct kmem_cache_node
*n
;
850 if (!(s
->flags
& SLAB_STORE_USER
))
853 n
= get_node(s
, page_to_nid(page
));
855 spin_lock(&n
->list_lock
);
856 list_del(&page
->lru
);
857 spin_unlock(&n
->list_lock
);
860 /* Tracking of the number of slabs for debugging purposes */
861 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
863 struct kmem_cache_node
*n
= get_node(s
, node
);
865 return atomic_long_read(&n
->nr_slabs
);
868 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
870 struct kmem_cache_node
*n
= get_node(s
, node
);
873 * May be called early in order to allocate a slab for the
874 * kmem_cache_node structure. Solve the chicken-egg
875 * dilemma by deferring the increment of the count during
876 * bootstrap (see early_kmem_cache_node_alloc).
878 if (!NUMA_BUILD
|| n
) {
879 atomic_long_inc(&n
->nr_slabs
);
880 atomic_long_add(objects
, &n
->total_objects
);
883 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
885 struct kmem_cache_node
*n
= get_node(s
, node
);
887 atomic_long_dec(&n
->nr_slabs
);
888 atomic_long_sub(objects
, &n
->total_objects
);
891 /* Object debug checks for alloc/free paths */
892 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
895 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
898 init_object(s
, object
, 0);
899 init_tracking(s
, object
);
902 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
903 void *object
, void *addr
)
905 if (!check_slab(s
, page
))
908 if (!on_freelist(s
, page
, object
)) {
909 object_err(s
, page
, object
, "Object already allocated");
913 if (!check_valid_pointer(s
, page
, object
)) {
914 object_err(s
, page
, object
, "Freelist Pointer check fails");
918 if (!check_object(s
, page
, object
, 0))
921 /* Success perform special debug activities for allocs */
922 if (s
->flags
& SLAB_STORE_USER
)
923 set_track(s
, object
, TRACK_ALLOC
, addr
);
924 trace(s
, page
, object
, 1);
925 init_object(s
, object
, 1);
929 if (PageSlab(page
)) {
931 * If this is a slab page then lets do the best we can
932 * to avoid issues in the future. Marking all objects
933 * as used avoids touching the remaining objects.
935 slab_fix(s
, "Marking all objects used");
936 page
->inuse
= page
->objects
;
937 page
->freelist
= NULL
;
942 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
943 void *object
, void *addr
)
945 if (!check_slab(s
, page
))
948 if (!check_valid_pointer(s
, page
, object
)) {
949 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
953 if (on_freelist(s
, page
, object
)) {
954 object_err(s
, page
, object
, "Object already free");
958 if (!check_object(s
, page
, object
, 1))
961 if (unlikely(s
!= page
->slab
)) {
962 if (!PageSlab(page
)) {
963 slab_err(s
, page
, "Attempt to free object(0x%p) "
964 "outside of slab", object
);
965 } else if (!page
->slab
) {
967 "SLUB <none>: no slab for object 0x%p.\n",
971 object_err(s
, page
, object
,
972 "page slab pointer corrupt.");
976 /* Special debug activities for freeing objects */
977 if (!SlabFrozen(page
) && !page
->freelist
)
978 remove_full(s
, page
);
979 if (s
->flags
& SLAB_STORE_USER
)
980 set_track(s
, object
, TRACK_FREE
, addr
);
981 trace(s
, page
, object
, 0);
982 init_object(s
, object
, 0);
986 slab_fix(s
, "Object at 0x%p not freed", object
);
990 static int __init
setup_slub_debug(char *str
)
992 slub_debug
= DEBUG_DEFAULT_FLAGS
;
993 if (*str
++ != '=' || !*str
)
995 * No options specified. Switch on full debugging.
1001 * No options but restriction on slabs. This means full
1002 * debugging for slabs matching a pattern.
1009 * Switch off all debugging measures.
1014 * Determine which debug features should be switched on
1016 for (; *str
&& *str
!= ','; str
++) {
1017 switch (tolower(*str
)) {
1019 slub_debug
|= SLAB_DEBUG_FREE
;
1022 slub_debug
|= SLAB_RED_ZONE
;
1025 slub_debug
|= SLAB_POISON
;
1028 slub_debug
|= SLAB_STORE_USER
;
1031 slub_debug
|= SLAB_TRACE
;
1034 printk(KERN_ERR
"slub_debug option '%c' "
1035 "unknown. skipped\n", *str
);
1041 slub_debug_slabs
= str
+ 1;
1046 __setup("slub_debug", setup_slub_debug
);
1048 static unsigned long kmem_cache_flags(unsigned long objsize
,
1049 unsigned long flags
, const char *name
,
1050 void (*ctor
)(struct kmem_cache
*, void *))
1053 * Enable debugging if selected on the kernel commandline.
1055 if (slub_debug
&& (!slub_debug_slabs
||
1056 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1057 flags
|= slub_debug
;
1062 static inline void setup_object_debug(struct kmem_cache
*s
,
1063 struct page
*page
, void *object
) {}
1065 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1066 struct page
*page
, void *object
, void *addr
) { return 0; }
1068 static inline int free_debug_processing(struct kmem_cache
*s
,
1069 struct page
*page
, void *object
, void *addr
) { return 0; }
1071 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1073 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1074 void *object
, int active
) { return 1; }
1075 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1076 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1077 unsigned long flags
, const char *name
,
1078 void (*ctor
)(struct kmem_cache
*, void *))
1082 #define slub_debug 0
1084 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1086 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1088 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1093 * Slab allocation and freeing
1095 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1096 struct kmem_cache_order_objects oo
)
1098 int order
= oo_order(oo
);
1101 return alloc_pages(flags
, order
);
1103 return alloc_pages_node(node
, flags
, order
);
1106 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1109 struct kmem_cache_order_objects oo
= s
->oo
;
1111 flags
|= s
->allocflags
;
1113 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1115 if (unlikely(!page
)) {
1118 * Allocation may have failed due to fragmentation.
1119 * Try a lower order alloc if possible
1121 page
= alloc_slab_page(flags
, node
, oo
);
1125 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1127 page
->objects
= oo_objects(oo
);
1128 mod_zone_page_state(page_zone(page
),
1129 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1130 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1136 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1139 setup_object_debug(s
, page
, object
);
1140 if (unlikely(s
->ctor
))
1144 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1151 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1153 page
= allocate_slab(s
,
1154 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1158 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1160 page
->flags
|= 1 << PG_slab
;
1161 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1162 SLAB_STORE_USER
| SLAB_TRACE
))
1165 start
= page_address(page
);
1167 if (unlikely(s
->flags
& SLAB_POISON
))
1168 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1171 for_each_object(p
, s
, start
, page
->objects
) {
1172 setup_object(s
, page
, last
);
1173 set_freepointer(s
, last
, p
);
1176 setup_object(s
, page
, last
);
1177 set_freepointer(s
, last
, NULL
);
1179 page
->freelist
= start
;
1185 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1187 int order
= compound_order(page
);
1188 int pages
= 1 << order
;
1190 if (unlikely(SlabDebug(page
))) {
1193 slab_pad_check(s
, page
);
1194 for_each_object(p
, s
, page_address(page
),
1196 check_object(s
, page
, p
, 0);
1197 ClearSlabDebug(page
);
1200 mod_zone_page_state(page_zone(page
),
1201 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1202 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1205 __ClearPageSlab(page
);
1206 reset_page_mapcount(page
);
1207 __free_pages(page
, order
);
1210 static void rcu_free_slab(struct rcu_head
*h
)
1214 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1215 __free_slab(page
->slab
, page
);
1218 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1220 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1222 * RCU free overloads the RCU head over the LRU
1224 struct rcu_head
*head
= (void *)&page
->lru
;
1226 call_rcu(head
, rcu_free_slab
);
1228 __free_slab(s
, page
);
1231 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1233 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1238 * Per slab locking using the pagelock
1240 static __always_inline
void slab_lock(struct page
*page
)
1242 bit_spin_lock(PG_locked
, &page
->flags
);
1245 static __always_inline
void slab_unlock(struct page
*page
)
1247 __bit_spin_unlock(PG_locked
, &page
->flags
);
1250 static __always_inline
int slab_trylock(struct page
*page
)
1254 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1259 * Management of partially allocated slabs
1261 static void add_partial(struct kmem_cache_node
*n
,
1262 struct page
*page
, int tail
)
1264 spin_lock(&n
->list_lock
);
1267 list_add_tail(&page
->lru
, &n
->partial
);
1269 list_add(&page
->lru
, &n
->partial
);
1270 spin_unlock(&n
->list_lock
);
1273 static void remove_partial(struct kmem_cache
*s
,
1276 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1278 spin_lock(&n
->list_lock
);
1279 list_del(&page
->lru
);
1281 spin_unlock(&n
->list_lock
);
1285 * Lock slab and remove from the partial list.
1287 * Must hold list_lock.
1289 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
, struct page
*page
)
1291 if (slab_trylock(page
)) {
1292 list_del(&page
->lru
);
1294 SetSlabFrozen(page
);
1301 * Try to allocate a partial slab from a specific node.
1303 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1308 * Racy check. If we mistakenly see no partial slabs then we
1309 * just allocate an empty slab. If we mistakenly try to get a
1310 * partial slab and there is none available then get_partials()
1313 if (!n
|| !n
->nr_partial
)
1316 spin_lock(&n
->list_lock
);
1317 list_for_each_entry(page
, &n
->partial
, lru
)
1318 if (lock_and_freeze_slab(n
, page
))
1322 spin_unlock(&n
->list_lock
);
1327 * Get a page from somewhere. Search in increasing NUMA distances.
1329 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1332 struct zonelist
*zonelist
;
1337 * The defrag ratio allows a configuration of the tradeoffs between
1338 * inter node defragmentation and node local allocations. A lower
1339 * defrag_ratio increases the tendency to do local allocations
1340 * instead of attempting to obtain partial slabs from other nodes.
1342 * If the defrag_ratio is set to 0 then kmalloc() always
1343 * returns node local objects. If the ratio is higher then kmalloc()
1344 * may return off node objects because partial slabs are obtained
1345 * from other nodes and filled up.
1347 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1348 * defrag_ratio = 1000) then every (well almost) allocation will
1349 * first attempt to defrag slab caches on other nodes. This means
1350 * scanning over all nodes to look for partial slabs which may be
1351 * expensive if we do it every time we are trying to find a slab
1352 * with available objects.
1354 if (!s
->remote_node_defrag_ratio
||
1355 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1358 zonelist
= &NODE_DATA(
1359 slab_node(current
->mempolicy
))->node_zonelists
[gfp_zone(flags
)];
1360 for (z
= zonelist
->zones
; *z
; z
++) {
1361 struct kmem_cache_node
*n
;
1363 n
= get_node(s
, zone_to_nid(*z
));
1365 if (n
&& cpuset_zone_allowed_hardwall(*z
, flags
) &&
1366 n
->nr_partial
> MIN_PARTIAL
) {
1367 page
= get_partial_node(n
);
1377 * Get a partial page, lock it and return it.
1379 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1382 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1384 page
= get_partial_node(get_node(s
, searchnode
));
1385 if (page
|| (flags
& __GFP_THISNODE
))
1388 return get_any_partial(s
, flags
);
1392 * Move a page back to the lists.
1394 * Must be called with the slab lock held.
1396 * On exit the slab lock will have been dropped.
1398 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1400 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1401 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1403 ClearSlabFrozen(page
);
1406 if (page
->freelist
) {
1407 add_partial(n
, page
, tail
);
1408 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1410 stat(c
, DEACTIVATE_FULL
);
1411 if (SlabDebug(page
) && (s
->flags
& SLAB_STORE_USER
))
1416 stat(c
, DEACTIVATE_EMPTY
);
1417 if (n
->nr_partial
< MIN_PARTIAL
) {
1419 * Adding an empty slab to the partial slabs in order
1420 * to avoid page allocator overhead. This slab needs
1421 * to come after the other slabs with objects in
1422 * so that the others get filled first. That way the
1423 * size of the partial list stays small.
1425 * kmem_cache_shrink can reclaim any empty slabs from the
1428 add_partial(n
, page
, 1);
1432 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1433 discard_slab(s
, page
);
1439 * Remove the cpu slab
1441 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1443 struct page
*page
= c
->page
;
1447 stat(c
, DEACTIVATE_REMOTE_FREES
);
1449 * Merge cpu freelist into slab freelist. Typically we get here
1450 * because both freelists are empty. So this is unlikely
1453 while (unlikely(c
->freelist
)) {
1456 tail
= 0; /* Hot objects. Put the slab first */
1458 /* Retrieve object from cpu_freelist */
1459 object
= c
->freelist
;
1460 c
->freelist
= c
->freelist
[c
->offset
];
1462 /* And put onto the regular freelist */
1463 object
[c
->offset
] = page
->freelist
;
1464 page
->freelist
= object
;
1468 unfreeze_slab(s
, page
, tail
);
1471 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1473 stat(c
, CPUSLAB_FLUSH
);
1475 deactivate_slab(s
, c
);
1481 * Called from IPI handler with interrupts disabled.
1483 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1485 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1487 if (likely(c
&& c
->page
))
1491 static void flush_cpu_slab(void *d
)
1493 struct kmem_cache
*s
= d
;
1495 __flush_cpu_slab(s
, smp_processor_id());
1498 static void flush_all(struct kmem_cache
*s
)
1501 on_each_cpu(flush_cpu_slab
, s
, 1, 1);
1503 unsigned long flags
;
1505 local_irq_save(flags
);
1507 local_irq_restore(flags
);
1512 * Check if the objects in a per cpu structure fit numa
1513 * locality expectations.
1515 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1518 if (node
!= -1 && c
->node
!= node
)
1525 * Slow path. The lockless freelist is empty or we need to perform
1528 * Interrupts are disabled.
1530 * Processing is still very fast if new objects have been freed to the
1531 * regular freelist. In that case we simply take over the regular freelist
1532 * as the lockless freelist and zap the regular freelist.
1534 * If that is not working then we fall back to the partial lists. We take the
1535 * first element of the freelist as the object to allocate now and move the
1536 * rest of the freelist to the lockless freelist.
1538 * And if we were unable to get a new slab from the partial slab lists then
1539 * we need to allocate a new slab. This is the slowest path since it involves
1540 * a call to the page allocator and the setup of a new slab.
1542 static void *__slab_alloc(struct kmem_cache
*s
,
1543 gfp_t gfpflags
, int node
, void *addr
, struct kmem_cache_cpu
*c
)
1548 /* We handle __GFP_ZERO in the caller */
1549 gfpflags
&= ~__GFP_ZERO
;
1555 if (unlikely(!node_match(c
, node
)))
1558 stat(c
, ALLOC_REFILL
);
1561 object
= c
->page
->freelist
;
1562 if (unlikely(!object
))
1564 if (unlikely(SlabDebug(c
->page
)))
1567 c
->freelist
= object
[c
->offset
];
1568 c
->page
->inuse
= c
->page
->objects
;
1569 c
->page
->freelist
= NULL
;
1570 c
->node
= page_to_nid(c
->page
);
1572 slab_unlock(c
->page
);
1573 stat(c
, ALLOC_SLOWPATH
);
1577 deactivate_slab(s
, c
);
1580 new = get_partial(s
, gfpflags
, node
);
1583 stat(c
, ALLOC_FROM_PARTIAL
);
1587 if (gfpflags
& __GFP_WAIT
)
1590 new = new_slab(s
, gfpflags
, node
);
1592 if (gfpflags
& __GFP_WAIT
)
1593 local_irq_disable();
1596 c
= get_cpu_slab(s
, smp_processor_id());
1597 stat(c
, ALLOC_SLAB
);
1607 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1611 c
->page
->freelist
= object
[c
->offset
];
1617 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1618 * have the fastpath folded into their functions. So no function call
1619 * overhead for requests that can be satisfied on the fastpath.
1621 * The fastpath works by first checking if the lockless freelist can be used.
1622 * If not then __slab_alloc is called for slow processing.
1624 * Otherwise we can simply pick the next object from the lockless free list.
1626 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1627 gfp_t gfpflags
, int node
, void *addr
)
1630 struct kmem_cache_cpu
*c
;
1631 unsigned long flags
;
1633 local_irq_save(flags
);
1634 c
= get_cpu_slab(s
, smp_processor_id());
1635 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1637 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1640 object
= c
->freelist
;
1641 c
->freelist
= object
[c
->offset
];
1642 stat(c
, ALLOC_FASTPATH
);
1644 local_irq_restore(flags
);
1646 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1647 memset(object
, 0, c
->objsize
);
1652 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1654 return slab_alloc(s
, gfpflags
, -1, __builtin_return_address(0));
1656 EXPORT_SYMBOL(kmem_cache_alloc
);
1659 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1661 return slab_alloc(s
, gfpflags
, node
, __builtin_return_address(0));
1663 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1667 * Slow patch handling. This may still be called frequently since objects
1668 * have a longer lifetime than the cpu slabs in most processing loads.
1670 * So we still attempt to reduce cache line usage. Just take the slab
1671 * lock and free the item. If there is no additional partial page
1672 * handling required then we can return immediately.
1674 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1675 void *x
, void *addr
, unsigned int offset
)
1678 void **object
= (void *)x
;
1679 struct kmem_cache_cpu
*c
;
1681 c
= get_cpu_slab(s
, raw_smp_processor_id());
1682 stat(c
, FREE_SLOWPATH
);
1685 if (unlikely(SlabDebug(page
)))
1689 prior
= object
[offset
] = page
->freelist
;
1690 page
->freelist
= object
;
1693 if (unlikely(SlabFrozen(page
))) {
1694 stat(c
, FREE_FROZEN
);
1698 if (unlikely(!page
->inuse
))
1702 * Objects left in the slab. If it was not on the partial list before
1705 if (unlikely(!prior
)) {
1706 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1707 stat(c
, FREE_ADD_PARTIAL
);
1717 * Slab still on the partial list.
1719 remove_partial(s
, page
);
1720 stat(c
, FREE_REMOVE_PARTIAL
);
1724 discard_slab(s
, page
);
1728 if (!free_debug_processing(s
, page
, x
, addr
))
1734 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1735 * can perform fastpath freeing without additional function calls.
1737 * The fastpath is only possible if we are freeing to the current cpu slab
1738 * of this processor. This typically the case if we have just allocated
1741 * If fastpath is not possible then fall back to __slab_free where we deal
1742 * with all sorts of special processing.
1744 static __always_inline
void slab_free(struct kmem_cache
*s
,
1745 struct page
*page
, void *x
, void *addr
)
1747 void **object
= (void *)x
;
1748 struct kmem_cache_cpu
*c
;
1749 unsigned long flags
;
1751 local_irq_save(flags
);
1752 c
= get_cpu_slab(s
, smp_processor_id());
1753 debug_check_no_locks_freed(object
, c
->objsize
);
1754 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1755 object
[c
->offset
] = c
->freelist
;
1756 c
->freelist
= object
;
1757 stat(c
, FREE_FASTPATH
);
1759 __slab_free(s
, page
, x
, addr
, c
->offset
);
1761 local_irq_restore(flags
);
1764 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1768 page
= virt_to_head_page(x
);
1770 slab_free(s
, page
, x
, __builtin_return_address(0));
1772 EXPORT_SYMBOL(kmem_cache_free
);
1774 /* Figure out on which slab object the object resides */
1775 static struct page
*get_object_page(const void *x
)
1777 struct page
*page
= virt_to_head_page(x
);
1779 if (!PageSlab(page
))
1786 * Object placement in a slab is made very easy because we always start at
1787 * offset 0. If we tune the size of the object to the alignment then we can
1788 * get the required alignment by putting one properly sized object after
1791 * Notice that the allocation order determines the sizes of the per cpu
1792 * caches. Each processor has always one slab available for allocations.
1793 * Increasing the allocation order reduces the number of times that slabs
1794 * must be moved on and off the partial lists and is therefore a factor in
1799 * Mininum / Maximum order of slab pages. This influences locking overhead
1800 * and slab fragmentation. A higher order reduces the number of partial slabs
1801 * and increases the number of allocations possible without having to
1802 * take the list_lock.
1804 static int slub_min_order
;
1805 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1806 static int slub_min_objects
;
1809 * Merge control. If this is set then no merging of slab caches will occur.
1810 * (Could be removed. This was introduced to pacify the merge skeptics.)
1812 static int slub_nomerge
;
1815 * Calculate the order of allocation given an slab object size.
1817 * The order of allocation has significant impact on performance and other
1818 * system components. Generally order 0 allocations should be preferred since
1819 * order 0 does not cause fragmentation in the page allocator. Larger objects
1820 * be problematic to put into order 0 slabs because there may be too much
1821 * unused space left. We go to a higher order if more than 1/16th of the slab
1824 * In order to reach satisfactory performance we must ensure that a minimum
1825 * number of objects is in one slab. Otherwise we may generate too much
1826 * activity on the partial lists which requires taking the list_lock. This is
1827 * less a concern for large slabs though which are rarely used.
1829 * slub_max_order specifies the order where we begin to stop considering the
1830 * number of objects in a slab as critical. If we reach slub_max_order then
1831 * we try to keep the page order as low as possible. So we accept more waste
1832 * of space in favor of a small page order.
1834 * Higher order allocations also allow the placement of more objects in a
1835 * slab and thereby reduce object handling overhead. If the user has
1836 * requested a higher mininum order then we start with that one instead of
1837 * the smallest order which will fit the object.
1839 static inline int slab_order(int size
, int min_objects
,
1840 int max_order
, int fract_leftover
)
1844 int min_order
= slub_min_order
;
1846 if ((PAGE_SIZE
<< min_order
) / size
> 65535)
1847 return get_order(size
* 65535) - 1;
1849 for (order
= max(min_order
,
1850 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1851 order
<= max_order
; order
++) {
1853 unsigned long slab_size
= PAGE_SIZE
<< order
;
1855 if (slab_size
< min_objects
* size
)
1858 rem
= slab_size
% size
;
1860 if (rem
<= slab_size
/ fract_leftover
)
1868 static inline int calculate_order(int size
)
1875 * Attempt to find best configuration for a slab. This
1876 * works by first attempting to generate a layout with
1877 * the best configuration and backing off gradually.
1879 * First we reduce the acceptable waste in a slab. Then
1880 * we reduce the minimum objects required in a slab.
1882 min_objects
= slub_min_objects
;
1884 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1885 while (min_objects
> 1) {
1887 while (fraction
>= 4) {
1888 order
= slab_order(size
, min_objects
,
1889 slub_max_order
, fraction
);
1890 if (order
<= slub_max_order
)
1898 * We were unable to place multiple objects in a slab. Now
1899 * lets see if we can place a single object there.
1901 order
= slab_order(size
, 1, slub_max_order
, 1);
1902 if (order
<= slub_max_order
)
1906 * Doh this slab cannot be placed using slub_max_order.
1908 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1909 if (order
<= MAX_ORDER
)
1915 * Figure out what the alignment of the objects will be.
1917 static unsigned long calculate_alignment(unsigned long flags
,
1918 unsigned long align
, unsigned long size
)
1921 * If the user wants hardware cache aligned objects then follow that
1922 * suggestion if the object is sufficiently large.
1924 * The hardware cache alignment cannot override the specified
1925 * alignment though. If that is greater then use it.
1927 if (flags
& SLAB_HWCACHE_ALIGN
) {
1928 unsigned long ralign
= cache_line_size();
1929 while (size
<= ralign
/ 2)
1931 align
= max(align
, ralign
);
1934 if (align
< ARCH_SLAB_MINALIGN
)
1935 align
= ARCH_SLAB_MINALIGN
;
1937 return ALIGN(align
, sizeof(void *));
1940 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1941 struct kmem_cache_cpu
*c
)
1946 c
->offset
= s
->offset
/ sizeof(void *);
1947 c
->objsize
= s
->objsize
;
1948 #ifdef CONFIG_SLUB_STATS
1949 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1953 static void init_kmem_cache_node(struct kmem_cache_node
*n
)
1956 spin_lock_init(&n
->list_lock
);
1957 INIT_LIST_HEAD(&n
->partial
);
1958 #ifdef CONFIG_SLUB_DEBUG
1959 atomic_long_set(&n
->nr_slabs
, 0);
1960 INIT_LIST_HEAD(&n
->full
);
1966 * Per cpu array for per cpu structures.
1968 * The per cpu array places all kmem_cache_cpu structures from one processor
1969 * close together meaning that it becomes possible that multiple per cpu
1970 * structures are contained in one cacheline. This may be particularly
1971 * beneficial for the kmalloc caches.
1973 * A desktop system typically has around 60-80 slabs. With 100 here we are
1974 * likely able to get per cpu structures for all caches from the array defined
1975 * here. We must be able to cover all kmalloc caches during bootstrap.
1977 * If the per cpu array is exhausted then fall back to kmalloc
1978 * of individual cachelines. No sharing is possible then.
1980 #define NR_KMEM_CACHE_CPU 100
1982 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
1983 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
1985 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
1986 static cpumask_t kmem_cach_cpu_free_init_once
= CPU_MASK_NONE
;
1988 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
1989 int cpu
, gfp_t flags
)
1991 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
1994 per_cpu(kmem_cache_cpu_free
, cpu
) =
1995 (void *)c
->freelist
;
1997 /* Table overflow: So allocate ourselves */
1999 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2000 flags
, cpu_to_node(cpu
));
2005 init_kmem_cache_cpu(s
, c
);
2009 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2011 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2012 c
> per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2016 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2017 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2020 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2024 for_each_online_cpu(cpu
) {
2025 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2028 s
->cpu_slab
[cpu
] = NULL
;
2029 free_kmem_cache_cpu(c
, cpu
);
2034 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2038 for_each_online_cpu(cpu
) {
2039 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2044 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2046 free_kmem_cache_cpus(s
);
2049 s
->cpu_slab
[cpu
] = c
;
2055 * Initialize the per cpu array.
2057 static void init_alloc_cpu_cpu(int cpu
)
2061 if (cpu_isset(cpu
, kmem_cach_cpu_free_init_once
))
2064 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2065 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2067 cpu_set(cpu
, kmem_cach_cpu_free_init_once
);
2070 static void __init
init_alloc_cpu(void)
2074 for_each_online_cpu(cpu
)
2075 init_alloc_cpu_cpu(cpu
);
2079 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2080 static inline void init_alloc_cpu(void) {}
2082 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2084 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2091 * No kmalloc_node yet so do it by hand. We know that this is the first
2092 * slab on the node for this slabcache. There are no concurrent accesses
2095 * Note that this function only works on the kmalloc_node_cache
2096 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2097 * memory on a fresh node that has no slab structures yet.
2099 static struct kmem_cache_node
*early_kmem_cache_node_alloc(gfp_t gfpflags
,
2103 struct kmem_cache_node
*n
;
2104 unsigned long flags
;
2106 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2108 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2111 if (page_to_nid(page
) != node
) {
2112 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2114 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2115 "in order to be able to continue\n");
2120 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2122 kmalloc_caches
->node
[node
] = n
;
2123 #ifdef CONFIG_SLUB_DEBUG
2124 init_object(kmalloc_caches
, n
, 1);
2125 init_tracking(kmalloc_caches
, n
);
2127 init_kmem_cache_node(n
);
2128 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2131 * lockdep requires consistent irq usage for each lock
2132 * so even though there cannot be a race this early in
2133 * the boot sequence, we still disable irqs.
2135 local_irq_save(flags
);
2136 add_partial(n
, page
, 0);
2137 local_irq_restore(flags
);
2141 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2145 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2146 struct kmem_cache_node
*n
= s
->node
[node
];
2147 if (n
&& n
!= &s
->local_node
)
2148 kmem_cache_free(kmalloc_caches
, n
);
2149 s
->node
[node
] = NULL
;
2153 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2158 if (slab_state
>= UP
)
2159 local_node
= page_to_nid(virt_to_page(s
));
2163 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2164 struct kmem_cache_node
*n
;
2166 if (local_node
== node
)
2169 if (slab_state
== DOWN
) {
2170 n
= early_kmem_cache_node_alloc(gfpflags
,
2174 n
= kmem_cache_alloc_node(kmalloc_caches
,
2178 free_kmem_cache_nodes(s
);
2184 init_kmem_cache_node(n
);
2189 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2193 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2195 init_kmem_cache_node(&s
->local_node
);
2201 * calculate_sizes() determines the order and the distribution of data within
2204 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2206 unsigned long flags
= s
->flags
;
2207 unsigned long size
= s
->objsize
;
2208 unsigned long align
= s
->align
;
2212 * Round up object size to the next word boundary. We can only
2213 * place the free pointer at word boundaries and this determines
2214 * the possible location of the free pointer.
2216 size
= ALIGN(size
, sizeof(void *));
2218 #ifdef CONFIG_SLUB_DEBUG
2220 * Determine if we can poison the object itself. If the user of
2221 * the slab may touch the object after free or before allocation
2222 * then we should never poison the object itself.
2224 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2226 s
->flags
|= __OBJECT_POISON
;
2228 s
->flags
&= ~__OBJECT_POISON
;
2232 * If we are Redzoning then check if there is some space between the
2233 * end of the object and the free pointer. If not then add an
2234 * additional word to have some bytes to store Redzone information.
2236 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2237 size
+= sizeof(void *);
2241 * With that we have determined the number of bytes in actual use
2242 * by the object. This is the potential offset to the free pointer.
2246 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2249 * Relocate free pointer after the object if it is not
2250 * permitted to overwrite the first word of the object on
2253 * This is the case if we do RCU, have a constructor or
2254 * destructor or are poisoning the objects.
2257 size
+= sizeof(void *);
2260 #ifdef CONFIG_SLUB_DEBUG
2261 if (flags
& SLAB_STORE_USER
)
2263 * Need to store information about allocs and frees after
2266 size
+= 2 * sizeof(struct track
);
2268 if (flags
& SLAB_RED_ZONE
)
2270 * Add some empty padding so that we can catch
2271 * overwrites from earlier objects rather than let
2272 * tracking information or the free pointer be
2273 * corrupted if an user writes before the start
2276 size
+= sizeof(void *);
2280 * Determine the alignment based on various parameters that the
2281 * user specified and the dynamic determination of cache line size
2284 align
= calculate_alignment(flags
, align
, s
->objsize
);
2287 * SLUB stores one object immediately after another beginning from
2288 * offset 0. In order to align the objects we have to simply size
2289 * each object to conform to the alignment.
2291 size
= ALIGN(size
, align
);
2293 if (forced_order
>= 0)
2294 order
= forced_order
;
2296 order
= calculate_order(size
);
2303 s
->allocflags
|= __GFP_COMP
;
2305 if (s
->flags
& SLAB_CACHE_DMA
)
2306 s
->allocflags
|= SLUB_DMA
;
2308 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2309 s
->allocflags
|= __GFP_RECLAIMABLE
;
2312 * Determine the number of objects per slab
2314 s
->oo
= oo_make(order
, size
);
2315 s
->min
= oo_make(get_order(size
), size
);
2316 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2319 return !!oo_objects(s
->oo
);
2323 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2324 const char *name
, size_t size
,
2325 size_t align
, unsigned long flags
,
2326 void (*ctor
)(struct kmem_cache
*, void *))
2328 memset(s
, 0, kmem_size
);
2333 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2335 if (!calculate_sizes(s
, -1))
2340 s
->remote_node_defrag_ratio
= 100;
2342 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2345 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2347 free_kmem_cache_nodes(s
);
2349 if (flags
& SLAB_PANIC
)
2350 panic("Cannot create slab %s size=%lu realsize=%u "
2351 "order=%u offset=%u flags=%lx\n",
2352 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2358 * Check if a given pointer is valid
2360 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2364 page
= get_object_page(object
);
2366 if (!page
|| s
!= page
->slab
)
2367 /* No slab or wrong slab */
2370 if (!check_valid_pointer(s
, page
, object
))
2374 * We could also check if the object is on the slabs freelist.
2375 * But this would be too expensive and it seems that the main
2376 * purpose of kmem_ptr_valid() is to check if the object belongs
2377 * to a certain slab.
2381 EXPORT_SYMBOL(kmem_ptr_validate
);
2384 * Determine the size of a slab object
2386 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2390 EXPORT_SYMBOL(kmem_cache_size
);
2392 const char *kmem_cache_name(struct kmem_cache
*s
)
2396 EXPORT_SYMBOL(kmem_cache_name
);
2398 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2401 #ifdef CONFIG_SLUB_DEBUG
2402 void *addr
= page_address(page
);
2404 DECLARE_BITMAP(map
, page
->objects
);
2406 bitmap_zero(map
, page
->objects
);
2407 slab_err(s
, page
, "%s", text
);
2409 for_each_free_object(p
, s
, page
->freelist
)
2410 set_bit(slab_index(p
, s
, addr
), map
);
2412 for_each_object(p
, s
, addr
, page
->objects
) {
2414 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2415 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2417 print_tracking(s
, p
);
2425 * Attempt to free all partial slabs on a node.
2427 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2429 unsigned long flags
;
2430 struct page
*page
, *h
;
2432 spin_lock_irqsave(&n
->list_lock
, flags
);
2433 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2435 list_del(&page
->lru
);
2436 discard_slab(s
, page
);
2439 list_slab_objects(s
, page
,
2440 "Objects remaining on kmem_cache_close()");
2443 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2447 * Release all resources used by a slab cache.
2449 static inline int kmem_cache_close(struct kmem_cache
*s
)
2455 /* Attempt to free all objects */
2456 free_kmem_cache_cpus(s
);
2457 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2458 struct kmem_cache_node
*n
= get_node(s
, node
);
2461 if (n
->nr_partial
|| slabs_node(s
, node
))
2464 free_kmem_cache_nodes(s
);
2469 * Close a cache and release the kmem_cache structure
2470 * (must be used for caches created using kmem_cache_create)
2472 void kmem_cache_destroy(struct kmem_cache
*s
)
2474 down_write(&slub_lock
);
2478 up_write(&slub_lock
);
2479 if (kmem_cache_close(s
)) {
2480 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2481 "still has objects.\n", s
->name
, __func__
);
2484 sysfs_slab_remove(s
);
2486 up_write(&slub_lock
);
2488 EXPORT_SYMBOL(kmem_cache_destroy
);
2490 /********************************************************************
2492 *******************************************************************/
2494 struct kmem_cache kmalloc_caches
[PAGE_SHIFT
+ 1] __cacheline_aligned
;
2495 EXPORT_SYMBOL(kmalloc_caches
);
2497 static int __init
setup_slub_min_order(char *str
)
2499 get_option(&str
, &slub_min_order
);
2504 __setup("slub_min_order=", setup_slub_min_order
);
2506 static int __init
setup_slub_max_order(char *str
)
2508 get_option(&str
, &slub_max_order
);
2513 __setup("slub_max_order=", setup_slub_max_order
);
2515 static int __init
setup_slub_min_objects(char *str
)
2517 get_option(&str
, &slub_min_objects
);
2522 __setup("slub_min_objects=", setup_slub_min_objects
);
2524 static int __init
setup_slub_nomerge(char *str
)
2530 __setup("slub_nomerge", setup_slub_nomerge
);
2532 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2533 const char *name
, int size
, gfp_t gfp_flags
)
2535 unsigned int flags
= 0;
2537 if (gfp_flags
& SLUB_DMA
)
2538 flags
= SLAB_CACHE_DMA
;
2540 down_write(&slub_lock
);
2541 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2545 list_add(&s
->list
, &slab_caches
);
2546 up_write(&slub_lock
);
2547 if (sysfs_slab_add(s
))
2552 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2555 #ifdef CONFIG_ZONE_DMA
2556 static struct kmem_cache
*kmalloc_caches_dma
[PAGE_SHIFT
+ 1];
2558 static void sysfs_add_func(struct work_struct
*w
)
2560 struct kmem_cache
*s
;
2562 down_write(&slub_lock
);
2563 list_for_each_entry(s
, &slab_caches
, list
) {
2564 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2565 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2569 up_write(&slub_lock
);
2572 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2574 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2576 struct kmem_cache
*s
;
2580 s
= kmalloc_caches_dma
[index
];
2584 /* Dynamically create dma cache */
2585 if (flags
& __GFP_WAIT
)
2586 down_write(&slub_lock
);
2588 if (!down_write_trylock(&slub_lock
))
2592 if (kmalloc_caches_dma
[index
])
2595 realsize
= kmalloc_caches
[index
].objsize
;
2596 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2597 (unsigned int)realsize
);
2598 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2600 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2601 realsize
, ARCH_KMALLOC_MINALIGN
,
2602 SLAB_CACHE_DMA
|__SYSFS_ADD_DEFERRED
, NULL
)) {
2608 list_add(&s
->list
, &slab_caches
);
2609 kmalloc_caches_dma
[index
] = s
;
2611 schedule_work(&sysfs_add_work
);
2614 up_write(&slub_lock
);
2616 return kmalloc_caches_dma
[index
];
2621 * Conversion table for small slabs sizes / 8 to the index in the
2622 * kmalloc array. This is necessary for slabs < 192 since we have non power
2623 * of two cache sizes there. The size of larger slabs can be determined using
2626 static s8 size_index
[24] = {
2653 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2659 return ZERO_SIZE_PTR
;
2661 index
= size_index
[(size
- 1) / 8];
2663 index
= fls(size
- 1);
2665 #ifdef CONFIG_ZONE_DMA
2666 if (unlikely((flags
& SLUB_DMA
)))
2667 return dma_kmalloc_cache(index
, flags
);
2670 return &kmalloc_caches
[index
];
2673 void *__kmalloc(size_t size
, gfp_t flags
)
2675 struct kmem_cache
*s
;
2677 if (unlikely(size
> PAGE_SIZE
))
2678 return kmalloc_large(size
, flags
);
2680 s
= get_slab(size
, flags
);
2682 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2685 return slab_alloc(s
, flags
, -1, __builtin_return_address(0));
2687 EXPORT_SYMBOL(__kmalloc
);
2689 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2691 struct page
*page
= alloc_pages_node(node
, flags
| __GFP_COMP
,
2695 return page_address(page
);
2701 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2703 struct kmem_cache
*s
;
2705 if (unlikely(size
> PAGE_SIZE
))
2706 return kmalloc_large_node(size
, flags
, node
);
2708 s
= get_slab(size
, flags
);
2710 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2713 return slab_alloc(s
, flags
, node
, __builtin_return_address(0));
2715 EXPORT_SYMBOL(__kmalloc_node
);
2718 size_t ksize(const void *object
)
2721 struct kmem_cache
*s
;
2723 if (unlikely(object
== ZERO_SIZE_PTR
))
2726 page
= virt_to_head_page(object
);
2728 if (unlikely(!PageSlab(page
)))
2729 return PAGE_SIZE
<< compound_order(page
);
2733 #ifdef CONFIG_SLUB_DEBUG
2735 * Debugging requires use of the padding between object
2736 * and whatever may come after it.
2738 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2743 * If we have the need to store the freelist pointer
2744 * back there or track user information then we can
2745 * only use the space before that information.
2747 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2750 * Else we can use all the padding etc for the allocation
2754 EXPORT_SYMBOL(ksize
);
2756 void kfree(const void *x
)
2759 void *object
= (void *)x
;
2761 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2764 page
= virt_to_head_page(x
);
2765 if (unlikely(!PageSlab(page
))) {
2769 slab_free(page
->slab
, page
, object
, __builtin_return_address(0));
2771 EXPORT_SYMBOL(kfree
);
2774 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2775 * the remaining slabs by the number of items in use. The slabs with the
2776 * most items in use come first. New allocations will then fill those up
2777 * and thus they can be removed from the partial lists.
2779 * The slabs with the least items are placed last. This results in them
2780 * being allocated from last increasing the chance that the last objects
2781 * are freed in them.
2783 int kmem_cache_shrink(struct kmem_cache
*s
)
2787 struct kmem_cache_node
*n
;
2790 int objects
= oo_objects(s
->max
);
2791 struct list_head
*slabs_by_inuse
=
2792 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2793 unsigned long flags
;
2795 if (!slabs_by_inuse
)
2799 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2800 n
= get_node(s
, node
);
2805 for (i
= 0; i
< objects
; i
++)
2806 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2808 spin_lock_irqsave(&n
->list_lock
, flags
);
2811 * Build lists indexed by the items in use in each slab.
2813 * Note that concurrent frees may occur while we hold the
2814 * list_lock. page->inuse here is the upper limit.
2816 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2817 if (!page
->inuse
&& slab_trylock(page
)) {
2819 * Must hold slab lock here because slab_free
2820 * may have freed the last object and be
2821 * waiting to release the slab.
2823 list_del(&page
->lru
);
2826 discard_slab(s
, page
);
2828 list_move(&page
->lru
,
2829 slabs_by_inuse
+ page
->inuse
);
2834 * Rebuild the partial list with the slabs filled up most
2835 * first and the least used slabs at the end.
2837 for (i
= objects
- 1; i
>= 0; i
--)
2838 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2840 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2843 kfree(slabs_by_inuse
);
2846 EXPORT_SYMBOL(kmem_cache_shrink
);
2848 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2849 static int slab_mem_going_offline_callback(void *arg
)
2851 struct kmem_cache
*s
;
2853 down_read(&slub_lock
);
2854 list_for_each_entry(s
, &slab_caches
, list
)
2855 kmem_cache_shrink(s
);
2856 up_read(&slub_lock
);
2861 static void slab_mem_offline_callback(void *arg
)
2863 struct kmem_cache_node
*n
;
2864 struct kmem_cache
*s
;
2865 struct memory_notify
*marg
= arg
;
2868 offline_node
= marg
->status_change_nid
;
2871 * If the node still has available memory. we need kmem_cache_node
2874 if (offline_node
< 0)
2877 down_read(&slub_lock
);
2878 list_for_each_entry(s
, &slab_caches
, list
) {
2879 n
= get_node(s
, offline_node
);
2882 * if n->nr_slabs > 0, slabs still exist on the node
2883 * that is going down. We were unable to free them,
2884 * and offline_pages() function shoudn't call this
2885 * callback. So, we must fail.
2887 BUG_ON(slabs_node(s
, offline_node
));
2889 s
->node
[offline_node
] = NULL
;
2890 kmem_cache_free(kmalloc_caches
, n
);
2893 up_read(&slub_lock
);
2896 static int slab_mem_going_online_callback(void *arg
)
2898 struct kmem_cache_node
*n
;
2899 struct kmem_cache
*s
;
2900 struct memory_notify
*marg
= arg
;
2901 int nid
= marg
->status_change_nid
;
2905 * If the node's memory is already available, then kmem_cache_node is
2906 * already created. Nothing to do.
2912 * We are bringing a node online. No memory is availabe yet. We must
2913 * allocate a kmem_cache_node structure in order to bring the node
2916 down_read(&slub_lock
);
2917 list_for_each_entry(s
, &slab_caches
, list
) {
2919 * XXX: kmem_cache_alloc_node will fallback to other nodes
2920 * since memory is not yet available from the node that
2923 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2928 init_kmem_cache_node(n
);
2932 up_read(&slub_lock
);
2936 static int slab_memory_callback(struct notifier_block
*self
,
2937 unsigned long action
, void *arg
)
2942 case MEM_GOING_ONLINE
:
2943 ret
= slab_mem_going_online_callback(arg
);
2945 case MEM_GOING_OFFLINE
:
2946 ret
= slab_mem_going_offline_callback(arg
);
2949 case MEM_CANCEL_ONLINE
:
2950 slab_mem_offline_callback(arg
);
2953 case MEM_CANCEL_OFFLINE
:
2957 ret
= notifier_from_errno(ret
);
2961 #endif /* CONFIG_MEMORY_HOTPLUG */
2963 /********************************************************************
2964 * Basic setup of slabs
2965 *******************************************************************/
2967 void __init
kmem_cache_init(void)
2976 * Must first have the slab cache available for the allocations of the
2977 * struct kmem_cache_node's. There is special bootstrap code in
2978 * kmem_cache_open for slab_state == DOWN.
2980 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
2981 sizeof(struct kmem_cache_node
), GFP_KERNEL
);
2982 kmalloc_caches
[0].refcount
= -1;
2985 hotplug_memory_notifier(slab_memory_callback
, 1);
2988 /* Able to allocate the per node structures */
2989 slab_state
= PARTIAL
;
2991 /* Caches that are not of the two-to-the-power-of size */
2992 if (KMALLOC_MIN_SIZE
<= 64) {
2993 create_kmalloc_cache(&kmalloc_caches
[1],
2994 "kmalloc-96", 96, GFP_KERNEL
);
2997 if (KMALLOC_MIN_SIZE
<= 128) {
2998 create_kmalloc_cache(&kmalloc_caches
[2],
2999 "kmalloc-192", 192, GFP_KERNEL
);
3003 for (i
= KMALLOC_SHIFT_LOW
; i
<= PAGE_SHIFT
; i
++) {
3004 create_kmalloc_cache(&kmalloc_caches
[i
],
3005 "kmalloc", 1 << i
, GFP_KERNEL
);
3011 * Patch up the size_index table if we have strange large alignment
3012 * requirements for the kmalloc array. This is only the case for
3013 * MIPS it seems. The standard arches will not generate any code here.
3015 * Largest permitted alignment is 256 bytes due to the way we
3016 * handle the index determination for the smaller caches.
3018 * Make sure that nothing crazy happens if someone starts tinkering
3019 * around with ARCH_KMALLOC_MINALIGN
3021 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3022 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3024 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3025 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3029 /* Provide the correct kmalloc names now that the caches are up */
3030 for (i
= KMALLOC_SHIFT_LOW
; i
<= PAGE_SHIFT
; i
++)
3031 kmalloc_caches
[i
]. name
=
3032 kasprintf(GFP_KERNEL
, "kmalloc-%d", 1 << i
);
3035 register_cpu_notifier(&slab_notifier
);
3036 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3037 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3039 kmem_size
= sizeof(struct kmem_cache
);
3043 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3044 " CPUs=%d, Nodes=%d\n",
3045 caches
, cache_line_size(),
3046 slub_min_order
, slub_max_order
, slub_min_objects
,
3047 nr_cpu_ids
, nr_node_ids
);
3051 * Find a mergeable slab cache
3053 static int slab_unmergeable(struct kmem_cache
*s
)
3055 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3062 * We may have set a slab to be unmergeable during bootstrap.
3064 if (s
->refcount
< 0)
3070 static struct kmem_cache
*find_mergeable(size_t size
,
3071 size_t align
, unsigned long flags
, const char *name
,
3072 void (*ctor
)(struct kmem_cache
*, void *))
3074 struct kmem_cache
*s
;
3076 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3082 size
= ALIGN(size
, sizeof(void *));
3083 align
= calculate_alignment(flags
, align
, size
);
3084 size
= ALIGN(size
, align
);
3085 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3087 list_for_each_entry(s
, &slab_caches
, list
) {
3088 if (slab_unmergeable(s
))
3094 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3097 * Check if alignment is compatible.
3098 * Courtesy of Adrian Drzewiecki
3100 if ((s
->size
& ~(align
- 1)) != s
->size
)
3103 if (s
->size
- size
>= sizeof(void *))
3111 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3112 size_t align
, unsigned long flags
,
3113 void (*ctor
)(struct kmem_cache
*, void *))
3115 struct kmem_cache
*s
;
3117 down_write(&slub_lock
);
3118 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3124 * Adjust the object sizes so that we clear
3125 * the complete object on kzalloc.
3127 s
->objsize
= max(s
->objsize
, (int)size
);
3130 * And then we need to update the object size in the
3131 * per cpu structures
3133 for_each_online_cpu(cpu
)
3134 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3136 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3137 up_write(&slub_lock
);
3139 if (sysfs_slab_alias(s
, name
))
3144 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3146 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3147 size
, align
, flags
, ctor
)) {
3148 list_add(&s
->list
, &slab_caches
);
3149 up_write(&slub_lock
);
3150 if (sysfs_slab_add(s
))
3156 up_write(&slub_lock
);
3159 if (flags
& SLAB_PANIC
)
3160 panic("Cannot create slabcache %s\n", name
);
3165 EXPORT_SYMBOL(kmem_cache_create
);
3169 * Use the cpu notifier to insure that the cpu slabs are flushed when
3172 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3173 unsigned long action
, void *hcpu
)
3175 long cpu
= (long)hcpu
;
3176 struct kmem_cache
*s
;
3177 unsigned long flags
;
3180 case CPU_UP_PREPARE
:
3181 case CPU_UP_PREPARE_FROZEN
:
3182 init_alloc_cpu_cpu(cpu
);
3183 down_read(&slub_lock
);
3184 list_for_each_entry(s
, &slab_caches
, list
)
3185 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3187 up_read(&slub_lock
);
3190 case CPU_UP_CANCELED
:
3191 case CPU_UP_CANCELED_FROZEN
:
3193 case CPU_DEAD_FROZEN
:
3194 down_read(&slub_lock
);
3195 list_for_each_entry(s
, &slab_caches
, list
) {
3196 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3198 local_irq_save(flags
);
3199 __flush_cpu_slab(s
, cpu
);
3200 local_irq_restore(flags
);
3201 free_kmem_cache_cpu(c
, cpu
);
3202 s
->cpu_slab
[cpu
] = NULL
;
3204 up_read(&slub_lock
);
3212 static struct notifier_block __cpuinitdata slab_notifier
= {
3213 .notifier_call
= slab_cpuup_callback
3218 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, void *caller
)
3220 struct kmem_cache
*s
;
3222 if (unlikely(size
> PAGE_SIZE
))
3223 return kmalloc_large(size
, gfpflags
);
3225 s
= get_slab(size
, gfpflags
);
3227 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3230 return slab_alloc(s
, gfpflags
, -1, caller
);
3233 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3234 int node
, void *caller
)
3236 struct kmem_cache
*s
;
3238 if (unlikely(size
> PAGE_SIZE
))
3239 return kmalloc_large_node(size
, gfpflags
, node
);
3241 s
= get_slab(size
, gfpflags
);
3243 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3246 return slab_alloc(s
, gfpflags
, node
, caller
);
3249 #if (defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)) || defined(CONFIG_SLABINFO)
3250 static unsigned long count_partial(struct kmem_cache_node
*n
,
3251 int (*get_count
)(struct page
*))
3253 unsigned long flags
;
3254 unsigned long x
= 0;
3257 spin_lock_irqsave(&n
->list_lock
, flags
);
3258 list_for_each_entry(page
, &n
->partial
, lru
)
3259 x
+= get_count(page
);
3260 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3264 static int count_inuse(struct page
*page
)
3269 static int count_total(struct page
*page
)
3271 return page
->objects
;
3274 static int count_free(struct page
*page
)
3276 return page
->objects
- page
->inuse
;
3280 #if defined(CONFIG_SYSFS) && defined(CONFIG_SLUB_DEBUG)
3281 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3285 void *addr
= page_address(page
);
3287 if (!check_slab(s
, page
) ||
3288 !on_freelist(s
, page
, NULL
))
3291 /* Now we know that a valid freelist exists */
3292 bitmap_zero(map
, page
->objects
);
3294 for_each_free_object(p
, s
, page
->freelist
) {
3295 set_bit(slab_index(p
, s
, addr
), map
);
3296 if (!check_object(s
, page
, p
, 0))
3300 for_each_object(p
, s
, addr
, page
->objects
)
3301 if (!test_bit(slab_index(p
, s
, addr
), map
))
3302 if (!check_object(s
, page
, p
, 1))
3307 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3310 if (slab_trylock(page
)) {
3311 validate_slab(s
, page
, map
);
3314 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3317 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3318 if (!SlabDebug(page
))
3319 printk(KERN_ERR
"SLUB %s: SlabDebug not set "
3320 "on slab 0x%p\n", s
->name
, page
);
3322 if (SlabDebug(page
))
3323 printk(KERN_ERR
"SLUB %s: SlabDebug set on "
3324 "slab 0x%p\n", s
->name
, page
);
3328 static int validate_slab_node(struct kmem_cache
*s
,
3329 struct kmem_cache_node
*n
, unsigned long *map
)
3331 unsigned long count
= 0;
3333 unsigned long flags
;
3335 spin_lock_irqsave(&n
->list_lock
, flags
);
3337 list_for_each_entry(page
, &n
->partial
, lru
) {
3338 validate_slab_slab(s
, page
, map
);
3341 if (count
!= n
->nr_partial
)
3342 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3343 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3345 if (!(s
->flags
& SLAB_STORE_USER
))
3348 list_for_each_entry(page
, &n
->full
, lru
) {
3349 validate_slab_slab(s
, page
, map
);
3352 if (count
!= atomic_long_read(&n
->nr_slabs
))
3353 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3354 "counter=%ld\n", s
->name
, count
,
3355 atomic_long_read(&n
->nr_slabs
));
3358 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3362 static long validate_slab_cache(struct kmem_cache
*s
)
3365 unsigned long count
= 0;
3366 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3367 sizeof(unsigned long), GFP_KERNEL
);
3373 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3374 struct kmem_cache_node
*n
= get_node(s
, node
);
3376 count
+= validate_slab_node(s
, n
, map
);
3382 #ifdef SLUB_RESILIENCY_TEST
3383 static void resiliency_test(void)
3387 printk(KERN_ERR
"SLUB resiliency testing\n");
3388 printk(KERN_ERR
"-----------------------\n");
3389 printk(KERN_ERR
"A. Corruption after allocation\n");
3391 p
= kzalloc(16, GFP_KERNEL
);
3393 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3394 " 0x12->0x%p\n\n", p
+ 16);
3396 validate_slab_cache(kmalloc_caches
+ 4);
3398 /* Hmmm... The next two are dangerous */
3399 p
= kzalloc(32, GFP_KERNEL
);
3400 p
[32 + sizeof(void *)] = 0x34;
3401 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3402 " 0x34 -> -0x%p\n", p
);
3404 "If allocated object is overwritten then not detectable\n\n");
3406 validate_slab_cache(kmalloc_caches
+ 5);
3407 p
= kzalloc(64, GFP_KERNEL
);
3408 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3410 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3413 "If allocated object is overwritten then not detectable\n\n");
3414 validate_slab_cache(kmalloc_caches
+ 6);
3416 printk(KERN_ERR
"\nB. Corruption after free\n");
3417 p
= kzalloc(128, GFP_KERNEL
);
3420 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3421 validate_slab_cache(kmalloc_caches
+ 7);
3423 p
= kzalloc(256, GFP_KERNEL
);
3426 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3428 validate_slab_cache(kmalloc_caches
+ 8);
3430 p
= kzalloc(512, GFP_KERNEL
);
3433 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3434 validate_slab_cache(kmalloc_caches
+ 9);
3437 static void resiliency_test(void) {};
3441 * Generate lists of code addresses where slabcache objects are allocated
3446 unsigned long count
;
3459 unsigned long count
;
3460 struct location
*loc
;
3463 static void free_loc_track(struct loc_track
*t
)
3466 free_pages((unsigned long)t
->loc
,
3467 get_order(sizeof(struct location
) * t
->max
));
3470 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3475 order
= get_order(sizeof(struct location
) * max
);
3477 l
= (void *)__get_free_pages(flags
, order
);
3482 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3490 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3491 const struct track
*track
)
3493 long start
, end
, pos
;
3496 unsigned long age
= jiffies
- track
->when
;
3502 pos
= start
+ (end
- start
+ 1) / 2;
3505 * There is nothing at "end". If we end up there
3506 * we need to add something to before end.
3511 caddr
= t
->loc
[pos
].addr
;
3512 if (track
->addr
== caddr
) {
3518 if (age
< l
->min_time
)
3520 if (age
> l
->max_time
)
3523 if (track
->pid
< l
->min_pid
)
3524 l
->min_pid
= track
->pid
;
3525 if (track
->pid
> l
->max_pid
)
3526 l
->max_pid
= track
->pid
;
3528 cpu_set(track
->cpu
, l
->cpus
);
3530 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3534 if (track
->addr
< caddr
)
3541 * Not found. Insert new tracking element.
3543 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3549 (t
->count
- pos
) * sizeof(struct location
));
3552 l
->addr
= track
->addr
;
3556 l
->min_pid
= track
->pid
;
3557 l
->max_pid
= track
->pid
;
3558 cpus_clear(l
->cpus
);
3559 cpu_set(track
->cpu
, l
->cpus
);
3560 nodes_clear(l
->nodes
);
3561 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3565 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3566 struct page
*page
, enum track_item alloc
)
3568 void *addr
= page_address(page
);
3569 DECLARE_BITMAP(map
, page
->objects
);
3572 bitmap_zero(map
, page
->objects
);
3573 for_each_free_object(p
, s
, page
->freelist
)
3574 set_bit(slab_index(p
, s
, addr
), map
);
3576 for_each_object(p
, s
, addr
, page
->objects
)
3577 if (!test_bit(slab_index(p
, s
, addr
), map
))
3578 add_location(t
, s
, get_track(s
, p
, alloc
));
3581 static int list_locations(struct kmem_cache
*s
, char *buf
,
3582 enum track_item alloc
)
3586 struct loc_track t
= { 0, 0, NULL
};
3589 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3591 return sprintf(buf
, "Out of memory\n");
3593 /* Push back cpu slabs */
3596 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3597 struct kmem_cache_node
*n
= get_node(s
, node
);
3598 unsigned long flags
;
3601 if (!atomic_long_read(&n
->nr_slabs
))
3604 spin_lock_irqsave(&n
->list_lock
, flags
);
3605 list_for_each_entry(page
, &n
->partial
, lru
)
3606 process_slab(&t
, s
, page
, alloc
);
3607 list_for_each_entry(page
, &n
->full
, lru
)
3608 process_slab(&t
, s
, page
, alloc
);
3609 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3612 for (i
= 0; i
< t
.count
; i
++) {
3613 struct location
*l
= &t
.loc
[i
];
3615 if (len
> PAGE_SIZE
- 100)
3617 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3620 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3622 len
+= sprintf(buf
+ len
, "<not-available>");
3624 if (l
->sum_time
!= l
->min_time
) {
3625 unsigned long remainder
;
3627 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3629 div_long_long_rem(l
->sum_time
, l
->count
, &remainder
),
3632 len
+= sprintf(buf
+ len
, " age=%ld",
3635 if (l
->min_pid
!= l
->max_pid
)
3636 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3637 l
->min_pid
, l
->max_pid
);
3639 len
+= sprintf(buf
+ len
, " pid=%ld",
3642 if (num_online_cpus() > 1 && !cpus_empty(l
->cpus
) &&
3643 len
< PAGE_SIZE
- 60) {
3644 len
+= sprintf(buf
+ len
, " cpus=");
3645 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3649 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3650 len
< PAGE_SIZE
- 60) {
3651 len
+= sprintf(buf
+ len
, " nodes=");
3652 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3656 len
+= sprintf(buf
+ len
, "\n");
3661 len
+= sprintf(buf
, "No data\n");
3665 enum slab_stat_type
{
3666 SL_ALL
, /* All slabs */
3667 SL_PARTIAL
, /* Only partially allocated slabs */
3668 SL_CPU
, /* Only slabs used for cpu caches */
3669 SL_OBJECTS
, /* Determine allocated objects not slabs */
3670 SL_TOTAL
/* Determine object capacity not slabs */
3673 #define SO_ALL (1 << SL_ALL)
3674 #define SO_PARTIAL (1 << SL_PARTIAL)
3675 #define SO_CPU (1 << SL_CPU)
3676 #define SO_OBJECTS (1 << SL_OBJECTS)
3677 #define SO_TOTAL (1 << SL_TOTAL)
3679 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3680 char *buf
, unsigned long flags
)
3682 unsigned long total
= 0;
3685 unsigned long *nodes
;
3686 unsigned long *per_cpu
;
3688 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3691 per_cpu
= nodes
+ nr_node_ids
;
3693 if (flags
& SO_CPU
) {
3696 for_each_possible_cpu(cpu
) {
3697 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3699 if (!c
|| c
->node
< 0)
3703 if (flags
& SO_TOTAL
)
3704 x
= c
->page
->objects
;
3705 else if (flags
& SO_OBJECTS
)
3711 nodes
[c
->node
] += x
;
3717 if (flags
& SO_ALL
) {
3718 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3719 struct kmem_cache_node
*n
= get_node(s
, node
);
3721 if (flags
& SO_TOTAL
)
3722 x
= atomic_long_read(&n
->total_objects
);
3723 else if (flags
& SO_OBJECTS
)
3724 x
= atomic_long_read(&n
->total_objects
) -
3725 count_partial(n
, count_free
);
3728 x
= atomic_long_read(&n
->nr_slabs
);
3733 } else if (flags
& SO_PARTIAL
) {
3734 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3735 struct kmem_cache_node
*n
= get_node(s
, node
);
3737 if (flags
& SO_TOTAL
)
3738 x
= count_partial(n
, count_total
);
3739 else if (flags
& SO_OBJECTS
)
3740 x
= count_partial(n
, count_inuse
);
3747 x
= sprintf(buf
, "%lu", total
);
3749 for_each_node_state(node
, N_NORMAL_MEMORY
)
3751 x
+= sprintf(buf
+ x
, " N%d=%lu",
3755 return x
+ sprintf(buf
+ x
, "\n");
3758 static int any_slab_objects(struct kmem_cache
*s
)
3762 for_each_online_node(node
) {
3763 struct kmem_cache_node
*n
= get_node(s
, node
);
3768 if (atomic_read(&n
->total_objects
))
3774 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3775 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3777 struct slab_attribute
{
3778 struct attribute attr
;
3779 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3780 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3783 #define SLAB_ATTR_RO(_name) \
3784 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3786 #define SLAB_ATTR(_name) \
3787 static struct slab_attribute _name##_attr = \
3788 __ATTR(_name, 0644, _name##_show, _name##_store)
3790 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3792 return sprintf(buf
, "%d\n", s
->size
);
3794 SLAB_ATTR_RO(slab_size
);
3796 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3798 return sprintf(buf
, "%d\n", s
->align
);
3800 SLAB_ATTR_RO(align
);
3802 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3804 return sprintf(buf
, "%d\n", s
->objsize
);
3806 SLAB_ATTR_RO(object_size
);
3808 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3810 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3812 SLAB_ATTR_RO(objs_per_slab
);
3814 static ssize_t
order_store(struct kmem_cache
*s
,
3815 const char *buf
, size_t length
)
3817 int order
= simple_strtoul(buf
, NULL
, 10);
3819 if (order
> slub_max_order
|| order
< slub_min_order
)
3822 calculate_sizes(s
, order
);
3826 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3828 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3832 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3835 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3837 return n
+ sprintf(buf
+ n
, "\n");
3843 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3845 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3847 SLAB_ATTR_RO(aliases
);
3849 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3851 return show_slab_objects(s
, buf
, SO_ALL
);
3853 SLAB_ATTR_RO(slabs
);
3855 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3857 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3859 SLAB_ATTR_RO(partial
);
3861 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3863 return show_slab_objects(s
, buf
, SO_CPU
);
3865 SLAB_ATTR_RO(cpu_slabs
);
3867 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3869 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3871 SLAB_ATTR_RO(objects
);
3873 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3875 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3877 SLAB_ATTR_RO(objects_partial
);
3879 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3881 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3883 SLAB_ATTR_RO(total_objects
);
3885 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3887 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3890 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3891 const char *buf
, size_t length
)
3893 s
->flags
&= ~SLAB_DEBUG_FREE
;
3895 s
->flags
|= SLAB_DEBUG_FREE
;
3898 SLAB_ATTR(sanity_checks
);
3900 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
3902 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
3905 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
3908 s
->flags
&= ~SLAB_TRACE
;
3910 s
->flags
|= SLAB_TRACE
;
3915 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
3917 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
3920 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
3921 const char *buf
, size_t length
)
3923 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
3925 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
3928 SLAB_ATTR(reclaim_account
);
3930 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
3932 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
3934 SLAB_ATTR_RO(hwcache_align
);
3936 #ifdef CONFIG_ZONE_DMA
3937 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
3939 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
3941 SLAB_ATTR_RO(cache_dma
);
3944 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
3946 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
3948 SLAB_ATTR_RO(destroy_by_rcu
);
3950 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
3952 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
3955 static ssize_t
red_zone_store(struct kmem_cache
*s
,
3956 const char *buf
, size_t length
)
3958 if (any_slab_objects(s
))
3961 s
->flags
&= ~SLAB_RED_ZONE
;
3963 s
->flags
|= SLAB_RED_ZONE
;
3964 calculate_sizes(s
, -1);
3967 SLAB_ATTR(red_zone
);
3969 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
3971 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
3974 static ssize_t
poison_store(struct kmem_cache
*s
,
3975 const char *buf
, size_t length
)
3977 if (any_slab_objects(s
))
3980 s
->flags
&= ~SLAB_POISON
;
3982 s
->flags
|= SLAB_POISON
;
3983 calculate_sizes(s
, -1);
3988 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
3990 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
3993 static ssize_t
store_user_store(struct kmem_cache
*s
,
3994 const char *buf
, size_t length
)
3996 if (any_slab_objects(s
))
3999 s
->flags
&= ~SLAB_STORE_USER
;
4001 s
->flags
|= SLAB_STORE_USER
;
4002 calculate_sizes(s
, -1);
4005 SLAB_ATTR(store_user
);
4007 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4012 static ssize_t
validate_store(struct kmem_cache
*s
,
4013 const char *buf
, size_t length
)
4017 if (buf
[0] == '1') {
4018 ret
= validate_slab_cache(s
);
4024 SLAB_ATTR(validate
);
4026 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4031 static ssize_t
shrink_store(struct kmem_cache
*s
,
4032 const char *buf
, size_t length
)
4034 if (buf
[0] == '1') {
4035 int rc
= kmem_cache_shrink(s
);
4045 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4047 if (!(s
->flags
& SLAB_STORE_USER
))
4049 return list_locations(s
, buf
, TRACK_ALLOC
);
4051 SLAB_ATTR_RO(alloc_calls
);
4053 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4055 if (!(s
->flags
& SLAB_STORE_USER
))
4057 return list_locations(s
, buf
, TRACK_FREE
);
4059 SLAB_ATTR_RO(free_calls
);
4062 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4064 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4067 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4068 const char *buf
, size_t length
)
4070 int n
= simple_strtoul(buf
, NULL
, 10);
4073 s
->remote_node_defrag_ratio
= n
* 10;
4076 SLAB_ATTR(remote_node_defrag_ratio
);
4079 #ifdef CONFIG_SLUB_STATS
4080 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4082 unsigned long sum
= 0;
4085 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4090 for_each_online_cpu(cpu
) {
4091 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4097 len
= sprintf(buf
, "%lu", sum
);
4100 for_each_online_cpu(cpu
) {
4101 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4102 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4106 return len
+ sprintf(buf
+ len
, "\n");
4109 #define STAT_ATTR(si, text) \
4110 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4112 return show_stat(s, buf, si); \
4114 SLAB_ATTR_RO(text); \
4116 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4117 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4118 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4119 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4120 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4121 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4122 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4123 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4124 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4125 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4126 STAT_ATTR(FREE_SLAB
, free_slab
);
4127 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4128 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4129 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4130 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4131 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4132 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4133 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4136 static struct attribute
*slab_attrs
[] = {
4137 &slab_size_attr
.attr
,
4138 &object_size_attr
.attr
,
4139 &objs_per_slab_attr
.attr
,
4142 &objects_partial_attr
.attr
,
4143 &total_objects_attr
.attr
,
4146 &cpu_slabs_attr
.attr
,
4150 &sanity_checks_attr
.attr
,
4152 &hwcache_align_attr
.attr
,
4153 &reclaim_account_attr
.attr
,
4154 &destroy_by_rcu_attr
.attr
,
4155 &red_zone_attr
.attr
,
4157 &store_user_attr
.attr
,
4158 &validate_attr
.attr
,
4160 &alloc_calls_attr
.attr
,
4161 &free_calls_attr
.attr
,
4162 #ifdef CONFIG_ZONE_DMA
4163 &cache_dma_attr
.attr
,
4166 &remote_node_defrag_ratio_attr
.attr
,
4168 #ifdef CONFIG_SLUB_STATS
4169 &alloc_fastpath_attr
.attr
,
4170 &alloc_slowpath_attr
.attr
,
4171 &free_fastpath_attr
.attr
,
4172 &free_slowpath_attr
.attr
,
4173 &free_frozen_attr
.attr
,
4174 &free_add_partial_attr
.attr
,
4175 &free_remove_partial_attr
.attr
,
4176 &alloc_from_partial_attr
.attr
,
4177 &alloc_slab_attr
.attr
,
4178 &alloc_refill_attr
.attr
,
4179 &free_slab_attr
.attr
,
4180 &cpuslab_flush_attr
.attr
,
4181 &deactivate_full_attr
.attr
,
4182 &deactivate_empty_attr
.attr
,
4183 &deactivate_to_head_attr
.attr
,
4184 &deactivate_to_tail_attr
.attr
,
4185 &deactivate_remote_frees_attr
.attr
,
4186 &order_fallback_attr
.attr
,
4191 static struct attribute_group slab_attr_group
= {
4192 .attrs
= slab_attrs
,
4195 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4196 struct attribute
*attr
,
4199 struct slab_attribute
*attribute
;
4200 struct kmem_cache
*s
;
4203 attribute
= to_slab_attr(attr
);
4206 if (!attribute
->show
)
4209 err
= attribute
->show(s
, buf
);
4214 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4215 struct attribute
*attr
,
4216 const char *buf
, size_t len
)
4218 struct slab_attribute
*attribute
;
4219 struct kmem_cache
*s
;
4222 attribute
= to_slab_attr(attr
);
4225 if (!attribute
->store
)
4228 err
= attribute
->store(s
, buf
, len
);
4233 static void kmem_cache_release(struct kobject
*kobj
)
4235 struct kmem_cache
*s
= to_slab(kobj
);
4240 static struct sysfs_ops slab_sysfs_ops
= {
4241 .show
= slab_attr_show
,
4242 .store
= slab_attr_store
,
4245 static struct kobj_type slab_ktype
= {
4246 .sysfs_ops
= &slab_sysfs_ops
,
4247 .release
= kmem_cache_release
4250 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4252 struct kobj_type
*ktype
= get_ktype(kobj
);
4254 if (ktype
== &slab_ktype
)
4259 static struct kset_uevent_ops slab_uevent_ops
= {
4260 .filter
= uevent_filter
,
4263 static struct kset
*slab_kset
;
4265 #define ID_STR_LENGTH 64
4267 /* Create a unique string id for a slab cache:
4269 * Format :[flags-]size
4271 static char *create_unique_id(struct kmem_cache
*s
)
4273 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4280 * First flags affecting slabcache operations. We will only
4281 * get here for aliasable slabs so we do not need to support
4282 * too many flags. The flags here must cover all flags that
4283 * are matched during merging to guarantee that the id is
4286 if (s
->flags
& SLAB_CACHE_DMA
)
4288 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4290 if (s
->flags
& SLAB_DEBUG_FREE
)
4294 p
+= sprintf(p
, "%07d", s
->size
);
4295 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4299 static int sysfs_slab_add(struct kmem_cache
*s
)
4305 if (slab_state
< SYSFS
)
4306 /* Defer until later */
4309 unmergeable
= slab_unmergeable(s
);
4312 * Slabcache can never be merged so we can use the name proper.
4313 * This is typically the case for debug situations. In that
4314 * case we can catch duplicate names easily.
4316 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4320 * Create a unique name for the slab as a target
4323 name
= create_unique_id(s
);
4326 s
->kobj
.kset
= slab_kset
;
4327 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4329 kobject_put(&s
->kobj
);
4333 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4336 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4338 /* Setup first alias */
4339 sysfs_slab_alias(s
, s
->name
);
4345 static void sysfs_slab_remove(struct kmem_cache
*s
)
4347 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4348 kobject_del(&s
->kobj
);
4349 kobject_put(&s
->kobj
);
4353 * Need to buffer aliases during bootup until sysfs becomes
4354 * available lest we loose that information.
4356 struct saved_alias
{
4357 struct kmem_cache
*s
;
4359 struct saved_alias
*next
;
4362 static struct saved_alias
*alias_list
;
4364 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4366 struct saved_alias
*al
;
4368 if (slab_state
== SYSFS
) {
4370 * If we have a leftover link then remove it.
4372 sysfs_remove_link(&slab_kset
->kobj
, name
);
4373 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4376 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4382 al
->next
= alias_list
;
4387 static int __init
slab_sysfs_init(void)
4389 struct kmem_cache
*s
;
4392 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4394 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4400 list_for_each_entry(s
, &slab_caches
, list
) {
4401 err
= sysfs_slab_add(s
);
4403 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4404 " to sysfs\n", s
->name
);
4407 while (alias_list
) {
4408 struct saved_alias
*al
= alias_list
;
4410 alias_list
= alias_list
->next
;
4411 err
= sysfs_slab_alias(al
->s
, al
->name
);
4413 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4414 " %s to sysfs\n", s
->name
);
4422 __initcall(slab_sysfs_init
);
4426 * The /proc/slabinfo ABI
4428 #ifdef CONFIG_SLABINFO
4430 ssize_t
slabinfo_write(struct file
*file
, const char __user
* buffer
,
4431 size_t count
, loff_t
*ppos
)
4437 static void print_slabinfo_header(struct seq_file
*m
)
4439 seq_puts(m
, "slabinfo - version: 2.1\n");
4440 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4441 "<objperslab> <pagesperslab>");
4442 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4443 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4447 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4451 down_read(&slub_lock
);
4453 print_slabinfo_header(m
);
4455 return seq_list_start(&slab_caches
, *pos
);
4458 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4460 return seq_list_next(p
, &slab_caches
, pos
);
4463 static void s_stop(struct seq_file
*m
, void *p
)
4465 up_read(&slub_lock
);
4468 static int s_show(struct seq_file
*m
, void *p
)
4470 unsigned long nr_partials
= 0;
4471 unsigned long nr_slabs
= 0;
4472 unsigned long nr_inuse
= 0;
4473 unsigned long nr_objs
= 0;
4474 unsigned long nr_free
= 0;
4475 struct kmem_cache
*s
;
4478 s
= list_entry(p
, struct kmem_cache
, list
);
4480 for_each_online_node(node
) {
4481 struct kmem_cache_node
*n
= get_node(s
, node
);
4486 nr_partials
+= n
->nr_partial
;
4487 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4488 nr_objs
+= atomic_long_read(&n
->total_objects
);
4489 nr_free
+= count_partial(n
, count_free
);
4492 nr_inuse
= nr_objs
- nr_free
;
4494 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4495 nr_objs
, s
->size
, oo_objects(s
->oo
),
4496 (1 << oo_order(s
->oo
)));
4497 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4498 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4504 const struct seq_operations slabinfo_op
= {
4511 #endif /* CONFIG_SLABINFO */