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
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/proc_fs.h>
18 #include <linux/seq_file.h>
19 #include <linux/cpu.h>
20 #include <linux/cpuset.h>
21 #include <linux/mempolicy.h>
22 #include <linux/ctype.h>
23 #include <linux/debugobjects.h>
24 #include <linux/kallsyms.h>
25 #include <linux/memory.h>
26 #include <linux/math64.h>
33 * The slab_lock protects operations on the object of a particular
34 * slab and its metadata in the page struct. If the slab lock
35 * has been taken then no allocations nor frees can be performed
36 * on the objects in the slab nor can the slab be added or removed
37 * from the partial or full lists since this would mean modifying
38 * the page_struct of the slab.
40 * The list_lock protects the partial and full list on each node and
41 * the partial slab counter. If taken then no new slabs may be added or
42 * removed from the lists nor make the number of partial slabs be modified.
43 * (Note that the total number of slabs is an atomic value that may be
44 * modified without taking the list lock).
46 * The list_lock is a centralized lock and thus we avoid taking it as
47 * much as possible. As long as SLUB does not have to handle partial
48 * slabs, operations can continue without any centralized lock. F.e.
49 * allocating a long series of objects that fill up slabs does not require
52 * The lock order is sometimes inverted when we are trying to get a slab
53 * off a list. We take the list_lock and then look for a page on the list
54 * to use. While we do that objects in the slabs may be freed. We can
55 * only operate on the slab if we have also taken the slab_lock. So we use
56 * a slab_trylock() on the slab. If trylock was successful then no frees
57 * can occur anymore and we can use the slab for allocations etc. If the
58 * slab_trylock() does not succeed then frees are in progress in the slab and
59 * we must stay away from it for a while since we may cause a bouncing
60 * cacheline if we try to acquire the lock. So go onto the next slab.
61 * If all pages are busy then we may allocate a new slab instead of reusing
62 * a partial slab. A new slab has noone operating on it and thus there is
63 * no danger of cacheline contention.
65 * Interrupts are disabled during allocation and deallocation in order to
66 * make the slab allocator safe to use in the context of an irq. In addition
67 * interrupts are disabled to ensure that the processor does not change
68 * while handling per_cpu slabs, due to kernel preemption.
70 * SLUB assigns one slab for allocation to each processor.
71 * Allocations only occur from these slabs called cpu slabs.
73 * Slabs with free elements are kept on a partial list and during regular
74 * operations no list for full slabs is used. If an object in a full slab is
75 * freed then the slab will show up again on the partial lists.
76 * We track full slabs for debugging purposes though because otherwise we
77 * cannot scan all objects.
79 * Slabs are freed when they become empty. Teardown and setup is
80 * minimal so we rely on the page allocators per cpu caches for
81 * fast frees and allocs.
83 * Overloading of page flags that are otherwise used for LRU management.
85 * PageActive The slab is frozen and exempt from list processing.
86 * This means that the slab is dedicated to a purpose
87 * such as satisfying allocations for a specific
88 * processor. Objects may be freed in the slab while
89 * it is frozen but slab_free will then skip the usual
90 * list operations. It is up to the processor holding
91 * the slab to integrate the slab into the slab lists
92 * when the slab is no longer needed.
94 * One use of this flag is to mark slabs that are
95 * used for allocations. Then such a slab becomes a cpu
96 * slab. The cpu slab may be equipped with an additional
97 * freelist that allows lockless access to
98 * free objects in addition to the regular freelist
99 * that requires the slab lock.
101 * PageError Slab requires special handling due to debug
102 * options set. This moves slab handling out of
103 * the fast path and disables lockless freelists.
106 #ifdef CONFIG_SLUB_DEBUG
113 * Issues still to be resolved:
115 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
117 * - Variable sizing of the per node arrays
120 /* Enable to test recovery from slab corruption on boot */
121 #undef SLUB_RESILIENCY_TEST
124 * Mininum number of partial slabs. These will be left on the partial
125 * lists even if they are empty. kmem_cache_shrink may reclaim them.
127 #define MIN_PARTIAL 5
130 * Maximum number of desirable partial slabs.
131 * The existence of more partial slabs makes kmem_cache_shrink
132 * sort the partial list by the number of objects in the.
134 #define MAX_PARTIAL 10
136 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
137 SLAB_POISON | SLAB_STORE_USER)
140 * Set of flags that will prevent slab merging
142 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
143 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
145 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
148 #ifndef ARCH_KMALLOC_MINALIGN
149 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
152 #ifndef ARCH_SLAB_MINALIGN
153 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
157 #define OO_MASK ((1 << OO_SHIFT) - 1)
158 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
160 /* Internal SLUB flags */
161 #define __OBJECT_POISON 0x80000000 /* Poison object */
162 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
164 static int kmem_size
= sizeof(struct kmem_cache
);
167 static struct notifier_block slab_notifier
;
171 DOWN
, /* No slab functionality available */
172 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
173 UP
, /* Everything works but does not show up in sysfs */
177 /* A list of all slab caches on the system */
178 static DECLARE_RWSEM(slub_lock
);
179 static LIST_HEAD(slab_caches
);
182 * Tracking user of a slab.
185 unsigned long addr
; /* Called from address */
186 int cpu
; /* Was running on cpu */
187 int pid
; /* Pid context */
188 unsigned long when
; /* When did the operation occur */
191 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
193 #ifdef CONFIG_SLUB_DEBUG
194 static int sysfs_slab_add(struct kmem_cache
*);
195 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
196 static void sysfs_slab_remove(struct kmem_cache
*);
199 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
200 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
202 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
209 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
211 #ifdef CONFIG_SLUB_STATS
216 /********************************************************************
217 * Core slab cache functions
218 *******************************************************************/
220 int slab_is_available(void)
222 return slab_state
>= UP
;
225 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
228 return s
->node
[node
];
230 return &s
->local_node
;
234 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
237 return s
->cpu_slab
[cpu
];
243 /* Verify that a pointer has an address that is valid within a slab page */
244 static inline int check_valid_pointer(struct kmem_cache
*s
,
245 struct page
*page
, const void *object
)
252 base
= page_address(page
);
253 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
254 (object
- base
) % s
->size
) {
262 * Slow version of get and set free pointer.
264 * This version requires touching the cache lines of kmem_cache which
265 * we avoid to do in the fast alloc free paths. There we obtain the offset
266 * from the page struct.
268 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
270 return *(void **)(object
+ s
->offset
);
273 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
275 *(void **)(object
+ s
->offset
) = fp
;
278 /* Loop over all objects in a slab */
279 #define for_each_object(__p, __s, __addr, __objects) \
280 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
284 #define for_each_free_object(__p, __s, __free) \
285 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
287 /* Determine object index from a given position */
288 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
290 return (p
- addr
) / s
->size
;
293 static inline struct kmem_cache_order_objects
oo_make(int order
,
296 struct kmem_cache_order_objects x
= {
297 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
303 static inline int oo_order(struct kmem_cache_order_objects x
)
305 return x
.x
>> OO_SHIFT
;
308 static inline int oo_objects(struct kmem_cache_order_objects x
)
310 return x
.x
& OO_MASK
;
313 #ifdef CONFIG_SLUB_DEBUG
317 #ifdef CONFIG_SLUB_DEBUG_ON
318 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
320 static int slub_debug
;
323 static char *slub_debug_slabs
;
328 static void print_section(char *text
, u8
*addr
, unsigned int length
)
336 for (i
= 0; i
< length
; i
++) {
338 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
341 printk(KERN_CONT
" %02x", addr
[i
]);
343 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
345 printk(KERN_CONT
" %s\n", ascii
);
352 printk(KERN_CONT
" ");
356 printk(KERN_CONT
" %s\n", ascii
);
360 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
361 enum track_item alloc
)
366 p
= object
+ s
->offset
+ sizeof(void *);
368 p
= object
+ s
->inuse
;
373 static void set_track(struct kmem_cache
*s
, void *object
,
374 enum track_item alloc
, unsigned long addr
)
379 p
= object
+ s
->offset
+ sizeof(void *);
381 p
= object
+ s
->inuse
;
386 p
->cpu
= smp_processor_id();
387 p
->pid
= current
->pid
;
390 memset(p
, 0, sizeof(struct track
));
393 static void init_tracking(struct kmem_cache
*s
, void *object
)
395 if (!(s
->flags
& SLAB_STORE_USER
))
398 set_track(s
, object
, TRACK_FREE
, 0UL);
399 set_track(s
, object
, TRACK_ALLOC
, 0UL);
402 static void print_track(const char *s
, struct track
*t
)
407 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
408 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
411 static void print_tracking(struct kmem_cache
*s
, void *object
)
413 if (!(s
->flags
& SLAB_STORE_USER
))
416 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
417 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
420 static void print_page_info(struct page
*page
)
422 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
423 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
427 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
433 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
435 printk(KERN_ERR
"========================================"
436 "=====================================\n");
437 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
438 printk(KERN_ERR
"----------------------------------------"
439 "-------------------------------------\n\n");
442 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
448 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
450 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
453 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
455 unsigned int off
; /* Offset of last byte */
456 u8
*addr
= page_address(page
);
458 print_tracking(s
, p
);
460 print_page_info(page
);
462 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
463 p
, p
- addr
, get_freepointer(s
, p
));
466 print_section("Bytes b4", p
- 16, 16);
468 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
470 if (s
->flags
& SLAB_RED_ZONE
)
471 print_section("Redzone", p
+ s
->objsize
,
472 s
->inuse
- s
->objsize
);
475 off
= s
->offset
+ sizeof(void *);
479 if (s
->flags
& SLAB_STORE_USER
)
480 off
+= 2 * sizeof(struct track
);
483 /* Beginning of the filler is the free pointer */
484 print_section("Padding", p
+ off
, s
->size
- off
);
489 static void object_err(struct kmem_cache
*s
, struct page
*page
,
490 u8
*object
, char *reason
)
492 slab_bug(s
, "%s", reason
);
493 print_trailer(s
, page
, object
);
496 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
502 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
504 slab_bug(s
, "%s", buf
);
505 print_page_info(page
);
509 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
513 if (s
->flags
& __OBJECT_POISON
) {
514 memset(p
, POISON_FREE
, s
->objsize
- 1);
515 p
[s
->objsize
- 1] = POISON_END
;
518 if (s
->flags
& SLAB_RED_ZONE
)
519 memset(p
+ s
->objsize
,
520 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
521 s
->inuse
- s
->objsize
);
524 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
527 if (*start
!= (u8
)value
)
535 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
536 void *from
, void *to
)
538 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
539 memset(from
, data
, to
- from
);
542 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
543 u8
*object
, char *what
,
544 u8
*start
, unsigned int value
, unsigned int bytes
)
549 fault
= check_bytes(start
, value
, bytes
);
554 while (end
> fault
&& end
[-1] == value
)
557 slab_bug(s
, "%s overwritten", what
);
558 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
559 fault
, end
- 1, fault
[0], value
);
560 print_trailer(s
, page
, object
);
562 restore_bytes(s
, what
, value
, fault
, end
);
570 * Bytes of the object to be managed.
571 * If the freepointer may overlay the object then the free
572 * pointer is the first word of the object.
574 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
577 * object + s->objsize
578 * Padding to reach word boundary. This is also used for Redzoning.
579 * Padding is extended by another word if Redzoning is enabled and
582 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
583 * 0xcc (RED_ACTIVE) for objects in use.
586 * Meta data starts here.
588 * A. Free pointer (if we cannot overwrite object on free)
589 * B. Tracking data for SLAB_STORE_USER
590 * C. Padding to reach required alignment boundary or at mininum
591 * one word if debugging is on to be able to detect writes
592 * before the word boundary.
594 * Padding is done using 0x5a (POISON_INUSE)
597 * Nothing is used beyond s->size.
599 * If slabcaches are merged then the objsize and inuse boundaries are mostly
600 * ignored. And therefore no slab options that rely on these boundaries
601 * may be used with merged slabcaches.
604 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
606 unsigned long off
= s
->inuse
; /* The end of info */
609 /* Freepointer is placed after the object. */
610 off
+= sizeof(void *);
612 if (s
->flags
& SLAB_STORE_USER
)
613 /* We also have user information there */
614 off
+= 2 * sizeof(struct track
);
619 return check_bytes_and_report(s
, page
, p
, "Object padding",
620 p
+ off
, POISON_INUSE
, s
->size
- off
);
623 /* Check the pad bytes at the end of a slab page */
624 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
632 if (!(s
->flags
& SLAB_POISON
))
635 start
= page_address(page
);
636 length
= (PAGE_SIZE
<< compound_order(page
));
637 end
= start
+ length
;
638 remainder
= length
% s
->size
;
642 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
645 while (end
> fault
&& end
[-1] == POISON_INUSE
)
648 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
649 print_section("Padding", end
- remainder
, remainder
);
651 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
655 static int check_object(struct kmem_cache
*s
, struct page
*page
,
656 void *object
, int active
)
659 u8
*endobject
= object
+ s
->objsize
;
661 if (s
->flags
& SLAB_RED_ZONE
) {
663 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
665 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
666 endobject
, red
, s
->inuse
- s
->objsize
))
669 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
670 check_bytes_and_report(s
, page
, p
, "Alignment padding",
671 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
675 if (s
->flags
& SLAB_POISON
) {
676 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
677 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
678 POISON_FREE
, s
->objsize
- 1) ||
679 !check_bytes_and_report(s
, page
, p
, "Poison",
680 p
+ s
->objsize
- 1, POISON_END
, 1)))
683 * check_pad_bytes cleans up on its own.
685 check_pad_bytes(s
, page
, p
);
688 if (!s
->offset
&& active
)
690 * Object and freepointer overlap. Cannot check
691 * freepointer while object is allocated.
695 /* Check free pointer validity */
696 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
697 object_err(s
, page
, p
, "Freepointer corrupt");
699 * No choice but to zap it and thus loose the remainder
700 * of the free objects in this slab. May cause
701 * another error because the object count is now wrong.
703 set_freepointer(s
, p
, NULL
);
709 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
713 VM_BUG_ON(!irqs_disabled());
715 if (!PageSlab(page
)) {
716 slab_err(s
, page
, "Not a valid slab page");
720 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
721 if (page
->objects
> maxobj
) {
722 slab_err(s
, page
, "objects %u > max %u",
723 s
->name
, page
->objects
, maxobj
);
726 if (page
->inuse
> page
->objects
) {
727 slab_err(s
, page
, "inuse %u > max %u",
728 s
->name
, page
->inuse
, page
->objects
);
731 /* Slab_pad_check fixes things up after itself */
732 slab_pad_check(s
, page
);
737 * Determine if a certain object on a page is on the freelist. Must hold the
738 * slab lock to guarantee that the chains are in a consistent state.
740 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
743 void *fp
= page
->freelist
;
745 unsigned long max_objects
;
747 while (fp
&& nr
<= page
->objects
) {
750 if (!check_valid_pointer(s
, page
, fp
)) {
752 object_err(s
, page
, object
,
753 "Freechain corrupt");
754 set_freepointer(s
, object
, NULL
);
757 slab_err(s
, page
, "Freepointer corrupt");
758 page
->freelist
= NULL
;
759 page
->inuse
= page
->objects
;
760 slab_fix(s
, "Freelist cleared");
766 fp
= get_freepointer(s
, object
);
770 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
771 if (max_objects
> MAX_OBJS_PER_PAGE
)
772 max_objects
= MAX_OBJS_PER_PAGE
;
774 if (page
->objects
!= max_objects
) {
775 slab_err(s
, page
, "Wrong number of objects. Found %d but "
776 "should be %d", page
->objects
, max_objects
);
777 page
->objects
= max_objects
;
778 slab_fix(s
, "Number of objects adjusted.");
780 if (page
->inuse
!= page
->objects
- nr
) {
781 slab_err(s
, page
, "Wrong object count. Counter is %d but "
782 "counted were %d", page
->inuse
, page
->objects
- nr
);
783 page
->inuse
= page
->objects
- nr
;
784 slab_fix(s
, "Object count adjusted.");
786 return search
== NULL
;
789 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
792 if (s
->flags
& SLAB_TRACE
) {
793 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
795 alloc
? "alloc" : "free",
800 print_section("Object", (void *)object
, s
->objsize
);
807 * Tracking of fully allocated slabs for debugging purposes.
809 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
811 spin_lock(&n
->list_lock
);
812 list_add(&page
->lru
, &n
->full
);
813 spin_unlock(&n
->list_lock
);
816 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
818 struct kmem_cache_node
*n
;
820 if (!(s
->flags
& SLAB_STORE_USER
))
823 n
= get_node(s
, page_to_nid(page
));
825 spin_lock(&n
->list_lock
);
826 list_del(&page
->lru
);
827 spin_unlock(&n
->list_lock
);
830 /* Tracking of the number of slabs for debugging purposes */
831 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
833 struct kmem_cache_node
*n
= get_node(s
, node
);
835 return atomic_long_read(&n
->nr_slabs
);
838 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
840 struct kmem_cache_node
*n
= get_node(s
, node
);
843 * May be called early in order to allocate a slab for the
844 * kmem_cache_node structure. Solve the chicken-egg
845 * dilemma by deferring the increment of the count during
846 * bootstrap (see early_kmem_cache_node_alloc).
848 if (!NUMA_BUILD
|| n
) {
849 atomic_long_inc(&n
->nr_slabs
);
850 atomic_long_add(objects
, &n
->total_objects
);
853 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
855 struct kmem_cache_node
*n
= get_node(s
, node
);
857 atomic_long_dec(&n
->nr_slabs
);
858 atomic_long_sub(objects
, &n
->total_objects
);
861 /* Object debug checks for alloc/free paths */
862 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
865 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
868 init_object(s
, object
, 0);
869 init_tracking(s
, object
);
872 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
873 void *object
, unsigned long addr
)
875 if (!check_slab(s
, page
))
878 if (!on_freelist(s
, page
, object
)) {
879 object_err(s
, page
, object
, "Object already allocated");
883 if (!check_valid_pointer(s
, page
, object
)) {
884 object_err(s
, page
, object
, "Freelist Pointer check fails");
888 if (!check_object(s
, page
, object
, 0))
891 /* Success perform special debug activities for allocs */
892 if (s
->flags
& SLAB_STORE_USER
)
893 set_track(s
, object
, TRACK_ALLOC
, addr
);
894 trace(s
, page
, object
, 1);
895 init_object(s
, object
, 1);
899 if (PageSlab(page
)) {
901 * If this is a slab page then lets do the best we can
902 * to avoid issues in the future. Marking all objects
903 * as used avoids touching the remaining objects.
905 slab_fix(s
, "Marking all objects used");
906 page
->inuse
= page
->objects
;
907 page
->freelist
= NULL
;
912 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
913 void *object
, unsigned long addr
)
915 if (!check_slab(s
, page
))
918 if (!check_valid_pointer(s
, page
, object
)) {
919 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
923 if (on_freelist(s
, page
, object
)) {
924 object_err(s
, page
, object
, "Object already free");
928 if (!check_object(s
, page
, object
, 1))
931 if (unlikely(s
!= page
->slab
)) {
932 if (!PageSlab(page
)) {
933 slab_err(s
, page
, "Attempt to free object(0x%p) "
934 "outside of slab", object
);
935 } else if (!page
->slab
) {
937 "SLUB <none>: no slab for object 0x%p.\n",
941 object_err(s
, page
, object
,
942 "page slab pointer corrupt.");
946 /* Special debug activities for freeing objects */
947 if (!PageSlubFrozen(page
) && !page
->freelist
)
948 remove_full(s
, page
);
949 if (s
->flags
& SLAB_STORE_USER
)
950 set_track(s
, object
, TRACK_FREE
, addr
);
951 trace(s
, page
, object
, 0);
952 init_object(s
, object
, 0);
956 slab_fix(s
, "Object at 0x%p not freed", object
);
960 static int __init
setup_slub_debug(char *str
)
962 slub_debug
= DEBUG_DEFAULT_FLAGS
;
963 if (*str
++ != '=' || !*str
)
965 * No options specified. Switch on full debugging.
971 * No options but restriction on slabs. This means full
972 * debugging for slabs matching a pattern.
979 * Switch off all debugging measures.
984 * Determine which debug features should be switched on
986 for (; *str
&& *str
!= ','; str
++) {
987 switch (tolower(*str
)) {
989 slub_debug
|= SLAB_DEBUG_FREE
;
992 slub_debug
|= SLAB_RED_ZONE
;
995 slub_debug
|= SLAB_POISON
;
998 slub_debug
|= SLAB_STORE_USER
;
1001 slub_debug
|= SLAB_TRACE
;
1004 printk(KERN_ERR
"slub_debug option '%c' "
1005 "unknown. skipped\n", *str
);
1011 slub_debug_slabs
= str
+ 1;
1016 __setup("slub_debug", setup_slub_debug
);
1018 static unsigned long kmem_cache_flags(unsigned long objsize
,
1019 unsigned long flags
, const char *name
,
1020 void (*ctor
)(void *))
1023 * Enable debugging if selected on the kernel commandline.
1025 if (slub_debug
&& (!slub_debug_slabs
||
1026 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1027 flags
|= slub_debug
;
1032 static inline void setup_object_debug(struct kmem_cache
*s
,
1033 struct page
*page
, void *object
) {}
1035 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1036 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1038 static inline int free_debug_processing(struct kmem_cache
*s
,
1039 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1041 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1043 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1044 void *object
, int active
) { return 1; }
1045 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1046 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1047 unsigned long flags
, const char *name
,
1048 void (*ctor
)(void *))
1052 #define slub_debug 0
1054 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1056 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1058 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1063 * Slab allocation and freeing
1065 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1066 struct kmem_cache_order_objects oo
)
1068 int order
= oo_order(oo
);
1071 return alloc_pages(flags
, order
);
1073 return alloc_pages_node(node
, flags
, order
);
1076 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1079 struct kmem_cache_order_objects oo
= s
->oo
;
1081 flags
|= s
->allocflags
;
1083 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1085 if (unlikely(!page
)) {
1088 * Allocation may have failed due to fragmentation.
1089 * Try a lower order alloc if possible
1091 page
= alloc_slab_page(flags
, node
, oo
);
1095 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1097 page
->objects
= oo_objects(oo
);
1098 mod_zone_page_state(page_zone(page
),
1099 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1100 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1106 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1109 setup_object_debug(s
, page
, object
);
1110 if (unlikely(s
->ctor
))
1114 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1121 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1123 page
= allocate_slab(s
,
1124 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1128 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1130 page
->flags
|= 1 << PG_slab
;
1131 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1132 SLAB_STORE_USER
| SLAB_TRACE
))
1133 __SetPageSlubDebug(page
);
1135 start
= page_address(page
);
1137 if (unlikely(s
->flags
& SLAB_POISON
))
1138 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1141 for_each_object(p
, s
, start
, page
->objects
) {
1142 setup_object(s
, page
, last
);
1143 set_freepointer(s
, last
, p
);
1146 setup_object(s
, page
, last
);
1147 set_freepointer(s
, last
, NULL
);
1149 page
->freelist
= start
;
1155 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1157 int order
= compound_order(page
);
1158 int pages
= 1 << order
;
1160 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1163 slab_pad_check(s
, page
);
1164 for_each_object(p
, s
, page_address(page
),
1166 check_object(s
, page
, p
, 0);
1167 __ClearPageSlubDebug(page
);
1170 mod_zone_page_state(page_zone(page
),
1171 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1172 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1175 __ClearPageSlab(page
);
1176 reset_page_mapcount(page
);
1177 __free_pages(page
, order
);
1180 static void rcu_free_slab(struct rcu_head
*h
)
1184 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1185 __free_slab(page
->slab
, page
);
1188 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1190 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1192 * RCU free overloads the RCU head over the LRU
1194 struct rcu_head
*head
= (void *)&page
->lru
;
1196 call_rcu(head
, rcu_free_slab
);
1198 __free_slab(s
, page
);
1201 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1203 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1208 * Per slab locking using the pagelock
1210 static __always_inline
void slab_lock(struct page
*page
)
1212 bit_spin_lock(PG_locked
, &page
->flags
);
1215 static __always_inline
void slab_unlock(struct page
*page
)
1217 __bit_spin_unlock(PG_locked
, &page
->flags
);
1220 static __always_inline
int slab_trylock(struct page
*page
)
1224 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1229 * Management of partially allocated slabs
1231 static void add_partial(struct kmem_cache_node
*n
,
1232 struct page
*page
, int tail
)
1234 spin_lock(&n
->list_lock
);
1237 list_add_tail(&page
->lru
, &n
->partial
);
1239 list_add(&page
->lru
, &n
->partial
);
1240 spin_unlock(&n
->list_lock
);
1243 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1245 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1247 spin_lock(&n
->list_lock
);
1248 list_del(&page
->lru
);
1250 spin_unlock(&n
->list_lock
);
1254 * Lock slab and remove from the partial list.
1256 * Must hold list_lock.
1258 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1261 if (slab_trylock(page
)) {
1262 list_del(&page
->lru
);
1264 __SetPageSlubFrozen(page
);
1271 * Try to allocate a partial slab from a specific node.
1273 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1278 * Racy check. If we mistakenly see no partial slabs then we
1279 * just allocate an empty slab. If we mistakenly try to get a
1280 * partial slab and there is none available then get_partials()
1283 if (!n
|| !n
->nr_partial
)
1286 spin_lock(&n
->list_lock
);
1287 list_for_each_entry(page
, &n
->partial
, lru
)
1288 if (lock_and_freeze_slab(n
, page
))
1292 spin_unlock(&n
->list_lock
);
1297 * Get a page from somewhere. Search in increasing NUMA distances.
1299 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1302 struct zonelist
*zonelist
;
1305 enum zone_type high_zoneidx
= gfp_zone(flags
);
1309 * The defrag ratio allows a configuration of the tradeoffs between
1310 * inter node defragmentation and node local allocations. A lower
1311 * defrag_ratio increases the tendency to do local allocations
1312 * instead of attempting to obtain partial slabs from other nodes.
1314 * If the defrag_ratio is set to 0 then kmalloc() always
1315 * returns node local objects. If the ratio is higher then kmalloc()
1316 * may return off node objects because partial slabs are obtained
1317 * from other nodes and filled up.
1319 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1320 * defrag_ratio = 1000) then every (well almost) allocation will
1321 * first attempt to defrag slab caches on other nodes. This means
1322 * scanning over all nodes to look for partial slabs which may be
1323 * expensive if we do it every time we are trying to find a slab
1324 * with available objects.
1326 if (!s
->remote_node_defrag_ratio
||
1327 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1330 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1331 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1332 struct kmem_cache_node
*n
;
1334 n
= get_node(s
, zone_to_nid(zone
));
1336 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1337 n
->nr_partial
> n
->min_partial
) {
1338 page
= get_partial_node(n
);
1348 * Get a partial page, lock it and return it.
1350 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1353 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1355 page
= get_partial_node(get_node(s
, searchnode
));
1356 if (page
|| (flags
& __GFP_THISNODE
))
1359 return get_any_partial(s
, flags
);
1363 * Move a page back to the lists.
1365 * Must be called with the slab lock held.
1367 * On exit the slab lock will have been dropped.
1369 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1371 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1372 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1374 __ClearPageSlubFrozen(page
);
1377 if (page
->freelist
) {
1378 add_partial(n
, page
, tail
);
1379 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1381 stat(c
, DEACTIVATE_FULL
);
1382 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1383 (s
->flags
& SLAB_STORE_USER
))
1388 stat(c
, DEACTIVATE_EMPTY
);
1389 if (n
->nr_partial
< n
->min_partial
) {
1391 * Adding an empty slab to the partial slabs in order
1392 * to avoid page allocator overhead. This slab needs
1393 * to come after the other slabs with objects in
1394 * so that the others get filled first. That way the
1395 * size of the partial list stays small.
1397 * kmem_cache_shrink can reclaim any empty slabs from
1400 add_partial(n
, page
, 1);
1404 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1405 discard_slab(s
, page
);
1411 * Remove the cpu slab
1413 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1415 struct page
*page
= c
->page
;
1419 stat(c
, DEACTIVATE_REMOTE_FREES
);
1421 * Merge cpu freelist into slab freelist. Typically we get here
1422 * because both freelists are empty. So this is unlikely
1425 while (unlikely(c
->freelist
)) {
1428 tail
= 0; /* Hot objects. Put the slab first */
1430 /* Retrieve object from cpu_freelist */
1431 object
= c
->freelist
;
1432 c
->freelist
= c
->freelist
[c
->offset
];
1434 /* And put onto the regular freelist */
1435 object
[c
->offset
] = page
->freelist
;
1436 page
->freelist
= object
;
1440 unfreeze_slab(s
, page
, tail
);
1443 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1445 stat(c
, CPUSLAB_FLUSH
);
1447 deactivate_slab(s
, c
);
1453 * Called from IPI handler with interrupts disabled.
1455 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1457 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1459 if (likely(c
&& c
->page
))
1463 static void flush_cpu_slab(void *d
)
1465 struct kmem_cache
*s
= d
;
1467 __flush_cpu_slab(s
, smp_processor_id());
1470 static void flush_all(struct kmem_cache
*s
)
1472 on_each_cpu(flush_cpu_slab
, s
, 1);
1476 * Check if the objects in a per cpu structure fit numa
1477 * locality expectations.
1479 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1482 if (node
!= -1 && c
->node
!= node
)
1489 * Slow path. The lockless freelist is empty or we need to perform
1492 * Interrupts are disabled.
1494 * Processing is still very fast if new objects have been freed to the
1495 * regular freelist. In that case we simply take over the regular freelist
1496 * as the lockless freelist and zap the regular freelist.
1498 * If that is not working then we fall back to the partial lists. We take the
1499 * first element of the freelist as the object to allocate now and move the
1500 * rest of the freelist to the lockless freelist.
1502 * And if we were unable to get a new slab from the partial slab lists then
1503 * we need to allocate a new slab. This is the slowest path since it involves
1504 * a call to the page allocator and the setup of a new slab.
1506 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1507 unsigned long addr
, struct kmem_cache_cpu
*c
)
1512 /* We handle __GFP_ZERO in the caller */
1513 gfpflags
&= ~__GFP_ZERO
;
1519 if (unlikely(!node_match(c
, node
)))
1522 stat(c
, ALLOC_REFILL
);
1525 object
= c
->page
->freelist
;
1526 if (unlikely(!object
))
1528 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1531 c
->freelist
= object
[c
->offset
];
1532 c
->page
->inuse
= c
->page
->objects
;
1533 c
->page
->freelist
= NULL
;
1534 c
->node
= page_to_nid(c
->page
);
1536 slab_unlock(c
->page
);
1537 stat(c
, ALLOC_SLOWPATH
);
1541 deactivate_slab(s
, c
);
1544 new = get_partial(s
, gfpflags
, node
);
1547 stat(c
, ALLOC_FROM_PARTIAL
);
1551 if (gfpflags
& __GFP_WAIT
)
1554 new = new_slab(s
, gfpflags
, node
);
1556 if (gfpflags
& __GFP_WAIT
)
1557 local_irq_disable();
1560 c
= get_cpu_slab(s
, smp_processor_id());
1561 stat(c
, ALLOC_SLAB
);
1565 __SetPageSlubFrozen(new);
1571 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1575 c
->page
->freelist
= object
[c
->offset
];
1581 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1582 * have the fastpath folded into their functions. So no function call
1583 * overhead for requests that can be satisfied on the fastpath.
1585 * The fastpath works by first checking if the lockless freelist can be used.
1586 * If not then __slab_alloc is called for slow processing.
1588 * Otherwise we can simply pick the next object from the lockless free list.
1590 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1591 gfp_t gfpflags
, int node
, unsigned long addr
)
1594 struct kmem_cache_cpu
*c
;
1595 unsigned long flags
;
1596 unsigned int objsize
;
1598 local_irq_save(flags
);
1599 c
= get_cpu_slab(s
, smp_processor_id());
1600 objsize
= c
->objsize
;
1601 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1603 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1606 object
= c
->freelist
;
1607 c
->freelist
= object
[c
->offset
];
1608 stat(c
, ALLOC_FASTPATH
);
1610 local_irq_restore(flags
);
1612 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1613 memset(object
, 0, objsize
);
1618 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1620 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1622 EXPORT_SYMBOL(kmem_cache_alloc
);
1625 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1627 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1629 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1633 * Slow patch handling. This may still be called frequently since objects
1634 * have a longer lifetime than the cpu slabs in most processing loads.
1636 * So we still attempt to reduce cache line usage. Just take the slab
1637 * lock and free the item. If there is no additional partial page
1638 * handling required then we can return immediately.
1640 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1641 void *x
, unsigned long addr
, unsigned int offset
)
1644 void **object
= (void *)x
;
1645 struct kmem_cache_cpu
*c
;
1647 c
= get_cpu_slab(s
, raw_smp_processor_id());
1648 stat(c
, FREE_SLOWPATH
);
1651 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1655 prior
= object
[offset
] = page
->freelist
;
1656 page
->freelist
= object
;
1659 if (unlikely(PageSlubFrozen(page
))) {
1660 stat(c
, FREE_FROZEN
);
1664 if (unlikely(!page
->inuse
))
1668 * Objects left in the slab. If it was not on the partial list before
1671 if (unlikely(!prior
)) {
1672 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1673 stat(c
, FREE_ADD_PARTIAL
);
1683 * Slab still on the partial list.
1685 remove_partial(s
, page
);
1686 stat(c
, FREE_REMOVE_PARTIAL
);
1690 discard_slab(s
, page
);
1694 if (!free_debug_processing(s
, page
, x
, addr
))
1700 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1701 * can perform fastpath freeing without additional function calls.
1703 * The fastpath is only possible if we are freeing to the current cpu slab
1704 * of this processor. This typically the case if we have just allocated
1707 * If fastpath is not possible then fall back to __slab_free where we deal
1708 * with all sorts of special processing.
1710 static __always_inline
void slab_free(struct kmem_cache
*s
,
1711 struct page
*page
, void *x
, unsigned long addr
)
1713 void **object
= (void *)x
;
1714 struct kmem_cache_cpu
*c
;
1715 unsigned long flags
;
1717 local_irq_save(flags
);
1718 c
= get_cpu_slab(s
, smp_processor_id());
1719 debug_check_no_locks_freed(object
, c
->objsize
);
1720 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1721 debug_check_no_obj_freed(object
, s
->objsize
);
1722 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1723 object
[c
->offset
] = c
->freelist
;
1724 c
->freelist
= object
;
1725 stat(c
, FREE_FASTPATH
);
1727 __slab_free(s
, page
, x
, addr
, c
->offset
);
1729 local_irq_restore(flags
);
1732 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1736 page
= virt_to_head_page(x
);
1738 slab_free(s
, page
, x
, _RET_IP_
);
1740 EXPORT_SYMBOL(kmem_cache_free
);
1742 /* Figure out on which slab page the object resides */
1743 static struct page
*get_object_page(const void *x
)
1745 struct page
*page
= virt_to_head_page(x
);
1747 if (!PageSlab(page
))
1754 * Object placement in a slab is made very easy because we always start at
1755 * offset 0. If we tune the size of the object to the alignment then we can
1756 * get the required alignment by putting one properly sized object after
1759 * Notice that the allocation order determines the sizes of the per cpu
1760 * caches. Each processor has always one slab available for allocations.
1761 * Increasing the allocation order reduces the number of times that slabs
1762 * must be moved on and off the partial lists and is therefore a factor in
1767 * Mininum / Maximum order of slab pages. This influences locking overhead
1768 * and slab fragmentation. A higher order reduces the number of partial slabs
1769 * and increases the number of allocations possible without having to
1770 * take the list_lock.
1772 static int slub_min_order
;
1773 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1774 static int slub_min_objects
;
1777 * Merge control. If this is set then no merging of slab caches will occur.
1778 * (Could be removed. This was introduced to pacify the merge skeptics.)
1780 static int slub_nomerge
;
1783 * Calculate the order of allocation given an slab object size.
1785 * The order of allocation has significant impact on performance and other
1786 * system components. Generally order 0 allocations should be preferred since
1787 * order 0 does not cause fragmentation in the page allocator. Larger objects
1788 * be problematic to put into order 0 slabs because there may be too much
1789 * unused space left. We go to a higher order if more than 1/16th of the slab
1792 * In order to reach satisfactory performance we must ensure that a minimum
1793 * number of objects is in one slab. Otherwise we may generate too much
1794 * activity on the partial lists which requires taking the list_lock. This is
1795 * less a concern for large slabs though which are rarely used.
1797 * slub_max_order specifies the order where we begin to stop considering the
1798 * number of objects in a slab as critical. If we reach slub_max_order then
1799 * we try to keep the page order as low as possible. So we accept more waste
1800 * of space in favor of a small page order.
1802 * Higher order allocations also allow the placement of more objects in a
1803 * slab and thereby reduce object handling overhead. If the user has
1804 * requested a higher mininum order then we start with that one instead of
1805 * the smallest order which will fit the object.
1807 static inline int slab_order(int size
, int min_objects
,
1808 int max_order
, int fract_leftover
)
1812 int min_order
= slub_min_order
;
1814 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1815 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1817 for (order
= max(min_order
,
1818 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1819 order
<= max_order
; order
++) {
1821 unsigned long slab_size
= PAGE_SIZE
<< order
;
1823 if (slab_size
< min_objects
* size
)
1826 rem
= slab_size
% size
;
1828 if (rem
<= slab_size
/ fract_leftover
)
1836 static inline int calculate_order(int size
)
1843 * Attempt to find best configuration for a slab. This
1844 * works by first attempting to generate a layout with
1845 * the best configuration and backing off gradually.
1847 * First we reduce the acceptable waste in a slab. Then
1848 * we reduce the minimum objects required in a slab.
1850 min_objects
= slub_min_objects
;
1852 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1853 while (min_objects
> 1) {
1855 while (fraction
>= 4) {
1856 order
= slab_order(size
, min_objects
,
1857 slub_max_order
, fraction
);
1858 if (order
<= slub_max_order
)
1866 * We were unable to place multiple objects in a slab. Now
1867 * lets see if we can place a single object there.
1869 order
= slab_order(size
, 1, slub_max_order
, 1);
1870 if (order
<= slub_max_order
)
1874 * Doh this slab cannot be placed using slub_max_order.
1876 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1877 if (order
<= MAX_ORDER
)
1883 * Figure out what the alignment of the objects will be.
1885 static unsigned long calculate_alignment(unsigned long flags
,
1886 unsigned long align
, unsigned long size
)
1889 * If the user wants hardware cache aligned objects then follow that
1890 * suggestion if the object is sufficiently large.
1892 * The hardware cache alignment cannot override the specified
1893 * alignment though. If that is greater then use it.
1895 if (flags
& SLAB_HWCACHE_ALIGN
) {
1896 unsigned long ralign
= cache_line_size();
1897 while (size
<= ralign
/ 2)
1899 align
= max(align
, ralign
);
1902 if (align
< ARCH_SLAB_MINALIGN
)
1903 align
= ARCH_SLAB_MINALIGN
;
1905 return ALIGN(align
, sizeof(void *));
1908 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1909 struct kmem_cache_cpu
*c
)
1914 c
->offset
= s
->offset
/ sizeof(void *);
1915 c
->objsize
= s
->objsize
;
1916 #ifdef CONFIG_SLUB_STATS
1917 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1922 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
1927 * The larger the object size is, the more pages we want on the partial
1928 * list to avoid pounding the page allocator excessively.
1930 n
->min_partial
= ilog2(s
->size
);
1931 if (n
->min_partial
< MIN_PARTIAL
)
1932 n
->min_partial
= MIN_PARTIAL
;
1933 else if (n
->min_partial
> MAX_PARTIAL
)
1934 n
->min_partial
= MAX_PARTIAL
;
1936 spin_lock_init(&n
->list_lock
);
1937 INIT_LIST_HEAD(&n
->partial
);
1938 #ifdef CONFIG_SLUB_DEBUG
1939 atomic_long_set(&n
->nr_slabs
, 0);
1940 atomic_long_set(&n
->total_objects
, 0);
1941 INIT_LIST_HEAD(&n
->full
);
1947 * Per cpu array for per cpu structures.
1949 * The per cpu array places all kmem_cache_cpu structures from one processor
1950 * close together meaning that it becomes possible that multiple per cpu
1951 * structures are contained in one cacheline. This may be particularly
1952 * beneficial for the kmalloc caches.
1954 * A desktop system typically has around 60-80 slabs. With 100 here we are
1955 * likely able to get per cpu structures for all caches from the array defined
1956 * here. We must be able to cover all kmalloc caches during bootstrap.
1958 * If the per cpu array is exhausted then fall back to kmalloc
1959 * of individual cachelines. No sharing is possible then.
1961 #define NR_KMEM_CACHE_CPU 100
1963 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
1964 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
1966 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
1967 static cpumask_t kmem_cach_cpu_free_init_once
= CPU_MASK_NONE
;
1969 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
1970 int cpu
, gfp_t flags
)
1972 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
1975 per_cpu(kmem_cache_cpu_free
, cpu
) =
1976 (void *)c
->freelist
;
1978 /* Table overflow: So allocate ourselves */
1980 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
1981 flags
, cpu_to_node(cpu
));
1986 init_kmem_cache_cpu(s
, c
);
1990 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
1992 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
1993 c
> per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
1997 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
1998 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2001 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2005 for_each_online_cpu(cpu
) {
2006 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2009 s
->cpu_slab
[cpu
] = NULL
;
2010 free_kmem_cache_cpu(c
, cpu
);
2015 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2019 for_each_online_cpu(cpu
) {
2020 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2025 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2027 free_kmem_cache_cpus(s
);
2030 s
->cpu_slab
[cpu
] = c
;
2036 * Initialize the per cpu array.
2038 static void init_alloc_cpu_cpu(int cpu
)
2042 if (cpu_isset(cpu
, kmem_cach_cpu_free_init_once
))
2045 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2046 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2048 cpu_set(cpu
, kmem_cach_cpu_free_init_once
);
2051 static void __init
init_alloc_cpu(void)
2055 for_each_online_cpu(cpu
)
2056 init_alloc_cpu_cpu(cpu
);
2060 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2061 static inline void init_alloc_cpu(void) {}
2063 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2065 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2072 * No kmalloc_node yet so do it by hand. We know that this is the first
2073 * slab on the node for this slabcache. There are no concurrent accesses
2076 * Note that this function only works on the kmalloc_node_cache
2077 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2078 * memory on a fresh node that has no slab structures yet.
2080 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2083 struct kmem_cache_node
*n
;
2084 unsigned long flags
;
2086 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2088 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2091 if (page_to_nid(page
) != node
) {
2092 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2094 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2095 "in order to be able to continue\n");
2100 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2102 kmalloc_caches
->node
[node
] = n
;
2103 #ifdef CONFIG_SLUB_DEBUG
2104 init_object(kmalloc_caches
, n
, 1);
2105 init_tracking(kmalloc_caches
, n
);
2107 init_kmem_cache_node(n
, kmalloc_caches
);
2108 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2111 * lockdep requires consistent irq usage for each lock
2112 * so even though there cannot be a race this early in
2113 * the boot sequence, we still disable irqs.
2115 local_irq_save(flags
);
2116 add_partial(n
, page
, 0);
2117 local_irq_restore(flags
);
2120 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2124 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2125 struct kmem_cache_node
*n
= s
->node
[node
];
2126 if (n
&& n
!= &s
->local_node
)
2127 kmem_cache_free(kmalloc_caches
, n
);
2128 s
->node
[node
] = NULL
;
2132 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2137 if (slab_state
>= UP
)
2138 local_node
= page_to_nid(virt_to_page(s
));
2142 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2143 struct kmem_cache_node
*n
;
2145 if (local_node
== node
)
2148 if (slab_state
== DOWN
) {
2149 early_kmem_cache_node_alloc(gfpflags
, node
);
2152 n
= kmem_cache_alloc_node(kmalloc_caches
,
2156 free_kmem_cache_nodes(s
);
2162 init_kmem_cache_node(n
, s
);
2167 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2171 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2173 init_kmem_cache_node(&s
->local_node
, s
);
2179 * calculate_sizes() determines the order and the distribution of data within
2182 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2184 unsigned long flags
= s
->flags
;
2185 unsigned long size
= s
->objsize
;
2186 unsigned long align
= s
->align
;
2190 * Round up object size to the next word boundary. We can only
2191 * place the free pointer at word boundaries and this determines
2192 * the possible location of the free pointer.
2194 size
= ALIGN(size
, sizeof(void *));
2196 #ifdef CONFIG_SLUB_DEBUG
2198 * Determine if we can poison the object itself. If the user of
2199 * the slab may touch the object after free or before allocation
2200 * then we should never poison the object itself.
2202 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2204 s
->flags
|= __OBJECT_POISON
;
2206 s
->flags
&= ~__OBJECT_POISON
;
2210 * If we are Redzoning then check if there is some space between the
2211 * end of the object and the free pointer. If not then add an
2212 * additional word to have some bytes to store Redzone information.
2214 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2215 size
+= sizeof(void *);
2219 * With that we have determined the number of bytes in actual use
2220 * by the object. This is the potential offset to the free pointer.
2224 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2227 * Relocate free pointer after the object if it is not
2228 * permitted to overwrite the first word of the object on
2231 * This is the case if we do RCU, have a constructor or
2232 * destructor or are poisoning the objects.
2235 size
+= sizeof(void *);
2238 #ifdef CONFIG_SLUB_DEBUG
2239 if (flags
& SLAB_STORE_USER
)
2241 * Need to store information about allocs and frees after
2244 size
+= 2 * sizeof(struct track
);
2246 if (flags
& SLAB_RED_ZONE
)
2248 * Add some empty padding so that we can catch
2249 * overwrites from earlier objects rather than let
2250 * tracking information or the free pointer be
2251 * corrupted if an user writes before the start
2254 size
+= sizeof(void *);
2258 * Determine the alignment based on various parameters that the
2259 * user specified and the dynamic determination of cache line size
2262 align
= calculate_alignment(flags
, align
, s
->objsize
);
2265 * SLUB stores one object immediately after another beginning from
2266 * offset 0. In order to align the objects we have to simply size
2267 * each object to conform to the alignment.
2269 size
= ALIGN(size
, align
);
2271 if (forced_order
>= 0)
2272 order
= forced_order
;
2274 order
= calculate_order(size
);
2281 s
->allocflags
|= __GFP_COMP
;
2283 if (s
->flags
& SLAB_CACHE_DMA
)
2284 s
->allocflags
|= SLUB_DMA
;
2286 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2287 s
->allocflags
|= __GFP_RECLAIMABLE
;
2290 * Determine the number of objects per slab
2292 s
->oo
= oo_make(order
, size
);
2293 s
->min
= oo_make(get_order(size
), size
);
2294 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2297 return !!oo_objects(s
->oo
);
2301 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2302 const char *name
, size_t size
,
2303 size_t align
, unsigned long flags
,
2304 void (*ctor
)(void *))
2306 memset(s
, 0, kmem_size
);
2311 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2313 if (!calculate_sizes(s
, -1))
2318 s
->remote_node_defrag_ratio
= 1000;
2320 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2323 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2325 free_kmem_cache_nodes(s
);
2327 if (flags
& SLAB_PANIC
)
2328 panic("Cannot create slab %s size=%lu realsize=%u "
2329 "order=%u offset=%u flags=%lx\n",
2330 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2336 * Check if a given pointer is valid
2338 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2342 page
= get_object_page(object
);
2344 if (!page
|| s
!= page
->slab
)
2345 /* No slab or wrong slab */
2348 if (!check_valid_pointer(s
, page
, object
))
2352 * We could also check if the object is on the slabs freelist.
2353 * But this would be too expensive and it seems that the main
2354 * purpose of kmem_ptr_valid() is to check if the object belongs
2355 * to a certain slab.
2359 EXPORT_SYMBOL(kmem_ptr_validate
);
2362 * Determine the size of a slab object
2364 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2368 EXPORT_SYMBOL(kmem_cache_size
);
2370 const char *kmem_cache_name(struct kmem_cache
*s
)
2374 EXPORT_SYMBOL(kmem_cache_name
);
2376 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2379 #ifdef CONFIG_SLUB_DEBUG
2380 void *addr
= page_address(page
);
2382 DECLARE_BITMAP(map
, page
->objects
);
2384 bitmap_zero(map
, page
->objects
);
2385 slab_err(s
, page
, "%s", text
);
2387 for_each_free_object(p
, s
, page
->freelist
)
2388 set_bit(slab_index(p
, s
, addr
), map
);
2390 for_each_object(p
, s
, addr
, page
->objects
) {
2392 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2393 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2395 print_tracking(s
, p
);
2403 * Attempt to free all partial slabs on a node.
2405 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2407 unsigned long flags
;
2408 struct page
*page
, *h
;
2410 spin_lock_irqsave(&n
->list_lock
, flags
);
2411 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2413 list_del(&page
->lru
);
2414 discard_slab(s
, page
);
2417 list_slab_objects(s
, page
,
2418 "Objects remaining on kmem_cache_close()");
2421 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2425 * Release all resources used by a slab cache.
2427 static inline int kmem_cache_close(struct kmem_cache
*s
)
2433 /* Attempt to free all objects */
2434 free_kmem_cache_cpus(s
);
2435 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2436 struct kmem_cache_node
*n
= get_node(s
, node
);
2439 if (n
->nr_partial
|| slabs_node(s
, node
))
2442 free_kmem_cache_nodes(s
);
2447 * Close a cache and release the kmem_cache structure
2448 * (must be used for caches created using kmem_cache_create)
2450 void kmem_cache_destroy(struct kmem_cache
*s
)
2452 down_write(&slub_lock
);
2456 up_write(&slub_lock
);
2457 if (kmem_cache_close(s
)) {
2458 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2459 "still has objects.\n", s
->name
, __func__
);
2462 sysfs_slab_remove(s
);
2464 up_write(&slub_lock
);
2466 EXPORT_SYMBOL(kmem_cache_destroy
);
2468 /********************************************************************
2470 *******************************************************************/
2472 struct kmem_cache kmalloc_caches
[PAGE_SHIFT
+ 1] __cacheline_aligned
;
2473 EXPORT_SYMBOL(kmalloc_caches
);
2475 static int __init
setup_slub_min_order(char *str
)
2477 get_option(&str
, &slub_min_order
);
2482 __setup("slub_min_order=", setup_slub_min_order
);
2484 static int __init
setup_slub_max_order(char *str
)
2486 get_option(&str
, &slub_max_order
);
2491 __setup("slub_max_order=", setup_slub_max_order
);
2493 static int __init
setup_slub_min_objects(char *str
)
2495 get_option(&str
, &slub_min_objects
);
2500 __setup("slub_min_objects=", setup_slub_min_objects
);
2502 static int __init
setup_slub_nomerge(char *str
)
2508 __setup("slub_nomerge", setup_slub_nomerge
);
2510 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2511 const char *name
, int size
, gfp_t gfp_flags
)
2513 unsigned int flags
= 0;
2515 if (gfp_flags
& SLUB_DMA
)
2516 flags
= SLAB_CACHE_DMA
;
2518 down_write(&slub_lock
);
2519 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2523 list_add(&s
->list
, &slab_caches
);
2524 up_write(&slub_lock
);
2525 if (sysfs_slab_add(s
))
2530 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2533 #ifdef CONFIG_ZONE_DMA
2534 static struct kmem_cache
*kmalloc_caches_dma
[PAGE_SHIFT
+ 1];
2536 static void sysfs_add_func(struct work_struct
*w
)
2538 struct kmem_cache
*s
;
2540 down_write(&slub_lock
);
2541 list_for_each_entry(s
, &slab_caches
, list
) {
2542 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2543 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2547 up_write(&slub_lock
);
2550 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2552 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2554 struct kmem_cache
*s
;
2558 s
= kmalloc_caches_dma
[index
];
2562 /* Dynamically create dma cache */
2563 if (flags
& __GFP_WAIT
)
2564 down_write(&slub_lock
);
2566 if (!down_write_trylock(&slub_lock
))
2570 if (kmalloc_caches_dma
[index
])
2573 realsize
= kmalloc_caches
[index
].objsize
;
2574 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2575 (unsigned int)realsize
);
2576 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2578 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2579 realsize
, ARCH_KMALLOC_MINALIGN
,
2580 SLAB_CACHE_DMA
|__SYSFS_ADD_DEFERRED
, NULL
)) {
2586 list_add(&s
->list
, &slab_caches
);
2587 kmalloc_caches_dma
[index
] = s
;
2589 schedule_work(&sysfs_add_work
);
2592 up_write(&slub_lock
);
2594 return kmalloc_caches_dma
[index
];
2599 * Conversion table for small slabs sizes / 8 to the index in the
2600 * kmalloc array. This is necessary for slabs < 192 since we have non power
2601 * of two cache sizes there. The size of larger slabs can be determined using
2604 static s8 size_index
[24] = {
2631 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2637 return ZERO_SIZE_PTR
;
2639 index
= size_index
[(size
- 1) / 8];
2641 index
= fls(size
- 1);
2643 #ifdef CONFIG_ZONE_DMA
2644 if (unlikely((flags
& SLUB_DMA
)))
2645 return dma_kmalloc_cache(index
, flags
);
2648 return &kmalloc_caches
[index
];
2651 void *__kmalloc(size_t size
, gfp_t flags
)
2653 struct kmem_cache
*s
;
2655 if (unlikely(size
> PAGE_SIZE
))
2656 return kmalloc_large(size
, flags
);
2658 s
= get_slab(size
, flags
);
2660 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2663 return slab_alloc(s
, flags
, -1, _RET_IP_
);
2665 EXPORT_SYMBOL(__kmalloc
);
2667 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2669 struct page
*page
= alloc_pages_node(node
, flags
| __GFP_COMP
,
2673 return page_address(page
);
2679 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2681 struct kmem_cache
*s
;
2683 if (unlikely(size
> PAGE_SIZE
))
2684 return kmalloc_large_node(size
, flags
, node
);
2686 s
= get_slab(size
, flags
);
2688 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2691 return slab_alloc(s
, flags
, node
, _RET_IP_
);
2693 EXPORT_SYMBOL(__kmalloc_node
);
2696 size_t ksize(const void *object
)
2699 struct kmem_cache
*s
;
2701 if (unlikely(object
== ZERO_SIZE_PTR
))
2704 page
= virt_to_head_page(object
);
2706 if (unlikely(!PageSlab(page
))) {
2707 WARN_ON(!PageCompound(page
));
2708 return PAGE_SIZE
<< compound_order(page
);
2712 #ifdef CONFIG_SLUB_DEBUG
2714 * Debugging requires use of the padding between object
2715 * and whatever may come after it.
2717 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2722 * If we have the need to store the freelist pointer
2723 * back there or track user information then we can
2724 * only use the space before that information.
2726 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2729 * Else we can use all the padding etc for the allocation
2734 void kfree(const void *x
)
2737 void *object
= (void *)x
;
2739 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2742 page
= virt_to_head_page(x
);
2743 if (unlikely(!PageSlab(page
))) {
2744 BUG_ON(!PageCompound(page
));
2748 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2750 EXPORT_SYMBOL(kfree
);
2753 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2754 * the remaining slabs by the number of items in use. The slabs with the
2755 * most items in use come first. New allocations will then fill those up
2756 * and thus they can be removed from the partial lists.
2758 * The slabs with the least items are placed last. This results in them
2759 * being allocated from last increasing the chance that the last objects
2760 * are freed in them.
2762 int kmem_cache_shrink(struct kmem_cache
*s
)
2766 struct kmem_cache_node
*n
;
2769 int objects
= oo_objects(s
->max
);
2770 struct list_head
*slabs_by_inuse
=
2771 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2772 unsigned long flags
;
2774 if (!slabs_by_inuse
)
2778 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2779 n
= get_node(s
, node
);
2784 for (i
= 0; i
< objects
; i
++)
2785 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2787 spin_lock_irqsave(&n
->list_lock
, flags
);
2790 * Build lists indexed by the items in use in each slab.
2792 * Note that concurrent frees may occur while we hold the
2793 * list_lock. page->inuse here is the upper limit.
2795 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2796 if (!page
->inuse
&& slab_trylock(page
)) {
2798 * Must hold slab lock here because slab_free
2799 * may have freed the last object and be
2800 * waiting to release the slab.
2802 list_del(&page
->lru
);
2805 discard_slab(s
, page
);
2807 list_move(&page
->lru
,
2808 slabs_by_inuse
+ page
->inuse
);
2813 * Rebuild the partial list with the slabs filled up most
2814 * first and the least used slabs at the end.
2816 for (i
= objects
- 1; i
>= 0; i
--)
2817 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2819 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2822 kfree(slabs_by_inuse
);
2825 EXPORT_SYMBOL(kmem_cache_shrink
);
2827 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2828 static int slab_mem_going_offline_callback(void *arg
)
2830 struct kmem_cache
*s
;
2832 down_read(&slub_lock
);
2833 list_for_each_entry(s
, &slab_caches
, list
)
2834 kmem_cache_shrink(s
);
2835 up_read(&slub_lock
);
2840 static void slab_mem_offline_callback(void *arg
)
2842 struct kmem_cache_node
*n
;
2843 struct kmem_cache
*s
;
2844 struct memory_notify
*marg
= arg
;
2847 offline_node
= marg
->status_change_nid
;
2850 * If the node still has available memory. we need kmem_cache_node
2853 if (offline_node
< 0)
2856 down_read(&slub_lock
);
2857 list_for_each_entry(s
, &slab_caches
, list
) {
2858 n
= get_node(s
, offline_node
);
2861 * if n->nr_slabs > 0, slabs still exist on the node
2862 * that is going down. We were unable to free them,
2863 * and offline_pages() function shoudn't call this
2864 * callback. So, we must fail.
2866 BUG_ON(slabs_node(s
, offline_node
));
2868 s
->node
[offline_node
] = NULL
;
2869 kmem_cache_free(kmalloc_caches
, n
);
2872 up_read(&slub_lock
);
2875 static int slab_mem_going_online_callback(void *arg
)
2877 struct kmem_cache_node
*n
;
2878 struct kmem_cache
*s
;
2879 struct memory_notify
*marg
= arg
;
2880 int nid
= marg
->status_change_nid
;
2884 * If the node's memory is already available, then kmem_cache_node is
2885 * already created. Nothing to do.
2891 * We are bringing a node online. No memory is available yet. We must
2892 * allocate a kmem_cache_node structure in order to bring the node
2895 down_read(&slub_lock
);
2896 list_for_each_entry(s
, &slab_caches
, list
) {
2898 * XXX: kmem_cache_alloc_node will fallback to other nodes
2899 * since memory is not yet available from the node that
2902 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2907 init_kmem_cache_node(n
, s
);
2911 up_read(&slub_lock
);
2915 static int slab_memory_callback(struct notifier_block
*self
,
2916 unsigned long action
, void *arg
)
2921 case MEM_GOING_ONLINE
:
2922 ret
= slab_mem_going_online_callback(arg
);
2924 case MEM_GOING_OFFLINE
:
2925 ret
= slab_mem_going_offline_callback(arg
);
2928 case MEM_CANCEL_ONLINE
:
2929 slab_mem_offline_callback(arg
);
2932 case MEM_CANCEL_OFFLINE
:
2936 ret
= notifier_from_errno(ret
);
2940 #endif /* CONFIG_MEMORY_HOTPLUG */
2942 /********************************************************************
2943 * Basic setup of slabs
2944 *******************************************************************/
2946 void __init
kmem_cache_init(void)
2955 * Must first have the slab cache available for the allocations of the
2956 * struct kmem_cache_node's. There is special bootstrap code in
2957 * kmem_cache_open for slab_state == DOWN.
2959 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
2960 sizeof(struct kmem_cache_node
), GFP_KERNEL
);
2961 kmalloc_caches
[0].refcount
= -1;
2964 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
2967 /* Able to allocate the per node structures */
2968 slab_state
= PARTIAL
;
2970 /* Caches that are not of the two-to-the-power-of size */
2971 if (KMALLOC_MIN_SIZE
<= 64) {
2972 create_kmalloc_cache(&kmalloc_caches
[1],
2973 "kmalloc-96", 96, GFP_KERNEL
);
2975 create_kmalloc_cache(&kmalloc_caches
[2],
2976 "kmalloc-192", 192, GFP_KERNEL
);
2980 for (i
= KMALLOC_SHIFT_LOW
; i
<= PAGE_SHIFT
; i
++) {
2981 create_kmalloc_cache(&kmalloc_caches
[i
],
2982 "kmalloc", 1 << i
, GFP_KERNEL
);
2988 * Patch up the size_index table if we have strange large alignment
2989 * requirements for the kmalloc array. This is only the case for
2990 * MIPS it seems. The standard arches will not generate any code here.
2992 * Largest permitted alignment is 256 bytes due to the way we
2993 * handle the index determination for the smaller caches.
2995 * Make sure that nothing crazy happens if someone starts tinkering
2996 * around with ARCH_KMALLOC_MINALIGN
2998 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
2999 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3001 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3002 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3004 if (KMALLOC_MIN_SIZE
== 128) {
3006 * The 192 byte sized cache is not used if the alignment
3007 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3010 for (i
= 128 + 8; i
<= 192; i
+= 8)
3011 size_index
[(i
- 1) / 8] = 8;
3016 /* Provide the correct kmalloc names now that the caches are up */
3017 for (i
= KMALLOC_SHIFT_LOW
; i
<= PAGE_SHIFT
; i
++)
3018 kmalloc_caches
[i
]. name
=
3019 kasprintf(GFP_KERNEL
, "kmalloc-%d", 1 << i
);
3022 register_cpu_notifier(&slab_notifier
);
3023 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3024 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3026 kmem_size
= sizeof(struct kmem_cache
);
3030 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3031 " CPUs=%d, Nodes=%d\n",
3032 caches
, cache_line_size(),
3033 slub_min_order
, slub_max_order
, slub_min_objects
,
3034 nr_cpu_ids
, nr_node_ids
);
3038 * Find a mergeable slab cache
3040 static int slab_unmergeable(struct kmem_cache
*s
)
3042 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3049 * We may have set a slab to be unmergeable during bootstrap.
3051 if (s
->refcount
< 0)
3057 static struct kmem_cache
*find_mergeable(size_t size
,
3058 size_t align
, unsigned long flags
, const char *name
,
3059 void (*ctor
)(void *))
3061 struct kmem_cache
*s
;
3063 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3069 size
= ALIGN(size
, sizeof(void *));
3070 align
= calculate_alignment(flags
, align
, size
);
3071 size
= ALIGN(size
, align
);
3072 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3074 list_for_each_entry(s
, &slab_caches
, list
) {
3075 if (slab_unmergeable(s
))
3081 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3084 * Check if alignment is compatible.
3085 * Courtesy of Adrian Drzewiecki
3087 if ((s
->size
& ~(align
- 1)) != s
->size
)
3090 if (s
->size
- size
>= sizeof(void *))
3098 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3099 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3101 struct kmem_cache
*s
;
3103 down_write(&slub_lock
);
3104 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3110 * Adjust the object sizes so that we clear
3111 * the complete object on kzalloc.
3113 s
->objsize
= max(s
->objsize
, (int)size
);
3116 * And then we need to update the object size in the
3117 * per cpu structures
3119 for_each_online_cpu(cpu
)
3120 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3122 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3123 up_write(&slub_lock
);
3125 if (sysfs_slab_alias(s
, name
))
3130 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3132 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3133 size
, align
, flags
, ctor
)) {
3134 list_add(&s
->list
, &slab_caches
);
3135 up_write(&slub_lock
);
3136 if (sysfs_slab_add(s
))
3142 up_write(&slub_lock
);
3145 if (flags
& SLAB_PANIC
)
3146 panic("Cannot create slabcache %s\n", name
);
3151 EXPORT_SYMBOL(kmem_cache_create
);
3155 * Use the cpu notifier to insure that the cpu slabs are flushed when
3158 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3159 unsigned long action
, void *hcpu
)
3161 long cpu
= (long)hcpu
;
3162 struct kmem_cache
*s
;
3163 unsigned long flags
;
3166 case CPU_UP_PREPARE
:
3167 case CPU_UP_PREPARE_FROZEN
:
3168 init_alloc_cpu_cpu(cpu
);
3169 down_read(&slub_lock
);
3170 list_for_each_entry(s
, &slab_caches
, list
)
3171 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3173 up_read(&slub_lock
);
3176 case CPU_UP_CANCELED
:
3177 case CPU_UP_CANCELED_FROZEN
:
3179 case CPU_DEAD_FROZEN
:
3180 down_read(&slub_lock
);
3181 list_for_each_entry(s
, &slab_caches
, list
) {
3182 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3184 local_irq_save(flags
);
3185 __flush_cpu_slab(s
, cpu
);
3186 local_irq_restore(flags
);
3187 free_kmem_cache_cpu(c
, cpu
);
3188 s
->cpu_slab
[cpu
] = NULL
;
3190 up_read(&slub_lock
);
3198 static struct notifier_block __cpuinitdata slab_notifier
= {
3199 .notifier_call
= slab_cpuup_callback
3204 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3206 struct kmem_cache
*s
;
3208 if (unlikely(size
> PAGE_SIZE
))
3209 return kmalloc_large(size
, gfpflags
);
3211 s
= get_slab(size
, gfpflags
);
3213 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3216 return slab_alloc(s
, gfpflags
, -1, caller
);
3219 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3220 int node
, unsigned long caller
)
3222 struct kmem_cache
*s
;
3224 if (unlikely(size
> PAGE_SIZE
))
3225 return kmalloc_large_node(size
, gfpflags
, node
);
3227 s
= get_slab(size
, gfpflags
);
3229 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3232 return slab_alloc(s
, gfpflags
, node
, caller
);
3235 #ifdef CONFIG_SLUB_DEBUG
3236 static unsigned long count_partial(struct kmem_cache_node
*n
,
3237 int (*get_count
)(struct page
*))
3239 unsigned long flags
;
3240 unsigned long x
= 0;
3243 spin_lock_irqsave(&n
->list_lock
, flags
);
3244 list_for_each_entry(page
, &n
->partial
, lru
)
3245 x
+= get_count(page
);
3246 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3250 static int count_inuse(struct page
*page
)
3255 static int count_total(struct page
*page
)
3257 return page
->objects
;
3260 static int count_free(struct page
*page
)
3262 return page
->objects
- page
->inuse
;
3265 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3269 void *addr
= page_address(page
);
3271 if (!check_slab(s
, page
) ||
3272 !on_freelist(s
, page
, NULL
))
3275 /* Now we know that a valid freelist exists */
3276 bitmap_zero(map
, page
->objects
);
3278 for_each_free_object(p
, s
, page
->freelist
) {
3279 set_bit(slab_index(p
, s
, addr
), map
);
3280 if (!check_object(s
, page
, p
, 0))
3284 for_each_object(p
, s
, addr
, page
->objects
)
3285 if (!test_bit(slab_index(p
, s
, addr
), map
))
3286 if (!check_object(s
, page
, p
, 1))
3291 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3294 if (slab_trylock(page
)) {
3295 validate_slab(s
, page
, map
);
3298 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3301 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3302 if (!PageSlubDebug(page
))
3303 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3304 "on slab 0x%p\n", s
->name
, page
);
3306 if (PageSlubDebug(page
))
3307 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3308 "slab 0x%p\n", s
->name
, page
);
3312 static int validate_slab_node(struct kmem_cache
*s
,
3313 struct kmem_cache_node
*n
, unsigned long *map
)
3315 unsigned long count
= 0;
3317 unsigned long flags
;
3319 spin_lock_irqsave(&n
->list_lock
, flags
);
3321 list_for_each_entry(page
, &n
->partial
, lru
) {
3322 validate_slab_slab(s
, page
, map
);
3325 if (count
!= n
->nr_partial
)
3326 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3327 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3329 if (!(s
->flags
& SLAB_STORE_USER
))
3332 list_for_each_entry(page
, &n
->full
, lru
) {
3333 validate_slab_slab(s
, page
, map
);
3336 if (count
!= atomic_long_read(&n
->nr_slabs
))
3337 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3338 "counter=%ld\n", s
->name
, count
,
3339 atomic_long_read(&n
->nr_slabs
));
3342 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3346 static long validate_slab_cache(struct kmem_cache
*s
)
3349 unsigned long count
= 0;
3350 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3351 sizeof(unsigned long), GFP_KERNEL
);
3357 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3358 struct kmem_cache_node
*n
= get_node(s
, node
);
3360 count
+= validate_slab_node(s
, n
, map
);
3366 #ifdef SLUB_RESILIENCY_TEST
3367 static void resiliency_test(void)
3371 printk(KERN_ERR
"SLUB resiliency testing\n");
3372 printk(KERN_ERR
"-----------------------\n");
3373 printk(KERN_ERR
"A. Corruption after allocation\n");
3375 p
= kzalloc(16, GFP_KERNEL
);
3377 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3378 " 0x12->0x%p\n\n", p
+ 16);
3380 validate_slab_cache(kmalloc_caches
+ 4);
3382 /* Hmmm... The next two are dangerous */
3383 p
= kzalloc(32, GFP_KERNEL
);
3384 p
[32 + sizeof(void *)] = 0x34;
3385 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3386 " 0x34 -> -0x%p\n", p
);
3388 "If allocated object is overwritten then not detectable\n\n");
3390 validate_slab_cache(kmalloc_caches
+ 5);
3391 p
= kzalloc(64, GFP_KERNEL
);
3392 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3394 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3397 "If allocated object is overwritten then not detectable\n\n");
3398 validate_slab_cache(kmalloc_caches
+ 6);
3400 printk(KERN_ERR
"\nB. Corruption after free\n");
3401 p
= kzalloc(128, GFP_KERNEL
);
3404 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3405 validate_slab_cache(kmalloc_caches
+ 7);
3407 p
= kzalloc(256, GFP_KERNEL
);
3410 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3412 validate_slab_cache(kmalloc_caches
+ 8);
3414 p
= kzalloc(512, GFP_KERNEL
);
3417 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3418 validate_slab_cache(kmalloc_caches
+ 9);
3421 static void resiliency_test(void) {};
3425 * Generate lists of code addresses where slabcache objects are allocated
3430 unsigned long count
;
3443 unsigned long count
;
3444 struct location
*loc
;
3447 static void free_loc_track(struct loc_track
*t
)
3450 free_pages((unsigned long)t
->loc
,
3451 get_order(sizeof(struct location
) * t
->max
));
3454 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3459 order
= get_order(sizeof(struct location
) * max
);
3461 l
= (void *)__get_free_pages(flags
, order
);
3466 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3474 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3475 const struct track
*track
)
3477 long start
, end
, pos
;
3479 unsigned long caddr
;
3480 unsigned long age
= jiffies
- track
->when
;
3486 pos
= start
+ (end
- start
+ 1) / 2;
3489 * There is nothing at "end". If we end up there
3490 * we need to add something to before end.
3495 caddr
= t
->loc
[pos
].addr
;
3496 if (track
->addr
== caddr
) {
3502 if (age
< l
->min_time
)
3504 if (age
> l
->max_time
)
3507 if (track
->pid
< l
->min_pid
)
3508 l
->min_pid
= track
->pid
;
3509 if (track
->pid
> l
->max_pid
)
3510 l
->max_pid
= track
->pid
;
3512 cpu_set(track
->cpu
, l
->cpus
);
3514 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3518 if (track
->addr
< caddr
)
3525 * Not found. Insert new tracking element.
3527 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3533 (t
->count
- pos
) * sizeof(struct location
));
3536 l
->addr
= track
->addr
;
3540 l
->min_pid
= track
->pid
;
3541 l
->max_pid
= track
->pid
;
3542 cpus_clear(l
->cpus
);
3543 cpu_set(track
->cpu
, l
->cpus
);
3544 nodes_clear(l
->nodes
);
3545 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3549 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3550 struct page
*page
, enum track_item alloc
)
3552 void *addr
= page_address(page
);
3553 DECLARE_BITMAP(map
, page
->objects
);
3556 bitmap_zero(map
, page
->objects
);
3557 for_each_free_object(p
, s
, page
->freelist
)
3558 set_bit(slab_index(p
, s
, addr
), map
);
3560 for_each_object(p
, s
, addr
, page
->objects
)
3561 if (!test_bit(slab_index(p
, s
, addr
), map
))
3562 add_location(t
, s
, get_track(s
, p
, alloc
));
3565 static int list_locations(struct kmem_cache
*s
, char *buf
,
3566 enum track_item alloc
)
3570 struct loc_track t
= { 0, 0, NULL
};
3573 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3575 return sprintf(buf
, "Out of memory\n");
3577 /* Push back cpu slabs */
3580 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3581 struct kmem_cache_node
*n
= get_node(s
, node
);
3582 unsigned long flags
;
3585 if (!atomic_long_read(&n
->nr_slabs
))
3588 spin_lock_irqsave(&n
->list_lock
, flags
);
3589 list_for_each_entry(page
, &n
->partial
, lru
)
3590 process_slab(&t
, s
, page
, alloc
);
3591 list_for_each_entry(page
, &n
->full
, lru
)
3592 process_slab(&t
, s
, page
, alloc
);
3593 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3596 for (i
= 0; i
< t
.count
; i
++) {
3597 struct location
*l
= &t
.loc
[i
];
3599 if (len
> PAGE_SIZE
- 100)
3601 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3604 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3606 len
+= sprintf(buf
+ len
, "<not-available>");
3608 if (l
->sum_time
!= l
->min_time
) {
3609 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3611 (long)div_u64(l
->sum_time
, l
->count
),
3614 len
+= sprintf(buf
+ len
, " age=%ld",
3617 if (l
->min_pid
!= l
->max_pid
)
3618 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3619 l
->min_pid
, l
->max_pid
);
3621 len
+= sprintf(buf
+ len
, " pid=%ld",
3624 if (num_online_cpus() > 1 && !cpus_empty(l
->cpus
) &&
3625 len
< PAGE_SIZE
- 60) {
3626 len
+= sprintf(buf
+ len
, " cpus=");
3627 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3631 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3632 len
< PAGE_SIZE
- 60) {
3633 len
+= sprintf(buf
+ len
, " nodes=");
3634 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3638 len
+= sprintf(buf
+ len
, "\n");
3643 len
+= sprintf(buf
, "No data\n");
3647 enum slab_stat_type
{
3648 SL_ALL
, /* All slabs */
3649 SL_PARTIAL
, /* Only partially allocated slabs */
3650 SL_CPU
, /* Only slabs used for cpu caches */
3651 SL_OBJECTS
, /* Determine allocated objects not slabs */
3652 SL_TOTAL
/* Determine object capacity not slabs */
3655 #define SO_ALL (1 << SL_ALL)
3656 #define SO_PARTIAL (1 << SL_PARTIAL)
3657 #define SO_CPU (1 << SL_CPU)
3658 #define SO_OBJECTS (1 << SL_OBJECTS)
3659 #define SO_TOTAL (1 << SL_TOTAL)
3661 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3662 char *buf
, unsigned long flags
)
3664 unsigned long total
= 0;
3667 unsigned long *nodes
;
3668 unsigned long *per_cpu
;
3670 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3673 per_cpu
= nodes
+ nr_node_ids
;
3675 if (flags
& SO_CPU
) {
3678 for_each_possible_cpu(cpu
) {
3679 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3681 if (!c
|| c
->node
< 0)
3685 if (flags
& SO_TOTAL
)
3686 x
= c
->page
->objects
;
3687 else if (flags
& SO_OBJECTS
)
3693 nodes
[c
->node
] += x
;
3699 if (flags
& SO_ALL
) {
3700 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3701 struct kmem_cache_node
*n
= get_node(s
, node
);
3703 if (flags
& SO_TOTAL
)
3704 x
= atomic_long_read(&n
->total_objects
);
3705 else if (flags
& SO_OBJECTS
)
3706 x
= atomic_long_read(&n
->total_objects
) -
3707 count_partial(n
, count_free
);
3710 x
= atomic_long_read(&n
->nr_slabs
);
3715 } else if (flags
& SO_PARTIAL
) {
3716 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3717 struct kmem_cache_node
*n
= get_node(s
, node
);
3719 if (flags
& SO_TOTAL
)
3720 x
= count_partial(n
, count_total
);
3721 else if (flags
& SO_OBJECTS
)
3722 x
= count_partial(n
, count_inuse
);
3729 x
= sprintf(buf
, "%lu", total
);
3731 for_each_node_state(node
, N_NORMAL_MEMORY
)
3733 x
+= sprintf(buf
+ x
, " N%d=%lu",
3737 return x
+ sprintf(buf
+ x
, "\n");
3740 static int any_slab_objects(struct kmem_cache
*s
)
3744 for_each_online_node(node
) {
3745 struct kmem_cache_node
*n
= get_node(s
, node
);
3750 if (atomic_long_read(&n
->total_objects
))
3756 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3757 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3759 struct slab_attribute
{
3760 struct attribute attr
;
3761 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3762 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3765 #define SLAB_ATTR_RO(_name) \
3766 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3768 #define SLAB_ATTR(_name) \
3769 static struct slab_attribute _name##_attr = \
3770 __ATTR(_name, 0644, _name##_show, _name##_store)
3772 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3774 return sprintf(buf
, "%d\n", s
->size
);
3776 SLAB_ATTR_RO(slab_size
);
3778 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3780 return sprintf(buf
, "%d\n", s
->align
);
3782 SLAB_ATTR_RO(align
);
3784 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3786 return sprintf(buf
, "%d\n", s
->objsize
);
3788 SLAB_ATTR_RO(object_size
);
3790 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3792 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3794 SLAB_ATTR_RO(objs_per_slab
);
3796 static ssize_t
order_store(struct kmem_cache
*s
,
3797 const char *buf
, size_t length
)
3799 unsigned long order
;
3802 err
= strict_strtoul(buf
, 10, &order
);
3806 if (order
> slub_max_order
|| order
< slub_min_order
)
3809 calculate_sizes(s
, order
);
3813 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3815 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3819 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3822 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3824 return n
+ sprintf(buf
+ n
, "\n");
3830 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3832 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3834 SLAB_ATTR_RO(aliases
);
3836 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3838 return show_slab_objects(s
, buf
, SO_ALL
);
3840 SLAB_ATTR_RO(slabs
);
3842 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3844 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3846 SLAB_ATTR_RO(partial
);
3848 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3850 return show_slab_objects(s
, buf
, SO_CPU
);
3852 SLAB_ATTR_RO(cpu_slabs
);
3854 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3856 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3858 SLAB_ATTR_RO(objects
);
3860 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3862 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3864 SLAB_ATTR_RO(objects_partial
);
3866 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3868 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3870 SLAB_ATTR_RO(total_objects
);
3872 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3874 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3877 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3878 const char *buf
, size_t length
)
3880 s
->flags
&= ~SLAB_DEBUG_FREE
;
3882 s
->flags
|= SLAB_DEBUG_FREE
;
3885 SLAB_ATTR(sanity_checks
);
3887 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
3889 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
3892 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
3895 s
->flags
&= ~SLAB_TRACE
;
3897 s
->flags
|= SLAB_TRACE
;
3902 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
3904 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
3907 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
3908 const char *buf
, size_t length
)
3910 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
3912 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
3915 SLAB_ATTR(reclaim_account
);
3917 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
3919 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
3921 SLAB_ATTR_RO(hwcache_align
);
3923 #ifdef CONFIG_ZONE_DMA
3924 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
3926 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
3928 SLAB_ATTR_RO(cache_dma
);
3931 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
3933 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
3935 SLAB_ATTR_RO(destroy_by_rcu
);
3937 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
3939 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
3942 static ssize_t
red_zone_store(struct kmem_cache
*s
,
3943 const char *buf
, size_t length
)
3945 if (any_slab_objects(s
))
3948 s
->flags
&= ~SLAB_RED_ZONE
;
3950 s
->flags
|= SLAB_RED_ZONE
;
3951 calculate_sizes(s
, -1);
3954 SLAB_ATTR(red_zone
);
3956 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
3958 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
3961 static ssize_t
poison_store(struct kmem_cache
*s
,
3962 const char *buf
, size_t length
)
3964 if (any_slab_objects(s
))
3967 s
->flags
&= ~SLAB_POISON
;
3969 s
->flags
|= SLAB_POISON
;
3970 calculate_sizes(s
, -1);
3975 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
3977 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
3980 static ssize_t
store_user_store(struct kmem_cache
*s
,
3981 const char *buf
, size_t length
)
3983 if (any_slab_objects(s
))
3986 s
->flags
&= ~SLAB_STORE_USER
;
3988 s
->flags
|= SLAB_STORE_USER
;
3989 calculate_sizes(s
, -1);
3992 SLAB_ATTR(store_user
);
3994 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
3999 static ssize_t
validate_store(struct kmem_cache
*s
,
4000 const char *buf
, size_t length
)
4004 if (buf
[0] == '1') {
4005 ret
= validate_slab_cache(s
);
4011 SLAB_ATTR(validate
);
4013 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4018 static ssize_t
shrink_store(struct kmem_cache
*s
,
4019 const char *buf
, size_t length
)
4021 if (buf
[0] == '1') {
4022 int rc
= kmem_cache_shrink(s
);
4032 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4034 if (!(s
->flags
& SLAB_STORE_USER
))
4036 return list_locations(s
, buf
, TRACK_ALLOC
);
4038 SLAB_ATTR_RO(alloc_calls
);
4040 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4042 if (!(s
->flags
& SLAB_STORE_USER
))
4044 return list_locations(s
, buf
, TRACK_FREE
);
4046 SLAB_ATTR_RO(free_calls
);
4049 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4051 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4054 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4055 const char *buf
, size_t length
)
4057 unsigned long ratio
;
4060 err
= strict_strtoul(buf
, 10, &ratio
);
4065 s
->remote_node_defrag_ratio
= ratio
* 10;
4069 SLAB_ATTR(remote_node_defrag_ratio
);
4072 #ifdef CONFIG_SLUB_STATS
4073 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4075 unsigned long sum
= 0;
4078 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4083 for_each_online_cpu(cpu
) {
4084 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4090 len
= sprintf(buf
, "%lu", sum
);
4093 for_each_online_cpu(cpu
) {
4094 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4095 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4099 return len
+ sprintf(buf
+ len
, "\n");
4102 #define STAT_ATTR(si, text) \
4103 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4105 return show_stat(s, buf, si); \
4107 SLAB_ATTR_RO(text); \
4109 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4110 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4111 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4112 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4113 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4114 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4115 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4116 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4117 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4118 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4119 STAT_ATTR(FREE_SLAB
, free_slab
);
4120 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4121 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4122 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4123 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4124 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4125 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4126 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4129 static struct attribute
*slab_attrs
[] = {
4130 &slab_size_attr
.attr
,
4131 &object_size_attr
.attr
,
4132 &objs_per_slab_attr
.attr
,
4135 &objects_partial_attr
.attr
,
4136 &total_objects_attr
.attr
,
4139 &cpu_slabs_attr
.attr
,
4143 &sanity_checks_attr
.attr
,
4145 &hwcache_align_attr
.attr
,
4146 &reclaim_account_attr
.attr
,
4147 &destroy_by_rcu_attr
.attr
,
4148 &red_zone_attr
.attr
,
4150 &store_user_attr
.attr
,
4151 &validate_attr
.attr
,
4153 &alloc_calls_attr
.attr
,
4154 &free_calls_attr
.attr
,
4155 #ifdef CONFIG_ZONE_DMA
4156 &cache_dma_attr
.attr
,
4159 &remote_node_defrag_ratio_attr
.attr
,
4161 #ifdef CONFIG_SLUB_STATS
4162 &alloc_fastpath_attr
.attr
,
4163 &alloc_slowpath_attr
.attr
,
4164 &free_fastpath_attr
.attr
,
4165 &free_slowpath_attr
.attr
,
4166 &free_frozen_attr
.attr
,
4167 &free_add_partial_attr
.attr
,
4168 &free_remove_partial_attr
.attr
,
4169 &alloc_from_partial_attr
.attr
,
4170 &alloc_slab_attr
.attr
,
4171 &alloc_refill_attr
.attr
,
4172 &free_slab_attr
.attr
,
4173 &cpuslab_flush_attr
.attr
,
4174 &deactivate_full_attr
.attr
,
4175 &deactivate_empty_attr
.attr
,
4176 &deactivate_to_head_attr
.attr
,
4177 &deactivate_to_tail_attr
.attr
,
4178 &deactivate_remote_frees_attr
.attr
,
4179 &order_fallback_attr
.attr
,
4184 static struct attribute_group slab_attr_group
= {
4185 .attrs
= slab_attrs
,
4188 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4189 struct attribute
*attr
,
4192 struct slab_attribute
*attribute
;
4193 struct kmem_cache
*s
;
4196 attribute
= to_slab_attr(attr
);
4199 if (!attribute
->show
)
4202 err
= attribute
->show(s
, buf
);
4207 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4208 struct attribute
*attr
,
4209 const char *buf
, size_t len
)
4211 struct slab_attribute
*attribute
;
4212 struct kmem_cache
*s
;
4215 attribute
= to_slab_attr(attr
);
4218 if (!attribute
->store
)
4221 err
= attribute
->store(s
, buf
, len
);
4226 static void kmem_cache_release(struct kobject
*kobj
)
4228 struct kmem_cache
*s
= to_slab(kobj
);
4233 static struct sysfs_ops slab_sysfs_ops
= {
4234 .show
= slab_attr_show
,
4235 .store
= slab_attr_store
,
4238 static struct kobj_type slab_ktype
= {
4239 .sysfs_ops
= &slab_sysfs_ops
,
4240 .release
= kmem_cache_release
4243 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4245 struct kobj_type
*ktype
= get_ktype(kobj
);
4247 if (ktype
== &slab_ktype
)
4252 static struct kset_uevent_ops slab_uevent_ops
= {
4253 .filter
= uevent_filter
,
4256 static struct kset
*slab_kset
;
4258 #define ID_STR_LENGTH 64
4260 /* Create a unique string id for a slab cache:
4262 * Format :[flags-]size
4264 static char *create_unique_id(struct kmem_cache
*s
)
4266 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4273 * First flags affecting slabcache operations. We will only
4274 * get here for aliasable slabs so we do not need to support
4275 * too many flags. The flags here must cover all flags that
4276 * are matched during merging to guarantee that the id is
4279 if (s
->flags
& SLAB_CACHE_DMA
)
4281 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4283 if (s
->flags
& SLAB_DEBUG_FREE
)
4287 p
+= sprintf(p
, "%07d", s
->size
);
4288 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4292 static int sysfs_slab_add(struct kmem_cache
*s
)
4298 if (slab_state
< SYSFS
)
4299 /* Defer until later */
4302 unmergeable
= slab_unmergeable(s
);
4305 * Slabcache can never be merged so we can use the name proper.
4306 * This is typically the case for debug situations. In that
4307 * case we can catch duplicate names easily.
4309 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4313 * Create a unique name for the slab as a target
4316 name
= create_unique_id(s
);
4319 s
->kobj
.kset
= slab_kset
;
4320 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4322 kobject_put(&s
->kobj
);
4326 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4329 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4331 /* Setup first alias */
4332 sysfs_slab_alias(s
, s
->name
);
4338 static void sysfs_slab_remove(struct kmem_cache
*s
)
4340 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4341 kobject_del(&s
->kobj
);
4342 kobject_put(&s
->kobj
);
4346 * Need to buffer aliases during bootup until sysfs becomes
4347 * available lest we loose that information.
4349 struct saved_alias
{
4350 struct kmem_cache
*s
;
4352 struct saved_alias
*next
;
4355 static struct saved_alias
*alias_list
;
4357 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4359 struct saved_alias
*al
;
4361 if (slab_state
== SYSFS
) {
4363 * If we have a leftover link then remove it.
4365 sysfs_remove_link(&slab_kset
->kobj
, name
);
4366 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4369 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4375 al
->next
= alias_list
;
4380 static int __init
slab_sysfs_init(void)
4382 struct kmem_cache
*s
;
4385 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4387 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4393 list_for_each_entry(s
, &slab_caches
, list
) {
4394 err
= sysfs_slab_add(s
);
4396 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4397 " to sysfs\n", s
->name
);
4400 while (alias_list
) {
4401 struct saved_alias
*al
= alias_list
;
4403 alias_list
= alias_list
->next
;
4404 err
= sysfs_slab_alias(al
->s
, al
->name
);
4406 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4407 " %s to sysfs\n", s
->name
);
4415 __initcall(slab_sysfs_init
);
4419 * The /proc/slabinfo ABI
4421 #ifdef CONFIG_SLABINFO
4422 static void print_slabinfo_header(struct seq_file
*m
)
4424 seq_puts(m
, "slabinfo - version: 2.1\n");
4425 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4426 "<objperslab> <pagesperslab>");
4427 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4428 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4432 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4436 down_read(&slub_lock
);
4438 print_slabinfo_header(m
);
4440 return seq_list_start(&slab_caches
, *pos
);
4443 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4445 return seq_list_next(p
, &slab_caches
, pos
);
4448 static void s_stop(struct seq_file
*m
, void *p
)
4450 up_read(&slub_lock
);
4453 static int s_show(struct seq_file
*m
, void *p
)
4455 unsigned long nr_partials
= 0;
4456 unsigned long nr_slabs
= 0;
4457 unsigned long nr_inuse
= 0;
4458 unsigned long nr_objs
= 0;
4459 unsigned long nr_free
= 0;
4460 struct kmem_cache
*s
;
4463 s
= list_entry(p
, struct kmem_cache
, list
);
4465 for_each_online_node(node
) {
4466 struct kmem_cache_node
*n
= get_node(s
, node
);
4471 nr_partials
+= n
->nr_partial
;
4472 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4473 nr_objs
+= atomic_long_read(&n
->total_objects
);
4474 nr_free
+= count_partial(n
, count_free
);
4477 nr_inuse
= nr_objs
- nr_free
;
4479 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4480 nr_objs
, s
->size
, oo_objects(s
->oo
),
4481 (1 << oo_order(s
->oo
)));
4482 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4483 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4489 static const struct seq_operations slabinfo_op
= {
4496 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4498 return seq_open(file
, &slabinfo_op
);
4501 static const struct file_operations proc_slabinfo_operations
= {
4502 .open
= slabinfo_open
,
4504 .llseek
= seq_lseek
,
4505 .release
= seq_release
,
4508 static int __init
slab_proc_init(void)
4510 proc_create("slabinfo",S_IWUSR
|S_IRUGO
,NULL
,&proc_slabinfo_operations
);
4513 module_init(slab_proc_init
);
4514 #endif /* CONFIG_SLABINFO */