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/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemtrace.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/kmemleak.h>
24 #include <linux/mempolicy.h>
25 #include <linux/ctype.h>
26 #include <linux/debugobjects.h>
27 #include <linux/kallsyms.h>
28 #include <linux/memory.h>
29 #include <linux/math64.h>
30 #include <linux/fault-inject.h>
37 * The slab_lock protects operations on the object of a particular
38 * slab and its metadata in the page struct. If the slab lock
39 * has been taken then no allocations nor frees can be performed
40 * on the objects in the slab nor can the slab be added or removed
41 * from the partial or full lists since this would mean modifying
42 * the page_struct of the slab.
44 * The list_lock protects the partial and full list on each node and
45 * the partial slab counter. If taken then no new slabs may be added or
46 * removed from the lists nor make the number of partial slabs be modified.
47 * (Note that the total number of slabs is an atomic value that may be
48 * modified without taking the list lock).
50 * The list_lock is a centralized lock and thus we avoid taking it as
51 * much as possible. As long as SLUB does not have to handle partial
52 * slabs, operations can continue without any centralized lock. F.e.
53 * allocating a long series of objects that fill up slabs does not require
56 * The lock order is sometimes inverted when we are trying to get a slab
57 * off a list. We take the list_lock and then look for a page on the list
58 * to use. While we do that objects in the slabs may be freed. We can
59 * only operate on the slab if we have also taken the slab_lock. So we use
60 * a slab_trylock() on the slab. If trylock was successful then no frees
61 * can occur anymore and we can use the slab for allocations etc. If the
62 * slab_trylock() does not succeed then frees are in progress in the slab and
63 * we must stay away from it for a while since we may cause a bouncing
64 * cacheline if we try to acquire the lock. So go onto the next slab.
65 * If all pages are busy then we may allocate a new slab instead of reusing
66 * a partial slab. A new slab has noone operating on it and thus there is
67 * no danger of cacheline contention.
69 * Interrupts are disabled during allocation and deallocation in order to
70 * make the slab allocator safe to use in the context of an irq. In addition
71 * interrupts are disabled to ensure that the processor does not change
72 * while handling per_cpu slabs, due to kernel preemption.
74 * SLUB assigns one slab for allocation to each processor.
75 * Allocations only occur from these slabs called cpu slabs.
77 * Slabs with free elements are kept on a partial list and during regular
78 * operations no list for full slabs is used. If an object in a full slab is
79 * freed then the slab will show up again on the partial lists.
80 * We track full slabs for debugging purposes though because otherwise we
81 * cannot scan all objects.
83 * Slabs are freed when they become empty. Teardown and setup is
84 * minimal so we rely on the page allocators per cpu caches for
85 * fast frees and allocs.
87 * Overloading of page flags that are otherwise used for LRU management.
89 * PageActive The slab is frozen and exempt from list processing.
90 * This means that the slab is dedicated to a purpose
91 * such as satisfying allocations for a specific
92 * processor. Objects may be freed in the slab while
93 * it is frozen but slab_free will then skip the usual
94 * list operations. It is up to the processor holding
95 * the slab to integrate the slab into the slab lists
96 * when the slab is no longer needed.
98 * One use of this flag is to mark slabs that are
99 * used for allocations. Then such a slab becomes a cpu
100 * slab. The cpu slab may be equipped with an additional
101 * freelist that allows lockless access to
102 * free objects in addition to the regular freelist
103 * that requires the slab lock.
105 * PageError Slab requires special handling due to debug
106 * options set. This moves slab handling out of
107 * the fast path and disables lockless freelists.
110 #ifdef CONFIG_SLUB_DEBUG
117 * Issues still to be resolved:
119 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
121 * - Variable sizing of the per node arrays
124 /* Enable to test recovery from slab corruption on boot */
125 #undef SLUB_RESILIENCY_TEST
128 * Mininum number of partial slabs. These will be left on the partial
129 * lists even if they are empty. kmem_cache_shrink may reclaim them.
131 #define MIN_PARTIAL 5
134 * Maximum number of desirable partial slabs.
135 * The existence of more partial slabs makes kmem_cache_shrink
136 * sort the partial list by the number of objects in the.
138 #define MAX_PARTIAL 10
140 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
141 SLAB_POISON | SLAB_STORE_USER)
144 * Set of flags that will prevent slab merging
146 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
147 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE)
149 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
152 #ifndef ARCH_KMALLOC_MINALIGN
153 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
156 #ifndef ARCH_SLAB_MINALIGN
157 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
161 #define OO_MASK ((1 << OO_SHIFT) - 1)
162 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
164 /* Internal SLUB flags */
165 #define __OBJECT_POISON 0x80000000 /* Poison object */
166 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
168 static int kmem_size
= sizeof(struct kmem_cache
);
171 static struct notifier_block slab_notifier
;
175 DOWN
, /* No slab functionality available */
176 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
177 UP
, /* Everything works but does not show up in sysfs */
181 /* A list of all slab caches on the system */
182 static DECLARE_RWSEM(slub_lock
);
183 static LIST_HEAD(slab_caches
);
186 * Tracking user of a slab.
189 unsigned long addr
; /* Called from address */
190 int cpu
; /* Was running on cpu */
191 int pid
; /* Pid context */
192 unsigned long when
; /* When did the operation occur */
195 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
197 #ifdef CONFIG_SLUB_DEBUG
198 static int sysfs_slab_add(struct kmem_cache
*);
199 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
200 static void sysfs_slab_remove(struct kmem_cache
*);
203 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
204 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
206 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
213 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
215 #ifdef CONFIG_SLUB_STATS
220 /********************************************************************
221 * Core slab cache functions
222 *******************************************************************/
224 int slab_is_available(void)
226 return slab_state
>= UP
;
229 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
232 return s
->node
[node
];
234 return &s
->local_node
;
238 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
241 return s
->cpu_slab
[cpu
];
247 /* Verify that a pointer has an address that is valid within a slab page */
248 static inline int check_valid_pointer(struct kmem_cache
*s
,
249 struct page
*page
, const void *object
)
256 base
= page_address(page
);
257 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
258 (object
- base
) % s
->size
) {
266 * Slow version of get and set free pointer.
268 * This version requires touching the cache lines of kmem_cache which
269 * we avoid to do in the fast alloc free paths. There we obtain the offset
270 * from the page struct.
272 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
274 return *(void **)(object
+ s
->offset
);
277 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
279 *(void **)(object
+ s
->offset
) = fp
;
282 /* Loop over all objects in a slab */
283 #define for_each_object(__p, __s, __addr, __objects) \
284 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
288 #define for_each_free_object(__p, __s, __free) \
289 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
291 /* Determine object index from a given position */
292 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
294 return (p
- addr
) / s
->size
;
297 static inline struct kmem_cache_order_objects
oo_make(int order
,
300 struct kmem_cache_order_objects x
= {
301 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
307 static inline int oo_order(struct kmem_cache_order_objects x
)
309 return x
.x
>> OO_SHIFT
;
312 static inline int oo_objects(struct kmem_cache_order_objects x
)
314 return x
.x
& OO_MASK
;
317 #ifdef CONFIG_SLUB_DEBUG
321 #ifdef CONFIG_SLUB_DEBUG_ON
322 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
324 static int slub_debug
;
327 static char *slub_debug_slabs
;
332 static void print_section(char *text
, u8
*addr
, unsigned int length
)
340 for (i
= 0; i
< length
; i
++) {
342 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
345 printk(KERN_CONT
" %02x", addr
[i
]);
347 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
349 printk(KERN_CONT
" %s\n", ascii
);
356 printk(KERN_CONT
" ");
360 printk(KERN_CONT
" %s\n", ascii
);
364 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
365 enum track_item alloc
)
370 p
= object
+ s
->offset
+ sizeof(void *);
372 p
= object
+ s
->inuse
;
377 static void set_track(struct kmem_cache
*s
, void *object
,
378 enum track_item alloc
, unsigned long addr
)
380 struct track
*p
= get_track(s
, object
, alloc
);
384 p
->cpu
= smp_processor_id();
385 p
->pid
= current
->pid
;
388 memset(p
, 0, sizeof(struct track
));
391 static void init_tracking(struct kmem_cache
*s
, void *object
)
393 if (!(s
->flags
& SLAB_STORE_USER
))
396 set_track(s
, object
, TRACK_FREE
, 0UL);
397 set_track(s
, object
, TRACK_ALLOC
, 0UL);
400 static void print_track(const char *s
, struct track
*t
)
405 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
406 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
409 static void print_tracking(struct kmem_cache
*s
, void *object
)
411 if (!(s
->flags
& SLAB_STORE_USER
))
414 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
415 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
418 static void print_page_info(struct page
*page
)
420 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
421 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
425 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
431 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
433 printk(KERN_ERR
"========================================"
434 "=====================================\n");
435 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
436 printk(KERN_ERR
"----------------------------------------"
437 "-------------------------------------\n\n");
440 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
446 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
448 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
451 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
453 unsigned int off
; /* Offset of last byte */
454 u8
*addr
= page_address(page
);
456 print_tracking(s
, p
);
458 print_page_info(page
);
460 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
461 p
, p
- addr
, get_freepointer(s
, p
));
464 print_section("Bytes b4", p
- 16, 16);
466 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
468 if (s
->flags
& SLAB_RED_ZONE
)
469 print_section("Redzone", p
+ s
->objsize
,
470 s
->inuse
- s
->objsize
);
473 off
= s
->offset
+ sizeof(void *);
477 if (s
->flags
& SLAB_STORE_USER
)
478 off
+= 2 * sizeof(struct track
);
481 /* Beginning of the filler is the free pointer */
482 print_section("Padding", p
+ off
, s
->size
- off
);
487 static void object_err(struct kmem_cache
*s
, struct page
*page
,
488 u8
*object
, char *reason
)
490 slab_bug(s
, "%s", reason
);
491 print_trailer(s
, page
, object
);
494 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
500 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
502 slab_bug(s
, "%s", buf
);
503 print_page_info(page
);
507 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
511 if (s
->flags
& __OBJECT_POISON
) {
512 memset(p
, POISON_FREE
, s
->objsize
- 1);
513 p
[s
->objsize
- 1] = POISON_END
;
516 if (s
->flags
& SLAB_RED_ZONE
)
517 memset(p
+ s
->objsize
,
518 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
519 s
->inuse
- s
->objsize
);
522 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
525 if (*start
!= (u8
)value
)
533 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
534 void *from
, void *to
)
536 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
537 memset(from
, data
, to
- from
);
540 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
541 u8
*object
, char *what
,
542 u8
*start
, unsigned int value
, unsigned int bytes
)
547 fault
= check_bytes(start
, value
, bytes
);
552 while (end
> fault
&& end
[-1] == value
)
555 slab_bug(s
, "%s overwritten", what
);
556 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
557 fault
, end
- 1, fault
[0], value
);
558 print_trailer(s
, page
, object
);
560 restore_bytes(s
, what
, value
, fault
, end
);
568 * Bytes of the object to be managed.
569 * If the freepointer may overlay the object then the free
570 * pointer is the first word of the object.
572 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
575 * object + s->objsize
576 * Padding to reach word boundary. This is also used for Redzoning.
577 * Padding is extended by another word if Redzoning is enabled and
580 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
581 * 0xcc (RED_ACTIVE) for objects in use.
584 * Meta data starts here.
586 * A. Free pointer (if we cannot overwrite object on free)
587 * B. Tracking data for SLAB_STORE_USER
588 * C. Padding to reach required alignment boundary or at mininum
589 * one word if debugging is on to be able to detect writes
590 * before the word boundary.
592 * Padding is done using 0x5a (POISON_INUSE)
595 * Nothing is used beyond s->size.
597 * If slabcaches are merged then the objsize and inuse boundaries are mostly
598 * ignored. And therefore no slab options that rely on these boundaries
599 * may be used with merged slabcaches.
602 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
604 unsigned long off
= s
->inuse
; /* The end of info */
607 /* Freepointer is placed after the object. */
608 off
+= sizeof(void *);
610 if (s
->flags
& SLAB_STORE_USER
)
611 /* We also have user information there */
612 off
+= 2 * sizeof(struct track
);
617 return check_bytes_and_report(s
, page
, p
, "Object padding",
618 p
+ off
, POISON_INUSE
, s
->size
- off
);
621 /* Check the pad bytes at the end of a slab page */
622 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
630 if (!(s
->flags
& SLAB_POISON
))
633 start
= page_address(page
);
634 length
= (PAGE_SIZE
<< compound_order(page
));
635 end
= start
+ length
;
636 remainder
= length
% s
->size
;
640 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
643 while (end
> fault
&& end
[-1] == POISON_INUSE
)
646 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
647 print_section("Padding", end
- remainder
, remainder
);
649 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
653 static int check_object(struct kmem_cache
*s
, struct page
*page
,
654 void *object
, int active
)
657 u8
*endobject
= object
+ s
->objsize
;
659 if (s
->flags
& SLAB_RED_ZONE
) {
661 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
663 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
664 endobject
, red
, s
->inuse
- s
->objsize
))
667 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
668 check_bytes_and_report(s
, page
, p
, "Alignment padding",
669 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
673 if (s
->flags
& SLAB_POISON
) {
674 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
675 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
676 POISON_FREE
, s
->objsize
- 1) ||
677 !check_bytes_and_report(s
, page
, p
, "Poison",
678 p
+ s
->objsize
- 1, POISON_END
, 1)))
681 * check_pad_bytes cleans up on its own.
683 check_pad_bytes(s
, page
, p
);
686 if (!s
->offset
&& active
)
688 * Object and freepointer overlap. Cannot check
689 * freepointer while object is allocated.
693 /* Check free pointer validity */
694 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
695 object_err(s
, page
, p
, "Freepointer corrupt");
697 * No choice but to zap it and thus lose the remainder
698 * of the free objects in this slab. May cause
699 * another error because the object count is now wrong.
701 set_freepointer(s
, p
, NULL
);
707 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
711 VM_BUG_ON(!irqs_disabled());
713 if (!PageSlab(page
)) {
714 slab_err(s
, page
, "Not a valid slab page");
718 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
719 if (page
->objects
> maxobj
) {
720 slab_err(s
, page
, "objects %u > max %u",
721 s
->name
, page
->objects
, maxobj
);
724 if (page
->inuse
> page
->objects
) {
725 slab_err(s
, page
, "inuse %u > max %u",
726 s
->name
, page
->inuse
, page
->objects
);
729 /* Slab_pad_check fixes things up after itself */
730 slab_pad_check(s
, page
);
735 * Determine if a certain object on a page is on the freelist. Must hold the
736 * slab lock to guarantee that the chains are in a consistent state.
738 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
741 void *fp
= page
->freelist
;
743 unsigned long max_objects
;
745 while (fp
&& nr
<= page
->objects
) {
748 if (!check_valid_pointer(s
, page
, fp
)) {
750 object_err(s
, page
, object
,
751 "Freechain corrupt");
752 set_freepointer(s
, object
, NULL
);
755 slab_err(s
, page
, "Freepointer corrupt");
756 page
->freelist
= NULL
;
757 page
->inuse
= page
->objects
;
758 slab_fix(s
, "Freelist cleared");
764 fp
= get_freepointer(s
, object
);
768 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
769 if (max_objects
> MAX_OBJS_PER_PAGE
)
770 max_objects
= MAX_OBJS_PER_PAGE
;
772 if (page
->objects
!= max_objects
) {
773 slab_err(s
, page
, "Wrong number of objects. Found %d but "
774 "should be %d", page
->objects
, max_objects
);
775 page
->objects
= max_objects
;
776 slab_fix(s
, "Number of objects adjusted.");
778 if (page
->inuse
!= page
->objects
- nr
) {
779 slab_err(s
, page
, "Wrong object count. Counter is %d but "
780 "counted were %d", page
->inuse
, page
->objects
- nr
);
781 page
->inuse
= page
->objects
- nr
;
782 slab_fix(s
, "Object count adjusted.");
784 return search
== NULL
;
787 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
790 if (s
->flags
& SLAB_TRACE
) {
791 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
793 alloc
? "alloc" : "free",
798 print_section("Object", (void *)object
, s
->objsize
);
805 * Tracking of fully allocated slabs for debugging purposes.
807 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
809 spin_lock(&n
->list_lock
);
810 list_add(&page
->lru
, &n
->full
);
811 spin_unlock(&n
->list_lock
);
814 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
816 struct kmem_cache_node
*n
;
818 if (!(s
->flags
& SLAB_STORE_USER
))
821 n
= get_node(s
, page_to_nid(page
));
823 spin_lock(&n
->list_lock
);
824 list_del(&page
->lru
);
825 spin_unlock(&n
->list_lock
);
828 /* Tracking of the number of slabs for debugging purposes */
829 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
831 struct kmem_cache_node
*n
= get_node(s
, node
);
833 return atomic_long_read(&n
->nr_slabs
);
836 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
838 struct kmem_cache_node
*n
= get_node(s
, node
);
841 * May be called early in order to allocate a slab for the
842 * kmem_cache_node structure. Solve the chicken-egg
843 * dilemma by deferring the increment of the count during
844 * bootstrap (see early_kmem_cache_node_alloc).
846 if (!NUMA_BUILD
|| n
) {
847 atomic_long_inc(&n
->nr_slabs
);
848 atomic_long_add(objects
, &n
->total_objects
);
851 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
853 struct kmem_cache_node
*n
= get_node(s
, node
);
855 atomic_long_dec(&n
->nr_slabs
);
856 atomic_long_sub(objects
, &n
->total_objects
);
859 /* Object debug checks for alloc/free paths */
860 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
863 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
866 init_object(s
, object
, 0);
867 init_tracking(s
, object
);
870 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
871 void *object
, unsigned long addr
)
873 if (!check_slab(s
, page
))
876 if (!on_freelist(s
, page
, object
)) {
877 object_err(s
, page
, object
, "Object already allocated");
881 if (!check_valid_pointer(s
, page
, object
)) {
882 object_err(s
, page
, object
, "Freelist Pointer check fails");
886 if (!check_object(s
, page
, object
, 0))
889 /* Success perform special debug activities for allocs */
890 if (s
->flags
& SLAB_STORE_USER
)
891 set_track(s
, object
, TRACK_ALLOC
, addr
);
892 trace(s
, page
, object
, 1);
893 init_object(s
, object
, 1);
897 if (PageSlab(page
)) {
899 * If this is a slab page then lets do the best we can
900 * to avoid issues in the future. Marking all objects
901 * as used avoids touching the remaining objects.
903 slab_fix(s
, "Marking all objects used");
904 page
->inuse
= page
->objects
;
905 page
->freelist
= NULL
;
910 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
911 void *object
, unsigned long addr
)
913 if (!check_slab(s
, page
))
916 if (!check_valid_pointer(s
, page
, object
)) {
917 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
921 if (on_freelist(s
, page
, object
)) {
922 object_err(s
, page
, object
, "Object already free");
926 if (!check_object(s
, page
, object
, 1))
929 if (unlikely(s
!= page
->slab
)) {
930 if (!PageSlab(page
)) {
931 slab_err(s
, page
, "Attempt to free object(0x%p) "
932 "outside of slab", object
);
933 } else if (!page
->slab
) {
935 "SLUB <none>: no slab for object 0x%p.\n",
939 object_err(s
, page
, object
,
940 "page slab pointer corrupt.");
944 /* Special debug activities for freeing objects */
945 if (!PageSlubFrozen(page
) && !page
->freelist
)
946 remove_full(s
, page
);
947 if (s
->flags
& SLAB_STORE_USER
)
948 set_track(s
, object
, TRACK_FREE
, addr
);
949 trace(s
, page
, object
, 0);
950 init_object(s
, object
, 0);
954 slab_fix(s
, "Object at 0x%p not freed", object
);
958 static int __init
setup_slub_debug(char *str
)
960 slub_debug
= DEBUG_DEFAULT_FLAGS
;
961 if (*str
++ != '=' || !*str
)
963 * No options specified. Switch on full debugging.
969 * No options but restriction on slabs. This means full
970 * debugging for slabs matching a pattern.
977 * Switch off all debugging measures.
982 * Determine which debug features should be switched on
984 for (; *str
&& *str
!= ','; str
++) {
985 switch (tolower(*str
)) {
987 slub_debug
|= SLAB_DEBUG_FREE
;
990 slub_debug
|= SLAB_RED_ZONE
;
993 slub_debug
|= SLAB_POISON
;
996 slub_debug
|= SLAB_STORE_USER
;
999 slub_debug
|= SLAB_TRACE
;
1002 printk(KERN_ERR
"slub_debug option '%c' "
1003 "unknown. skipped\n", *str
);
1009 slub_debug_slabs
= str
+ 1;
1014 __setup("slub_debug", setup_slub_debug
);
1016 static unsigned long kmem_cache_flags(unsigned long objsize
,
1017 unsigned long flags
, const char *name
,
1018 void (*ctor
)(void *))
1021 * Enable debugging if selected on the kernel commandline.
1023 if (slub_debug
&& (!slub_debug_slabs
||
1024 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1025 flags
|= slub_debug
;
1030 static inline void setup_object_debug(struct kmem_cache
*s
,
1031 struct page
*page
, void *object
) {}
1033 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1034 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1036 static inline int free_debug_processing(struct kmem_cache
*s
,
1037 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1039 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1041 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1042 void *object
, int active
) { return 1; }
1043 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1044 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1045 unsigned long flags
, const char *name
,
1046 void (*ctor
)(void *))
1050 #define slub_debug 0
1052 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1054 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1056 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1061 * Slab allocation and freeing
1063 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1064 struct kmem_cache_order_objects oo
)
1066 int order
= oo_order(oo
);
1069 return alloc_pages(flags
, order
);
1071 return alloc_pages_node(node
, flags
, order
);
1074 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1077 struct kmem_cache_order_objects oo
= s
->oo
;
1079 flags
|= s
->allocflags
;
1081 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1083 if (unlikely(!page
)) {
1086 * Allocation may have failed due to fragmentation.
1087 * Try a lower order alloc if possible
1089 page
= alloc_slab_page(flags
, node
, oo
);
1093 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1095 page
->objects
= oo_objects(oo
);
1096 mod_zone_page_state(page_zone(page
),
1097 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1098 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1104 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1107 setup_object_debug(s
, page
, object
);
1108 if (unlikely(s
->ctor
))
1112 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1119 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1121 page
= allocate_slab(s
,
1122 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1126 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1128 page
->flags
|= 1 << PG_slab
;
1129 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1130 SLAB_STORE_USER
| SLAB_TRACE
))
1131 __SetPageSlubDebug(page
);
1133 start
= page_address(page
);
1135 if (unlikely(s
->flags
& SLAB_POISON
))
1136 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1139 for_each_object(p
, s
, start
, page
->objects
) {
1140 setup_object(s
, page
, last
);
1141 set_freepointer(s
, last
, p
);
1144 setup_object(s
, page
, last
);
1145 set_freepointer(s
, last
, NULL
);
1147 page
->freelist
= start
;
1153 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1155 int order
= compound_order(page
);
1156 int pages
= 1 << order
;
1158 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1161 slab_pad_check(s
, page
);
1162 for_each_object(p
, s
, page_address(page
),
1164 check_object(s
, page
, p
, 0);
1165 __ClearPageSlubDebug(page
);
1168 mod_zone_page_state(page_zone(page
),
1169 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1170 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1173 __ClearPageSlab(page
);
1174 reset_page_mapcount(page
);
1175 if (current
->reclaim_state
)
1176 current
->reclaim_state
->reclaimed_slab
+= pages
;
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
> s
->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
< s
->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 lockdep_trace_alloc(gfpflags
);
1599 might_sleep_if(gfpflags
& __GFP_WAIT
);
1601 if (should_failslab(s
->objsize
, gfpflags
))
1604 local_irq_save(flags
);
1605 c
= get_cpu_slab(s
, smp_processor_id());
1606 objsize
= c
->objsize
;
1607 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1609 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1612 object
= c
->freelist
;
1613 c
->freelist
= object
[c
->offset
];
1614 stat(c
, ALLOC_FASTPATH
);
1616 local_irq_restore(flags
);
1618 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1619 memset(object
, 0, objsize
);
1621 kmemleak_alloc_recursive(object
, objsize
, 1, s
->flags
, gfpflags
);
1625 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1627 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1629 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1633 EXPORT_SYMBOL(kmem_cache_alloc
);
1635 #ifdef CONFIG_KMEMTRACE
1636 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1638 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1640 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1644 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1646 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1648 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1649 s
->objsize
, s
->size
, gfpflags
, node
);
1653 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1656 #ifdef CONFIG_KMEMTRACE
1657 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1661 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1663 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1667 * Slow patch handling. This may still be called frequently since objects
1668 * have a longer lifetime than the cpu slabs in most processing loads.
1670 * So we still attempt to reduce cache line usage. Just take the slab
1671 * lock and free the item. If there is no additional partial page
1672 * handling required then we can return immediately.
1674 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1675 void *x
, unsigned long addr
, unsigned int offset
)
1678 void **object
= (void *)x
;
1679 struct kmem_cache_cpu
*c
;
1681 c
= get_cpu_slab(s
, raw_smp_processor_id());
1682 stat(c
, FREE_SLOWPATH
);
1685 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1689 prior
= object
[offset
] = page
->freelist
;
1690 page
->freelist
= object
;
1693 if (unlikely(PageSlubFrozen(page
))) {
1694 stat(c
, FREE_FROZEN
);
1698 if (unlikely(!page
->inuse
))
1702 * Objects left in the slab. If it was not on the partial list before
1705 if (unlikely(!prior
)) {
1706 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1707 stat(c
, FREE_ADD_PARTIAL
);
1717 * Slab still on the partial list.
1719 remove_partial(s
, page
);
1720 stat(c
, FREE_REMOVE_PARTIAL
);
1724 discard_slab(s
, page
);
1728 if (!free_debug_processing(s
, page
, x
, addr
))
1734 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1735 * can perform fastpath freeing without additional function calls.
1737 * The fastpath is only possible if we are freeing to the current cpu slab
1738 * of this processor. This typically the case if we have just allocated
1741 * If fastpath is not possible then fall back to __slab_free where we deal
1742 * with all sorts of special processing.
1744 static __always_inline
void slab_free(struct kmem_cache
*s
,
1745 struct page
*page
, void *x
, unsigned long addr
)
1747 void **object
= (void *)x
;
1748 struct kmem_cache_cpu
*c
;
1749 unsigned long flags
;
1751 kmemleak_free_recursive(x
, s
->flags
);
1752 local_irq_save(flags
);
1753 c
= get_cpu_slab(s
, smp_processor_id());
1754 debug_check_no_locks_freed(object
, c
->objsize
);
1755 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1756 debug_check_no_obj_freed(object
, c
->objsize
);
1757 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1758 object
[c
->offset
] = c
->freelist
;
1759 c
->freelist
= object
;
1760 stat(c
, FREE_FASTPATH
);
1762 __slab_free(s
, page
, x
, addr
, c
->offset
);
1764 local_irq_restore(flags
);
1767 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1771 page
= virt_to_head_page(x
);
1773 slab_free(s
, page
, x
, _RET_IP_
);
1775 trace_kmem_cache_free(_RET_IP_
, x
);
1777 EXPORT_SYMBOL(kmem_cache_free
);
1779 /* Figure out on which slab page the object resides */
1780 static struct page
*get_object_page(const void *x
)
1782 struct page
*page
= virt_to_head_page(x
);
1784 if (!PageSlab(page
))
1791 * Object placement in a slab is made very easy because we always start at
1792 * offset 0. If we tune the size of the object to the alignment then we can
1793 * get the required alignment by putting one properly sized object after
1796 * Notice that the allocation order determines the sizes of the per cpu
1797 * caches. Each processor has always one slab available for allocations.
1798 * Increasing the allocation order reduces the number of times that slabs
1799 * must be moved on and off the partial lists and is therefore a factor in
1804 * Mininum / Maximum order of slab pages. This influences locking overhead
1805 * and slab fragmentation. A higher order reduces the number of partial slabs
1806 * and increases the number of allocations possible without having to
1807 * take the list_lock.
1809 static int slub_min_order
;
1810 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1811 static int slub_min_objects
;
1814 * Merge control. If this is set then no merging of slab caches will occur.
1815 * (Could be removed. This was introduced to pacify the merge skeptics.)
1817 static int slub_nomerge
;
1820 * Calculate the order of allocation given an slab object size.
1822 * The order of allocation has significant impact on performance and other
1823 * system components. Generally order 0 allocations should be preferred since
1824 * order 0 does not cause fragmentation in the page allocator. Larger objects
1825 * be problematic to put into order 0 slabs because there may be too much
1826 * unused space left. We go to a higher order if more than 1/16th of the slab
1829 * In order to reach satisfactory performance we must ensure that a minimum
1830 * number of objects is in one slab. Otherwise we may generate too much
1831 * activity on the partial lists which requires taking the list_lock. This is
1832 * less a concern for large slabs though which are rarely used.
1834 * slub_max_order specifies the order where we begin to stop considering the
1835 * number of objects in a slab as critical. If we reach slub_max_order then
1836 * we try to keep the page order as low as possible. So we accept more waste
1837 * of space in favor of a small page order.
1839 * Higher order allocations also allow the placement of more objects in a
1840 * slab and thereby reduce object handling overhead. If the user has
1841 * requested a higher mininum order then we start with that one instead of
1842 * the smallest order which will fit the object.
1844 static inline int slab_order(int size
, int min_objects
,
1845 int max_order
, int fract_leftover
)
1849 int min_order
= slub_min_order
;
1851 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1852 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1854 for (order
= max(min_order
,
1855 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1856 order
<= max_order
; order
++) {
1858 unsigned long slab_size
= PAGE_SIZE
<< order
;
1860 if (slab_size
< min_objects
* size
)
1863 rem
= slab_size
% size
;
1865 if (rem
<= slab_size
/ fract_leftover
)
1873 static inline int calculate_order(int size
)
1881 * Attempt to find best configuration for a slab. This
1882 * works by first attempting to generate a layout with
1883 * the best configuration and backing off gradually.
1885 * First we reduce the acceptable waste in a slab. Then
1886 * we reduce the minimum objects required in a slab.
1888 min_objects
= slub_min_objects
;
1890 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1891 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1892 min_objects
= min(min_objects
, max_objects
);
1894 while (min_objects
> 1) {
1896 while (fraction
>= 4) {
1897 order
= slab_order(size
, min_objects
,
1898 slub_max_order
, fraction
);
1899 if (order
<= slub_max_order
)
1907 * We were unable to place multiple objects in a slab. Now
1908 * lets see if we can place a single object there.
1910 order
= slab_order(size
, 1, slub_max_order
, 1);
1911 if (order
<= slub_max_order
)
1915 * Doh this slab cannot be placed using slub_max_order.
1917 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1918 if (order
< MAX_ORDER
)
1924 * Figure out what the alignment of the objects will be.
1926 static unsigned long calculate_alignment(unsigned long flags
,
1927 unsigned long align
, unsigned long size
)
1930 * If the user wants hardware cache aligned objects then follow that
1931 * suggestion if the object is sufficiently large.
1933 * The hardware cache alignment cannot override the specified
1934 * alignment though. If that is greater then use it.
1936 if (flags
& SLAB_HWCACHE_ALIGN
) {
1937 unsigned long ralign
= cache_line_size();
1938 while (size
<= ralign
/ 2)
1940 align
= max(align
, ralign
);
1943 if (align
< ARCH_SLAB_MINALIGN
)
1944 align
= ARCH_SLAB_MINALIGN
;
1946 return ALIGN(align
, sizeof(void *));
1949 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1950 struct kmem_cache_cpu
*c
)
1955 c
->offset
= s
->offset
/ sizeof(void *);
1956 c
->objsize
= s
->objsize
;
1957 #ifdef CONFIG_SLUB_STATS
1958 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1963 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
1966 spin_lock_init(&n
->list_lock
);
1967 INIT_LIST_HEAD(&n
->partial
);
1968 #ifdef CONFIG_SLUB_DEBUG
1969 atomic_long_set(&n
->nr_slabs
, 0);
1970 atomic_long_set(&n
->total_objects
, 0);
1971 INIT_LIST_HEAD(&n
->full
);
1977 * Per cpu array for per cpu structures.
1979 * The per cpu array places all kmem_cache_cpu structures from one processor
1980 * close together meaning that it becomes possible that multiple per cpu
1981 * structures are contained in one cacheline. This may be particularly
1982 * beneficial for the kmalloc caches.
1984 * A desktop system typically has around 60-80 slabs. With 100 here we are
1985 * likely able to get per cpu structures for all caches from the array defined
1986 * here. We must be able to cover all kmalloc caches during bootstrap.
1988 * If the per cpu array is exhausted then fall back to kmalloc
1989 * of individual cachelines. No sharing is possible then.
1991 #define NR_KMEM_CACHE_CPU 100
1993 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
1994 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
1996 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
1997 static DECLARE_BITMAP(kmem_cach_cpu_free_init_once
, CONFIG_NR_CPUS
);
1999 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
2000 int cpu
, gfp_t flags
)
2002 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
2005 per_cpu(kmem_cache_cpu_free
, cpu
) =
2006 (void *)c
->freelist
;
2008 /* Table overflow: So allocate ourselves */
2010 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2011 flags
, cpu_to_node(cpu
));
2016 init_kmem_cache_cpu(s
, c
);
2020 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2022 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2023 c
>= per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2027 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2028 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2031 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2035 for_each_online_cpu(cpu
) {
2036 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2039 s
->cpu_slab
[cpu
] = NULL
;
2040 free_kmem_cache_cpu(c
, cpu
);
2045 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2049 for_each_online_cpu(cpu
) {
2050 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2055 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2057 free_kmem_cache_cpus(s
);
2060 s
->cpu_slab
[cpu
] = c
;
2066 * Initialize the per cpu array.
2068 static void init_alloc_cpu_cpu(int cpu
)
2072 if (cpumask_test_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
)))
2075 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2076 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2078 cpumask_set_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
));
2081 static void __init
init_alloc_cpu(void)
2085 for_each_online_cpu(cpu
)
2086 init_alloc_cpu_cpu(cpu
);
2090 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2091 static inline void init_alloc_cpu(void) {}
2093 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2095 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2102 * No kmalloc_node yet so do it by hand. We know that this is the first
2103 * slab on the node for this slabcache. There are no concurrent accesses
2106 * Note that this function only works on the kmalloc_node_cache
2107 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2108 * memory on a fresh node that has no slab structures yet.
2110 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2113 struct kmem_cache_node
*n
;
2114 unsigned long flags
;
2116 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2118 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2121 if (page_to_nid(page
) != node
) {
2122 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2124 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2125 "in order to be able to continue\n");
2130 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2132 kmalloc_caches
->node
[node
] = n
;
2133 #ifdef CONFIG_SLUB_DEBUG
2134 init_object(kmalloc_caches
, n
, 1);
2135 init_tracking(kmalloc_caches
, n
);
2137 init_kmem_cache_node(n
, kmalloc_caches
);
2138 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2141 * lockdep requires consistent irq usage for each lock
2142 * so even though there cannot be a race this early in
2143 * the boot sequence, we still disable irqs.
2145 local_irq_save(flags
);
2146 add_partial(n
, page
, 0);
2147 local_irq_restore(flags
);
2150 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2154 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2155 struct kmem_cache_node
*n
= s
->node
[node
];
2156 if (n
&& n
!= &s
->local_node
)
2157 kmem_cache_free(kmalloc_caches
, n
);
2158 s
->node
[node
] = NULL
;
2162 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2167 if (slab_state
>= UP
)
2168 local_node
= page_to_nid(virt_to_page(s
));
2172 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2173 struct kmem_cache_node
*n
;
2175 if (local_node
== node
)
2178 if (slab_state
== DOWN
) {
2179 early_kmem_cache_node_alloc(gfpflags
, node
);
2182 n
= kmem_cache_alloc_node(kmalloc_caches
,
2186 free_kmem_cache_nodes(s
);
2192 init_kmem_cache_node(n
, s
);
2197 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2201 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2203 init_kmem_cache_node(&s
->local_node
, s
);
2208 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2210 if (min
< MIN_PARTIAL
)
2212 else if (min
> MAX_PARTIAL
)
2214 s
->min_partial
= min
;
2218 * calculate_sizes() determines the order and the distribution of data within
2221 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2223 unsigned long flags
= s
->flags
;
2224 unsigned long size
= s
->objsize
;
2225 unsigned long align
= s
->align
;
2229 * Round up object size to the next word boundary. We can only
2230 * place the free pointer at word boundaries and this determines
2231 * the possible location of the free pointer.
2233 size
= ALIGN(size
, sizeof(void *));
2235 #ifdef CONFIG_SLUB_DEBUG
2237 * Determine if we can poison the object itself. If the user of
2238 * the slab may touch the object after free or before allocation
2239 * then we should never poison the object itself.
2241 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2243 s
->flags
|= __OBJECT_POISON
;
2245 s
->flags
&= ~__OBJECT_POISON
;
2249 * If we are Redzoning then check if there is some space between the
2250 * end of the object and the free pointer. If not then add an
2251 * additional word to have some bytes to store Redzone information.
2253 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2254 size
+= sizeof(void *);
2258 * With that we have determined the number of bytes in actual use
2259 * by the object. This is the potential offset to the free pointer.
2263 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2266 * Relocate free pointer after the object if it is not
2267 * permitted to overwrite the first word of the object on
2270 * This is the case if we do RCU, have a constructor or
2271 * destructor or are poisoning the objects.
2274 size
+= sizeof(void *);
2277 #ifdef CONFIG_SLUB_DEBUG
2278 if (flags
& SLAB_STORE_USER
)
2280 * Need to store information about allocs and frees after
2283 size
+= 2 * sizeof(struct track
);
2285 if (flags
& SLAB_RED_ZONE
)
2287 * Add some empty padding so that we can catch
2288 * overwrites from earlier objects rather than let
2289 * tracking information or the free pointer be
2290 * corrupted if a user writes before the start
2293 size
+= sizeof(void *);
2297 * Determine the alignment based on various parameters that the
2298 * user specified and the dynamic determination of cache line size
2301 align
= calculate_alignment(flags
, align
, s
->objsize
);
2304 * SLUB stores one object immediately after another beginning from
2305 * offset 0. In order to align the objects we have to simply size
2306 * each object to conform to the alignment.
2308 size
= ALIGN(size
, align
);
2310 if (forced_order
>= 0)
2311 order
= forced_order
;
2313 order
= calculate_order(size
);
2320 s
->allocflags
|= __GFP_COMP
;
2322 if (s
->flags
& SLAB_CACHE_DMA
)
2323 s
->allocflags
|= SLUB_DMA
;
2325 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2326 s
->allocflags
|= __GFP_RECLAIMABLE
;
2329 * Determine the number of objects per slab
2331 s
->oo
= oo_make(order
, size
);
2332 s
->min
= oo_make(get_order(size
), size
);
2333 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2336 return !!oo_objects(s
->oo
);
2340 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2341 const char *name
, size_t size
,
2342 size_t align
, unsigned long flags
,
2343 void (*ctor
)(void *))
2345 memset(s
, 0, kmem_size
);
2350 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2352 if (!calculate_sizes(s
, -1))
2356 * The larger the object size is, the more pages we want on the partial
2357 * list to avoid pounding the page allocator excessively.
2359 set_min_partial(s
, ilog2(s
->size
));
2362 s
->remote_node_defrag_ratio
= 1000;
2364 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2367 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2369 free_kmem_cache_nodes(s
);
2371 if (flags
& SLAB_PANIC
)
2372 panic("Cannot create slab %s size=%lu realsize=%u "
2373 "order=%u offset=%u flags=%lx\n",
2374 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2380 * Check if a given pointer is valid
2382 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2386 page
= get_object_page(object
);
2388 if (!page
|| s
!= page
->slab
)
2389 /* No slab or wrong slab */
2392 if (!check_valid_pointer(s
, page
, object
))
2396 * We could also check if the object is on the slabs freelist.
2397 * But this would be too expensive and it seems that the main
2398 * purpose of kmem_ptr_valid() is to check if the object belongs
2399 * to a certain slab.
2403 EXPORT_SYMBOL(kmem_ptr_validate
);
2406 * Determine the size of a slab object
2408 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2412 EXPORT_SYMBOL(kmem_cache_size
);
2414 const char *kmem_cache_name(struct kmem_cache
*s
)
2418 EXPORT_SYMBOL(kmem_cache_name
);
2420 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2423 #ifdef CONFIG_SLUB_DEBUG
2424 void *addr
= page_address(page
);
2426 DECLARE_BITMAP(map
, page
->objects
);
2428 bitmap_zero(map
, page
->objects
);
2429 slab_err(s
, page
, "%s", text
);
2431 for_each_free_object(p
, s
, page
->freelist
)
2432 set_bit(slab_index(p
, s
, addr
), map
);
2434 for_each_object(p
, s
, addr
, page
->objects
) {
2436 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2437 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2439 print_tracking(s
, p
);
2447 * Attempt to free all partial slabs on a node.
2449 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2451 unsigned long flags
;
2452 struct page
*page
, *h
;
2454 spin_lock_irqsave(&n
->list_lock
, flags
);
2455 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2457 list_del(&page
->lru
);
2458 discard_slab(s
, page
);
2461 list_slab_objects(s
, page
,
2462 "Objects remaining on kmem_cache_close()");
2465 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2469 * Release all resources used by a slab cache.
2471 static inline int kmem_cache_close(struct kmem_cache
*s
)
2477 /* Attempt to free all objects */
2478 free_kmem_cache_cpus(s
);
2479 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2480 struct kmem_cache_node
*n
= get_node(s
, node
);
2483 if (n
->nr_partial
|| slabs_node(s
, node
))
2486 free_kmem_cache_nodes(s
);
2491 * Close a cache and release the kmem_cache structure
2492 * (must be used for caches created using kmem_cache_create)
2494 void kmem_cache_destroy(struct kmem_cache
*s
)
2496 down_write(&slub_lock
);
2500 up_write(&slub_lock
);
2501 if (kmem_cache_close(s
)) {
2502 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2503 "still has objects.\n", s
->name
, __func__
);
2506 sysfs_slab_remove(s
);
2508 up_write(&slub_lock
);
2510 EXPORT_SYMBOL(kmem_cache_destroy
);
2512 /********************************************************************
2514 *******************************************************************/
2516 struct kmem_cache kmalloc_caches
[SLUB_PAGE_SHIFT
] __cacheline_aligned
;
2517 EXPORT_SYMBOL(kmalloc_caches
);
2519 static int __init
setup_slub_min_order(char *str
)
2521 get_option(&str
, &slub_min_order
);
2526 __setup("slub_min_order=", setup_slub_min_order
);
2528 static int __init
setup_slub_max_order(char *str
)
2530 get_option(&str
, &slub_max_order
);
2531 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2536 __setup("slub_max_order=", setup_slub_max_order
);
2538 static int __init
setup_slub_min_objects(char *str
)
2540 get_option(&str
, &slub_min_objects
);
2545 __setup("slub_min_objects=", setup_slub_min_objects
);
2547 static int __init
setup_slub_nomerge(char *str
)
2553 __setup("slub_nomerge", setup_slub_nomerge
);
2555 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2556 const char *name
, int size
, gfp_t gfp_flags
)
2558 unsigned int flags
= 0;
2560 if (gfp_flags
& SLUB_DMA
)
2561 flags
= SLAB_CACHE_DMA
;
2564 * This function is called with IRQs disabled during early-boot on
2565 * single CPU so there's no need to take slub_lock here.
2567 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2571 list_add(&s
->list
, &slab_caches
);
2573 if (sysfs_slab_add(s
))
2578 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2581 #ifdef CONFIG_ZONE_DMA
2582 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2584 static void sysfs_add_func(struct work_struct
*w
)
2586 struct kmem_cache
*s
;
2588 down_write(&slub_lock
);
2589 list_for_each_entry(s
, &slab_caches
, list
) {
2590 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2591 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2595 up_write(&slub_lock
);
2598 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2600 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2602 struct kmem_cache
*s
;
2606 s
= kmalloc_caches_dma
[index
];
2610 /* Dynamically create dma cache */
2611 if (flags
& __GFP_WAIT
)
2612 down_write(&slub_lock
);
2614 if (!down_write_trylock(&slub_lock
))
2618 if (kmalloc_caches_dma
[index
])
2621 realsize
= kmalloc_caches
[index
].objsize
;
2622 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2623 (unsigned int)realsize
);
2624 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2626 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2627 realsize
, ARCH_KMALLOC_MINALIGN
,
2628 SLAB_CACHE_DMA
|__SYSFS_ADD_DEFERRED
, NULL
)) {
2634 list_add(&s
->list
, &slab_caches
);
2635 kmalloc_caches_dma
[index
] = s
;
2637 schedule_work(&sysfs_add_work
);
2640 up_write(&slub_lock
);
2642 return kmalloc_caches_dma
[index
];
2647 * Conversion table for small slabs sizes / 8 to the index in the
2648 * kmalloc array. This is necessary for slabs < 192 since we have non power
2649 * of two cache sizes there. The size of larger slabs can be determined using
2652 static s8 size_index
[24] = {
2679 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2685 return ZERO_SIZE_PTR
;
2687 index
= size_index
[(size
- 1) / 8];
2689 index
= fls(size
- 1);
2691 #ifdef CONFIG_ZONE_DMA
2692 if (unlikely((flags
& SLUB_DMA
)))
2693 return dma_kmalloc_cache(index
, flags
);
2696 return &kmalloc_caches
[index
];
2699 void *__kmalloc(size_t size
, gfp_t flags
)
2701 struct kmem_cache
*s
;
2704 if (unlikely(size
> SLUB_MAX_SIZE
))
2705 return kmalloc_large(size
, flags
);
2707 s
= get_slab(size
, flags
);
2709 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2712 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2714 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2718 EXPORT_SYMBOL(__kmalloc
);
2720 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2722 struct page
*page
= alloc_pages_node(node
, flags
| __GFP_COMP
,
2726 return page_address(page
);
2732 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2734 struct kmem_cache
*s
;
2737 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2738 ret
= kmalloc_large_node(size
, flags
, node
);
2740 trace_kmalloc_node(_RET_IP_
, ret
,
2741 size
, PAGE_SIZE
<< get_order(size
),
2747 s
= get_slab(size
, flags
);
2749 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2752 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2754 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2758 EXPORT_SYMBOL(__kmalloc_node
);
2761 size_t ksize(const void *object
)
2764 struct kmem_cache
*s
;
2766 if (unlikely(object
== ZERO_SIZE_PTR
))
2769 page
= virt_to_head_page(object
);
2771 if (unlikely(!PageSlab(page
))) {
2772 WARN_ON(!PageCompound(page
));
2773 return PAGE_SIZE
<< compound_order(page
);
2777 #ifdef CONFIG_SLUB_DEBUG
2779 * Debugging requires use of the padding between object
2780 * and whatever may come after it.
2782 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2787 * If we have the need to store the freelist pointer
2788 * back there or track user information then we can
2789 * only use the space before that information.
2791 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2794 * Else we can use all the padding etc for the allocation
2798 EXPORT_SYMBOL(ksize
);
2800 void kfree(const void *x
)
2803 void *object
= (void *)x
;
2805 trace_kfree(_RET_IP_
, x
);
2807 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2810 page
= virt_to_head_page(x
);
2811 if (unlikely(!PageSlab(page
))) {
2812 BUG_ON(!PageCompound(page
));
2816 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2818 EXPORT_SYMBOL(kfree
);
2821 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2822 * the remaining slabs by the number of items in use. The slabs with the
2823 * most items in use come first. New allocations will then fill those up
2824 * and thus they can be removed from the partial lists.
2826 * The slabs with the least items are placed last. This results in them
2827 * being allocated from last increasing the chance that the last objects
2828 * are freed in them.
2830 int kmem_cache_shrink(struct kmem_cache
*s
)
2834 struct kmem_cache_node
*n
;
2837 int objects
= oo_objects(s
->max
);
2838 struct list_head
*slabs_by_inuse
=
2839 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2840 unsigned long flags
;
2842 if (!slabs_by_inuse
)
2846 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2847 n
= get_node(s
, node
);
2852 for (i
= 0; i
< objects
; i
++)
2853 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2855 spin_lock_irqsave(&n
->list_lock
, flags
);
2858 * Build lists indexed by the items in use in each slab.
2860 * Note that concurrent frees may occur while we hold the
2861 * list_lock. page->inuse here is the upper limit.
2863 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2864 if (!page
->inuse
&& slab_trylock(page
)) {
2866 * Must hold slab lock here because slab_free
2867 * may have freed the last object and be
2868 * waiting to release the slab.
2870 list_del(&page
->lru
);
2873 discard_slab(s
, page
);
2875 list_move(&page
->lru
,
2876 slabs_by_inuse
+ page
->inuse
);
2881 * Rebuild the partial list with the slabs filled up most
2882 * first and the least used slabs at the end.
2884 for (i
= objects
- 1; i
>= 0; i
--)
2885 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2887 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2890 kfree(slabs_by_inuse
);
2893 EXPORT_SYMBOL(kmem_cache_shrink
);
2895 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2896 static int slab_mem_going_offline_callback(void *arg
)
2898 struct kmem_cache
*s
;
2900 down_read(&slub_lock
);
2901 list_for_each_entry(s
, &slab_caches
, list
)
2902 kmem_cache_shrink(s
);
2903 up_read(&slub_lock
);
2908 static void slab_mem_offline_callback(void *arg
)
2910 struct kmem_cache_node
*n
;
2911 struct kmem_cache
*s
;
2912 struct memory_notify
*marg
= arg
;
2915 offline_node
= marg
->status_change_nid
;
2918 * If the node still has available memory. we need kmem_cache_node
2921 if (offline_node
< 0)
2924 down_read(&slub_lock
);
2925 list_for_each_entry(s
, &slab_caches
, list
) {
2926 n
= get_node(s
, offline_node
);
2929 * if n->nr_slabs > 0, slabs still exist on the node
2930 * that is going down. We were unable to free them,
2931 * and offline_pages() function shoudn't call this
2932 * callback. So, we must fail.
2934 BUG_ON(slabs_node(s
, offline_node
));
2936 s
->node
[offline_node
] = NULL
;
2937 kmem_cache_free(kmalloc_caches
, n
);
2940 up_read(&slub_lock
);
2943 static int slab_mem_going_online_callback(void *arg
)
2945 struct kmem_cache_node
*n
;
2946 struct kmem_cache
*s
;
2947 struct memory_notify
*marg
= arg
;
2948 int nid
= marg
->status_change_nid
;
2952 * If the node's memory is already available, then kmem_cache_node is
2953 * already created. Nothing to do.
2959 * We are bringing a node online. No memory is available yet. We must
2960 * allocate a kmem_cache_node structure in order to bring the node
2963 down_read(&slub_lock
);
2964 list_for_each_entry(s
, &slab_caches
, list
) {
2966 * XXX: kmem_cache_alloc_node will fallback to other nodes
2967 * since memory is not yet available from the node that
2970 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2975 init_kmem_cache_node(n
, s
);
2979 up_read(&slub_lock
);
2983 static int slab_memory_callback(struct notifier_block
*self
,
2984 unsigned long action
, void *arg
)
2989 case MEM_GOING_ONLINE
:
2990 ret
= slab_mem_going_online_callback(arg
);
2992 case MEM_GOING_OFFLINE
:
2993 ret
= slab_mem_going_offline_callback(arg
);
2996 case MEM_CANCEL_ONLINE
:
2997 slab_mem_offline_callback(arg
);
3000 case MEM_CANCEL_OFFLINE
:
3004 ret
= notifier_from_errno(ret
);
3010 #endif /* CONFIG_MEMORY_HOTPLUG */
3012 /********************************************************************
3013 * Basic setup of slabs
3014 *******************************************************************/
3016 void __init
kmem_cache_init(void)
3025 * Must first have the slab cache available for the allocations of the
3026 * struct kmem_cache_node's. There is special bootstrap code in
3027 * kmem_cache_open for slab_state == DOWN.
3029 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3030 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3031 kmalloc_caches
[0].refcount
= -1;
3034 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3037 /* Able to allocate the per node structures */
3038 slab_state
= PARTIAL
;
3040 /* Caches that are not of the two-to-the-power-of size */
3041 if (KMALLOC_MIN_SIZE
<= 64) {
3042 create_kmalloc_cache(&kmalloc_caches
[1],
3043 "kmalloc-96", 96, GFP_NOWAIT
);
3045 create_kmalloc_cache(&kmalloc_caches
[2],
3046 "kmalloc-192", 192, GFP_NOWAIT
);
3050 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3051 create_kmalloc_cache(&kmalloc_caches
[i
],
3052 "kmalloc", 1 << i
, GFP_NOWAIT
);
3058 * Patch up the size_index table if we have strange large alignment
3059 * requirements for the kmalloc array. This is only the case for
3060 * MIPS it seems. The standard arches will not generate any code here.
3062 * Largest permitted alignment is 256 bytes due to the way we
3063 * handle the index determination for the smaller caches.
3065 * Make sure that nothing crazy happens if someone starts tinkering
3066 * around with ARCH_KMALLOC_MINALIGN
3068 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3069 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3071 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3072 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3074 if (KMALLOC_MIN_SIZE
== 128) {
3076 * The 192 byte sized cache is not used if the alignment
3077 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3080 for (i
= 128 + 8; i
<= 192; i
+= 8)
3081 size_index
[(i
- 1) / 8] = 8;
3086 /* Provide the correct kmalloc names now that the caches are up */
3087 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3088 kmalloc_caches
[i
]. name
=
3089 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3092 register_cpu_notifier(&slab_notifier
);
3093 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3094 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3096 kmem_size
= sizeof(struct kmem_cache
);
3100 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3101 " CPUs=%d, Nodes=%d\n",
3102 caches
, cache_line_size(),
3103 slub_min_order
, slub_max_order
, slub_min_objects
,
3104 nr_cpu_ids
, nr_node_ids
);
3108 * Find a mergeable slab cache
3110 static int slab_unmergeable(struct kmem_cache
*s
)
3112 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3119 * We may have set a slab to be unmergeable during bootstrap.
3121 if (s
->refcount
< 0)
3127 static struct kmem_cache
*find_mergeable(size_t size
,
3128 size_t align
, unsigned long flags
, const char *name
,
3129 void (*ctor
)(void *))
3131 struct kmem_cache
*s
;
3133 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3139 size
= ALIGN(size
, sizeof(void *));
3140 align
= calculate_alignment(flags
, align
, size
);
3141 size
= ALIGN(size
, align
);
3142 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3144 list_for_each_entry(s
, &slab_caches
, list
) {
3145 if (slab_unmergeable(s
))
3151 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3154 * Check if alignment is compatible.
3155 * Courtesy of Adrian Drzewiecki
3157 if ((s
->size
& ~(align
- 1)) != s
->size
)
3160 if (s
->size
- size
>= sizeof(void *))
3168 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3169 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3171 struct kmem_cache
*s
;
3173 down_write(&slub_lock
);
3174 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3180 * Adjust the object sizes so that we clear
3181 * the complete object on kzalloc.
3183 s
->objsize
= max(s
->objsize
, (int)size
);
3186 * And then we need to update the object size in the
3187 * per cpu structures
3189 for_each_online_cpu(cpu
)
3190 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3192 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3193 up_write(&slub_lock
);
3195 if (sysfs_slab_alias(s
, name
)) {
3196 down_write(&slub_lock
);
3198 up_write(&slub_lock
);
3204 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3206 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3207 size
, align
, flags
, ctor
)) {
3208 list_add(&s
->list
, &slab_caches
);
3209 up_write(&slub_lock
);
3210 if (sysfs_slab_add(s
)) {
3211 down_write(&slub_lock
);
3213 up_write(&slub_lock
);
3221 up_write(&slub_lock
);
3224 if (flags
& SLAB_PANIC
)
3225 panic("Cannot create slabcache %s\n", name
);
3230 EXPORT_SYMBOL(kmem_cache_create
);
3234 * Use the cpu notifier to insure that the cpu slabs are flushed when
3237 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3238 unsigned long action
, void *hcpu
)
3240 long cpu
= (long)hcpu
;
3241 struct kmem_cache
*s
;
3242 unsigned long flags
;
3245 case CPU_UP_PREPARE
:
3246 case CPU_UP_PREPARE_FROZEN
:
3247 init_alloc_cpu_cpu(cpu
);
3248 down_read(&slub_lock
);
3249 list_for_each_entry(s
, &slab_caches
, list
)
3250 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3252 up_read(&slub_lock
);
3255 case CPU_UP_CANCELED
:
3256 case CPU_UP_CANCELED_FROZEN
:
3258 case CPU_DEAD_FROZEN
:
3259 down_read(&slub_lock
);
3260 list_for_each_entry(s
, &slab_caches
, list
) {
3261 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3263 local_irq_save(flags
);
3264 __flush_cpu_slab(s
, cpu
);
3265 local_irq_restore(flags
);
3266 free_kmem_cache_cpu(c
, cpu
);
3267 s
->cpu_slab
[cpu
] = NULL
;
3269 up_read(&slub_lock
);
3277 static struct notifier_block __cpuinitdata slab_notifier
= {
3278 .notifier_call
= slab_cpuup_callback
3283 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3285 struct kmem_cache
*s
;
3288 if (unlikely(size
> SLUB_MAX_SIZE
))
3289 return kmalloc_large(size
, gfpflags
);
3291 s
= get_slab(size
, gfpflags
);
3293 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3296 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3298 /* Honor the call site pointer we recieved. */
3299 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3304 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3305 int node
, unsigned long caller
)
3307 struct kmem_cache
*s
;
3310 if (unlikely(size
> SLUB_MAX_SIZE
))
3311 return kmalloc_large_node(size
, gfpflags
, node
);
3313 s
= get_slab(size
, gfpflags
);
3315 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3318 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3320 /* Honor the call site pointer we recieved. */
3321 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3326 #ifdef CONFIG_SLUB_DEBUG
3327 static unsigned long count_partial(struct kmem_cache_node
*n
,
3328 int (*get_count
)(struct page
*))
3330 unsigned long flags
;
3331 unsigned long x
= 0;
3334 spin_lock_irqsave(&n
->list_lock
, flags
);
3335 list_for_each_entry(page
, &n
->partial
, lru
)
3336 x
+= get_count(page
);
3337 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3341 static int count_inuse(struct page
*page
)
3346 static int count_total(struct page
*page
)
3348 return page
->objects
;
3351 static int count_free(struct page
*page
)
3353 return page
->objects
- page
->inuse
;
3356 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3360 void *addr
= page_address(page
);
3362 if (!check_slab(s
, page
) ||
3363 !on_freelist(s
, page
, NULL
))
3366 /* Now we know that a valid freelist exists */
3367 bitmap_zero(map
, page
->objects
);
3369 for_each_free_object(p
, s
, page
->freelist
) {
3370 set_bit(slab_index(p
, s
, addr
), map
);
3371 if (!check_object(s
, page
, p
, 0))
3375 for_each_object(p
, s
, addr
, page
->objects
)
3376 if (!test_bit(slab_index(p
, s
, addr
), map
))
3377 if (!check_object(s
, page
, p
, 1))
3382 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3385 if (slab_trylock(page
)) {
3386 validate_slab(s
, page
, map
);
3389 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3392 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3393 if (!PageSlubDebug(page
))
3394 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3395 "on slab 0x%p\n", s
->name
, page
);
3397 if (PageSlubDebug(page
))
3398 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3399 "slab 0x%p\n", s
->name
, page
);
3403 static int validate_slab_node(struct kmem_cache
*s
,
3404 struct kmem_cache_node
*n
, unsigned long *map
)
3406 unsigned long count
= 0;
3408 unsigned long flags
;
3410 spin_lock_irqsave(&n
->list_lock
, flags
);
3412 list_for_each_entry(page
, &n
->partial
, lru
) {
3413 validate_slab_slab(s
, page
, map
);
3416 if (count
!= n
->nr_partial
)
3417 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3418 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3420 if (!(s
->flags
& SLAB_STORE_USER
))
3423 list_for_each_entry(page
, &n
->full
, lru
) {
3424 validate_slab_slab(s
, page
, map
);
3427 if (count
!= atomic_long_read(&n
->nr_slabs
))
3428 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3429 "counter=%ld\n", s
->name
, count
,
3430 atomic_long_read(&n
->nr_slabs
));
3433 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3437 static long validate_slab_cache(struct kmem_cache
*s
)
3440 unsigned long count
= 0;
3441 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3442 sizeof(unsigned long), GFP_KERNEL
);
3448 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3449 struct kmem_cache_node
*n
= get_node(s
, node
);
3451 count
+= validate_slab_node(s
, n
, map
);
3457 #ifdef SLUB_RESILIENCY_TEST
3458 static void resiliency_test(void)
3462 printk(KERN_ERR
"SLUB resiliency testing\n");
3463 printk(KERN_ERR
"-----------------------\n");
3464 printk(KERN_ERR
"A. Corruption after allocation\n");
3466 p
= kzalloc(16, GFP_KERNEL
);
3468 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3469 " 0x12->0x%p\n\n", p
+ 16);
3471 validate_slab_cache(kmalloc_caches
+ 4);
3473 /* Hmmm... The next two are dangerous */
3474 p
= kzalloc(32, GFP_KERNEL
);
3475 p
[32 + sizeof(void *)] = 0x34;
3476 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3477 " 0x34 -> -0x%p\n", p
);
3479 "If allocated object is overwritten then not detectable\n\n");
3481 validate_slab_cache(kmalloc_caches
+ 5);
3482 p
= kzalloc(64, GFP_KERNEL
);
3483 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3485 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3488 "If allocated object is overwritten then not detectable\n\n");
3489 validate_slab_cache(kmalloc_caches
+ 6);
3491 printk(KERN_ERR
"\nB. Corruption after free\n");
3492 p
= kzalloc(128, GFP_KERNEL
);
3495 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3496 validate_slab_cache(kmalloc_caches
+ 7);
3498 p
= kzalloc(256, GFP_KERNEL
);
3501 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3503 validate_slab_cache(kmalloc_caches
+ 8);
3505 p
= kzalloc(512, GFP_KERNEL
);
3508 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3509 validate_slab_cache(kmalloc_caches
+ 9);
3512 static void resiliency_test(void) {};
3516 * Generate lists of code addresses where slabcache objects are allocated
3521 unsigned long count
;
3528 DECLARE_BITMAP(cpus
, NR_CPUS
);
3534 unsigned long count
;
3535 struct location
*loc
;
3538 static void free_loc_track(struct loc_track
*t
)
3541 free_pages((unsigned long)t
->loc
,
3542 get_order(sizeof(struct location
) * t
->max
));
3545 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3550 order
= get_order(sizeof(struct location
) * max
);
3552 l
= (void *)__get_free_pages(flags
, order
);
3557 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3565 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3566 const struct track
*track
)
3568 long start
, end
, pos
;
3570 unsigned long caddr
;
3571 unsigned long age
= jiffies
- track
->when
;
3577 pos
= start
+ (end
- start
+ 1) / 2;
3580 * There is nothing at "end". If we end up there
3581 * we need to add something to before end.
3586 caddr
= t
->loc
[pos
].addr
;
3587 if (track
->addr
== caddr
) {
3593 if (age
< l
->min_time
)
3595 if (age
> l
->max_time
)
3598 if (track
->pid
< l
->min_pid
)
3599 l
->min_pid
= track
->pid
;
3600 if (track
->pid
> l
->max_pid
)
3601 l
->max_pid
= track
->pid
;
3603 cpumask_set_cpu(track
->cpu
,
3604 to_cpumask(l
->cpus
));
3606 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3610 if (track
->addr
< caddr
)
3617 * Not found. Insert new tracking element.
3619 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3625 (t
->count
- pos
) * sizeof(struct location
));
3628 l
->addr
= track
->addr
;
3632 l
->min_pid
= track
->pid
;
3633 l
->max_pid
= track
->pid
;
3634 cpumask_clear(to_cpumask(l
->cpus
));
3635 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3636 nodes_clear(l
->nodes
);
3637 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3641 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3642 struct page
*page
, enum track_item alloc
)
3644 void *addr
= page_address(page
);
3645 DECLARE_BITMAP(map
, page
->objects
);
3648 bitmap_zero(map
, page
->objects
);
3649 for_each_free_object(p
, s
, page
->freelist
)
3650 set_bit(slab_index(p
, s
, addr
), map
);
3652 for_each_object(p
, s
, addr
, page
->objects
)
3653 if (!test_bit(slab_index(p
, s
, addr
), map
))
3654 add_location(t
, s
, get_track(s
, p
, alloc
));
3657 static int list_locations(struct kmem_cache
*s
, char *buf
,
3658 enum track_item alloc
)
3662 struct loc_track t
= { 0, 0, NULL
};
3665 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3667 return sprintf(buf
, "Out of memory\n");
3669 /* Push back cpu slabs */
3672 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3673 struct kmem_cache_node
*n
= get_node(s
, node
);
3674 unsigned long flags
;
3677 if (!atomic_long_read(&n
->nr_slabs
))
3680 spin_lock_irqsave(&n
->list_lock
, flags
);
3681 list_for_each_entry(page
, &n
->partial
, lru
)
3682 process_slab(&t
, s
, page
, alloc
);
3683 list_for_each_entry(page
, &n
->full
, lru
)
3684 process_slab(&t
, s
, page
, alloc
);
3685 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3688 for (i
= 0; i
< t
.count
; i
++) {
3689 struct location
*l
= &t
.loc
[i
];
3691 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3693 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3696 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3698 len
+= sprintf(buf
+ len
, "<not-available>");
3700 if (l
->sum_time
!= l
->min_time
) {
3701 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3703 (long)div_u64(l
->sum_time
, l
->count
),
3706 len
+= sprintf(buf
+ len
, " age=%ld",
3709 if (l
->min_pid
!= l
->max_pid
)
3710 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3711 l
->min_pid
, l
->max_pid
);
3713 len
+= sprintf(buf
+ len
, " pid=%ld",
3716 if (num_online_cpus() > 1 &&
3717 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3718 len
< PAGE_SIZE
- 60) {
3719 len
+= sprintf(buf
+ len
, " cpus=");
3720 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3721 to_cpumask(l
->cpus
));
3724 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3725 len
< PAGE_SIZE
- 60) {
3726 len
+= sprintf(buf
+ len
, " nodes=");
3727 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3731 len
+= sprintf(buf
+ len
, "\n");
3736 len
+= sprintf(buf
, "No data\n");
3740 enum slab_stat_type
{
3741 SL_ALL
, /* All slabs */
3742 SL_PARTIAL
, /* Only partially allocated slabs */
3743 SL_CPU
, /* Only slabs used for cpu caches */
3744 SL_OBJECTS
, /* Determine allocated objects not slabs */
3745 SL_TOTAL
/* Determine object capacity not slabs */
3748 #define SO_ALL (1 << SL_ALL)
3749 #define SO_PARTIAL (1 << SL_PARTIAL)
3750 #define SO_CPU (1 << SL_CPU)
3751 #define SO_OBJECTS (1 << SL_OBJECTS)
3752 #define SO_TOTAL (1 << SL_TOTAL)
3754 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3755 char *buf
, unsigned long flags
)
3757 unsigned long total
= 0;
3760 unsigned long *nodes
;
3761 unsigned long *per_cpu
;
3763 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3766 per_cpu
= nodes
+ nr_node_ids
;
3768 if (flags
& SO_CPU
) {
3771 for_each_possible_cpu(cpu
) {
3772 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3774 if (!c
|| c
->node
< 0)
3778 if (flags
& SO_TOTAL
)
3779 x
= c
->page
->objects
;
3780 else if (flags
& SO_OBJECTS
)
3786 nodes
[c
->node
] += x
;
3792 if (flags
& SO_ALL
) {
3793 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3794 struct kmem_cache_node
*n
= get_node(s
, node
);
3796 if (flags
& SO_TOTAL
)
3797 x
= atomic_long_read(&n
->total_objects
);
3798 else if (flags
& SO_OBJECTS
)
3799 x
= atomic_long_read(&n
->total_objects
) -
3800 count_partial(n
, count_free
);
3803 x
= atomic_long_read(&n
->nr_slabs
);
3808 } else if (flags
& SO_PARTIAL
) {
3809 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3810 struct kmem_cache_node
*n
= get_node(s
, node
);
3812 if (flags
& SO_TOTAL
)
3813 x
= count_partial(n
, count_total
);
3814 else if (flags
& SO_OBJECTS
)
3815 x
= count_partial(n
, count_inuse
);
3822 x
= sprintf(buf
, "%lu", total
);
3824 for_each_node_state(node
, N_NORMAL_MEMORY
)
3826 x
+= sprintf(buf
+ x
, " N%d=%lu",
3830 return x
+ sprintf(buf
+ x
, "\n");
3833 static int any_slab_objects(struct kmem_cache
*s
)
3837 for_each_online_node(node
) {
3838 struct kmem_cache_node
*n
= get_node(s
, node
);
3843 if (atomic_long_read(&n
->total_objects
))
3849 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3850 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3852 struct slab_attribute
{
3853 struct attribute attr
;
3854 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3855 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3858 #define SLAB_ATTR_RO(_name) \
3859 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3861 #define SLAB_ATTR(_name) \
3862 static struct slab_attribute _name##_attr = \
3863 __ATTR(_name, 0644, _name##_show, _name##_store)
3865 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3867 return sprintf(buf
, "%d\n", s
->size
);
3869 SLAB_ATTR_RO(slab_size
);
3871 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3873 return sprintf(buf
, "%d\n", s
->align
);
3875 SLAB_ATTR_RO(align
);
3877 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3879 return sprintf(buf
, "%d\n", s
->objsize
);
3881 SLAB_ATTR_RO(object_size
);
3883 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3885 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3887 SLAB_ATTR_RO(objs_per_slab
);
3889 static ssize_t
order_store(struct kmem_cache
*s
,
3890 const char *buf
, size_t length
)
3892 unsigned long order
;
3895 err
= strict_strtoul(buf
, 10, &order
);
3899 if (order
> slub_max_order
|| order
< slub_min_order
)
3902 calculate_sizes(s
, order
);
3906 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3908 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3912 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3914 return sprintf(buf
, "%lu\n", s
->min_partial
);
3917 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3923 err
= strict_strtoul(buf
, 10, &min
);
3927 set_min_partial(s
, min
);
3930 SLAB_ATTR(min_partial
);
3932 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3935 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3937 return n
+ sprintf(buf
+ n
, "\n");
3943 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3945 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3947 SLAB_ATTR_RO(aliases
);
3949 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3951 return show_slab_objects(s
, buf
, SO_ALL
);
3953 SLAB_ATTR_RO(slabs
);
3955 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3957 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3959 SLAB_ATTR_RO(partial
);
3961 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3963 return show_slab_objects(s
, buf
, SO_CPU
);
3965 SLAB_ATTR_RO(cpu_slabs
);
3967 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3969 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3971 SLAB_ATTR_RO(objects
);
3973 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3975 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3977 SLAB_ATTR_RO(objects_partial
);
3979 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3981 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3983 SLAB_ATTR_RO(total_objects
);
3985 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3987 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3990 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3991 const char *buf
, size_t length
)
3993 s
->flags
&= ~SLAB_DEBUG_FREE
;
3995 s
->flags
|= SLAB_DEBUG_FREE
;
3998 SLAB_ATTR(sanity_checks
);
4000 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4002 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4005 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4008 s
->flags
&= ~SLAB_TRACE
;
4010 s
->flags
|= SLAB_TRACE
;
4015 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4017 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4020 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4021 const char *buf
, size_t length
)
4023 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4025 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4028 SLAB_ATTR(reclaim_account
);
4030 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4032 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4034 SLAB_ATTR_RO(hwcache_align
);
4036 #ifdef CONFIG_ZONE_DMA
4037 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4039 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4041 SLAB_ATTR_RO(cache_dma
);
4044 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4046 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4048 SLAB_ATTR_RO(destroy_by_rcu
);
4050 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4052 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4055 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4056 const char *buf
, size_t length
)
4058 if (any_slab_objects(s
))
4061 s
->flags
&= ~SLAB_RED_ZONE
;
4063 s
->flags
|= SLAB_RED_ZONE
;
4064 calculate_sizes(s
, -1);
4067 SLAB_ATTR(red_zone
);
4069 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4071 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4074 static ssize_t
poison_store(struct kmem_cache
*s
,
4075 const char *buf
, size_t length
)
4077 if (any_slab_objects(s
))
4080 s
->flags
&= ~SLAB_POISON
;
4082 s
->flags
|= SLAB_POISON
;
4083 calculate_sizes(s
, -1);
4088 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4090 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4093 static ssize_t
store_user_store(struct kmem_cache
*s
,
4094 const char *buf
, size_t length
)
4096 if (any_slab_objects(s
))
4099 s
->flags
&= ~SLAB_STORE_USER
;
4101 s
->flags
|= SLAB_STORE_USER
;
4102 calculate_sizes(s
, -1);
4105 SLAB_ATTR(store_user
);
4107 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4112 static ssize_t
validate_store(struct kmem_cache
*s
,
4113 const char *buf
, size_t length
)
4117 if (buf
[0] == '1') {
4118 ret
= validate_slab_cache(s
);
4124 SLAB_ATTR(validate
);
4126 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4131 static ssize_t
shrink_store(struct kmem_cache
*s
,
4132 const char *buf
, size_t length
)
4134 if (buf
[0] == '1') {
4135 int rc
= kmem_cache_shrink(s
);
4145 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4147 if (!(s
->flags
& SLAB_STORE_USER
))
4149 return list_locations(s
, buf
, TRACK_ALLOC
);
4151 SLAB_ATTR_RO(alloc_calls
);
4153 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4155 if (!(s
->flags
& SLAB_STORE_USER
))
4157 return list_locations(s
, buf
, TRACK_FREE
);
4159 SLAB_ATTR_RO(free_calls
);
4162 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4164 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4167 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4168 const char *buf
, size_t length
)
4170 unsigned long ratio
;
4173 err
= strict_strtoul(buf
, 10, &ratio
);
4178 s
->remote_node_defrag_ratio
= ratio
* 10;
4182 SLAB_ATTR(remote_node_defrag_ratio
);
4185 #ifdef CONFIG_SLUB_STATS
4186 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4188 unsigned long sum
= 0;
4191 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4196 for_each_online_cpu(cpu
) {
4197 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4203 len
= sprintf(buf
, "%lu", sum
);
4206 for_each_online_cpu(cpu
) {
4207 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4208 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4212 return len
+ sprintf(buf
+ len
, "\n");
4215 #define STAT_ATTR(si, text) \
4216 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4218 return show_stat(s, buf, si); \
4220 SLAB_ATTR_RO(text); \
4222 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4223 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4224 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4225 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4226 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4227 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4228 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4229 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4230 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4231 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4232 STAT_ATTR(FREE_SLAB
, free_slab
);
4233 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4234 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4235 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4236 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4237 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4238 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4239 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4242 static struct attribute
*slab_attrs
[] = {
4243 &slab_size_attr
.attr
,
4244 &object_size_attr
.attr
,
4245 &objs_per_slab_attr
.attr
,
4247 &min_partial_attr
.attr
,
4249 &objects_partial_attr
.attr
,
4250 &total_objects_attr
.attr
,
4253 &cpu_slabs_attr
.attr
,
4257 &sanity_checks_attr
.attr
,
4259 &hwcache_align_attr
.attr
,
4260 &reclaim_account_attr
.attr
,
4261 &destroy_by_rcu_attr
.attr
,
4262 &red_zone_attr
.attr
,
4264 &store_user_attr
.attr
,
4265 &validate_attr
.attr
,
4267 &alloc_calls_attr
.attr
,
4268 &free_calls_attr
.attr
,
4269 #ifdef CONFIG_ZONE_DMA
4270 &cache_dma_attr
.attr
,
4273 &remote_node_defrag_ratio_attr
.attr
,
4275 #ifdef CONFIG_SLUB_STATS
4276 &alloc_fastpath_attr
.attr
,
4277 &alloc_slowpath_attr
.attr
,
4278 &free_fastpath_attr
.attr
,
4279 &free_slowpath_attr
.attr
,
4280 &free_frozen_attr
.attr
,
4281 &free_add_partial_attr
.attr
,
4282 &free_remove_partial_attr
.attr
,
4283 &alloc_from_partial_attr
.attr
,
4284 &alloc_slab_attr
.attr
,
4285 &alloc_refill_attr
.attr
,
4286 &free_slab_attr
.attr
,
4287 &cpuslab_flush_attr
.attr
,
4288 &deactivate_full_attr
.attr
,
4289 &deactivate_empty_attr
.attr
,
4290 &deactivate_to_head_attr
.attr
,
4291 &deactivate_to_tail_attr
.attr
,
4292 &deactivate_remote_frees_attr
.attr
,
4293 &order_fallback_attr
.attr
,
4298 static struct attribute_group slab_attr_group
= {
4299 .attrs
= slab_attrs
,
4302 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4303 struct attribute
*attr
,
4306 struct slab_attribute
*attribute
;
4307 struct kmem_cache
*s
;
4310 attribute
= to_slab_attr(attr
);
4313 if (!attribute
->show
)
4316 err
= attribute
->show(s
, buf
);
4321 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4322 struct attribute
*attr
,
4323 const char *buf
, size_t len
)
4325 struct slab_attribute
*attribute
;
4326 struct kmem_cache
*s
;
4329 attribute
= to_slab_attr(attr
);
4332 if (!attribute
->store
)
4335 err
= attribute
->store(s
, buf
, len
);
4340 static void kmem_cache_release(struct kobject
*kobj
)
4342 struct kmem_cache
*s
= to_slab(kobj
);
4347 static struct sysfs_ops slab_sysfs_ops
= {
4348 .show
= slab_attr_show
,
4349 .store
= slab_attr_store
,
4352 static struct kobj_type slab_ktype
= {
4353 .sysfs_ops
= &slab_sysfs_ops
,
4354 .release
= kmem_cache_release
4357 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4359 struct kobj_type
*ktype
= get_ktype(kobj
);
4361 if (ktype
== &slab_ktype
)
4366 static struct kset_uevent_ops slab_uevent_ops
= {
4367 .filter
= uevent_filter
,
4370 static struct kset
*slab_kset
;
4372 #define ID_STR_LENGTH 64
4374 /* Create a unique string id for a slab cache:
4376 * Format :[flags-]size
4378 static char *create_unique_id(struct kmem_cache
*s
)
4380 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4387 * First flags affecting slabcache operations. We will only
4388 * get here for aliasable slabs so we do not need to support
4389 * too many flags. The flags here must cover all flags that
4390 * are matched during merging to guarantee that the id is
4393 if (s
->flags
& SLAB_CACHE_DMA
)
4395 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4397 if (s
->flags
& SLAB_DEBUG_FREE
)
4401 p
+= sprintf(p
, "%07d", s
->size
);
4402 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4406 static int sysfs_slab_add(struct kmem_cache
*s
)
4412 if (slab_state
< SYSFS
)
4413 /* Defer until later */
4416 unmergeable
= slab_unmergeable(s
);
4419 * Slabcache can never be merged so we can use the name proper.
4420 * This is typically the case for debug situations. In that
4421 * case we can catch duplicate names easily.
4423 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4427 * Create a unique name for the slab as a target
4430 name
= create_unique_id(s
);
4433 s
->kobj
.kset
= slab_kset
;
4434 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4436 kobject_put(&s
->kobj
);
4440 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4443 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4445 /* Setup first alias */
4446 sysfs_slab_alias(s
, s
->name
);
4452 static void sysfs_slab_remove(struct kmem_cache
*s
)
4454 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4455 kobject_del(&s
->kobj
);
4456 kobject_put(&s
->kobj
);
4460 * Need to buffer aliases during bootup until sysfs becomes
4461 * available lest we lose that information.
4463 struct saved_alias
{
4464 struct kmem_cache
*s
;
4466 struct saved_alias
*next
;
4469 static struct saved_alias
*alias_list
;
4471 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4473 struct saved_alias
*al
;
4475 if (slab_state
== SYSFS
) {
4477 * If we have a leftover link then remove it.
4479 sysfs_remove_link(&slab_kset
->kobj
, name
);
4480 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4483 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4489 al
->next
= alias_list
;
4494 static int __init
slab_sysfs_init(void)
4496 struct kmem_cache
*s
;
4499 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4501 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4507 list_for_each_entry(s
, &slab_caches
, list
) {
4508 err
= sysfs_slab_add(s
);
4510 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4511 " to sysfs\n", s
->name
);
4514 while (alias_list
) {
4515 struct saved_alias
*al
= alias_list
;
4517 alias_list
= alias_list
->next
;
4518 err
= sysfs_slab_alias(al
->s
, al
->name
);
4520 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4521 " %s to sysfs\n", s
->name
);
4529 __initcall(slab_sysfs_init
);
4533 * The /proc/slabinfo ABI
4535 #ifdef CONFIG_SLABINFO
4536 static void print_slabinfo_header(struct seq_file
*m
)
4538 seq_puts(m
, "slabinfo - version: 2.1\n");
4539 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4540 "<objperslab> <pagesperslab>");
4541 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4542 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4546 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4550 down_read(&slub_lock
);
4552 print_slabinfo_header(m
);
4554 return seq_list_start(&slab_caches
, *pos
);
4557 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4559 return seq_list_next(p
, &slab_caches
, pos
);
4562 static void s_stop(struct seq_file
*m
, void *p
)
4564 up_read(&slub_lock
);
4567 static int s_show(struct seq_file
*m
, void *p
)
4569 unsigned long nr_partials
= 0;
4570 unsigned long nr_slabs
= 0;
4571 unsigned long nr_inuse
= 0;
4572 unsigned long nr_objs
= 0;
4573 unsigned long nr_free
= 0;
4574 struct kmem_cache
*s
;
4577 s
= list_entry(p
, struct kmem_cache
, list
);
4579 for_each_online_node(node
) {
4580 struct kmem_cache_node
*n
= get_node(s
, node
);
4585 nr_partials
+= n
->nr_partial
;
4586 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4587 nr_objs
+= atomic_long_read(&n
->total_objects
);
4588 nr_free
+= count_partial(n
, count_free
);
4591 nr_inuse
= nr_objs
- nr_free
;
4593 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4594 nr_objs
, s
->size
, oo_objects(s
->oo
),
4595 (1 << oo_order(s
->oo
)));
4596 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4597 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4603 static const struct seq_operations slabinfo_op
= {
4610 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4612 return seq_open(file
, &slabinfo_op
);
4615 static const struct file_operations proc_slabinfo_operations
= {
4616 .open
= slabinfo_open
,
4618 .llseek
= seq_lseek
,
4619 .release
= seq_release
,
4622 static int __init
slab_proc_init(void)
4624 proc_create("slabinfo",S_IWUSR
|S_IRUGO
,NULL
,&proc_slabinfo_operations
);
4627 module_init(slab_proc_init
);
4628 #endif /* CONFIG_SLABINFO */