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/kmemcheck.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/mempolicy.h>
24 #include <linux/ctype.h>
25 #include <linux/debugobjects.h>
26 #include <linux/kallsyms.h>
27 #include <linux/memory.h>
28 #include <linux/math64.h>
29 #include <linux/fault-inject.h>
36 * The slab_lock protects operations on the object of a particular
37 * slab and its metadata in the page struct. If the slab lock
38 * has been taken then no allocations nor frees can be performed
39 * on the objects in the slab nor can the slab be added or removed
40 * from the partial or full lists since this would mean modifying
41 * the page_struct of the slab.
43 * The list_lock protects the partial and full list on each node and
44 * the partial slab counter. If taken then no new slabs may be added or
45 * removed from the lists nor make the number of partial slabs be modified.
46 * (Note that the total number of slabs is an atomic value that may be
47 * modified without taking the list lock).
49 * The list_lock is a centralized lock and thus we avoid taking it as
50 * much as possible. As long as SLUB does not have to handle partial
51 * slabs, operations can continue without any centralized lock. F.e.
52 * allocating a long series of objects that fill up slabs does not require
55 * The lock order is sometimes inverted when we are trying to get a slab
56 * off a list. We take the list_lock and then look for a page on the list
57 * to use. While we do that objects in the slabs may be freed. We can
58 * only operate on the slab if we have also taken the slab_lock. So we use
59 * a slab_trylock() on the slab. If trylock was successful then no frees
60 * can occur anymore and we can use the slab for allocations etc. If the
61 * slab_trylock() does not succeed then frees are in progress in the slab and
62 * we must stay away from it for a while since we may cause a bouncing
63 * cacheline if we try to acquire the lock. So go onto the next slab.
64 * If all pages are busy then we may allocate a new slab instead of reusing
65 * a partial slab. A new slab has noone operating on it and thus there is
66 * no danger of cacheline contention.
68 * Interrupts are disabled during allocation and deallocation in order to
69 * make the slab allocator safe to use in the context of an irq. In addition
70 * interrupts are disabled to ensure that the processor does not change
71 * while handling per_cpu slabs, due to kernel preemption.
73 * SLUB assigns one slab for allocation to each processor.
74 * Allocations only occur from these slabs called cpu slabs.
76 * Slabs with free elements are kept on a partial list and during regular
77 * operations no list for full slabs is used. If an object in a full slab is
78 * freed then the slab will show up again on the partial lists.
79 * We track full slabs for debugging purposes though because otherwise we
80 * cannot scan all objects.
82 * Slabs are freed when they become empty. Teardown and setup is
83 * minimal so we rely on the page allocators per cpu caches for
84 * fast frees and allocs.
86 * Overloading of page flags that are otherwise used for LRU management.
88 * PageActive The slab is frozen and exempt from list processing.
89 * This means that the slab is dedicated to a purpose
90 * such as satisfying allocations for a specific
91 * processor. Objects may be freed in the slab while
92 * it is frozen but slab_free will then skip the usual
93 * list operations. It is up to the processor holding
94 * the slab to integrate the slab into the slab lists
95 * when the slab is no longer needed.
97 * One use of this flag is to mark slabs that are
98 * used for allocations. Then such a slab becomes a cpu
99 * slab. The cpu slab may be equipped with an additional
100 * freelist that allows lockless access to
101 * free objects in addition to the regular freelist
102 * that requires the slab lock.
104 * PageError Slab requires special handling due to debug
105 * options set. This moves slab handling out of
106 * the fast path and disables lockless freelists.
109 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
110 SLAB_TRACE | SLAB_DEBUG_FREE)
112 static inline int kmem_cache_debug(struct kmem_cache
*s
)
114 #ifdef CONFIG_SLUB_DEBUG
115 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
122 * Issues still to be resolved:
124 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
126 * - Variable sizing of the per node arrays
129 /* Enable to test recovery from slab corruption on boot */
130 #undef SLUB_RESILIENCY_TEST
133 * Mininum number of partial slabs. These will be left on the partial
134 * lists even if they are empty. kmem_cache_shrink may reclaim them.
136 #define MIN_PARTIAL 5
139 * Maximum number of desirable partial slabs.
140 * The existence of more partial slabs makes kmem_cache_shrink
141 * sort the partial list by the number of objects in the.
143 #define MAX_PARTIAL 10
145 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
146 SLAB_POISON | SLAB_STORE_USER)
149 * Debugging flags that require metadata to be stored in the slab. These get
150 * disabled when slub_debug=O is used and a cache's min order increases with
153 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
156 * Set of flags that will prevent slab merging
158 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
159 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
162 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
163 SLAB_CACHE_DMA | SLAB_NOTRACK)
166 #define OO_MASK ((1 << OO_SHIFT) - 1)
167 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
169 /* Internal SLUB flags */
170 #define __OBJECT_POISON 0x80000000UL /* Poison object */
172 static int kmem_size
= sizeof(struct kmem_cache
);
175 static struct notifier_block slab_notifier
;
179 DOWN
, /* No slab functionality available */
180 PARTIAL
, /* Kmem_cache_node works */
181 UP
, /* Everything works but does not show up in sysfs */
185 /* A list of all slab caches on the system */
186 static DECLARE_RWSEM(slub_lock
);
187 static LIST_HEAD(slab_caches
);
190 * Tracking user of a slab.
193 unsigned long addr
; /* Called from address */
194 int cpu
; /* Was running on cpu */
195 int pid
; /* Pid context */
196 unsigned long when
; /* When did the operation occur */
199 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
201 #ifdef CONFIG_SLUB_DEBUG
202 static int sysfs_slab_add(struct kmem_cache
*);
203 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
204 static void sysfs_slab_remove(struct kmem_cache
*);
207 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
208 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
210 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
217 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
219 #ifdef CONFIG_SLUB_STATS
220 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
224 /********************************************************************
225 * Core slab cache functions
226 *******************************************************************/
228 int slab_is_available(void)
230 return slab_state
>= UP
;
233 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
236 return s
->node
[node
];
238 return &s
->local_node
;
242 /* Verify that a pointer has an address that is valid within a slab page */
243 static inline int check_valid_pointer(struct kmem_cache
*s
,
244 struct page
*page
, const void *object
)
251 base
= page_address(page
);
252 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
253 (object
- base
) % s
->size
) {
260 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
262 return *(void **)(object
+ s
->offset
);
265 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
267 *(void **)(object
+ s
->offset
) = fp
;
270 /* Loop over all objects in a slab */
271 #define for_each_object(__p, __s, __addr, __objects) \
272 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
276 #define for_each_free_object(__p, __s, __free) \
277 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
279 /* Determine object index from a given position */
280 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
282 return (p
- addr
) / s
->size
;
285 static inline struct kmem_cache_order_objects
oo_make(int order
,
288 struct kmem_cache_order_objects x
= {
289 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
295 static inline int oo_order(struct kmem_cache_order_objects x
)
297 return x
.x
>> OO_SHIFT
;
300 static inline int oo_objects(struct kmem_cache_order_objects x
)
302 return x
.x
& OO_MASK
;
305 #ifdef CONFIG_SLUB_DEBUG
309 #ifdef CONFIG_SLUB_DEBUG_ON
310 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
312 static int slub_debug
;
315 static char *slub_debug_slabs
;
316 static int disable_higher_order_debug
;
321 static void print_section(char *text
, u8
*addr
, unsigned int length
)
329 for (i
= 0; i
< length
; i
++) {
331 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
334 printk(KERN_CONT
" %02x", addr
[i
]);
336 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
338 printk(KERN_CONT
" %s\n", ascii
);
345 printk(KERN_CONT
" ");
349 printk(KERN_CONT
" %s\n", ascii
);
353 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
354 enum track_item alloc
)
359 p
= object
+ s
->offset
+ sizeof(void *);
361 p
= object
+ s
->inuse
;
366 static void set_track(struct kmem_cache
*s
, void *object
,
367 enum track_item alloc
, unsigned long addr
)
369 struct track
*p
= get_track(s
, object
, alloc
);
373 p
->cpu
= smp_processor_id();
374 p
->pid
= current
->pid
;
377 memset(p
, 0, sizeof(struct track
));
380 static void init_tracking(struct kmem_cache
*s
, void *object
)
382 if (!(s
->flags
& SLAB_STORE_USER
))
385 set_track(s
, object
, TRACK_FREE
, 0UL);
386 set_track(s
, object
, TRACK_ALLOC
, 0UL);
389 static void print_track(const char *s
, struct track
*t
)
394 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
395 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
398 static void print_tracking(struct kmem_cache
*s
, void *object
)
400 if (!(s
->flags
& SLAB_STORE_USER
))
403 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
404 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
407 static void print_page_info(struct page
*page
)
409 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
410 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
414 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
420 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
422 printk(KERN_ERR
"========================================"
423 "=====================================\n");
424 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
425 printk(KERN_ERR
"----------------------------------------"
426 "-------------------------------------\n\n");
429 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
435 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
437 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
440 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
442 unsigned int off
; /* Offset of last byte */
443 u8
*addr
= page_address(page
);
445 print_tracking(s
, p
);
447 print_page_info(page
);
449 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
450 p
, p
- addr
, get_freepointer(s
, p
));
453 print_section("Bytes b4", p
- 16, 16);
455 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
457 if (s
->flags
& SLAB_RED_ZONE
)
458 print_section("Redzone", p
+ s
->objsize
,
459 s
->inuse
- s
->objsize
);
462 off
= s
->offset
+ sizeof(void *);
466 if (s
->flags
& SLAB_STORE_USER
)
467 off
+= 2 * sizeof(struct track
);
470 /* Beginning of the filler is the free pointer */
471 print_section("Padding", p
+ off
, s
->size
- off
);
476 static void object_err(struct kmem_cache
*s
, struct page
*page
,
477 u8
*object
, char *reason
)
479 slab_bug(s
, "%s", reason
);
480 print_trailer(s
, page
, object
);
483 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
489 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
491 slab_bug(s
, "%s", buf
);
492 print_page_info(page
);
496 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
500 if (s
->flags
& __OBJECT_POISON
) {
501 memset(p
, POISON_FREE
, s
->objsize
- 1);
502 p
[s
->objsize
- 1] = POISON_END
;
505 if (s
->flags
& SLAB_RED_ZONE
)
506 memset(p
+ s
->objsize
,
507 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
508 s
->inuse
- s
->objsize
);
511 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
514 if (*start
!= (u8
)value
)
522 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
523 void *from
, void *to
)
525 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
526 memset(from
, data
, to
- from
);
529 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
530 u8
*object
, char *what
,
531 u8
*start
, unsigned int value
, unsigned int bytes
)
536 fault
= check_bytes(start
, value
, bytes
);
541 while (end
> fault
&& end
[-1] == value
)
544 slab_bug(s
, "%s overwritten", what
);
545 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
546 fault
, end
- 1, fault
[0], value
);
547 print_trailer(s
, page
, object
);
549 restore_bytes(s
, what
, value
, fault
, end
);
557 * Bytes of the object to be managed.
558 * If the freepointer may overlay the object then the free
559 * pointer is the first word of the object.
561 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
564 * object + s->objsize
565 * Padding to reach word boundary. This is also used for Redzoning.
566 * Padding is extended by another word if Redzoning is enabled and
569 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
570 * 0xcc (RED_ACTIVE) for objects in use.
573 * Meta data starts here.
575 * A. Free pointer (if we cannot overwrite object on free)
576 * B. Tracking data for SLAB_STORE_USER
577 * C. Padding to reach required alignment boundary or at mininum
578 * one word if debugging is on to be able to detect writes
579 * before the word boundary.
581 * Padding is done using 0x5a (POISON_INUSE)
584 * Nothing is used beyond s->size.
586 * If slabcaches are merged then the objsize and inuse boundaries are mostly
587 * ignored. And therefore no slab options that rely on these boundaries
588 * may be used with merged slabcaches.
591 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
593 unsigned long off
= s
->inuse
; /* The end of info */
596 /* Freepointer is placed after the object. */
597 off
+= sizeof(void *);
599 if (s
->flags
& SLAB_STORE_USER
)
600 /* We also have user information there */
601 off
+= 2 * sizeof(struct track
);
606 return check_bytes_and_report(s
, page
, p
, "Object padding",
607 p
+ off
, POISON_INUSE
, s
->size
- off
);
610 /* Check the pad bytes at the end of a slab page */
611 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
619 if (!(s
->flags
& SLAB_POISON
))
622 start
= page_address(page
);
623 length
= (PAGE_SIZE
<< compound_order(page
));
624 end
= start
+ length
;
625 remainder
= length
% s
->size
;
629 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
632 while (end
> fault
&& end
[-1] == POISON_INUSE
)
635 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
636 print_section("Padding", end
- remainder
, remainder
);
638 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
642 static int check_object(struct kmem_cache
*s
, struct page
*page
,
643 void *object
, int active
)
646 u8
*endobject
= object
+ s
->objsize
;
648 if (s
->flags
& SLAB_RED_ZONE
) {
650 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
652 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
653 endobject
, red
, s
->inuse
- s
->objsize
))
656 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
657 check_bytes_and_report(s
, page
, p
, "Alignment padding",
658 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
662 if (s
->flags
& SLAB_POISON
) {
663 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
664 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
665 POISON_FREE
, s
->objsize
- 1) ||
666 !check_bytes_and_report(s
, page
, p
, "Poison",
667 p
+ s
->objsize
- 1, POISON_END
, 1)))
670 * check_pad_bytes cleans up on its own.
672 check_pad_bytes(s
, page
, p
);
675 if (!s
->offset
&& active
)
677 * Object and freepointer overlap. Cannot check
678 * freepointer while object is allocated.
682 /* Check free pointer validity */
683 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
684 object_err(s
, page
, p
, "Freepointer corrupt");
686 * No choice but to zap it and thus lose the remainder
687 * of the free objects in this slab. May cause
688 * another error because the object count is now wrong.
690 set_freepointer(s
, p
, NULL
);
696 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
700 VM_BUG_ON(!irqs_disabled());
702 if (!PageSlab(page
)) {
703 slab_err(s
, page
, "Not a valid slab page");
707 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
708 if (page
->objects
> maxobj
) {
709 slab_err(s
, page
, "objects %u > max %u",
710 s
->name
, page
->objects
, maxobj
);
713 if (page
->inuse
> page
->objects
) {
714 slab_err(s
, page
, "inuse %u > max %u",
715 s
->name
, page
->inuse
, page
->objects
);
718 /* Slab_pad_check fixes things up after itself */
719 slab_pad_check(s
, page
);
724 * Determine if a certain object on a page is on the freelist. Must hold the
725 * slab lock to guarantee that the chains are in a consistent state.
727 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
730 void *fp
= page
->freelist
;
732 unsigned long max_objects
;
734 while (fp
&& nr
<= page
->objects
) {
737 if (!check_valid_pointer(s
, page
, fp
)) {
739 object_err(s
, page
, object
,
740 "Freechain corrupt");
741 set_freepointer(s
, object
, NULL
);
744 slab_err(s
, page
, "Freepointer corrupt");
745 page
->freelist
= NULL
;
746 page
->inuse
= page
->objects
;
747 slab_fix(s
, "Freelist cleared");
753 fp
= get_freepointer(s
, object
);
757 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
758 if (max_objects
> MAX_OBJS_PER_PAGE
)
759 max_objects
= MAX_OBJS_PER_PAGE
;
761 if (page
->objects
!= max_objects
) {
762 slab_err(s
, page
, "Wrong number of objects. Found %d but "
763 "should be %d", page
->objects
, max_objects
);
764 page
->objects
= max_objects
;
765 slab_fix(s
, "Number of objects adjusted.");
767 if (page
->inuse
!= page
->objects
- nr
) {
768 slab_err(s
, page
, "Wrong object count. Counter is %d but "
769 "counted were %d", page
->inuse
, page
->objects
- nr
);
770 page
->inuse
= page
->objects
- nr
;
771 slab_fix(s
, "Object count adjusted.");
773 return search
== NULL
;
776 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
779 if (s
->flags
& SLAB_TRACE
) {
780 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
782 alloc
? "alloc" : "free",
787 print_section("Object", (void *)object
, s
->objsize
);
794 * Hooks for other subsystems that check memory allocations. In a typical
795 * production configuration these hooks all should produce no code at all.
797 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
799 flags
&= gfp_allowed_mask
;
800 lockdep_trace_alloc(flags
);
801 might_sleep_if(flags
& __GFP_WAIT
);
803 return should_failslab(s
->objsize
, flags
, s
->flags
);
806 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
808 flags
&= gfp_allowed_mask
;
809 kmemcheck_slab_alloc(s
, flags
, object
, s
->objsize
);
810 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
813 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
815 kmemleak_free_recursive(x
, s
->flags
);
818 static inline void slab_free_hook_irq(struct kmem_cache
*s
, void *object
)
820 kmemcheck_slab_free(s
, object
, s
->objsize
);
821 debug_check_no_locks_freed(object
, s
->objsize
);
822 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
823 debug_check_no_obj_freed(object
, s
->objsize
);
827 * Tracking of fully allocated slabs for debugging purposes.
829 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
831 spin_lock(&n
->list_lock
);
832 list_add(&page
->lru
, &n
->full
);
833 spin_unlock(&n
->list_lock
);
836 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
838 struct kmem_cache_node
*n
;
840 if (!(s
->flags
& SLAB_STORE_USER
))
843 n
= get_node(s
, page_to_nid(page
));
845 spin_lock(&n
->list_lock
);
846 list_del(&page
->lru
);
847 spin_unlock(&n
->list_lock
);
850 /* Tracking of the number of slabs for debugging purposes */
851 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
853 struct kmem_cache_node
*n
= get_node(s
, node
);
855 return atomic_long_read(&n
->nr_slabs
);
858 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
860 return atomic_long_read(&n
->nr_slabs
);
863 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
865 struct kmem_cache_node
*n
= get_node(s
, node
);
868 * May be called early in order to allocate a slab for the
869 * kmem_cache_node structure. Solve the chicken-egg
870 * dilemma by deferring the increment of the count during
871 * bootstrap (see early_kmem_cache_node_alloc).
873 if (!NUMA_BUILD
|| n
) {
874 atomic_long_inc(&n
->nr_slabs
);
875 atomic_long_add(objects
, &n
->total_objects
);
878 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
880 struct kmem_cache_node
*n
= get_node(s
, node
);
882 atomic_long_dec(&n
->nr_slabs
);
883 atomic_long_sub(objects
, &n
->total_objects
);
886 /* Object debug checks for alloc/free paths */
887 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
890 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
893 init_object(s
, object
, 0);
894 init_tracking(s
, object
);
897 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
898 void *object
, unsigned long addr
)
900 if (!check_slab(s
, page
))
903 if (!on_freelist(s
, page
, object
)) {
904 object_err(s
, page
, object
, "Object already allocated");
908 if (!check_valid_pointer(s
, page
, object
)) {
909 object_err(s
, page
, object
, "Freelist Pointer check fails");
913 if (!check_object(s
, page
, object
, 0))
916 /* Success perform special debug activities for allocs */
917 if (s
->flags
& SLAB_STORE_USER
)
918 set_track(s
, object
, TRACK_ALLOC
, addr
);
919 trace(s
, page
, object
, 1);
920 init_object(s
, object
, 1);
924 if (PageSlab(page
)) {
926 * If this is a slab page then lets do the best we can
927 * to avoid issues in the future. Marking all objects
928 * as used avoids touching the remaining objects.
930 slab_fix(s
, "Marking all objects used");
931 page
->inuse
= page
->objects
;
932 page
->freelist
= NULL
;
937 static noinline
int free_debug_processing(struct kmem_cache
*s
,
938 struct page
*page
, void *object
, unsigned long addr
)
940 if (!check_slab(s
, page
))
943 if (!check_valid_pointer(s
, page
, object
)) {
944 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
948 if (on_freelist(s
, page
, object
)) {
949 object_err(s
, page
, object
, "Object already free");
953 if (!check_object(s
, page
, object
, 1))
956 if (unlikely(s
!= page
->slab
)) {
957 if (!PageSlab(page
)) {
958 slab_err(s
, page
, "Attempt to free object(0x%p) "
959 "outside of slab", object
);
960 } else if (!page
->slab
) {
962 "SLUB <none>: no slab for object 0x%p.\n",
966 object_err(s
, page
, object
,
967 "page slab pointer corrupt.");
971 /* Special debug activities for freeing objects */
972 if (!PageSlubFrozen(page
) && !page
->freelist
)
973 remove_full(s
, page
);
974 if (s
->flags
& SLAB_STORE_USER
)
975 set_track(s
, object
, TRACK_FREE
, addr
);
976 trace(s
, page
, object
, 0);
977 init_object(s
, object
, 0);
981 slab_fix(s
, "Object at 0x%p not freed", object
);
985 static int __init
setup_slub_debug(char *str
)
987 slub_debug
= DEBUG_DEFAULT_FLAGS
;
988 if (*str
++ != '=' || !*str
)
990 * No options specified. Switch on full debugging.
996 * No options but restriction on slabs. This means full
997 * debugging for slabs matching a pattern.
1001 if (tolower(*str
) == 'o') {
1003 * Avoid enabling debugging on caches if its minimum order
1004 * would increase as a result.
1006 disable_higher_order_debug
= 1;
1013 * Switch off all debugging measures.
1018 * Determine which debug features should be switched on
1020 for (; *str
&& *str
!= ','; str
++) {
1021 switch (tolower(*str
)) {
1023 slub_debug
|= SLAB_DEBUG_FREE
;
1026 slub_debug
|= SLAB_RED_ZONE
;
1029 slub_debug
|= SLAB_POISON
;
1032 slub_debug
|= SLAB_STORE_USER
;
1035 slub_debug
|= SLAB_TRACE
;
1038 slub_debug
|= SLAB_FAILSLAB
;
1041 printk(KERN_ERR
"slub_debug option '%c' "
1042 "unknown. skipped\n", *str
);
1048 slub_debug_slabs
= str
+ 1;
1053 __setup("slub_debug", setup_slub_debug
);
1055 static unsigned long kmem_cache_flags(unsigned long objsize
,
1056 unsigned long flags
, const char *name
,
1057 void (*ctor
)(void *))
1060 * Enable debugging if selected on the kernel commandline.
1062 if (slub_debug
&& (!slub_debug_slabs
||
1063 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1064 flags
|= slub_debug
;
1069 static inline void setup_object_debug(struct kmem_cache
*s
,
1070 struct page
*page
, void *object
) {}
1072 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1073 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1075 static inline int free_debug_processing(struct kmem_cache
*s
,
1076 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1078 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1080 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1081 void *object
, int active
) { return 1; }
1082 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1083 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1084 unsigned long flags
, const char *name
,
1085 void (*ctor
)(void *))
1089 #define slub_debug 0
1091 #define disable_higher_order_debug 0
1093 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1095 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1097 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1099 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1102 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1105 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1108 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1110 static inline void slab_free_hook_irq(struct kmem_cache
*s
,
1116 * Slab allocation and freeing
1118 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1119 struct kmem_cache_order_objects oo
)
1121 int order
= oo_order(oo
);
1123 flags
|= __GFP_NOTRACK
;
1125 if (node
== NUMA_NO_NODE
)
1126 return alloc_pages(flags
, order
);
1128 return alloc_pages_exact_node(node
, flags
, order
);
1131 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1134 struct kmem_cache_order_objects oo
= s
->oo
;
1137 flags
|= s
->allocflags
;
1140 * Let the initial higher-order allocation fail under memory pressure
1141 * so we fall-back to the minimum order allocation.
1143 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1145 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1146 if (unlikely(!page
)) {
1149 * Allocation may have failed due to fragmentation.
1150 * Try a lower order alloc if possible
1152 page
= alloc_slab_page(flags
, node
, oo
);
1156 stat(s
, ORDER_FALLBACK
);
1159 if (kmemcheck_enabled
1160 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1161 int pages
= 1 << oo_order(oo
);
1163 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1166 * Objects from caches that have a constructor don't get
1167 * cleared when they're allocated, so we need to do it here.
1170 kmemcheck_mark_uninitialized_pages(page
, pages
);
1172 kmemcheck_mark_unallocated_pages(page
, pages
);
1175 page
->objects
= oo_objects(oo
);
1176 mod_zone_page_state(page_zone(page
),
1177 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1178 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1184 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1187 setup_object_debug(s
, page
, object
);
1188 if (unlikely(s
->ctor
))
1192 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1199 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1201 page
= allocate_slab(s
,
1202 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1206 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1208 page
->flags
|= 1 << PG_slab
;
1210 start
= page_address(page
);
1212 if (unlikely(s
->flags
& SLAB_POISON
))
1213 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1216 for_each_object(p
, s
, start
, page
->objects
) {
1217 setup_object(s
, page
, last
);
1218 set_freepointer(s
, last
, p
);
1221 setup_object(s
, page
, last
);
1222 set_freepointer(s
, last
, NULL
);
1224 page
->freelist
= start
;
1230 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1232 int order
= compound_order(page
);
1233 int pages
= 1 << order
;
1235 if (kmem_cache_debug(s
)) {
1238 slab_pad_check(s
, page
);
1239 for_each_object(p
, s
, page_address(page
),
1241 check_object(s
, page
, p
, 0);
1244 kmemcheck_free_shadow(page
, compound_order(page
));
1246 mod_zone_page_state(page_zone(page
),
1247 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1248 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1251 __ClearPageSlab(page
);
1252 reset_page_mapcount(page
);
1253 if (current
->reclaim_state
)
1254 current
->reclaim_state
->reclaimed_slab
+= pages
;
1255 __free_pages(page
, order
);
1258 static void rcu_free_slab(struct rcu_head
*h
)
1262 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1263 __free_slab(page
->slab
, page
);
1266 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1268 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1270 * RCU free overloads the RCU head over the LRU
1272 struct rcu_head
*head
= (void *)&page
->lru
;
1274 call_rcu(head
, rcu_free_slab
);
1276 __free_slab(s
, page
);
1279 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1281 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1286 * Per slab locking using the pagelock
1288 static __always_inline
void slab_lock(struct page
*page
)
1290 bit_spin_lock(PG_locked
, &page
->flags
);
1293 static __always_inline
void slab_unlock(struct page
*page
)
1295 __bit_spin_unlock(PG_locked
, &page
->flags
);
1298 static __always_inline
int slab_trylock(struct page
*page
)
1302 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1307 * Management of partially allocated slabs
1309 static void add_partial(struct kmem_cache_node
*n
,
1310 struct page
*page
, int tail
)
1312 spin_lock(&n
->list_lock
);
1315 list_add_tail(&page
->lru
, &n
->partial
);
1317 list_add(&page
->lru
, &n
->partial
);
1318 spin_unlock(&n
->list_lock
);
1321 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1323 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1325 spin_lock(&n
->list_lock
);
1326 list_del(&page
->lru
);
1328 spin_unlock(&n
->list_lock
);
1332 * Lock slab and remove from the partial list.
1334 * Must hold list_lock.
1336 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1339 if (slab_trylock(page
)) {
1340 list_del(&page
->lru
);
1342 __SetPageSlubFrozen(page
);
1349 * Try to allocate a partial slab from a specific node.
1351 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1356 * Racy check. If we mistakenly see no partial slabs then we
1357 * just allocate an empty slab. If we mistakenly try to get a
1358 * partial slab and there is none available then get_partials()
1361 if (!n
|| !n
->nr_partial
)
1364 spin_lock(&n
->list_lock
);
1365 list_for_each_entry(page
, &n
->partial
, lru
)
1366 if (lock_and_freeze_slab(n
, page
))
1370 spin_unlock(&n
->list_lock
);
1375 * Get a page from somewhere. Search in increasing NUMA distances.
1377 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1380 struct zonelist
*zonelist
;
1383 enum zone_type high_zoneidx
= gfp_zone(flags
);
1387 * The defrag ratio allows a configuration of the tradeoffs between
1388 * inter node defragmentation and node local allocations. A lower
1389 * defrag_ratio increases the tendency to do local allocations
1390 * instead of attempting to obtain partial slabs from other nodes.
1392 * If the defrag_ratio is set to 0 then kmalloc() always
1393 * returns node local objects. If the ratio is higher then kmalloc()
1394 * may return off node objects because partial slabs are obtained
1395 * from other nodes and filled up.
1397 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1398 * defrag_ratio = 1000) then every (well almost) allocation will
1399 * first attempt to defrag slab caches on other nodes. This means
1400 * scanning over all nodes to look for partial slabs which may be
1401 * expensive if we do it every time we are trying to find a slab
1402 * with available objects.
1404 if (!s
->remote_node_defrag_ratio
||
1405 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1409 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1410 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1411 struct kmem_cache_node
*n
;
1413 n
= get_node(s
, zone_to_nid(zone
));
1415 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1416 n
->nr_partial
> s
->min_partial
) {
1417 page
= get_partial_node(n
);
1430 * Get a partial page, lock it and return it.
1432 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1435 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1437 page
= get_partial_node(get_node(s
, searchnode
));
1438 if (page
|| node
!= -1)
1441 return get_any_partial(s
, flags
);
1445 * Move a page back to the lists.
1447 * Must be called with the slab lock held.
1449 * On exit the slab lock will have been dropped.
1451 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1453 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1455 __ClearPageSlubFrozen(page
);
1458 if (page
->freelist
) {
1459 add_partial(n
, page
, tail
);
1460 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1462 stat(s
, DEACTIVATE_FULL
);
1463 if (kmem_cache_debug(s
) && (s
->flags
& SLAB_STORE_USER
))
1468 stat(s
, DEACTIVATE_EMPTY
);
1469 if (n
->nr_partial
< s
->min_partial
) {
1471 * Adding an empty slab to the partial slabs in order
1472 * to avoid page allocator overhead. This slab needs
1473 * to come after the other slabs with objects in
1474 * so that the others get filled first. That way the
1475 * size of the partial list stays small.
1477 * kmem_cache_shrink can reclaim any empty slabs from
1480 add_partial(n
, page
, 1);
1485 discard_slab(s
, page
);
1491 * Remove the cpu slab
1493 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1495 struct page
*page
= c
->page
;
1499 stat(s
, DEACTIVATE_REMOTE_FREES
);
1501 * Merge cpu freelist into slab freelist. Typically we get here
1502 * because both freelists are empty. So this is unlikely
1505 while (unlikely(c
->freelist
)) {
1508 tail
= 0; /* Hot objects. Put the slab first */
1510 /* Retrieve object from cpu_freelist */
1511 object
= c
->freelist
;
1512 c
->freelist
= get_freepointer(s
, c
->freelist
);
1514 /* And put onto the regular freelist */
1515 set_freepointer(s
, object
, page
->freelist
);
1516 page
->freelist
= object
;
1520 unfreeze_slab(s
, page
, tail
);
1523 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1525 stat(s
, CPUSLAB_FLUSH
);
1527 deactivate_slab(s
, c
);
1533 * Called from IPI handler with interrupts disabled.
1535 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1537 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1539 if (likely(c
&& c
->page
))
1543 static void flush_cpu_slab(void *d
)
1545 struct kmem_cache
*s
= d
;
1547 __flush_cpu_slab(s
, smp_processor_id());
1550 static void flush_all(struct kmem_cache
*s
)
1552 on_each_cpu(flush_cpu_slab
, s
, 1);
1556 * Check if the objects in a per cpu structure fit numa
1557 * locality expectations.
1559 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1562 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
1568 static int count_free(struct page
*page
)
1570 return page
->objects
- page
->inuse
;
1573 static unsigned long count_partial(struct kmem_cache_node
*n
,
1574 int (*get_count
)(struct page
*))
1576 unsigned long flags
;
1577 unsigned long x
= 0;
1580 spin_lock_irqsave(&n
->list_lock
, flags
);
1581 list_for_each_entry(page
, &n
->partial
, lru
)
1582 x
+= get_count(page
);
1583 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1587 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1589 #ifdef CONFIG_SLUB_DEBUG
1590 return atomic_long_read(&n
->total_objects
);
1596 static noinline
void
1597 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1602 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1604 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1605 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1606 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1608 if (oo_order(s
->min
) > get_order(s
->objsize
))
1609 printk(KERN_WARNING
" %s debugging increased min order, use "
1610 "slub_debug=O to disable.\n", s
->name
);
1612 for_each_online_node(node
) {
1613 struct kmem_cache_node
*n
= get_node(s
, node
);
1614 unsigned long nr_slabs
;
1615 unsigned long nr_objs
;
1616 unsigned long nr_free
;
1621 nr_free
= count_partial(n
, count_free
);
1622 nr_slabs
= node_nr_slabs(n
);
1623 nr_objs
= node_nr_objs(n
);
1626 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1627 node
, nr_slabs
, nr_objs
, nr_free
);
1632 * Slow path. The lockless freelist is empty or we need to perform
1635 * Interrupts are disabled.
1637 * Processing is still very fast if new objects have been freed to the
1638 * regular freelist. In that case we simply take over the regular freelist
1639 * as the lockless freelist and zap the regular freelist.
1641 * If that is not working then we fall back to the partial lists. We take the
1642 * first element of the freelist as the object to allocate now and move the
1643 * rest of the freelist to the lockless freelist.
1645 * And if we were unable to get a new slab from the partial slab lists then
1646 * we need to allocate a new slab. This is the slowest path since it involves
1647 * a call to the page allocator and the setup of a new slab.
1649 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1650 unsigned long addr
, struct kmem_cache_cpu
*c
)
1655 /* We handle __GFP_ZERO in the caller */
1656 gfpflags
&= ~__GFP_ZERO
;
1662 if (unlikely(!node_match(c
, node
)))
1665 stat(s
, ALLOC_REFILL
);
1668 object
= c
->page
->freelist
;
1669 if (unlikely(!object
))
1671 if (kmem_cache_debug(s
))
1674 c
->freelist
= get_freepointer(s
, object
);
1675 c
->page
->inuse
= c
->page
->objects
;
1676 c
->page
->freelist
= NULL
;
1677 c
->node
= page_to_nid(c
->page
);
1679 slab_unlock(c
->page
);
1680 stat(s
, ALLOC_SLOWPATH
);
1684 deactivate_slab(s
, c
);
1687 new = get_partial(s
, gfpflags
, node
);
1690 stat(s
, ALLOC_FROM_PARTIAL
);
1694 gfpflags
&= gfp_allowed_mask
;
1695 if (gfpflags
& __GFP_WAIT
)
1698 new = new_slab(s
, gfpflags
, node
);
1700 if (gfpflags
& __GFP_WAIT
)
1701 local_irq_disable();
1704 c
= __this_cpu_ptr(s
->cpu_slab
);
1705 stat(s
, ALLOC_SLAB
);
1709 __SetPageSlubFrozen(new);
1713 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1714 slab_out_of_memory(s
, gfpflags
, node
);
1717 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1721 c
->page
->freelist
= get_freepointer(s
, object
);
1727 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1728 * have the fastpath folded into their functions. So no function call
1729 * overhead for requests that can be satisfied on the fastpath.
1731 * The fastpath works by first checking if the lockless freelist can be used.
1732 * If not then __slab_alloc is called for slow processing.
1734 * Otherwise we can simply pick the next object from the lockless free list.
1736 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1737 gfp_t gfpflags
, int node
, unsigned long addr
)
1740 struct kmem_cache_cpu
*c
;
1741 unsigned long flags
;
1743 if (slab_pre_alloc_hook(s
, gfpflags
))
1746 local_irq_save(flags
);
1747 c
= __this_cpu_ptr(s
->cpu_slab
);
1748 object
= c
->freelist
;
1749 if (unlikely(!object
|| !node_match(c
, node
)))
1751 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1754 c
->freelist
= get_freepointer(s
, object
);
1755 stat(s
, ALLOC_FASTPATH
);
1757 local_irq_restore(flags
);
1759 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1760 memset(object
, 0, s
->objsize
);
1762 slab_post_alloc_hook(s
, gfpflags
, object
);
1767 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1769 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1771 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1775 EXPORT_SYMBOL(kmem_cache_alloc
);
1777 #ifdef CONFIG_TRACING
1778 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1780 return slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1782 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1786 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1788 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1790 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1791 s
->objsize
, s
->size
, gfpflags
, node
);
1795 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1798 #ifdef CONFIG_TRACING
1799 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1803 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1805 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1809 * Slow patch handling. This may still be called frequently since objects
1810 * have a longer lifetime than the cpu slabs in most processing loads.
1812 * So we still attempt to reduce cache line usage. Just take the slab
1813 * lock and free the item. If there is no additional partial page
1814 * handling required then we can return immediately.
1816 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1817 void *x
, unsigned long addr
)
1820 void **object
= (void *)x
;
1822 stat(s
, FREE_SLOWPATH
);
1825 if (kmem_cache_debug(s
))
1829 prior
= page
->freelist
;
1830 set_freepointer(s
, object
, prior
);
1831 page
->freelist
= object
;
1834 if (unlikely(PageSlubFrozen(page
))) {
1835 stat(s
, FREE_FROZEN
);
1839 if (unlikely(!page
->inuse
))
1843 * Objects left in the slab. If it was not on the partial list before
1846 if (unlikely(!prior
)) {
1847 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1848 stat(s
, FREE_ADD_PARTIAL
);
1858 * Slab still on the partial list.
1860 remove_partial(s
, page
);
1861 stat(s
, FREE_REMOVE_PARTIAL
);
1865 discard_slab(s
, page
);
1869 if (!free_debug_processing(s
, page
, x
, addr
))
1875 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1876 * can perform fastpath freeing without additional function calls.
1878 * The fastpath is only possible if we are freeing to the current cpu slab
1879 * of this processor. This typically the case if we have just allocated
1882 * If fastpath is not possible then fall back to __slab_free where we deal
1883 * with all sorts of special processing.
1885 static __always_inline
void slab_free(struct kmem_cache
*s
,
1886 struct page
*page
, void *x
, unsigned long addr
)
1888 void **object
= (void *)x
;
1889 struct kmem_cache_cpu
*c
;
1890 unsigned long flags
;
1892 slab_free_hook(s
, x
);
1894 local_irq_save(flags
);
1895 c
= __this_cpu_ptr(s
->cpu_slab
);
1897 slab_free_hook_irq(s
, x
);
1899 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1900 set_freepointer(s
, object
, c
->freelist
);
1901 c
->freelist
= object
;
1902 stat(s
, FREE_FASTPATH
);
1904 __slab_free(s
, page
, x
, addr
);
1906 local_irq_restore(flags
);
1909 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1913 page
= virt_to_head_page(x
);
1915 slab_free(s
, page
, x
, _RET_IP_
);
1917 trace_kmem_cache_free(_RET_IP_
, x
);
1919 EXPORT_SYMBOL(kmem_cache_free
);
1921 /* Figure out on which slab page the object resides */
1922 static struct page
*get_object_page(const void *x
)
1924 struct page
*page
= virt_to_head_page(x
);
1926 if (!PageSlab(page
))
1933 * Object placement in a slab is made very easy because we always start at
1934 * offset 0. If we tune the size of the object to the alignment then we can
1935 * get the required alignment by putting one properly sized object after
1938 * Notice that the allocation order determines the sizes of the per cpu
1939 * caches. Each processor has always one slab available for allocations.
1940 * Increasing the allocation order reduces the number of times that slabs
1941 * must be moved on and off the partial lists and is therefore a factor in
1946 * Mininum / Maximum order of slab pages. This influences locking overhead
1947 * and slab fragmentation. A higher order reduces the number of partial slabs
1948 * and increases the number of allocations possible without having to
1949 * take the list_lock.
1951 static int slub_min_order
;
1952 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1953 static int slub_min_objects
;
1956 * Merge control. If this is set then no merging of slab caches will occur.
1957 * (Could be removed. This was introduced to pacify the merge skeptics.)
1959 static int slub_nomerge
;
1962 * Calculate the order of allocation given an slab object size.
1964 * The order of allocation has significant impact on performance and other
1965 * system components. Generally order 0 allocations should be preferred since
1966 * order 0 does not cause fragmentation in the page allocator. Larger objects
1967 * be problematic to put into order 0 slabs because there may be too much
1968 * unused space left. We go to a higher order if more than 1/16th of the slab
1971 * In order to reach satisfactory performance we must ensure that a minimum
1972 * number of objects is in one slab. Otherwise we may generate too much
1973 * activity on the partial lists which requires taking the list_lock. This is
1974 * less a concern for large slabs though which are rarely used.
1976 * slub_max_order specifies the order where we begin to stop considering the
1977 * number of objects in a slab as critical. If we reach slub_max_order then
1978 * we try to keep the page order as low as possible. So we accept more waste
1979 * of space in favor of a small page order.
1981 * Higher order allocations also allow the placement of more objects in a
1982 * slab and thereby reduce object handling overhead. If the user has
1983 * requested a higher mininum order then we start with that one instead of
1984 * the smallest order which will fit the object.
1986 static inline int slab_order(int size
, int min_objects
,
1987 int max_order
, int fract_leftover
)
1991 int min_order
= slub_min_order
;
1993 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1994 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1996 for (order
= max(min_order
,
1997 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1998 order
<= max_order
; order
++) {
2000 unsigned long slab_size
= PAGE_SIZE
<< order
;
2002 if (slab_size
< min_objects
* size
)
2005 rem
= slab_size
% size
;
2007 if (rem
<= slab_size
/ fract_leftover
)
2015 static inline int calculate_order(int size
)
2023 * Attempt to find best configuration for a slab. This
2024 * works by first attempting to generate a layout with
2025 * the best configuration and backing off gradually.
2027 * First we reduce the acceptable waste in a slab. Then
2028 * we reduce the minimum objects required in a slab.
2030 min_objects
= slub_min_objects
;
2032 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2033 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
2034 min_objects
= min(min_objects
, max_objects
);
2036 while (min_objects
> 1) {
2038 while (fraction
>= 4) {
2039 order
= slab_order(size
, min_objects
,
2040 slub_max_order
, fraction
);
2041 if (order
<= slub_max_order
)
2049 * We were unable to place multiple objects in a slab. Now
2050 * lets see if we can place a single object there.
2052 order
= slab_order(size
, 1, slub_max_order
, 1);
2053 if (order
<= slub_max_order
)
2057 * Doh this slab cannot be placed using slub_max_order.
2059 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2060 if (order
< MAX_ORDER
)
2066 * Figure out what the alignment of the objects will be.
2068 static unsigned long calculate_alignment(unsigned long flags
,
2069 unsigned long align
, unsigned long size
)
2072 * If the user wants hardware cache aligned objects then follow that
2073 * suggestion if the object is sufficiently large.
2075 * The hardware cache alignment cannot override the specified
2076 * alignment though. If that is greater then use it.
2078 if (flags
& SLAB_HWCACHE_ALIGN
) {
2079 unsigned long ralign
= cache_line_size();
2080 while (size
<= ralign
/ 2)
2082 align
= max(align
, ralign
);
2085 if (align
< ARCH_SLAB_MINALIGN
)
2086 align
= ARCH_SLAB_MINALIGN
;
2088 return ALIGN(align
, sizeof(void *));
2092 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2095 spin_lock_init(&n
->list_lock
);
2096 INIT_LIST_HEAD(&n
->partial
);
2097 #ifdef CONFIG_SLUB_DEBUG
2098 atomic_long_set(&n
->nr_slabs
, 0);
2099 atomic_long_set(&n
->total_objects
, 0);
2100 INIT_LIST_HEAD(&n
->full
);
2104 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2106 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2107 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2109 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2111 return s
->cpu_slab
!= NULL
;
2115 static struct kmem_cache
*kmem_cache_node
;
2118 * No kmalloc_node yet so do it by hand. We know that this is the first
2119 * slab on the node for this slabcache. There are no concurrent accesses
2122 * Note that this function only works on the kmalloc_node_cache
2123 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2124 * memory on a fresh node that has no slab structures yet.
2126 static void early_kmem_cache_node_alloc(int node
)
2129 struct kmem_cache_node
*n
;
2130 unsigned long flags
;
2132 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2134 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2137 if (page_to_nid(page
) != node
) {
2138 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2140 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2141 "in order to be able to continue\n");
2146 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2148 kmem_cache_node
->node
[node
] = n
;
2149 #ifdef CONFIG_SLUB_DEBUG
2150 init_object(kmem_cache_node
, n
, 1);
2151 init_tracking(kmem_cache_node
, n
);
2153 init_kmem_cache_node(n
, kmem_cache_node
);
2154 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2157 * lockdep requires consistent irq usage for each lock
2158 * so even though there cannot be a race this early in
2159 * the boot sequence, we still disable irqs.
2161 local_irq_save(flags
);
2162 add_partial(n
, page
, 0);
2163 local_irq_restore(flags
);
2166 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2170 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2171 struct kmem_cache_node
*n
= s
->node
[node
];
2174 kmem_cache_free(kmem_cache_node
, n
);
2176 s
->node
[node
] = NULL
;
2180 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2184 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2185 struct kmem_cache_node
*n
;
2187 if (slab_state
== DOWN
) {
2188 early_kmem_cache_node_alloc(node
);
2191 n
= kmem_cache_alloc_node(kmem_cache_node
,
2195 free_kmem_cache_nodes(s
);
2200 init_kmem_cache_node(n
, s
);
2205 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2209 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2211 init_kmem_cache_node(&s
->local_node
, s
);
2216 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2218 if (min
< MIN_PARTIAL
)
2220 else if (min
> MAX_PARTIAL
)
2222 s
->min_partial
= min
;
2226 * calculate_sizes() determines the order and the distribution of data within
2229 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2231 unsigned long flags
= s
->flags
;
2232 unsigned long size
= s
->objsize
;
2233 unsigned long align
= s
->align
;
2237 * Round up object size to the next word boundary. We can only
2238 * place the free pointer at word boundaries and this determines
2239 * the possible location of the free pointer.
2241 size
= ALIGN(size
, sizeof(void *));
2243 #ifdef CONFIG_SLUB_DEBUG
2245 * Determine if we can poison the object itself. If the user of
2246 * the slab may touch the object after free or before allocation
2247 * then we should never poison the object itself.
2249 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2251 s
->flags
|= __OBJECT_POISON
;
2253 s
->flags
&= ~__OBJECT_POISON
;
2257 * If we are Redzoning then check if there is some space between the
2258 * end of the object and the free pointer. If not then add an
2259 * additional word to have some bytes to store Redzone information.
2261 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2262 size
+= sizeof(void *);
2266 * With that we have determined the number of bytes in actual use
2267 * by the object. This is the potential offset to the free pointer.
2271 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2274 * Relocate free pointer after the object if it is not
2275 * permitted to overwrite the first word of the object on
2278 * This is the case if we do RCU, have a constructor or
2279 * destructor or are poisoning the objects.
2282 size
+= sizeof(void *);
2285 #ifdef CONFIG_SLUB_DEBUG
2286 if (flags
& SLAB_STORE_USER
)
2288 * Need to store information about allocs and frees after
2291 size
+= 2 * sizeof(struct track
);
2293 if (flags
& SLAB_RED_ZONE
)
2295 * Add some empty padding so that we can catch
2296 * overwrites from earlier objects rather than let
2297 * tracking information or the free pointer be
2298 * corrupted if a user writes before the start
2301 size
+= sizeof(void *);
2305 * Determine the alignment based on various parameters that the
2306 * user specified and the dynamic determination of cache line size
2309 align
= calculate_alignment(flags
, align
, s
->objsize
);
2313 * SLUB stores one object immediately after another beginning from
2314 * offset 0. In order to align the objects we have to simply size
2315 * each object to conform to the alignment.
2317 size
= ALIGN(size
, align
);
2319 if (forced_order
>= 0)
2320 order
= forced_order
;
2322 order
= calculate_order(size
);
2329 s
->allocflags
|= __GFP_COMP
;
2331 if (s
->flags
& SLAB_CACHE_DMA
)
2332 s
->allocflags
|= SLUB_DMA
;
2334 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2335 s
->allocflags
|= __GFP_RECLAIMABLE
;
2338 * Determine the number of objects per slab
2340 s
->oo
= oo_make(order
, size
);
2341 s
->min
= oo_make(get_order(size
), size
);
2342 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2345 return !!oo_objects(s
->oo
);
2349 static int kmem_cache_open(struct kmem_cache
*s
,
2350 const char *name
, size_t size
,
2351 size_t align
, unsigned long flags
,
2352 void (*ctor
)(void *))
2354 memset(s
, 0, kmem_size
);
2359 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2361 if (!calculate_sizes(s
, -1))
2363 if (disable_higher_order_debug
) {
2365 * Disable debugging flags that store metadata if the min slab
2368 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2369 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2371 if (!calculate_sizes(s
, -1))
2377 * The larger the object size is, the more pages we want on the partial
2378 * list to avoid pounding the page allocator excessively.
2380 set_min_partial(s
, ilog2(s
->size
));
2383 s
->remote_node_defrag_ratio
= 1000;
2385 if (!init_kmem_cache_nodes(s
))
2388 if (alloc_kmem_cache_cpus(s
))
2391 free_kmem_cache_nodes(s
);
2393 if (flags
& SLAB_PANIC
)
2394 panic("Cannot create slab %s size=%lu realsize=%u "
2395 "order=%u offset=%u flags=%lx\n",
2396 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2402 * Check if a given pointer is valid
2404 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2408 if (!kern_ptr_validate(object
, s
->size
))
2411 page
= get_object_page(object
);
2413 if (!page
|| s
!= page
->slab
)
2414 /* No slab or wrong slab */
2417 if (!check_valid_pointer(s
, page
, object
))
2421 * We could also check if the object is on the slabs freelist.
2422 * But this would be too expensive and it seems that the main
2423 * purpose of kmem_ptr_valid() is to check if the object belongs
2424 * to a certain slab.
2428 EXPORT_SYMBOL(kmem_ptr_validate
);
2431 * Determine the size of a slab object
2433 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2437 EXPORT_SYMBOL(kmem_cache_size
);
2439 const char *kmem_cache_name(struct kmem_cache
*s
)
2443 EXPORT_SYMBOL(kmem_cache_name
);
2445 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2448 #ifdef CONFIG_SLUB_DEBUG
2449 void *addr
= page_address(page
);
2451 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2456 slab_err(s
, page
, "%s", text
);
2458 for_each_free_object(p
, s
, page
->freelist
)
2459 set_bit(slab_index(p
, s
, addr
), map
);
2461 for_each_object(p
, s
, addr
, page
->objects
) {
2463 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2464 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2466 print_tracking(s
, p
);
2475 * Attempt to free all partial slabs on a node.
2477 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2479 unsigned long flags
;
2480 struct page
*page
, *h
;
2482 spin_lock_irqsave(&n
->list_lock
, flags
);
2483 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2485 list_del(&page
->lru
);
2486 discard_slab(s
, page
);
2489 list_slab_objects(s
, page
,
2490 "Objects remaining on kmem_cache_close()");
2493 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2497 * Release all resources used by a slab cache.
2499 static inline int kmem_cache_close(struct kmem_cache
*s
)
2504 free_percpu(s
->cpu_slab
);
2505 /* Attempt to free all objects */
2506 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2507 struct kmem_cache_node
*n
= get_node(s
, node
);
2510 if (n
->nr_partial
|| slabs_node(s
, node
))
2513 free_kmem_cache_nodes(s
);
2518 * Close a cache and release the kmem_cache structure
2519 * (must be used for caches created using kmem_cache_create)
2521 void kmem_cache_destroy(struct kmem_cache
*s
)
2523 down_write(&slub_lock
);
2527 if (kmem_cache_close(s
)) {
2528 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2529 "still has objects.\n", s
->name
, __func__
);
2532 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2534 sysfs_slab_remove(s
);
2536 up_write(&slub_lock
);
2538 EXPORT_SYMBOL(kmem_cache_destroy
);
2540 /********************************************************************
2542 *******************************************************************/
2544 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
2545 EXPORT_SYMBOL(kmalloc_caches
);
2547 static struct kmem_cache
*kmem_cache
;
2549 #ifdef CONFIG_ZONE_DMA
2550 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
2553 static int __init
setup_slub_min_order(char *str
)
2555 get_option(&str
, &slub_min_order
);
2560 __setup("slub_min_order=", setup_slub_min_order
);
2562 static int __init
setup_slub_max_order(char *str
)
2564 get_option(&str
, &slub_max_order
);
2565 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2570 __setup("slub_max_order=", setup_slub_max_order
);
2572 static int __init
setup_slub_min_objects(char *str
)
2574 get_option(&str
, &slub_min_objects
);
2579 __setup("slub_min_objects=", setup_slub_min_objects
);
2581 static int __init
setup_slub_nomerge(char *str
)
2587 __setup("slub_nomerge", setup_slub_nomerge
);
2589 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
2590 int size
, unsigned int flags
)
2592 struct kmem_cache
*s
;
2594 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
2597 * This function is called with IRQs disabled during early-boot on
2598 * single CPU so there's no need to take slub_lock here.
2600 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2604 list_add(&s
->list
, &slab_caches
);
2608 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2613 * Conversion table for small slabs sizes / 8 to the index in the
2614 * kmalloc array. This is necessary for slabs < 192 since we have non power
2615 * of two cache sizes there. The size of larger slabs can be determined using
2618 static s8 size_index
[24] = {
2645 static inline int size_index_elem(size_t bytes
)
2647 return (bytes
- 1) / 8;
2650 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2656 return ZERO_SIZE_PTR
;
2658 index
= size_index
[size_index_elem(size
)];
2660 index
= fls(size
- 1);
2662 #ifdef CONFIG_ZONE_DMA
2663 if (unlikely((flags
& SLUB_DMA
)))
2664 return kmalloc_dma_caches
[index
];
2667 return kmalloc_caches
[index
];
2670 void *__kmalloc(size_t size
, gfp_t flags
)
2672 struct kmem_cache
*s
;
2675 if (unlikely(size
> SLUB_MAX_SIZE
))
2676 return kmalloc_large(size
, flags
);
2678 s
= get_slab(size
, flags
);
2680 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2683 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
2685 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2689 EXPORT_SYMBOL(__kmalloc
);
2691 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2696 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2697 page
= alloc_pages_node(node
, flags
, get_order(size
));
2699 ptr
= page_address(page
);
2701 kmemleak_alloc(ptr
, size
, 1, flags
);
2706 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2708 struct kmem_cache
*s
;
2711 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2712 ret
= kmalloc_large_node(size
, flags
, node
);
2714 trace_kmalloc_node(_RET_IP_
, ret
,
2715 size
, PAGE_SIZE
<< get_order(size
),
2721 s
= get_slab(size
, flags
);
2723 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2726 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2728 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2732 EXPORT_SYMBOL(__kmalloc_node
);
2735 size_t ksize(const void *object
)
2738 struct kmem_cache
*s
;
2740 if (unlikely(object
== ZERO_SIZE_PTR
))
2743 page
= virt_to_head_page(object
);
2745 if (unlikely(!PageSlab(page
))) {
2746 WARN_ON(!PageCompound(page
));
2747 return PAGE_SIZE
<< compound_order(page
);
2751 #ifdef CONFIG_SLUB_DEBUG
2753 * Debugging requires use of the padding between object
2754 * and whatever may come after it.
2756 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2761 * If we have the need to store the freelist pointer
2762 * back there or track user information then we can
2763 * only use the space before that information.
2765 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2768 * Else we can use all the padding etc for the allocation
2772 EXPORT_SYMBOL(ksize
);
2774 void kfree(const void *x
)
2777 void *object
= (void *)x
;
2779 trace_kfree(_RET_IP_
, x
);
2781 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2784 page
= virt_to_head_page(x
);
2785 if (unlikely(!PageSlab(page
))) {
2786 BUG_ON(!PageCompound(page
));
2791 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2793 EXPORT_SYMBOL(kfree
);
2796 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2797 * the remaining slabs by the number of items in use. The slabs with the
2798 * most items in use come first. New allocations will then fill those up
2799 * and thus they can be removed from the partial lists.
2801 * The slabs with the least items are placed last. This results in them
2802 * being allocated from last increasing the chance that the last objects
2803 * are freed in them.
2805 int kmem_cache_shrink(struct kmem_cache
*s
)
2809 struct kmem_cache_node
*n
;
2812 int objects
= oo_objects(s
->max
);
2813 struct list_head
*slabs_by_inuse
=
2814 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2815 unsigned long flags
;
2817 if (!slabs_by_inuse
)
2821 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2822 n
= get_node(s
, node
);
2827 for (i
= 0; i
< objects
; i
++)
2828 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2830 spin_lock_irqsave(&n
->list_lock
, flags
);
2833 * Build lists indexed by the items in use in each slab.
2835 * Note that concurrent frees may occur while we hold the
2836 * list_lock. page->inuse here is the upper limit.
2838 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2839 if (!page
->inuse
&& slab_trylock(page
)) {
2841 * Must hold slab lock here because slab_free
2842 * may have freed the last object and be
2843 * waiting to release the slab.
2845 list_del(&page
->lru
);
2848 discard_slab(s
, page
);
2850 list_move(&page
->lru
,
2851 slabs_by_inuse
+ page
->inuse
);
2856 * Rebuild the partial list with the slabs filled up most
2857 * first and the least used slabs at the end.
2859 for (i
= objects
- 1; i
>= 0; i
--)
2860 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2862 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2865 kfree(slabs_by_inuse
);
2868 EXPORT_SYMBOL(kmem_cache_shrink
);
2870 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2871 static int slab_mem_going_offline_callback(void *arg
)
2873 struct kmem_cache
*s
;
2875 down_read(&slub_lock
);
2876 list_for_each_entry(s
, &slab_caches
, list
)
2877 kmem_cache_shrink(s
);
2878 up_read(&slub_lock
);
2883 static void slab_mem_offline_callback(void *arg
)
2885 struct kmem_cache_node
*n
;
2886 struct kmem_cache
*s
;
2887 struct memory_notify
*marg
= arg
;
2890 offline_node
= marg
->status_change_nid
;
2893 * If the node still has available memory. we need kmem_cache_node
2896 if (offline_node
< 0)
2899 down_read(&slub_lock
);
2900 list_for_each_entry(s
, &slab_caches
, list
) {
2901 n
= get_node(s
, offline_node
);
2904 * if n->nr_slabs > 0, slabs still exist on the node
2905 * that is going down. We were unable to free them,
2906 * and offline_pages() function shouldn't call this
2907 * callback. So, we must fail.
2909 BUG_ON(slabs_node(s
, offline_node
));
2911 s
->node
[offline_node
] = NULL
;
2912 kmem_cache_free(kmalloc_caches
, n
);
2915 up_read(&slub_lock
);
2918 static int slab_mem_going_online_callback(void *arg
)
2920 struct kmem_cache_node
*n
;
2921 struct kmem_cache
*s
;
2922 struct memory_notify
*marg
= arg
;
2923 int nid
= marg
->status_change_nid
;
2927 * If the node's memory is already available, then kmem_cache_node is
2928 * already created. Nothing to do.
2934 * We are bringing a node online. No memory is available yet. We must
2935 * allocate a kmem_cache_node structure in order to bring the node
2938 down_read(&slub_lock
);
2939 list_for_each_entry(s
, &slab_caches
, list
) {
2941 * XXX: kmem_cache_alloc_node will fallback to other nodes
2942 * since memory is not yet available from the node that
2945 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2950 init_kmem_cache_node(n
, s
);
2954 up_read(&slub_lock
);
2958 static int slab_memory_callback(struct notifier_block
*self
,
2959 unsigned long action
, void *arg
)
2964 case MEM_GOING_ONLINE
:
2965 ret
= slab_mem_going_online_callback(arg
);
2967 case MEM_GOING_OFFLINE
:
2968 ret
= slab_mem_going_offline_callback(arg
);
2971 case MEM_CANCEL_ONLINE
:
2972 slab_mem_offline_callback(arg
);
2975 case MEM_CANCEL_OFFLINE
:
2979 ret
= notifier_from_errno(ret
);
2985 #endif /* CONFIG_MEMORY_HOTPLUG */
2987 /********************************************************************
2988 * Basic setup of slabs
2989 *******************************************************************/
2992 * Used for early kmem_cache structures that were allocated using
2993 * the page allocator
2996 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3000 list_add(&s
->list
, &slab_caches
);
3003 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3004 struct kmem_cache_node
*n
= get_node(s
, node
);
3008 list_for_each_entry(p
, &n
->partial
, lru
)
3011 #ifdef CONFIG_SLAB_DEBUG
3012 list_for_each_entry(p
, &n
->full
, lru
)
3019 void __init
kmem_cache_init(void)
3023 struct kmem_cache
*temp_kmem_cache
;
3027 struct kmem_cache
*temp_kmem_cache_node
;
3028 unsigned long kmalloc_size
;
3030 kmem_size
= offsetof(struct kmem_cache
, node
) +
3031 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3033 /* Allocate two kmem_caches from the page allocator */
3034 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3035 order
= get_order(2 * kmalloc_size
);
3036 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3039 * Must first have the slab cache available for the allocations of the
3040 * struct kmem_cache_node's. There is special bootstrap code in
3041 * kmem_cache_open for slab_state == DOWN.
3043 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3045 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3046 sizeof(struct kmem_cache_node
),
3047 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3049 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3051 /* Allocate a single kmem_cache from the page allocator */
3052 kmem_size
= sizeof(struct kmem_cache
);
3053 order
= get_order(kmem_size
);
3054 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3057 /* Able to allocate the per node structures */
3058 slab_state
= PARTIAL
;
3060 temp_kmem_cache
= kmem_cache
;
3061 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3062 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3063 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3064 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3068 * Allocate kmem_cache_node properly from the kmem_cache slab.
3069 * kmem_cache_node is separately allocated so no need to
3070 * update any list pointers.
3072 temp_kmem_cache_node
= kmem_cache_node
;
3074 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3075 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3077 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3082 * kmem_cache has kmem_cache_node embedded and we moved it!
3083 * Update the list heads
3085 INIT_LIST_HEAD(&kmem_cache
->local_node
.partial
);
3086 list_splice(&temp_kmem_cache
->local_node
.partial
, &kmem_cache
->local_node
.partial
);
3087 #ifdef CONFIG_SLUB_DEBUG
3088 INIT_LIST_HEAD(&kmem_cache
->local_node
.full
);
3089 list_splice(&temp_kmem_cache
->local_node
.full
, &kmem_cache
->local_node
.full
);
3092 kmem_cache_bootstrap_fixup(kmem_cache
);
3094 /* Free temporary boot structure */
3095 free_pages((unsigned long)temp_kmem_cache
, order
);
3097 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3100 * Patch up the size_index table if we have strange large alignment
3101 * requirements for the kmalloc array. This is only the case for
3102 * MIPS it seems. The standard arches will not generate any code here.
3104 * Largest permitted alignment is 256 bytes due to the way we
3105 * handle the index determination for the smaller caches.
3107 * Make sure that nothing crazy happens if someone starts tinkering
3108 * around with ARCH_KMALLOC_MINALIGN
3110 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3111 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3113 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3114 int elem
= size_index_elem(i
);
3115 if (elem
>= ARRAY_SIZE(size_index
))
3117 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3120 if (KMALLOC_MIN_SIZE
== 64) {
3122 * The 96 byte size cache is not used if the alignment
3125 for (i
= 64 + 8; i
<= 96; i
+= 8)
3126 size_index
[size_index_elem(i
)] = 7;
3127 } else if (KMALLOC_MIN_SIZE
== 128) {
3129 * The 192 byte sized cache is not used if the alignment
3130 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3133 for (i
= 128 + 8; i
<= 192; i
+= 8)
3134 size_index
[size_index_elem(i
)] = 8;
3137 /* Caches that are not of the two-to-the-power-of size */
3138 if (KMALLOC_MIN_SIZE
<= 32) {
3139 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3143 if (KMALLOC_MIN_SIZE
<= 64) {
3144 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3148 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3149 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3155 /* Provide the correct kmalloc names now that the caches are up */
3156 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3157 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3160 kmalloc_caches
[i
]->name
= s
;
3164 register_cpu_notifier(&slab_notifier
);
3167 #ifdef CONFIG_ZONE_DMA
3168 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3169 struct kmem_cache
*s
= kmalloc_caches
[i
];
3172 char *name
= kasprintf(GFP_NOWAIT
,
3173 "dma-kmalloc-%d", s
->objsize
);
3176 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3177 s
->objsize
, SLAB_CACHE_DMA
);
3182 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3183 " CPUs=%d, Nodes=%d\n",
3184 caches
, cache_line_size(),
3185 slub_min_order
, slub_max_order
, slub_min_objects
,
3186 nr_cpu_ids
, nr_node_ids
);
3189 void __init
kmem_cache_init_late(void)
3194 * Find a mergeable slab cache
3196 static int slab_unmergeable(struct kmem_cache
*s
)
3198 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3205 * We may have set a slab to be unmergeable during bootstrap.
3207 if (s
->refcount
< 0)
3213 static struct kmem_cache
*find_mergeable(size_t size
,
3214 size_t align
, unsigned long flags
, const char *name
,
3215 void (*ctor
)(void *))
3217 struct kmem_cache
*s
;
3219 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3225 size
= ALIGN(size
, sizeof(void *));
3226 align
= calculate_alignment(flags
, align
, size
);
3227 size
= ALIGN(size
, align
);
3228 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3230 list_for_each_entry(s
, &slab_caches
, list
) {
3231 if (slab_unmergeable(s
))
3237 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3240 * Check if alignment is compatible.
3241 * Courtesy of Adrian Drzewiecki
3243 if ((s
->size
& ~(align
- 1)) != s
->size
)
3246 if (s
->size
- size
>= sizeof(void *))
3254 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3255 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3257 struct kmem_cache
*s
;
3262 down_write(&slub_lock
);
3263 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3267 * Adjust the object sizes so that we clear
3268 * the complete object on kzalloc.
3270 s
->objsize
= max(s
->objsize
, (int)size
);
3271 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3273 if (sysfs_slab_alias(s
, name
)) {
3277 up_write(&slub_lock
);
3281 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3283 if (kmem_cache_open(s
, name
,
3284 size
, align
, flags
, ctor
)) {
3285 list_add(&s
->list
, &slab_caches
);
3286 if (sysfs_slab_add(s
)) {
3291 up_write(&slub_lock
);
3296 up_write(&slub_lock
);
3299 if (flags
& SLAB_PANIC
)
3300 panic("Cannot create slabcache %s\n", name
);
3305 EXPORT_SYMBOL(kmem_cache_create
);
3309 * Use the cpu notifier to insure that the cpu slabs are flushed when
3312 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3313 unsigned long action
, void *hcpu
)
3315 long cpu
= (long)hcpu
;
3316 struct kmem_cache
*s
;
3317 unsigned long flags
;
3320 case CPU_UP_CANCELED
:
3321 case CPU_UP_CANCELED_FROZEN
:
3323 case CPU_DEAD_FROZEN
:
3324 down_read(&slub_lock
);
3325 list_for_each_entry(s
, &slab_caches
, list
) {
3326 local_irq_save(flags
);
3327 __flush_cpu_slab(s
, cpu
);
3328 local_irq_restore(flags
);
3330 up_read(&slub_lock
);
3338 static struct notifier_block __cpuinitdata slab_notifier
= {
3339 .notifier_call
= slab_cpuup_callback
3344 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3346 struct kmem_cache
*s
;
3349 if (unlikely(size
> SLUB_MAX_SIZE
))
3350 return kmalloc_large(size
, gfpflags
);
3352 s
= get_slab(size
, gfpflags
);
3354 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3357 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
3359 /* Honor the call site pointer we recieved. */
3360 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3365 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3366 int node
, unsigned long caller
)
3368 struct kmem_cache
*s
;
3371 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3372 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3374 trace_kmalloc_node(caller
, ret
,
3375 size
, PAGE_SIZE
<< get_order(size
),
3381 s
= get_slab(size
, gfpflags
);
3383 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3386 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3388 /* Honor the call site pointer we recieved. */
3389 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3394 #ifdef CONFIG_SLUB_DEBUG
3395 static int count_inuse(struct page
*page
)
3400 static int count_total(struct page
*page
)
3402 return page
->objects
;
3405 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3409 void *addr
= page_address(page
);
3411 if (!check_slab(s
, page
) ||
3412 !on_freelist(s
, page
, NULL
))
3415 /* Now we know that a valid freelist exists */
3416 bitmap_zero(map
, page
->objects
);
3418 for_each_free_object(p
, s
, page
->freelist
) {
3419 set_bit(slab_index(p
, s
, addr
), map
);
3420 if (!check_object(s
, page
, p
, 0))
3424 for_each_object(p
, s
, addr
, page
->objects
)
3425 if (!test_bit(slab_index(p
, s
, addr
), map
))
3426 if (!check_object(s
, page
, p
, 1))
3431 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3434 if (slab_trylock(page
)) {
3435 validate_slab(s
, page
, map
);
3438 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3442 static int validate_slab_node(struct kmem_cache
*s
,
3443 struct kmem_cache_node
*n
, unsigned long *map
)
3445 unsigned long count
= 0;
3447 unsigned long flags
;
3449 spin_lock_irqsave(&n
->list_lock
, flags
);
3451 list_for_each_entry(page
, &n
->partial
, lru
) {
3452 validate_slab_slab(s
, page
, map
);
3455 if (count
!= n
->nr_partial
)
3456 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3457 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3459 if (!(s
->flags
& SLAB_STORE_USER
))
3462 list_for_each_entry(page
, &n
->full
, lru
) {
3463 validate_slab_slab(s
, page
, map
);
3466 if (count
!= atomic_long_read(&n
->nr_slabs
))
3467 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3468 "counter=%ld\n", s
->name
, count
,
3469 atomic_long_read(&n
->nr_slabs
));
3472 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3476 static long validate_slab_cache(struct kmem_cache
*s
)
3479 unsigned long count
= 0;
3480 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3481 sizeof(unsigned long), GFP_KERNEL
);
3487 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3488 struct kmem_cache_node
*n
= get_node(s
, node
);
3490 count
+= validate_slab_node(s
, n
, map
);
3496 #ifdef SLUB_RESILIENCY_TEST
3497 static void resiliency_test(void)
3501 printk(KERN_ERR
"SLUB resiliency testing\n");
3502 printk(KERN_ERR
"-----------------------\n");
3503 printk(KERN_ERR
"A. Corruption after allocation\n");
3505 p
= kzalloc(16, GFP_KERNEL
);
3507 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3508 " 0x12->0x%p\n\n", p
+ 16);
3510 validate_slab_cache(kmalloc_caches
+ 4);
3512 /* Hmmm... The next two are dangerous */
3513 p
= kzalloc(32, GFP_KERNEL
);
3514 p
[32 + sizeof(void *)] = 0x34;
3515 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3516 " 0x34 -> -0x%p\n", p
);
3518 "If allocated object is overwritten then not detectable\n\n");
3520 validate_slab_cache(kmalloc_caches
+ 5);
3521 p
= kzalloc(64, GFP_KERNEL
);
3522 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3524 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3527 "If allocated object is overwritten then not detectable\n\n");
3528 validate_slab_cache(kmalloc_caches
+ 6);
3530 printk(KERN_ERR
"\nB. Corruption after free\n");
3531 p
= kzalloc(128, GFP_KERNEL
);
3534 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3535 validate_slab_cache(kmalloc_caches
+ 7);
3537 p
= kzalloc(256, GFP_KERNEL
);
3540 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3542 validate_slab_cache(kmalloc_caches
+ 8);
3544 p
= kzalloc(512, GFP_KERNEL
);
3547 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3548 validate_slab_cache(kmalloc_caches
+ 9);
3551 static void resiliency_test(void) {};
3555 * Generate lists of code addresses where slabcache objects are allocated
3560 unsigned long count
;
3567 DECLARE_BITMAP(cpus
, NR_CPUS
);
3573 unsigned long count
;
3574 struct location
*loc
;
3577 static void free_loc_track(struct loc_track
*t
)
3580 free_pages((unsigned long)t
->loc
,
3581 get_order(sizeof(struct location
) * t
->max
));
3584 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3589 order
= get_order(sizeof(struct location
) * max
);
3591 l
= (void *)__get_free_pages(flags
, order
);
3596 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3604 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3605 const struct track
*track
)
3607 long start
, end
, pos
;
3609 unsigned long caddr
;
3610 unsigned long age
= jiffies
- track
->when
;
3616 pos
= start
+ (end
- start
+ 1) / 2;
3619 * There is nothing at "end". If we end up there
3620 * we need to add something to before end.
3625 caddr
= t
->loc
[pos
].addr
;
3626 if (track
->addr
== caddr
) {
3632 if (age
< l
->min_time
)
3634 if (age
> l
->max_time
)
3637 if (track
->pid
< l
->min_pid
)
3638 l
->min_pid
= track
->pid
;
3639 if (track
->pid
> l
->max_pid
)
3640 l
->max_pid
= track
->pid
;
3642 cpumask_set_cpu(track
->cpu
,
3643 to_cpumask(l
->cpus
));
3645 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3649 if (track
->addr
< caddr
)
3656 * Not found. Insert new tracking element.
3658 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3664 (t
->count
- pos
) * sizeof(struct location
));
3667 l
->addr
= track
->addr
;
3671 l
->min_pid
= track
->pid
;
3672 l
->max_pid
= track
->pid
;
3673 cpumask_clear(to_cpumask(l
->cpus
));
3674 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3675 nodes_clear(l
->nodes
);
3676 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3680 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3681 struct page
*page
, enum track_item alloc
,
3684 void *addr
= page_address(page
);
3687 bitmap_zero(map
, page
->objects
);
3688 for_each_free_object(p
, s
, page
->freelist
)
3689 set_bit(slab_index(p
, s
, addr
), map
);
3691 for_each_object(p
, s
, addr
, page
->objects
)
3692 if (!test_bit(slab_index(p
, s
, addr
), map
))
3693 add_location(t
, s
, get_track(s
, p
, alloc
));
3696 static int list_locations(struct kmem_cache
*s
, char *buf
,
3697 enum track_item alloc
)
3701 struct loc_track t
= { 0, 0, NULL
};
3703 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3704 sizeof(unsigned long), GFP_KERNEL
);
3706 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3709 return sprintf(buf
, "Out of memory\n");
3711 /* Push back cpu slabs */
3714 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3715 struct kmem_cache_node
*n
= get_node(s
, node
);
3716 unsigned long flags
;
3719 if (!atomic_long_read(&n
->nr_slabs
))
3722 spin_lock_irqsave(&n
->list_lock
, flags
);
3723 list_for_each_entry(page
, &n
->partial
, lru
)
3724 process_slab(&t
, s
, page
, alloc
, map
);
3725 list_for_each_entry(page
, &n
->full
, lru
)
3726 process_slab(&t
, s
, page
, alloc
, map
);
3727 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3730 for (i
= 0; i
< t
.count
; i
++) {
3731 struct location
*l
= &t
.loc
[i
];
3733 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3735 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3738 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3740 len
+= sprintf(buf
+ len
, "<not-available>");
3742 if (l
->sum_time
!= l
->min_time
) {
3743 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3745 (long)div_u64(l
->sum_time
, l
->count
),
3748 len
+= sprintf(buf
+ len
, " age=%ld",
3751 if (l
->min_pid
!= l
->max_pid
)
3752 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3753 l
->min_pid
, l
->max_pid
);
3755 len
+= sprintf(buf
+ len
, " pid=%ld",
3758 if (num_online_cpus() > 1 &&
3759 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3760 len
< PAGE_SIZE
- 60) {
3761 len
+= sprintf(buf
+ len
, " cpus=");
3762 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3763 to_cpumask(l
->cpus
));
3766 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3767 len
< PAGE_SIZE
- 60) {
3768 len
+= sprintf(buf
+ len
, " nodes=");
3769 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3773 len
+= sprintf(buf
+ len
, "\n");
3779 len
+= sprintf(buf
, "No data\n");
3783 enum slab_stat_type
{
3784 SL_ALL
, /* All slabs */
3785 SL_PARTIAL
, /* Only partially allocated slabs */
3786 SL_CPU
, /* Only slabs used for cpu caches */
3787 SL_OBJECTS
, /* Determine allocated objects not slabs */
3788 SL_TOTAL
/* Determine object capacity not slabs */
3791 #define SO_ALL (1 << SL_ALL)
3792 #define SO_PARTIAL (1 << SL_PARTIAL)
3793 #define SO_CPU (1 << SL_CPU)
3794 #define SO_OBJECTS (1 << SL_OBJECTS)
3795 #define SO_TOTAL (1 << SL_TOTAL)
3797 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3798 char *buf
, unsigned long flags
)
3800 unsigned long total
= 0;
3803 unsigned long *nodes
;
3804 unsigned long *per_cpu
;
3806 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3809 per_cpu
= nodes
+ nr_node_ids
;
3811 if (flags
& SO_CPU
) {
3814 for_each_possible_cpu(cpu
) {
3815 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3817 if (!c
|| c
->node
< 0)
3821 if (flags
& SO_TOTAL
)
3822 x
= c
->page
->objects
;
3823 else if (flags
& SO_OBJECTS
)
3829 nodes
[c
->node
] += x
;
3835 if (flags
& SO_ALL
) {
3836 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3837 struct kmem_cache_node
*n
= get_node(s
, node
);
3839 if (flags
& SO_TOTAL
)
3840 x
= atomic_long_read(&n
->total_objects
);
3841 else if (flags
& SO_OBJECTS
)
3842 x
= atomic_long_read(&n
->total_objects
) -
3843 count_partial(n
, count_free
);
3846 x
= atomic_long_read(&n
->nr_slabs
);
3851 } else if (flags
& SO_PARTIAL
) {
3852 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3853 struct kmem_cache_node
*n
= get_node(s
, node
);
3855 if (flags
& SO_TOTAL
)
3856 x
= count_partial(n
, count_total
);
3857 else if (flags
& SO_OBJECTS
)
3858 x
= count_partial(n
, count_inuse
);
3865 x
= sprintf(buf
, "%lu", total
);
3867 for_each_node_state(node
, N_NORMAL_MEMORY
)
3869 x
+= sprintf(buf
+ x
, " N%d=%lu",
3873 return x
+ sprintf(buf
+ x
, "\n");
3876 static int any_slab_objects(struct kmem_cache
*s
)
3880 for_each_online_node(node
) {
3881 struct kmem_cache_node
*n
= get_node(s
, node
);
3886 if (atomic_long_read(&n
->total_objects
))
3892 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3893 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3895 struct slab_attribute
{
3896 struct attribute attr
;
3897 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3898 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3901 #define SLAB_ATTR_RO(_name) \
3902 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3904 #define SLAB_ATTR(_name) \
3905 static struct slab_attribute _name##_attr = \
3906 __ATTR(_name, 0644, _name##_show, _name##_store)
3908 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3910 return sprintf(buf
, "%d\n", s
->size
);
3912 SLAB_ATTR_RO(slab_size
);
3914 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3916 return sprintf(buf
, "%d\n", s
->align
);
3918 SLAB_ATTR_RO(align
);
3920 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3922 return sprintf(buf
, "%d\n", s
->objsize
);
3924 SLAB_ATTR_RO(object_size
);
3926 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3928 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3930 SLAB_ATTR_RO(objs_per_slab
);
3932 static ssize_t
order_store(struct kmem_cache
*s
,
3933 const char *buf
, size_t length
)
3935 unsigned long order
;
3938 err
= strict_strtoul(buf
, 10, &order
);
3942 if (order
> slub_max_order
|| order
< slub_min_order
)
3945 calculate_sizes(s
, order
);
3949 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3951 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3955 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3957 return sprintf(buf
, "%lu\n", s
->min_partial
);
3960 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3966 err
= strict_strtoul(buf
, 10, &min
);
3970 set_min_partial(s
, min
);
3973 SLAB_ATTR(min_partial
);
3975 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3978 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3980 return n
+ sprintf(buf
+ n
, "\n");
3986 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3988 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3990 SLAB_ATTR_RO(aliases
);
3992 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3994 return show_slab_objects(s
, buf
, SO_ALL
);
3996 SLAB_ATTR_RO(slabs
);
3998 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4000 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4002 SLAB_ATTR_RO(partial
);
4004 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4006 return show_slab_objects(s
, buf
, SO_CPU
);
4008 SLAB_ATTR_RO(cpu_slabs
);
4010 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4012 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4014 SLAB_ATTR_RO(objects
);
4016 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4018 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4020 SLAB_ATTR_RO(objects_partial
);
4022 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4024 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4026 SLAB_ATTR_RO(total_objects
);
4028 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4030 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4033 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4034 const char *buf
, size_t length
)
4036 s
->flags
&= ~SLAB_DEBUG_FREE
;
4038 s
->flags
|= SLAB_DEBUG_FREE
;
4041 SLAB_ATTR(sanity_checks
);
4043 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4045 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4048 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4051 s
->flags
&= ~SLAB_TRACE
;
4053 s
->flags
|= SLAB_TRACE
;
4058 #ifdef CONFIG_FAILSLAB
4059 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4061 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4064 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4067 s
->flags
&= ~SLAB_FAILSLAB
;
4069 s
->flags
|= SLAB_FAILSLAB
;
4072 SLAB_ATTR(failslab
);
4075 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4077 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4080 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4081 const char *buf
, size_t length
)
4083 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4085 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4088 SLAB_ATTR(reclaim_account
);
4090 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4092 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4094 SLAB_ATTR_RO(hwcache_align
);
4096 #ifdef CONFIG_ZONE_DMA
4097 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4099 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4101 SLAB_ATTR_RO(cache_dma
);
4104 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4106 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4108 SLAB_ATTR_RO(destroy_by_rcu
);
4110 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4112 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4115 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4116 const char *buf
, size_t length
)
4118 if (any_slab_objects(s
))
4121 s
->flags
&= ~SLAB_RED_ZONE
;
4123 s
->flags
|= SLAB_RED_ZONE
;
4124 calculate_sizes(s
, -1);
4127 SLAB_ATTR(red_zone
);
4129 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4131 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4134 static ssize_t
poison_store(struct kmem_cache
*s
,
4135 const char *buf
, size_t length
)
4137 if (any_slab_objects(s
))
4140 s
->flags
&= ~SLAB_POISON
;
4142 s
->flags
|= SLAB_POISON
;
4143 calculate_sizes(s
, -1);
4148 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4150 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4153 static ssize_t
store_user_store(struct kmem_cache
*s
,
4154 const char *buf
, size_t length
)
4156 if (any_slab_objects(s
))
4159 s
->flags
&= ~SLAB_STORE_USER
;
4161 s
->flags
|= SLAB_STORE_USER
;
4162 calculate_sizes(s
, -1);
4165 SLAB_ATTR(store_user
);
4167 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4172 static ssize_t
validate_store(struct kmem_cache
*s
,
4173 const char *buf
, size_t length
)
4177 if (buf
[0] == '1') {
4178 ret
= validate_slab_cache(s
);
4184 SLAB_ATTR(validate
);
4186 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4191 static ssize_t
shrink_store(struct kmem_cache
*s
,
4192 const char *buf
, size_t length
)
4194 if (buf
[0] == '1') {
4195 int rc
= kmem_cache_shrink(s
);
4205 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4207 if (!(s
->flags
& SLAB_STORE_USER
))
4209 return list_locations(s
, buf
, TRACK_ALLOC
);
4211 SLAB_ATTR_RO(alloc_calls
);
4213 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4215 if (!(s
->flags
& SLAB_STORE_USER
))
4217 return list_locations(s
, buf
, TRACK_FREE
);
4219 SLAB_ATTR_RO(free_calls
);
4222 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4224 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4227 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4228 const char *buf
, size_t length
)
4230 unsigned long ratio
;
4233 err
= strict_strtoul(buf
, 10, &ratio
);
4238 s
->remote_node_defrag_ratio
= ratio
* 10;
4242 SLAB_ATTR(remote_node_defrag_ratio
);
4245 #ifdef CONFIG_SLUB_STATS
4246 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4248 unsigned long sum
= 0;
4251 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4256 for_each_online_cpu(cpu
) {
4257 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4263 len
= sprintf(buf
, "%lu", sum
);
4266 for_each_online_cpu(cpu
) {
4267 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4268 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4272 return len
+ sprintf(buf
+ len
, "\n");
4275 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4279 for_each_online_cpu(cpu
)
4280 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4283 #define STAT_ATTR(si, text) \
4284 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4286 return show_stat(s, buf, si); \
4288 static ssize_t text##_store(struct kmem_cache *s, \
4289 const char *buf, size_t length) \
4291 if (buf[0] != '0') \
4293 clear_stat(s, si); \
4298 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4299 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4300 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4301 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4302 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4303 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4304 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4305 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4306 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4307 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4308 STAT_ATTR(FREE_SLAB
, free_slab
);
4309 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4310 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4311 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4312 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4313 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4314 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4315 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4318 static struct attribute
*slab_attrs
[] = {
4319 &slab_size_attr
.attr
,
4320 &object_size_attr
.attr
,
4321 &objs_per_slab_attr
.attr
,
4323 &min_partial_attr
.attr
,
4325 &objects_partial_attr
.attr
,
4326 &total_objects_attr
.attr
,
4329 &cpu_slabs_attr
.attr
,
4333 &sanity_checks_attr
.attr
,
4335 &hwcache_align_attr
.attr
,
4336 &reclaim_account_attr
.attr
,
4337 &destroy_by_rcu_attr
.attr
,
4338 &red_zone_attr
.attr
,
4340 &store_user_attr
.attr
,
4341 &validate_attr
.attr
,
4343 &alloc_calls_attr
.attr
,
4344 &free_calls_attr
.attr
,
4345 #ifdef CONFIG_ZONE_DMA
4346 &cache_dma_attr
.attr
,
4349 &remote_node_defrag_ratio_attr
.attr
,
4351 #ifdef CONFIG_SLUB_STATS
4352 &alloc_fastpath_attr
.attr
,
4353 &alloc_slowpath_attr
.attr
,
4354 &free_fastpath_attr
.attr
,
4355 &free_slowpath_attr
.attr
,
4356 &free_frozen_attr
.attr
,
4357 &free_add_partial_attr
.attr
,
4358 &free_remove_partial_attr
.attr
,
4359 &alloc_from_partial_attr
.attr
,
4360 &alloc_slab_attr
.attr
,
4361 &alloc_refill_attr
.attr
,
4362 &free_slab_attr
.attr
,
4363 &cpuslab_flush_attr
.attr
,
4364 &deactivate_full_attr
.attr
,
4365 &deactivate_empty_attr
.attr
,
4366 &deactivate_to_head_attr
.attr
,
4367 &deactivate_to_tail_attr
.attr
,
4368 &deactivate_remote_frees_attr
.attr
,
4369 &order_fallback_attr
.attr
,
4371 #ifdef CONFIG_FAILSLAB
4372 &failslab_attr
.attr
,
4378 static struct attribute_group slab_attr_group
= {
4379 .attrs
= slab_attrs
,
4382 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4383 struct attribute
*attr
,
4386 struct slab_attribute
*attribute
;
4387 struct kmem_cache
*s
;
4390 attribute
= to_slab_attr(attr
);
4393 if (!attribute
->show
)
4396 err
= attribute
->show(s
, buf
);
4401 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4402 struct attribute
*attr
,
4403 const char *buf
, size_t len
)
4405 struct slab_attribute
*attribute
;
4406 struct kmem_cache
*s
;
4409 attribute
= to_slab_attr(attr
);
4412 if (!attribute
->store
)
4415 err
= attribute
->store(s
, buf
, len
);
4420 static void kmem_cache_release(struct kobject
*kobj
)
4422 struct kmem_cache
*s
= to_slab(kobj
);
4427 static const struct sysfs_ops slab_sysfs_ops
= {
4428 .show
= slab_attr_show
,
4429 .store
= slab_attr_store
,
4432 static struct kobj_type slab_ktype
= {
4433 .sysfs_ops
= &slab_sysfs_ops
,
4434 .release
= kmem_cache_release
4437 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4439 struct kobj_type
*ktype
= get_ktype(kobj
);
4441 if (ktype
== &slab_ktype
)
4446 static const struct kset_uevent_ops slab_uevent_ops
= {
4447 .filter
= uevent_filter
,
4450 static struct kset
*slab_kset
;
4452 #define ID_STR_LENGTH 64
4454 /* Create a unique string id for a slab cache:
4456 * Format :[flags-]size
4458 static char *create_unique_id(struct kmem_cache
*s
)
4460 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4467 * First flags affecting slabcache operations. We will only
4468 * get here for aliasable slabs so we do not need to support
4469 * too many flags. The flags here must cover all flags that
4470 * are matched during merging to guarantee that the id is
4473 if (s
->flags
& SLAB_CACHE_DMA
)
4475 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4477 if (s
->flags
& SLAB_DEBUG_FREE
)
4479 if (!(s
->flags
& SLAB_NOTRACK
))
4483 p
+= sprintf(p
, "%07d", s
->size
);
4484 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4488 static int sysfs_slab_add(struct kmem_cache
*s
)
4494 if (slab_state
< SYSFS
)
4495 /* Defer until later */
4498 unmergeable
= slab_unmergeable(s
);
4501 * Slabcache can never be merged so we can use the name proper.
4502 * This is typically the case for debug situations. In that
4503 * case we can catch duplicate names easily.
4505 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4509 * Create a unique name for the slab as a target
4512 name
= create_unique_id(s
);
4515 s
->kobj
.kset
= slab_kset
;
4516 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4518 kobject_put(&s
->kobj
);
4522 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4524 kobject_del(&s
->kobj
);
4525 kobject_put(&s
->kobj
);
4528 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4530 /* Setup first alias */
4531 sysfs_slab_alias(s
, s
->name
);
4537 static void sysfs_slab_remove(struct kmem_cache
*s
)
4539 if (slab_state
< SYSFS
)
4541 * Sysfs has not been setup yet so no need to remove the
4546 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4547 kobject_del(&s
->kobj
);
4548 kobject_put(&s
->kobj
);
4552 * Need to buffer aliases during bootup until sysfs becomes
4553 * available lest we lose that information.
4555 struct saved_alias
{
4556 struct kmem_cache
*s
;
4558 struct saved_alias
*next
;
4561 static struct saved_alias
*alias_list
;
4563 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4565 struct saved_alias
*al
;
4567 if (slab_state
== SYSFS
) {
4569 * If we have a leftover link then remove it.
4571 sysfs_remove_link(&slab_kset
->kobj
, name
);
4572 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4575 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4581 al
->next
= alias_list
;
4586 static int __init
slab_sysfs_init(void)
4588 struct kmem_cache
*s
;
4591 down_write(&slub_lock
);
4593 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4595 up_write(&slub_lock
);
4596 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4602 list_for_each_entry(s
, &slab_caches
, list
) {
4603 err
= sysfs_slab_add(s
);
4605 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4606 " to sysfs\n", s
->name
);
4609 while (alias_list
) {
4610 struct saved_alias
*al
= alias_list
;
4612 alias_list
= alias_list
->next
;
4613 err
= sysfs_slab_alias(al
->s
, al
->name
);
4615 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4616 " %s to sysfs\n", s
->name
);
4620 up_write(&slub_lock
);
4625 __initcall(slab_sysfs_init
);
4629 * The /proc/slabinfo ABI
4631 #ifdef CONFIG_SLABINFO
4632 static void print_slabinfo_header(struct seq_file
*m
)
4634 seq_puts(m
, "slabinfo - version: 2.1\n");
4635 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4636 "<objperslab> <pagesperslab>");
4637 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4638 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4642 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4646 down_read(&slub_lock
);
4648 print_slabinfo_header(m
);
4650 return seq_list_start(&slab_caches
, *pos
);
4653 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4655 return seq_list_next(p
, &slab_caches
, pos
);
4658 static void s_stop(struct seq_file
*m
, void *p
)
4660 up_read(&slub_lock
);
4663 static int s_show(struct seq_file
*m
, void *p
)
4665 unsigned long nr_partials
= 0;
4666 unsigned long nr_slabs
= 0;
4667 unsigned long nr_inuse
= 0;
4668 unsigned long nr_objs
= 0;
4669 unsigned long nr_free
= 0;
4670 struct kmem_cache
*s
;
4673 s
= list_entry(p
, struct kmem_cache
, list
);
4675 for_each_online_node(node
) {
4676 struct kmem_cache_node
*n
= get_node(s
, node
);
4681 nr_partials
+= n
->nr_partial
;
4682 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4683 nr_objs
+= atomic_long_read(&n
->total_objects
);
4684 nr_free
+= count_partial(n
, count_free
);
4687 nr_inuse
= nr_objs
- nr_free
;
4689 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4690 nr_objs
, s
->size
, oo_objects(s
->oo
),
4691 (1 << oo_order(s
->oo
)));
4692 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4693 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4699 static const struct seq_operations slabinfo_op
= {
4706 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4708 return seq_open(file
, &slabinfo_op
);
4711 static const struct file_operations proc_slabinfo_operations
= {
4712 .open
= slabinfo_open
,
4714 .llseek
= seq_lseek
,
4715 .release
= seq_release
,
4718 static int __init
slab_proc_init(void)
4720 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4723 module_init(slab_proc_init
);
4724 #endif /* CONFIG_SLABINFO */