2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/module.h>
13 #include <linux/bit_spinlock.h>
14 #include <linux/interrupt.h>
15 #include <linux/bitops.h>
16 #include <linux/slab.h>
17 #include <linux/proc_fs.h>
18 #include <linux/seq_file.h>
19 #include <trace/kmemtrace.h>
20 #include <linux/cpu.h>
21 #include <linux/cpuset.h>
22 #include <linux/mempolicy.h>
23 #include <linux/ctype.h>
24 #include <linux/debugobjects.h>
25 #include <linux/kallsyms.h>
26 #include <linux/memory.h>
27 #include <linux/math64.h>
28 #include <linux/fault-inject.h>
35 * The slab_lock protects operations on the object of a particular
36 * slab and its metadata in the page struct. If the slab lock
37 * has been taken then no allocations nor frees can be performed
38 * on the objects in the slab nor can the slab be added or removed
39 * from the partial or full lists since this would mean modifying
40 * the page_struct of the slab.
42 * The list_lock protects the partial and full list on each node and
43 * the partial slab counter. If taken then no new slabs may be added or
44 * removed from the lists nor make the number of partial slabs be modified.
45 * (Note that the total number of slabs is an atomic value that may be
46 * modified without taking the list lock).
48 * The list_lock is a centralized lock and thus we avoid taking it as
49 * much as possible. As long as SLUB does not have to handle partial
50 * slabs, operations can continue without any centralized lock. F.e.
51 * allocating a long series of objects that fill up slabs does not require
54 * The lock order is sometimes inverted when we are trying to get a slab
55 * off a list. We take the list_lock and then look for a page on the list
56 * to use. While we do that objects in the slabs may be freed. We can
57 * only operate on the slab if we have also taken the slab_lock. So we use
58 * a slab_trylock() on the slab. If trylock was successful then no frees
59 * can occur anymore and we can use the slab for allocations etc. If the
60 * slab_trylock() does not succeed then frees are in progress in the slab and
61 * we must stay away from it for a while since we may cause a bouncing
62 * cacheline if we try to acquire the lock. So go onto the next slab.
63 * If all pages are busy then we may allocate a new slab instead of reusing
64 * a partial slab. A new slab has noone operating on it and thus there is
65 * no danger of cacheline contention.
67 * Interrupts are disabled during allocation and deallocation in order to
68 * make the slab allocator safe to use in the context of an irq. In addition
69 * interrupts are disabled to ensure that the processor does not change
70 * while handling per_cpu slabs, due to kernel preemption.
72 * SLUB assigns one slab for allocation to each processor.
73 * Allocations only occur from these slabs called cpu slabs.
75 * Slabs with free elements are kept on a partial list and during regular
76 * operations no list for full slabs is used. If an object in a full slab is
77 * freed then the slab will show up again on the partial lists.
78 * We track full slabs for debugging purposes though because otherwise we
79 * cannot scan all objects.
81 * Slabs are freed when they become empty. Teardown and setup is
82 * minimal so we rely on the page allocators per cpu caches for
83 * fast frees and allocs.
85 * Overloading of page flags that are otherwise used for LRU management.
87 * PageActive The slab is frozen and exempt from list processing.
88 * This means that the slab is dedicated to a purpose
89 * such as satisfying allocations for a specific
90 * processor. Objects may be freed in the slab while
91 * it is frozen but slab_free will then skip the usual
92 * list operations. It is up to the processor holding
93 * the slab to integrate the slab into the slab lists
94 * when the slab is no longer needed.
96 * One use of this flag is to mark slabs that are
97 * used for allocations. Then such a slab becomes a cpu
98 * slab. The cpu slab may be equipped with an additional
99 * freelist that allows lockless access to
100 * free objects in addition to the regular freelist
101 * that requires the slab lock.
103 * PageError Slab requires special handling due to debug
104 * options set. This moves slab handling out of
105 * the fast path and disables lockless freelists.
108 #ifdef CONFIG_SLUB_DEBUG
115 * Issues still to be resolved:
117 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
119 * - Variable sizing of the per node arrays
122 /* Enable to test recovery from slab corruption on boot */
123 #undef SLUB_RESILIENCY_TEST
126 * Mininum number of partial slabs. These will be left on the partial
127 * lists even if they are empty. kmem_cache_shrink may reclaim them.
129 #define MIN_PARTIAL 5
132 * Maximum number of desirable partial slabs.
133 * The existence of more partial slabs makes kmem_cache_shrink
134 * sort the partial list by the number of objects in the.
136 #define MAX_PARTIAL 10
138 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
139 SLAB_POISON | SLAB_STORE_USER)
142 * Set of flags that will prevent slab merging
144 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
145 SLAB_TRACE | SLAB_DESTROY_BY_RCU)
147 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
150 #ifndef ARCH_KMALLOC_MINALIGN
151 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
154 #ifndef ARCH_SLAB_MINALIGN
155 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
159 #define OO_MASK ((1 << OO_SHIFT) - 1)
160 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
162 /* Internal SLUB flags */
163 #define __OBJECT_POISON 0x80000000 /* Poison object */
164 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
166 static int kmem_size
= sizeof(struct kmem_cache
);
169 static struct notifier_block slab_notifier
;
173 DOWN
, /* No slab functionality available */
174 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
175 UP
, /* Everything works but does not show up in sysfs */
179 /* A list of all slab caches on the system */
180 static DECLARE_RWSEM(slub_lock
);
181 static LIST_HEAD(slab_caches
);
184 * Tracking user of a slab.
187 unsigned long addr
; /* Called from address */
188 int cpu
; /* Was running on cpu */
189 int pid
; /* Pid context */
190 unsigned long when
; /* When did the operation occur */
193 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
195 #ifdef CONFIG_SLUB_DEBUG
196 static int sysfs_slab_add(struct kmem_cache
*);
197 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
198 static void sysfs_slab_remove(struct kmem_cache
*);
201 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
202 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
204 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
211 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
213 #ifdef CONFIG_SLUB_STATS
218 /********************************************************************
219 * Core slab cache functions
220 *******************************************************************/
222 int slab_is_available(void)
224 return slab_state
>= UP
;
227 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
230 return s
->node
[node
];
232 return &s
->local_node
;
236 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
239 return s
->cpu_slab
[cpu
];
245 /* Verify that a pointer has an address that is valid within a slab page */
246 static inline int check_valid_pointer(struct kmem_cache
*s
,
247 struct page
*page
, const void *object
)
254 base
= page_address(page
);
255 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
256 (object
- base
) % s
->size
) {
264 * Slow version of get and set free pointer.
266 * This version requires touching the cache lines of kmem_cache which
267 * we avoid to do in the fast alloc free paths. There we obtain the offset
268 * from the page struct.
270 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
272 return *(void **)(object
+ s
->offset
);
275 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
277 *(void **)(object
+ s
->offset
) = fp
;
280 /* Loop over all objects in a slab */
281 #define for_each_object(__p, __s, __addr, __objects) \
282 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
286 #define for_each_free_object(__p, __s, __free) \
287 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
289 /* Determine object index from a given position */
290 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
292 return (p
- addr
) / s
->size
;
295 static inline struct kmem_cache_order_objects
oo_make(int order
,
298 struct kmem_cache_order_objects x
= {
299 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
305 static inline int oo_order(struct kmem_cache_order_objects x
)
307 return x
.x
>> OO_SHIFT
;
310 static inline int oo_objects(struct kmem_cache_order_objects x
)
312 return x
.x
& OO_MASK
;
315 #ifdef CONFIG_SLUB_DEBUG
319 #ifdef CONFIG_SLUB_DEBUG_ON
320 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
322 static int slub_debug
;
325 static char *slub_debug_slabs
;
330 static void print_section(char *text
, u8
*addr
, unsigned int length
)
338 for (i
= 0; i
< length
; i
++) {
340 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
343 printk(KERN_CONT
" %02x", addr
[i
]);
345 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
347 printk(KERN_CONT
" %s\n", ascii
);
354 printk(KERN_CONT
" ");
358 printk(KERN_CONT
" %s\n", ascii
);
362 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
363 enum track_item alloc
)
368 p
= object
+ s
->offset
+ sizeof(void *);
370 p
= object
+ s
->inuse
;
375 static void set_track(struct kmem_cache
*s
, void *object
,
376 enum track_item alloc
, unsigned long addr
)
378 struct track
*p
= get_track(s
, object
, alloc
);
382 p
->cpu
= smp_processor_id();
383 p
->pid
= current
->pid
;
386 memset(p
, 0, sizeof(struct track
));
389 static void init_tracking(struct kmem_cache
*s
, void *object
)
391 if (!(s
->flags
& SLAB_STORE_USER
))
394 set_track(s
, object
, TRACK_FREE
, 0UL);
395 set_track(s
, object
, TRACK_ALLOC
, 0UL);
398 static void print_track(const char *s
, struct track
*t
)
403 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
404 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
407 static void print_tracking(struct kmem_cache
*s
, void *object
)
409 if (!(s
->flags
& SLAB_STORE_USER
))
412 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
413 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
416 static void print_page_info(struct page
*page
)
418 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
419 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
423 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
429 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
431 printk(KERN_ERR
"========================================"
432 "=====================================\n");
433 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
434 printk(KERN_ERR
"----------------------------------------"
435 "-------------------------------------\n\n");
438 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
444 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
446 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
449 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
451 unsigned int off
; /* Offset of last byte */
452 u8
*addr
= page_address(page
);
454 print_tracking(s
, p
);
456 print_page_info(page
);
458 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
459 p
, p
- addr
, get_freepointer(s
, p
));
462 print_section("Bytes b4", p
- 16, 16);
464 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
466 if (s
->flags
& SLAB_RED_ZONE
)
467 print_section("Redzone", p
+ s
->objsize
,
468 s
->inuse
- s
->objsize
);
471 off
= s
->offset
+ sizeof(void *);
475 if (s
->flags
& SLAB_STORE_USER
)
476 off
+= 2 * sizeof(struct track
);
479 /* Beginning of the filler is the free pointer */
480 print_section("Padding", p
+ off
, s
->size
- off
);
485 static void object_err(struct kmem_cache
*s
, struct page
*page
,
486 u8
*object
, char *reason
)
488 slab_bug(s
, "%s", reason
);
489 print_trailer(s
, page
, object
);
492 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
498 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
500 slab_bug(s
, "%s", buf
);
501 print_page_info(page
);
505 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
509 if (s
->flags
& __OBJECT_POISON
) {
510 memset(p
, POISON_FREE
, s
->objsize
- 1);
511 p
[s
->objsize
- 1] = POISON_END
;
514 if (s
->flags
& SLAB_RED_ZONE
)
515 memset(p
+ s
->objsize
,
516 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
517 s
->inuse
- s
->objsize
);
520 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
523 if (*start
!= (u8
)value
)
531 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
532 void *from
, void *to
)
534 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
535 memset(from
, data
, to
- from
);
538 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
539 u8
*object
, char *what
,
540 u8
*start
, unsigned int value
, unsigned int bytes
)
545 fault
= check_bytes(start
, value
, bytes
);
550 while (end
> fault
&& end
[-1] == value
)
553 slab_bug(s
, "%s overwritten", what
);
554 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
555 fault
, end
- 1, fault
[0], value
);
556 print_trailer(s
, page
, object
);
558 restore_bytes(s
, what
, value
, fault
, end
);
566 * Bytes of the object to be managed.
567 * If the freepointer may overlay the object then the free
568 * pointer is the first word of the object.
570 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
573 * object + s->objsize
574 * Padding to reach word boundary. This is also used for Redzoning.
575 * Padding is extended by another word if Redzoning is enabled and
578 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
579 * 0xcc (RED_ACTIVE) for objects in use.
582 * Meta data starts here.
584 * A. Free pointer (if we cannot overwrite object on free)
585 * B. Tracking data for SLAB_STORE_USER
586 * C. Padding to reach required alignment boundary or at mininum
587 * one word if debugging is on to be able to detect writes
588 * before the word boundary.
590 * Padding is done using 0x5a (POISON_INUSE)
593 * Nothing is used beyond s->size.
595 * If slabcaches are merged then the objsize and inuse boundaries are mostly
596 * ignored. And therefore no slab options that rely on these boundaries
597 * may be used with merged slabcaches.
600 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
602 unsigned long off
= s
->inuse
; /* The end of info */
605 /* Freepointer is placed after the object. */
606 off
+= sizeof(void *);
608 if (s
->flags
& SLAB_STORE_USER
)
609 /* We also have user information there */
610 off
+= 2 * sizeof(struct track
);
615 return check_bytes_and_report(s
, page
, p
, "Object padding",
616 p
+ off
, POISON_INUSE
, s
->size
- off
);
619 /* Check the pad bytes at the end of a slab page */
620 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
628 if (!(s
->flags
& SLAB_POISON
))
631 start
= page_address(page
);
632 length
= (PAGE_SIZE
<< compound_order(page
));
633 end
= start
+ length
;
634 remainder
= length
% s
->size
;
638 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
641 while (end
> fault
&& end
[-1] == POISON_INUSE
)
644 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
645 print_section("Padding", end
- remainder
, remainder
);
647 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
651 static int check_object(struct kmem_cache
*s
, struct page
*page
,
652 void *object
, int active
)
655 u8
*endobject
= object
+ s
->objsize
;
657 if (s
->flags
& SLAB_RED_ZONE
) {
659 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
661 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
662 endobject
, red
, s
->inuse
- s
->objsize
))
665 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
666 check_bytes_and_report(s
, page
, p
, "Alignment padding",
667 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
671 if (s
->flags
& SLAB_POISON
) {
672 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
673 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
674 POISON_FREE
, s
->objsize
- 1) ||
675 !check_bytes_and_report(s
, page
, p
, "Poison",
676 p
+ s
->objsize
- 1, POISON_END
, 1)))
679 * check_pad_bytes cleans up on its own.
681 check_pad_bytes(s
, page
, p
);
684 if (!s
->offset
&& active
)
686 * Object and freepointer overlap. Cannot check
687 * freepointer while object is allocated.
691 /* Check free pointer validity */
692 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
693 object_err(s
, page
, p
, "Freepointer corrupt");
695 * No choice but to zap it and thus lose the remainder
696 * of the free objects in this slab. May cause
697 * another error because the object count is now wrong.
699 set_freepointer(s
, p
, NULL
);
705 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
709 VM_BUG_ON(!irqs_disabled());
711 if (!PageSlab(page
)) {
712 slab_err(s
, page
, "Not a valid slab page");
716 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
717 if (page
->objects
> maxobj
) {
718 slab_err(s
, page
, "objects %u > max %u",
719 s
->name
, page
->objects
, maxobj
);
722 if (page
->inuse
> page
->objects
) {
723 slab_err(s
, page
, "inuse %u > max %u",
724 s
->name
, page
->inuse
, page
->objects
);
727 /* Slab_pad_check fixes things up after itself */
728 slab_pad_check(s
, page
);
733 * Determine if a certain object on a page is on the freelist. Must hold the
734 * slab lock to guarantee that the chains are in a consistent state.
736 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
739 void *fp
= page
->freelist
;
741 unsigned long max_objects
;
743 while (fp
&& nr
<= page
->objects
) {
746 if (!check_valid_pointer(s
, page
, fp
)) {
748 object_err(s
, page
, object
,
749 "Freechain corrupt");
750 set_freepointer(s
, object
, NULL
);
753 slab_err(s
, page
, "Freepointer corrupt");
754 page
->freelist
= NULL
;
755 page
->inuse
= page
->objects
;
756 slab_fix(s
, "Freelist cleared");
762 fp
= get_freepointer(s
, object
);
766 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
767 if (max_objects
> MAX_OBJS_PER_PAGE
)
768 max_objects
= MAX_OBJS_PER_PAGE
;
770 if (page
->objects
!= max_objects
) {
771 slab_err(s
, page
, "Wrong number of objects. Found %d but "
772 "should be %d", page
->objects
, max_objects
);
773 page
->objects
= max_objects
;
774 slab_fix(s
, "Number of objects adjusted.");
776 if (page
->inuse
!= page
->objects
- nr
) {
777 slab_err(s
, page
, "Wrong object count. Counter is %d but "
778 "counted were %d", page
->inuse
, page
->objects
- nr
);
779 page
->inuse
= page
->objects
- nr
;
780 slab_fix(s
, "Object count adjusted.");
782 return search
== NULL
;
785 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
788 if (s
->flags
& SLAB_TRACE
) {
789 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
791 alloc
? "alloc" : "free",
796 print_section("Object", (void *)object
, s
->objsize
);
803 * Tracking of fully allocated slabs for debugging purposes.
805 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
807 spin_lock(&n
->list_lock
);
808 list_add(&page
->lru
, &n
->full
);
809 spin_unlock(&n
->list_lock
);
812 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
814 struct kmem_cache_node
*n
;
816 if (!(s
->flags
& SLAB_STORE_USER
))
819 n
= get_node(s
, page_to_nid(page
));
821 spin_lock(&n
->list_lock
);
822 list_del(&page
->lru
);
823 spin_unlock(&n
->list_lock
);
826 /* Tracking of the number of slabs for debugging purposes */
827 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
829 struct kmem_cache_node
*n
= get_node(s
, node
);
831 return atomic_long_read(&n
->nr_slabs
);
834 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
836 struct kmem_cache_node
*n
= get_node(s
, node
);
839 * May be called early in order to allocate a slab for the
840 * kmem_cache_node structure. Solve the chicken-egg
841 * dilemma by deferring the increment of the count during
842 * bootstrap (see early_kmem_cache_node_alloc).
844 if (!NUMA_BUILD
|| n
) {
845 atomic_long_inc(&n
->nr_slabs
);
846 atomic_long_add(objects
, &n
->total_objects
);
849 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
851 struct kmem_cache_node
*n
= get_node(s
, node
);
853 atomic_long_dec(&n
->nr_slabs
);
854 atomic_long_sub(objects
, &n
->total_objects
);
857 /* Object debug checks for alloc/free paths */
858 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
861 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
864 init_object(s
, object
, 0);
865 init_tracking(s
, object
);
868 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
869 void *object
, unsigned long addr
)
871 if (!check_slab(s
, page
))
874 if (!on_freelist(s
, page
, object
)) {
875 object_err(s
, page
, object
, "Object already allocated");
879 if (!check_valid_pointer(s
, page
, object
)) {
880 object_err(s
, page
, object
, "Freelist Pointer check fails");
884 if (!check_object(s
, page
, object
, 0))
887 /* Success perform special debug activities for allocs */
888 if (s
->flags
& SLAB_STORE_USER
)
889 set_track(s
, object
, TRACK_ALLOC
, addr
);
890 trace(s
, page
, object
, 1);
891 init_object(s
, object
, 1);
895 if (PageSlab(page
)) {
897 * If this is a slab page then lets do the best we can
898 * to avoid issues in the future. Marking all objects
899 * as used avoids touching the remaining objects.
901 slab_fix(s
, "Marking all objects used");
902 page
->inuse
= page
->objects
;
903 page
->freelist
= NULL
;
908 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
909 void *object
, unsigned long addr
)
911 if (!check_slab(s
, page
))
914 if (!check_valid_pointer(s
, page
, object
)) {
915 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
919 if (on_freelist(s
, page
, object
)) {
920 object_err(s
, page
, object
, "Object already free");
924 if (!check_object(s
, page
, object
, 1))
927 if (unlikely(s
!= page
->slab
)) {
928 if (!PageSlab(page
)) {
929 slab_err(s
, page
, "Attempt to free object(0x%p) "
930 "outside of slab", object
);
931 } else if (!page
->slab
) {
933 "SLUB <none>: no slab for object 0x%p.\n",
937 object_err(s
, page
, object
,
938 "page slab pointer corrupt.");
942 /* Special debug activities for freeing objects */
943 if (!PageSlubFrozen(page
) && !page
->freelist
)
944 remove_full(s
, page
);
945 if (s
->flags
& SLAB_STORE_USER
)
946 set_track(s
, object
, TRACK_FREE
, addr
);
947 trace(s
, page
, object
, 0);
948 init_object(s
, object
, 0);
952 slab_fix(s
, "Object at 0x%p not freed", object
);
956 static int __init
setup_slub_debug(char *str
)
958 slub_debug
= DEBUG_DEFAULT_FLAGS
;
959 if (*str
++ != '=' || !*str
)
961 * No options specified. Switch on full debugging.
967 * No options but restriction on slabs. This means full
968 * debugging for slabs matching a pattern.
975 * Switch off all debugging measures.
980 * Determine which debug features should be switched on
982 for (; *str
&& *str
!= ','; str
++) {
983 switch (tolower(*str
)) {
985 slub_debug
|= SLAB_DEBUG_FREE
;
988 slub_debug
|= SLAB_RED_ZONE
;
991 slub_debug
|= SLAB_POISON
;
994 slub_debug
|= SLAB_STORE_USER
;
997 slub_debug
|= SLAB_TRACE
;
1000 printk(KERN_ERR
"slub_debug option '%c' "
1001 "unknown. skipped\n", *str
);
1007 slub_debug_slabs
= str
+ 1;
1012 __setup("slub_debug", setup_slub_debug
);
1014 static unsigned long kmem_cache_flags(unsigned long objsize
,
1015 unsigned long flags
, const char *name
,
1016 void (*ctor
)(void *))
1019 * Enable debugging if selected on the kernel commandline.
1021 if (slub_debug
&& (!slub_debug_slabs
||
1022 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1023 flags
|= slub_debug
;
1028 static inline void setup_object_debug(struct kmem_cache
*s
,
1029 struct page
*page
, void *object
) {}
1031 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1032 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1034 static inline int free_debug_processing(struct kmem_cache
*s
,
1035 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1037 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1039 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1040 void *object
, int active
) { return 1; }
1041 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1042 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1043 unsigned long flags
, const char *name
,
1044 void (*ctor
)(void *))
1048 #define slub_debug 0
1050 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1052 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1054 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1059 * Slab allocation and freeing
1061 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1062 struct kmem_cache_order_objects oo
)
1064 int order
= oo_order(oo
);
1067 return alloc_pages(flags
, order
);
1069 return alloc_pages_node(node
, flags
, order
);
1072 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1075 struct kmem_cache_order_objects oo
= s
->oo
;
1077 flags
|= s
->allocflags
;
1079 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1081 if (unlikely(!page
)) {
1084 * Allocation may have failed due to fragmentation.
1085 * Try a lower order alloc if possible
1087 page
= alloc_slab_page(flags
, node
, oo
);
1091 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1093 page
->objects
= oo_objects(oo
);
1094 mod_zone_page_state(page_zone(page
),
1095 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1096 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1102 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1105 setup_object_debug(s
, page
, object
);
1106 if (unlikely(s
->ctor
))
1110 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1117 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1119 page
= allocate_slab(s
,
1120 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1124 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1126 page
->flags
|= 1 << PG_slab
;
1127 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1128 SLAB_STORE_USER
| SLAB_TRACE
))
1129 __SetPageSlubDebug(page
);
1131 start
= page_address(page
);
1133 if (unlikely(s
->flags
& SLAB_POISON
))
1134 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1137 for_each_object(p
, s
, start
, page
->objects
) {
1138 setup_object(s
, page
, last
);
1139 set_freepointer(s
, last
, p
);
1142 setup_object(s
, page
, last
);
1143 set_freepointer(s
, last
, NULL
);
1145 page
->freelist
= start
;
1151 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1153 int order
= compound_order(page
);
1154 int pages
= 1 << order
;
1156 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1159 slab_pad_check(s
, page
);
1160 for_each_object(p
, s
, page_address(page
),
1162 check_object(s
, page
, p
, 0);
1163 __ClearPageSlubDebug(page
);
1166 mod_zone_page_state(page_zone(page
),
1167 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1168 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1171 __ClearPageSlab(page
);
1172 reset_page_mapcount(page
);
1173 __free_pages(page
, order
);
1176 static void rcu_free_slab(struct rcu_head
*h
)
1180 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1181 __free_slab(page
->slab
, page
);
1184 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1186 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1188 * RCU free overloads the RCU head over the LRU
1190 struct rcu_head
*head
= (void *)&page
->lru
;
1192 call_rcu(head
, rcu_free_slab
);
1194 __free_slab(s
, page
);
1197 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1199 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1204 * Per slab locking using the pagelock
1206 static __always_inline
void slab_lock(struct page
*page
)
1208 bit_spin_lock(PG_locked
, &page
->flags
);
1211 static __always_inline
void slab_unlock(struct page
*page
)
1213 __bit_spin_unlock(PG_locked
, &page
->flags
);
1216 static __always_inline
int slab_trylock(struct page
*page
)
1220 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1225 * Management of partially allocated slabs
1227 static void add_partial(struct kmem_cache_node
*n
,
1228 struct page
*page
, int tail
)
1230 spin_lock(&n
->list_lock
);
1233 list_add_tail(&page
->lru
, &n
->partial
);
1235 list_add(&page
->lru
, &n
->partial
);
1236 spin_unlock(&n
->list_lock
);
1239 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1241 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1243 spin_lock(&n
->list_lock
);
1244 list_del(&page
->lru
);
1246 spin_unlock(&n
->list_lock
);
1250 * Lock slab and remove from the partial list.
1252 * Must hold list_lock.
1254 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1257 if (slab_trylock(page
)) {
1258 list_del(&page
->lru
);
1260 __SetPageSlubFrozen(page
);
1267 * Try to allocate a partial slab from a specific node.
1269 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1274 * Racy check. If we mistakenly see no partial slabs then we
1275 * just allocate an empty slab. If we mistakenly try to get a
1276 * partial slab and there is none available then get_partials()
1279 if (!n
|| !n
->nr_partial
)
1282 spin_lock(&n
->list_lock
);
1283 list_for_each_entry(page
, &n
->partial
, lru
)
1284 if (lock_and_freeze_slab(n
, page
))
1288 spin_unlock(&n
->list_lock
);
1293 * Get a page from somewhere. Search in increasing NUMA distances.
1295 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1298 struct zonelist
*zonelist
;
1301 enum zone_type high_zoneidx
= gfp_zone(flags
);
1305 * The defrag ratio allows a configuration of the tradeoffs between
1306 * inter node defragmentation and node local allocations. A lower
1307 * defrag_ratio increases the tendency to do local allocations
1308 * instead of attempting to obtain partial slabs from other nodes.
1310 * If the defrag_ratio is set to 0 then kmalloc() always
1311 * returns node local objects. If the ratio is higher then kmalloc()
1312 * may return off node objects because partial slabs are obtained
1313 * from other nodes and filled up.
1315 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1316 * defrag_ratio = 1000) then every (well almost) allocation will
1317 * first attempt to defrag slab caches on other nodes. This means
1318 * scanning over all nodes to look for partial slabs which may be
1319 * expensive if we do it every time we are trying to find a slab
1320 * with available objects.
1322 if (!s
->remote_node_defrag_ratio
||
1323 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1326 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1327 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1328 struct kmem_cache_node
*n
;
1330 n
= get_node(s
, zone_to_nid(zone
));
1332 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1333 n
->nr_partial
> s
->min_partial
) {
1334 page
= get_partial_node(n
);
1344 * Get a partial page, lock it and return it.
1346 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1349 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1351 page
= get_partial_node(get_node(s
, searchnode
));
1352 if (page
|| (flags
& __GFP_THISNODE
))
1355 return get_any_partial(s
, flags
);
1359 * Move a page back to the lists.
1361 * Must be called with the slab lock held.
1363 * On exit the slab lock will have been dropped.
1365 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1367 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1368 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1370 __ClearPageSlubFrozen(page
);
1373 if (page
->freelist
) {
1374 add_partial(n
, page
, tail
);
1375 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1377 stat(c
, DEACTIVATE_FULL
);
1378 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1379 (s
->flags
& SLAB_STORE_USER
))
1384 stat(c
, DEACTIVATE_EMPTY
);
1385 if (n
->nr_partial
< s
->min_partial
) {
1387 * Adding an empty slab to the partial slabs in order
1388 * to avoid page allocator overhead. This slab needs
1389 * to come after the other slabs with objects in
1390 * so that the others get filled first. That way the
1391 * size of the partial list stays small.
1393 * kmem_cache_shrink can reclaim any empty slabs from
1396 add_partial(n
, page
, 1);
1400 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1401 discard_slab(s
, page
);
1407 * Remove the cpu slab
1409 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1411 struct page
*page
= c
->page
;
1415 stat(c
, DEACTIVATE_REMOTE_FREES
);
1417 * Merge cpu freelist into slab freelist. Typically we get here
1418 * because both freelists are empty. So this is unlikely
1421 while (unlikely(c
->freelist
)) {
1424 tail
= 0; /* Hot objects. Put the slab first */
1426 /* Retrieve object from cpu_freelist */
1427 object
= c
->freelist
;
1428 c
->freelist
= c
->freelist
[c
->offset
];
1430 /* And put onto the regular freelist */
1431 object
[c
->offset
] = page
->freelist
;
1432 page
->freelist
= object
;
1436 unfreeze_slab(s
, page
, tail
);
1439 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1441 stat(c
, CPUSLAB_FLUSH
);
1443 deactivate_slab(s
, c
);
1449 * Called from IPI handler with interrupts disabled.
1451 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1453 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1455 if (likely(c
&& c
->page
))
1459 static void flush_cpu_slab(void *d
)
1461 struct kmem_cache
*s
= d
;
1463 __flush_cpu_slab(s
, smp_processor_id());
1466 static void flush_all(struct kmem_cache
*s
)
1468 on_each_cpu(flush_cpu_slab
, s
, 1);
1472 * Check if the objects in a per cpu structure fit numa
1473 * locality expectations.
1475 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1478 if (node
!= -1 && c
->node
!= node
)
1485 * Slow path. The lockless freelist is empty or we need to perform
1488 * Interrupts are disabled.
1490 * Processing is still very fast if new objects have been freed to the
1491 * regular freelist. In that case we simply take over the regular freelist
1492 * as the lockless freelist and zap the regular freelist.
1494 * If that is not working then we fall back to the partial lists. We take the
1495 * first element of the freelist as the object to allocate now and move the
1496 * rest of the freelist to the lockless freelist.
1498 * And if we were unable to get a new slab from the partial slab lists then
1499 * we need to allocate a new slab. This is the slowest path since it involves
1500 * a call to the page allocator and the setup of a new slab.
1502 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1503 unsigned long addr
, struct kmem_cache_cpu
*c
)
1508 /* We handle __GFP_ZERO in the caller */
1509 gfpflags
&= ~__GFP_ZERO
;
1515 if (unlikely(!node_match(c
, node
)))
1518 stat(c
, ALLOC_REFILL
);
1521 object
= c
->page
->freelist
;
1522 if (unlikely(!object
))
1524 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1527 c
->freelist
= object
[c
->offset
];
1528 c
->page
->inuse
= c
->page
->objects
;
1529 c
->page
->freelist
= NULL
;
1530 c
->node
= page_to_nid(c
->page
);
1532 slab_unlock(c
->page
);
1533 stat(c
, ALLOC_SLOWPATH
);
1537 deactivate_slab(s
, c
);
1540 new = get_partial(s
, gfpflags
, node
);
1543 stat(c
, ALLOC_FROM_PARTIAL
);
1547 if (gfpflags
& __GFP_WAIT
)
1550 new = new_slab(s
, gfpflags
, node
);
1552 if (gfpflags
& __GFP_WAIT
)
1553 local_irq_disable();
1556 c
= get_cpu_slab(s
, smp_processor_id());
1557 stat(c
, ALLOC_SLAB
);
1561 __SetPageSlubFrozen(new);
1567 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1571 c
->page
->freelist
= object
[c
->offset
];
1577 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1578 * have the fastpath folded into their functions. So no function call
1579 * overhead for requests that can be satisfied on the fastpath.
1581 * The fastpath works by first checking if the lockless freelist can be used.
1582 * If not then __slab_alloc is called for slow processing.
1584 * Otherwise we can simply pick the next object from the lockless free list.
1586 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1587 gfp_t gfpflags
, int node
, unsigned long addr
)
1590 struct kmem_cache_cpu
*c
;
1591 unsigned long flags
;
1592 unsigned int objsize
;
1594 lockdep_trace_alloc(gfpflags
);
1595 might_sleep_if(gfpflags
& __GFP_WAIT
);
1597 if (should_failslab(s
->objsize
, gfpflags
))
1600 local_irq_save(flags
);
1601 c
= get_cpu_slab(s
, smp_processor_id());
1602 objsize
= c
->objsize
;
1603 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1605 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1608 object
= c
->freelist
;
1609 c
->freelist
= object
[c
->offset
];
1610 stat(c
, ALLOC_FASTPATH
);
1612 local_irq_restore(flags
);
1614 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1615 memset(object
, 0, objsize
);
1620 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1622 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1624 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1628 EXPORT_SYMBOL(kmem_cache_alloc
);
1630 #ifdef CONFIG_KMEMTRACE
1631 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1633 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1635 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1639 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1641 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1643 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1644 s
->objsize
, s
->size
, gfpflags
, node
);
1648 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1651 #ifdef CONFIG_KMEMTRACE
1652 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1656 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1658 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1662 * Slow patch handling. This may still be called frequently since objects
1663 * have a longer lifetime than the cpu slabs in most processing loads.
1665 * So we still attempt to reduce cache line usage. Just take the slab
1666 * lock and free the item. If there is no additional partial page
1667 * handling required then we can return immediately.
1669 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1670 void *x
, unsigned long addr
, unsigned int offset
)
1673 void **object
= (void *)x
;
1674 struct kmem_cache_cpu
*c
;
1676 c
= get_cpu_slab(s
, raw_smp_processor_id());
1677 stat(c
, FREE_SLOWPATH
);
1680 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1684 prior
= object
[offset
] = page
->freelist
;
1685 page
->freelist
= object
;
1688 if (unlikely(PageSlubFrozen(page
))) {
1689 stat(c
, FREE_FROZEN
);
1693 if (unlikely(!page
->inuse
))
1697 * Objects left in the slab. If it was not on the partial list before
1700 if (unlikely(!prior
)) {
1701 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1702 stat(c
, FREE_ADD_PARTIAL
);
1712 * Slab still on the partial list.
1714 remove_partial(s
, page
);
1715 stat(c
, FREE_REMOVE_PARTIAL
);
1719 discard_slab(s
, page
);
1723 if (!free_debug_processing(s
, page
, x
, addr
))
1729 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1730 * can perform fastpath freeing without additional function calls.
1732 * The fastpath is only possible if we are freeing to the current cpu slab
1733 * of this processor. This typically the case if we have just allocated
1736 * If fastpath is not possible then fall back to __slab_free where we deal
1737 * with all sorts of special processing.
1739 static __always_inline
void slab_free(struct kmem_cache
*s
,
1740 struct page
*page
, void *x
, unsigned long addr
)
1742 void **object
= (void *)x
;
1743 struct kmem_cache_cpu
*c
;
1744 unsigned long flags
;
1746 local_irq_save(flags
);
1747 c
= get_cpu_slab(s
, smp_processor_id());
1748 debug_check_no_locks_freed(object
, c
->objsize
);
1749 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1750 debug_check_no_obj_freed(object
, c
->objsize
);
1751 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1752 object
[c
->offset
] = c
->freelist
;
1753 c
->freelist
= object
;
1754 stat(c
, FREE_FASTPATH
);
1756 __slab_free(s
, page
, x
, addr
, c
->offset
);
1758 local_irq_restore(flags
);
1761 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1765 page
= virt_to_head_page(x
);
1767 slab_free(s
, page
, x
, _RET_IP_
);
1769 trace_kmem_cache_free(_RET_IP_
, x
);
1771 EXPORT_SYMBOL(kmem_cache_free
);
1773 /* Figure out on which slab page the object resides */
1774 static struct page
*get_object_page(const void *x
)
1776 struct page
*page
= virt_to_head_page(x
);
1778 if (!PageSlab(page
))
1785 * Object placement in a slab is made very easy because we always start at
1786 * offset 0. If we tune the size of the object to the alignment then we can
1787 * get the required alignment by putting one properly sized object after
1790 * Notice that the allocation order determines the sizes of the per cpu
1791 * caches. Each processor has always one slab available for allocations.
1792 * Increasing the allocation order reduces the number of times that slabs
1793 * must be moved on and off the partial lists and is therefore a factor in
1798 * Mininum / Maximum order of slab pages. This influences locking overhead
1799 * and slab fragmentation. A higher order reduces the number of partial slabs
1800 * and increases the number of allocations possible without having to
1801 * take the list_lock.
1803 static int slub_min_order
;
1804 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1805 static int slub_min_objects
;
1808 * Merge control. If this is set then no merging of slab caches will occur.
1809 * (Could be removed. This was introduced to pacify the merge skeptics.)
1811 static int slub_nomerge
;
1814 * Calculate the order of allocation given an slab object size.
1816 * The order of allocation has significant impact on performance and other
1817 * system components. Generally order 0 allocations should be preferred since
1818 * order 0 does not cause fragmentation in the page allocator. Larger objects
1819 * be problematic to put into order 0 slabs because there may be too much
1820 * unused space left. We go to a higher order if more than 1/16th of the slab
1823 * In order to reach satisfactory performance we must ensure that a minimum
1824 * number of objects is in one slab. Otherwise we may generate too much
1825 * activity on the partial lists which requires taking the list_lock. This is
1826 * less a concern for large slabs though which are rarely used.
1828 * slub_max_order specifies the order where we begin to stop considering the
1829 * number of objects in a slab as critical. If we reach slub_max_order then
1830 * we try to keep the page order as low as possible. So we accept more waste
1831 * of space in favor of a small page order.
1833 * Higher order allocations also allow the placement of more objects in a
1834 * slab and thereby reduce object handling overhead. If the user has
1835 * requested a higher mininum order then we start with that one instead of
1836 * the smallest order which will fit the object.
1838 static inline int slab_order(int size
, int min_objects
,
1839 int max_order
, int fract_leftover
)
1843 int min_order
= slub_min_order
;
1845 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1846 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1848 for (order
= max(min_order
,
1849 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1850 order
<= max_order
; order
++) {
1852 unsigned long slab_size
= PAGE_SIZE
<< order
;
1854 if (slab_size
< min_objects
* size
)
1857 rem
= slab_size
% size
;
1859 if (rem
<= slab_size
/ fract_leftover
)
1867 static inline int calculate_order(int size
)
1875 * Attempt to find best configuration for a slab. This
1876 * works by first attempting to generate a layout with
1877 * the best configuration and backing off gradually.
1879 * First we reduce the acceptable waste in a slab. Then
1880 * we reduce the minimum objects required in a slab.
1882 min_objects
= slub_min_objects
;
1884 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1885 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1886 min_objects
= min(min_objects
, max_objects
);
1888 while (min_objects
> 1) {
1890 while (fraction
>= 4) {
1891 order
= slab_order(size
, min_objects
,
1892 slub_max_order
, fraction
);
1893 if (order
<= slub_max_order
)
1901 * We were unable to place multiple objects in a slab. Now
1902 * lets see if we can place a single object there.
1904 order
= slab_order(size
, 1, slub_max_order
, 1);
1905 if (order
<= slub_max_order
)
1909 * Doh this slab cannot be placed using slub_max_order.
1911 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1912 if (order
< MAX_ORDER
)
1918 * Figure out what the alignment of the objects will be.
1920 static unsigned long calculate_alignment(unsigned long flags
,
1921 unsigned long align
, unsigned long size
)
1924 * If the user wants hardware cache aligned objects then follow that
1925 * suggestion if the object is sufficiently large.
1927 * The hardware cache alignment cannot override the specified
1928 * alignment though. If that is greater then use it.
1930 if (flags
& SLAB_HWCACHE_ALIGN
) {
1931 unsigned long ralign
= cache_line_size();
1932 while (size
<= ralign
/ 2)
1934 align
= max(align
, ralign
);
1937 if (align
< ARCH_SLAB_MINALIGN
)
1938 align
= ARCH_SLAB_MINALIGN
;
1940 return ALIGN(align
, sizeof(void *));
1943 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1944 struct kmem_cache_cpu
*c
)
1949 c
->offset
= s
->offset
/ sizeof(void *);
1950 c
->objsize
= s
->objsize
;
1951 #ifdef CONFIG_SLUB_STATS
1952 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1957 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
1960 spin_lock_init(&n
->list_lock
);
1961 INIT_LIST_HEAD(&n
->partial
);
1962 #ifdef CONFIG_SLUB_DEBUG
1963 atomic_long_set(&n
->nr_slabs
, 0);
1964 atomic_long_set(&n
->total_objects
, 0);
1965 INIT_LIST_HEAD(&n
->full
);
1971 * Per cpu array for per cpu structures.
1973 * The per cpu array places all kmem_cache_cpu structures from one processor
1974 * close together meaning that it becomes possible that multiple per cpu
1975 * structures are contained in one cacheline. This may be particularly
1976 * beneficial for the kmalloc caches.
1978 * A desktop system typically has around 60-80 slabs. With 100 here we are
1979 * likely able to get per cpu structures for all caches from the array defined
1980 * here. We must be able to cover all kmalloc caches during bootstrap.
1982 * If the per cpu array is exhausted then fall back to kmalloc
1983 * of individual cachelines. No sharing is possible then.
1985 #define NR_KMEM_CACHE_CPU 100
1987 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
1988 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
1990 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
1991 static DECLARE_BITMAP(kmem_cach_cpu_free_init_once
, CONFIG_NR_CPUS
);
1993 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
1994 int cpu
, gfp_t flags
)
1996 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
1999 per_cpu(kmem_cache_cpu_free
, cpu
) =
2000 (void *)c
->freelist
;
2002 /* Table overflow: So allocate ourselves */
2004 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2005 flags
, cpu_to_node(cpu
));
2010 init_kmem_cache_cpu(s
, c
);
2014 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2016 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2017 c
>= per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2021 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2022 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2025 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2029 for_each_online_cpu(cpu
) {
2030 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2033 s
->cpu_slab
[cpu
] = NULL
;
2034 free_kmem_cache_cpu(c
, cpu
);
2039 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2043 for_each_online_cpu(cpu
) {
2044 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2049 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2051 free_kmem_cache_cpus(s
);
2054 s
->cpu_slab
[cpu
] = c
;
2060 * Initialize the per cpu array.
2062 static void init_alloc_cpu_cpu(int cpu
)
2066 if (cpumask_test_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
)))
2069 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2070 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2072 cpumask_set_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
));
2075 static void __init
init_alloc_cpu(void)
2079 for_each_online_cpu(cpu
)
2080 init_alloc_cpu_cpu(cpu
);
2084 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2085 static inline void init_alloc_cpu(void) {}
2087 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2089 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2096 * No kmalloc_node yet so do it by hand. We know that this is the first
2097 * slab on the node for this slabcache. There are no concurrent accesses
2100 * Note that this function only works on the kmalloc_node_cache
2101 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2102 * memory on a fresh node that has no slab structures yet.
2104 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2107 struct kmem_cache_node
*n
;
2108 unsigned long flags
;
2110 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2112 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2115 if (page_to_nid(page
) != node
) {
2116 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2118 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2119 "in order to be able to continue\n");
2124 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2126 kmalloc_caches
->node
[node
] = n
;
2127 #ifdef CONFIG_SLUB_DEBUG
2128 init_object(kmalloc_caches
, n
, 1);
2129 init_tracking(kmalloc_caches
, n
);
2131 init_kmem_cache_node(n
, kmalloc_caches
);
2132 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2135 * lockdep requires consistent irq usage for each lock
2136 * so even though there cannot be a race this early in
2137 * the boot sequence, we still disable irqs.
2139 local_irq_save(flags
);
2140 add_partial(n
, page
, 0);
2141 local_irq_restore(flags
);
2144 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2148 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2149 struct kmem_cache_node
*n
= s
->node
[node
];
2150 if (n
&& n
!= &s
->local_node
)
2151 kmem_cache_free(kmalloc_caches
, n
);
2152 s
->node
[node
] = NULL
;
2156 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2161 if (slab_state
>= UP
)
2162 local_node
= page_to_nid(virt_to_page(s
));
2166 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2167 struct kmem_cache_node
*n
;
2169 if (local_node
== node
)
2172 if (slab_state
== DOWN
) {
2173 early_kmem_cache_node_alloc(gfpflags
, node
);
2176 n
= kmem_cache_alloc_node(kmalloc_caches
,
2180 free_kmem_cache_nodes(s
);
2186 init_kmem_cache_node(n
, s
);
2191 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2195 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2197 init_kmem_cache_node(&s
->local_node
, s
);
2202 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2204 if (min
< MIN_PARTIAL
)
2206 else if (min
> MAX_PARTIAL
)
2208 s
->min_partial
= min
;
2212 * calculate_sizes() determines the order and the distribution of data within
2215 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2217 unsigned long flags
= s
->flags
;
2218 unsigned long size
= s
->objsize
;
2219 unsigned long align
= s
->align
;
2223 * Round up object size to the next word boundary. We can only
2224 * place the free pointer at word boundaries and this determines
2225 * the possible location of the free pointer.
2227 size
= ALIGN(size
, sizeof(void *));
2229 #ifdef CONFIG_SLUB_DEBUG
2231 * Determine if we can poison the object itself. If the user of
2232 * the slab may touch the object after free or before allocation
2233 * then we should never poison the object itself.
2235 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2237 s
->flags
|= __OBJECT_POISON
;
2239 s
->flags
&= ~__OBJECT_POISON
;
2243 * If we are Redzoning then check if there is some space between the
2244 * end of the object and the free pointer. If not then add an
2245 * additional word to have some bytes to store Redzone information.
2247 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2248 size
+= sizeof(void *);
2252 * With that we have determined the number of bytes in actual use
2253 * by the object. This is the potential offset to the free pointer.
2257 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2260 * Relocate free pointer after the object if it is not
2261 * permitted to overwrite the first word of the object on
2264 * This is the case if we do RCU, have a constructor or
2265 * destructor or are poisoning the objects.
2268 size
+= sizeof(void *);
2271 #ifdef CONFIG_SLUB_DEBUG
2272 if (flags
& SLAB_STORE_USER
)
2274 * Need to store information about allocs and frees after
2277 size
+= 2 * sizeof(struct track
);
2279 if (flags
& SLAB_RED_ZONE
)
2281 * Add some empty padding so that we can catch
2282 * overwrites from earlier objects rather than let
2283 * tracking information or the free pointer be
2284 * corrupted if a user writes before the start
2287 size
+= sizeof(void *);
2291 * Determine the alignment based on various parameters that the
2292 * user specified and the dynamic determination of cache line size
2295 align
= calculate_alignment(flags
, align
, s
->objsize
);
2298 * SLUB stores one object immediately after another beginning from
2299 * offset 0. In order to align the objects we have to simply size
2300 * each object to conform to the alignment.
2302 size
= ALIGN(size
, align
);
2304 if (forced_order
>= 0)
2305 order
= forced_order
;
2307 order
= calculate_order(size
);
2314 s
->allocflags
|= __GFP_COMP
;
2316 if (s
->flags
& SLAB_CACHE_DMA
)
2317 s
->allocflags
|= SLUB_DMA
;
2319 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2320 s
->allocflags
|= __GFP_RECLAIMABLE
;
2323 * Determine the number of objects per slab
2325 s
->oo
= oo_make(order
, size
);
2326 s
->min
= oo_make(get_order(size
), size
);
2327 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2330 return !!oo_objects(s
->oo
);
2334 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2335 const char *name
, size_t size
,
2336 size_t align
, unsigned long flags
,
2337 void (*ctor
)(void *))
2339 memset(s
, 0, kmem_size
);
2344 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2346 if (!calculate_sizes(s
, -1))
2350 * The larger the object size is, the more pages we want on the partial
2351 * list to avoid pounding the page allocator excessively.
2353 set_min_partial(s
, ilog2(s
->size
));
2356 s
->remote_node_defrag_ratio
= 1000;
2358 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2361 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2363 free_kmem_cache_nodes(s
);
2365 if (flags
& SLAB_PANIC
)
2366 panic("Cannot create slab %s size=%lu realsize=%u "
2367 "order=%u offset=%u flags=%lx\n",
2368 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2374 * Check if a given pointer is valid
2376 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2380 page
= get_object_page(object
);
2382 if (!page
|| s
!= page
->slab
)
2383 /* No slab or wrong slab */
2386 if (!check_valid_pointer(s
, page
, object
))
2390 * We could also check if the object is on the slabs freelist.
2391 * But this would be too expensive and it seems that the main
2392 * purpose of kmem_ptr_valid() is to check if the object belongs
2393 * to a certain slab.
2397 EXPORT_SYMBOL(kmem_ptr_validate
);
2400 * Determine the size of a slab object
2402 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2406 EXPORT_SYMBOL(kmem_cache_size
);
2408 const char *kmem_cache_name(struct kmem_cache
*s
)
2412 EXPORT_SYMBOL(kmem_cache_name
);
2414 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2417 #ifdef CONFIG_SLUB_DEBUG
2418 void *addr
= page_address(page
);
2420 DECLARE_BITMAP(map
, page
->objects
);
2422 bitmap_zero(map
, page
->objects
);
2423 slab_err(s
, page
, "%s", text
);
2425 for_each_free_object(p
, s
, page
->freelist
)
2426 set_bit(slab_index(p
, s
, addr
), map
);
2428 for_each_object(p
, s
, addr
, page
->objects
) {
2430 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2431 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2433 print_tracking(s
, p
);
2441 * Attempt to free all partial slabs on a node.
2443 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2445 unsigned long flags
;
2446 struct page
*page
, *h
;
2448 spin_lock_irqsave(&n
->list_lock
, flags
);
2449 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2451 list_del(&page
->lru
);
2452 discard_slab(s
, page
);
2455 list_slab_objects(s
, page
,
2456 "Objects remaining on kmem_cache_close()");
2459 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2463 * Release all resources used by a slab cache.
2465 static inline int kmem_cache_close(struct kmem_cache
*s
)
2471 /* Attempt to free all objects */
2472 free_kmem_cache_cpus(s
);
2473 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2474 struct kmem_cache_node
*n
= get_node(s
, node
);
2477 if (n
->nr_partial
|| slabs_node(s
, node
))
2480 free_kmem_cache_nodes(s
);
2485 * Close a cache and release the kmem_cache structure
2486 * (must be used for caches created using kmem_cache_create)
2488 void kmem_cache_destroy(struct kmem_cache
*s
)
2490 down_write(&slub_lock
);
2494 up_write(&slub_lock
);
2495 if (kmem_cache_close(s
)) {
2496 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2497 "still has objects.\n", s
->name
, __func__
);
2500 sysfs_slab_remove(s
);
2502 up_write(&slub_lock
);
2504 EXPORT_SYMBOL(kmem_cache_destroy
);
2506 /********************************************************************
2508 *******************************************************************/
2510 struct kmem_cache kmalloc_caches
[SLUB_PAGE_SHIFT
] __cacheline_aligned
;
2511 EXPORT_SYMBOL(kmalloc_caches
);
2513 static int __init
setup_slub_min_order(char *str
)
2515 get_option(&str
, &slub_min_order
);
2520 __setup("slub_min_order=", setup_slub_min_order
);
2522 static int __init
setup_slub_max_order(char *str
)
2524 get_option(&str
, &slub_max_order
);
2525 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2530 __setup("slub_max_order=", setup_slub_max_order
);
2532 static int __init
setup_slub_min_objects(char *str
)
2534 get_option(&str
, &slub_min_objects
);
2539 __setup("slub_min_objects=", setup_slub_min_objects
);
2541 static int __init
setup_slub_nomerge(char *str
)
2547 __setup("slub_nomerge", setup_slub_nomerge
);
2549 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2550 const char *name
, int size
, gfp_t gfp_flags
)
2552 unsigned int flags
= 0;
2554 if (gfp_flags
& SLUB_DMA
)
2555 flags
= SLAB_CACHE_DMA
;
2557 down_write(&slub_lock
);
2558 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2562 list_add(&s
->list
, &slab_caches
);
2563 up_write(&slub_lock
);
2564 if (sysfs_slab_add(s
))
2569 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2572 #ifdef CONFIG_ZONE_DMA
2573 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2575 static void sysfs_add_func(struct work_struct
*w
)
2577 struct kmem_cache
*s
;
2579 down_write(&slub_lock
);
2580 list_for_each_entry(s
, &slab_caches
, list
) {
2581 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2582 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2586 up_write(&slub_lock
);
2589 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2591 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2593 struct kmem_cache
*s
;
2597 s
= kmalloc_caches_dma
[index
];
2601 /* Dynamically create dma cache */
2602 if (flags
& __GFP_WAIT
)
2603 down_write(&slub_lock
);
2605 if (!down_write_trylock(&slub_lock
))
2609 if (kmalloc_caches_dma
[index
])
2612 realsize
= kmalloc_caches
[index
].objsize
;
2613 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2614 (unsigned int)realsize
);
2615 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2617 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2618 realsize
, ARCH_KMALLOC_MINALIGN
,
2619 SLAB_CACHE_DMA
|__SYSFS_ADD_DEFERRED
, NULL
)) {
2625 list_add(&s
->list
, &slab_caches
);
2626 kmalloc_caches_dma
[index
] = s
;
2628 schedule_work(&sysfs_add_work
);
2631 up_write(&slub_lock
);
2633 return kmalloc_caches_dma
[index
];
2638 * Conversion table for small slabs sizes / 8 to the index in the
2639 * kmalloc array. This is necessary for slabs < 192 since we have non power
2640 * of two cache sizes there. The size of larger slabs can be determined using
2643 static s8 size_index
[24] = {
2670 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2676 return ZERO_SIZE_PTR
;
2678 index
= size_index
[(size
- 1) / 8];
2680 index
= fls(size
- 1);
2682 #ifdef CONFIG_ZONE_DMA
2683 if (unlikely((flags
& SLUB_DMA
)))
2684 return dma_kmalloc_cache(index
, flags
);
2687 return &kmalloc_caches
[index
];
2690 void *__kmalloc(size_t size
, gfp_t flags
)
2692 struct kmem_cache
*s
;
2695 if (unlikely(size
> SLUB_MAX_SIZE
))
2696 return kmalloc_large(size
, flags
);
2698 s
= get_slab(size
, flags
);
2700 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2703 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2705 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2709 EXPORT_SYMBOL(__kmalloc
);
2711 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2713 struct page
*page
= alloc_pages_node(node
, flags
| __GFP_COMP
,
2717 return page_address(page
);
2723 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2725 struct kmem_cache
*s
;
2728 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2729 ret
= kmalloc_large_node(size
, flags
, node
);
2731 trace_kmalloc_node(_RET_IP_
, ret
,
2732 size
, PAGE_SIZE
<< get_order(size
),
2738 s
= get_slab(size
, flags
);
2740 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2743 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2745 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2749 EXPORT_SYMBOL(__kmalloc_node
);
2752 size_t ksize(const void *object
)
2755 struct kmem_cache
*s
;
2757 if (unlikely(object
== ZERO_SIZE_PTR
))
2760 page
= virt_to_head_page(object
);
2762 if (unlikely(!PageSlab(page
))) {
2763 WARN_ON(!PageCompound(page
));
2764 return PAGE_SIZE
<< compound_order(page
);
2768 #ifdef CONFIG_SLUB_DEBUG
2770 * Debugging requires use of the padding between object
2771 * and whatever may come after it.
2773 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2778 * If we have the need to store the freelist pointer
2779 * back there or track user information then we can
2780 * only use the space before that information.
2782 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2785 * Else we can use all the padding etc for the allocation
2789 EXPORT_SYMBOL(ksize
);
2791 void kfree(const void *x
)
2794 void *object
= (void *)x
;
2796 trace_kfree(_RET_IP_
, x
);
2798 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2801 page
= virt_to_head_page(x
);
2802 if (unlikely(!PageSlab(page
))) {
2803 BUG_ON(!PageCompound(page
));
2807 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2809 EXPORT_SYMBOL(kfree
);
2812 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2813 * the remaining slabs by the number of items in use. The slabs with the
2814 * most items in use come first. New allocations will then fill those up
2815 * and thus they can be removed from the partial lists.
2817 * The slabs with the least items are placed last. This results in them
2818 * being allocated from last increasing the chance that the last objects
2819 * are freed in them.
2821 int kmem_cache_shrink(struct kmem_cache
*s
)
2825 struct kmem_cache_node
*n
;
2828 int objects
= oo_objects(s
->max
);
2829 struct list_head
*slabs_by_inuse
=
2830 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2831 unsigned long flags
;
2833 if (!slabs_by_inuse
)
2837 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2838 n
= get_node(s
, node
);
2843 for (i
= 0; i
< objects
; i
++)
2844 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2846 spin_lock_irqsave(&n
->list_lock
, flags
);
2849 * Build lists indexed by the items in use in each slab.
2851 * Note that concurrent frees may occur while we hold the
2852 * list_lock. page->inuse here is the upper limit.
2854 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2855 if (!page
->inuse
&& slab_trylock(page
)) {
2857 * Must hold slab lock here because slab_free
2858 * may have freed the last object and be
2859 * waiting to release the slab.
2861 list_del(&page
->lru
);
2864 discard_slab(s
, page
);
2866 list_move(&page
->lru
,
2867 slabs_by_inuse
+ page
->inuse
);
2872 * Rebuild the partial list with the slabs filled up most
2873 * first and the least used slabs at the end.
2875 for (i
= objects
- 1; i
>= 0; i
--)
2876 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2878 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2881 kfree(slabs_by_inuse
);
2884 EXPORT_SYMBOL(kmem_cache_shrink
);
2886 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2887 static int slab_mem_going_offline_callback(void *arg
)
2889 struct kmem_cache
*s
;
2891 down_read(&slub_lock
);
2892 list_for_each_entry(s
, &slab_caches
, list
)
2893 kmem_cache_shrink(s
);
2894 up_read(&slub_lock
);
2899 static void slab_mem_offline_callback(void *arg
)
2901 struct kmem_cache_node
*n
;
2902 struct kmem_cache
*s
;
2903 struct memory_notify
*marg
= arg
;
2906 offline_node
= marg
->status_change_nid
;
2909 * If the node still has available memory. we need kmem_cache_node
2912 if (offline_node
< 0)
2915 down_read(&slub_lock
);
2916 list_for_each_entry(s
, &slab_caches
, list
) {
2917 n
= get_node(s
, offline_node
);
2920 * if n->nr_slabs > 0, slabs still exist on the node
2921 * that is going down. We were unable to free them,
2922 * and offline_pages() function shoudn't call this
2923 * callback. So, we must fail.
2925 BUG_ON(slabs_node(s
, offline_node
));
2927 s
->node
[offline_node
] = NULL
;
2928 kmem_cache_free(kmalloc_caches
, n
);
2931 up_read(&slub_lock
);
2934 static int slab_mem_going_online_callback(void *arg
)
2936 struct kmem_cache_node
*n
;
2937 struct kmem_cache
*s
;
2938 struct memory_notify
*marg
= arg
;
2939 int nid
= marg
->status_change_nid
;
2943 * If the node's memory is already available, then kmem_cache_node is
2944 * already created. Nothing to do.
2950 * We are bringing a node online. No memory is available yet. We must
2951 * allocate a kmem_cache_node structure in order to bring the node
2954 down_read(&slub_lock
);
2955 list_for_each_entry(s
, &slab_caches
, list
) {
2957 * XXX: kmem_cache_alloc_node will fallback to other nodes
2958 * since memory is not yet available from the node that
2961 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2966 init_kmem_cache_node(n
, s
);
2970 up_read(&slub_lock
);
2974 static int slab_memory_callback(struct notifier_block
*self
,
2975 unsigned long action
, void *arg
)
2980 case MEM_GOING_ONLINE
:
2981 ret
= slab_mem_going_online_callback(arg
);
2983 case MEM_GOING_OFFLINE
:
2984 ret
= slab_mem_going_offline_callback(arg
);
2987 case MEM_CANCEL_ONLINE
:
2988 slab_mem_offline_callback(arg
);
2991 case MEM_CANCEL_OFFLINE
:
2995 ret
= notifier_from_errno(ret
);
3001 #endif /* CONFIG_MEMORY_HOTPLUG */
3003 /********************************************************************
3004 * Basic setup of slabs
3005 *******************************************************************/
3007 void __init
kmem_cache_init(void)
3016 * Must first have the slab cache available for the allocations of the
3017 * struct kmem_cache_node's. There is special bootstrap code in
3018 * kmem_cache_open for slab_state == DOWN.
3020 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3021 sizeof(struct kmem_cache_node
), GFP_KERNEL
);
3022 kmalloc_caches
[0].refcount
= -1;
3025 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3028 /* Able to allocate the per node structures */
3029 slab_state
= PARTIAL
;
3031 /* Caches that are not of the two-to-the-power-of size */
3032 if (KMALLOC_MIN_SIZE
<= 64) {
3033 create_kmalloc_cache(&kmalloc_caches
[1],
3034 "kmalloc-96", 96, GFP_KERNEL
);
3036 create_kmalloc_cache(&kmalloc_caches
[2],
3037 "kmalloc-192", 192, GFP_KERNEL
);
3041 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3042 create_kmalloc_cache(&kmalloc_caches
[i
],
3043 "kmalloc", 1 << i
, GFP_KERNEL
);
3049 * Patch up the size_index table if we have strange large alignment
3050 * requirements for the kmalloc array. This is only the case for
3051 * MIPS it seems. The standard arches will not generate any code here.
3053 * Largest permitted alignment is 256 bytes due to the way we
3054 * handle the index determination for the smaller caches.
3056 * Make sure that nothing crazy happens if someone starts tinkering
3057 * around with ARCH_KMALLOC_MINALIGN
3059 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3060 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3062 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3063 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3065 if (KMALLOC_MIN_SIZE
== 128) {
3067 * The 192 byte sized cache is not used if the alignment
3068 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3071 for (i
= 128 + 8; i
<= 192; i
+= 8)
3072 size_index
[(i
- 1) / 8] = 8;
3077 /* Provide the correct kmalloc names now that the caches are up */
3078 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3079 kmalloc_caches
[i
]. name
=
3080 kasprintf(GFP_KERNEL
, "kmalloc-%d", 1 << i
);
3083 register_cpu_notifier(&slab_notifier
);
3084 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3085 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3087 kmem_size
= sizeof(struct kmem_cache
);
3091 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3092 " CPUs=%d, Nodes=%d\n",
3093 caches
, cache_line_size(),
3094 slub_min_order
, slub_max_order
, slub_min_objects
,
3095 nr_cpu_ids
, nr_node_ids
);
3099 * Find a mergeable slab cache
3101 static int slab_unmergeable(struct kmem_cache
*s
)
3103 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3110 * We may have set a slab to be unmergeable during bootstrap.
3112 if (s
->refcount
< 0)
3118 static struct kmem_cache
*find_mergeable(size_t size
,
3119 size_t align
, unsigned long flags
, const char *name
,
3120 void (*ctor
)(void *))
3122 struct kmem_cache
*s
;
3124 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3130 size
= ALIGN(size
, sizeof(void *));
3131 align
= calculate_alignment(flags
, align
, size
);
3132 size
= ALIGN(size
, align
);
3133 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3135 list_for_each_entry(s
, &slab_caches
, list
) {
3136 if (slab_unmergeable(s
))
3142 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3145 * Check if alignment is compatible.
3146 * Courtesy of Adrian Drzewiecki
3148 if ((s
->size
& ~(align
- 1)) != s
->size
)
3151 if (s
->size
- size
>= sizeof(void *))
3159 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3160 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3162 struct kmem_cache
*s
;
3164 down_write(&slub_lock
);
3165 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3171 * Adjust the object sizes so that we clear
3172 * the complete object on kzalloc.
3174 s
->objsize
= max(s
->objsize
, (int)size
);
3177 * And then we need to update the object size in the
3178 * per cpu structures
3180 for_each_online_cpu(cpu
)
3181 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3183 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3184 up_write(&slub_lock
);
3186 if (sysfs_slab_alias(s
, name
)) {
3187 down_write(&slub_lock
);
3189 up_write(&slub_lock
);
3195 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3197 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3198 size
, align
, flags
, ctor
)) {
3199 list_add(&s
->list
, &slab_caches
);
3200 up_write(&slub_lock
);
3201 if (sysfs_slab_add(s
)) {
3202 down_write(&slub_lock
);
3204 up_write(&slub_lock
);
3212 up_write(&slub_lock
);
3215 if (flags
& SLAB_PANIC
)
3216 panic("Cannot create slabcache %s\n", name
);
3221 EXPORT_SYMBOL(kmem_cache_create
);
3225 * Use the cpu notifier to insure that the cpu slabs are flushed when
3228 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3229 unsigned long action
, void *hcpu
)
3231 long cpu
= (long)hcpu
;
3232 struct kmem_cache
*s
;
3233 unsigned long flags
;
3236 case CPU_UP_PREPARE
:
3237 case CPU_UP_PREPARE_FROZEN
:
3238 init_alloc_cpu_cpu(cpu
);
3239 down_read(&slub_lock
);
3240 list_for_each_entry(s
, &slab_caches
, list
)
3241 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3243 up_read(&slub_lock
);
3246 case CPU_UP_CANCELED
:
3247 case CPU_UP_CANCELED_FROZEN
:
3249 case CPU_DEAD_FROZEN
:
3250 down_read(&slub_lock
);
3251 list_for_each_entry(s
, &slab_caches
, list
) {
3252 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3254 local_irq_save(flags
);
3255 __flush_cpu_slab(s
, cpu
);
3256 local_irq_restore(flags
);
3257 free_kmem_cache_cpu(c
, cpu
);
3258 s
->cpu_slab
[cpu
] = NULL
;
3260 up_read(&slub_lock
);
3268 static struct notifier_block __cpuinitdata slab_notifier
= {
3269 .notifier_call
= slab_cpuup_callback
3274 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3276 struct kmem_cache
*s
;
3279 if (unlikely(size
> SLUB_MAX_SIZE
))
3280 return kmalloc_large(size
, gfpflags
);
3282 s
= get_slab(size
, gfpflags
);
3284 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3287 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3289 /* Honor the call site pointer we recieved. */
3290 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3295 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3296 int node
, unsigned long caller
)
3298 struct kmem_cache
*s
;
3301 if (unlikely(size
> SLUB_MAX_SIZE
))
3302 return kmalloc_large_node(size
, gfpflags
, node
);
3304 s
= get_slab(size
, gfpflags
);
3306 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3309 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3311 /* Honor the call site pointer we recieved. */
3312 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3317 #ifdef CONFIG_SLUB_DEBUG
3318 static unsigned long count_partial(struct kmem_cache_node
*n
,
3319 int (*get_count
)(struct page
*))
3321 unsigned long flags
;
3322 unsigned long x
= 0;
3325 spin_lock_irqsave(&n
->list_lock
, flags
);
3326 list_for_each_entry(page
, &n
->partial
, lru
)
3327 x
+= get_count(page
);
3328 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3332 static int count_inuse(struct page
*page
)
3337 static int count_total(struct page
*page
)
3339 return page
->objects
;
3342 static int count_free(struct page
*page
)
3344 return page
->objects
- page
->inuse
;
3347 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3351 void *addr
= page_address(page
);
3353 if (!check_slab(s
, page
) ||
3354 !on_freelist(s
, page
, NULL
))
3357 /* Now we know that a valid freelist exists */
3358 bitmap_zero(map
, page
->objects
);
3360 for_each_free_object(p
, s
, page
->freelist
) {
3361 set_bit(slab_index(p
, s
, addr
), map
);
3362 if (!check_object(s
, page
, p
, 0))
3366 for_each_object(p
, s
, addr
, page
->objects
)
3367 if (!test_bit(slab_index(p
, s
, addr
), map
))
3368 if (!check_object(s
, page
, p
, 1))
3373 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3376 if (slab_trylock(page
)) {
3377 validate_slab(s
, page
, map
);
3380 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3383 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3384 if (!PageSlubDebug(page
))
3385 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3386 "on slab 0x%p\n", s
->name
, page
);
3388 if (PageSlubDebug(page
))
3389 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3390 "slab 0x%p\n", s
->name
, page
);
3394 static int validate_slab_node(struct kmem_cache
*s
,
3395 struct kmem_cache_node
*n
, unsigned long *map
)
3397 unsigned long count
= 0;
3399 unsigned long flags
;
3401 spin_lock_irqsave(&n
->list_lock
, flags
);
3403 list_for_each_entry(page
, &n
->partial
, lru
) {
3404 validate_slab_slab(s
, page
, map
);
3407 if (count
!= n
->nr_partial
)
3408 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3409 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3411 if (!(s
->flags
& SLAB_STORE_USER
))
3414 list_for_each_entry(page
, &n
->full
, lru
) {
3415 validate_slab_slab(s
, page
, map
);
3418 if (count
!= atomic_long_read(&n
->nr_slabs
))
3419 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3420 "counter=%ld\n", s
->name
, count
,
3421 atomic_long_read(&n
->nr_slabs
));
3424 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3428 static long validate_slab_cache(struct kmem_cache
*s
)
3431 unsigned long count
= 0;
3432 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3433 sizeof(unsigned long), GFP_KERNEL
);
3439 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3440 struct kmem_cache_node
*n
= get_node(s
, node
);
3442 count
+= validate_slab_node(s
, n
, map
);
3448 #ifdef SLUB_RESILIENCY_TEST
3449 static void resiliency_test(void)
3453 printk(KERN_ERR
"SLUB resiliency testing\n");
3454 printk(KERN_ERR
"-----------------------\n");
3455 printk(KERN_ERR
"A. Corruption after allocation\n");
3457 p
= kzalloc(16, GFP_KERNEL
);
3459 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3460 " 0x12->0x%p\n\n", p
+ 16);
3462 validate_slab_cache(kmalloc_caches
+ 4);
3464 /* Hmmm... The next two are dangerous */
3465 p
= kzalloc(32, GFP_KERNEL
);
3466 p
[32 + sizeof(void *)] = 0x34;
3467 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3468 " 0x34 -> -0x%p\n", p
);
3470 "If allocated object is overwritten then not detectable\n\n");
3472 validate_slab_cache(kmalloc_caches
+ 5);
3473 p
= kzalloc(64, GFP_KERNEL
);
3474 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3476 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3479 "If allocated object is overwritten then not detectable\n\n");
3480 validate_slab_cache(kmalloc_caches
+ 6);
3482 printk(KERN_ERR
"\nB. Corruption after free\n");
3483 p
= kzalloc(128, GFP_KERNEL
);
3486 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3487 validate_slab_cache(kmalloc_caches
+ 7);
3489 p
= kzalloc(256, GFP_KERNEL
);
3492 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3494 validate_slab_cache(kmalloc_caches
+ 8);
3496 p
= kzalloc(512, GFP_KERNEL
);
3499 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3500 validate_slab_cache(kmalloc_caches
+ 9);
3503 static void resiliency_test(void) {};
3507 * Generate lists of code addresses where slabcache objects are allocated
3512 unsigned long count
;
3519 DECLARE_BITMAP(cpus
, NR_CPUS
);
3525 unsigned long count
;
3526 struct location
*loc
;
3529 static void free_loc_track(struct loc_track
*t
)
3532 free_pages((unsigned long)t
->loc
,
3533 get_order(sizeof(struct location
) * t
->max
));
3536 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3541 order
= get_order(sizeof(struct location
) * max
);
3543 l
= (void *)__get_free_pages(flags
, order
);
3548 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3556 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3557 const struct track
*track
)
3559 long start
, end
, pos
;
3561 unsigned long caddr
;
3562 unsigned long age
= jiffies
- track
->when
;
3568 pos
= start
+ (end
- start
+ 1) / 2;
3571 * There is nothing at "end". If we end up there
3572 * we need to add something to before end.
3577 caddr
= t
->loc
[pos
].addr
;
3578 if (track
->addr
== caddr
) {
3584 if (age
< l
->min_time
)
3586 if (age
> l
->max_time
)
3589 if (track
->pid
< l
->min_pid
)
3590 l
->min_pid
= track
->pid
;
3591 if (track
->pid
> l
->max_pid
)
3592 l
->max_pid
= track
->pid
;
3594 cpumask_set_cpu(track
->cpu
,
3595 to_cpumask(l
->cpus
));
3597 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3601 if (track
->addr
< caddr
)
3608 * Not found. Insert new tracking element.
3610 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3616 (t
->count
- pos
) * sizeof(struct location
));
3619 l
->addr
= track
->addr
;
3623 l
->min_pid
= track
->pid
;
3624 l
->max_pid
= track
->pid
;
3625 cpumask_clear(to_cpumask(l
->cpus
));
3626 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3627 nodes_clear(l
->nodes
);
3628 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3632 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3633 struct page
*page
, enum track_item alloc
)
3635 void *addr
= page_address(page
);
3636 DECLARE_BITMAP(map
, page
->objects
);
3639 bitmap_zero(map
, page
->objects
);
3640 for_each_free_object(p
, s
, page
->freelist
)
3641 set_bit(slab_index(p
, s
, addr
), map
);
3643 for_each_object(p
, s
, addr
, page
->objects
)
3644 if (!test_bit(slab_index(p
, s
, addr
), map
))
3645 add_location(t
, s
, get_track(s
, p
, alloc
));
3648 static int list_locations(struct kmem_cache
*s
, char *buf
,
3649 enum track_item alloc
)
3653 struct loc_track t
= { 0, 0, NULL
};
3656 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3658 return sprintf(buf
, "Out of memory\n");
3660 /* Push back cpu slabs */
3663 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3664 struct kmem_cache_node
*n
= get_node(s
, node
);
3665 unsigned long flags
;
3668 if (!atomic_long_read(&n
->nr_slabs
))
3671 spin_lock_irqsave(&n
->list_lock
, flags
);
3672 list_for_each_entry(page
, &n
->partial
, lru
)
3673 process_slab(&t
, s
, page
, alloc
);
3674 list_for_each_entry(page
, &n
->full
, lru
)
3675 process_slab(&t
, s
, page
, alloc
);
3676 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3679 for (i
= 0; i
< t
.count
; i
++) {
3680 struct location
*l
= &t
.loc
[i
];
3682 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3684 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3687 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3689 len
+= sprintf(buf
+ len
, "<not-available>");
3691 if (l
->sum_time
!= l
->min_time
) {
3692 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3694 (long)div_u64(l
->sum_time
, l
->count
),
3697 len
+= sprintf(buf
+ len
, " age=%ld",
3700 if (l
->min_pid
!= l
->max_pid
)
3701 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3702 l
->min_pid
, l
->max_pid
);
3704 len
+= sprintf(buf
+ len
, " pid=%ld",
3707 if (num_online_cpus() > 1 &&
3708 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3709 len
< PAGE_SIZE
- 60) {
3710 len
+= sprintf(buf
+ len
, " cpus=");
3711 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3712 to_cpumask(l
->cpus
));
3715 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3716 len
< PAGE_SIZE
- 60) {
3717 len
+= sprintf(buf
+ len
, " nodes=");
3718 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3722 len
+= sprintf(buf
+ len
, "\n");
3727 len
+= sprintf(buf
, "No data\n");
3731 enum slab_stat_type
{
3732 SL_ALL
, /* All slabs */
3733 SL_PARTIAL
, /* Only partially allocated slabs */
3734 SL_CPU
, /* Only slabs used for cpu caches */
3735 SL_OBJECTS
, /* Determine allocated objects not slabs */
3736 SL_TOTAL
/* Determine object capacity not slabs */
3739 #define SO_ALL (1 << SL_ALL)
3740 #define SO_PARTIAL (1 << SL_PARTIAL)
3741 #define SO_CPU (1 << SL_CPU)
3742 #define SO_OBJECTS (1 << SL_OBJECTS)
3743 #define SO_TOTAL (1 << SL_TOTAL)
3745 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3746 char *buf
, unsigned long flags
)
3748 unsigned long total
= 0;
3751 unsigned long *nodes
;
3752 unsigned long *per_cpu
;
3754 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3757 per_cpu
= nodes
+ nr_node_ids
;
3759 if (flags
& SO_CPU
) {
3762 for_each_possible_cpu(cpu
) {
3763 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3765 if (!c
|| c
->node
< 0)
3769 if (flags
& SO_TOTAL
)
3770 x
= c
->page
->objects
;
3771 else if (flags
& SO_OBJECTS
)
3777 nodes
[c
->node
] += x
;
3783 if (flags
& SO_ALL
) {
3784 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3785 struct kmem_cache_node
*n
= get_node(s
, node
);
3787 if (flags
& SO_TOTAL
)
3788 x
= atomic_long_read(&n
->total_objects
);
3789 else if (flags
& SO_OBJECTS
)
3790 x
= atomic_long_read(&n
->total_objects
) -
3791 count_partial(n
, count_free
);
3794 x
= atomic_long_read(&n
->nr_slabs
);
3799 } else if (flags
& SO_PARTIAL
) {
3800 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3801 struct kmem_cache_node
*n
= get_node(s
, node
);
3803 if (flags
& SO_TOTAL
)
3804 x
= count_partial(n
, count_total
);
3805 else if (flags
& SO_OBJECTS
)
3806 x
= count_partial(n
, count_inuse
);
3813 x
= sprintf(buf
, "%lu", total
);
3815 for_each_node_state(node
, N_NORMAL_MEMORY
)
3817 x
+= sprintf(buf
+ x
, " N%d=%lu",
3821 return x
+ sprintf(buf
+ x
, "\n");
3824 static int any_slab_objects(struct kmem_cache
*s
)
3828 for_each_online_node(node
) {
3829 struct kmem_cache_node
*n
= get_node(s
, node
);
3834 if (atomic_long_read(&n
->total_objects
))
3840 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3841 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3843 struct slab_attribute
{
3844 struct attribute attr
;
3845 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3846 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3849 #define SLAB_ATTR_RO(_name) \
3850 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3852 #define SLAB_ATTR(_name) \
3853 static struct slab_attribute _name##_attr = \
3854 __ATTR(_name, 0644, _name##_show, _name##_store)
3856 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3858 return sprintf(buf
, "%d\n", s
->size
);
3860 SLAB_ATTR_RO(slab_size
);
3862 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3864 return sprintf(buf
, "%d\n", s
->align
);
3866 SLAB_ATTR_RO(align
);
3868 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3870 return sprintf(buf
, "%d\n", s
->objsize
);
3872 SLAB_ATTR_RO(object_size
);
3874 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3876 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3878 SLAB_ATTR_RO(objs_per_slab
);
3880 static ssize_t
order_store(struct kmem_cache
*s
,
3881 const char *buf
, size_t length
)
3883 unsigned long order
;
3886 err
= strict_strtoul(buf
, 10, &order
);
3890 if (order
> slub_max_order
|| order
< slub_min_order
)
3893 calculate_sizes(s
, order
);
3897 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3899 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3903 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3905 return sprintf(buf
, "%lu\n", s
->min_partial
);
3908 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3914 err
= strict_strtoul(buf
, 10, &min
);
3918 set_min_partial(s
, min
);
3921 SLAB_ATTR(min_partial
);
3923 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3926 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3928 return n
+ sprintf(buf
+ n
, "\n");
3934 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3936 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3938 SLAB_ATTR_RO(aliases
);
3940 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3942 return show_slab_objects(s
, buf
, SO_ALL
);
3944 SLAB_ATTR_RO(slabs
);
3946 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3948 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3950 SLAB_ATTR_RO(partial
);
3952 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3954 return show_slab_objects(s
, buf
, SO_CPU
);
3956 SLAB_ATTR_RO(cpu_slabs
);
3958 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3960 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3962 SLAB_ATTR_RO(objects
);
3964 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3966 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3968 SLAB_ATTR_RO(objects_partial
);
3970 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3972 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3974 SLAB_ATTR_RO(total_objects
);
3976 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3978 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3981 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3982 const char *buf
, size_t length
)
3984 s
->flags
&= ~SLAB_DEBUG_FREE
;
3986 s
->flags
|= SLAB_DEBUG_FREE
;
3989 SLAB_ATTR(sanity_checks
);
3991 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
3993 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
3996 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
3999 s
->flags
&= ~SLAB_TRACE
;
4001 s
->flags
|= SLAB_TRACE
;
4006 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4008 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4011 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4012 const char *buf
, size_t length
)
4014 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4016 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4019 SLAB_ATTR(reclaim_account
);
4021 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4023 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4025 SLAB_ATTR_RO(hwcache_align
);
4027 #ifdef CONFIG_ZONE_DMA
4028 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4030 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4032 SLAB_ATTR_RO(cache_dma
);
4035 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4037 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4039 SLAB_ATTR_RO(destroy_by_rcu
);
4041 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4043 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4046 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4047 const char *buf
, size_t length
)
4049 if (any_slab_objects(s
))
4052 s
->flags
&= ~SLAB_RED_ZONE
;
4054 s
->flags
|= SLAB_RED_ZONE
;
4055 calculate_sizes(s
, -1);
4058 SLAB_ATTR(red_zone
);
4060 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4062 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4065 static ssize_t
poison_store(struct kmem_cache
*s
,
4066 const char *buf
, size_t length
)
4068 if (any_slab_objects(s
))
4071 s
->flags
&= ~SLAB_POISON
;
4073 s
->flags
|= SLAB_POISON
;
4074 calculate_sizes(s
, -1);
4079 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4081 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4084 static ssize_t
store_user_store(struct kmem_cache
*s
,
4085 const char *buf
, size_t length
)
4087 if (any_slab_objects(s
))
4090 s
->flags
&= ~SLAB_STORE_USER
;
4092 s
->flags
|= SLAB_STORE_USER
;
4093 calculate_sizes(s
, -1);
4096 SLAB_ATTR(store_user
);
4098 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4103 static ssize_t
validate_store(struct kmem_cache
*s
,
4104 const char *buf
, size_t length
)
4108 if (buf
[0] == '1') {
4109 ret
= validate_slab_cache(s
);
4115 SLAB_ATTR(validate
);
4117 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4122 static ssize_t
shrink_store(struct kmem_cache
*s
,
4123 const char *buf
, size_t length
)
4125 if (buf
[0] == '1') {
4126 int rc
= kmem_cache_shrink(s
);
4136 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4138 if (!(s
->flags
& SLAB_STORE_USER
))
4140 return list_locations(s
, buf
, TRACK_ALLOC
);
4142 SLAB_ATTR_RO(alloc_calls
);
4144 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4146 if (!(s
->flags
& SLAB_STORE_USER
))
4148 return list_locations(s
, buf
, TRACK_FREE
);
4150 SLAB_ATTR_RO(free_calls
);
4153 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4155 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4158 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4159 const char *buf
, size_t length
)
4161 unsigned long ratio
;
4164 err
= strict_strtoul(buf
, 10, &ratio
);
4169 s
->remote_node_defrag_ratio
= ratio
* 10;
4173 SLAB_ATTR(remote_node_defrag_ratio
);
4176 #ifdef CONFIG_SLUB_STATS
4177 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4179 unsigned long sum
= 0;
4182 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4187 for_each_online_cpu(cpu
) {
4188 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4194 len
= sprintf(buf
, "%lu", sum
);
4197 for_each_online_cpu(cpu
) {
4198 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4199 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4203 return len
+ sprintf(buf
+ len
, "\n");
4206 #define STAT_ATTR(si, text) \
4207 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4209 return show_stat(s, buf, si); \
4211 SLAB_ATTR_RO(text); \
4213 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4214 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4215 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4216 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4217 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4218 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4219 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4220 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4221 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4222 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4223 STAT_ATTR(FREE_SLAB
, free_slab
);
4224 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4225 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4226 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4227 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4228 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4229 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4230 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4233 static struct attribute
*slab_attrs
[] = {
4234 &slab_size_attr
.attr
,
4235 &object_size_attr
.attr
,
4236 &objs_per_slab_attr
.attr
,
4238 &min_partial_attr
.attr
,
4240 &objects_partial_attr
.attr
,
4241 &total_objects_attr
.attr
,
4244 &cpu_slabs_attr
.attr
,
4248 &sanity_checks_attr
.attr
,
4250 &hwcache_align_attr
.attr
,
4251 &reclaim_account_attr
.attr
,
4252 &destroy_by_rcu_attr
.attr
,
4253 &red_zone_attr
.attr
,
4255 &store_user_attr
.attr
,
4256 &validate_attr
.attr
,
4258 &alloc_calls_attr
.attr
,
4259 &free_calls_attr
.attr
,
4260 #ifdef CONFIG_ZONE_DMA
4261 &cache_dma_attr
.attr
,
4264 &remote_node_defrag_ratio_attr
.attr
,
4266 #ifdef CONFIG_SLUB_STATS
4267 &alloc_fastpath_attr
.attr
,
4268 &alloc_slowpath_attr
.attr
,
4269 &free_fastpath_attr
.attr
,
4270 &free_slowpath_attr
.attr
,
4271 &free_frozen_attr
.attr
,
4272 &free_add_partial_attr
.attr
,
4273 &free_remove_partial_attr
.attr
,
4274 &alloc_from_partial_attr
.attr
,
4275 &alloc_slab_attr
.attr
,
4276 &alloc_refill_attr
.attr
,
4277 &free_slab_attr
.attr
,
4278 &cpuslab_flush_attr
.attr
,
4279 &deactivate_full_attr
.attr
,
4280 &deactivate_empty_attr
.attr
,
4281 &deactivate_to_head_attr
.attr
,
4282 &deactivate_to_tail_attr
.attr
,
4283 &deactivate_remote_frees_attr
.attr
,
4284 &order_fallback_attr
.attr
,
4289 static struct attribute_group slab_attr_group
= {
4290 .attrs
= slab_attrs
,
4293 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4294 struct attribute
*attr
,
4297 struct slab_attribute
*attribute
;
4298 struct kmem_cache
*s
;
4301 attribute
= to_slab_attr(attr
);
4304 if (!attribute
->show
)
4307 err
= attribute
->show(s
, buf
);
4312 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4313 struct attribute
*attr
,
4314 const char *buf
, size_t len
)
4316 struct slab_attribute
*attribute
;
4317 struct kmem_cache
*s
;
4320 attribute
= to_slab_attr(attr
);
4323 if (!attribute
->store
)
4326 err
= attribute
->store(s
, buf
, len
);
4331 static void kmem_cache_release(struct kobject
*kobj
)
4333 struct kmem_cache
*s
= to_slab(kobj
);
4338 static struct sysfs_ops slab_sysfs_ops
= {
4339 .show
= slab_attr_show
,
4340 .store
= slab_attr_store
,
4343 static struct kobj_type slab_ktype
= {
4344 .sysfs_ops
= &slab_sysfs_ops
,
4345 .release
= kmem_cache_release
4348 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4350 struct kobj_type
*ktype
= get_ktype(kobj
);
4352 if (ktype
== &slab_ktype
)
4357 static struct kset_uevent_ops slab_uevent_ops
= {
4358 .filter
= uevent_filter
,
4361 static struct kset
*slab_kset
;
4363 #define ID_STR_LENGTH 64
4365 /* Create a unique string id for a slab cache:
4367 * Format :[flags-]size
4369 static char *create_unique_id(struct kmem_cache
*s
)
4371 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4378 * First flags affecting slabcache operations. We will only
4379 * get here for aliasable slabs so we do not need to support
4380 * too many flags. The flags here must cover all flags that
4381 * are matched during merging to guarantee that the id is
4384 if (s
->flags
& SLAB_CACHE_DMA
)
4386 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4388 if (s
->flags
& SLAB_DEBUG_FREE
)
4392 p
+= sprintf(p
, "%07d", s
->size
);
4393 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4397 static int sysfs_slab_add(struct kmem_cache
*s
)
4403 if (slab_state
< SYSFS
)
4404 /* Defer until later */
4407 unmergeable
= slab_unmergeable(s
);
4410 * Slabcache can never be merged so we can use the name proper.
4411 * This is typically the case for debug situations. In that
4412 * case we can catch duplicate names easily.
4414 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4418 * Create a unique name for the slab as a target
4421 name
= create_unique_id(s
);
4424 s
->kobj
.kset
= slab_kset
;
4425 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4427 kobject_put(&s
->kobj
);
4431 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4434 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4436 /* Setup first alias */
4437 sysfs_slab_alias(s
, s
->name
);
4443 static void sysfs_slab_remove(struct kmem_cache
*s
)
4445 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4446 kobject_del(&s
->kobj
);
4447 kobject_put(&s
->kobj
);
4451 * Need to buffer aliases during bootup until sysfs becomes
4452 * available lest we lose that information.
4454 struct saved_alias
{
4455 struct kmem_cache
*s
;
4457 struct saved_alias
*next
;
4460 static struct saved_alias
*alias_list
;
4462 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4464 struct saved_alias
*al
;
4466 if (slab_state
== SYSFS
) {
4468 * If we have a leftover link then remove it.
4470 sysfs_remove_link(&slab_kset
->kobj
, name
);
4471 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4474 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4480 al
->next
= alias_list
;
4485 static int __init
slab_sysfs_init(void)
4487 struct kmem_cache
*s
;
4490 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4492 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4498 list_for_each_entry(s
, &slab_caches
, list
) {
4499 err
= sysfs_slab_add(s
);
4501 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4502 " to sysfs\n", s
->name
);
4505 while (alias_list
) {
4506 struct saved_alias
*al
= alias_list
;
4508 alias_list
= alias_list
->next
;
4509 err
= sysfs_slab_alias(al
->s
, al
->name
);
4511 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4512 " %s to sysfs\n", s
->name
);
4520 __initcall(slab_sysfs_init
);
4524 * The /proc/slabinfo ABI
4526 #ifdef CONFIG_SLABINFO
4527 static void print_slabinfo_header(struct seq_file
*m
)
4529 seq_puts(m
, "slabinfo - version: 2.1\n");
4530 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4531 "<objperslab> <pagesperslab>");
4532 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4533 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4537 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4541 down_read(&slub_lock
);
4543 print_slabinfo_header(m
);
4545 return seq_list_start(&slab_caches
, *pos
);
4548 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4550 return seq_list_next(p
, &slab_caches
, pos
);
4553 static void s_stop(struct seq_file
*m
, void *p
)
4555 up_read(&slub_lock
);
4558 static int s_show(struct seq_file
*m
, void *p
)
4560 unsigned long nr_partials
= 0;
4561 unsigned long nr_slabs
= 0;
4562 unsigned long nr_inuse
= 0;
4563 unsigned long nr_objs
= 0;
4564 unsigned long nr_free
= 0;
4565 struct kmem_cache
*s
;
4568 s
= list_entry(p
, struct kmem_cache
, list
);
4570 for_each_online_node(node
) {
4571 struct kmem_cache_node
*n
= get_node(s
, node
);
4576 nr_partials
+= n
->nr_partial
;
4577 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4578 nr_objs
+= atomic_long_read(&n
->total_objects
);
4579 nr_free
+= count_partial(n
, count_free
);
4582 nr_inuse
= nr_objs
- nr_free
;
4584 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4585 nr_objs
, s
->size
, oo_objects(s
->oo
),
4586 (1 << oo_order(s
->oo
)));
4587 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4588 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4594 static const struct seq_operations slabinfo_op
= {
4601 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4603 return seq_open(file
, &slabinfo_op
);
4606 static const struct file_operations proc_slabinfo_operations
= {
4607 .open
= slabinfo_open
,
4609 .llseek
= seq_lseek
,
4610 .release
= seq_release
,
4613 static int __init
slab_proc_init(void)
4615 proc_create("slabinfo",S_IWUSR
|S_IRUGO
,NULL
,&proc_slabinfo_operations
);
4618 module_init(slab_proc_init
);
4619 #endif /* CONFIG_SLABINFO */