2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemcheck.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/mempolicy.h>
24 #include <linux/ctype.h>
25 #include <linux/debugobjects.h>
26 #include <linux/kallsyms.h>
27 #include <linux/memory.h>
28 #include <linux/math64.h>
29 #include <linux/fault-inject.h>
36 * The slab_lock protects operations on the object of a particular
37 * slab and its metadata in the page struct. If the slab lock
38 * has been taken then no allocations nor frees can be performed
39 * on the objects in the slab nor can the slab be added or removed
40 * from the partial or full lists since this would mean modifying
41 * the page_struct of the slab.
43 * The list_lock protects the partial and full list on each node and
44 * the partial slab counter. If taken then no new slabs may be added or
45 * removed from the lists nor make the number of partial slabs be modified.
46 * (Note that the total number of slabs is an atomic value that may be
47 * modified without taking the list lock).
49 * The list_lock is a centralized lock and thus we avoid taking it as
50 * much as possible. As long as SLUB does not have to handle partial
51 * slabs, operations can continue without any centralized lock. F.e.
52 * allocating a long series of objects that fill up slabs does not require
55 * The lock order is sometimes inverted when we are trying to get a slab
56 * off a list. We take the list_lock and then look for a page on the list
57 * to use. While we do that objects in the slabs may be freed. We can
58 * only operate on the slab if we have also taken the slab_lock. So we use
59 * a slab_trylock() on the slab. If trylock was successful then no frees
60 * can occur anymore and we can use the slab for allocations etc. If the
61 * slab_trylock() does not succeed then frees are in progress in the slab and
62 * we must stay away from it for a while since we may cause a bouncing
63 * cacheline if we try to acquire the lock. So go onto the next slab.
64 * If all pages are busy then we may allocate a new slab instead of reusing
65 * a partial slab. A new slab has noone operating on it and thus there is
66 * no danger of cacheline contention.
68 * Interrupts are disabled during allocation and deallocation in order to
69 * make the slab allocator safe to use in the context of an irq. In addition
70 * interrupts are disabled to ensure that the processor does not change
71 * while handling per_cpu slabs, due to kernel preemption.
73 * SLUB assigns one slab for allocation to each processor.
74 * Allocations only occur from these slabs called cpu slabs.
76 * Slabs with free elements are kept on a partial list and during regular
77 * operations no list for full slabs is used. If an object in a full slab is
78 * freed then the slab will show up again on the partial lists.
79 * We track full slabs for debugging purposes though because otherwise we
80 * cannot scan all objects.
82 * Slabs are freed when they become empty. Teardown and setup is
83 * minimal so we rely on the page allocators per cpu caches for
84 * fast frees and allocs.
86 * Overloading of page flags that are otherwise used for LRU management.
88 * PageActive The slab is frozen and exempt from list processing.
89 * This means that the slab is dedicated to a purpose
90 * such as satisfying allocations for a specific
91 * processor. Objects may be freed in the slab while
92 * it is frozen but slab_free will then skip the usual
93 * list operations. It is up to the processor holding
94 * the slab to integrate the slab into the slab lists
95 * when the slab is no longer needed.
97 * One use of this flag is to mark slabs that are
98 * used for allocations. Then such a slab becomes a cpu
99 * slab. The cpu slab may be equipped with an additional
100 * freelist that allows lockless access to
101 * free objects in addition to the regular freelist
102 * that requires the slab lock.
104 * PageError Slab requires special handling due to debug
105 * options set. This moves slab handling out of
106 * the fast path and disables lockless freelists.
109 #ifdef CONFIG_SLUB_DEBUG
116 * Issues still to be resolved:
118 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
120 * - Variable sizing of the per node arrays
123 /* Enable to test recovery from slab corruption on boot */
124 #undef SLUB_RESILIENCY_TEST
127 * Mininum number of partial slabs. These will be left on the partial
128 * lists even if they are empty. kmem_cache_shrink may reclaim them.
130 #define MIN_PARTIAL 5
133 * Maximum number of desirable partial slabs.
134 * The existence of more partial slabs makes kmem_cache_shrink
135 * sort the partial list by the number of objects in the.
137 #define MAX_PARTIAL 10
139 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
140 SLAB_POISON | SLAB_STORE_USER)
143 * Debugging flags that require metadata to be stored in the slab. These get
144 * disabled when slub_debug=O is used and a cache's min order increases with
147 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
150 * Set of flags that will prevent slab merging
152 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
153 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
156 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
157 SLAB_CACHE_DMA | SLAB_NOTRACK)
160 #define OO_MASK ((1 << OO_SHIFT) - 1)
161 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
163 /* Internal SLUB flags */
164 #define __OBJECT_POISON 0x80000000 /* Poison object */
165 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
167 static int kmem_size
= sizeof(struct kmem_cache
);
170 static struct notifier_block slab_notifier
;
174 DOWN
, /* No slab functionality available */
175 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
176 UP
, /* Everything works but does not show up in sysfs */
180 /* A list of all slab caches on the system */
181 static DECLARE_RWSEM(slub_lock
);
182 static LIST_HEAD(slab_caches
);
185 * Tracking user of a slab.
188 unsigned long addr
; /* Called from address */
189 int cpu
; /* Was running on cpu */
190 int pid
; /* Pid context */
191 unsigned long when
; /* When did the operation occur */
194 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
196 #ifdef CONFIG_SLUB_DEBUG
197 static int sysfs_slab_add(struct kmem_cache
*);
198 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
199 static void sysfs_slab_remove(struct kmem_cache
*);
202 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
203 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
205 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
212 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
214 #ifdef CONFIG_SLUB_STATS
215 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
219 /********************************************************************
220 * Core slab cache functions
221 *******************************************************************/
223 int slab_is_available(void)
225 return slab_state
>= UP
;
228 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
231 return s
->node
[node
];
233 return &s
->local_node
;
237 /* Verify that a pointer has an address that is valid within a slab page */
238 static inline int check_valid_pointer(struct kmem_cache
*s
,
239 struct page
*page
, const void *object
)
246 base
= page_address(page
);
247 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
248 (object
- base
) % s
->size
) {
255 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
257 return *(void **)(object
+ s
->offset
);
260 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
262 *(void **)(object
+ s
->offset
) = fp
;
265 /* Loop over all objects in a slab */
266 #define for_each_object(__p, __s, __addr, __objects) \
267 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
271 #define for_each_free_object(__p, __s, __free) \
272 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
274 /* Determine object index from a given position */
275 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
277 return (p
- addr
) / s
->size
;
280 static inline struct kmem_cache_order_objects
oo_make(int order
,
283 struct kmem_cache_order_objects x
= {
284 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
290 static inline int oo_order(struct kmem_cache_order_objects x
)
292 return x
.x
>> OO_SHIFT
;
295 static inline int oo_objects(struct kmem_cache_order_objects x
)
297 return x
.x
& OO_MASK
;
300 #ifdef CONFIG_SLUB_DEBUG
304 #ifdef CONFIG_SLUB_DEBUG_ON
305 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
307 static int slub_debug
;
310 static char *slub_debug_slabs
;
311 static int disable_higher_order_debug
;
316 static void print_section(char *text
, u8
*addr
, unsigned int length
)
324 for (i
= 0; i
< length
; i
++) {
326 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
329 printk(KERN_CONT
" %02x", addr
[i
]);
331 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
333 printk(KERN_CONT
" %s\n", ascii
);
340 printk(KERN_CONT
" ");
344 printk(KERN_CONT
" %s\n", ascii
);
348 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
349 enum track_item alloc
)
354 p
= object
+ s
->offset
+ sizeof(void *);
356 p
= object
+ s
->inuse
;
361 static void set_track(struct kmem_cache
*s
, void *object
,
362 enum track_item alloc
, unsigned long addr
)
364 struct track
*p
= get_track(s
, object
, alloc
);
368 p
->cpu
= smp_processor_id();
369 p
->pid
= current
->pid
;
372 memset(p
, 0, sizeof(struct track
));
375 static void init_tracking(struct kmem_cache
*s
, void *object
)
377 if (!(s
->flags
& SLAB_STORE_USER
))
380 set_track(s
, object
, TRACK_FREE
, 0UL);
381 set_track(s
, object
, TRACK_ALLOC
, 0UL);
384 static void print_track(const char *s
, struct track
*t
)
389 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
390 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
393 static void print_tracking(struct kmem_cache
*s
, void *object
)
395 if (!(s
->flags
& SLAB_STORE_USER
))
398 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
399 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
402 static void print_page_info(struct page
*page
)
404 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
405 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
409 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
415 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
417 printk(KERN_ERR
"========================================"
418 "=====================================\n");
419 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
420 printk(KERN_ERR
"----------------------------------------"
421 "-------------------------------------\n\n");
424 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
430 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
432 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
435 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
437 unsigned int off
; /* Offset of last byte */
438 u8
*addr
= page_address(page
);
440 print_tracking(s
, p
);
442 print_page_info(page
);
444 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
445 p
, p
- addr
, get_freepointer(s
, p
));
448 print_section("Bytes b4", p
- 16, 16);
450 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
452 if (s
->flags
& SLAB_RED_ZONE
)
453 print_section("Redzone", p
+ s
->objsize
,
454 s
->inuse
- s
->objsize
);
457 off
= s
->offset
+ sizeof(void *);
461 if (s
->flags
& SLAB_STORE_USER
)
462 off
+= 2 * sizeof(struct track
);
465 /* Beginning of the filler is the free pointer */
466 print_section("Padding", p
+ off
, s
->size
- off
);
471 static void object_err(struct kmem_cache
*s
, struct page
*page
,
472 u8
*object
, char *reason
)
474 slab_bug(s
, "%s", reason
);
475 print_trailer(s
, page
, object
);
478 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
484 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
486 slab_bug(s
, "%s", buf
);
487 print_page_info(page
);
491 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
495 if (s
->flags
& __OBJECT_POISON
) {
496 memset(p
, POISON_FREE
, s
->objsize
- 1);
497 p
[s
->objsize
- 1] = POISON_END
;
500 if (s
->flags
& SLAB_RED_ZONE
)
501 memset(p
+ s
->objsize
,
502 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
503 s
->inuse
- s
->objsize
);
506 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
509 if (*start
!= (u8
)value
)
517 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
518 void *from
, void *to
)
520 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
521 memset(from
, data
, to
- from
);
524 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
525 u8
*object
, char *what
,
526 u8
*start
, unsigned int value
, unsigned int bytes
)
531 fault
= check_bytes(start
, value
, bytes
);
536 while (end
> fault
&& end
[-1] == value
)
539 slab_bug(s
, "%s overwritten", what
);
540 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
541 fault
, end
- 1, fault
[0], value
);
542 print_trailer(s
, page
, object
);
544 restore_bytes(s
, what
, value
, fault
, end
);
552 * Bytes of the object to be managed.
553 * If the freepointer may overlay the object then the free
554 * pointer is the first word of the object.
556 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
559 * object + s->objsize
560 * Padding to reach word boundary. This is also used for Redzoning.
561 * Padding is extended by another word if Redzoning is enabled and
564 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
565 * 0xcc (RED_ACTIVE) for objects in use.
568 * Meta data starts here.
570 * A. Free pointer (if we cannot overwrite object on free)
571 * B. Tracking data for SLAB_STORE_USER
572 * C. Padding to reach required alignment boundary or at mininum
573 * one word if debugging is on to be able to detect writes
574 * before the word boundary.
576 * Padding is done using 0x5a (POISON_INUSE)
579 * Nothing is used beyond s->size.
581 * If slabcaches are merged then the objsize and inuse boundaries are mostly
582 * ignored. And therefore no slab options that rely on these boundaries
583 * may be used with merged slabcaches.
586 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
588 unsigned long off
= s
->inuse
; /* The end of info */
591 /* Freepointer is placed after the object. */
592 off
+= sizeof(void *);
594 if (s
->flags
& SLAB_STORE_USER
)
595 /* We also have user information there */
596 off
+= 2 * sizeof(struct track
);
601 return check_bytes_and_report(s
, page
, p
, "Object padding",
602 p
+ off
, POISON_INUSE
, s
->size
- off
);
605 /* Check the pad bytes at the end of a slab page */
606 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
614 if (!(s
->flags
& SLAB_POISON
))
617 start
= page_address(page
);
618 length
= (PAGE_SIZE
<< compound_order(page
));
619 end
= start
+ length
;
620 remainder
= length
% s
->size
;
624 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
627 while (end
> fault
&& end
[-1] == POISON_INUSE
)
630 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
631 print_section("Padding", end
- remainder
, remainder
);
633 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
637 static int check_object(struct kmem_cache
*s
, struct page
*page
,
638 void *object
, int active
)
641 u8
*endobject
= object
+ s
->objsize
;
643 if (s
->flags
& SLAB_RED_ZONE
) {
645 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
647 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
648 endobject
, red
, s
->inuse
- s
->objsize
))
651 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
652 check_bytes_and_report(s
, page
, p
, "Alignment padding",
653 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
657 if (s
->flags
& SLAB_POISON
) {
658 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
659 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
660 POISON_FREE
, s
->objsize
- 1) ||
661 !check_bytes_and_report(s
, page
, p
, "Poison",
662 p
+ s
->objsize
- 1, POISON_END
, 1)))
665 * check_pad_bytes cleans up on its own.
667 check_pad_bytes(s
, page
, p
);
670 if (!s
->offset
&& active
)
672 * Object and freepointer overlap. Cannot check
673 * freepointer while object is allocated.
677 /* Check free pointer validity */
678 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
679 object_err(s
, page
, p
, "Freepointer corrupt");
681 * No choice but to zap it and thus lose the remainder
682 * of the free objects in this slab. May cause
683 * another error because the object count is now wrong.
685 set_freepointer(s
, p
, NULL
);
691 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
695 VM_BUG_ON(!irqs_disabled());
697 if (!PageSlab(page
)) {
698 slab_err(s
, page
, "Not a valid slab page");
702 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
703 if (page
->objects
> maxobj
) {
704 slab_err(s
, page
, "objects %u > max %u",
705 s
->name
, page
->objects
, maxobj
);
708 if (page
->inuse
> page
->objects
) {
709 slab_err(s
, page
, "inuse %u > max %u",
710 s
->name
, page
->inuse
, page
->objects
);
713 /* Slab_pad_check fixes things up after itself */
714 slab_pad_check(s
, page
);
719 * Determine if a certain object on a page is on the freelist. Must hold the
720 * slab lock to guarantee that the chains are in a consistent state.
722 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
725 void *fp
= page
->freelist
;
727 unsigned long max_objects
;
729 while (fp
&& nr
<= page
->objects
) {
732 if (!check_valid_pointer(s
, page
, fp
)) {
734 object_err(s
, page
, object
,
735 "Freechain corrupt");
736 set_freepointer(s
, object
, NULL
);
739 slab_err(s
, page
, "Freepointer corrupt");
740 page
->freelist
= NULL
;
741 page
->inuse
= page
->objects
;
742 slab_fix(s
, "Freelist cleared");
748 fp
= get_freepointer(s
, object
);
752 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
753 if (max_objects
> MAX_OBJS_PER_PAGE
)
754 max_objects
= MAX_OBJS_PER_PAGE
;
756 if (page
->objects
!= max_objects
) {
757 slab_err(s
, page
, "Wrong number of objects. Found %d but "
758 "should be %d", page
->objects
, max_objects
);
759 page
->objects
= max_objects
;
760 slab_fix(s
, "Number of objects adjusted.");
762 if (page
->inuse
!= page
->objects
- nr
) {
763 slab_err(s
, page
, "Wrong object count. Counter is %d but "
764 "counted were %d", page
->inuse
, page
->objects
- nr
);
765 page
->inuse
= page
->objects
- nr
;
766 slab_fix(s
, "Object count adjusted.");
768 return search
== NULL
;
771 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
774 if (s
->flags
& SLAB_TRACE
) {
775 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
777 alloc
? "alloc" : "free",
782 print_section("Object", (void *)object
, s
->objsize
);
789 * Tracking of fully allocated slabs for debugging purposes.
791 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
793 spin_lock(&n
->list_lock
);
794 list_add(&page
->lru
, &n
->full
);
795 spin_unlock(&n
->list_lock
);
798 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
800 struct kmem_cache_node
*n
;
802 if (!(s
->flags
& SLAB_STORE_USER
))
805 n
= get_node(s
, page_to_nid(page
));
807 spin_lock(&n
->list_lock
);
808 list_del(&page
->lru
);
809 spin_unlock(&n
->list_lock
);
812 /* Tracking of the number of slabs for debugging purposes */
813 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
815 struct kmem_cache_node
*n
= get_node(s
, node
);
817 return atomic_long_read(&n
->nr_slabs
);
820 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
822 return atomic_long_read(&n
->nr_slabs
);
825 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
827 struct kmem_cache_node
*n
= get_node(s
, node
);
830 * May be called early in order to allocate a slab for the
831 * kmem_cache_node structure. Solve the chicken-egg
832 * dilemma by deferring the increment of the count during
833 * bootstrap (see early_kmem_cache_node_alloc).
835 if (!NUMA_BUILD
|| n
) {
836 atomic_long_inc(&n
->nr_slabs
);
837 atomic_long_add(objects
, &n
->total_objects
);
840 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
842 struct kmem_cache_node
*n
= get_node(s
, node
);
844 atomic_long_dec(&n
->nr_slabs
);
845 atomic_long_sub(objects
, &n
->total_objects
);
848 /* Object debug checks for alloc/free paths */
849 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
852 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
855 init_object(s
, object
, 0);
856 init_tracking(s
, object
);
859 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
860 void *object
, unsigned long addr
)
862 if (!check_slab(s
, page
))
865 if (!on_freelist(s
, page
, object
)) {
866 object_err(s
, page
, object
, "Object already allocated");
870 if (!check_valid_pointer(s
, page
, object
)) {
871 object_err(s
, page
, object
, "Freelist Pointer check fails");
875 if (!check_object(s
, page
, object
, 0))
878 /* Success perform special debug activities for allocs */
879 if (s
->flags
& SLAB_STORE_USER
)
880 set_track(s
, object
, TRACK_ALLOC
, addr
);
881 trace(s
, page
, object
, 1);
882 init_object(s
, object
, 1);
886 if (PageSlab(page
)) {
888 * If this is a slab page then lets do the best we can
889 * to avoid issues in the future. Marking all objects
890 * as used avoids touching the remaining objects.
892 slab_fix(s
, "Marking all objects used");
893 page
->inuse
= page
->objects
;
894 page
->freelist
= NULL
;
899 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
900 void *object
, unsigned long addr
)
902 if (!check_slab(s
, page
))
905 if (!check_valid_pointer(s
, page
, object
)) {
906 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
910 if (on_freelist(s
, page
, object
)) {
911 object_err(s
, page
, object
, "Object already free");
915 if (!check_object(s
, page
, object
, 1))
918 if (unlikely(s
!= page
->slab
)) {
919 if (!PageSlab(page
)) {
920 slab_err(s
, page
, "Attempt to free object(0x%p) "
921 "outside of slab", object
);
922 } else if (!page
->slab
) {
924 "SLUB <none>: no slab for object 0x%p.\n",
928 object_err(s
, page
, object
,
929 "page slab pointer corrupt.");
933 /* Special debug activities for freeing objects */
934 if (!PageSlubFrozen(page
) && !page
->freelist
)
935 remove_full(s
, page
);
936 if (s
->flags
& SLAB_STORE_USER
)
937 set_track(s
, object
, TRACK_FREE
, addr
);
938 trace(s
, page
, object
, 0);
939 init_object(s
, object
, 0);
943 slab_fix(s
, "Object at 0x%p not freed", object
);
947 static int __init
setup_slub_debug(char *str
)
949 slub_debug
= DEBUG_DEFAULT_FLAGS
;
950 if (*str
++ != '=' || !*str
)
952 * No options specified. Switch on full debugging.
958 * No options but restriction on slabs. This means full
959 * debugging for slabs matching a pattern.
963 if (tolower(*str
) == 'o') {
965 * Avoid enabling debugging on caches if its minimum order
966 * would increase as a result.
968 disable_higher_order_debug
= 1;
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 slub_debug
|= SLAB_FAILSLAB
;
1003 printk(KERN_ERR
"slub_debug option '%c' "
1004 "unknown. skipped\n", *str
);
1010 slub_debug_slabs
= str
+ 1;
1015 __setup("slub_debug", setup_slub_debug
);
1017 static unsigned long kmem_cache_flags(unsigned long objsize
,
1018 unsigned long flags
, const char *name
,
1019 void (*ctor
)(void *))
1022 * Enable debugging if selected on the kernel commandline.
1024 if (slub_debug
&& (!slub_debug_slabs
||
1025 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1026 flags
|= slub_debug
;
1031 static inline void setup_object_debug(struct kmem_cache
*s
,
1032 struct page
*page
, void *object
) {}
1034 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1035 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1037 static inline int free_debug_processing(struct kmem_cache
*s
,
1038 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1040 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1042 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1043 void *object
, int active
) { return 1; }
1044 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1045 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1046 unsigned long flags
, const char *name
,
1047 void (*ctor
)(void *))
1051 #define slub_debug 0
1053 #define disable_higher_order_debug 0
1055 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1057 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1059 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1061 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1066 * Slab allocation and freeing
1068 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1069 struct kmem_cache_order_objects oo
)
1071 int order
= oo_order(oo
);
1073 flags
|= __GFP_NOTRACK
;
1076 return alloc_pages(flags
, order
);
1078 return alloc_pages_exact_node(node
, flags
, order
);
1081 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1084 struct kmem_cache_order_objects oo
= s
->oo
;
1087 flags
|= s
->allocflags
;
1090 * Let the initial higher-order allocation fail under memory pressure
1091 * so we fall-back to the minimum order allocation.
1093 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1095 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1096 if (unlikely(!page
)) {
1099 * Allocation may have failed due to fragmentation.
1100 * Try a lower order alloc if possible
1102 page
= alloc_slab_page(flags
, node
, oo
);
1106 stat(s
, ORDER_FALLBACK
);
1109 if (kmemcheck_enabled
1110 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1111 int pages
= 1 << oo_order(oo
);
1113 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1116 * Objects from caches that have a constructor don't get
1117 * cleared when they're allocated, so we need to do it here.
1120 kmemcheck_mark_uninitialized_pages(page
, pages
);
1122 kmemcheck_mark_unallocated_pages(page
, pages
);
1125 page
->objects
= oo_objects(oo
);
1126 mod_zone_page_state(page_zone(page
),
1127 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1128 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1134 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1137 setup_object_debug(s
, page
, object
);
1138 if (unlikely(s
->ctor
))
1142 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1149 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1151 page
= allocate_slab(s
,
1152 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1156 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1158 page
->flags
|= 1 << PG_slab
;
1159 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1160 SLAB_STORE_USER
| SLAB_TRACE
))
1161 __SetPageSlubDebug(page
);
1163 start
= page_address(page
);
1165 if (unlikely(s
->flags
& SLAB_POISON
))
1166 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1169 for_each_object(p
, s
, start
, page
->objects
) {
1170 setup_object(s
, page
, last
);
1171 set_freepointer(s
, last
, p
);
1174 setup_object(s
, page
, last
);
1175 set_freepointer(s
, last
, NULL
);
1177 page
->freelist
= start
;
1183 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1185 int order
= compound_order(page
);
1186 int pages
= 1 << order
;
1188 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1191 slab_pad_check(s
, page
);
1192 for_each_object(p
, s
, page_address(page
),
1194 check_object(s
, page
, p
, 0);
1195 __ClearPageSlubDebug(page
);
1198 kmemcheck_free_shadow(page
, compound_order(page
));
1200 mod_zone_page_state(page_zone(page
),
1201 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1202 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1205 __ClearPageSlab(page
);
1206 reset_page_mapcount(page
);
1207 if (current
->reclaim_state
)
1208 current
->reclaim_state
->reclaimed_slab
+= pages
;
1209 __free_pages(page
, order
);
1212 static void rcu_free_slab(struct rcu_head
*h
)
1216 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1217 __free_slab(page
->slab
, page
);
1220 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1222 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1224 * RCU free overloads the RCU head over the LRU
1226 struct rcu_head
*head
= (void *)&page
->lru
;
1228 call_rcu(head
, rcu_free_slab
);
1230 __free_slab(s
, page
);
1233 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1235 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1240 * Per slab locking using the pagelock
1242 static __always_inline
void slab_lock(struct page
*page
)
1244 bit_spin_lock(PG_locked
, &page
->flags
);
1247 static __always_inline
void slab_unlock(struct page
*page
)
1249 __bit_spin_unlock(PG_locked
, &page
->flags
);
1252 static __always_inline
int slab_trylock(struct page
*page
)
1256 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1261 * Management of partially allocated slabs
1263 static void add_partial(struct kmem_cache_node
*n
,
1264 struct page
*page
, int tail
)
1266 spin_lock(&n
->list_lock
);
1269 list_add_tail(&page
->lru
, &n
->partial
);
1271 list_add(&page
->lru
, &n
->partial
);
1272 spin_unlock(&n
->list_lock
);
1275 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1277 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1279 spin_lock(&n
->list_lock
);
1280 list_del(&page
->lru
);
1282 spin_unlock(&n
->list_lock
);
1286 * Lock slab and remove from the partial list.
1288 * Must hold list_lock.
1290 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1293 if (slab_trylock(page
)) {
1294 list_del(&page
->lru
);
1296 __SetPageSlubFrozen(page
);
1303 * Try to allocate a partial slab from a specific node.
1305 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1310 * Racy check. If we mistakenly see no partial slabs then we
1311 * just allocate an empty slab. If we mistakenly try to get a
1312 * partial slab and there is none available then get_partials()
1315 if (!n
|| !n
->nr_partial
)
1318 spin_lock(&n
->list_lock
);
1319 list_for_each_entry(page
, &n
->partial
, lru
)
1320 if (lock_and_freeze_slab(n
, page
))
1324 spin_unlock(&n
->list_lock
);
1329 * Get a page from somewhere. Search in increasing NUMA distances.
1331 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1334 struct zonelist
*zonelist
;
1337 enum zone_type high_zoneidx
= gfp_zone(flags
);
1341 * The defrag ratio allows a configuration of the tradeoffs between
1342 * inter node defragmentation and node local allocations. A lower
1343 * defrag_ratio increases the tendency to do local allocations
1344 * instead of attempting to obtain partial slabs from other nodes.
1346 * If the defrag_ratio is set to 0 then kmalloc() always
1347 * returns node local objects. If the ratio is higher then kmalloc()
1348 * may return off node objects because partial slabs are obtained
1349 * from other nodes and filled up.
1351 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1352 * defrag_ratio = 1000) then every (well almost) allocation will
1353 * first attempt to defrag slab caches on other nodes. This means
1354 * scanning over all nodes to look for partial slabs which may be
1355 * expensive if we do it every time we are trying to find a slab
1356 * with available objects.
1358 if (!s
->remote_node_defrag_ratio
||
1359 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1363 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1364 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1365 struct kmem_cache_node
*n
;
1367 n
= get_node(s
, zone_to_nid(zone
));
1369 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1370 n
->nr_partial
> s
->min_partial
) {
1371 page
= get_partial_node(n
);
1384 * Get a partial page, lock it and return it.
1386 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1389 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1391 page
= get_partial_node(get_node(s
, searchnode
));
1392 if (page
|| (flags
& __GFP_THISNODE
))
1395 return get_any_partial(s
, flags
);
1399 * Move a page back to the lists.
1401 * Must be called with the slab lock held.
1403 * On exit the slab lock will have been dropped.
1405 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1407 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1409 __ClearPageSlubFrozen(page
);
1412 if (page
->freelist
) {
1413 add_partial(n
, page
, tail
);
1414 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1416 stat(s
, DEACTIVATE_FULL
);
1417 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1418 (s
->flags
& SLAB_STORE_USER
))
1423 stat(s
, DEACTIVATE_EMPTY
);
1424 if (n
->nr_partial
< s
->min_partial
) {
1426 * Adding an empty slab to the partial slabs in order
1427 * to avoid page allocator overhead. This slab needs
1428 * to come after the other slabs with objects in
1429 * so that the others get filled first. That way the
1430 * size of the partial list stays small.
1432 * kmem_cache_shrink can reclaim any empty slabs from
1435 add_partial(n
, page
, 1);
1440 discard_slab(s
, page
);
1446 * Remove the cpu slab
1448 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1450 struct page
*page
= c
->page
;
1454 stat(s
, DEACTIVATE_REMOTE_FREES
);
1456 * Merge cpu freelist into slab freelist. Typically we get here
1457 * because both freelists are empty. So this is unlikely
1460 while (unlikely(c
->freelist
)) {
1463 tail
= 0; /* Hot objects. Put the slab first */
1465 /* Retrieve object from cpu_freelist */
1466 object
= c
->freelist
;
1467 c
->freelist
= get_freepointer(s
, c
->freelist
);
1469 /* And put onto the regular freelist */
1470 set_freepointer(s
, object
, page
->freelist
);
1471 page
->freelist
= object
;
1475 unfreeze_slab(s
, page
, tail
);
1478 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1480 stat(s
, CPUSLAB_FLUSH
);
1482 deactivate_slab(s
, c
);
1488 * Called from IPI handler with interrupts disabled.
1490 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1492 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1494 if (likely(c
&& c
->page
))
1498 static void flush_cpu_slab(void *d
)
1500 struct kmem_cache
*s
= d
;
1502 __flush_cpu_slab(s
, smp_processor_id());
1505 static void flush_all(struct kmem_cache
*s
)
1507 on_each_cpu(flush_cpu_slab
, s
, 1);
1511 * Check if the objects in a per cpu structure fit numa
1512 * locality expectations.
1514 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1517 if (node
!= -1 && c
->node
!= node
)
1523 static int count_free(struct page
*page
)
1525 return page
->objects
- page
->inuse
;
1528 static unsigned long count_partial(struct kmem_cache_node
*n
,
1529 int (*get_count
)(struct page
*))
1531 unsigned long flags
;
1532 unsigned long x
= 0;
1535 spin_lock_irqsave(&n
->list_lock
, flags
);
1536 list_for_each_entry(page
, &n
->partial
, lru
)
1537 x
+= get_count(page
);
1538 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1542 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1544 #ifdef CONFIG_SLUB_DEBUG
1545 return atomic_long_read(&n
->total_objects
);
1551 static noinline
void
1552 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1557 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1559 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1560 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1561 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1563 if (oo_order(s
->min
) > get_order(s
->objsize
))
1564 printk(KERN_WARNING
" %s debugging increased min order, use "
1565 "slub_debug=O to disable.\n", s
->name
);
1567 for_each_online_node(node
) {
1568 struct kmem_cache_node
*n
= get_node(s
, node
);
1569 unsigned long nr_slabs
;
1570 unsigned long nr_objs
;
1571 unsigned long nr_free
;
1576 nr_free
= count_partial(n
, count_free
);
1577 nr_slabs
= node_nr_slabs(n
);
1578 nr_objs
= node_nr_objs(n
);
1581 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1582 node
, nr_slabs
, nr_objs
, nr_free
);
1587 * Slow path. The lockless freelist is empty or we need to perform
1590 * Interrupts are disabled.
1592 * Processing is still very fast if new objects have been freed to the
1593 * regular freelist. In that case we simply take over the regular freelist
1594 * as the lockless freelist and zap the regular freelist.
1596 * If that is not working then we fall back to the partial lists. We take the
1597 * first element of the freelist as the object to allocate now and move the
1598 * rest of the freelist to the lockless freelist.
1600 * And if we were unable to get a new slab from the partial slab lists then
1601 * we need to allocate a new slab. This is the slowest path since it involves
1602 * a call to the page allocator and the setup of a new slab.
1604 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1605 unsigned long addr
, struct kmem_cache_cpu
*c
)
1610 /* We handle __GFP_ZERO in the caller */
1611 gfpflags
&= ~__GFP_ZERO
;
1617 if (unlikely(!node_match(c
, node
)))
1620 stat(s
, ALLOC_REFILL
);
1623 object
= c
->page
->freelist
;
1624 if (unlikely(!object
))
1626 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1629 c
->freelist
= get_freepointer(s
, object
);
1630 c
->page
->inuse
= c
->page
->objects
;
1631 c
->page
->freelist
= NULL
;
1632 c
->node
= page_to_nid(c
->page
);
1634 slab_unlock(c
->page
);
1635 stat(s
, ALLOC_SLOWPATH
);
1639 deactivate_slab(s
, c
);
1642 new = get_partial(s
, gfpflags
, node
);
1645 stat(s
, ALLOC_FROM_PARTIAL
);
1649 if (gfpflags
& __GFP_WAIT
)
1652 new = new_slab(s
, gfpflags
, node
);
1654 if (gfpflags
& __GFP_WAIT
)
1655 local_irq_disable();
1658 c
= __this_cpu_ptr(s
->cpu_slab
);
1659 stat(s
, ALLOC_SLAB
);
1663 __SetPageSlubFrozen(new);
1667 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1668 slab_out_of_memory(s
, gfpflags
, node
);
1671 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1675 c
->page
->freelist
= get_freepointer(s
, object
);
1681 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1682 * have the fastpath folded into their functions. So no function call
1683 * overhead for requests that can be satisfied on the fastpath.
1685 * The fastpath works by first checking if the lockless freelist can be used.
1686 * If not then __slab_alloc is called for slow processing.
1688 * Otherwise we can simply pick the next object from the lockless free list.
1690 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1691 gfp_t gfpflags
, int node
, unsigned long addr
)
1694 struct kmem_cache_cpu
*c
;
1695 unsigned long flags
;
1697 gfpflags
&= gfp_allowed_mask
;
1699 lockdep_trace_alloc(gfpflags
);
1700 might_sleep_if(gfpflags
& __GFP_WAIT
);
1702 if (should_failslab(s
->objsize
, gfpflags
, s
->flags
))
1705 local_irq_save(flags
);
1706 c
= __this_cpu_ptr(s
->cpu_slab
);
1707 object
= c
->freelist
;
1708 if (unlikely(!object
|| !node_match(c
, node
)))
1710 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1713 c
->freelist
= get_freepointer(s
, object
);
1714 stat(s
, ALLOC_FASTPATH
);
1716 local_irq_restore(flags
);
1718 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1719 memset(object
, 0, s
->objsize
);
1721 kmemcheck_slab_alloc(s
, gfpflags
, object
, s
->objsize
);
1722 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, gfpflags
);
1727 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1729 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1731 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1735 EXPORT_SYMBOL(kmem_cache_alloc
);
1737 #ifdef CONFIG_TRACING
1738 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1740 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1742 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1746 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1748 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1750 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1751 s
->objsize
, s
->size
, gfpflags
, node
);
1755 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1758 #ifdef CONFIG_TRACING
1759 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1763 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1765 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1769 * Slow patch handling. This may still be called frequently since objects
1770 * have a longer lifetime than the cpu slabs in most processing loads.
1772 * So we still attempt to reduce cache line usage. Just take the slab
1773 * lock and free the item. If there is no additional partial page
1774 * handling required then we can return immediately.
1776 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1777 void *x
, unsigned long addr
)
1780 void **object
= (void *)x
;
1782 stat(s
, FREE_SLOWPATH
);
1785 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1789 prior
= page
->freelist
;
1790 set_freepointer(s
, object
, prior
);
1791 page
->freelist
= object
;
1794 if (unlikely(PageSlubFrozen(page
))) {
1795 stat(s
, FREE_FROZEN
);
1799 if (unlikely(!page
->inuse
))
1803 * Objects left in the slab. If it was not on the partial list before
1806 if (unlikely(!prior
)) {
1807 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1808 stat(s
, FREE_ADD_PARTIAL
);
1818 * Slab still on the partial list.
1820 remove_partial(s
, page
);
1821 stat(s
, FREE_REMOVE_PARTIAL
);
1825 discard_slab(s
, page
);
1829 if (!free_debug_processing(s
, page
, x
, addr
))
1835 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1836 * can perform fastpath freeing without additional function calls.
1838 * The fastpath is only possible if we are freeing to the current cpu slab
1839 * of this processor. This typically the case if we have just allocated
1842 * If fastpath is not possible then fall back to __slab_free where we deal
1843 * with all sorts of special processing.
1845 static __always_inline
void slab_free(struct kmem_cache
*s
,
1846 struct page
*page
, void *x
, unsigned long addr
)
1848 void **object
= (void *)x
;
1849 struct kmem_cache_cpu
*c
;
1850 unsigned long flags
;
1852 kmemleak_free_recursive(x
, s
->flags
);
1853 local_irq_save(flags
);
1854 c
= __this_cpu_ptr(s
->cpu_slab
);
1855 kmemcheck_slab_free(s
, object
, s
->objsize
);
1856 debug_check_no_locks_freed(object
, s
->objsize
);
1857 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1858 debug_check_no_obj_freed(object
, s
->objsize
);
1859 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1860 set_freepointer(s
, object
, c
->freelist
);
1861 c
->freelist
= object
;
1862 stat(s
, FREE_FASTPATH
);
1864 __slab_free(s
, page
, x
, addr
);
1866 local_irq_restore(flags
);
1869 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1873 page
= virt_to_head_page(x
);
1875 slab_free(s
, page
, x
, _RET_IP_
);
1877 trace_kmem_cache_free(_RET_IP_
, x
);
1879 EXPORT_SYMBOL(kmem_cache_free
);
1881 /* Figure out on which slab page the object resides */
1882 static struct page
*get_object_page(const void *x
)
1884 struct page
*page
= virt_to_head_page(x
);
1886 if (!PageSlab(page
))
1893 * Object placement in a slab is made very easy because we always start at
1894 * offset 0. If we tune the size of the object to the alignment then we can
1895 * get the required alignment by putting one properly sized object after
1898 * Notice that the allocation order determines the sizes of the per cpu
1899 * caches. Each processor has always one slab available for allocations.
1900 * Increasing the allocation order reduces the number of times that slabs
1901 * must be moved on and off the partial lists and is therefore a factor in
1906 * Mininum / Maximum order of slab pages. This influences locking overhead
1907 * and slab fragmentation. A higher order reduces the number of partial slabs
1908 * and increases the number of allocations possible without having to
1909 * take the list_lock.
1911 static int slub_min_order
;
1912 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1913 static int slub_min_objects
;
1916 * Merge control. If this is set then no merging of slab caches will occur.
1917 * (Could be removed. This was introduced to pacify the merge skeptics.)
1919 static int slub_nomerge
;
1922 * Calculate the order of allocation given an slab object size.
1924 * The order of allocation has significant impact on performance and other
1925 * system components. Generally order 0 allocations should be preferred since
1926 * order 0 does not cause fragmentation in the page allocator. Larger objects
1927 * be problematic to put into order 0 slabs because there may be too much
1928 * unused space left. We go to a higher order if more than 1/16th of the slab
1931 * In order to reach satisfactory performance we must ensure that a minimum
1932 * number of objects is in one slab. Otherwise we may generate too much
1933 * activity on the partial lists which requires taking the list_lock. This is
1934 * less a concern for large slabs though which are rarely used.
1936 * slub_max_order specifies the order where we begin to stop considering the
1937 * number of objects in a slab as critical. If we reach slub_max_order then
1938 * we try to keep the page order as low as possible. So we accept more waste
1939 * of space in favor of a small page order.
1941 * Higher order allocations also allow the placement of more objects in a
1942 * slab and thereby reduce object handling overhead. If the user has
1943 * requested a higher mininum order then we start with that one instead of
1944 * the smallest order which will fit the object.
1946 static inline int slab_order(int size
, int min_objects
,
1947 int max_order
, int fract_leftover
)
1951 int min_order
= slub_min_order
;
1953 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1954 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1956 for (order
= max(min_order
,
1957 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1958 order
<= max_order
; order
++) {
1960 unsigned long slab_size
= PAGE_SIZE
<< order
;
1962 if (slab_size
< min_objects
* size
)
1965 rem
= slab_size
% size
;
1967 if (rem
<= slab_size
/ fract_leftover
)
1975 static inline int calculate_order(int size
)
1983 * Attempt to find best configuration for a slab. This
1984 * works by first attempting to generate a layout with
1985 * the best configuration and backing off gradually.
1987 * First we reduce the acceptable waste in a slab. Then
1988 * we reduce the minimum objects required in a slab.
1990 min_objects
= slub_min_objects
;
1992 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1993 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1994 min_objects
= min(min_objects
, max_objects
);
1996 while (min_objects
> 1) {
1998 while (fraction
>= 4) {
1999 order
= slab_order(size
, min_objects
,
2000 slub_max_order
, fraction
);
2001 if (order
<= slub_max_order
)
2009 * We were unable to place multiple objects in a slab. Now
2010 * lets see if we can place a single object there.
2012 order
= slab_order(size
, 1, slub_max_order
, 1);
2013 if (order
<= slub_max_order
)
2017 * Doh this slab cannot be placed using slub_max_order.
2019 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2020 if (order
< MAX_ORDER
)
2026 * Figure out what the alignment of the objects will be.
2028 static unsigned long calculate_alignment(unsigned long flags
,
2029 unsigned long align
, unsigned long size
)
2032 * If the user wants hardware cache aligned objects then follow that
2033 * suggestion if the object is sufficiently large.
2035 * The hardware cache alignment cannot override the specified
2036 * alignment though. If that is greater then use it.
2038 if (flags
& SLAB_HWCACHE_ALIGN
) {
2039 unsigned long ralign
= cache_line_size();
2040 while (size
<= ralign
/ 2)
2042 align
= max(align
, ralign
);
2045 if (align
< ARCH_SLAB_MINALIGN
)
2046 align
= ARCH_SLAB_MINALIGN
;
2048 return ALIGN(align
, sizeof(void *));
2052 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2055 spin_lock_init(&n
->list_lock
);
2056 INIT_LIST_HEAD(&n
->partial
);
2057 #ifdef CONFIG_SLUB_DEBUG
2058 atomic_long_set(&n
->nr_slabs
, 0);
2059 atomic_long_set(&n
->total_objects
, 0);
2060 INIT_LIST_HEAD(&n
->full
);
2064 static DEFINE_PER_CPU(struct kmem_cache_cpu
, kmalloc_percpu
[KMALLOC_CACHES
]);
2066 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2068 if (s
< kmalloc_caches
+ KMALLOC_CACHES
&& s
>= kmalloc_caches
)
2070 * Boot time creation of the kmalloc array. Use static per cpu data
2071 * since the per cpu allocator is not available yet.
2073 s
->cpu_slab
= kmalloc_percpu
+ (s
- kmalloc_caches
);
2075 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2085 * No kmalloc_node yet so do it by hand. We know that this is the first
2086 * slab on the node for this slabcache. There are no concurrent accesses
2089 * Note that this function only works on the kmalloc_node_cache
2090 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2091 * memory on a fresh node that has no slab structures yet.
2093 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2096 struct kmem_cache_node
*n
;
2097 unsigned long flags
;
2099 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2101 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2104 if (page_to_nid(page
) != node
) {
2105 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2107 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2108 "in order to be able to continue\n");
2113 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2115 kmalloc_caches
->node
[node
] = n
;
2116 #ifdef CONFIG_SLUB_DEBUG
2117 init_object(kmalloc_caches
, n
, 1);
2118 init_tracking(kmalloc_caches
, n
);
2120 init_kmem_cache_node(n
, kmalloc_caches
);
2121 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2124 * lockdep requires consistent irq usage for each lock
2125 * so even though there cannot be a race this early in
2126 * the boot sequence, we still disable irqs.
2128 local_irq_save(flags
);
2129 add_partial(n
, page
, 0);
2130 local_irq_restore(flags
);
2133 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2137 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2138 struct kmem_cache_node
*n
= s
->node
[node
];
2139 if (n
&& n
!= &s
->local_node
)
2140 kmem_cache_free(kmalloc_caches
, n
);
2141 s
->node
[node
] = NULL
;
2145 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2150 if (slab_state
>= UP
&& (s
< kmalloc_caches
||
2151 s
>= kmalloc_caches
+ KMALLOC_CACHES
))
2152 local_node
= page_to_nid(virt_to_page(s
));
2156 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2157 struct kmem_cache_node
*n
;
2159 if (local_node
== node
)
2162 if (slab_state
== DOWN
) {
2163 early_kmem_cache_node_alloc(gfpflags
, node
);
2166 n
= kmem_cache_alloc_node(kmalloc_caches
,
2170 free_kmem_cache_nodes(s
);
2176 init_kmem_cache_node(n
, s
);
2181 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2185 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2187 init_kmem_cache_node(&s
->local_node
, s
);
2192 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2194 if (min
< MIN_PARTIAL
)
2196 else if (min
> MAX_PARTIAL
)
2198 s
->min_partial
= min
;
2202 * calculate_sizes() determines the order and the distribution of data within
2205 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2207 unsigned long flags
= s
->flags
;
2208 unsigned long size
= s
->objsize
;
2209 unsigned long align
= s
->align
;
2213 * Round up object size to the next word boundary. We can only
2214 * place the free pointer at word boundaries and this determines
2215 * the possible location of the free pointer.
2217 size
= ALIGN(size
, sizeof(void *));
2219 #ifdef CONFIG_SLUB_DEBUG
2221 * Determine if we can poison the object itself. If the user of
2222 * the slab may touch the object after free or before allocation
2223 * then we should never poison the object itself.
2225 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2227 s
->flags
|= __OBJECT_POISON
;
2229 s
->flags
&= ~__OBJECT_POISON
;
2233 * If we are Redzoning then check if there is some space between the
2234 * end of the object and the free pointer. If not then add an
2235 * additional word to have some bytes to store Redzone information.
2237 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2238 size
+= sizeof(void *);
2242 * With that we have determined the number of bytes in actual use
2243 * by the object. This is the potential offset to the free pointer.
2247 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2250 * Relocate free pointer after the object if it is not
2251 * permitted to overwrite the first word of the object on
2254 * This is the case if we do RCU, have a constructor or
2255 * destructor or are poisoning the objects.
2258 size
+= sizeof(void *);
2261 #ifdef CONFIG_SLUB_DEBUG
2262 if (flags
& SLAB_STORE_USER
)
2264 * Need to store information about allocs and frees after
2267 size
+= 2 * sizeof(struct track
);
2269 if (flags
& SLAB_RED_ZONE
)
2271 * Add some empty padding so that we can catch
2272 * overwrites from earlier objects rather than let
2273 * tracking information or the free pointer be
2274 * corrupted if a user writes before the start
2277 size
+= sizeof(void *);
2281 * Determine the alignment based on various parameters that the
2282 * user specified and the dynamic determination of cache line size
2285 align
= calculate_alignment(flags
, align
, s
->objsize
);
2289 * SLUB stores one object immediately after another beginning from
2290 * offset 0. In order to align the objects we have to simply size
2291 * each object to conform to the alignment.
2293 size
= ALIGN(size
, align
);
2295 if (forced_order
>= 0)
2296 order
= forced_order
;
2298 order
= calculate_order(size
);
2305 s
->allocflags
|= __GFP_COMP
;
2307 if (s
->flags
& SLAB_CACHE_DMA
)
2308 s
->allocflags
|= SLUB_DMA
;
2310 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2311 s
->allocflags
|= __GFP_RECLAIMABLE
;
2314 * Determine the number of objects per slab
2316 s
->oo
= oo_make(order
, size
);
2317 s
->min
= oo_make(get_order(size
), size
);
2318 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2321 return !!oo_objects(s
->oo
);
2325 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2326 const char *name
, size_t size
,
2327 size_t align
, unsigned long flags
,
2328 void (*ctor
)(void *))
2330 memset(s
, 0, kmem_size
);
2335 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2337 if (!calculate_sizes(s
, -1))
2339 if (disable_higher_order_debug
) {
2341 * Disable debugging flags that store metadata if the min slab
2344 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2345 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2347 if (!calculate_sizes(s
, -1))
2353 * The larger the object size is, the more pages we want on the partial
2354 * list to avoid pounding the page allocator excessively.
2356 set_min_partial(s
, ilog2(s
->size
));
2359 s
->remote_node_defrag_ratio
= 1000;
2361 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2364 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2367 free_kmem_cache_nodes(s
);
2369 if (flags
& SLAB_PANIC
)
2370 panic("Cannot create slab %s size=%lu realsize=%u "
2371 "order=%u offset=%u flags=%lx\n",
2372 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2378 * Check if a given pointer is valid
2380 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2384 if (!kern_ptr_validate(object
, s
->size
))
2387 page
= get_object_page(object
);
2389 if (!page
|| s
!= page
->slab
)
2390 /* No slab or wrong slab */
2393 if (!check_valid_pointer(s
, page
, object
))
2397 * We could also check if the object is on the slabs freelist.
2398 * But this would be too expensive and it seems that the main
2399 * purpose of kmem_ptr_valid() is to check if the object belongs
2400 * to a certain slab.
2404 EXPORT_SYMBOL(kmem_ptr_validate
);
2407 * Determine the size of a slab object
2409 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2413 EXPORT_SYMBOL(kmem_cache_size
);
2415 const char *kmem_cache_name(struct kmem_cache
*s
)
2419 EXPORT_SYMBOL(kmem_cache_name
);
2421 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2424 #ifdef CONFIG_SLUB_DEBUG
2425 void *addr
= page_address(page
);
2427 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2432 slab_err(s
, page
, "%s", text
);
2434 for_each_free_object(p
, s
, page
->freelist
)
2435 set_bit(slab_index(p
, s
, addr
), map
);
2437 for_each_object(p
, s
, addr
, page
->objects
) {
2439 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2440 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2442 print_tracking(s
, p
);
2451 * Attempt to free all partial slabs on a node.
2453 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2455 unsigned long flags
;
2456 struct page
*page
, *h
;
2458 spin_lock_irqsave(&n
->list_lock
, flags
);
2459 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2461 list_del(&page
->lru
);
2462 discard_slab(s
, page
);
2465 list_slab_objects(s
, page
,
2466 "Objects remaining on kmem_cache_close()");
2469 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2473 * Release all resources used by a slab cache.
2475 static inline int kmem_cache_close(struct kmem_cache
*s
)
2480 free_percpu(s
->cpu_slab
);
2481 /* Attempt to free all objects */
2482 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2483 struct kmem_cache_node
*n
= get_node(s
, node
);
2486 if (n
->nr_partial
|| slabs_node(s
, node
))
2489 free_kmem_cache_nodes(s
);
2494 * Close a cache and release the kmem_cache structure
2495 * (must be used for caches created using kmem_cache_create)
2497 void kmem_cache_destroy(struct kmem_cache
*s
)
2499 down_write(&slub_lock
);
2503 up_write(&slub_lock
);
2504 if (kmem_cache_close(s
)) {
2505 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2506 "still has objects.\n", s
->name
, __func__
);
2509 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2511 sysfs_slab_remove(s
);
2513 up_write(&slub_lock
);
2515 EXPORT_SYMBOL(kmem_cache_destroy
);
2517 /********************************************************************
2519 *******************************************************************/
2521 struct kmem_cache kmalloc_caches
[KMALLOC_CACHES
] __cacheline_aligned
;
2522 EXPORT_SYMBOL(kmalloc_caches
);
2524 static int __init
setup_slub_min_order(char *str
)
2526 get_option(&str
, &slub_min_order
);
2531 __setup("slub_min_order=", setup_slub_min_order
);
2533 static int __init
setup_slub_max_order(char *str
)
2535 get_option(&str
, &slub_max_order
);
2536 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2541 __setup("slub_max_order=", setup_slub_max_order
);
2543 static int __init
setup_slub_min_objects(char *str
)
2545 get_option(&str
, &slub_min_objects
);
2550 __setup("slub_min_objects=", setup_slub_min_objects
);
2552 static int __init
setup_slub_nomerge(char *str
)
2558 __setup("slub_nomerge", setup_slub_nomerge
);
2560 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2561 const char *name
, int size
, gfp_t gfp_flags
)
2563 unsigned int flags
= 0;
2565 if (gfp_flags
& SLUB_DMA
)
2566 flags
= SLAB_CACHE_DMA
;
2569 * This function is called with IRQs disabled during early-boot on
2570 * single CPU so there's no need to take slub_lock here.
2572 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2576 list_add(&s
->list
, &slab_caches
);
2578 if (sysfs_slab_add(s
))
2583 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2586 #ifdef CONFIG_ZONE_DMA
2587 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2589 static void sysfs_add_func(struct work_struct
*w
)
2591 struct kmem_cache
*s
;
2593 down_write(&slub_lock
);
2594 list_for_each_entry(s
, &slab_caches
, list
) {
2595 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2596 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2600 up_write(&slub_lock
);
2603 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2605 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2607 struct kmem_cache
*s
;
2610 unsigned long slabflags
;
2613 s
= kmalloc_caches_dma
[index
];
2617 /* Dynamically create dma cache */
2618 if (flags
& __GFP_WAIT
)
2619 down_write(&slub_lock
);
2621 if (!down_write_trylock(&slub_lock
))
2625 if (kmalloc_caches_dma
[index
])
2628 realsize
= kmalloc_caches
[index
].objsize
;
2629 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2630 (unsigned int)realsize
);
2633 for (i
= 0; i
< KMALLOC_CACHES
; i
++)
2634 if (!kmalloc_caches
[i
].size
)
2637 BUG_ON(i
>= KMALLOC_CACHES
);
2638 s
= kmalloc_caches
+ i
;
2641 * Must defer sysfs creation to a workqueue because we don't know
2642 * what context we are called from. Before sysfs comes up, we don't
2643 * need to do anything because our sysfs initcall will start by
2644 * adding all existing slabs to sysfs.
2646 slabflags
= SLAB_CACHE_DMA
|SLAB_NOTRACK
;
2647 if (slab_state
>= SYSFS
)
2648 slabflags
|= __SYSFS_ADD_DEFERRED
;
2650 if (!text
|| !kmem_cache_open(s
, flags
, text
,
2651 realsize
, ARCH_KMALLOC_MINALIGN
, slabflags
, NULL
)) {
2657 list_add(&s
->list
, &slab_caches
);
2658 kmalloc_caches_dma
[index
] = s
;
2660 if (slab_state
>= SYSFS
)
2661 schedule_work(&sysfs_add_work
);
2664 up_write(&slub_lock
);
2666 return kmalloc_caches_dma
[index
];
2671 * Conversion table for small slabs sizes / 8 to the index in the
2672 * kmalloc array. This is necessary for slabs < 192 since we have non power
2673 * of two cache sizes there. The size of larger slabs can be determined using
2676 static s8 size_index
[24] = {
2703 static inline int size_index_elem(size_t bytes
)
2705 return (bytes
- 1) / 8;
2708 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2714 return ZERO_SIZE_PTR
;
2716 index
= size_index
[size_index_elem(size
)];
2718 index
= fls(size
- 1);
2720 #ifdef CONFIG_ZONE_DMA
2721 if (unlikely((flags
& SLUB_DMA
)))
2722 return dma_kmalloc_cache(index
, flags
);
2725 return &kmalloc_caches
[index
];
2728 void *__kmalloc(size_t size
, gfp_t flags
)
2730 struct kmem_cache
*s
;
2733 if (unlikely(size
> SLUB_MAX_SIZE
))
2734 return kmalloc_large(size
, flags
);
2736 s
= get_slab(size
, flags
);
2738 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2741 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2743 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2747 EXPORT_SYMBOL(__kmalloc
);
2749 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2754 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2755 page
= alloc_pages_node(node
, flags
, get_order(size
));
2757 ptr
= page_address(page
);
2759 kmemleak_alloc(ptr
, size
, 1, flags
);
2764 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2766 struct kmem_cache
*s
;
2769 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2770 ret
= kmalloc_large_node(size
, flags
, node
);
2772 trace_kmalloc_node(_RET_IP_
, ret
,
2773 size
, PAGE_SIZE
<< get_order(size
),
2779 s
= get_slab(size
, flags
);
2781 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2784 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2786 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2790 EXPORT_SYMBOL(__kmalloc_node
);
2793 size_t ksize(const void *object
)
2796 struct kmem_cache
*s
;
2798 if (unlikely(object
== ZERO_SIZE_PTR
))
2801 page
= virt_to_head_page(object
);
2803 if (unlikely(!PageSlab(page
))) {
2804 WARN_ON(!PageCompound(page
));
2805 return PAGE_SIZE
<< compound_order(page
);
2809 #ifdef CONFIG_SLUB_DEBUG
2811 * Debugging requires use of the padding between object
2812 * and whatever may come after it.
2814 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2819 * If we have the need to store the freelist pointer
2820 * back there or track user information then we can
2821 * only use the space before that information.
2823 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2826 * Else we can use all the padding etc for the allocation
2830 EXPORT_SYMBOL(ksize
);
2832 void kfree(const void *x
)
2835 void *object
= (void *)x
;
2837 trace_kfree(_RET_IP_
, x
);
2839 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2842 page
= virt_to_head_page(x
);
2843 if (unlikely(!PageSlab(page
))) {
2844 BUG_ON(!PageCompound(page
));
2849 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2851 EXPORT_SYMBOL(kfree
);
2854 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2855 * the remaining slabs by the number of items in use. The slabs with the
2856 * most items in use come first. New allocations will then fill those up
2857 * and thus they can be removed from the partial lists.
2859 * The slabs with the least items are placed last. This results in them
2860 * being allocated from last increasing the chance that the last objects
2861 * are freed in them.
2863 int kmem_cache_shrink(struct kmem_cache
*s
)
2867 struct kmem_cache_node
*n
;
2870 int objects
= oo_objects(s
->max
);
2871 struct list_head
*slabs_by_inuse
=
2872 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2873 unsigned long flags
;
2875 if (!slabs_by_inuse
)
2879 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2880 n
= get_node(s
, node
);
2885 for (i
= 0; i
< objects
; i
++)
2886 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2888 spin_lock_irqsave(&n
->list_lock
, flags
);
2891 * Build lists indexed by the items in use in each slab.
2893 * Note that concurrent frees may occur while we hold the
2894 * list_lock. page->inuse here is the upper limit.
2896 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2897 if (!page
->inuse
&& slab_trylock(page
)) {
2899 * Must hold slab lock here because slab_free
2900 * may have freed the last object and be
2901 * waiting to release the slab.
2903 list_del(&page
->lru
);
2906 discard_slab(s
, page
);
2908 list_move(&page
->lru
,
2909 slabs_by_inuse
+ page
->inuse
);
2914 * Rebuild the partial list with the slabs filled up most
2915 * first and the least used slabs at the end.
2917 for (i
= objects
- 1; i
>= 0; i
--)
2918 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2920 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2923 kfree(slabs_by_inuse
);
2926 EXPORT_SYMBOL(kmem_cache_shrink
);
2928 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2929 static int slab_mem_going_offline_callback(void *arg
)
2931 struct kmem_cache
*s
;
2933 down_read(&slub_lock
);
2934 list_for_each_entry(s
, &slab_caches
, list
)
2935 kmem_cache_shrink(s
);
2936 up_read(&slub_lock
);
2941 static void slab_mem_offline_callback(void *arg
)
2943 struct kmem_cache_node
*n
;
2944 struct kmem_cache
*s
;
2945 struct memory_notify
*marg
= arg
;
2948 offline_node
= marg
->status_change_nid
;
2951 * If the node still has available memory. we need kmem_cache_node
2954 if (offline_node
< 0)
2957 down_read(&slub_lock
);
2958 list_for_each_entry(s
, &slab_caches
, list
) {
2959 n
= get_node(s
, offline_node
);
2962 * if n->nr_slabs > 0, slabs still exist on the node
2963 * that is going down. We were unable to free them,
2964 * and offline_pages() function shouldn't call this
2965 * callback. So, we must fail.
2967 BUG_ON(slabs_node(s
, offline_node
));
2969 s
->node
[offline_node
] = NULL
;
2970 kmem_cache_free(kmalloc_caches
, n
);
2973 up_read(&slub_lock
);
2976 static int slab_mem_going_online_callback(void *arg
)
2978 struct kmem_cache_node
*n
;
2979 struct kmem_cache
*s
;
2980 struct memory_notify
*marg
= arg
;
2981 int nid
= marg
->status_change_nid
;
2985 * If the node's memory is already available, then kmem_cache_node is
2986 * already created. Nothing to do.
2992 * We are bringing a node online. No memory is available yet. We must
2993 * allocate a kmem_cache_node structure in order to bring the node
2996 down_read(&slub_lock
);
2997 list_for_each_entry(s
, &slab_caches
, list
) {
2999 * XXX: kmem_cache_alloc_node will fallback to other nodes
3000 * since memory is not yet available from the node that
3003 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
3008 init_kmem_cache_node(n
, s
);
3012 up_read(&slub_lock
);
3016 static int slab_memory_callback(struct notifier_block
*self
,
3017 unsigned long action
, void *arg
)
3022 case MEM_GOING_ONLINE
:
3023 ret
= slab_mem_going_online_callback(arg
);
3025 case MEM_GOING_OFFLINE
:
3026 ret
= slab_mem_going_offline_callback(arg
);
3029 case MEM_CANCEL_ONLINE
:
3030 slab_mem_offline_callback(arg
);
3033 case MEM_CANCEL_OFFLINE
:
3037 ret
= notifier_from_errno(ret
);
3043 #endif /* CONFIG_MEMORY_HOTPLUG */
3045 /********************************************************************
3046 * Basic setup of slabs
3047 *******************************************************************/
3049 void __init
kmem_cache_init(void)
3056 * Must first have the slab cache available for the allocations of the
3057 * struct kmem_cache_node's. There is special bootstrap code in
3058 * kmem_cache_open for slab_state == DOWN.
3060 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3061 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3062 kmalloc_caches
[0].refcount
= -1;
3065 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3068 /* Able to allocate the per node structures */
3069 slab_state
= PARTIAL
;
3071 /* Caches that are not of the two-to-the-power-of size */
3072 if (KMALLOC_MIN_SIZE
<= 32) {
3073 create_kmalloc_cache(&kmalloc_caches
[1],
3074 "kmalloc-96", 96, GFP_NOWAIT
);
3077 if (KMALLOC_MIN_SIZE
<= 64) {
3078 create_kmalloc_cache(&kmalloc_caches
[2],
3079 "kmalloc-192", 192, GFP_NOWAIT
);
3083 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3084 create_kmalloc_cache(&kmalloc_caches
[i
],
3085 "kmalloc", 1 << i
, GFP_NOWAIT
);
3091 * Patch up the size_index table if we have strange large alignment
3092 * requirements for the kmalloc array. This is only the case for
3093 * MIPS it seems. The standard arches will not generate any code here.
3095 * Largest permitted alignment is 256 bytes due to the way we
3096 * handle the index determination for the smaller caches.
3098 * Make sure that nothing crazy happens if someone starts tinkering
3099 * around with ARCH_KMALLOC_MINALIGN
3101 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3102 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3104 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3105 int elem
= size_index_elem(i
);
3106 if (elem
>= ARRAY_SIZE(size_index
))
3108 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3111 if (KMALLOC_MIN_SIZE
== 64) {
3113 * The 96 byte size cache is not used if the alignment
3116 for (i
= 64 + 8; i
<= 96; i
+= 8)
3117 size_index
[size_index_elem(i
)] = 7;
3118 } else if (KMALLOC_MIN_SIZE
== 128) {
3120 * The 192 byte sized cache is not used if the alignment
3121 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3124 for (i
= 128 + 8; i
<= 192; i
+= 8)
3125 size_index
[size_index_elem(i
)] = 8;
3130 /* Provide the correct kmalloc names now that the caches are up */
3131 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3132 kmalloc_caches
[i
]. name
=
3133 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3136 register_cpu_notifier(&slab_notifier
);
3139 kmem_size
= offsetof(struct kmem_cache
, node
) +
3140 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3142 kmem_size
= sizeof(struct kmem_cache
);
3146 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3147 " CPUs=%d, Nodes=%d\n",
3148 caches
, cache_line_size(),
3149 slub_min_order
, slub_max_order
, slub_min_objects
,
3150 nr_cpu_ids
, nr_node_ids
);
3153 void __init
kmem_cache_init_late(void)
3158 * Find a mergeable slab cache
3160 static int slab_unmergeable(struct kmem_cache
*s
)
3162 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3169 * We may have set a slab to be unmergeable during bootstrap.
3171 if (s
->refcount
< 0)
3177 static struct kmem_cache
*find_mergeable(size_t size
,
3178 size_t align
, unsigned long flags
, const char *name
,
3179 void (*ctor
)(void *))
3181 struct kmem_cache
*s
;
3183 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3189 size
= ALIGN(size
, sizeof(void *));
3190 align
= calculate_alignment(flags
, align
, size
);
3191 size
= ALIGN(size
, align
);
3192 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3194 list_for_each_entry(s
, &slab_caches
, list
) {
3195 if (slab_unmergeable(s
))
3201 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3204 * Check if alignment is compatible.
3205 * Courtesy of Adrian Drzewiecki
3207 if ((s
->size
& ~(align
- 1)) != s
->size
)
3210 if (s
->size
- size
>= sizeof(void *))
3218 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3219 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3221 struct kmem_cache
*s
;
3226 down_write(&slub_lock
);
3227 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3231 * Adjust the object sizes so that we clear
3232 * the complete object on kzalloc.
3234 s
->objsize
= max(s
->objsize
, (int)size
);
3235 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3236 up_write(&slub_lock
);
3238 if (sysfs_slab_alias(s
, name
)) {
3239 down_write(&slub_lock
);
3241 up_write(&slub_lock
);
3247 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3249 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3250 size
, align
, flags
, ctor
)) {
3251 list_add(&s
->list
, &slab_caches
);
3252 up_write(&slub_lock
);
3253 if (sysfs_slab_add(s
)) {
3254 down_write(&slub_lock
);
3256 up_write(&slub_lock
);
3264 up_write(&slub_lock
);
3267 if (flags
& SLAB_PANIC
)
3268 panic("Cannot create slabcache %s\n", name
);
3273 EXPORT_SYMBOL(kmem_cache_create
);
3277 * Use the cpu notifier to insure that the cpu slabs are flushed when
3280 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3281 unsigned long action
, void *hcpu
)
3283 long cpu
= (long)hcpu
;
3284 struct kmem_cache
*s
;
3285 unsigned long flags
;
3288 case CPU_UP_CANCELED
:
3289 case CPU_UP_CANCELED_FROZEN
:
3291 case CPU_DEAD_FROZEN
:
3292 down_read(&slub_lock
);
3293 list_for_each_entry(s
, &slab_caches
, list
) {
3294 local_irq_save(flags
);
3295 __flush_cpu_slab(s
, cpu
);
3296 local_irq_restore(flags
);
3298 up_read(&slub_lock
);
3306 static struct notifier_block __cpuinitdata slab_notifier
= {
3307 .notifier_call
= slab_cpuup_callback
3312 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3314 struct kmem_cache
*s
;
3317 if (unlikely(size
> SLUB_MAX_SIZE
))
3318 return kmalloc_large(size
, gfpflags
);
3320 s
= get_slab(size
, gfpflags
);
3322 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3325 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3327 /* Honor the call site pointer we recieved. */
3328 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3333 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3334 int node
, unsigned long caller
)
3336 struct kmem_cache
*s
;
3339 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3340 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3342 trace_kmalloc_node(caller
, ret
,
3343 size
, PAGE_SIZE
<< get_order(size
),
3349 s
= get_slab(size
, gfpflags
);
3351 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3354 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3356 /* Honor the call site pointer we recieved. */
3357 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3362 #ifdef CONFIG_SLUB_DEBUG
3363 static int count_inuse(struct page
*page
)
3368 static int count_total(struct page
*page
)
3370 return page
->objects
;
3373 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3377 void *addr
= page_address(page
);
3379 if (!check_slab(s
, page
) ||
3380 !on_freelist(s
, page
, NULL
))
3383 /* Now we know that a valid freelist exists */
3384 bitmap_zero(map
, page
->objects
);
3386 for_each_free_object(p
, s
, page
->freelist
) {
3387 set_bit(slab_index(p
, s
, addr
), map
);
3388 if (!check_object(s
, page
, p
, 0))
3392 for_each_object(p
, s
, addr
, page
->objects
)
3393 if (!test_bit(slab_index(p
, s
, addr
), map
))
3394 if (!check_object(s
, page
, p
, 1))
3399 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3402 if (slab_trylock(page
)) {
3403 validate_slab(s
, page
, map
);
3406 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3409 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3410 if (!PageSlubDebug(page
))
3411 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3412 "on slab 0x%p\n", s
->name
, page
);
3414 if (PageSlubDebug(page
))
3415 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3416 "slab 0x%p\n", s
->name
, page
);
3420 static int validate_slab_node(struct kmem_cache
*s
,
3421 struct kmem_cache_node
*n
, unsigned long *map
)
3423 unsigned long count
= 0;
3425 unsigned long flags
;
3427 spin_lock_irqsave(&n
->list_lock
, flags
);
3429 list_for_each_entry(page
, &n
->partial
, lru
) {
3430 validate_slab_slab(s
, page
, map
);
3433 if (count
!= n
->nr_partial
)
3434 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3435 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3437 if (!(s
->flags
& SLAB_STORE_USER
))
3440 list_for_each_entry(page
, &n
->full
, lru
) {
3441 validate_slab_slab(s
, page
, map
);
3444 if (count
!= atomic_long_read(&n
->nr_slabs
))
3445 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3446 "counter=%ld\n", s
->name
, count
,
3447 atomic_long_read(&n
->nr_slabs
));
3450 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3454 static long validate_slab_cache(struct kmem_cache
*s
)
3457 unsigned long count
= 0;
3458 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3459 sizeof(unsigned long), GFP_KERNEL
);
3465 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3466 struct kmem_cache_node
*n
= get_node(s
, node
);
3468 count
+= validate_slab_node(s
, n
, map
);
3474 #ifdef SLUB_RESILIENCY_TEST
3475 static void resiliency_test(void)
3479 printk(KERN_ERR
"SLUB resiliency testing\n");
3480 printk(KERN_ERR
"-----------------------\n");
3481 printk(KERN_ERR
"A. Corruption after allocation\n");
3483 p
= kzalloc(16, GFP_KERNEL
);
3485 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3486 " 0x12->0x%p\n\n", p
+ 16);
3488 validate_slab_cache(kmalloc_caches
+ 4);
3490 /* Hmmm... The next two are dangerous */
3491 p
= kzalloc(32, GFP_KERNEL
);
3492 p
[32 + sizeof(void *)] = 0x34;
3493 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3494 " 0x34 -> -0x%p\n", p
);
3496 "If allocated object is overwritten then not detectable\n\n");
3498 validate_slab_cache(kmalloc_caches
+ 5);
3499 p
= kzalloc(64, GFP_KERNEL
);
3500 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3502 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3505 "If allocated object is overwritten then not detectable\n\n");
3506 validate_slab_cache(kmalloc_caches
+ 6);
3508 printk(KERN_ERR
"\nB. Corruption after free\n");
3509 p
= kzalloc(128, GFP_KERNEL
);
3512 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3513 validate_slab_cache(kmalloc_caches
+ 7);
3515 p
= kzalloc(256, GFP_KERNEL
);
3518 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3520 validate_slab_cache(kmalloc_caches
+ 8);
3522 p
= kzalloc(512, GFP_KERNEL
);
3525 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3526 validate_slab_cache(kmalloc_caches
+ 9);
3529 static void resiliency_test(void) {};
3533 * Generate lists of code addresses where slabcache objects are allocated
3538 unsigned long count
;
3545 DECLARE_BITMAP(cpus
, NR_CPUS
);
3551 unsigned long count
;
3552 struct location
*loc
;
3555 static void free_loc_track(struct loc_track
*t
)
3558 free_pages((unsigned long)t
->loc
,
3559 get_order(sizeof(struct location
) * t
->max
));
3562 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3567 order
= get_order(sizeof(struct location
) * max
);
3569 l
= (void *)__get_free_pages(flags
, order
);
3574 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3582 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3583 const struct track
*track
)
3585 long start
, end
, pos
;
3587 unsigned long caddr
;
3588 unsigned long age
= jiffies
- track
->when
;
3594 pos
= start
+ (end
- start
+ 1) / 2;
3597 * There is nothing at "end". If we end up there
3598 * we need to add something to before end.
3603 caddr
= t
->loc
[pos
].addr
;
3604 if (track
->addr
== caddr
) {
3610 if (age
< l
->min_time
)
3612 if (age
> l
->max_time
)
3615 if (track
->pid
< l
->min_pid
)
3616 l
->min_pid
= track
->pid
;
3617 if (track
->pid
> l
->max_pid
)
3618 l
->max_pid
= track
->pid
;
3620 cpumask_set_cpu(track
->cpu
,
3621 to_cpumask(l
->cpus
));
3623 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3627 if (track
->addr
< caddr
)
3634 * Not found. Insert new tracking element.
3636 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3642 (t
->count
- pos
) * sizeof(struct location
));
3645 l
->addr
= track
->addr
;
3649 l
->min_pid
= track
->pid
;
3650 l
->max_pid
= track
->pid
;
3651 cpumask_clear(to_cpumask(l
->cpus
));
3652 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3653 nodes_clear(l
->nodes
);
3654 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3658 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3659 struct page
*page
, enum track_item alloc
,
3662 void *addr
= page_address(page
);
3665 bitmap_zero(map
, page
->objects
);
3666 for_each_free_object(p
, s
, page
->freelist
)
3667 set_bit(slab_index(p
, s
, addr
), map
);
3669 for_each_object(p
, s
, addr
, page
->objects
)
3670 if (!test_bit(slab_index(p
, s
, addr
), map
))
3671 add_location(t
, s
, get_track(s
, p
, alloc
));
3674 static int list_locations(struct kmem_cache
*s
, char *buf
,
3675 enum track_item alloc
)
3679 struct loc_track t
= { 0, 0, NULL
};
3681 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3682 sizeof(unsigned long), GFP_KERNEL
);
3684 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3687 return sprintf(buf
, "Out of memory\n");
3689 /* Push back cpu slabs */
3692 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3693 struct kmem_cache_node
*n
= get_node(s
, node
);
3694 unsigned long flags
;
3697 if (!atomic_long_read(&n
->nr_slabs
))
3700 spin_lock_irqsave(&n
->list_lock
, flags
);
3701 list_for_each_entry(page
, &n
->partial
, lru
)
3702 process_slab(&t
, s
, page
, alloc
, map
);
3703 list_for_each_entry(page
, &n
->full
, lru
)
3704 process_slab(&t
, s
, page
, alloc
, map
);
3705 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3708 for (i
= 0; i
< t
.count
; i
++) {
3709 struct location
*l
= &t
.loc
[i
];
3711 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3713 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3716 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3718 len
+= sprintf(buf
+ len
, "<not-available>");
3720 if (l
->sum_time
!= l
->min_time
) {
3721 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3723 (long)div_u64(l
->sum_time
, l
->count
),
3726 len
+= sprintf(buf
+ len
, " age=%ld",
3729 if (l
->min_pid
!= l
->max_pid
)
3730 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3731 l
->min_pid
, l
->max_pid
);
3733 len
+= sprintf(buf
+ len
, " pid=%ld",
3736 if (num_online_cpus() > 1 &&
3737 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3738 len
< PAGE_SIZE
- 60) {
3739 len
+= sprintf(buf
+ len
, " cpus=");
3740 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3741 to_cpumask(l
->cpus
));
3744 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3745 len
< PAGE_SIZE
- 60) {
3746 len
+= sprintf(buf
+ len
, " nodes=");
3747 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3751 len
+= sprintf(buf
+ len
, "\n");
3757 len
+= sprintf(buf
, "No data\n");
3761 enum slab_stat_type
{
3762 SL_ALL
, /* All slabs */
3763 SL_PARTIAL
, /* Only partially allocated slabs */
3764 SL_CPU
, /* Only slabs used for cpu caches */
3765 SL_OBJECTS
, /* Determine allocated objects not slabs */
3766 SL_TOTAL
/* Determine object capacity not slabs */
3769 #define SO_ALL (1 << SL_ALL)
3770 #define SO_PARTIAL (1 << SL_PARTIAL)
3771 #define SO_CPU (1 << SL_CPU)
3772 #define SO_OBJECTS (1 << SL_OBJECTS)
3773 #define SO_TOTAL (1 << SL_TOTAL)
3775 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3776 char *buf
, unsigned long flags
)
3778 unsigned long total
= 0;
3781 unsigned long *nodes
;
3782 unsigned long *per_cpu
;
3784 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3787 per_cpu
= nodes
+ nr_node_ids
;
3789 if (flags
& SO_CPU
) {
3792 for_each_possible_cpu(cpu
) {
3793 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3795 if (!c
|| c
->node
< 0)
3799 if (flags
& SO_TOTAL
)
3800 x
= c
->page
->objects
;
3801 else if (flags
& SO_OBJECTS
)
3807 nodes
[c
->node
] += x
;
3813 if (flags
& SO_ALL
) {
3814 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3815 struct kmem_cache_node
*n
= get_node(s
, node
);
3817 if (flags
& SO_TOTAL
)
3818 x
= atomic_long_read(&n
->total_objects
);
3819 else if (flags
& SO_OBJECTS
)
3820 x
= atomic_long_read(&n
->total_objects
) -
3821 count_partial(n
, count_free
);
3824 x
= atomic_long_read(&n
->nr_slabs
);
3829 } else if (flags
& SO_PARTIAL
) {
3830 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3831 struct kmem_cache_node
*n
= get_node(s
, node
);
3833 if (flags
& SO_TOTAL
)
3834 x
= count_partial(n
, count_total
);
3835 else if (flags
& SO_OBJECTS
)
3836 x
= count_partial(n
, count_inuse
);
3843 x
= sprintf(buf
, "%lu", total
);
3845 for_each_node_state(node
, N_NORMAL_MEMORY
)
3847 x
+= sprintf(buf
+ x
, " N%d=%lu",
3851 return x
+ sprintf(buf
+ x
, "\n");
3854 static int any_slab_objects(struct kmem_cache
*s
)
3858 for_each_online_node(node
) {
3859 struct kmem_cache_node
*n
= get_node(s
, node
);
3864 if (atomic_long_read(&n
->total_objects
))
3870 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3871 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3873 struct slab_attribute
{
3874 struct attribute attr
;
3875 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3876 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3879 #define SLAB_ATTR_RO(_name) \
3880 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3882 #define SLAB_ATTR(_name) \
3883 static struct slab_attribute _name##_attr = \
3884 __ATTR(_name, 0644, _name##_show, _name##_store)
3886 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3888 return sprintf(buf
, "%d\n", s
->size
);
3890 SLAB_ATTR_RO(slab_size
);
3892 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3894 return sprintf(buf
, "%d\n", s
->align
);
3896 SLAB_ATTR_RO(align
);
3898 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3900 return sprintf(buf
, "%d\n", s
->objsize
);
3902 SLAB_ATTR_RO(object_size
);
3904 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3906 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3908 SLAB_ATTR_RO(objs_per_slab
);
3910 static ssize_t
order_store(struct kmem_cache
*s
,
3911 const char *buf
, size_t length
)
3913 unsigned long order
;
3916 err
= strict_strtoul(buf
, 10, &order
);
3920 if (order
> slub_max_order
|| order
< slub_min_order
)
3923 calculate_sizes(s
, order
);
3927 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3929 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3933 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3935 return sprintf(buf
, "%lu\n", s
->min_partial
);
3938 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3944 err
= strict_strtoul(buf
, 10, &min
);
3948 set_min_partial(s
, min
);
3951 SLAB_ATTR(min_partial
);
3953 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3956 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3958 return n
+ sprintf(buf
+ n
, "\n");
3964 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3966 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3968 SLAB_ATTR_RO(aliases
);
3970 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3972 return show_slab_objects(s
, buf
, SO_ALL
);
3974 SLAB_ATTR_RO(slabs
);
3976 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3978 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3980 SLAB_ATTR_RO(partial
);
3982 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3984 return show_slab_objects(s
, buf
, SO_CPU
);
3986 SLAB_ATTR_RO(cpu_slabs
);
3988 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3990 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3992 SLAB_ATTR_RO(objects
);
3994 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3996 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3998 SLAB_ATTR_RO(objects_partial
);
4000 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4002 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4004 SLAB_ATTR_RO(total_objects
);
4006 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4008 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4011 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4012 const char *buf
, size_t length
)
4014 s
->flags
&= ~SLAB_DEBUG_FREE
;
4016 s
->flags
|= SLAB_DEBUG_FREE
;
4019 SLAB_ATTR(sanity_checks
);
4021 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4023 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4026 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4029 s
->flags
&= ~SLAB_TRACE
;
4031 s
->flags
|= SLAB_TRACE
;
4036 #ifdef CONFIG_FAILSLAB
4037 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4039 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4042 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4045 s
->flags
&= ~SLAB_FAILSLAB
;
4047 s
->flags
|= SLAB_FAILSLAB
;
4050 SLAB_ATTR(failslab
);
4053 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4055 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4058 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4059 const char *buf
, size_t length
)
4061 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4063 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4066 SLAB_ATTR(reclaim_account
);
4068 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4070 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4072 SLAB_ATTR_RO(hwcache_align
);
4074 #ifdef CONFIG_ZONE_DMA
4075 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4077 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4079 SLAB_ATTR_RO(cache_dma
);
4082 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4084 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4086 SLAB_ATTR_RO(destroy_by_rcu
);
4088 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4090 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4093 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4094 const char *buf
, size_t length
)
4096 if (any_slab_objects(s
))
4099 s
->flags
&= ~SLAB_RED_ZONE
;
4101 s
->flags
|= SLAB_RED_ZONE
;
4102 calculate_sizes(s
, -1);
4105 SLAB_ATTR(red_zone
);
4107 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4109 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4112 static ssize_t
poison_store(struct kmem_cache
*s
,
4113 const char *buf
, size_t length
)
4115 if (any_slab_objects(s
))
4118 s
->flags
&= ~SLAB_POISON
;
4120 s
->flags
|= SLAB_POISON
;
4121 calculate_sizes(s
, -1);
4126 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4128 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4131 static ssize_t
store_user_store(struct kmem_cache
*s
,
4132 const char *buf
, size_t length
)
4134 if (any_slab_objects(s
))
4137 s
->flags
&= ~SLAB_STORE_USER
;
4139 s
->flags
|= SLAB_STORE_USER
;
4140 calculate_sizes(s
, -1);
4143 SLAB_ATTR(store_user
);
4145 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4150 static ssize_t
validate_store(struct kmem_cache
*s
,
4151 const char *buf
, size_t length
)
4155 if (buf
[0] == '1') {
4156 ret
= validate_slab_cache(s
);
4162 SLAB_ATTR(validate
);
4164 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4169 static ssize_t
shrink_store(struct kmem_cache
*s
,
4170 const char *buf
, size_t length
)
4172 if (buf
[0] == '1') {
4173 int rc
= kmem_cache_shrink(s
);
4183 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4185 if (!(s
->flags
& SLAB_STORE_USER
))
4187 return list_locations(s
, buf
, TRACK_ALLOC
);
4189 SLAB_ATTR_RO(alloc_calls
);
4191 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4193 if (!(s
->flags
& SLAB_STORE_USER
))
4195 return list_locations(s
, buf
, TRACK_FREE
);
4197 SLAB_ATTR_RO(free_calls
);
4200 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4202 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4205 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4206 const char *buf
, size_t length
)
4208 unsigned long ratio
;
4211 err
= strict_strtoul(buf
, 10, &ratio
);
4216 s
->remote_node_defrag_ratio
= ratio
* 10;
4220 SLAB_ATTR(remote_node_defrag_ratio
);
4223 #ifdef CONFIG_SLUB_STATS
4224 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4226 unsigned long sum
= 0;
4229 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4234 for_each_online_cpu(cpu
) {
4235 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4241 len
= sprintf(buf
, "%lu", sum
);
4244 for_each_online_cpu(cpu
) {
4245 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4246 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4250 return len
+ sprintf(buf
+ len
, "\n");
4253 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4257 for_each_online_cpu(cpu
)
4258 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4261 #define STAT_ATTR(si, text) \
4262 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4264 return show_stat(s, buf, si); \
4266 static ssize_t text##_store(struct kmem_cache *s, \
4267 const char *buf, size_t length) \
4269 if (buf[0] != '0') \
4271 clear_stat(s, si); \
4276 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4277 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4278 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4279 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4280 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4281 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4282 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4283 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4284 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4285 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4286 STAT_ATTR(FREE_SLAB
, free_slab
);
4287 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4288 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4289 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4290 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4291 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4292 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4293 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4296 static struct attribute
*slab_attrs
[] = {
4297 &slab_size_attr
.attr
,
4298 &object_size_attr
.attr
,
4299 &objs_per_slab_attr
.attr
,
4301 &min_partial_attr
.attr
,
4303 &objects_partial_attr
.attr
,
4304 &total_objects_attr
.attr
,
4307 &cpu_slabs_attr
.attr
,
4311 &sanity_checks_attr
.attr
,
4313 &hwcache_align_attr
.attr
,
4314 &reclaim_account_attr
.attr
,
4315 &destroy_by_rcu_attr
.attr
,
4316 &red_zone_attr
.attr
,
4318 &store_user_attr
.attr
,
4319 &validate_attr
.attr
,
4321 &alloc_calls_attr
.attr
,
4322 &free_calls_attr
.attr
,
4323 #ifdef CONFIG_ZONE_DMA
4324 &cache_dma_attr
.attr
,
4327 &remote_node_defrag_ratio_attr
.attr
,
4329 #ifdef CONFIG_SLUB_STATS
4330 &alloc_fastpath_attr
.attr
,
4331 &alloc_slowpath_attr
.attr
,
4332 &free_fastpath_attr
.attr
,
4333 &free_slowpath_attr
.attr
,
4334 &free_frozen_attr
.attr
,
4335 &free_add_partial_attr
.attr
,
4336 &free_remove_partial_attr
.attr
,
4337 &alloc_from_partial_attr
.attr
,
4338 &alloc_slab_attr
.attr
,
4339 &alloc_refill_attr
.attr
,
4340 &free_slab_attr
.attr
,
4341 &cpuslab_flush_attr
.attr
,
4342 &deactivate_full_attr
.attr
,
4343 &deactivate_empty_attr
.attr
,
4344 &deactivate_to_head_attr
.attr
,
4345 &deactivate_to_tail_attr
.attr
,
4346 &deactivate_remote_frees_attr
.attr
,
4347 &order_fallback_attr
.attr
,
4349 #ifdef CONFIG_FAILSLAB
4350 &failslab_attr
.attr
,
4356 static struct attribute_group slab_attr_group
= {
4357 .attrs
= slab_attrs
,
4360 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4361 struct attribute
*attr
,
4364 struct slab_attribute
*attribute
;
4365 struct kmem_cache
*s
;
4368 attribute
= to_slab_attr(attr
);
4371 if (!attribute
->show
)
4374 err
= attribute
->show(s
, buf
);
4379 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4380 struct attribute
*attr
,
4381 const char *buf
, size_t len
)
4383 struct slab_attribute
*attribute
;
4384 struct kmem_cache
*s
;
4387 attribute
= to_slab_attr(attr
);
4390 if (!attribute
->store
)
4393 err
= attribute
->store(s
, buf
, len
);
4398 static void kmem_cache_release(struct kobject
*kobj
)
4400 struct kmem_cache
*s
= to_slab(kobj
);
4405 static const struct sysfs_ops slab_sysfs_ops
= {
4406 .show
= slab_attr_show
,
4407 .store
= slab_attr_store
,
4410 static struct kobj_type slab_ktype
= {
4411 .sysfs_ops
= &slab_sysfs_ops
,
4412 .release
= kmem_cache_release
4415 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4417 struct kobj_type
*ktype
= get_ktype(kobj
);
4419 if (ktype
== &slab_ktype
)
4424 static const struct kset_uevent_ops slab_uevent_ops
= {
4425 .filter
= uevent_filter
,
4428 static struct kset
*slab_kset
;
4430 #define ID_STR_LENGTH 64
4432 /* Create a unique string id for a slab cache:
4434 * Format :[flags-]size
4436 static char *create_unique_id(struct kmem_cache
*s
)
4438 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4445 * First flags affecting slabcache operations. We will only
4446 * get here for aliasable slabs so we do not need to support
4447 * too many flags. The flags here must cover all flags that
4448 * are matched during merging to guarantee that the id is
4451 if (s
->flags
& SLAB_CACHE_DMA
)
4453 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4455 if (s
->flags
& SLAB_DEBUG_FREE
)
4457 if (!(s
->flags
& SLAB_NOTRACK
))
4461 p
+= sprintf(p
, "%07d", s
->size
);
4462 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4466 static int sysfs_slab_add(struct kmem_cache
*s
)
4472 if (slab_state
< SYSFS
)
4473 /* Defer until later */
4476 unmergeable
= slab_unmergeable(s
);
4479 * Slabcache can never be merged so we can use the name proper.
4480 * This is typically the case for debug situations. In that
4481 * case we can catch duplicate names easily.
4483 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4487 * Create a unique name for the slab as a target
4490 name
= create_unique_id(s
);
4493 s
->kobj
.kset
= slab_kset
;
4494 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4496 kobject_put(&s
->kobj
);
4500 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4502 kobject_del(&s
->kobj
);
4503 kobject_put(&s
->kobj
);
4506 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4508 /* Setup first alias */
4509 sysfs_slab_alias(s
, s
->name
);
4515 static void sysfs_slab_remove(struct kmem_cache
*s
)
4517 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4518 kobject_del(&s
->kobj
);
4519 kobject_put(&s
->kobj
);
4523 * Need to buffer aliases during bootup until sysfs becomes
4524 * available lest we lose that information.
4526 struct saved_alias
{
4527 struct kmem_cache
*s
;
4529 struct saved_alias
*next
;
4532 static struct saved_alias
*alias_list
;
4534 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4536 struct saved_alias
*al
;
4538 if (slab_state
== SYSFS
) {
4540 * If we have a leftover link then remove it.
4542 sysfs_remove_link(&slab_kset
->kobj
, name
);
4543 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4546 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4552 al
->next
= alias_list
;
4557 static int __init
slab_sysfs_init(void)
4559 struct kmem_cache
*s
;
4562 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4564 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4570 list_for_each_entry(s
, &slab_caches
, list
) {
4571 err
= sysfs_slab_add(s
);
4573 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4574 " to sysfs\n", s
->name
);
4577 while (alias_list
) {
4578 struct saved_alias
*al
= alias_list
;
4580 alias_list
= alias_list
->next
;
4581 err
= sysfs_slab_alias(al
->s
, al
->name
);
4583 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4584 " %s to sysfs\n", s
->name
);
4592 __initcall(slab_sysfs_init
);
4596 * The /proc/slabinfo ABI
4598 #ifdef CONFIG_SLABINFO
4599 static void print_slabinfo_header(struct seq_file
*m
)
4601 seq_puts(m
, "slabinfo - version: 2.1\n");
4602 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4603 "<objperslab> <pagesperslab>");
4604 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4605 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4609 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4613 down_read(&slub_lock
);
4615 print_slabinfo_header(m
);
4617 return seq_list_start(&slab_caches
, *pos
);
4620 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4622 return seq_list_next(p
, &slab_caches
, pos
);
4625 static void s_stop(struct seq_file
*m
, void *p
)
4627 up_read(&slub_lock
);
4630 static int s_show(struct seq_file
*m
, void *p
)
4632 unsigned long nr_partials
= 0;
4633 unsigned long nr_slabs
= 0;
4634 unsigned long nr_inuse
= 0;
4635 unsigned long nr_objs
= 0;
4636 unsigned long nr_free
= 0;
4637 struct kmem_cache
*s
;
4640 s
= list_entry(p
, struct kmem_cache
, list
);
4642 for_each_online_node(node
) {
4643 struct kmem_cache_node
*n
= get_node(s
, node
);
4648 nr_partials
+= n
->nr_partial
;
4649 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4650 nr_objs
+= atomic_long_read(&n
->total_objects
);
4651 nr_free
+= count_partial(n
, count_free
);
4654 nr_inuse
= nr_objs
- nr_free
;
4656 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4657 nr_objs
, s
->size
, oo_objects(s
->oo
),
4658 (1 << oo_order(s
->oo
)));
4659 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4660 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4666 static const struct seq_operations slabinfo_op
= {
4673 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4675 return seq_open(file
, &slabinfo_op
);
4678 static const struct file_operations proc_slabinfo_operations
= {
4679 .open
= slabinfo_open
,
4681 .llseek
= seq_lseek
,
4682 .release
= seq_release
,
4685 static int __init
slab_proc_init(void)
4687 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4690 module_init(slab_proc_init
);
4691 #endif /* CONFIG_SLABINFO */