2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemtrace.h>
21 #include <linux/kmemcheck.h>
22 #include <linux/cpu.h>
23 #include <linux/cpuset.h>
24 #include <linux/kmemleak.h>
25 #include <linux/mempolicy.h>
26 #include <linux/ctype.h>
27 #include <linux/debugobjects.h>
28 #include <linux/kallsyms.h>
29 #include <linux/memory.h>
30 #include <linux/math64.h>
31 #include <linux/fault-inject.h>
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #ifdef CONFIG_SLUB_DEBUG
118 * Issues still to be resolved:
120 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
122 * - Variable sizing of the per node arrays
125 /* Enable to test recovery from slab corruption on boot */
126 #undef SLUB_RESILIENCY_TEST
129 * Mininum number of partial slabs. These will be left on the partial
130 * lists even if they are empty. kmem_cache_shrink may reclaim them.
132 #define MIN_PARTIAL 5
135 * Maximum number of desirable partial slabs.
136 * The existence of more partial slabs makes kmem_cache_shrink
137 * sort the partial list by the number of objects in the.
139 #define MAX_PARTIAL 10
141 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
142 SLAB_POISON | SLAB_STORE_USER)
145 * Set of flags that will prevent slab merging
147 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
148 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE)
150 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
151 SLAB_CACHE_DMA | SLAB_NOTRACK)
153 #ifndef ARCH_KMALLOC_MINALIGN
154 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
157 #ifndef ARCH_SLAB_MINALIGN
158 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
162 #define OO_MASK ((1 << OO_SHIFT) - 1)
163 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
165 /* Internal SLUB flags */
166 #define __OBJECT_POISON 0x80000000 /* Poison object */
167 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
169 static int kmem_size
= sizeof(struct kmem_cache
);
172 static struct notifier_block slab_notifier
;
176 DOWN
, /* No slab functionality available */
177 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
178 UP
, /* Everything works but does not show up in sysfs */
183 * The slab allocator is initialized with interrupts disabled. Therefore, make
184 * sure early boot allocations don't accidentally enable interrupts.
186 static gfp_t slab_gfp_mask __read_mostly
= SLAB_GFP_BOOT_MASK
;
188 /* A list of all slab caches on the system */
189 static DECLARE_RWSEM(slub_lock
);
190 static LIST_HEAD(slab_caches
);
193 * Tracking user of a slab.
196 unsigned long addr
; /* Called from address */
197 int cpu
; /* Was running on cpu */
198 int pid
; /* Pid context */
199 unsigned long when
; /* When did the operation occur */
202 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
204 #ifdef CONFIG_SLUB_DEBUG
205 static int sysfs_slab_add(struct kmem_cache
*);
206 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
207 static void sysfs_slab_remove(struct kmem_cache
*);
210 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
211 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
213 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
220 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
222 #ifdef CONFIG_SLUB_STATS
227 /********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
231 int slab_is_available(void)
233 return slab_state
>= UP
;
236 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
239 return s
->node
[node
];
241 return &s
->local_node
;
245 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
248 return s
->cpu_slab
[cpu
];
254 /* Verify that a pointer has an address that is valid within a slab page */
255 static inline int check_valid_pointer(struct kmem_cache
*s
,
256 struct page
*page
, const void *object
)
263 base
= page_address(page
);
264 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
265 (object
- base
) % s
->size
) {
273 * Slow version of get and set free pointer.
275 * This version requires touching the cache lines of kmem_cache which
276 * we avoid to do in the fast alloc free paths. There we obtain the offset
277 * from the page struct.
279 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
281 return *(void **)(object
+ s
->offset
);
284 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
286 *(void **)(object
+ s
->offset
) = fp
;
289 /* Loop over all objects in a slab */
290 #define for_each_object(__p, __s, __addr, __objects) \
291 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
295 #define for_each_free_object(__p, __s, __free) \
296 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
298 /* Determine object index from a given position */
299 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
301 return (p
- addr
) / s
->size
;
304 static inline struct kmem_cache_order_objects
oo_make(int order
,
307 struct kmem_cache_order_objects x
= {
308 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
314 static inline int oo_order(struct kmem_cache_order_objects x
)
316 return x
.x
>> OO_SHIFT
;
319 static inline int oo_objects(struct kmem_cache_order_objects x
)
321 return x
.x
& OO_MASK
;
324 #ifdef CONFIG_SLUB_DEBUG
328 #ifdef CONFIG_SLUB_DEBUG_ON
329 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
331 static int slub_debug
;
334 static char *slub_debug_slabs
;
339 static void print_section(char *text
, u8
*addr
, unsigned int length
)
347 for (i
= 0; i
< length
; i
++) {
349 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
352 printk(KERN_CONT
" %02x", addr
[i
]);
354 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
356 printk(KERN_CONT
" %s\n", ascii
);
363 printk(KERN_CONT
" ");
367 printk(KERN_CONT
" %s\n", ascii
);
371 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
372 enum track_item alloc
)
377 p
= object
+ s
->offset
+ sizeof(void *);
379 p
= object
+ s
->inuse
;
384 static void set_track(struct kmem_cache
*s
, void *object
,
385 enum track_item alloc
, unsigned long addr
)
387 struct track
*p
= get_track(s
, object
, alloc
);
391 p
->cpu
= smp_processor_id();
392 p
->pid
= current
->pid
;
395 memset(p
, 0, sizeof(struct track
));
398 static void init_tracking(struct kmem_cache
*s
, void *object
)
400 if (!(s
->flags
& SLAB_STORE_USER
))
403 set_track(s
, object
, TRACK_FREE
, 0UL);
404 set_track(s
, object
, TRACK_ALLOC
, 0UL);
407 static void print_track(const char *s
, struct track
*t
)
412 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
413 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
416 static void print_tracking(struct kmem_cache
*s
, void *object
)
418 if (!(s
->flags
& SLAB_STORE_USER
))
421 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
422 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
425 static void print_page_info(struct page
*page
)
427 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
428 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
432 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
438 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
440 printk(KERN_ERR
"========================================"
441 "=====================================\n");
442 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
443 printk(KERN_ERR
"----------------------------------------"
444 "-------------------------------------\n\n");
447 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
453 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
455 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
458 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
460 unsigned int off
; /* Offset of last byte */
461 u8
*addr
= page_address(page
);
463 print_tracking(s
, p
);
465 print_page_info(page
);
467 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
468 p
, p
- addr
, get_freepointer(s
, p
));
471 print_section("Bytes b4", p
- 16, 16);
473 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
475 if (s
->flags
& SLAB_RED_ZONE
)
476 print_section("Redzone", p
+ s
->objsize
,
477 s
->inuse
- s
->objsize
);
480 off
= s
->offset
+ sizeof(void *);
484 if (s
->flags
& SLAB_STORE_USER
)
485 off
+= 2 * sizeof(struct track
);
488 /* Beginning of the filler is the free pointer */
489 print_section("Padding", p
+ off
, s
->size
- off
);
494 static void object_err(struct kmem_cache
*s
, struct page
*page
,
495 u8
*object
, char *reason
)
497 slab_bug(s
, "%s", reason
);
498 print_trailer(s
, page
, object
);
501 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
507 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
509 slab_bug(s
, "%s", buf
);
510 print_page_info(page
);
514 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
518 if (s
->flags
& __OBJECT_POISON
) {
519 memset(p
, POISON_FREE
, s
->objsize
- 1);
520 p
[s
->objsize
- 1] = POISON_END
;
523 if (s
->flags
& SLAB_RED_ZONE
)
524 memset(p
+ s
->objsize
,
525 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
526 s
->inuse
- s
->objsize
);
529 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
532 if (*start
!= (u8
)value
)
540 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
541 void *from
, void *to
)
543 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
544 memset(from
, data
, to
- from
);
547 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
548 u8
*object
, char *what
,
549 u8
*start
, unsigned int value
, unsigned int bytes
)
554 fault
= check_bytes(start
, value
, bytes
);
559 while (end
> fault
&& end
[-1] == value
)
562 slab_bug(s
, "%s overwritten", what
);
563 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
564 fault
, end
- 1, fault
[0], value
);
565 print_trailer(s
, page
, object
);
567 restore_bytes(s
, what
, value
, fault
, end
);
575 * Bytes of the object to be managed.
576 * If the freepointer may overlay the object then the free
577 * pointer is the first word of the object.
579 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
582 * object + s->objsize
583 * Padding to reach word boundary. This is also used for Redzoning.
584 * Padding is extended by another word if Redzoning is enabled and
587 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
588 * 0xcc (RED_ACTIVE) for objects in use.
591 * Meta data starts here.
593 * A. Free pointer (if we cannot overwrite object on free)
594 * B. Tracking data for SLAB_STORE_USER
595 * C. Padding to reach required alignment boundary or at mininum
596 * one word if debugging is on to be able to detect writes
597 * before the word boundary.
599 * Padding is done using 0x5a (POISON_INUSE)
602 * Nothing is used beyond s->size.
604 * If slabcaches are merged then the objsize and inuse boundaries are mostly
605 * ignored. And therefore no slab options that rely on these boundaries
606 * may be used with merged slabcaches.
609 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
611 unsigned long off
= s
->inuse
; /* The end of info */
614 /* Freepointer is placed after the object. */
615 off
+= sizeof(void *);
617 if (s
->flags
& SLAB_STORE_USER
)
618 /* We also have user information there */
619 off
+= 2 * sizeof(struct track
);
624 return check_bytes_and_report(s
, page
, p
, "Object padding",
625 p
+ off
, POISON_INUSE
, s
->size
- off
);
628 /* Check the pad bytes at the end of a slab page */
629 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
637 if (!(s
->flags
& SLAB_POISON
))
640 start
= page_address(page
);
641 length
= (PAGE_SIZE
<< compound_order(page
));
642 end
= start
+ length
;
643 remainder
= length
% s
->size
;
647 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
650 while (end
> fault
&& end
[-1] == POISON_INUSE
)
653 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
654 print_section("Padding", end
- remainder
, remainder
);
656 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
660 static int check_object(struct kmem_cache
*s
, struct page
*page
,
661 void *object
, int active
)
664 u8
*endobject
= object
+ s
->objsize
;
666 if (s
->flags
& SLAB_RED_ZONE
) {
668 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
670 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
671 endobject
, red
, s
->inuse
- s
->objsize
))
674 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
675 check_bytes_and_report(s
, page
, p
, "Alignment padding",
676 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
680 if (s
->flags
& SLAB_POISON
) {
681 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
682 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
683 POISON_FREE
, s
->objsize
- 1) ||
684 !check_bytes_and_report(s
, page
, p
, "Poison",
685 p
+ s
->objsize
- 1, POISON_END
, 1)))
688 * check_pad_bytes cleans up on its own.
690 check_pad_bytes(s
, page
, p
);
693 if (!s
->offset
&& active
)
695 * Object and freepointer overlap. Cannot check
696 * freepointer while object is allocated.
700 /* Check free pointer validity */
701 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
702 object_err(s
, page
, p
, "Freepointer corrupt");
704 * No choice but to zap it and thus lose the remainder
705 * of the free objects in this slab. May cause
706 * another error because the object count is now wrong.
708 set_freepointer(s
, p
, NULL
);
714 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
718 VM_BUG_ON(!irqs_disabled());
720 if (!PageSlab(page
)) {
721 slab_err(s
, page
, "Not a valid slab page");
725 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
726 if (page
->objects
> maxobj
) {
727 slab_err(s
, page
, "objects %u > max %u",
728 s
->name
, page
->objects
, maxobj
);
731 if (page
->inuse
> page
->objects
) {
732 slab_err(s
, page
, "inuse %u > max %u",
733 s
->name
, page
->inuse
, page
->objects
);
736 /* Slab_pad_check fixes things up after itself */
737 slab_pad_check(s
, page
);
742 * Determine if a certain object on a page is on the freelist. Must hold the
743 * slab lock to guarantee that the chains are in a consistent state.
745 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
748 void *fp
= page
->freelist
;
750 unsigned long max_objects
;
752 while (fp
&& nr
<= page
->objects
) {
755 if (!check_valid_pointer(s
, page
, fp
)) {
757 object_err(s
, page
, object
,
758 "Freechain corrupt");
759 set_freepointer(s
, object
, NULL
);
762 slab_err(s
, page
, "Freepointer corrupt");
763 page
->freelist
= NULL
;
764 page
->inuse
= page
->objects
;
765 slab_fix(s
, "Freelist cleared");
771 fp
= get_freepointer(s
, object
);
775 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
776 if (max_objects
> MAX_OBJS_PER_PAGE
)
777 max_objects
= MAX_OBJS_PER_PAGE
;
779 if (page
->objects
!= max_objects
) {
780 slab_err(s
, page
, "Wrong number of objects. Found %d but "
781 "should be %d", page
->objects
, max_objects
);
782 page
->objects
= max_objects
;
783 slab_fix(s
, "Number of objects adjusted.");
785 if (page
->inuse
!= page
->objects
- nr
) {
786 slab_err(s
, page
, "Wrong object count. Counter is %d but "
787 "counted were %d", page
->inuse
, page
->objects
- nr
);
788 page
->inuse
= page
->objects
- nr
;
789 slab_fix(s
, "Object count adjusted.");
791 return search
== NULL
;
794 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
797 if (s
->flags
& SLAB_TRACE
) {
798 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
800 alloc
? "alloc" : "free",
805 print_section("Object", (void *)object
, s
->objsize
);
812 * Tracking of fully allocated slabs for debugging purposes.
814 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
816 spin_lock(&n
->list_lock
);
817 list_add(&page
->lru
, &n
->full
);
818 spin_unlock(&n
->list_lock
);
821 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
823 struct kmem_cache_node
*n
;
825 if (!(s
->flags
& SLAB_STORE_USER
))
828 n
= get_node(s
, page_to_nid(page
));
830 spin_lock(&n
->list_lock
);
831 list_del(&page
->lru
);
832 spin_unlock(&n
->list_lock
);
835 /* Tracking of the number of slabs for debugging purposes */
836 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
838 struct kmem_cache_node
*n
= get_node(s
, node
);
840 return atomic_long_read(&n
->nr_slabs
);
843 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
845 struct kmem_cache_node
*n
= get_node(s
, node
);
848 * May be called early in order to allocate a slab for the
849 * kmem_cache_node structure. Solve the chicken-egg
850 * dilemma by deferring the increment of the count during
851 * bootstrap (see early_kmem_cache_node_alloc).
853 if (!NUMA_BUILD
|| n
) {
854 atomic_long_inc(&n
->nr_slabs
);
855 atomic_long_add(objects
, &n
->total_objects
);
858 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
860 struct kmem_cache_node
*n
= get_node(s
, node
);
862 atomic_long_dec(&n
->nr_slabs
);
863 atomic_long_sub(objects
, &n
->total_objects
);
866 /* Object debug checks for alloc/free paths */
867 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
870 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
873 init_object(s
, object
, 0);
874 init_tracking(s
, object
);
877 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
878 void *object
, unsigned long addr
)
880 if (!check_slab(s
, page
))
883 if (!on_freelist(s
, page
, object
)) {
884 object_err(s
, page
, object
, "Object already allocated");
888 if (!check_valid_pointer(s
, page
, object
)) {
889 object_err(s
, page
, object
, "Freelist Pointer check fails");
893 if (!check_object(s
, page
, object
, 0))
896 /* Success perform special debug activities for allocs */
897 if (s
->flags
& SLAB_STORE_USER
)
898 set_track(s
, object
, TRACK_ALLOC
, addr
);
899 trace(s
, page
, object
, 1);
900 init_object(s
, object
, 1);
904 if (PageSlab(page
)) {
906 * If this is a slab page then lets do the best we can
907 * to avoid issues in the future. Marking all objects
908 * as used avoids touching the remaining objects.
910 slab_fix(s
, "Marking all objects used");
911 page
->inuse
= page
->objects
;
912 page
->freelist
= NULL
;
917 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
918 void *object
, unsigned long addr
)
920 if (!check_slab(s
, page
))
923 if (!check_valid_pointer(s
, page
, object
)) {
924 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
928 if (on_freelist(s
, page
, object
)) {
929 object_err(s
, page
, object
, "Object already free");
933 if (!check_object(s
, page
, object
, 1))
936 if (unlikely(s
!= page
->slab
)) {
937 if (!PageSlab(page
)) {
938 slab_err(s
, page
, "Attempt to free object(0x%p) "
939 "outside of slab", object
);
940 } else if (!page
->slab
) {
942 "SLUB <none>: no slab for object 0x%p.\n",
946 object_err(s
, page
, object
,
947 "page slab pointer corrupt.");
951 /* Special debug activities for freeing objects */
952 if (!PageSlubFrozen(page
) && !page
->freelist
)
953 remove_full(s
, page
);
954 if (s
->flags
& SLAB_STORE_USER
)
955 set_track(s
, object
, TRACK_FREE
, addr
);
956 trace(s
, page
, object
, 0);
957 init_object(s
, object
, 0);
961 slab_fix(s
, "Object at 0x%p not freed", object
);
965 static int __init
setup_slub_debug(char *str
)
967 slub_debug
= DEBUG_DEFAULT_FLAGS
;
968 if (*str
++ != '=' || !*str
)
970 * No options specified. Switch on full debugging.
976 * No options but restriction on slabs. This means full
977 * debugging for slabs matching a pattern.
984 * Switch off all debugging measures.
989 * Determine which debug features should be switched on
991 for (; *str
&& *str
!= ','; str
++) {
992 switch (tolower(*str
)) {
994 slub_debug
|= SLAB_DEBUG_FREE
;
997 slub_debug
|= SLAB_RED_ZONE
;
1000 slub_debug
|= SLAB_POISON
;
1003 slub_debug
|= SLAB_STORE_USER
;
1006 slub_debug
|= SLAB_TRACE
;
1009 printk(KERN_ERR
"slub_debug option '%c' "
1010 "unknown. skipped\n", *str
);
1016 slub_debug_slabs
= str
+ 1;
1021 __setup("slub_debug", setup_slub_debug
);
1023 static unsigned long kmem_cache_flags(unsigned long objsize
,
1024 unsigned long flags
, const char *name
,
1025 void (*ctor
)(void *))
1028 * Enable debugging if selected on the kernel commandline.
1030 if (slub_debug
&& (!slub_debug_slabs
||
1031 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1032 flags
|= slub_debug
;
1037 static inline void setup_object_debug(struct kmem_cache
*s
,
1038 struct page
*page
, void *object
) {}
1040 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1041 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1043 static inline int free_debug_processing(struct kmem_cache
*s
,
1044 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1046 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1048 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1049 void *object
, int active
) { return 1; }
1050 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1051 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1052 unsigned long flags
, const char *name
,
1053 void (*ctor
)(void *))
1057 #define slub_debug 0
1059 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1061 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1063 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1068 * Slab allocation and freeing
1070 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1071 struct kmem_cache_order_objects oo
)
1073 int order
= oo_order(oo
);
1075 flags
|= __GFP_NOTRACK
;
1078 return alloc_pages(flags
, order
);
1080 return alloc_pages_node(node
, flags
, order
);
1083 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1086 struct kmem_cache_order_objects oo
= s
->oo
;
1088 flags
|= s
->allocflags
;
1090 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1092 if (unlikely(!page
)) {
1095 * Allocation may have failed due to fragmentation.
1096 * Try a lower order alloc if possible
1098 page
= alloc_slab_page(flags
, node
, oo
);
1102 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1105 if (kmemcheck_enabled
1106 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
)))
1108 int pages
= 1 << oo_order(oo
);
1110 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1113 * Objects from caches that have a constructor don't get
1114 * cleared when they're allocated, so we need to do it here.
1117 kmemcheck_mark_uninitialized_pages(page
, pages
);
1119 kmemcheck_mark_unallocated_pages(page
, pages
);
1122 page
->objects
= oo_objects(oo
);
1123 mod_zone_page_state(page_zone(page
),
1124 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1125 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1131 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1134 setup_object_debug(s
, page
, object
);
1135 if (unlikely(s
->ctor
))
1139 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1146 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1148 page
= allocate_slab(s
,
1149 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1153 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1155 page
->flags
|= 1 << PG_slab
;
1156 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1157 SLAB_STORE_USER
| SLAB_TRACE
))
1158 __SetPageSlubDebug(page
);
1160 start
= page_address(page
);
1162 if (unlikely(s
->flags
& SLAB_POISON
))
1163 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1166 for_each_object(p
, s
, start
, page
->objects
) {
1167 setup_object(s
, page
, last
);
1168 set_freepointer(s
, last
, p
);
1171 setup_object(s
, page
, last
);
1172 set_freepointer(s
, last
, NULL
);
1174 page
->freelist
= start
;
1180 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1182 int order
= compound_order(page
);
1183 int pages
= 1 << order
;
1185 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1188 slab_pad_check(s
, page
);
1189 for_each_object(p
, s
, page_address(page
),
1191 check_object(s
, page
, p
, 0);
1192 __ClearPageSlubDebug(page
);
1195 kmemcheck_free_shadow(page
, compound_order(page
));
1197 mod_zone_page_state(page_zone(page
),
1198 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1199 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1202 __ClearPageSlab(page
);
1203 reset_page_mapcount(page
);
1204 if (current
->reclaim_state
)
1205 current
->reclaim_state
->reclaimed_slab
+= pages
;
1206 __free_pages(page
, order
);
1209 static void rcu_free_slab(struct rcu_head
*h
)
1213 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1214 __free_slab(page
->slab
, page
);
1217 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1219 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1221 * RCU free overloads the RCU head over the LRU
1223 struct rcu_head
*head
= (void *)&page
->lru
;
1225 call_rcu(head
, rcu_free_slab
);
1227 __free_slab(s
, page
);
1230 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1232 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1237 * Per slab locking using the pagelock
1239 static __always_inline
void slab_lock(struct page
*page
)
1241 bit_spin_lock(PG_locked
, &page
->flags
);
1244 static __always_inline
void slab_unlock(struct page
*page
)
1246 __bit_spin_unlock(PG_locked
, &page
->flags
);
1249 static __always_inline
int slab_trylock(struct page
*page
)
1253 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1258 * Management of partially allocated slabs
1260 static void add_partial(struct kmem_cache_node
*n
,
1261 struct page
*page
, int tail
)
1263 spin_lock(&n
->list_lock
);
1266 list_add_tail(&page
->lru
, &n
->partial
);
1268 list_add(&page
->lru
, &n
->partial
);
1269 spin_unlock(&n
->list_lock
);
1272 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1274 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1276 spin_lock(&n
->list_lock
);
1277 list_del(&page
->lru
);
1279 spin_unlock(&n
->list_lock
);
1283 * Lock slab and remove from the partial list.
1285 * Must hold list_lock.
1287 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1290 if (slab_trylock(page
)) {
1291 list_del(&page
->lru
);
1293 __SetPageSlubFrozen(page
);
1300 * Try to allocate a partial slab from a specific node.
1302 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1307 * Racy check. If we mistakenly see no partial slabs then we
1308 * just allocate an empty slab. If we mistakenly try to get a
1309 * partial slab and there is none available then get_partials()
1312 if (!n
|| !n
->nr_partial
)
1315 spin_lock(&n
->list_lock
);
1316 list_for_each_entry(page
, &n
->partial
, lru
)
1317 if (lock_and_freeze_slab(n
, page
))
1321 spin_unlock(&n
->list_lock
);
1326 * Get a page from somewhere. Search in increasing NUMA distances.
1328 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1331 struct zonelist
*zonelist
;
1334 enum zone_type high_zoneidx
= gfp_zone(flags
);
1338 * The defrag ratio allows a configuration of the tradeoffs between
1339 * inter node defragmentation and node local allocations. A lower
1340 * defrag_ratio increases the tendency to do local allocations
1341 * instead of attempting to obtain partial slabs from other nodes.
1343 * If the defrag_ratio is set to 0 then kmalloc() always
1344 * returns node local objects. If the ratio is higher then kmalloc()
1345 * may return off node objects because partial slabs are obtained
1346 * from other nodes and filled up.
1348 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1349 * defrag_ratio = 1000) then every (well almost) allocation will
1350 * first attempt to defrag slab caches on other nodes. This means
1351 * scanning over all nodes to look for partial slabs which may be
1352 * expensive if we do it every time we are trying to find a slab
1353 * with available objects.
1355 if (!s
->remote_node_defrag_ratio
||
1356 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1359 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1360 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1361 struct kmem_cache_node
*n
;
1363 n
= get_node(s
, zone_to_nid(zone
));
1365 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1366 n
->nr_partial
> s
->min_partial
) {
1367 page
= get_partial_node(n
);
1377 * Get a partial page, lock it and return it.
1379 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1382 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1384 page
= get_partial_node(get_node(s
, searchnode
));
1385 if (page
|| (flags
& __GFP_THISNODE
))
1388 return get_any_partial(s
, flags
);
1392 * Move a page back to the lists.
1394 * Must be called with the slab lock held.
1396 * On exit the slab lock will have been dropped.
1398 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1400 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1401 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1403 __ClearPageSlubFrozen(page
);
1406 if (page
->freelist
) {
1407 add_partial(n
, page
, tail
);
1408 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1410 stat(c
, DEACTIVATE_FULL
);
1411 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1412 (s
->flags
& SLAB_STORE_USER
))
1417 stat(c
, DEACTIVATE_EMPTY
);
1418 if (n
->nr_partial
< s
->min_partial
) {
1420 * Adding an empty slab to the partial slabs in order
1421 * to avoid page allocator overhead. This slab needs
1422 * to come after the other slabs with objects in
1423 * so that the others get filled first. That way the
1424 * size of the partial list stays small.
1426 * kmem_cache_shrink can reclaim any empty slabs from
1429 add_partial(n
, page
, 1);
1433 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1434 discard_slab(s
, page
);
1440 * Remove the cpu slab
1442 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1444 struct page
*page
= c
->page
;
1448 stat(c
, DEACTIVATE_REMOTE_FREES
);
1450 * Merge cpu freelist into slab freelist. Typically we get here
1451 * because both freelists are empty. So this is unlikely
1454 while (unlikely(c
->freelist
)) {
1457 tail
= 0; /* Hot objects. Put the slab first */
1459 /* Retrieve object from cpu_freelist */
1460 object
= c
->freelist
;
1461 c
->freelist
= c
->freelist
[c
->offset
];
1463 /* And put onto the regular freelist */
1464 object
[c
->offset
] = page
->freelist
;
1465 page
->freelist
= object
;
1469 unfreeze_slab(s
, page
, tail
);
1472 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1474 stat(c
, CPUSLAB_FLUSH
);
1476 deactivate_slab(s
, c
);
1482 * Called from IPI handler with interrupts disabled.
1484 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1486 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1488 if (likely(c
&& c
->page
))
1492 static void flush_cpu_slab(void *d
)
1494 struct kmem_cache
*s
= d
;
1496 __flush_cpu_slab(s
, smp_processor_id());
1499 static void flush_all(struct kmem_cache
*s
)
1501 on_each_cpu(flush_cpu_slab
, s
, 1);
1505 * Check if the objects in a per cpu structure fit numa
1506 * locality expectations.
1508 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1511 if (node
!= -1 && c
->node
!= node
)
1518 * Slow path. The lockless freelist is empty or we need to perform
1521 * Interrupts are disabled.
1523 * Processing is still very fast if new objects have been freed to the
1524 * regular freelist. In that case we simply take over the regular freelist
1525 * as the lockless freelist and zap the regular freelist.
1527 * If that is not working then we fall back to the partial lists. We take the
1528 * first element of the freelist as the object to allocate now and move the
1529 * rest of the freelist to the lockless freelist.
1531 * And if we were unable to get a new slab from the partial slab lists then
1532 * we need to allocate a new slab. This is the slowest path since it involves
1533 * a call to the page allocator and the setup of a new slab.
1535 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1536 unsigned long addr
, struct kmem_cache_cpu
*c
)
1541 /* We handle __GFP_ZERO in the caller */
1542 gfpflags
&= ~__GFP_ZERO
;
1548 if (unlikely(!node_match(c
, node
)))
1551 stat(c
, ALLOC_REFILL
);
1554 object
= c
->page
->freelist
;
1555 if (unlikely(!object
))
1557 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1560 c
->freelist
= object
[c
->offset
];
1561 c
->page
->inuse
= c
->page
->objects
;
1562 c
->page
->freelist
= NULL
;
1563 c
->node
= page_to_nid(c
->page
);
1565 slab_unlock(c
->page
);
1566 stat(c
, ALLOC_SLOWPATH
);
1570 deactivate_slab(s
, c
);
1573 new = get_partial(s
, gfpflags
, node
);
1576 stat(c
, ALLOC_FROM_PARTIAL
);
1580 if (gfpflags
& __GFP_WAIT
)
1583 new = new_slab(s
, gfpflags
, node
);
1585 if (gfpflags
& __GFP_WAIT
)
1586 local_irq_disable();
1589 c
= get_cpu_slab(s
, smp_processor_id());
1590 stat(c
, ALLOC_SLAB
);
1594 __SetPageSlubFrozen(new);
1600 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1604 c
->page
->freelist
= object
[c
->offset
];
1610 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1611 * have the fastpath folded into their functions. So no function call
1612 * overhead for requests that can be satisfied on the fastpath.
1614 * The fastpath works by first checking if the lockless freelist can be used.
1615 * If not then __slab_alloc is called for slow processing.
1617 * Otherwise we can simply pick the next object from the lockless free list.
1619 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1620 gfp_t gfpflags
, int node
, unsigned long addr
)
1623 struct kmem_cache_cpu
*c
;
1624 unsigned long flags
;
1625 unsigned int objsize
;
1627 gfpflags
&= slab_gfp_mask
;
1629 lockdep_trace_alloc(gfpflags
);
1630 might_sleep_if(gfpflags
& __GFP_WAIT
);
1632 if (should_failslab(s
->objsize
, gfpflags
))
1635 local_irq_save(flags
);
1636 c
= get_cpu_slab(s
, smp_processor_id());
1637 objsize
= c
->objsize
;
1638 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1640 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1643 object
= c
->freelist
;
1644 c
->freelist
= object
[c
->offset
];
1645 stat(c
, ALLOC_FASTPATH
);
1647 local_irq_restore(flags
);
1649 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1650 memset(object
, 0, objsize
);
1652 kmemcheck_slab_alloc(s
, gfpflags
, object
, c
->objsize
);
1653 kmemleak_alloc_recursive(object
, objsize
, 1, s
->flags
, gfpflags
);
1658 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1660 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1662 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1666 EXPORT_SYMBOL(kmem_cache_alloc
);
1668 #ifdef CONFIG_KMEMTRACE
1669 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1671 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1673 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1677 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1679 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1681 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1682 s
->objsize
, s
->size
, gfpflags
, node
);
1686 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1689 #ifdef CONFIG_KMEMTRACE
1690 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1694 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1696 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1700 * Slow patch handling. This may still be called frequently since objects
1701 * have a longer lifetime than the cpu slabs in most processing loads.
1703 * So we still attempt to reduce cache line usage. Just take the slab
1704 * lock and free the item. If there is no additional partial page
1705 * handling required then we can return immediately.
1707 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1708 void *x
, unsigned long addr
, unsigned int offset
)
1711 void **object
= (void *)x
;
1712 struct kmem_cache_cpu
*c
;
1714 c
= get_cpu_slab(s
, raw_smp_processor_id());
1715 stat(c
, FREE_SLOWPATH
);
1718 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1722 prior
= object
[offset
] = page
->freelist
;
1723 page
->freelist
= object
;
1726 if (unlikely(PageSlubFrozen(page
))) {
1727 stat(c
, FREE_FROZEN
);
1731 if (unlikely(!page
->inuse
))
1735 * Objects left in the slab. If it was not on the partial list before
1738 if (unlikely(!prior
)) {
1739 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1740 stat(c
, FREE_ADD_PARTIAL
);
1750 * Slab still on the partial list.
1752 remove_partial(s
, page
);
1753 stat(c
, FREE_REMOVE_PARTIAL
);
1757 discard_slab(s
, page
);
1761 if (!free_debug_processing(s
, page
, x
, addr
))
1767 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1768 * can perform fastpath freeing without additional function calls.
1770 * The fastpath is only possible if we are freeing to the current cpu slab
1771 * of this processor. This typically the case if we have just allocated
1774 * If fastpath is not possible then fall back to __slab_free where we deal
1775 * with all sorts of special processing.
1777 static __always_inline
void slab_free(struct kmem_cache
*s
,
1778 struct page
*page
, void *x
, unsigned long addr
)
1780 void **object
= (void *)x
;
1781 struct kmem_cache_cpu
*c
;
1782 unsigned long flags
;
1784 kmemleak_free_recursive(x
, s
->flags
);
1785 local_irq_save(flags
);
1786 c
= get_cpu_slab(s
, smp_processor_id());
1787 kmemcheck_slab_free(s
, object
, c
->objsize
);
1788 debug_check_no_locks_freed(object
, c
->objsize
);
1789 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1790 debug_check_no_obj_freed(object
, c
->objsize
);
1791 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1792 object
[c
->offset
] = c
->freelist
;
1793 c
->freelist
= object
;
1794 stat(c
, FREE_FASTPATH
);
1796 __slab_free(s
, page
, x
, addr
, c
->offset
);
1798 local_irq_restore(flags
);
1801 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1805 page
= virt_to_head_page(x
);
1807 slab_free(s
, page
, x
, _RET_IP_
);
1809 trace_kmem_cache_free(_RET_IP_
, x
);
1811 EXPORT_SYMBOL(kmem_cache_free
);
1813 /* Figure out on which slab page the object resides */
1814 static struct page
*get_object_page(const void *x
)
1816 struct page
*page
= virt_to_head_page(x
);
1818 if (!PageSlab(page
))
1825 * Object placement in a slab is made very easy because we always start at
1826 * offset 0. If we tune the size of the object to the alignment then we can
1827 * get the required alignment by putting one properly sized object after
1830 * Notice that the allocation order determines the sizes of the per cpu
1831 * caches. Each processor has always one slab available for allocations.
1832 * Increasing the allocation order reduces the number of times that slabs
1833 * must be moved on and off the partial lists and is therefore a factor in
1838 * Mininum / Maximum order of slab pages. This influences locking overhead
1839 * and slab fragmentation. A higher order reduces the number of partial slabs
1840 * and increases the number of allocations possible without having to
1841 * take the list_lock.
1843 static int slub_min_order
;
1844 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1845 static int slub_min_objects
;
1848 * Merge control. If this is set then no merging of slab caches will occur.
1849 * (Could be removed. This was introduced to pacify the merge skeptics.)
1851 static int slub_nomerge
;
1854 * Calculate the order of allocation given an slab object size.
1856 * The order of allocation has significant impact on performance and other
1857 * system components. Generally order 0 allocations should be preferred since
1858 * order 0 does not cause fragmentation in the page allocator. Larger objects
1859 * be problematic to put into order 0 slabs because there may be too much
1860 * unused space left. We go to a higher order if more than 1/16th of the slab
1863 * In order to reach satisfactory performance we must ensure that a minimum
1864 * number of objects is in one slab. Otherwise we may generate too much
1865 * activity on the partial lists which requires taking the list_lock. This is
1866 * less a concern for large slabs though which are rarely used.
1868 * slub_max_order specifies the order where we begin to stop considering the
1869 * number of objects in a slab as critical. If we reach slub_max_order then
1870 * we try to keep the page order as low as possible. So we accept more waste
1871 * of space in favor of a small page order.
1873 * Higher order allocations also allow the placement of more objects in a
1874 * slab and thereby reduce object handling overhead. If the user has
1875 * requested a higher mininum order then we start with that one instead of
1876 * the smallest order which will fit the object.
1878 static inline int slab_order(int size
, int min_objects
,
1879 int max_order
, int fract_leftover
)
1883 int min_order
= slub_min_order
;
1885 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1886 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1888 for (order
= max(min_order
,
1889 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1890 order
<= max_order
; order
++) {
1892 unsigned long slab_size
= PAGE_SIZE
<< order
;
1894 if (slab_size
< min_objects
* size
)
1897 rem
= slab_size
% size
;
1899 if (rem
<= slab_size
/ fract_leftover
)
1907 static inline int calculate_order(int size
)
1915 * Attempt to find best configuration for a slab. This
1916 * works by first attempting to generate a layout with
1917 * the best configuration and backing off gradually.
1919 * First we reduce the acceptable waste in a slab. Then
1920 * we reduce the minimum objects required in a slab.
1922 min_objects
= slub_min_objects
;
1924 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1925 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1926 min_objects
= min(min_objects
, max_objects
);
1928 while (min_objects
> 1) {
1930 while (fraction
>= 4) {
1931 order
= slab_order(size
, min_objects
,
1932 slub_max_order
, fraction
);
1933 if (order
<= slub_max_order
)
1941 * We were unable to place multiple objects in a slab. Now
1942 * lets see if we can place a single object there.
1944 order
= slab_order(size
, 1, slub_max_order
, 1);
1945 if (order
<= slub_max_order
)
1949 * Doh this slab cannot be placed using slub_max_order.
1951 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1952 if (order
< MAX_ORDER
)
1958 * Figure out what the alignment of the objects will be.
1960 static unsigned long calculate_alignment(unsigned long flags
,
1961 unsigned long align
, unsigned long size
)
1964 * If the user wants hardware cache aligned objects then follow that
1965 * suggestion if the object is sufficiently large.
1967 * The hardware cache alignment cannot override the specified
1968 * alignment though. If that is greater then use it.
1970 if (flags
& SLAB_HWCACHE_ALIGN
) {
1971 unsigned long ralign
= cache_line_size();
1972 while (size
<= ralign
/ 2)
1974 align
= max(align
, ralign
);
1977 if (align
< ARCH_SLAB_MINALIGN
)
1978 align
= ARCH_SLAB_MINALIGN
;
1980 return ALIGN(align
, sizeof(void *));
1983 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1984 struct kmem_cache_cpu
*c
)
1989 c
->offset
= s
->offset
/ sizeof(void *);
1990 c
->objsize
= s
->objsize
;
1991 #ifdef CONFIG_SLUB_STATS
1992 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1997 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2000 spin_lock_init(&n
->list_lock
);
2001 INIT_LIST_HEAD(&n
->partial
);
2002 #ifdef CONFIG_SLUB_DEBUG
2003 atomic_long_set(&n
->nr_slabs
, 0);
2004 atomic_long_set(&n
->total_objects
, 0);
2005 INIT_LIST_HEAD(&n
->full
);
2011 * Per cpu array for per cpu structures.
2013 * The per cpu array places all kmem_cache_cpu structures from one processor
2014 * close together meaning that it becomes possible that multiple per cpu
2015 * structures are contained in one cacheline. This may be particularly
2016 * beneficial for the kmalloc caches.
2018 * A desktop system typically has around 60-80 slabs. With 100 here we are
2019 * likely able to get per cpu structures for all caches from the array defined
2020 * here. We must be able to cover all kmalloc caches during bootstrap.
2022 * If the per cpu array is exhausted then fall back to kmalloc
2023 * of individual cachelines. No sharing is possible then.
2025 #define NR_KMEM_CACHE_CPU 100
2027 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
2028 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
2030 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
2031 static DECLARE_BITMAP(kmem_cach_cpu_free_init_once
, CONFIG_NR_CPUS
);
2033 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
2034 int cpu
, gfp_t flags
)
2036 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
2039 per_cpu(kmem_cache_cpu_free
, cpu
) =
2040 (void *)c
->freelist
;
2042 /* Table overflow: So allocate ourselves */
2044 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2045 flags
, cpu_to_node(cpu
));
2050 init_kmem_cache_cpu(s
, c
);
2054 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2056 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2057 c
>= per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2061 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2062 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2065 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2069 for_each_online_cpu(cpu
) {
2070 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2073 s
->cpu_slab
[cpu
] = NULL
;
2074 free_kmem_cache_cpu(c
, cpu
);
2079 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2083 for_each_online_cpu(cpu
) {
2084 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2089 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2091 free_kmem_cache_cpus(s
);
2094 s
->cpu_slab
[cpu
] = c
;
2100 * Initialize the per cpu array.
2102 static void init_alloc_cpu_cpu(int cpu
)
2106 if (cpumask_test_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
)))
2109 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2110 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2112 cpumask_set_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
));
2115 static void __init
init_alloc_cpu(void)
2119 for_each_online_cpu(cpu
)
2120 init_alloc_cpu_cpu(cpu
);
2124 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2125 static inline void init_alloc_cpu(void) {}
2127 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2129 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2136 * No kmalloc_node yet so do it by hand. We know that this is the first
2137 * slab on the node for this slabcache. There are no concurrent accesses
2140 * Note that this function only works on the kmalloc_node_cache
2141 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2142 * memory on a fresh node that has no slab structures yet.
2144 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2147 struct kmem_cache_node
*n
;
2148 unsigned long flags
;
2150 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2152 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2155 if (page_to_nid(page
) != node
) {
2156 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2158 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2159 "in order to be able to continue\n");
2164 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2166 kmalloc_caches
->node
[node
] = n
;
2167 #ifdef CONFIG_SLUB_DEBUG
2168 init_object(kmalloc_caches
, n
, 1);
2169 init_tracking(kmalloc_caches
, n
);
2171 init_kmem_cache_node(n
, kmalloc_caches
);
2172 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2175 * lockdep requires consistent irq usage for each lock
2176 * so even though there cannot be a race this early in
2177 * the boot sequence, we still disable irqs.
2179 local_irq_save(flags
);
2180 add_partial(n
, page
, 0);
2181 local_irq_restore(flags
);
2184 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2188 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2189 struct kmem_cache_node
*n
= s
->node
[node
];
2190 if (n
&& n
!= &s
->local_node
)
2191 kmem_cache_free(kmalloc_caches
, n
);
2192 s
->node
[node
] = NULL
;
2196 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2201 if (slab_state
>= UP
)
2202 local_node
= page_to_nid(virt_to_page(s
));
2206 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2207 struct kmem_cache_node
*n
;
2209 if (local_node
== node
)
2212 if (slab_state
== DOWN
) {
2213 early_kmem_cache_node_alloc(gfpflags
, node
);
2216 n
= kmem_cache_alloc_node(kmalloc_caches
,
2220 free_kmem_cache_nodes(s
);
2226 init_kmem_cache_node(n
, s
);
2231 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2235 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2237 init_kmem_cache_node(&s
->local_node
, s
);
2242 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2244 if (min
< MIN_PARTIAL
)
2246 else if (min
> MAX_PARTIAL
)
2248 s
->min_partial
= min
;
2252 * calculate_sizes() determines the order and the distribution of data within
2255 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2257 unsigned long flags
= s
->flags
;
2258 unsigned long size
= s
->objsize
;
2259 unsigned long align
= s
->align
;
2263 * Round up object size to the next word boundary. We can only
2264 * place the free pointer at word boundaries and this determines
2265 * the possible location of the free pointer.
2267 size
= ALIGN(size
, sizeof(void *));
2269 #ifdef CONFIG_SLUB_DEBUG
2271 * Determine if we can poison the object itself. If the user of
2272 * the slab may touch the object after free or before allocation
2273 * then we should never poison the object itself.
2275 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2277 s
->flags
|= __OBJECT_POISON
;
2279 s
->flags
&= ~__OBJECT_POISON
;
2283 * If we are Redzoning then check if there is some space between the
2284 * end of the object and the free pointer. If not then add an
2285 * additional word to have some bytes to store Redzone information.
2287 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2288 size
+= sizeof(void *);
2292 * With that we have determined the number of bytes in actual use
2293 * by the object. This is the potential offset to the free pointer.
2297 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2300 * Relocate free pointer after the object if it is not
2301 * permitted to overwrite the first word of the object on
2304 * This is the case if we do RCU, have a constructor or
2305 * destructor or are poisoning the objects.
2308 size
+= sizeof(void *);
2311 #ifdef CONFIG_SLUB_DEBUG
2312 if (flags
& SLAB_STORE_USER
)
2314 * Need to store information about allocs and frees after
2317 size
+= 2 * sizeof(struct track
);
2319 if (flags
& SLAB_RED_ZONE
)
2321 * Add some empty padding so that we can catch
2322 * overwrites from earlier objects rather than let
2323 * tracking information or the free pointer be
2324 * corrupted if a user writes before the start
2327 size
+= sizeof(void *);
2331 * Determine the alignment based on various parameters that the
2332 * user specified and the dynamic determination of cache line size
2335 align
= calculate_alignment(flags
, align
, s
->objsize
);
2338 * SLUB stores one object immediately after another beginning from
2339 * offset 0. In order to align the objects we have to simply size
2340 * each object to conform to the alignment.
2342 size
= ALIGN(size
, align
);
2344 if (forced_order
>= 0)
2345 order
= forced_order
;
2347 order
= calculate_order(size
);
2354 s
->allocflags
|= __GFP_COMP
;
2356 if (s
->flags
& SLAB_CACHE_DMA
)
2357 s
->allocflags
|= SLUB_DMA
;
2359 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2360 s
->allocflags
|= __GFP_RECLAIMABLE
;
2363 * Determine the number of objects per slab
2365 s
->oo
= oo_make(order
, size
);
2366 s
->min
= oo_make(get_order(size
), size
);
2367 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2370 return !!oo_objects(s
->oo
);
2374 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2375 const char *name
, size_t size
,
2376 size_t align
, unsigned long flags
,
2377 void (*ctor
)(void *))
2379 memset(s
, 0, kmem_size
);
2384 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2386 if (!calculate_sizes(s
, -1))
2390 * The larger the object size is, the more pages we want on the partial
2391 * list to avoid pounding the page allocator excessively.
2393 set_min_partial(s
, ilog2(s
->size
));
2396 s
->remote_node_defrag_ratio
= 1000;
2398 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2401 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2403 free_kmem_cache_nodes(s
);
2405 if (flags
& SLAB_PANIC
)
2406 panic("Cannot create slab %s size=%lu realsize=%u "
2407 "order=%u offset=%u flags=%lx\n",
2408 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2414 * Check if a given pointer is valid
2416 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2420 page
= get_object_page(object
);
2422 if (!page
|| s
!= page
->slab
)
2423 /* No slab or wrong slab */
2426 if (!check_valid_pointer(s
, page
, object
))
2430 * We could also check if the object is on the slabs freelist.
2431 * But this would be too expensive and it seems that the main
2432 * purpose of kmem_ptr_valid() is to check if the object belongs
2433 * to a certain slab.
2437 EXPORT_SYMBOL(kmem_ptr_validate
);
2440 * Determine the size of a slab object
2442 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2446 EXPORT_SYMBOL(kmem_cache_size
);
2448 const char *kmem_cache_name(struct kmem_cache
*s
)
2452 EXPORT_SYMBOL(kmem_cache_name
);
2454 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2457 #ifdef CONFIG_SLUB_DEBUG
2458 void *addr
= page_address(page
);
2460 DECLARE_BITMAP(map
, page
->objects
);
2462 bitmap_zero(map
, page
->objects
);
2463 slab_err(s
, page
, "%s", text
);
2465 for_each_free_object(p
, s
, page
->freelist
)
2466 set_bit(slab_index(p
, s
, addr
), map
);
2468 for_each_object(p
, s
, addr
, page
->objects
) {
2470 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2471 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2473 print_tracking(s
, p
);
2481 * Attempt to free all partial slabs on a node.
2483 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2485 unsigned long flags
;
2486 struct page
*page
, *h
;
2488 spin_lock_irqsave(&n
->list_lock
, flags
);
2489 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2491 list_del(&page
->lru
);
2492 discard_slab(s
, page
);
2495 list_slab_objects(s
, page
,
2496 "Objects remaining on kmem_cache_close()");
2499 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2503 * Release all resources used by a slab cache.
2505 static inline int kmem_cache_close(struct kmem_cache
*s
)
2511 /* Attempt to free all objects */
2512 free_kmem_cache_cpus(s
);
2513 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2514 struct kmem_cache_node
*n
= get_node(s
, node
);
2517 if (n
->nr_partial
|| slabs_node(s
, node
))
2520 free_kmem_cache_nodes(s
);
2525 * Close a cache and release the kmem_cache structure
2526 * (must be used for caches created using kmem_cache_create)
2528 void kmem_cache_destroy(struct kmem_cache
*s
)
2530 down_write(&slub_lock
);
2534 up_write(&slub_lock
);
2535 if (kmem_cache_close(s
)) {
2536 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2537 "still has objects.\n", s
->name
, __func__
);
2540 sysfs_slab_remove(s
);
2542 up_write(&slub_lock
);
2544 EXPORT_SYMBOL(kmem_cache_destroy
);
2546 /********************************************************************
2548 *******************************************************************/
2550 struct kmem_cache kmalloc_caches
[SLUB_PAGE_SHIFT
] __cacheline_aligned
;
2551 EXPORT_SYMBOL(kmalloc_caches
);
2553 static int __init
setup_slub_min_order(char *str
)
2555 get_option(&str
, &slub_min_order
);
2560 __setup("slub_min_order=", setup_slub_min_order
);
2562 static int __init
setup_slub_max_order(char *str
)
2564 get_option(&str
, &slub_max_order
);
2565 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2570 __setup("slub_max_order=", setup_slub_max_order
);
2572 static int __init
setup_slub_min_objects(char *str
)
2574 get_option(&str
, &slub_min_objects
);
2579 __setup("slub_min_objects=", setup_slub_min_objects
);
2581 static int __init
setup_slub_nomerge(char *str
)
2587 __setup("slub_nomerge", setup_slub_nomerge
);
2589 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2590 const char *name
, int size
, gfp_t gfp_flags
)
2592 unsigned int flags
= 0;
2594 if (gfp_flags
& SLUB_DMA
)
2595 flags
= SLAB_CACHE_DMA
;
2598 * This function is called with IRQs disabled during early-boot on
2599 * single CPU so there's no need to take slub_lock here.
2601 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2605 list_add(&s
->list
, &slab_caches
);
2607 if (sysfs_slab_add(s
))
2612 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2615 #ifdef CONFIG_ZONE_DMA
2616 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2618 static void sysfs_add_func(struct work_struct
*w
)
2620 struct kmem_cache
*s
;
2622 down_write(&slub_lock
);
2623 list_for_each_entry(s
, &slab_caches
, list
) {
2624 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2625 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2629 up_write(&slub_lock
);
2632 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2634 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2636 struct kmem_cache
*s
;
2640 s
= kmalloc_caches_dma
[index
];
2644 /* Dynamically create dma cache */
2645 if (flags
& __GFP_WAIT
)
2646 down_write(&slub_lock
);
2648 if (!down_write_trylock(&slub_lock
))
2652 if (kmalloc_caches_dma
[index
])
2655 realsize
= kmalloc_caches
[index
].objsize
;
2656 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2657 (unsigned int)realsize
);
2658 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2660 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2661 realsize
, ARCH_KMALLOC_MINALIGN
,
2662 SLAB_CACHE_DMA
|SLAB_NOTRACK
|__SYSFS_ADD_DEFERRED
,
2669 list_add(&s
->list
, &slab_caches
);
2670 kmalloc_caches_dma
[index
] = s
;
2672 schedule_work(&sysfs_add_work
);
2675 up_write(&slub_lock
);
2677 return kmalloc_caches_dma
[index
];
2682 * Conversion table for small slabs sizes / 8 to the index in the
2683 * kmalloc array. This is necessary for slabs < 192 since we have non power
2684 * of two cache sizes there. The size of larger slabs can be determined using
2687 static s8 size_index
[24] = {
2714 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2720 return ZERO_SIZE_PTR
;
2722 index
= size_index
[(size
- 1) / 8];
2724 index
= fls(size
- 1);
2726 #ifdef CONFIG_ZONE_DMA
2727 if (unlikely((flags
& SLUB_DMA
)))
2728 return dma_kmalloc_cache(index
, flags
);
2731 return &kmalloc_caches
[index
];
2734 void *__kmalloc(size_t size
, gfp_t flags
)
2736 struct kmem_cache
*s
;
2739 if (unlikely(size
> SLUB_MAX_SIZE
))
2740 return kmalloc_large(size
, flags
);
2742 s
= get_slab(size
, flags
);
2744 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2747 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2749 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2753 EXPORT_SYMBOL(__kmalloc
);
2755 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2759 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2760 page
= alloc_pages_node(node
, flags
, get_order(size
));
2762 return page_address(page
);
2768 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2770 struct kmem_cache
*s
;
2773 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2774 ret
= kmalloc_large_node(size
, flags
, node
);
2776 trace_kmalloc_node(_RET_IP_
, ret
,
2777 size
, PAGE_SIZE
<< get_order(size
),
2783 s
= get_slab(size
, flags
);
2785 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2788 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2790 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2794 EXPORT_SYMBOL(__kmalloc_node
);
2797 size_t ksize(const void *object
)
2800 struct kmem_cache
*s
;
2802 if (unlikely(object
== ZERO_SIZE_PTR
))
2805 page
= virt_to_head_page(object
);
2807 if (unlikely(!PageSlab(page
))) {
2808 WARN_ON(!PageCompound(page
));
2809 return PAGE_SIZE
<< compound_order(page
);
2813 #ifdef CONFIG_SLUB_DEBUG
2815 * Debugging requires use of the padding between object
2816 * and whatever may come after it.
2818 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2823 * If we have the need to store the freelist pointer
2824 * back there or track user information then we can
2825 * only use the space before that information.
2827 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2830 * Else we can use all the padding etc for the allocation
2834 EXPORT_SYMBOL(ksize
);
2836 void kfree(const void *x
)
2839 void *object
= (void *)x
;
2841 trace_kfree(_RET_IP_
, x
);
2843 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2846 page
= virt_to_head_page(x
);
2847 if (unlikely(!PageSlab(page
))) {
2848 BUG_ON(!PageCompound(page
));
2852 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2854 EXPORT_SYMBOL(kfree
);
2857 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2858 * the remaining slabs by the number of items in use. The slabs with the
2859 * most items in use come first. New allocations will then fill those up
2860 * and thus they can be removed from the partial lists.
2862 * The slabs with the least items are placed last. This results in them
2863 * being allocated from last increasing the chance that the last objects
2864 * are freed in them.
2866 int kmem_cache_shrink(struct kmem_cache
*s
)
2870 struct kmem_cache_node
*n
;
2873 int objects
= oo_objects(s
->max
);
2874 struct list_head
*slabs_by_inuse
=
2875 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2876 unsigned long flags
;
2878 if (!slabs_by_inuse
)
2882 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2883 n
= get_node(s
, node
);
2888 for (i
= 0; i
< objects
; i
++)
2889 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2891 spin_lock_irqsave(&n
->list_lock
, flags
);
2894 * Build lists indexed by the items in use in each slab.
2896 * Note that concurrent frees may occur while we hold the
2897 * list_lock. page->inuse here is the upper limit.
2899 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2900 if (!page
->inuse
&& slab_trylock(page
)) {
2902 * Must hold slab lock here because slab_free
2903 * may have freed the last object and be
2904 * waiting to release the slab.
2906 list_del(&page
->lru
);
2909 discard_slab(s
, page
);
2911 list_move(&page
->lru
,
2912 slabs_by_inuse
+ page
->inuse
);
2917 * Rebuild the partial list with the slabs filled up most
2918 * first and the least used slabs at the end.
2920 for (i
= objects
- 1; i
>= 0; i
--)
2921 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2923 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2926 kfree(slabs_by_inuse
);
2929 EXPORT_SYMBOL(kmem_cache_shrink
);
2931 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2932 static int slab_mem_going_offline_callback(void *arg
)
2934 struct kmem_cache
*s
;
2936 down_read(&slub_lock
);
2937 list_for_each_entry(s
, &slab_caches
, list
)
2938 kmem_cache_shrink(s
);
2939 up_read(&slub_lock
);
2944 static void slab_mem_offline_callback(void *arg
)
2946 struct kmem_cache_node
*n
;
2947 struct kmem_cache
*s
;
2948 struct memory_notify
*marg
= arg
;
2951 offline_node
= marg
->status_change_nid
;
2954 * If the node still has available memory. we need kmem_cache_node
2957 if (offline_node
< 0)
2960 down_read(&slub_lock
);
2961 list_for_each_entry(s
, &slab_caches
, list
) {
2962 n
= get_node(s
, offline_node
);
2965 * if n->nr_slabs > 0, slabs still exist on the node
2966 * that is going down. We were unable to free them,
2967 * and offline_pages() function shoudn't call this
2968 * callback. So, we must fail.
2970 BUG_ON(slabs_node(s
, offline_node
));
2972 s
->node
[offline_node
] = NULL
;
2973 kmem_cache_free(kmalloc_caches
, n
);
2976 up_read(&slub_lock
);
2979 static int slab_mem_going_online_callback(void *arg
)
2981 struct kmem_cache_node
*n
;
2982 struct kmem_cache
*s
;
2983 struct memory_notify
*marg
= arg
;
2984 int nid
= marg
->status_change_nid
;
2988 * If the node's memory is already available, then kmem_cache_node is
2989 * already created. Nothing to do.
2995 * We are bringing a node online. No memory is available yet. We must
2996 * allocate a kmem_cache_node structure in order to bring the node
2999 down_read(&slub_lock
);
3000 list_for_each_entry(s
, &slab_caches
, list
) {
3002 * XXX: kmem_cache_alloc_node will fallback to other nodes
3003 * since memory is not yet available from the node that
3006 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
3011 init_kmem_cache_node(n
, s
);
3015 up_read(&slub_lock
);
3019 static int slab_memory_callback(struct notifier_block
*self
,
3020 unsigned long action
, void *arg
)
3025 case MEM_GOING_ONLINE
:
3026 ret
= slab_mem_going_online_callback(arg
);
3028 case MEM_GOING_OFFLINE
:
3029 ret
= slab_mem_going_offline_callback(arg
);
3032 case MEM_CANCEL_ONLINE
:
3033 slab_mem_offline_callback(arg
);
3036 case MEM_CANCEL_OFFLINE
:
3040 ret
= notifier_from_errno(ret
);
3046 #endif /* CONFIG_MEMORY_HOTPLUG */
3048 /********************************************************************
3049 * Basic setup of slabs
3050 *******************************************************************/
3052 void __init
kmem_cache_init(void)
3061 * Must first have the slab cache available for the allocations of the
3062 * struct kmem_cache_node's. There is special bootstrap code in
3063 * kmem_cache_open for slab_state == DOWN.
3065 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3066 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3067 kmalloc_caches
[0].refcount
= -1;
3070 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3073 /* Able to allocate the per node structures */
3074 slab_state
= PARTIAL
;
3076 /* Caches that are not of the two-to-the-power-of size */
3077 if (KMALLOC_MIN_SIZE
<= 64) {
3078 create_kmalloc_cache(&kmalloc_caches
[1],
3079 "kmalloc-96", 96, GFP_NOWAIT
);
3081 create_kmalloc_cache(&kmalloc_caches
[2],
3082 "kmalloc-192", 192, GFP_NOWAIT
);
3086 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3087 create_kmalloc_cache(&kmalloc_caches
[i
],
3088 "kmalloc", 1 << i
, GFP_NOWAIT
);
3094 * Patch up the size_index table if we have strange large alignment
3095 * requirements for the kmalloc array. This is only the case for
3096 * MIPS it seems. The standard arches will not generate any code here.
3098 * Largest permitted alignment is 256 bytes due to the way we
3099 * handle the index determination for the smaller caches.
3101 * Make sure that nothing crazy happens if someone starts tinkering
3102 * around with ARCH_KMALLOC_MINALIGN
3104 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3105 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3107 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3108 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3110 if (KMALLOC_MIN_SIZE
== 128) {
3112 * The 192 byte sized cache is not used if the alignment
3113 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3116 for (i
= 128 + 8; i
<= 192; i
+= 8)
3117 size_index
[(i
- 1) / 8] = 8;
3122 /* Provide the correct kmalloc names now that the caches are up */
3123 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3124 kmalloc_caches
[i
]. name
=
3125 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3128 register_cpu_notifier(&slab_notifier
);
3129 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3130 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3132 kmem_size
= sizeof(struct kmem_cache
);
3136 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3137 " CPUs=%d, Nodes=%d\n",
3138 caches
, cache_line_size(),
3139 slub_min_order
, slub_max_order
, slub_min_objects
,
3140 nr_cpu_ids
, nr_node_ids
);
3143 void __init
kmem_cache_init_late(void)
3146 * Interrupts are enabled now so all GFP allocations are safe.
3148 slab_gfp_mask
= __GFP_BITS_MASK
;
3152 * Find a mergeable slab cache
3154 static int slab_unmergeable(struct kmem_cache
*s
)
3156 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3163 * We may have set a slab to be unmergeable during bootstrap.
3165 if (s
->refcount
< 0)
3171 static struct kmem_cache
*find_mergeable(size_t size
,
3172 size_t align
, unsigned long flags
, const char *name
,
3173 void (*ctor
)(void *))
3175 struct kmem_cache
*s
;
3177 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3183 size
= ALIGN(size
, sizeof(void *));
3184 align
= calculate_alignment(flags
, align
, size
);
3185 size
= ALIGN(size
, align
);
3186 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3188 list_for_each_entry(s
, &slab_caches
, list
) {
3189 if (slab_unmergeable(s
))
3195 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3198 * Check if alignment is compatible.
3199 * Courtesy of Adrian Drzewiecki
3201 if ((s
->size
& ~(align
- 1)) != s
->size
)
3204 if (s
->size
- size
>= sizeof(void *))
3212 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3213 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3215 struct kmem_cache
*s
;
3217 down_write(&slub_lock
);
3218 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3224 * Adjust the object sizes so that we clear
3225 * the complete object on kzalloc.
3227 s
->objsize
= max(s
->objsize
, (int)size
);
3230 * And then we need to update the object size in the
3231 * per cpu structures
3233 for_each_online_cpu(cpu
)
3234 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3236 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3237 up_write(&slub_lock
);
3239 if (sysfs_slab_alias(s
, name
)) {
3240 down_write(&slub_lock
);
3242 up_write(&slub_lock
);
3248 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3250 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3251 size
, align
, flags
, ctor
)) {
3252 list_add(&s
->list
, &slab_caches
);
3253 up_write(&slub_lock
);
3254 if (sysfs_slab_add(s
)) {
3255 down_write(&slub_lock
);
3257 up_write(&slub_lock
);
3265 up_write(&slub_lock
);
3268 if (flags
& SLAB_PANIC
)
3269 panic("Cannot create slabcache %s\n", name
);
3274 EXPORT_SYMBOL(kmem_cache_create
);
3278 * Use the cpu notifier to insure that the cpu slabs are flushed when
3281 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3282 unsigned long action
, void *hcpu
)
3284 long cpu
= (long)hcpu
;
3285 struct kmem_cache
*s
;
3286 unsigned long flags
;
3289 case CPU_UP_PREPARE
:
3290 case CPU_UP_PREPARE_FROZEN
:
3291 init_alloc_cpu_cpu(cpu
);
3292 down_read(&slub_lock
);
3293 list_for_each_entry(s
, &slab_caches
, list
)
3294 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3296 up_read(&slub_lock
);
3299 case CPU_UP_CANCELED
:
3300 case CPU_UP_CANCELED_FROZEN
:
3302 case CPU_DEAD_FROZEN
:
3303 down_read(&slub_lock
);
3304 list_for_each_entry(s
, &slab_caches
, list
) {
3305 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3307 local_irq_save(flags
);
3308 __flush_cpu_slab(s
, cpu
);
3309 local_irq_restore(flags
);
3310 free_kmem_cache_cpu(c
, cpu
);
3311 s
->cpu_slab
[cpu
] = NULL
;
3313 up_read(&slub_lock
);
3321 static struct notifier_block __cpuinitdata slab_notifier
= {
3322 .notifier_call
= slab_cpuup_callback
3327 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3329 struct kmem_cache
*s
;
3332 if (unlikely(size
> SLUB_MAX_SIZE
))
3333 return kmalloc_large(size
, gfpflags
);
3335 s
= get_slab(size
, gfpflags
);
3337 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3340 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3342 /* Honor the call site pointer we recieved. */
3343 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3348 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3349 int node
, unsigned long caller
)
3351 struct kmem_cache
*s
;
3354 if (unlikely(size
> SLUB_MAX_SIZE
))
3355 return kmalloc_large_node(size
, gfpflags
, node
);
3357 s
= get_slab(size
, gfpflags
);
3359 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3362 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3364 /* Honor the call site pointer we recieved. */
3365 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3370 #ifdef CONFIG_SLUB_DEBUG
3371 static unsigned long count_partial(struct kmem_cache_node
*n
,
3372 int (*get_count
)(struct page
*))
3374 unsigned long flags
;
3375 unsigned long x
= 0;
3378 spin_lock_irqsave(&n
->list_lock
, flags
);
3379 list_for_each_entry(page
, &n
->partial
, lru
)
3380 x
+= get_count(page
);
3381 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3385 static int count_inuse(struct page
*page
)
3390 static int count_total(struct page
*page
)
3392 return page
->objects
;
3395 static int count_free(struct page
*page
)
3397 return page
->objects
- page
->inuse
;
3400 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3404 void *addr
= page_address(page
);
3406 if (!check_slab(s
, page
) ||
3407 !on_freelist(s
, page
, NULL
))
3410 /* Now we know that a valid freelist exists */
3411 bitmap_zero(map
, page
->objects
);
3413 for_each_free_object(p
, s
, page
->freelist
) {
3414 set_bit(slab_index(p
, s
, addr
), map
);
3415 if (!check_object(s
, page
, p
, 0))
3419 for_each_object(p
, s
, addr
, page
->objects
)
3420 if (!test_bit(slab_index(p
, s
, addr
), map
))
3421 if (!check_object(s
, page
, p
, 1))
3426 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3429 if (slab_trylock(page
)) {
3430 validate_slab(s
, page
, map
);
3433 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3436 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3437 if (!PageSlubDebug(page
))
3438 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3439 "on slab 0x%p\n", s
->name
, page
);
3441 if (PageSlubDebug(page
))
3442 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3443 "slab 0x%p\n", s
->name
, page
);
3447 static int validate_slab_node(struct kmem_cache
*s
,
3448 struct kmem_cache_node
*n
, unsigned long *map
)
3450 unsigned long count
= 0;
3452 unsigned long flags
;
3454 spin_lock_irqsave(&n
->list_lock
, flags
);
3456 list_for_each_entry(page
, &n
->partial
, lru
) {
3457 validate_slab_slab(s
, page
, map
);
3460 if (count
!= n
->nr_partial
)
3461 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3462 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3464 if (!(s
->flags
& SLAB_STORE_USER
))
3467 list_for_each_entry(page
, &n
->full
, lru
) {
3468 validate_slab_slab(s
, page
, map
);
3471 if (count
!= atomic_long_read(&n
->nr_slabs
))
3472 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3473 "counter=%ld\n", s
->name
, count
,
3474 atomic_long_read(&n
->nr_slabs
));
3477 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3481 static long validate_slab_cache(struct kmem_cache
*s
)
3484 unsigned long count
= 0;
3485 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3486 sizeof(unsigned long), GFP_KERNEL
);
3492 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3493 struct kmem_cache_node
*n
= get_node(s
, node
);
3495 count
+= validate_slab_node(s
, n
, map
);
3501 #ifdef SLUB_RESILIENCY_TEST
3502 static void resiliency_test(void)
3506 printk(KERN_ERR
"SLUB resiliency testing\n");
3507 printk(KERN_ERR
"-----------------------\n");
3508 printk(KERN_ERR
"A. Corruption after allocation\n");
3510 p
= kzalloc(16, GFP_KERNEL
);
3512 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3513 " 0x12->0x%p\n\n", p
+ 16);
3515 validate_slab_cache(kmalloc_caches
+ 4);
3517 /* Hmmm... The next two are dangerous */
3518 p
= kzalloc(32, GFP_KERNEL
);
3519 p
[32 + sizeof(void *)] = 0x34;
3520 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3521 " 0x34 -> -0x%p\n", p
);
3523 "If allocated object is overwritten then not detectable\n\n");
3525 validate_slab_cache(kmalloc_caches
+ 5);
3526 p
= kzalloc(64, GFP_KERNEL
);
3527 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3529 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3532 "If allocated object is overwritten then not detectable\n\n");
3533 validate_slab_cache(kmalloc_caches
+ 6);
3535 printk(KERN_ERR
"\nB. Corruption after free\n");
3536 p
= kzalloc(128, GFP_KERNEL
);
3539 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3540 validate_slab_cache(kmalloc_caches
+ 7);
3542 p
= kzalloc(256, GFP_KERNEL
);
3545 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3547 validate_slab_cache(kmalloc_caches
+ 8);
3549 p
= kzalloc(512, GFP_KERNEL
);
3552 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3553 validate_slab_cache(kmalloc_caches
+ 9);
3556 static void resiliency_test(void) {};
3560 * Generate lists of code addresses where slabcache objects are allocated
3565 unsigned long count
;
3572 DECLARE_BITMAP(cpus
, NR_CPUS
);
3578 unsigned long count
;
3579 struct location
*loc
;
3582 static void free_loc_track(struct loc_track
*t
)
3585 free_pages((unsigned long)t
->loc
,
3586 get_order(sizeof(struct location
) * t
->max
));
3589 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3594 order
= get_order(sizeof(struct location
) * max
);
3596 l
= (void *)__get_free_pages(flags
, order
);
3601 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3609 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3610 const struct track
*track
)
3612 long start
, end
, pos
;
3614 unsigned long caddr
;
3615 unsigned long age
= jiffies
- track
->when
;
3621 pos
= start
+ (end
- start
+ 1) / 2;
3624 * There is nothing at "end". If we end up there
3625 * we need to add something to before end.
3630 caddr
= t
->loc
[pos
].addr
;
3631 if (track
->addr
== caddr
) {
3637 if (age
< l
->min_time
)
3639 if (age
> l
->max_time
)
3642 if (track
->pid
< l
->min_pid
)
3643 l
->min_pid
= track
->pid
;
3644 if (track
->pid
> l
->max_pid
)
3645 l
->max_pid
= track
->pid
;
3647 cpumask_set_cpu(track
->cpu
,
3648 to_cpumask(l
->cpus
));
3650 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3654 if (track
->addr
< caddr
)
3661 * Not found. Insert new tracking element.
3663 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3669 (t
->count
- pos
) * sizeof(struct location
));
3672 l
->addr
= track
->addr
;
3676 l
->min_pid
= track
->pid
;
3677 l
->max_pid
= track
->pid
;
3678 cpumask_clear(to_cpumask(l
->cpus
));
3679 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3680 nodes_clear(l
->nodes
);
3681 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3685 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3686 struct page
*page
, enum track_item alloc
)
3688 void *addr
= page_address(page
);
3689 DECLARE_BITMAP(map
, page
->objects
);
3692 bitmap_zero(map
, page
->objects
);
3693 for_each_free_object(p
, s
, page
->freelist
)
3694 set_bit(slab_index(p
, s
, addr
), map
);
3696 for_each_object(p
, s
, addr
, page
->objects
)
3697 if (!test_bit(slab_index(p
, s
, addr
), map
))
3698 add_location(t
, s
, get_track(s
, p
, alloc
));
3701 static int list_locations(struct kmem_cache
*s
, char *buf
,
3702 enum track_item alloc
)
3706 struct loc_track t
= { 0, 0, NULL
};
3709 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3711 return sprintf(buf
, "Out of memory\n");
3713 /* Push back cpu slabs */
3716 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3717 struct kmem_cache_node
*n
= get_node(s
, node
);
3718 unsigned long flags
;
3721 if (!atomic_long_read(&n
->nr_slabs
))
3724 spin_lock_irqsave(&n
->list_lock
, flags
);
3725 list_for_each_entry(page
, &n
->partial
, lru
)
3726 process_slab(&t
, s
, page
, alloc
);
3727 list_for_each_entry(page
, &n
->full
, lru
)
3728 process_slab(&t
, s
, page
, alloc
);
3729 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3732 for (i
= 0; i
< t
.count
; i
++) {
3733 struct location
*l
= &t
.loc
[i
];
3735 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3737 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3740 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3742 len
+= sprintf(buf
+ len
, "<not-available>");
3744 if (l
->sum_time
!= l
->min_time
) {
3745 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3747 (long)div_u64(l
->sum_time
, l
->count
),
3750 len
+= sprintf(buf
+ len
, " age=%ld",
3753 if (l
->min_pid
!= l
->max_pid
)
3754 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3755 l
->min_pid
, l
->max_pid
);
3757 len
+= sprintf(buf
+ len
, " pid=%ld",
3760 if (num_online_cpus() > 1 &&
3761 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3762 len
< PAGE_SIZE
- 60) {
3763 len
+= sprintf(buf
+ len
, " cpus=");
3764 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3765 to_cpumask(l
->cpus
));
3768 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3769 len
< PAGE_SIZE
- 60) {
3770 len
+= sprintf(buf
+ len
, " nodes=");
3771 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3775 len
+= sprintf(buf
+ len
, "\n");
3780 len
+= sprintf(buf
, "No data\n");
3784 enum slab_stat_type
{
3785 SL_ALL
, /* All slabs */
3786 SL_PARTIAL
, /* Only partially allocated slabs */
3787 SL_CPU
, /* Only slabs used for cpu caches */
3788 SL_OBJECTS
, /* Determine allocated objects not slabs */
3789 SL_TOTAL
/* Determine object capacity not slabs */
3792 #define SO_ALL (1 << SL_ALL)
3793 #define SO_PARTIAL (1 << SL_PARTIAL)
3794 #define SO_CPU (1 << SL_CPU)
3795 #define SO_OBJECTS (1 << SL_OBJECTS)
3796 #define SO_TOTAL (1 << SL_TOTAL)
3798 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3799 char *buf
, unsigned long flags
)
3801 unsigned long total
= 0;
3804 unsigned long *nodes
;
3805 unsigned long *per_cpu
;
3807 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3810 per_cpu
= nodes
+ nr_node_ids
;
3812 if (flags
& SO_CPU
) {
3815 for_each_possible_cpu(cpu
) {
3816 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3818 if (!c
|| c
->node
< 0)
3822 if (flags
& SO_TOTAL
)
3823 x
= c
->page
->objects
;
3824 else if (flags
& SO_OBJECTS
)
3830 nodes
[c
->node
] += x
;
3836 if (flags
& SO_ALL
) {
3837 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3838 struct kmem_cache_node
*n
= get_node(s
, node
);
3840 if (flags
& SO_TOTAL
)
3841 x
= atomic_long_read(&n
->total_objects
);
3842 else if (flags
& SO_OBJECTS
)
3843 x
= atomic_long_read(&n
->total_objects
) -
3844 count_partial(n
, count_free
);
3847 x
= atomic_long_read(&n
->nr_slabs
);
3852 } else if (flags
& SO_PARTIAL
) {
3853 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3854 struct kmem_cache_node
*n
= get_node(s
, node
);
3856 if (flags
& SO_TOTAL
)
3857 x
= count_partial(n
, count_total
);
3858 else if (flags
& SO_OBJECTS
)
3859 x
= count_partial(n
, count_inuse
);
3866 x
= sprintf(buf
, "%lu", total
);
3868 for_each_node_state(node
, N_NORMAL_MEMORY
)
3870 x
+= sprintf(buf
+ x
, " N%d=%lu",
3874 return x
+ sprintf(buf
+ x
, "\n");
3877 static int any_slab_objects(struct kmem_cache
*s
)
3881 for_each_online_node(node
) {
3882 struct kmem_cache_node
*n
= get_node(s
, node
);
3887 if (atomic_long_read(&n
->total_objects
))
3893 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3894 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3896 struct slab_attribute
{
3897 struct attribute attr
;
3898 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3899 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3902 #define SLAB_ATTR_RO(_name) \
3903 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3905 #define SLAB_ATTR(_name) \
3906 static struct slab_attribute _name##_attr = \
3907 __ATTR(_name, 0644, _name##_show, _name##_store)
3909 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3911 return sprintf(buf
, "%d\n", s
->size
);
3913 SLAB_ATTR_RO(slab_size
);
3915 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3917 return sprintf(buf
, "%d\n", s
->align
);
3919 SLAB_ATTR_RO(align
);
3921 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3923 return sprintf(buf
, "%d\n", s
->objsize
);
3925 SLAB_ATTR_RO(object_size
);
3927 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3929 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3931 SLAB_ATTR_RO(objs_per_slab
);
3933 static ssize_t
order_store(struct kmem_cache
*s
,
3934 const char *buf
, size_t length
)
3936 unsigned long order
;
3939 err
= strict_strtoul(buf
, 10, &order
);
3943 if (order
> slub_max_order
|| order
< slub_min_order
)
3946 calculate_sizes(s
, order
);
3950 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3952 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3956 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3958 return sprintf(buf
, "%lu\n", s
->min_partial
);
3961 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3967 err
= strict_strtoul(buf
, 10, &min
);
3971 set_min_partial(s
, min
);
3974 SLAB_ATTR(min_partial
);
3976 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3979 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3981 return n
+ sprintf(buf
+ n
, "\n");
3987 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3989 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3991 SLAB_ATTR_RO(aliases
);
3993 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3995 return show_slab_objects(s
, buf
, SO_ALL
);
3997 SLAB_ATTR_RO(slabs
);
3999 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4001 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4003 SLAB_ATTR_RO(partial
);
4005 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4007 return show_slab_objects(s
, buf
, SO_CPU
);
4009 SLAB_ATTR_RO(cpu_slabs
);
4011 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4013 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4015 SLAB_ATTR_RO(objects
);
4017 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4019 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4021 SLAB_ATTR_RO(objects_partial
);
4023 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4025 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4027 SLAB_ATTR_RO(total_objects
);
4029 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4031 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4034 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4035 const char *buf
, size_t length
)
4037 s
->flags
&= ~SLAB_DEBUG_FREE
;
4039 s
->flags
|= SLAB_DEBUG_FREE
;
4042 SLAB_ATTR(sanity_checks
);
4044 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4046 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4049 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4052 s
->flags
&= ~SLAB_TRACE
;
4054 s
->flags
|= SLAB_TRACE
;
4059 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4061 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4064 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4065 const char *buf
, size_t length
)
4067 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4069 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4072 SLAB_ATTR(reclaim_account
);
4074 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4076 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4078 SLAB_ATTR_RO(hwcache_align
);
4080 #ifdef CONFIG_ZONE_DMA
4081 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4083 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4085 SLAB_ATTR_RO(cache_dma
);
4088 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4090 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4092 SLAB_ATTR_RO(destroy_by_rcu
);
4094 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4096 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4099 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4100 const char *buf
, size_t length
)
4102 if (any_slab_objects(s
))
4105 s
->flags
&= ~SLAB_RED_ZONE
;
4107 s
->flags
|= SLAB_RED_ZONE
;
4108 calculate_sizes(s
, -1);
4111 SLAB_ATTR(red_zone
);
4113 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4115 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4118 static ssize_t
poison_store(struct kmem_cache
*s
,
4119 const char *buf
, size_t length
)
4121 if (any_slab_objects(s
))
4124 s
->flags
&= ~SLAB_POISON
;
4126 s
->flags
|= SLAB_POISON
;
4127 calculate_sizes(s
, -1);
4132 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4134 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4137 static ssize_t
store_user_store(struct kmem_cache
*s
,
4138 const char *buf
, size_t length
)
4140 if (any_slab_objects(s
))
4143 s
->flags
&= ~SLAB_STORE_USER
;
4145 s
->flags
|= SLAB_STORE_USER
;
4146 calculate_sizes(s
, -1);
4149 SLAB_ATTR(store_user
);
4151 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4156 static ssize_t
validate_store(struct kmem_cache
*s
,
4157 const char *buf
, size_t length
)
4161 if (buf
[0] == '1') {
4162 ret
= validate_slab_cache(s
);
4168 SLAB_ATTR(validate
);
4170 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4175 static ssize_t
shrink_store(struct kmem_cache
*s
,
4176 const char *buf
, size_t length
)
4178 if (buf
[0] == '1') {
4179 int rc
= kmem_cache_shrink(s
);
4189 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4191 if (!(s
->flags
& SLAB_STORE_USER
))
4193 return list_locations(s
, buf
, TRACK_ALLOC
);
4195 SLAB_ATTR_RO(alloc_calls
);
4197 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4199 if (!(s
->flags
& SLAB_STORE_USER
))
4201 return list_locations(s
, buf
, TRACK_FREE
);
4203 SLAB_ATTR_RO(free_calls
);
4206 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4208 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4211 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4212 const char *buf
, size_t length
)
4214 unsigned long ratio
;
4217 err
= strict_strtoul(buf
, 10, &ratio
);
4222 s
->remote_node_defrag_ratio
= ratio
* 10;
4226 SLAB_ATTR(remote_node_defrag_ratio
);
4229 #ifdef CONFIG_SLUB_STATS
4230 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4232 unsigned long sum
= 0;
4235 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4240 for_each_online_cpu(cpu
) {
4241 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4247 len
= sprintf(buf
, "%lu", sum
);
4250 for_each_online_cpu(cpu
) {
4251 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4252 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4256 return len
+ sprintf(buf
+ len
, "\n");
4259 #define STAT_ATTR(si, text) \
4260 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4262 return show_stat(s, buf, si); \
4264 SLAB_ATTR_RO(text); \
4266 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4267 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4268 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4269 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4270 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4271 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4272 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4273 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4274 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4275 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4276 STAT_ATTR(FREE_SLAB
, free_slab
);
4277 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4278 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4279 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4280 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4281 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4282 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4283 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4286 static struct attribute
*slab_attrs
[] = {
4287 &slab_size_attr
.attr
,
4288 &object_size_attr
.attr
,
4289 &objs_per_slab_attr
.attr
,
4291 &min_partial_attr
.attr
,
4293 &objects_partial_attr
.attr
,
4294 &total_objects_attr
.attr
,
4297 &cpu_slabs_attr
.attr
,
4301 &sanity_checks_attr
.attr
,
4303 &hwcache_align_attr
.attr
,
4304 &reclaim_account_attr
.attr
,
4305 &destroy_by_rcu_attr
.attr
,
4306 &red_zone_attr
.attr
,
4308 &store_user_attr
.attr
,
4309 &validate_attr
.attr
,
4311 &alloc_calls_attr
.attr
,
4312 &free_calls_attr
.attr
,
4313 #ifdef CONFIG_ZONE_DMA
4314 &cache_dma_attr
.attr
,
4317 &remote_node_defrag_ratio_attr
.attr
,
4319 #ifdef CONFIG_SLUB_STATS
4320 &alloc_fastpath_attr
.attr
,
4321 &alloc_slowpath_attr
.attr
,
4322 &free_fastpath_attr
.attr
,
4323 &free_slowpath_attr
.attr
,
4324 &free_frozen_attr
.attr
,
4325 &free_add_partial_attr
.attr
,
4326 &free_remove_partial_attr
.attr
,
4327 &alloc_from_partial_attr
.attr
,
4328 &alloc_slab_attr
.attr
,
4329 &alloc_refill_attr
.attr
,
4330 &free_slab_attr
.attr
,
4331 &cpuslab_flush_attr
.attr
,
4332 &deactivate_full_attr
.attr
,
4333 &deactivate_empty_attr
.attr
,
4334 &deactivate_to_head_attr
.attr
,
4335 &deactivate_to_tail_attr
.attr
,
4336 &deactivate_remote_frees_attr
.attr
,
4337 &order_fallback_attr
.attr
,
4342 static struct attribute_group slab_attr_group
= {
4343 .attrs
= slab_attrs
,
4346 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4347 struct attribute
*attr
,
4350 struct slab_attribute
*attribute
;
4351 struct kmem_cache
*s
;
4354 attribute
= to_slab_attr(attr
);
4357 if (!attribute
->show
)
4360 err
= attribute
->show(s
, buf
);
4365 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4366 struct attribute
*attr
,
4367 const char *buf
, size_t len
)
4369 struct slab_attribute
*attribute
;
4370 struct kmem_cache
*s
;
4373 attribute
= to_slab_attr(attr
);
4376 if (!attribute
->store
)
4379 err
= attribute
->store(s
, buf
, len
);
4384 static void kmem_cache_release(struct kobject
*kobj
)
4386 struct kmem_cache
*s
= to_slab(kobj
);
4391 static struct sysfs_ops slab_sysfs_ops
= {
4392 .show
= slab_attr_show
,
4393 .store
= slab_attr_store
,
4396 static struct kobj_type slab_ktype
= {
4397 .sysfs_ops
= &slab_sysfs_ops
,
4398 .release
= kmem_cache_release
4401 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4403 struct kobj_type
*ktype
= get_ktype(kobj
);
4405 if (ktype
== &slab_ktype
)
4410 static struct kset_uevent_ops slab_uevent_ops
= {
4411 .filter
= uevent_filter
,
4414 static struct kset
*slab_kset
;
4416 #define ID_STR_LENGTH 64
4418 /* Create a unique string id for a slab cache:
4420 * Format :[flags-]size
4422 static char *create_unique_id(struct kmem_cache
*s
)
4424 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4431 * First flags affecting slabcache operations. We will only
4432 * get here for aliasable slabs so we do not need to support
4433 * too many flags. The flags here must cover all flags that
4434 * are matched during merging to guarantee that the id is
4437 if (s
->flags
& SLAB_CACHE_DMA
)
4439 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4441 if (s
->flags
& SLAB_DEBUG_FREE
)
4443 if (!(s
->flags
& SLAB_NOTRACK
))
4447 p
+= sprintf(p
, "%07d", s
->size
);
4448 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4452 static int sysfs_slab_add(struct kmem_cache
*s
)
4458 if (slab_state
< SYSFS
)
4459 /* Defer until later */
4462 unmergeable
= slab_unmergeable(s
);
4465 * Slabcache can never be merged so we can use the name proper.
4466 * This is typically the case for debug situations. In that
4467 * case we can catch duplicate names easily.
4469 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4473 * Create a unique name for the slab as a target
4476 name
= create_unique_id(s
);
4479 s
->kobj
.kset
= slab_kset
;
4480 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4482 kobject_put(&s
->kobj
);
4486 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4489 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4491 /* Setup first alias */
4492 sysfs_slab_alias(s
, s
->name
);
4498 static void sysfs_slab_remove(struct kmem_cache
*s
)
4500 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4501 kobject_del(&s
->kobj
);
4502 kobject_put(&s
->kobj
);
4506 * Need to buffer aliases during bootup until sysfs becomes
4507 * available lest we lose that information.
4509 struct saved_alias
{
4510 struct kmem_cache
*s
;
4512 struct saved_alias
*next
;
4515 static struct saved_alias
*alias_list
;
4517 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4519 struct saved_alias
*al
;
4521 if (slab_state
== SYSFS
) {
4523 * If we have a leftover link then remove it.
4525 sysfs_remove_link(&slab_kset
->kobj
, name
);
4526 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4529 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4535 al
->next
= alias_list
;
4540 static int __init
slab_sysfs_init(void)
4542 struct kmem_cache
*s
;
4545 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4547 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4553 list_for_each_entry(s
, &slab_caches
, list
) {
4554 err
= sysfs_slab_add(s
);
4556 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4557 " to sysfs\n", s
->name
);
4560 while (alias_list
) {
4561 struct saved_alias
*al
= alias_list
;
4563 alias_list
= alias_list
->next
;
4564 err
= sysfs_slab_alias(al
->s
, al
->name
);
4566 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4567 " %s to sysfs\n", s
->name
);
4575 __initcall(slab_sysfs_init
);
4579 * The /proc/slabinfo ABI
4581 #ifdef CONFIG_SLABINFO
4582 static void print_slabinfo_header(struct seq_file
*m
)
4584 seq_puts(m
, "slabinfo - version: 2.1\n");
4585 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4586 "<objperslab> <pagesperslab>");
4587 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4588 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4592 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4596 down_read(&slub_lock
);
4598 print_slabinfo_header(m
);
4600 return seq_list_start(&slab_caches
, *pos
);
4603 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4605 return seq_list_next(p
, &slab_caches
, pos
);
4608 static void s_stop(struct seq_file
*m
, void *p
)
4610 up_read(&slub_lock
);
4613 static int s_show(struct seq_file
*m
, void *p
)
4615 unsigned long nr_partials
= 0;
4616 unsigned long nr_slabs
= 0;
4617 unsigned long nr_inuse
= 0;
4618 unsigned long nr_objs
= 0;
4619 unsigned long nr_free
= 0;
4620 struct kmem_cache
*s
;
4623 s
= list_entry(p
, struct kmem_cache
, list
);
4625 for_each_online_node(node
) {
4626 struct kmem_cache_node
*n
= get_node(s
, node
);
4631 nr_partials
+= n
->nr_partial
;
4632 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4633 nr_objs
+= atomic_long_read(&n
->total_objects
);
4634 nr_free
+= count_partial(n
, count_free
);
4637 nr_inuse
= nr_objs
- nr_free
;
4639 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4640 nr_objs
, s
->size
, oo_objects(s
->oo
),
4641 (1 << oo_order(s
->oo
)));
4642 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4643 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4649 static const struct seq_operations slabinfo_op
= {
4656 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4658 return seq_open(file
, &slabinfo_op
);
4661 static const struct file_operations proc_slabinfo_operations
= {
4662 .open
= slabinfo_open
,
4664 .llseek
= seq_lseek
,
4665 .release
= seq_release
,
4668 static int __init
slab_proc_init(void)
4670 proc_create("slabinfo",S_IWUSR
|S_IRUGO
,NULL
,&proc_slabinfo_operations
);
4673 module_init(slab_proc_init
);
4674 #endif /* CONFIG_SLABINFO */