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/mempolicy.h>
25 #include <linux/ctype.h>
26 #include <linux/debugobjects.h>
27 #include <linux/kallsyms.h>
28 #include <linux/memory.h>
29 #include <linux/math64.h>
30 #include <linux/fault-inject.h>
37 * The slab_lock protects operations on the object of a particular
38 * slab and its metadata in the page struct. If the slab lock
39 * has been taken then no allocations nor frees can be performed
40 * on the objects in the slab nor can the slab be added or removed
41 * from the partial or full lists since this would mean modifying
42 * the page_struct of the slab.
44 * The list_lock protects the partial and full list on each node and
45 * the partial slab counter. If taken then no new slabs may be added or
46 * removed from the lists nor make the number of partial slabs be modified.
47 * (Note that the total number of slabs is an atomic value that may be
48 * modified without taking the list lock).
50 * The list_lock is a centralized lock and thus we avoid taking it as
51 * much as possible. As long as SLUB does not have to handle partial
52 * slabs, operations can continue without any centralized lock. F.e.
53 * allocating a long series of objects that fill up slabs does not require
56 * The lock order is sometimes inverted when we are trying to get a slab
57 * off a list. We take the list_lock and then look for a page on the list
58 * to use. While we do that objects in the slabs may be freed. We can
59 * only operate on the slab if we have also taken the slab_lock. So we use
60 * a slab_trylock() on the slab. If trylock was successful then no frees
61 * can occur anymore and we can use the slab for allocations etc. If the
62 * slab_trylock() does not succeed then frees are in progress in the slab and
63 * we must stay away from it for a while since we may cause a bouncing
64 * cacheline if we try to acquire the lock. So go onto the next slab.
65 * If all pages are busy then we may allocate a new slab instead of reusing
66 * a partial slab. A new slab has noone operating on it and thus there is
67 * no danger of cacheline contention.
69 * Interrupts are disabled during allocation and deallocation in order to
70 * make the slab allocator safe to use in the context of an irq. In addition
71 * interrupts are disabled to ensure that the processor does not change
72 * while handling per_cpu slabs, due to kernel preemption.
74 * SLUB assigns one slab for allocation to each processor.
75 * Allocations only occur from these slabs called cpu slabs.
77 * Slabs with free elements are kept on a partial list and during regular
78 * operations no list for full slabs is used. If an object in a full slab is
79 * freed then the slab will show up again on the partial lists.
80 * We track full slabs for debugging purposes though because otherwise we
81 * cannot scan all objects.
83 * Slabs are freed when they become empty. Teardown and setup is
84 * minimal so we rely on the page allocators per cpu caches for
85 * fast frees and allocs.
87 * Overloading of page flags that are otherwise used for LRU management.
89 * PageActive The slab is frozen and exempt from list processing.
90 * This means that the slab is dedicated to a purpose
91 * such as satisfying allocations for a specific
92 * processor. Objects may be freed in the slab while
93 * it is frozen but slab_free will then skip the usual
94 * list operations. It is up to the processor holding
95 * the slab to integrate the slab into the slab lists
96 * when the slab is no longer needed.
98 * One use of this flag is to mark slabs that are
99 * used for allocations. Then such a slab becomes a cpu
100 * slab. The cpu slab may be equipped with an additional
101 * freelist that allows lockless access to
102 * free objects in addition to the regular freelist
103 * that requires the slab lock.
105 * PageError Slab requires special handling due to debug
106 * options set. This moves slab handling out of
107 * the fast path and disables lockless freelists.
110 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
111 SLAB_TRACE | SLAB_DEBUG_FREE)
113 static inline int kmem_cache_debug(struct kmem_cache
*s
)
115 #ifdef CONFIG_SLUB_DEBUG
116 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
123 * Issues still to be resolved:
125 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
127 * - Variable sizing of the per node arrays
130 /* Enable to test recovery from slab corruption on boot */
131 #undef SLUB_RESILIENCY_TEST
134 * Mininum number of partial slabs. These will be left on the partial
135 * lists even if they are empty. kmem_cache_shrink may reclaim them.
137 #define MIN_PARTIAL 5
140 * Maximum number of desirable partial slabs.
141 * The existence of more partial slabs makes kmem_cache_shrink
142 * sort the partial list by the number of objects in the.
144 #define MAX_PARTIAL 10
146 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
147 SLAB_POISON | SLAB_STORE_USER)
150 * Debugging flags that require metadata to be stored in the slab. These get
151 * disabled when slub_debug=O is used and a cache's min order increases with
154 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
157 * Set of flags that will prevent slab merging
159 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
160 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
163 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
164 SLAB_CACHE_DMA | SLAB_NOTRACK)
167 #define OO_MASK ((1 << OO_SHIFT) - 1)
168 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
170 /* Internal SLUB flags */
171 #define __OBJECT_POISON 0x80000000UL /* Poison object */
172 #define __SYSFS_ADD_DEFERRED 0x40000000UL /* Not yet visible via sysfs */
174 static int kmem_size
= sizeof(struct kmem_cache
);
177 static struct notifier_block slab_notifier
;
181 DOWN
, /* No slab functionality available */
182 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
183 UP
, /* Everything works but does not show up in sysfs */
187 /* A list of all slab caches on the system */
188 static DECLARE_RWSEM(slub_lock
);
189 static LIST_HEAD(slab_caches
);
192 * Tracking user of a slab.
195 unsigned long addr
; /* Called from address */
196 int cpu
; /* Was running on cpu */
197 int pid
; /* Pid context */
198 unsigned long when
; /* When did the operation occur */
201 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
203 #ifdef CONFIG_SLUB_DEBUG
204 static int sysfs_slab_add(struct kmem_cache
*);
205 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
206 static void sysfs_slab_remove(struct kmem_cache
*);
209 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
210 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
212 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
219 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
221 #ifdef CONFIG_SLUB_STATS
222 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
226 /********************************************************************
227 * Core slab cache functions
228 *******************************************************************/
230 int slab_is_available(void)
232 return slab_state
>= UP
;
235 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
238 return s
->node
[node
];
240 return &s
->local_node
;
244 /* Verify that a pointer has an address that is valid within a slab page */
245 static inline int check_valid_pointer(struct kmem_cache
*s
,
246 struct page
*page
, const void *object
)
253 base
= page_address(page
);
254 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
255 (object
- base
) % s
->size
) {
262 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
264 return *(void **)(object
+ s
->offset
);
267 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
269 *(void **)(object
+ s
->offset
) = fp
;
272 /* Loop over all objects in a slab */
273 #define for_each_object(__p, __s, __addr, __objects) \
274 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
278 #define for_each_free_object(__p, __s, __free) \
279 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
281 /* Determine object index from a given position */
282 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
284 return (p
- addr
) / s
->size
;
287 static inline struct kmem_cache_order_objects
oo_make(int order
,
290 struct kmem_cache_order_objects x
= {
291 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
297 static inline int oo_order(struct kmem_cache_order_objects x
)
299 return x
.x
>> OO_SHIFT
;
302 static inline int oo_objects(struct kmem_cache_order_objects x
)
304 return x
.x
& OO_MASK
;
307 #ifdef CONFIG_SLUB_DEBUG
311 #ifdef CONFIG_SLUB_DEBUG_ON
312 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
314 static int slub_debug
;
317 static char *slub_debug_slabs
;
318 static int disable_higher_order_debug
;
323 static void print_section(char *text
, u8
*addr
, unsigned int length
)
331 for (i
= 0; i
< length
; i
++) {
333 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
336 printk(KERN_CONT
" %02x", addr
[i
]);
338 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
340 printk(KERN_CONT
" %s\n", ascii
);
347 printk(KERN_CONT
" ");
351 printk(KERN_CONT
" %s\n", ascii
);
355 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
356 enum track_item alloc
)
361 p
= object
+ s
->offset
+ sizeof(void *);
363 p
= object
+ s
->inuse
;
368 static void set_track(struct kmem_cache
*s
, void *object
,
369 enum track_item alloc
, unsigned long addr
)
371 struct track
*p
= get_track(s
, object
, alloc
);
375 p
->cpu
= smp_processor_id();
376 p
->pid
= current
->pid
;
379 memset(p
, 0, sizeof(struct track
));
382 static void init_tracking(struct kmem_cache
*s
, void *object
)
384 if (!(s
->flags
& SLAB_STORE_USER
))
387 set_track(s
, object
, TRACK_FREE
, 0UL);
388 set_track(s
, object
, TRACK_ALLOC
, 0UL);
391 static void print_track(const char *s
, struct track
*t
)
396 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
397 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
400 static void print_tracking(struct kmem_cache
*s
, void *object
)
402 if (!(s
->flags
& SLAB_STORE_USER
))
405 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
406 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
409 static void print_page_info(struct page
*page
)
411 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
412 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
416 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
422 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
424 printk(KERN_ERR
"========================================"
425 "=====================================\n");
426 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
427 printk(KERN_ERR
"----------------------------------------"
428 "-------------------------------------\n\n");
431 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
437 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
439 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
442 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
444 unsigned int off
; /* Offset of last byte */
445 u8
*addr
= page_address(page
);
447 print_tracking(s
, p
);
449 print_page_info(page
);
451 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
452 p
, p
- addr
, get_freepointer(s
, p
));
455 print_section("Bytes b4", p
- 16, 16);
457 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
459 if (s
->flags
& SLAB_RED_ZONE
)
460 print_section("Redzone", p
+ s
->objsize
,
461 s
->inuse
- s
->objsize
);
464 off
= s
->offset
+ sizeof(void *);
468 if (s
->flags
& SLAB_STORE_USER
)
469 off
+= 2 * sizeof(struct track
);
472 /* Beginning of the filler is the free pointer */
473 print_section("Padding", p
+ off
, s
->size
- off
);
478 static void object_err(struct kmem_cache
*s
, struct page
*page
,
479 u8
*object
, char *reason
)
481 slab_bug(s
, "%s", reason
);
482 print_trailer(s
, page
, object
);
485 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
491 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
493 slab_bug(s
, "%s", buf
);
494 print_page_info(page
);
498 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
502 if (s
->flags
& __OBJECT_POISON
) {
503 memset(p
, POISON_FREE
, s
->objsize
- 1);
504 p
[s
->objsize
- 1] = POISON_END
;
507 if (s
->flags
& SLAB_RED_ZONE
)
508 memset(p
+ s
->objsize
,
509 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
510 s
->inuse
- s
->objsize
);
513 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
516 if (*start
!= (u8
)value
)
524 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
525 void *from
, void *to
)
527 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
528 memset(from
, data
, to
- from
);
531 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
532 u8
*object
, char *what
,
533 u8
*start
, unsigned int value
, unsigned int bytes
)
538 fault
= check_bytes(start
, value
, bytes
);
543 while (end
> fault
&& end
[-1] == value
)
546 slab_bug(s
, "%s overwritten", what
);
547 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
548 fault
, end
- 1, fault
[0], value
);
549 print_trailer(s
, page
, object
);
551 restore_bytes(s
, what
, value
, fault
, end
);
559 * Bytes of the object to be managed.
560 * If the freepointer may overlay the object then the free
561 * pointer is the first word of the object.
563 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
566 * object + s->objsize
567 * Padding to reach word boundary. This is also used for Redzoning.
568 * Padding is extended by another word if Redzoning is enabled and
571 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
572 * 0xcc (RED_ACTIVE) for objects in use.
575 * Meta data starts here.
577 * A. Free pointer (if we cannot overwrite object on free)
578 * B. Tracking data for SLAB_STORE_USER
579 * C. Padding to reach required alignment boundary or at mininum
580 * one word if debugging is on to be able to detect writes
581 * before the word boundary.
583 * Padding is done using 0x5a (POISON_INUSE)
586 * Nothing is used beyond s->size.
588 * If slabcaches are merged then the objsize and inuse boundaries are mostly
589 * ignored. And therefore no slab options that rely on these boundaries
590 * may be used with merged slabcaches.
593 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
595 unsigned long off
= s
->inuse
; /* The end of info */
598 /* Freepointer is placed after the object. */
599 off
+= sizeof(void *);
601 if (s
->flags
& SLAB_STORE_USER
)
602 /* We also have user information there */
603 off
+= 2 * sizeof(struct track
);
608 return check_bytes_and_report(s
, page
, p
, "Object padding",
609 p
+ off
, POISON_INUSE
, s
->size
- off
);
612 /* Check the pad bytes at the end of a slab page */
613 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
621 if (!(s
->flags
& SLAB_POISON
))
624 start
= page_address(page
);
625 length
= (PAGE_SIZE
<< compound_order(page
));
626 end
= start
+ length
;
627 remainder
= length
% s
->size
;
631 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
634 while (end
> fault
&& end
[-1] == POISON_INUSE
)
637 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
638 print_section("Padding", end
- remainder
, remainder
);
640 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
644 static int check_object(struct kmem_cache
*s
, struct page
*page
,
645 void *object
, int active
)
648 u8
*endobject
= object
+ s
->objsize
;
650 if (s
->flags
& SLAB_RED_ZONE
) {
652 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
654 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
655 endobject
, red
, s
->inuse
- s
->objsize
))
658 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
659 check_bytes_and_report(s
, page
, p
, "Alignment padding",
660 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
664 if (s
->flags
& SLAB_POISON
) {
665 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
666 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
667 POISON_FREE
, s
->objsize
- 1) ||
668 !check_bytes_and_report(s
, page
, p
, "Poison",
669 p
+ s
->objsize
- 1, POISON_END
, 1)))
672 * check_pad_bytes cleans up on its own.
674 check_pad_bytes(s
, page
, p
);
677 if (!s
->offset
&& active
)
679 * Object and freepointer overlap. Cannot check
680 * freepointer while object is allocated.
684 /* Check free pointer validity */
685 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
686 object_err(s
, page
, p
, "Freepointer corrupt");
688 * No choice but to zap it and thus lose the remainder
689 * of the free objects in this slab. May cause
690 * another error because the object count is now wrong.
692 set_freepointer(s
, p
, NULL
);
698 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
702 VM_BUG_ON(!irqs_disabled());
704 if (!PageSlab(page
)) {
705 slab_err(s
, page
, "Not a valid slab page");
709 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
710 if (page
->objects
> maxobj
) {
711 slab_err(s
, page
, "objects %u > max %u",
712 s
->name
, page
->objects
, maxobj
);
715 if (page
->inuse
> page
->objects
) {
716 slab_err(s
, page
, "inuse %u > max %u",
717 s
->name
, page
->inuse
, page
->objects
);
720 /* Slab_pad_check fixes things up after itself */
721 slab_pad_check(s
, page
);
726 * Determine if a certain object on a page is on the freelist. Must hold the
727 * slab lock to guarantee that the chains are in a consistent state.
729 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
732 void *fp
= page
->freelist
;
734 unsigned long max_objects
;
736 while (fp
&& nr
<= page
->objects
) {
739 if (!check_valid_pointer(s
, page
, fp
)) {
741 object_err(s
, page
, object
,
742 "Freechain corrupt");
743 set_freepointer(s
, object
, NULL
);
746 slab_err(s
, page
, "Freepointer corrupt");
747 page
->freelist
= NULL
;
748 page
->inuse
= page
->objects
;
749 slab_fix(s
, "Freelist cleared");
755 fp
= get_freepointer(s
, object
);
759 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
760 if (max_objects
> MAX_OBJS_PER_PAGE
)
761 max_objects
= MAX_OBJS_PER_PAGE
;
763 if (page
->objects
!= max_objects
) {
764 slab_err(s
, page
, "Wrong number of objects. Found %d but "
765 "should be %d", page
->objects
, max_objects
);
766 page
->objects
= max_objects
;
767 slab_fix(s
, "Number of objects adjusted.");
769 if (page
->inuse
!= page
->objects
- nr
) {
770 slab_err(s
, page
, "Wrong object count. Counter is %d but "
771 "counted were %d", page
->inuse
, page
->objects
- nr
);
772 page
->inuse
= page
->objects
- nr
;
773 slab_fix(s
, "Object count adjusted.");
775 return search
== NULL
;
778 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
781 if (s
->flags
& SLAB_TRACE
) {
782 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
784 alloc
? "alloc" : "free",
789 print_section("Object", (void *)object
, s
->objsize
);
796 * Tracking of fully allocated slabs for debugging purposes.
798 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
800 spin_lock(&n
->list_lock
);
801 list_add(&page
->lru
, &n
->full
);
802 spin_unlock(&n
->list_lock
);
805 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
807 struct kmem_cache_node
*n
;
809 if (!(s
->flags
& SLAB_STORE_USER
))
812 n
= get_node(s
, page_to_nid(page
));
814 spin_lock(&n
->list_lock
);
815 list_del(&page
->lru
);
816 spin_unlock(&n
->list_lock
);
819 /* Tracking of the number of slabs for debugging purposes */
820 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
822 struct kmem_cache_node
*n
= get_node(s
, node
);
824 return atomic_long_read(&n
->nr_slabs
);
827 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
829 return atomic_long_read(&n
->nr_slabs
);
832 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
834 struct kmem_cache_node
*n
= get_node(s
, node
);
837 * May be called early in order to allocate a slab for the
838 * kmem_cache_node structure. Solve the chicken-egg
839 * dilemma by deferring the increment of the count during
840 * bootstrap (see early_kmem_cache_node_alloc).
842 if (!NUMA_BUILD
|| n
) {
843 atomic_long_inc(&n
->nr_slabs
);
844 atomic_long_add(objects
, &n
->total_objects
);
847 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
849 struct kmem_cache_node
*n
= get_node(s
, node
);
851 atomic_long_dec(&n
->nr_slabs
);
852 atomic_long_sub(objects
, &n
->total_objects
);
855 /* Object debug checks for alloc/free paths */
856 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
859 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
862 init_object(s
, object
, 0);
863 init_tracking(s
, object
);
866 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
867 void *object
, unsigned long addr
)
869 if (!check_slab(s
, page
))
872 if (!on_freelist(s
, page
, object
)) {
873 object_err(s
, page
, object
, "Object already allocated");
877 if (!check_valid_pointer(s
, page
, object
)) {
878 object_err(s
, page
, object
, "Freelist Pointer check fails");
882 if (!check_object(s
, page
, object
, 0))
885 /* Success perform special debug activities for allocs */
886 if (s
->flags
& SLAB_STORE_USER
)
887 set_track(s
, object
, TRACK_ALLOC
, addr
);
888 trace(s
, page
, object
, 1);
889 init_object(s
, object
, 1);
893 if (PageSlab(page
)) {
895 * If this is a slab page then lets do the best we can
896 * to avoid issues in the future. Marking all objects
897 * as used avoids touching the remaining objects.
899 slab_fix(s
, "Marking all objects used");
900 page
->inuse
= page
->objects
;
901 page
->freelist
= NULL
;
906 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
907 void *object
, unsigned long addr
)
909 if (!check_slab(s
, page
))
912 if (!check_valid_pointer(s
, page
, object
)) {
913 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
917 if (on_freelist(s
, page
, object
)) {
918 object_err(s
, page
, object
, "Object already free");
922 if (!check_object(s
, page
, object
, 1))
925 if (unlikely(s
!= page
->slab
)) {
926 if (!PageSlab(page
)) {
927 slab_err(s
, page
, "Attempt to free object(0x%p) "
928 "outside of slab", object
);
929 } else if (!page
->slab
) {
931 "SLUB <none>: no slab for object 0x%p.\n",
935 object_err(s
, page
, object
,
936 "page slab pointer corrupt.");
940 /* Special debug activities for freeing objects */
941 if (!PageSlubFrozen(page
) && !page
->freelist
)
942 remove_full(s
, page
);
943 if (s
->flags
& SLAB_STORE_USER
)
944 set_track(s
, object
, TRACK_FREE
, addr
);
945 trace(s
, page
, object
, 0);
946 init_object(s
, object
, 0);
950 slab_fix(s
, "Object at 0x%p not freed", object
);
954 static int __init
setup_slub_debug(char *str
)
956 slub_debug
= DEBUG_DEFAULT_FLAGS
;
957 if (*str
++ != '=' || !*str
)
959 * No options specified. Switch on full debugging.
965 * No options but restriction on slabs. This means full
966 * debugging for slabs matching a pattern.
970 if (tolower(*str
) == 'o') {
972 * Avoid enabling debugging on caches if its minimum order
973 * would increase as a result.
975 disable_higher_order_debug
= 1;
982 * Switch off all debugging measures.
987 * Determine which debug features should be switched on
989 for (; *str
&& *str
!= ','; str
++) {
990 switch (tolower(*str
)) {
992 slub_debug
|= SLAB_DEBUG_FREE
;
995 slub_debug
|= SLAB_RED_ZONE
;
998 slub_debug
|= SLAB_POISON
;
1001 slub_debug
|= SLAB_STORE_USER
;
1004 slub_debug
|= SLAB_TRACE
;
1007 slub_debug
|= SLAB_FAILSLAB
;
1010 printk(KERN_ERR
"slub_debug option '%c' "
1011 "unknown. skipped\n", *str
);
1017 slub_debug_slabs
= str
+ 1;
1022 __setup("slub_debug", setup_slub_debug
);
1024 static unsigned long kmem_cache_flags(unsigned long objsize
,
1025 unsigned long flags
, const char *name
,
1026 void (*ctor
)(void *))
1029 * Enable debugging if selected on the kernel commandline.
1031 if (slub_debug
&& (!slub_debug_slabs
||
1032 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1033 flags
|= slub_debug
;
1038 static inline void setup_object_debug(struct kmem_cache
*s
,
1039 struct page
*page
, void *object
) {}
1041 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1042 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1044 static inline int free_debug_processing(struct kmem_cache
*s
,
1045 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1047 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1049 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1050 void *object
, int active
) { return 1; }
1051 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1052 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1053 unsigned long flags
, const char *name
,
1054 void (*ctor
)(void *))
1058 #define slub_debug 0
1060 #define disable_higher_order_debug 0
1062 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1064 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1066 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1068 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1073 * Slab allocation and freeing
1075 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1076 struct kmem_cache_order_objects oo
)
1078 int order
= oo_order(oo
);
1080 flags
|= __GFP_NOTRACK
;
1082 if (node
== NUMA_NO_NODE
)
1083 return alloc_pages(flags
, order
);
1085 return alloc_pages_exact_node(node
, flags
, order
);
1088 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1091 struct kmem_cache_order_objects oo
= s
->oo
;
1094 flags
|= s
->allocflags
;
1097 * Let the initial higher-order allocation fail under memory pressure
1098 * so we fall-back to the minimum order allocation.
1100 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1102 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1103 if (unlikely(!page
)) {
1106 * Allocation may have failed due to fragmentation.
1107 * Try a lower order alloc if possible
1109 page
= alloc_slab_page(flags
, node
, oo
);
1113 stat(s
, ORDER_FALLBACK
);
1116 if (kmemcheck_enabled
1117 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1118 int pages
= 1 << oo_order(oo
);
1120 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1123 * Objects from caches that have a constructor don't get
1124 * cleared when they're allocated, so we need to do it here.
1127 kmemcheck_mark_uninitialized_pages(page
, pages
);
1129 kmemcheck_mark_unallocated_pages(page
, pages
);
1132 page
->objects
= oo_objects(oo
);
1133 mod_zone_page_state(page_zone(page
),
1134 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1135 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1141 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1144 setup_object_debug(s
, page
, object
);
1145 if (unlikely(s
->ctor
))
1149 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1156 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1158 page
= allocate_slab(s
,
1159 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1163 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1165 page
->flags
|= 1 << PG_slab
;
1167 start
= page_address(page
);
1169 if (unlikely(s
->flags
& SLAB_POISON
))
1170 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1173 for_each_object(p
, s
, start
, page
->objects
) {
1174 setup_object(s
, page
, last
);
1175 set_freepointer(s
, last
, p
);
1178 setup_object(s
, page
, last
);
1179 set_freepointer(s
, last
, NULL
);
1181 page
->freelist
= start
;
1187 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1189 int order
= compound_order(page
);
1190 int pages
= 1 << order
;
1192 if (kmem_cache_debug(s
)) {
1195 slab_pad_check(s
, page
);
1196 for_each_object(p
, s
, page_address(page
),
1198 check_object(s
, page
, p
, 0);
1201 kmemcheck_free_shadow(page
, compound_order(page
));
1203 mod_zone_page_state(page_zone(page
),
1204 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1205 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1208 __ClearPageSlab(page
);
1209 reset_page_mapcount(page
);
1210 if (current
->reclaim_state
)
1211 current
->reclaim_state
->reclaimed_slab
+= pages
;
1212 __free_pages(page
, order
);
1215 static void rcu_free_slab(struct rcu_head
*h
)
1219 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1220 __free_slab(page
->slab
, page
);
1223 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1225 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1227 * RCU free overloads the RCU head over the LRU
1229 struct rcu_head
*head
= (void *)&page
->lru
;
1231 call_rcu(head
, rcu_free_slab
);
1233 __free_slab(s
, page
);
1236 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1238 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1243 * Per slab locking using the pagelock
1245 static __always_inline
void slab_lock(struct page
*page
)
1247 bit_spin_lock(PG_locked
, &page
->flags
);
1250 static __always_inline
void slab_unlock(struct page
*page
)
1252 __bit_spin_unlock(PG_locked
, &page
->flags
);
1255 static __always_inline
int slab_trylock(struct page
*page
)
1259 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1264 * Management of partially allocated slabs
1266 static void add_partial(struct kmem_cache_node
*n
,
1267 struct page
*page
, int tail
)
1269 spin_lock(&n
->list_lock
);
1272 list_add_tail(&page
->lru
, &n
->partial
);
1274 list_add(&page
->lru
, &n
->partial
);
1275 spin_unlock(&n
->list_lock
);
1278 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1280 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1282 spin_lock(&n
->list_lock
);
1283 list_del(&page
->lru
);
1285 spin_unlock(&n
->list_lock
);
1289 * Lock slab and remove from the partial list.
1291 * Must hold list_lock.
1293 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1296 if (slab_trylock(page
)) {
1297 list_del(&page
->lru
);
1299 __SetPageSlubFrozen(page
);
1306 * Try to allocate a partial slab from a specific node.
1308 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1313 * Racy check. If we mistakenly see no partial slabs then we
1314 * just allocate an empty slab. If we mistakenly try to get a
1315 * partial slab and there is none available then get_partials()
1318 if (!n
|| !n
->nr_partial
)
1321 spin_lock(&n
->list_lock
);
1322 list_for_each_entry(page
, &n
->partial
, lru
)
1323 if (lock_and_freeze_slab(n
, page
))
1327 spin_unlock(&n
->list_lock
);
1332 * Get a page from somewhere. Search in increasing NUMA distances.
1334 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1337 struct zonelist
*zonelist
;
1340 enum zone_type high_zoneidx
= gfp_zone(flags
);
1344 * The defrag ratio allows a configuration of the tradeoffs between
1345 * inter node defragmentation and node local allocations. A lower
1346 * defrag_ratio increases the tendency to do local allocations
1347 * instead of attempting to obtain partial slabs from other nodes.
1349 * If the defrag_ratio is set to 0 then kmalloc() always
1350 * returns node local objects. If the ratio is higher then kmalloc()
1351 * may return off node objects because partial slabs are obtained
1352 * from other nodes and filled up.
1354 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1355 * defrag_ratio = 1000) then every (well almost) allocation will
1356 * first attempt to defrag slab caches on other nodes. This means
1357 * scanning over all nodes to look for partial slabs which may be
1358 * expensive if we do it every time we are trying to find a slab
1359 * with available objects.
1361 if (!s
->remote_node_defrag_ratio
||
1362 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1366 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1367 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1368 struct kmem_cache_node
*n
;
1370 n
= get_node(s
, zone_to_nid(zone
));
1372 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1373 n
->nr_partial
> s
->min_partial
) {
1374 page
= get_partial_node(n
);
1387 * Get a partial page, lock it and return it.
1389 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1392 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1394 page
= get_partial_node(get_node(s
, searchnode
));
1395 if (page
|| (flags
& __GFP_THISNODE
))
1398 return get_any_partial(s
, flags
);
1402 * Move a page back to the lists.
1404 * Must be called with the slab lock held.
1406 * On exit the slab lock will have been dropped.
1408 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1410 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1412 __ClearPageSlubFrozen(page
);
1415 if (page
->freelist
) {
1416 add_partial(n
, page
, tail
);
1417 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1419 stat(s
, DEACTIVATE_FULL
);
1420 if (kmem_cache_debug(s
) && (s
->flags
& SLAB_STORE_USER
))
1425 stat(s
, DEACTIVATE_EMPTY
);
1426 if (n
->nr_partial
< s
->min_partial
) {
1428 * Adding an empty slab to the partial slabs in order
1429 * to avoid page allocator overhead. This slab needs
1430 * to come after the other slabs with objects in
1431 * so that the others get filled first. That way the
1432 * size of the partial list stays small.
1434 * kmem_cache_shrink can reclaim any empty slabs from
1437 add_partial(n
, page
, 1);
1442 discard_slab(s
, page
);
1448 * Remove the cpu slab
1450 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1452 struct page
*page
= c
->page
;
1456 stat(s
, DEACTIVATE_REMOTE_FREES
);
1458 * Merge cpu freelist into slab freelist. Typically we get here
1459 * because both freelists are empty. So this is unlikely
1462 while (unlikely(c
->freelist
)) {
1465 tail
= 0; /* Hot objects. Put the slab first */
1467 /* Retrieve object from cpu_freelist */
1468 object
= c
->freelist
;
1469 c
->freelist
= get_freepointer(s
, c
->freelist
);
1471 /* And put onto the regular freelist */
1472 set_freepointer(s
, object
, page
->freelist
);
1473 page
->freelist
= object
;
1477 unfreeze_slab(s
, page
, tail
);
1480 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1482 stat(s
, CPUSLAB_FLUSH
);
1484 deactivate_slab(s
, c
);
1490 * Called from IPI handler with interrupts disabled.
1492 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1494 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1496 if (likely(c
&& c
->page
))
1500 static void flush_cpu_slab(void *d
)
1502 struct kmem_cache
*s
= d
;
1504 __flush_cpu_slab(s
, smp_processor_id());
1507 static void flush_all(struct kmem_cache
*s
)
1509 on_each_cpu(flush_cpu_slab
, s
, 1);
1513 * Check if the objects in a per cpu structure fit numa
1514 * locality expectations.
1516 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1519 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
1525 static int count_free(struct page
*page
)
1527 return page
->objects
- page
->inuse
;
1530 static unsigned long count_partial(struct kmem_cache_node
*n
,
1531 int (*get_count
)(struct page
*))
1533 unsigned long flags
;
1534 unsigned long x
= 0;
1537 spin_lock_irqsave(&n
->list_lock
, flags
);
1538 list_for_each_entry(page
, &n
->partial
, lru
)
1539 x
+= get_count(page
);
1540 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1544 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1546 #ifdef CONFIG_SLUB_DEBUG
1547 return atomic_long_read(&n
->total_objects
);
1553 static noinline
void
1554 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1559 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1561 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1562 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1563 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1565 if (oo_order(s
->min
) > get_order(s
->objsize
))
1566 printk(KERN_WARNING
" %s debugging increased min order, use "
1567 "slub_debug=O to disable.\n", s
->name
);
1569 for_each_online_node(node
) {
1570 struct kmem_cache_node
*n
= get_node(s
, node
);
1571 unsigned long nr_slabs
;
1572 unsigned long nr_objs
;
1573 unsigned long nr_free
;
1578 nr_free
= count_partial(n
, count_free
);
1579 nr_slabs
= node_nr_slabs(n
);
1580 nr_objs
= node_nr_objs(n
);
1583 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1584 node
, nr_slabs
, nr_objs
, nr_free
);
1589 * Slow path. The lockless freelist is empty or we need to perform
1592 * Interrupts are disabled.
1594 * Processing is still very fast if new objects have been freed to the
1595 * regular freelist. In that case we simply take over the regular freelist
1596 * as the lockless freelist and zap the regular freelist.
1598 * If that is not working then we fall back to the partial lists. We take the
1599 * first element of the freelist as the object to allocate now and move the
1600 * rest of the freelist to the lockless freelist.
1602 * And if we were unable to get a new slab from the partial slab lists then
1603 * we need to allocate a new slab. This is the slowest path since it involves
1604 * a call to the page allocator and the setup of a new slab.
1606 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1607 unsigned long addr
, struct kmem_cache_cpu
*c
)
1612 /* We handle __GFP_ZERO in the caller */
1613 gfpflags
&= ~__GFP_ZERO
;
1619 if (unlikely(!node_match(c
, node
)))
1622 stat(s
, ALLOC_REFILL
);
1625 object
= c
->page
->freelist
;
1626 if (unlikely(!object
))
1628 if (kmem_cache_debug(s
))
1631 c
->freelist
= get_freepointer(s
, object
);
1632 c
->page
->inuse
= c
->page
->objects
;
1633 c
->page
->freelist
= NULL
;
1634 c
->node
= page_to_nid(c
->page
);
1636 slab_unlock(c
->page
);
1637 stat(s
, ALLOC_SLOWPATH
);
1641 deactivate_slab(s
, c
);
1644 new = get_partial(s
, gfpflags
, node
);
1647 stat(s
, ALLOC_FROM_PARTIAL
);
1651 if (gfpflags
& __GFP_WAIT
)
1654 new = new_slab(s
, gfpflags
, node
);
1656 if (gfpflags
& __GFP_WAIT
)
1657 local_irq_disable();
1660 c
= __this_cpu_ptr(s
->cpu_slab
);
1661 stat(s
, ALLOC_SLAB
);
1665 __SetPageSlubFrozen(new);
1669 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1670 slab_out_of_memory(s
, gfpflags
, node
);
1673 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1677 c
->page
->freelist
= get_freepointer(s
, object
);
1683 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1684 * have the fastpath folded into their functions. So no function call
1685 * overhead for requests that can be satisfied on the fastpath.
1687 * The fastpath works by first checking if the lockless freelist can be used.
1688 * If not then __slab_alloc is called for slow processing.
1690 * Otherwise we can simply pick the next object from the lockless free list.
1692 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1693 gfp_t gfpflags
, int node
, unsigned long addr
)
1696 struct kmem_cache_cpu
*c
;
1697 unsigned long flags
;
1699 gfpflags
&= gfp_allowed_mask
;
1701 lockdep_trace_alloc(gfpflags
);
1702 might_sleep_if(gfpflags
& __GFP_WAIT
);
1704 if (should_failslab(s
->objsize
, gfpflags
, s
->flags
))
1707 local_irq_save(flags
);
1708 c
= __this_cpu_ptr(s
->cpu_slab
);
1709 object
= c
->freelist
;
1710 if (unlikely(!object
|| !node_match(c
, node
)))
1712 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1715 c
->freelist
= get_freepointer(s
, object
);
1716 stat(s
, ALLOC_FASTPATH
);
1718 local_irq_restore(flags
);
1720 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1721 memset(object
, 0, s
->objsize
);
1723 kmemcheck_slab_alloc(s
, gfpflags
, object
, s
->objsize
);
1724 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, gfpflags
);
1729 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1731 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1733 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1737 EXPORT_SYMBOL(kmem_cache_alloc
);
1739 #ifdef CONFIG_TRACING
1740 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1742 return slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1744 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1748 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1750 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1752 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1753 s
->objsize
, s
->size
, gfpflags
, node
);
1757 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1760 #ifdef CONFIG_TRACING
1761 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1765 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1767 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1771 * Slow patch handling. This may still be called frequently since objects
1772 * have a longer lifetime than the cpu slabs in most processing loads.
1774 * So we still attempt to reduce cache line usage. Just take the slab
1775 * lock and free the item. If there is no additional partial page
1776 * handling required then we can return immediately.
1778 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1779 void *x
, unsigned long addr
)
1782 void **object
= (void *)x
;
1784 stat(s
, FREE_SLOWPATH
);
1787 if (kmem_cache_debug(s
))
1791 prior
= page
->freelist
;
1792 set_freepointer(s
, object
, prior
);
1793 page
->freelist
= object
;
1796 if (unlikely(PageSlubFrozen(page
))) {
1797 stat(s
, FREE_FROZEN
);
1801 if (unlikely(!page
->inuse
))
1805 * Objects left in the slab. If it was not on the partial list before
1808 if (unlikely(!prior
)) {
1809 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1810 stat(s
, FREE_ADD_PARTIAL
);
1820 * Slab still on the partial list.
1822 remove_partial(s
, page
);
1823 stat(s
, FREE_REMOVE_PARTIAL
);
1827 discard_slab(s
, page
);
1831 if (!free_debug_processing(s
, page
, x
, addr
))
1837 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1838 * can perform fastpath freeing without additional function calls.
1840 * The fastpath is only possible if we are freeing to the current cpu slab
1841 * of this processor. This typically the case if we have just allocated
1844 * If fastpath is not possible then fall back to __slab_free where we deal
1845 * with all sorts of special processing.
1847 static __always_inline
void slab_free(struct kmem_cache
*s
,
1848 struct page
*page
, void *x
, unsigned long addr
)
1850 void **object
= (void *)x
;
1851 struct kmem_cache_cpu
*c
;
1852 unsigned long flags
;
1854 kmemleak_free_recursive(x
, s
->flags
);
1855 local_irq_save(flags
);
1856 c
= __this_cpu_ptr(s
->cpu_slab
);
1857 kmemcheck_slab_free(s
, object
, s
->objsize
);
1858 debug_check_no_locks_freed(object
, s
->objsize
);
1859 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1860 debug_check_no_obj_freed(object
, s
->objsize
);
1861 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1862 set_freepointer(s
, object
, c
->freelist
);
1863 c
->freelist
= object
;
1864 stat(s
, FREE_FASTPATH
);
1866 __slab_free(s
, page
, x
, addr
);
1868 local_irq_restore(flags
);
1871 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1875 page
= virt_to_head_page(x
);
1877 slab_free(s
, page
, x
, _RET_IP_
);
1879 trace_kmem_cache_free(_RET_IP_
, x
);
1881 EXPORT_SYMBOL(kmem_cache_free
);
1883 /* Figure out on which slab page the object resides */
1884 static struct page
*get_object_page(const void *x
)
1886 struct page
*page
= virt_to_head_page(x
);
1888 if (!PageSlab(page
))
1895 * Object placement in a slab is made very easy because we always start at
1896 * offset 0. If we tune the size of the object to the alignment then we can
1897 * get the required alignment by putting one properly sized object after
1900 * Notice that the allocation order determines the sizes of the per cpu
1901 * caches. Each processor has always one slab available for allocations.
1902 * Increasing the allocation order reduces the number of times that slabs
1903 * must be moved on and off the partial lists and is therefore a factor in
1908 * Mininum / Maximum order of slab pages. This influences locking overhead
1909 * and slab fragmentation. A higher order reduces the number of partial slabs
1910 * and increases the number of allocations possible without having to
1911 * take the list_lock.
1913 static int slub_min_order
;
1914 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1915 static int slub_min_objects
;
1918 * Merge control. If this is set then no merging of slab caches will occur.
1919 * (Could be removed. This was introduced to pacify the merge skeptics.)
1921 static int slub_nomerge
;
1924 * Calculate the order of allocation given an slab object size.
1926 * The order of allocation has significant impact on performance and other
1927 * system components. Generally order 0 allocations should be preferred since
1928 * order 0 does not cause fragmentation in the page allocator. Larger objects
1929 * be problematic to put into order 0 slabs because there may be too much
1930 * unused space left. We go to a higher order if more than 1/16th of the slab
1933 * In order to reach satisfactory performance we must ensure that a minimum
1934 * number of objects is in one slab. Otherwise we may generate too much
1935 * activity on the partial lists which requires taking the list_lock. This is
1936 * less a concern for large slabs though which are rarely used.
1938 * slub_max_order specifies the order where we begin to stop considering the
1939 * number of objects in a slab as critical. If we reach slub_max_order then
1940 * we try to keep the page order as low as possible. So we accept more waste
1941 * of space in favor of a small page order.
1943 * Higher order allocations also allow the placement of more objects in a
1944 * slab and thereby reduce object handling overhead. If the user has
1945 * requested a higher mininum order then we start with that one instead of
1946 * the smallest order which will fit the object.
1948 static inline int slab_order(int size
, int min_objects
,
1949 int max_order
, int fract_leftover
)
1953 int min_order
= slub_min_order
;
1955 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1956 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1958 for (order
= max(min_order
,
1959 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1960 order
<= max_order
; order
++) {
1962 unsigned long slab_size
= PAGE_SIZE
<< order
;
1964 if (slab_size
< min_objects
* size
)
1967 rem
= slab_size
% size
;
1969 if (rem
<= slab_size
/ fract_leftover
)
1977 static inline int calculate_order(int size
)
1985 * Attempt to find best configuration for a slab. This
1986 * works by first attempting to generate a layout with
1987 * the best configuration and backing off gradually.
1989 * First we reduce the acceptable waste in a slab. Then
1990 * we reduce the minimum objects required in a slab.
1992 min_objects
= slub_min_objects
;
1994 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1995 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1996 min_objects
= min(min_objects
, max_objects
);
1998 while (min_objects
> 1) {
2000 while (fraction
>= 4) {
2001 order
= slab_order(size
, min_objects
,
2002 slub_max_order
, fraction
);
2003 if (order
<= slub_max_order
)
2011 * We were unable to place multiple objects in a slab. Now
2012 * lets see if we can place a single object there.
2014 order
= slab_order(size
, 1, slub_max_order
, 1);
2015 if (order
<= slub_max_order
)
2019 * Doh this slab cannot be placed using slub_max_order.
2021 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2022 if (order
< MAX_ORDER
)
2028 * Figure out what the alignment of the objects will be.
2030 static unsigned long calculate_alignment(unsigned long flags
,
2031 unsigned long align
, unsigned long size
)
2034 * If the user wants hardware cache aligned objects then follow that
2035 * suggestion if the object is sufficiently large.
2037 * The hardware cache alignment cannot override the specified
2038 * alignment though. If that is greater then use it.
2040 if (flags
& SLAB_HWCACHE_ALIGN
) {
2041 unsigned long ralign
= cache_line_size();
2042 while (size
<= ralign
/ 2)
2044 align
= max(align
, ralign
);
2047 if (align
< ARCH_SLAB_MINALIGN
)
2048 align
= ARCH_SLAB_MINALIGN
;
2050 return ALIGN(align
, sizeof(void *));
2054 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2057 spin_lock_init(&n
->list_lock
);
2058 INIT_LIST_HEAD(&n
->partial
);
2059 #ifdef CONFIG_SLUB_DEBUG
2060 atomic_long_set(&n
->nr_slabs
, 0);
2061 atomic_long_set(&n
->total_objects
, 0);
2062 INIT_LIST_HEAD(&n
->full
);
2066 static DEFINE_PER_CPU(struct kmem_cache_cpu
, kmalloc_percpu
[KMALLOC_CACHES
]);
2068 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2070 if (s
< kmalloc_caches
+ KMALLOC_CACHES
&& s
>= kmalloc_caches
)
2072 * Boot time creation of the kmalloc array. Use static per cpu data
2073 * since the per cpu allocator is not available yet.
2075 s
->cpu_slab
= kmalloc_percpu
+ (s
- kmalloc_caches
);
2077 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2087 * No kmalloc_node yet so do it by hand. We know that this is the first
2088 * slab on the node for this slabcache. There are no concurrent accesses
2091 * Note that this function only works on the kmalloc_node_cache
2092 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2093 * memory on a fresh node that has no slab structures yet.
2095 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2098 struct kmem_cache_node
*n
;
2099 unsigned long flags
;
2101 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2103 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2106 if (page_to_nid(page
) != node
) {
2107 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2109 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2110 "in order to be able to continue\n");
2115 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2117 kmalloc_caches
->node
[node
] = n
;
2118 #ifdef CONFIG_SLUB_DEBUG
2119 init_object(kmalloc_caches
, n
, 1);
2120 init_tracking(kmalloc_caches
, n
);
2122 init_kmem_cache_node(n
, kmalloc_caches
);
2123 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2126 * lockdep requires consistent irq usage for each lock
2127 * so even though there cannot be a race this early in
2128 * the boot sequence, we still disable irqs.
2130 local_irq_save(flags
);
2131 add_partial(n
, page
, 0);
2132 local_irq_restore(flags
);
2135 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2139 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2140 struct kmem_cache_node
*n
= s
->node
[node
];
2142 kmem_cache_free(kmalloc_caches
, n
);
2143 s
->node
[node
] = NULL
;
2147 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2151 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2152 struct kmem_cache_node
*n
;
2154 if (slab_state
== DOWN
) {
2155 early_kmem_cache_node_alloc(gfpflags
, node
);
2158 n
= kmem_cache_alloc_node(kmalloc_caches
,
2162 free_kmem_cache_nodes(s
);
2167 init_kmem_cache_node(n
, s
);
2172 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2176 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2178 init_kmem_cache_node(&s
->local_node
, s
);
2183 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2185 if (min
< MIN_PARTIAL
)
2187 else if (min
> MAX_PARTIAL
)
2189 s
->min_partial
= min
;
2193 * calculate_sizes() determines the order and the distribution of data within
2196 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2198 unsigned long flags
= s
->flags
;
2199 unsigned long size
= s
->objsize
;
2200 unsigned long align
= s
->align
;
2204 * Round up object size to the next word boundary. We can only
2205 * place the free pointer at word boundaries and this determines
2206 * the possible location of the free pointer.
2208 size
= ALIGN(size
, sizeof(void *));
2210 #ifdef CONFIG_SLUB_DEBUG
2212 * Determine if we can poison the object itself. If the user of
2213 * the slab may touch the object after free or before allocation
2214 * then we should never poison the object itself.
2216 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2218 s
->flags
|= __OBJECT_POISON
;
2220 s
->flags
&= ~__OBJECT_POISON
;
2224 * If we are Redzoning then check if there is some space between the
2225 * end of the object and the free pointer. If not then add an
2226 * additional word to have some bytes to store Redzone information.
2228 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2229 size
+= sizeof(void *);
2233 * With that we have determined the number of bytes in actual use
2234 * by the object. This is the potential offset to the free pointer.
2238 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2241 * Relocate free pointer after the object if it is not
2242 * permitted to overwrite the first word of the object on
2245 * This is the case if we do RCU, have a constructor or
2246 * destructor or are poisoning the objects.
2249 size
+= sizeof(void *);
2252 #ifdef CONFIG_SLUB_DEBUG
2253 if (flags
& SLAB_STORE_USER
)
2255 * Need to store information about allocs and frees after
2258 size
+= 2 * sizeof(struct track
);
2260 if (flags
& SLAB_RED_ZONE
)
2262 * Add some empty padding so that we can catch
2263 * overwrites from earlier objects rather than let
2264 * tracking information or the free pointer be
2265 * corrupted if a user writes before the start
2268 size
+= sizeof(void *);
2272 * Determine the alignment based on various parameters that the
2273 * user specified and the dynamic determination of cache line size
2276 align
= calculate_alignment(flags
, align
, s
->objsize
);
2280 * SLUB stores one object immediately after another beginning from
2281 * offset 0. In order to align the objects we have to simply size
2282 * each object to conform to the alignment.
2284 size
= ALIGN(size
, align
);
2286 if (forced_order
>= 0)
2287 order
= forced_order
;
2289 order
= calculate_order(size
);
2296 s
->allocflags
|= __GFP_COMP
;
2298 if (s
->flags
& SLAB_CACHE_DMA
)
2299 s
->allocflags
|= SLUB_DMA
;
2301 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2302 s
->allocflags
|= __GFP_RECLAIMABLE
;
2305 * Determine the number of objects per slab
2307 s
->oo
= oo_make(order
, size
);
2308 s
->min
= oo_make(get_order(size
), size
);
2309 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2312 return !!oo_objects(s
->oo
);
2316 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2317 const char *name
, size_t size
,
2318 size_t align
, unsigned long flags
,
2319 void (*ctor
)(void *))
2321 memset(s
, 0, kmem_size
);
2326 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2328 if (!calculate_sizes(s
, -1))
2330 if (disable_higher_order_debug
) {
2332 * Disable debugging flags that store metadata if the min slab
2335 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2336 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2338 if (!calculate_sizes(s
, -1))
2344 * The larger the object size is, the more pages we want on the partial
2345 * list to avoid pounding the page allocator excessively.
2347 set_min_partial(s
, ilog2(s
->size
));
2350 s
->remote_node_defrag_ratio
= 1000;
2352 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2355 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2358 free_kmem_cache_nodes(s
);
2360 if (flags
& SLAB_PANIC
)
2361 panic("Cannot create slab %s size=%lu realsize=%u "
2362 "order=%u offset=%u flags=%lx\n",
2363 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2369 * Check if a given pointer is valid
2371 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2375 if (!kern_ptr_validate(object
, s
->size
))
2378 page
= get_object_page(object
);
2380 if (!page
|| s
!= page
->slab
)
2381 /* No slab or wrong slab */
2384 if (!check_valid_pointer(s
, page
, object
))
2388 * We could also check if the object is on the slabs freelist.
2389 * But this would be too expensive and it seems that the main
2390 * purpose of kmem_ptr_valid() is to check if the object belongs
2391 * to a certain slab.
2395 EXPORT_SYMBOL(kmem_ptr_validate
);
2398 * Determine the size of a slab object
2400 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2404 EXPORT_SYMBOL(kmem_cache_size
);
2406 const char *kmem_cache_name(struct kmem_cache
*s
)
2410 EXPORT_SYMBOL(kmem_cache_name
);
2412 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2415 #ifdef CONFIG_SLUB_DEBUG
2416 void *addr
= page_address(page
);
2418 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2423 slab_err(s
, page
, "%s", text
);
2425 for_each_free_object(p
, s
, page
->freelist
)
2426 set_bit(slab_index(p
, s
, addr
), map
);
2428 for_each_object(p
, s
, addr
, page
->objects
) {
2430 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2431 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2433 print_tracking(s
, p
);
2442 * Attempt to free all partial slabs on a node.
2444 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2446 unsigned long flags
;
2447 struct page
*page
, *h
;
2449 spin_lock_irqsave(&n
->list_lock
, flags
);
2450 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2452 list_del(&page
->lru
);
2453 discard_slab(s
, page
);
2456 list_slab_objects(s
, page
,
2457 "Objects remaining on kmem_cache_close()");
2460 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2464 * Release all resources used by a slab cache.
2466 static inline int kmem_cache_close(struct kmem_cache
*s
)
2471 free_percpu(s
->cpu_slab
);
2472 /* Attempt to free all objects */
2473 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2474 struct kmem_cache_node
*n
= get_node(s
, node
);
2477 if (n
->nr_partial
|| slabs_node(s
, node
))
2480 free_kmem_cache_nodes(s
);
2485 * Close a cache and release the kmem_cache structure
2486 * (must be used for caches created using kmem_cache_create)
2488 void kmem_cache_destroy(struct kmem_cache
*s
)
2490 down_write(&slub_lock
);
2494 if (kmem_cache_close(s
)) {
2495 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2496 "still has objects.\n", s
->name
, __func__
);
2499 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2501 sysfs_slab_remove(s
);
2503 up_write(&slub_lock
);
2505 EXPORT_SYMBOL(kmem_cache_destroy
);
2507 /********************************************************************
2509 *******************************************************************/
2511 struct kmem_cache kmalloc_caches
[KMALLOC_CACHES
] __cacheline_aligned
;
2512 EXPORT_SYMBOL(kmalloc_caches
);
2514 static int __init
setup_slub_min_order(char *str
)
2516 get_option(&str
, &slub_min_order
);
2521 __setup("slub_min_order=", setup_slub_min_order
);
2523 static int __init
setup_slub_max_order(char *str
)
2525 get_option(&str
, &slub_max_order
);
2526 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2531 __setup("slub_max_order=", setup_slub_max_order
);
2533 static int __init
setup_slub_min_objects(char *str
)
2535 get_option(&str
, &slub_min_objects
);
2540 __setup("slub_min_objects=", setup_slub_min_objects
);
2542 static int __init
setup_slub_nomerge(char *str
)
2548 __setup("slub_nomerge", setup_slub_nomerge
);
2550 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2551 const char *name
, int size
, gfp_t gfp_flags
)
2553 unsigned int flags
= 0;
2555 if (gfp_flags
& SLUB_DMA
)
2556 flags
= SLAB_CACHE_DMA
;
2559 * This function is called with IRQs disabled during early-boot on
2560 * single CPU so there's no need to take slub_lock here.
2562 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2566 list_add(&s
->list
, &slab_caches
);
2568 if (sysfs_slab_add(s
))
2573 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2576 #ifdef CONFIG_ZONE_DMA
2577 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2579 static void sysfs_add_func(struct work_struct
*w
)
2581 struct kmem_cache
*s
;
2583 down_write(&slub_lock
);
2584 list_for_each_entry(s
, &slab_caches
, list
) {
2585 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2586 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2590 up_write(&slub_lock
);
2593 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2595 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2597 struct kmem_cache
*s
;
2600 unsigned long slabflags
;
2603 s
= kmalloc_caches_dma
[index
];
2607 /* Dynamically create dma cache */
2608 if (flags
& __GFP_WAIT
)
2609 down_write(&slub_lock
);
2611 if (!down_write_trylock(&slub_lock
))
2615 if (kmalloc_caches_dma
[index
])
2618 realsize
= kmalloc_caches
[index
].objsize
;
2619 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2620 (unsigned int)realsize
);
2623 for (i
= 0; i
< KMALLOC_CACHES
; i
++)
2624 if (!kmalloc_caches
[i
].size
)
2627 BUG_ON(i
>= KMALLOC_CACHES
);
2628 s
= kmalloc_caches
+ i
;
2631 * Must defer sysfs creation to a workqueue because we don't know
2632 * what context we are called from. Before sysfs comes up, we don't
2633 * need to do anything because our sysfs initcall will start by
2634 * adding all existing slabs to sysfs.
2636 slabflags
= SLAB_CACHE_DMA
|SLAB_NOTRACK
;
2637 if (slab_state
>= SYSFS
)
2638 slabflags
|= __SYSFS_ADD_DEFERRED
;
2640 if (!text
|| !kmem_cache_open(s
, flags
, text
,
2641 realsize
, ARCH_KMALLOC_MINALIGN
, slabflags
, NULL
)) {
2647 list_add(&s
->list
, &slab_caches
);
2648 kmalloc_caches_dma
[index
] = s
;
2650 if (slab_state
>= SYSFS
)
2651 schedule_work(&sysfs_add_work
);
2654 up_write(&slub_lock
);
2656 return kmalloc_caches_dma
[index
];
2661 * Conversion table for small slabs sizes / 8 to the index in the
2662 * kmalloc array. This is necessary for slabs < 192 since we have non power
2663 * of two cache sizes there. The size of larger slabs can be determined using
2666 static s8 size_index
[24] = {
2693 static inline int size_index_elem(size_t bytes
)
2695 return (bytes
- 1) / 8;
2698 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2704 return ZERO_SIZE_PTR
;
2706 index
= size_index
[size_index_elem(size
)];
2708 index
= fls(size
- 1);
2710 #ifdef CONFIG_ZONE_DMA
2711 if (unlikely((flags
& SLUB_DMA
)))
2712 return dma_kmalloc_cache(index
, flags
);
2715 return &kmalloc_caches
[index
];
2718 void *__kmalloc(size_t size
, gfp_t flags
)
2720 struct kmem_cache
*s
;
2723 if (unlikely(size
> SLUB_MAX_SIZE
))
2724 return kmalloc_large(size
, flags
);
2726 s
= get_slab(size
, flags
);
2728 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2731 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
2733 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2737 EXPORT_SYMBOL(__kmalloc
);
2739 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2744 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2745 page
= alloc_pages_node(node
, flags
, get_order(size
));
2747 ptr
= page_address(page
);
2749 kmemleak_alloc(ptr
, size
, 1, flags
);
2754 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2756 struct kmem_cache
*s
;
2759 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2760 ret
= kmalloc_large_node(size
, flags
, node
);
2762 trace_kmalloc_node(_RET_IP_
, ret
,
2763 size
, PAGE_SIZE
<< get_order(size
),
2769 s
= get_slab(size
, flags
);
2771 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2774 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2776 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2780 EXPORT_SYMBOL(__kmalloc_node
);
2783 size_t ksize(const void *object
)
2786 struct kmem_cache
*s
;
2788 if (unlikely(object
== ZERO_SIZE_PTR
))
2791 page
= virt_to_head_page(object
);
2793 if (unlikely(!PageSlab(page
))) {
2794 WARN_ON(!PageCompound(page
));
2795 return PAGE_SIZE
<< compound_order(page
);
2799 #ifdef CONFIG_SLUB_DEBUG
2801 * Debugging requires use of the padding between object
2802 * and whatever may come after it.
2804 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2809 * If we have the need to store the freelist pointer
2810 * back there or track user information then we can
2811 * only use the space before that information.
2813 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2816 * Else we can use all the padding etc for the allocation
2820 EXPORT_SYMBOL(ksize
);
2822 void kfree(const void *x
)
2825 void *object
= (void *)x
;
2827 trace_kfree(_RET_IP_
, x
);
2829 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2832 page
= virt_to_head_page(x
);
2833 if (unlikely(!PageSlab(page
))) {
2834 BUG_ON(!PageCompound(page
));
2839 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2841 EXPORT_SYMBOL(kfree
);
2844 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2845 * the remaining slabs by the number of items in use. The slabs with the
2846 * most items in use come first. New allocations will then fill those up
2847 * and thus they can be removed from the partial lists.
2849 * The slabs with the least items are placed last. This results in them
2850 * being allocated from last increasing the chance that the last objects
2851 * are freed in them.
2853 int kmem_cache_shrink(struct kmem_cache
*s
)
2857 struct kmem_cache_node
*n
;
2860 int objects
= oo_objects(s
->max
);
2861 struct list_head
*slabs_by_inuse
=
2862 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2863 unsigned long flags
;
2865 if (!slabs_by_inuse
)
2869 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2870 n
= get_node(s
, node
);
2875 for (i
= 0; i
< objects
; i
++)
2876 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2878 spin_lock_irqsave(&n
->list_lock
, flags
);
2881 * Build lists indexed by the items in use in each slab.
2883 * Note that concurrent frees may occur while we hold the
2884 * list_lock. page->inuse here is the upper limit.
2886 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2887 if (!page
->inuse
&& slab_trylock(page
)) {
2889 * Must hold slab lock here because slab_free
2890 * may have freed the last object and be
2891 * waiting to release the slab.
2893 list_del(&page
->lru
);
2896 discard_slab(s
, page
);
2898 list_move(&page
->lru
,
2899 slabs_by_inuse
+ page
->inuse
);
2904 * Rebuild the partial list with the slabs filled up most
2905 * first and the least used slabs at the end.
2907 for (i
= objects
- 1; i
>= 0; i
--)
2908 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2910 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2913 kfree(slabs_by_inuse
);
2916 EXPORT_SYMBOL(kmem_cache_shrink
);
2918 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2919 static int slab_mem_going_offline_callback(void *arg
)
2921 struct kmem_cache
*s
;
2923 down_read(&slub_lock
);
2924 list_for_each_entry(s
, &slab_caches
, list
)
2925 kmem_cache_shrink(s
);
2926 up_read(&slub_lock
);
2931 static void slab_mem_offline_callback(void *arg
)
2933 struct kmem_cache_node
*n
;
2934 struct kmem_cache
*s
;
2935 struct memory_notify
*marg
= arg
;
2938 offline_node
= marg
->status_change_nid
;
2941 * If the node still has available memory. we need kmem_cache_node
2944 if (offline_node
< 0)
2947 down_read(&slub_lock
);
2948 list_for_each_entry(s
, &slab_caches
, list
) {
2949 n
= get_node(s
, offline_node
);
2952 * if n->nr_slabs > 0, slabs still exist on the node
2953 * that is going down. We were unable to free them,
2954 * and offline_pages() function shouldn't call this
2955 * callback. So, we must fail.
2957 BUG_ON(slabs_node(s
, offline_node
));
2959 s
->node
[offline_node
] = NULL
;
2960 kmem_cache_free(kmalloc_caches
, n
);
2963 up_read(&slub_lock
);
2966 static int slab_mem_going_online_callback(void *arg
)
2968 struct kmem_cache_node
*n
;
2969 struct kmem_cache
*s
;
2970 struct memory_notify
*marg
= arg
;
2971 int nid
= marg
->status_change_nid
;
2975 * If the node's memory is already available, then kmem_cache_node is
2976 * already created. Nothing to do.
2982 * We are bringing a node online. No memory is available yet. We must
2983 * allocate a kmem_cache_node structure in order to bring the node
2986 down_read(&slub_lock
);
2987 list_for_each_entry(s
, &slab_caches
, list
) {
2989 * XXX: kmem_cache_alloc_node will fallback to other nodes
2990 * since memory is not yet available from the node that
2993 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2998 init_kmem_cache_node(n
, s
);
3002 up_read(&slub_lock
);
3006 static int slab_memory_callback(struct notifier_block
*self
,
3007 unsigned long action
, void *arg
)
3012 case MEM_GOING_ONLINE
:
3013 ret
= slab_mem_going_online_callback(arg
);
3015 case MEM_GOING_OFFLINE
:
3016 ret
= slab_mem_going_offline_callback(arg
);
3019 case MEM_CANCEL_ONLINE
:
3020 slab_mem_offline_callback(arg
);
3023 case MEM_CANCEL_OFFLINE
:
3027 ret
= notifier_from_errno(ret
);
3033 #endif /* CONFIG_MEMORY_HOTPLUG */
3035 /********************************************************************
3036 * Basic setup of slabs
3037 *******************************************************************/
3039 void __init
kmem_cache_init(void)
3046 * Must first have the slab cache available for the allocations of the
3047 * struct kmem_cache_node's. There is special bootstrap code in
3048 * kmem_cache_open for slab_state == DOWN.
3050 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3051 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3052 kmalloc_caches
[0].refcount
= -1;
3055 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3058 /* Able to allocate the per node structures */
3059 slab_state
= PARTIAL
;
3061 /* Caches that are not of the two-to-the-power-of size */
3062 if (KMALLOC_MIN_SIZE
<= 32) {
3063 create_kmalloc_cache(&kmalloc_caches
[1],
3064 "kmalloc-96", 96, GFP_NOWAIT
);
3067 if (KMALLOC_MIN_SIZE
<= 64) {
3068 create_kmalloc_cache(&kmalloc_caches
[2],
3069 "kmalloc-192", 192, GFP_NOWAIT
);
3073 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3074 create_kmalloc_cache(&kmalloc_caches
[i
],
3075 "kmalloc", 1 << i
, GFP_NOWAIT
);
3081 * Patch up the size_index table if we have strange large alignment
3082 * requirements for the kmalloc array. This is only the case for
3083 * MIPS it seems. The standard arches will not generate any code here.
3085 * Largest permitted alignment is 256 bytes due to the way we
3086 * handle the index determination for the smaller caches.
3088 * Make sure that nothing crazy happens if someone starts tinkering
3089 * around with ARCH_KMALLOC_MINALIGN
3091 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3092 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3094 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3095 int elem
= size_index_elem(i
);
3096 if (elem
>= ARRAY_SIZE(size_index
))
3098 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3101 if (KMALLOC_MIN_SIZE
== 64) {
3103 * The 96 byte size cache is not used if the alignment
3106 for (i
= 64 + 8; i
<= 96; i
+= 8)
3107 size_index
[size_index_elem(i
)] = 7;
3108 } else if (KMALLOC_MIN_SIZE
== 128) {
3110 * The 192 byte sized cache is not used if the alignment
3111 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3114 for (i
= 128 + 8; i
<= 192; i
+= 8)
3115 size_index
[size_index_elem(i
)] = 8;
3120 /* Provide the correct kmalloc names now that the caches are up */
3121 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3122 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3125 kmalloc_caches
[i
].name
= s
;
3129 register_cpu_notifier(&slab_notifier
);
3132 kmem_size
= offsetof(struct kmem_cache
, node
) +
3133 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3135 kmem_size
= sizeof(struct kmem_cache
);
3139 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3140 " CPUs=%d, Nodes=%d\n",
3141 caches
, cache_line_size(),
3142 slub_min_order
, slub_max_order
, slub_min_objects
,
3143 nr_cpu_ids
, nr_node_ids
);
3146 void __init
kmem_cache_init_late(void)
3151 * Find a mergeable slab cache
3153 static int slab_unmergeable(struct kmem_cache
*s
)
3155 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3162 * We may have set a slab to be unmergeable during bootstrap.
3164 if (s
->refcount
< 0)
3170 static struct kmem_cache
*find_mergeable(size_t size
,
3171 size_t align
, unsigned long flags
, const char *name
,
3172 void (*ctor
)(void *))
3174 struct kmem_cache
*s
;
3176 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3182 size
= ALIGN(size
, sizeof(void *));
3183 align
= calculate_alignment(flags
, align
, size
);
3184 size
= ALIGN(size
, align
);
3185 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3187 list_for_each_entry(s
, &slab_caches
, list
) {
3188 if (slab_unmergeable(s
))
3194 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3197 * Check if alignment is compatible.
3198 * Courtesy of Adrian Drzewiecki
3200 if ((s
->size
& ~(align
- 1)) != s
->size
)
3203 if (s
->size
- size
>= sizeof(void *))
3211 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3212 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3214 struct kmem_cache
*s
;
3219 down_write(&slub_lock
);
3220 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
);
3228 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3230 if (sysfs_slab_alias(s
, name
)) {
3234 up_write(&slub_lock
);
3238 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3240 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3241 size
, align
, flags
, ctor
)) {
3242 list_add(&s
->list
, &slab_caches
);
3243 if (sysfs_slab_add(s
)) {
3248 up_write(&slub_lock
);
3253 up_write(&slub_lock
);
3256 if (flags
& SLAB_PANIC
)
3257 panic("Cannot create slabcache %s\n", name
);
3262 EXPORT_SYMBOL(kmem_cache_create
);
3266 * Use the cpu notifier to insure that the cpu slabs are flushed when
3269 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3270 unsigned long action
, void *hcpu
)
3272 long cpu
= (long)hcpu
;
3273 struct kmem_cache
*s
;
3274 unsigned long flags
;
3277 case CPU_UP_CANCELED
:
3278 case CPU_UP_CANCELED_FROZEN
:
3280 case CPU_DEAD_FROZEN
:
3281 down_read(&slub_lock
);
3282 list_for_each_entry(s
, &slab_caches
, list
) {
3283 local_irq_save(flags
);
3284 __flush_cpu_slab(s
, cpu
);
3285 local_irq_restore(flags
);
3287 up_read(&slub_lock
);
3295 static struct notifier_block __cpuinitdata slab_notifier
= {
3296 .notifier_call
= slab_cpuup_callback
3301 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3303 struct kmem_cache
*s
;
3306 if (unlikely(size
> SLUB_MAX_SIZE
))
3307 return kmalloc_large(size
, gfpflags
);
3309 s
= get_slab(size
, gfpflags
);
3311 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3314 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
3316 /* Honor the call site pointer we recieved. */
3317 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3322 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3323 int node
, unsigned long caller
)
3325 struct kmem_cache
*s
;
3328 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3329 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3331 trace_kmalloc_node(caller
, ret
,
3332 size
, PAGE_SIZE
<< get_order(size
),
3338 s
= get_slab(size
, gfpflags
);
3340 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3343 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3345 /* Honor the call site pointer we recieved. */
3346 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3351 #ifdef CONFIG_SLUB_DEBUG
3352 static int count_inuse(struct page
*page
)
3357 static int count_total(struct page
*page
)
3359 return page
->objects
;
3362 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3366 void *addr
= page_address(page
);
3368 if (!check_slab(s
, page
) ||
3369 !on_freelist(s
, page
, NULL
))
3372 /* Now we know that a valid freelist exists */
3373 bitmap_zero(map
, page
->objects
);
3375 for_each_free_object(p
, s
, page
->freelist
) {
3376 set_bit(slab_index(p
, s
, addr
), map
);
3377 if (!check_object(s
, page
, p
, 0))
3381 for_each_object(p
, s
, addr
, page
->objects
)
3382 if (!test_bit(slab_index(p
, s
, addr
), map
))
3383 if (!check_object(s
, page
, p
, 1))
3388 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3391 if (slab_trylock(page
)) {
3392 validate_slab(s
, page
, map
);
3395 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3399 static int validate_slab_node(struct kmem_cache
*s
,
3400 struct kmem_cache_node
*n
, unsigned long *map
)
3402 unsigned long count
= 0;
3404 unsigned long flags
;
3406 spin_lock_irqsave(&n
->list_lock
, flags
);
3408 list_for_each_entry(page
, &n
->partial
, lru
) {
3409 validate_slab_slab(s
, page
, map
);
3412 if (count
!= n
->nr_partial
)
3413 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3414 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3416 if (!(s
->flags
& SLAB_STORE_USER
))
3419 list_for_each_entry(page
, &n
->full
, lru
) {
3420 validate_slab_slab(s
, page
, map
);
3423 if (count
!= atomic_long_read(&n
->nr_slabs
))
3424 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3425 "counter=%ld\n", s
->name
, count
,
3426 atomic_long_read(&n
->nr_slabs
));
3429 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3433 static long validate_slab_cache(struct kmem_cache
*s
)
3436 unsigned long count
= 0;
3437 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3438 sizeof(unsigned long), GFP_KERNEL
);
3444 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3445 struct kmem_cache_node
*n
= get_node(s
, node
);
3447 count
+= validate_slab_node(s
, n
, map
);
3453 #ifdef SLUB_RESILIENCY_TEST
3454 static void resiliency_test(void)
3458 printk(KERN_ERR
"SLUB resiliency testing\n");
3459 printk(KERN_ERR
"-----------------------\n");
3460 printk(KERN_ERR
"A. Corruption after allocation\n");
3462 p
= kzalloc(16, GFP_KERNEL
);
3464 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3465 " 0x12->0x%p\n\n", p
+ 16);
3467 validate_slab_cache(kmalloc_caches
+ 4);
3469 /* Hmmm... The next two are dangerous */
3470 p
= kzalloc(32, GFP_KERNEL
);
3471 p
[32 + sizeof(void *)] = 0x34;
3472 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3473 " 0x34 -> -0x%p\n", p
);
3475 "If allocated object is overwritten then not detectable\n\n");
3477 validate_slab_cache(kmalloc_caches
+ 5);
3478 p
= kzalloc(64, GFP_KERNEL
);
3479 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3481 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3484 "If allocated object is overwritten then not detectable\n\n");
3485 validate_slab_cache(kmalloc_caches
+ 6);
3487 printk(KERN_ERR
"\nB. Corruption after free\n");
3488 p
= kzalloc(128, GFP_KERNEL
);
3491 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3492 validate_slab_cache(kmalloc_caches
+ 7);
3494 p
= kzalloc(256, GFP_KERNEL
);
3497 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3499 validate_slab_cache(kmalloc_caches
+ 8);
3501 p
= kzalloc(512, GFP_KERNEL
);
3504 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3505 validate_slab_cache(kmalloc_caches
+ 9);
3508 static void resiliency_test(void) {};
3512 * Generate lists of code addresses where slabcache objects are allocated
3517 unsigned long count
;
3524 DECLARE_BITMAP(cpus
, NR_CPUS
);
3530 unsigned long count
;
3531 struct location
*loc
;
3534 static void free_loc_track(struct loc_track
*t
)
3537 free_pages((unsigned long)t
->loc
,
3538 get_order(sizeof(struct location
) * t
->max
));
3541 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3546 order
= get_order(sizeof(struct location
) * max
);
3548 l
= (void *)__get_free_pages(flags
, order
);
3553 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3561 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3562 const struct track
*track
)
3564 long start
, end
, pos
;
3566 unsigned long caddr
;
3567 unsigned long age
= jiffies
- track
->when
;
3573 pos
= start
+ (end
- start
+ 1) / 2;
3576 * There is nothing at "end". If we end up there
3577 * we need to add something to before end.
3582 caddr
= t
->loc
[pos
].addr
;
3583 if (track
->addr
== caddr
) {
3589 if (age
< l
->min_time
)
3591 if (age
> l
->max_time
)
3594 if (track
->pid
< l
->min_pid
)
3595 l
->min_pid
= track
->pid
;
3596 if (track
->pid
> l
->max_pid
)
3597 l
->max_pid
= track
->pid
;
3599 cpumask_set_cpu(track
->cpu
,
3600 to_cpumask(l
->cpus
));
3602 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3606 if (track
->addr
< caddr
)
3613 * Not found. Insert new tracking element.
3615 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3621 (t
->count
- pos
) * sizeof(struct location
));
3624 l
->addr
= track
->addr
;
3628 l
->min_pid
= track
->pid
;
3629 l
->max_pid
= track
->pid
;
3630 cpumask_clear(to_cpumask(l
->cpus
));
3631 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3632 nodes_clear(l
->nodes
);
3633 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3637 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3638 struct page
*page
, enum track_item alloc
,
3641 void *addr
= page_address(page
);
3644 bitmap_zero(map
, page
->objects
);
3645 for_each_free_object(p
, s
, page
->freelist
)
3646 set_bit(slab_index(p
, s
, addr
), map
);
3648 for_each_object(p
, s
, addr
, page
->objects
)
3649 if (!test_bit(slab_index(p
, s
, addr
), map
))
3650 add_location(t
, s
, get_track(s
, p
, alloc
));
3653 static int list_locations(struct kmem_cache
*s
, char *buf
,
3654 enum track_item alloc
)
3658 struct loc_track t
= { 0, 0, NULL
};
3660 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3661 sizeof(unsigned long), GFP_KERNEL
);
3663 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3666 return sprintf(buf
, "Out of memory\n");
3668 /* Push back cpu slabs */
3671 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3672 struct kmem_cache_node
*n
= get_node(s
, node
);
3673 unsigned long flags
;
3676 if (!atomic_long_read(&n
->nr_slabs
))
3679 spin_lock_irqsave(&n
->list_lock
, flags
);
3680 list_for_each_entry(page
, &n
->partial
, lru
)
3681 process_slab(&t
, s
, page
, alloc
, map
);
3682 list_for_each_entry(page
, &n
->full
, lru
)
3683 process_slab(&t
, s
, page
, alloc
, map
);
3684 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3687 for (i
= 0; i
< t
.count
; i
++) {
3688 struct location
*l
= &t
.loc
[i
];
3690 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3692 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3695 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3697 len
+= sprintf(buf
+ len
, "<not-available>");
3699 if (l
->sum_time
!= l
->min_time
) {
3700 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3702 (long)div_u64(l
->sum_time
, l
->count
),
3705 len
+= sprintf(buf
+ len
, " age=%ld",
3708 if (l
->min_pid
!= l
->max_pid
)
3709 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3710 l
->min_pid
, l
->max_pid
);
3712 len
+= sprintf(buf
+ len
, " pid=%ld",
3715 if (num_online_cpus() > 1 &&
3716 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3717 len
< PAGE_SIZE
- 60) {
3718 len
+= sprintf(buf
+ len
, " cpus=");
3719 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3720 to_cpumask(l
->cpus
));
3723 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3724 len
< PAGE_SIZE
- 60) {
3725 len
+= sprintf(buf
+ len
, " nodes=");
3726 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3730 len
+= sprintf(buf
+ len
, "\n");
3736 len
+= sprintf(buf
, "No data\n");
3740 enum slab_stat_type
{
3741 SL_ALL
, /* All slabs */
3742 SL_PARTIAL
, /* Only partially allocated slabs */
3743 SL_CPU
, /* Only slabs used for cpu caches */
3744 SL_OBJECTS
, /* Determine allocated objects not slabs */
3745 SL_TOTAL
/* Determine object capacity not slabs */
3748 #define SO_ALL (1 << SL_ALL)
3749 #define SO_PARTIAL (1 << SL_PARTIAL)
3750 #define SO_CPU (1 << SL_CPU)
3751 #define SO_OBJECTS (1 << SL_OBJECTS)
3752 #define SO_TOTAL (1 << SL_TOTAL)
3754 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3755 char *buf
, unsigned long flags
)
3757 unsigned long total
= 0;
3760 unsigned long *nodes
;
3761 unsigned long *per_cpu
;
3763 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3766 per_cpu
= nodes
+ nr_node_ids
;
3768 if (flags
& SO_CPU
) {
3771 for_each_possible_cpu(cpu
) {
3772 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3774 if (!c
|| c
->node
< 0)
3778 if (flags
& SO_TOTAL
)
3779 x
= c
->page
->objects
;
3780 else if (flags
& SO_OBJECTS
)
3786 nodes
[c
->node
] += x
;
3792 if (flags
& SO_ALL
) {
3793 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3794 struct kmem_cache_node
*n
= get_node(s
, node
);
3796 if (flags
& SO_TOTAL
)
3797 x
= atomic_long_read(&n
->total_objects
);
3798 else if (flags
& SO_OBJECTS
)
3799 x
= atomic_long_read(&n
->total_objects
) -
3800 count_partial(n
, count_free
);
3803 x
= atomic_long_read(&n
->nr_slabs
);
3808 } else if (flags
& SO_PARTIAL
) {
3809 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3810 struct kmem_cache_node
*n
= get_node(s
, node
);
3812 if (flags
& SO_TOTAL
)
3813 x
= count_partial(n
, count_total
);
3814 else if (flags
& SO_OBJECTS
)
3815 x
= count_partial(n
, count_inuse
);
3822 x
= sprintf(buf
, "%lu", total
);
3824 for_each_node_state(node
, N_NORMAL_MEMORY
)
3826 x
+= sprintf(buf
+ x
, " N%d=%lu",
3830 return x
+ sprintf(buf
+ x
, "\n");
3833 static int any_slab_objects(struct kmem_cache
*s
)
3837 for_each_online_node(node
) {
3838 struct kmem_cache_node
*n
= get_node(s
, node
);
3843 if (atomic_long_read(&n
->total_objects
))
3849 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3850 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3852 struct slab_attribute
{
3853 struct attribute attr
;
3854 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3855 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3858 #define SLAB_ATTR_RO(_name) \
3859 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3861 #define SLAB_ATTR(_name) \
3862 static struct slab_attribute _name##_attr = \
3863 __ATTR(_name, 0644, _name##_show, _name##_store)
3865 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3867 return sprintf(buf
, "%d\n", s
->size
);
3869 SLAB_ATTR_RO(slab_size
);
3871 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3873 return sprintf(buf
, "%d\n", s
->align
);
3875 SLAB_ATTR_RO(align
);
3877 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3879 return sprintf(buf
, "%d\n", s
->objsize
);
3881 SLAB_ATTR_RO(object_size
);
3883 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3885 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3887 SLAB_ATTR_RO(objs_per_slab
);
3889 static ssize_t
order_store(struct kmem_cache
*s
,
3890 const char *buf
, size_t length
)
3892 unsigned long order
;
3895 err
= strict_strtoul(buf
, 10, &order
);
3899 if (order
> slub_max_order
|| order
< slub_min_order
)
3902 calculate_sizes(s
, order
);
3906 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3908 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3912 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3914 return sprintf(buf
, "%lu\n", s
->min_partial
);
3917 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3923 err
= strict_strtoul(buf
, 10, &min
);
3927 set_min_partial(s
, min
);
3930 SLAB_ATTR(min_partial
);
3932 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3935 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3937 return n
+ sprintf(buf
+ n
, "\n");
3943 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3945 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3947 SLAB_ATTR_RO(aliases
);
3949 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3951 return show_slab_objects(s
, buf
, SO_ALL
);
3953 SLAB_ATTR_RO(slabs
);
3955 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3957 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3959 SLAB_ATTR_RO(partial
);
3961 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3963 return show_slab_objects(s
, buf
, SO_CPU
);
3965 SLAB_ATTR_RO(cpu_slabs
);
3967 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3969 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3971 SLAB_ATTR_RO(objects
);
3973 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3975 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3977 SLAB_ATTR_RO(objects_partial
);
3979 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3981 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3983 SLAB_ATTR_RO(total_objects
);
3985 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3987 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3990 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3991 const char *buf
, size_t length
)
3993 s
->flags
&= ~SLAB_DEBUG_FREE
;
3995 s
->flags
|= SLAB_DEBUG_FREE
;
3998 SLAB_ATTR(sanity_checks
);
4000 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4002 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4005 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4008 s
->flags
&= ~SLAB_TRACE
;
4010 s
->flags
|= SLAB_TRACE
;
4015 #ifdef CONFIG_FAILSLAB
4016 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4018 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4021 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4024 s
->flags
&= ~SLAB_FAILSLAB
;
4026 s
->flags
|= SLAB_FAILSLAB
;
4029 SLAB_ATTR(failslab
);
4032 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4034 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4037 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4038 const char *buf
, size_t length
)
4040 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4042 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4045 SLAB_ATTR(reclaim_account
);
4047 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4049 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4051 SLAB_ATTR_RO(hwcache_align
);
4053 #ifdef CONFIG_ZONE_DMA
4054 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4056 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4058 SLAB_ATTR_RO(cache_dma
);
4061 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4063 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4065 SLAB_ATTR_RO(destroy_by_rcu
);
4067 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4069 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4072 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4073 const char *buf
, size_t length
)
4075 if (any_slab_objects(s
))
4078 s
->flags
&= ~SLAB_RED_ZONE
;
4080 s
->flags
|= SLAB_RED_ZONE
;
4081 calculate_sizes(s
, -1);
4084 SLAB_ATTR(red_zone
);
4086 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4088 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4091 static ssize_t
poison_store(struct kmem_cache
*s
,
4092 const char *buf
, size_t length
)
4094 if (any_slab_objects(s
))
4097 s
->flags
&= ~SLAB_POISON
;
4099 s
->flags
|= SLAB_POISON
;
4100 calculate_sizes(s
, -1);
4105 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4107 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4110 static ssize_t
store_user_store(struct kmem_cache
*s
,
4111 const char *buf
, size_t length
)
4113 if (any_slab_objects(s
))
4116 s
->flags
&= ~SLAB_STORE_USER
;
4118 s
->flags
|= SLAB_STORE_USER
;
4119 calculate_sizes(s
, -1);
4122 SLAB_ATTR(store_user
);
4124 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4129 static ssize_t
validate_store(struct kmem_cache
*s
,
4130 const char *buf
, size_t length
)
4134 if (buf
[0] == '1') {
4135 ret
= validate_slab_cache(s
);
4141 SLAB_ATTR(validate
);
4143 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4148 static ssize_t
shrink_store(struct kmem_cache
*s
,
4149 const char *buf
, size_t length
)
4151 if (buf
[0] == '1') {
4152 int rc
= kmem_cache_shrink(s
);
4162 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4164 if (!(s
->flags
& SLAB_STORE_USER
))
4166 return list_locations(s
, buf
, TRACK_ALLOC
);
4168 SLAB_ATTR_RO(alloc_calls
);
4170 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4172 if (!(s
->flags
& SLAB_STORE_USER
))
4174 return list_locations(s
, buf
, TRACK_FREE
);
4176 SLAB_ATTR_RO(free_calls
);
4179 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4181 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4184 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4185 const char *buf
, size_t length
)
4187 unsigned long ratio
;
4190 err
= strict_strtoul(buf
, 10, &ratio
);
4195 s
->remote_node_defrag_ratio
= ratio
* 10;
4199 SLAB_ATTR(remote_node_defrag_ratio
);
4202 #ifdef CONFIG_SLUB_STATS
4203 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4205 unsigned long sum
= 0;
4208 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4213 for_each_online_cpu(cpu
) {
4214 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4220 len
= sprintf(buf
, "%lu", sum
);
4223 for_each_online_cpu(cpu
) {
4224 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4225 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4229 return len
+ sprintf(buf
+ len
, "\n");
4232 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4236 for_each_online_cpu(cpu
)
4237 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4240 #define STAT_ATTR(si, text) \
4241 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4243 return show_stat(s, buf, si); \
4245 static ssize_t text##_store(struct kmem_cache *s, \
4246 const char *buf, size_t length) \
4248 if (buf[0] != '0') \
4250 clear_stat(s, si); \
4255 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4256 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4257 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4258 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4259 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4260 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4261 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4262 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4263 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4264 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4265 STAT_ATTR(FREE_SLAB
, free_slab
);
4266 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4267 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4268 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4269 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4270 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4271 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4272 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4275 static struct attribute
*slab_attrs
[] = {
4276 &slab_size_attr
.attr
,
4277 &object_size_attr
.attr
,
4278 &objs_per_slab_attr
.attr
,
4280 &min_partial_attr
.attr
,
4282 &objects_partial_attr
.attr
,
4283 &total_objects_attr
.attr
,
4286 &cpu_slabs_attr
.attr
,
4290 &sanity_checks_attr
.attr
,
4292 &hwcache_align_attr
.attr
,
4293 &reclaim_account_attr
.attr
,
4294 &destroy_by_rcu_attr
.attr
,
4295 &red_zone_attr
.attr
,
4297 &store_user_attr
.attr
,
4298 &validate_attr
.attr
,
4300 &alloc_calls_attr
.attr
,
4301 &free_calls_attr
.attr
,
4302 #ifdef CONFIG_ZONE_DMA
4303 &cache_dma_attr
.attr
,
4306 &remote_node_defrag_ratio_attr
.attr
,
4308 #ifdef CONFIG_SLUB_STATS
4309 &alloc_fastpath_attr
.attr
,
4310 &alloc_slowpath_attr
.attr
,
4311 &free_fastpath_attr
.attr
,
4312 &free_slowpath_attr
.attr
,
4313 &free_frozen_attr
.attr
,
4314 &free_add_partial_attr
.attr
,
4315 &free_remove_partial_attr
.attr
,
4316 &alloc_from_partial_attr
.attr
,
4317 &alloc_slab_attr
.attr
,
4318 &alloc_refill_attr
.attr
,
4319 &free_slab_attr
.attr
,
4320 &cpuslab_flush_attr
.attr
,
4321 &deactivate_full_attr
.attr
,
4322 &deactivate_empty_attr
.attr
,
4323 &deactivate_to_head_attr
.attr
,
4324 &deactivate_to_tail_attr
.attr
,
4325 &deactivate_remote_frees_attr
.attr
,
4326 &order_fallback_attr
.attr
,
4328 #ifdef CONFIG_FAILSLAB
4329 &failslab_attr
.attr
,
4335 static struct attribute_group slab_attr_group
= {
4336 .attrs
= slab_attrs
,
4339 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4340 struct attribute
*attr
,
4343 struct slab_attribute
*attribute
;
4344 struct kmem_cache
*s
;
4347 attribute
= to_slab_attr(attr
);
4350 if (!attribute
->show
)
4353 err
= attribute
->show(s
, buf
);
4358 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4359 struct attribute
*attr
,
4360 const char *buf
, size_t len
)
4362 struct slab_attribute
*attribute
;
4363 struct kmem_cache
*s
;
4366 attribute
= to_slab_attr(attr
);
4369 if (!attribute
->store
)
4372 err
= attribute
->store(s
, buf
, len
);
4377 static void kmem_cache_release(struct kobject
*kobj
)
4379 struct kmem_cache
*s
= to_slab(kobj
);
4384 static const struct sysfs_ops slab_sysfs_ops
= {
4385 .show
= slab_attr_show
,
4386 .store
= slab_attr_store
,
4389 static struct kobj_type slab_ktype
= {
4390 .sysfs_ops
= &slab_sysfs_ops
,
4391 .release
= kmem_cache_release
4394 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4396 struct kobj_type
*ktype
= get_ktype(kobj
);
4398 if (ktype
== &slab_ktype
)
4403 static const struct kset_uevent_ops slab_uevent_ops
= {
4404 .filter
= uevent_filter
,
4407 static struct kset
*slab_kset
;
4409 #define ID_STR_LENGTH 64
4411 /* Create a unique string id for a slab cache:
4413 * Format :[flags-]size
4415 static char *create_unique_id(struct kmem_cache
*s
)
4417 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4424 * First flags affecting slabcache operations. We will only
4425 * get here for aliasable slabs so we do not need to support
4426 * too many flags. The flags here must cover all flags that
4427 * are matched during merging to guarantee that the id is
4430 if (s
->flags
& SLAB_CACHE_DMA
)
4432 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4434 if (s
->flags
& SLAB_DEBUG_FREE
)
4436 if (!(s
->flags
& SLAB_NOTRACK
))
4440 p
+= sprintf(p
, "%07d", s
->size
);
4441 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4445 static int sysfs_slab_add(struct kmem_cache
*s
)
4451 if (slab_state
< SYSFS
)
4452 /* Defer until later */
4455 unmergeable
= slab_unmergeable(s
);
4458 * Slabcache can never be merged so we can use the name proper.
4459 * This is typically the case for debug situations. In that
4460 * case we can catch duplicate names easily.
4462 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4466 * Create a unique name for the slab as a target
4469 name
= create_unique_id(s
);
4472 s
->kobj
.kset
= slab_kset
;
4473 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4475 kobject_put(&s
->kobj
);
4479 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4481 kobject_del(&s
->kobj
);
4482 kobject_put(&s
->kobj
);
4485 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4487 /* Setup first alias */
4488 sysfs_slab_alias(s
, s
->name
);
4494 static void sysfs_slab_remove(struct kmem_cache
*s
)
4496 if (slab_state
< SYSFS
)
4498 * Sysfs has not been setup yet so no need to remove the
4503 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4504 kobject_del(&s
->kobj
);
4505 kobject_put(&s
->kobj
);
4509 * Need to buffer aliases during bootup until sysfs becomes
4510 * available lest we lose that information.
4512 struct saved_alias
{
4513 struct kmem_cache
*s
;
4515 struct saved_alias
*next
;
4518 static struct saved_alias
*alias_list
;
4520 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4522 struct saved_alias
*al
;
4524 if (slab_state
== SYSFS
) {
4526 * If we have a leftover link then remove it.
4528 sysfs_remove_link(&slab_kset
->kobj
, name
);
4529 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4532 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4538 al
->next
= alias_list
;
4543 static int __init
slab_sysfs_init(void)
4545 struct kmem_cache
*s
;
4548 down_write(&slub_lock
);
4550 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4552 up_write(&slub_lock
);
4553 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4559 list_for_each_entry(s
, &slab_caches
, list
) {
4560 err
= sysfs_slab_add(s
);
4562 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4563 " to sysfs\n", s
->name
);
4566 while (alias_list
) {
4567 struct saved_alias
*al
= alias_list
;
4569 alias_list
= alias_list
->next
;
4570 err
= sysfs_slab_alias(al
->s
, al
->name
);
4572 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4573 " %s to sysfs\n", s
->name
);
4577 up_write(&slub_lock
);
4582 __initcall(slab_sysfs_init
);
4586 * The /proc/slabinfo ABI
4588 #ifdef CONFIG_SLABINFO
4589 static void print_slabinfo_header(struct seq_file
*m
)
4591 seq_puts(m
, "slabinfo - version: 2.1\n");
4592 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4593 "<objperslab> <pagesperslab>");
4594 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4595 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4599 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4603 down_read(&slub_lock
);
4605 print_slabinfo_header(m
);
4607 return seq_list_start(&slab_caches
, *pos
);
4610 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4612 return seq_list_next(p
, &slab_caches
, pos
);
4615 static void s_stop(struct seq_file
*m
, void *p
)
4617 up_read(&slub_lock
);
4620 static int s_show(struct seq_file
*m
, void *p
)
4622 unsigned long nr_partials
= 0;
4623 unsigned long nr_slabs
= 0;
4624 unsigned long nr_inuse
= 0;
4625 unsigned long nr_objs
= 0;
4626 unsigned long nr_free
= 0;
4627 struct kmem_cache
*s
;
4630 s
= list_entry(p
, struct kmem_cache
, list
);
4632 for_each_online_node(node
) {
4633 struct kmem_cache_node
*n
= get_node(s
, node
);
4638 nr_partials
+= n
->nr_partial
;
4639 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4640 nr_objs
+= atomic_long_read(&n
->total_objects
);
4641 nr_free
+= count_partial(n
, count_free
);
4644 nr_inuse
= nr_objs
- nr_free
;
4646 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4647 nr_objs
, s
->size
, oo_objects(s
->oo
),
4648 (1 << oo_order(s
->oo
)));
4649 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4650 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4656 static const struct seq_operations slabinfo_op
= {
4663 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4665 return seq_open(file
, &slabinfo_op
);
4668 static const struct file_operations proc_slabinfo_operations
= {
4669 .open
= slabinfo_open
,
4671 .llseek
= seq_lseek
,
4672 .release
= seq_release
,
4675 static int __init
slab_proc_init(void)
4677 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4680 module_init(slab_proc_init
);
4681 #endif /* CONFIG_SLABINFO */