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 #ifdef CONFIG_SLUB_DEBUG
117 * Issues still to be resolved:
119 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
121 * - Variable sizing of the per node arrays
124 /* Enable to test recovery from slab corruption on boot */
125 #undef SLUB_RESILIENCY_TEST
128 * Mininum number of partial slabs. These will be left on the partial
129 * lists even if they are empty. kmem_cache_shrink may reclaim them.
131 #define MIN_PARTIAL 5
134 * Maximum number of desirable partial slabs.
135 * The existence of more partial slabs makes kmem_cache_shrink
136 * sort the partial list by the number of objects in the.
138 #define MAX_PARTIAL 10
140 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
141 SLAB_POISON | SLAB_STORE_USER)
144 * Debugging flags that require metadata to be stored in the slab. These get
145 * disabled when slub_debug=O is used and a cache's min order increases with
148 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
151 * Set of flags that will prevent slab merging
153 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
154 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE)
156 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
157 SLAB_CACHE_DMA | SLAB_NOTRACK)
159 #ifndef ARCH_KMALLOC_MINALIGN
160 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
163 #ifndef ARCH_SLAB_MINALIGN
164 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
168 #define OO_MASK ((1 << OO_SHIFT) - 1)
169 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
171 /* Internal SLUB flags */
172 #define __OBJECT_POISON 0x80000000 /* Poison object */
173 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
175 static int kmem_size
= sizeof(struct kmem_cache
);
178 static struct notifier_block slab_notifier
;
182 DOWN
, /* No slab functionality available */
183 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
184 UP
, /* Everything works but does not show up in sysfs */
188 /* A list of all slab caches on the system */
189 static DECLARE_RWSEM(slub_lock
);
190 static LIST_HEAD(slab_caches
);
193 * Tracking user of a slab.
196 unsigned long addr
; /* Called from address */
197 int cpu
; /* Was running on cpu */
198 int pid
; /* Pid context */
199 unsigned long when
; /* When did the operation occur */
202 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
204 #ifdef CONFIG_SLUB_DEBUG
205 static int sysfs_slab_add(struct kmem_cache
*);
206 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
207 static void sysfs_slab_remove(struct kmem_cache
*);
210 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
211 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
213 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
220 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
222 #ifdef CONFIG_SLUB_STATS
223 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
227 /********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
231 int slab_is_available(void)
233 return slab_state
>= UP
;
236 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
239 return s
->node
[node
];
241 return &s
->local_node
;
245 /* Verify that a pointer has an address that is valid within a slab page */
246 static inline int check_valid_pointer(struct kmem_cache
*s
,
247 struct page
*page
, const void *object
)
254 base
= page_address(page
);
255 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
256 (object
- base
) % s
->size
) {
263 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
265 return *(void **)(object
+ s
->offset
);
268 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
270 *(void **)(object
+ s
->offset
) = fp
;
273 /* Loop over all objects in a slab */
274 #define for_each_object(__p, __s, __addr, __objects) \
275 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
279 #define for_each_free_object(__p, __s, __free) \
280 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
282 /* Determine object index from a given position */
283 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
285 return (p
- addr
) / s
->size
;
288 static inline struct kmem_cache_order_objects
oo_make(int order
,
291 struct kmem_cache_order_objects x
= {
292 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
298 static inline int oo_order(struct kmem_cache_order_objects x
)
300 return x
.x
>> OO_SHIFT
;
303 static inline int oo_objects(struct kmem_cache_order_objects x
)
305 return x
.x
& OO_MASK
;
308 #ifdef CONFIG_SLUB_DEBUG
312 #ifdef CONFIG_SLUB_DEBUG_ON
313 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
315 static int slub_debug
;
318 static char *slub_debug_slabs
;
319 static int disable_higher_order_debug
;
324 static void print_section(char *text
, u8
*addr
, unsigned int length
)
332 for (i
= 0; i
< length
; i
++) {
334 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
337 printk(KERN_CONT
" %02x", addr
[i
]);
339 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
341 printk(KERN_CONT
" %s\n", ascii
);
348 printk(KERN_CONT
" ");
352 printk(KERN_CONT
" %s\n", ascii
);
356 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
357 enum track_item alloc
)
362 p
= object
+ s
->offset
+ sizeof(void *);
364 p
= object
+ s
->inuse
;
369 static void set_track(struct kmem_cache
*s
, void *object
,
370 enum track_item alloc
, unsigned long addr
)
372 struct track
*p
= get_track(s
, object
, alloc
);
376 p
->cpu
= smp_processor_id();
377 p
->pid
= current
->pid
;
380 memset(p
, 0, sizeof(struct track
));
383 static void init_tracking(struct kmem_cache
*s
, void *object
)
385 if (!(s
->flags
& SLAB_STORE_USER
))
388 set_track(s
, object
, TRACK_FREE
, 0UL);
389 set_track(s
, object
, TRACK_ALLOC
, 0UL);
392 static void print_track(const char *s
, struct track
*t
)
397 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
398 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
401 static void print_tracking(struct kmem_cache
*s
, void *object
)
403 if (!(s
->flags
& SLAB_STORE_USER
))
406 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
407 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
410 static void print_page_info(struct page
*page
)
412 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
413 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
417 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
423 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
425 printk(KERN_ERR
"========================================"
426 "=====================================\n");
427 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
428 printk(KERN_ERR
"----------------------------------------"
429 "-------------------------------------\n\n");
432 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
438 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
440 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
443 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
445 unsigned int off
; /* Offset of last byte */
446 u8
*addr
= page_address(page
);
448 print_tracking(s
, p
);
450 print_page_info(page
);
452 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
453 p
, p
- addr
, get_freepointer(s
, p
));
456 print_section("Bytes b4", p
- 16, 16);
458 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
460 if (s
->flags
& SLAB_RED_ZONE
)
461 print_section("Redzone", p
+ s
->objsize
,
462 s
->inuse
- s
->objsize
);
465 off
= s
->offset
+ sizeof(void *);
469 if (s
->flags
& SLAB_STORE_USER
)
470 off
+= 2 * sizeof(struct track
);
473 /* Beginning of the filler is the free pointer */
474 print_section("Padding", p
+ off
, s
->size
- off
);
479 static void object_err(struct kmem_cache
*s
, struct page
*page
,
480 u8
*object
, char *reason
)
482 slab_bug(s
, "%s", reason
);
483 print_trailer(s
, page
, object
);
486 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
492 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
494 slab_bug(s
, "%s", buf
);
495 print_page_info(page
);
499 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
503 if (s
->flags
& __OBJECT_POISON
) {
504 memset(p
, POISON_FREE
, s
->objsize
- 1);
505 p
[s
->objsize
- 1] = POISON_END
;
508 if (s
->flags
& SLAB_RED_ZONE
)
509 memset(p
+ s
->objsize
,
510 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
511 s
->inuse
- s
->objsize
);
514 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
517 if (*start
!= (u8
)value
)
525 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
526 void *from
, void *to
)
528 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
529 memset(from
, data
, to
- from
);
532 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
533 u8
*object
, char *what
,
534 u8
*start
, unsigned int value
, unsigned int bytes
)
539 fault
= check_bytes(start
, value
, bytes
);
544 while (end
> fault
&& end
[-1] == value
)
547 slab_bug(s
, "%s overwritten", what
);
548 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
549 fault
, end
- 1, fault
[0], value
);
550 print_trailer(s
, page
, object
);
552 restore_bytes(s
, what
, value
, fault
, end
);
560 * Bytes of the object to be managed.
561 * If the freepointer may overlay the object then the free
562 * pointer is the first word of the object.
564 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
567 * object + s->objsize
568 * Padding to reach word boundary. This is also used for Redzoning.
569 * Padding is extended by another word if Redzoning is enabled and
572 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
573 * 0xcc (RED_ACTIVE) for objects in use.
576 * Meta data starts here.
578 * A. Free pointer (if we cannot overwrite object on free)
579 * B. Tracking data for SLAB_STORE_USER
580 * C. Padding to reach required alignment boundary or at mininum
581 * one word if debugging is on to be able to detect writes
582 * before the word boundary.
584 * Padding is done using 0x5a (POISON_INUSE)
587 * Nothing is used beyond s->size.
589 * If slabcaches are merged then the objsize and inuse boundaries are mostly
590 * ignored. And therefore no slab options that rely on these boundaries
591 * may be used with merged slabcaches.
594 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
596 unsigned long off
= s
->inuse
; /* The end of info */
599 /* Freepointer is placed after the object. */
600 off
+= sizeof(void *);
602 if (s
->flags
& SLAB_STORE_USER
)
603 /* We also have user information there */
604 off
+= 2 * sizeof(struct track
);
609 return check_bytes_and_report(s
, page
, p
, "Object padding",
610 p
+ off
, POISON_INUSE
, s
->size
- off
);
613 /* Check the pad bytes at the end of a slab page */
614 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
622 if (!(s
->flags
& SLAB_POISON
))
625 start
= page_address(page
);
626 length
= (PAGE_SIZE
<< compound_order(page
));
627 end
= start
+ length
;
628 remainder
= length
% s
->size
;
632 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
635 while (end
> fault
&& end
[-1] == POISON_INUSE
)
638 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
639 print_section("Padding", end
- remainder
, remainder
);
641 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
645 static int check_object(struct kmem_cache
*s
, struct page
*page
,
646 void *object
, int active
)
649 u8
*endobject
= object
+ s
->objsize
;
651 if (s
->flags
& SLAB_RED_ZONE
) {
653 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
655 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
656 endobject
, red
, s
->inuse
- s
->objsize
))
659 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
660 check_bytes_and_report(s
, page
, p
, "Alignment padding",
661 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
665 if (s
->flags
& SLAB_POISON
) {
666 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
667 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
668 POISON_FREE
, s
->objsize
- 1) ||
669 !check_bytes_and_report(s
, page
, p
, "Poison",
670 p
+ s
->objsize
- 1, POISON_END
, 1)))
673 * check_pad_bytes cleans up on its own.
675 check_pad_bytes(s
, page
, p
);
678 if (!s
->offset
&& active
)
680 * Object and freepointer overlap. Cannot check
681 * freepointer while object is allocated.
685 /* Check free pointer validity */
686 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
687 object_err(s
, page
, p
, "Freepointer corrupt");
689 * No choice but to zap it and thus lose the remainder
690 * of the free objects in this slab. May cause
691 * another error because the object count is now wrong.
693 set_freepointer(s
, p
, NULL
);
699 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
703 VM_BUG_ON(!irqs_disabled());
705 if (!PageSlab(page
)) {
706 slab_err(s
, page
, "Not a valid slab page");
710 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
711 if (page
->objects
> maxobj
) {
712 slab_err(s
, page
, "objects %u > max %u",
713 s
->name
, page
->objects
, maxobj
);
716 if (page
->inuse
> page
->objects
) {
717 slab_err(s
, page
, "inuse %u > max %u",
718 s
->name
, page
->inuse
, page
->objects
);
721 /* Slab_pad_check fixes things up after itself */
722 slab_pad_check(s
, page
);
727 * Determine if a certain object on a page is on the freelist. Must hold the
728 * slab lock to guarantee that the chains are in a consistent state.
730 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
733 void *fp
= page
->freelist
;
735 unsigned long max_objects
;
737 while (fp
&& nr
<= page
->objects
) {
740 if (!check_valid_pointer(s
, page
, fp
)) {
742 object_err(s
, page
, object
,
743 "Freechain corrupt");
744 set_freepointer(s
, object
, NULL
);
747 slab_err(s
, page
, "Freepointer corrupt");
748 page
->freelist
= NULL
;
749 page
->inuse
= page
->objects
;
750 slab_fix(s
, "Freelist cleared");
756 fp
= get_freepointer(s
, object
);
760 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
761 if (max_objects
> MAX_OBJS_PER_PAGE
)
762 max_objects
= MAX_OBJS_PER_PAGE
;
764 if (page
->objects
!= max_objects
) {
765 slab_err(s
, page
, "Wrong number of objects. Found %d but "
766 "should be %d", page
->objects
, max_objects
);
767 page
->objects
= max_objects
;
768 slab_fix(s
, "Number of objects adjusted.");
770 if (page
->inuse
!= page
->objects
- nr
) {
771 slab_err(s
, page
, "Wrong object count. Counter is %d but "
772 "counted were %d", page
->inuse
, page
->objects
- nr
);
773 page
->inuse
= page
->objects
- nr
;
774 slab_fix(s
, "Object count adjusted.");
776 return search
== NULL
;
779 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
782 if (s
->flags
& SLAB_TRACE
) {
783 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
785 alloc
? "alloc" : "free",
790 print_section("Object", (void *)object
, s
->objsize
);
797 * Tracking of fully allocated slabs for debugging purposes.
799 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
801 spin_lock(&n
->list_lock
);
802 list_add(&page
->lru
, &n
->full
);
803 spin_unlock(&n
->list_lock
);
806 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
808 struct kmem_cache_node
*n
;
810 if (!(s
->flags
& SLAB_STORE_USER
))
813 n
= get_node(s
, page_to_nid(page
));
815 spin_lock(&n
->list_lock
);
816 list_del(&page
->lru
);
817 spin_unlock(&n
->list_lock
);
820 /* Tracking of the number of slabs for debugging purposes */
821 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
823 struct kmem_cache_node
*n
= get_node(s
, node
);
825 return atomic_long_read(&n
->nr_slabs
);
828 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
830 return atomic_long_read(&n
->nr_slabs
);
833 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
835 struct kmem_cache_node
*n
= get_node(s
, node
);
838 * May be called early in order to allocate a slab for the
839 * kmem_cache_node structure. Solve the chicken-egg
840 * dilemma by deferring the increment of the count during
841 * bootstrap (see early_kmem_cache_node_alloc).
843 if (!NUMA_BUILD
|| n
) {
844 atomic_long_inc(&n
->nr_slabs
);
845 atomic_long_add(objects
, &n
->total_objects
);
848 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
850 struct kmem_cache_node
*n
= get_node(s
, node
);
852 atomic_long_dec(&n
->nr_slabs
);
853 atomic_long_sub(objects
, &n
->total_objects
);
856 /* Object debug checks for alloc/free paths */
857 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
860 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
863 init_object(s
, object
, 0);
864 init_tracking(s
, object
);
867 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
868 void *object
, unsigned long addr
)
870 if (!check_slab(s
, page
))
873 if (!on_freelist(s
, page
, object
)) {
874 object_err(s
, page
, object
, "Object already allocated");
878 if (!check_valid_pointer(s
, page
, object
)) {
879 object_err(s
, page
, object
, "Freelist Pointer check fails");
883 if (!check_object(s
, page
, object
, 0))
886 /* Success perform special debug activities for allocs */
887 if (s
->flags
& SLAB_STORE_USER
)
888 set_track(s
, object
, TRACK_ALLOC
, addr
);
889 trace(s
, page
, object
, 1);
890 init_object(s
, object
, 1);
894 if (PageSlab(page
)) {
896 * If this is a slab page then lets do the best we can
897 * to avoid issues in the future. Marking all objects
898 * as used avoids touching the remaining objects.
900 slab_fix(s
, "Marking all objects used");
901 page
->inuse
= page
->objects
;
902 page
->freelist
= NULL
;
907 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
908 void *object
, unsigned long addr
)
910 if (!check_slab(s
, page
))
913 if (!check_valid_pointer(s
, page
, object
)) {
914 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
918 if (on_freelist(s
, page
, object
)) {
919 object_err(s
, page
, object
, "Object already free");
923 if (!check_object(s
, page
, object
, 1))
926 if (unlikely(s
!= page
->slab
)) {
927 if (!PageSlab(page
)) {
928 slab_err(s
, page
, "Attempt to free object(0x%p) "
929 "outside of slab", object
);
930 } else if (!page
->slab
) {
932 "SLUB <none>: no slab for object 0x%p.\n",
936 object_err(s
, page
, object
,
937 "page slab pointer corrupt.");
941 /* Special debug activities for freeing objects */
942 if (!PageSlubFrozen(page
) && !page
->freelist
)
943 remove_full(s
, page
);
944 if (s
->flags
& SLAB_STORE_USER
)
945 set_track(s
, object
, TRACK_FREE
, addr
);
946 trace(s
, page
, object
, 0);
947 init_object(s
, object
, 0);
951 slab_fix(s
, "Object at 0x%p not freed", object
);
955 static int __init
setup_slub_debug(char *str
)
957 slub_debug
= DEBUG_DEFAULT_FLAGS
;
958 if (*str
++ != '=' || !*str
)
960 * No options specified. Switch on full debugging.
966 * No options but restriction on slabs. This means full
967 * debugging for slabs matching a pattern.
971 if (tolower(*str
) == 'o') {
973 * Avoid enabling debugging on caches if its minimum order
974 * would increase as a result.
976 disable_higher_order_debug
= 1;
983 * Switch off all debugging measures.
988 * Determine which debug features should be switched on
990 for (; *str
&& *str
!= ','; str
++) {
991 switch (tolower(*str
)) {
993 slub_debug
|= SLAB_DEBUG_FREE
;
996 slub_debug
|= SLAB_RED_ZONE
;
999 slub_debug
|= SLAB_POISON
;
1002 slub_debug
|= SLAB_STORE_USER
;
1005 slub_debug
|= SLAB_TRACE
;
1008 printk(KERN_ERR
"slub_debug option '%c' "
1009 "unknown. skipped\n", *str
);
1015 slub_debug_slabs
= str
+ 1;
1020 __setup("slub_debug", setup_slub_debug
);
1022 static unsigned long kmem_cache_flags(unsigned long objsize
,
1023 unsigned long flags
, const char *name
,
1024 void (*ctor
)(void *))
1027 * Enable debugging if selected on the kernel commandline.
1029 if (slub_debug
&& (!slub_debug_slabs
||
1030 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1031 flags
|= slub_debug
;
1036 static inline void setup_object_debug(struct kmem_cache
*s
,
1037 struct page
*page
, void *object
) {}
1039 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1040 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1042 static inline int free_debug_processing(struct kmem_cache
*s
,
1043 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1045 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1047 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1048 void *object
, int active
) { return 1; }
1049 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1050 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1051 unsigned long flags
, const char *name
,
1052 void (*ctor
)(void *))
1056 #define slub_debug 0
1058 #define disable_higher_order_debug 0
1060 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1062 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1064 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1066 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1071 * Slab allocation and freeing
1073 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1074 struct kmem_cache_order_objects oo
)
1076 int order
= oo_order(oo
);
1078 flags
|= __GFP_NOTRACK
;
1081 return alloc_pages(flags
, order
);
1083 return alloc_pages_node(node
, flags
, order
);
1086 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1089 struct kmem_cache_order_objects oo
= s
->oo
;
1092 flags
|= s
->allocflags
;
1095 * Let the initial higher-order allocation fail under memory pressure
1096 * so we fall-back to the minimum order allocation.
1098 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1100 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1101 if (unlikely(!page
)) {
1104 * Allocation may have failed due to fragmentation.
1105 * Try a lower order alloc if possible
1107 page
= alloc_slab_page(flags
, node
, oo
);
1111 stat(s
, ORDER_FALLBACK
);
1114 if (kmemcheck_enabled
1115 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1116 int pages
= 1 << oo_order(oo
);
1118 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1121 * Objects from caches that have a constructor don't get
1122 * cleared when they're allocated, so we need to do it here.
1125 kmemcheck_mark_uninitialized_pages(page
, pages
);
1127 kmemcheck_mark_unallocated_pages(page
, pages
);
1130 page
->objects
= oo_objects(oo
);
1131 mod_zone_page_state(page_zone(page
),
1132 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1133 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1139 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1142 setup_object_debug(s
, page
, object
);
1143 if (unlikely(s
->ctor
))
1147 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1154 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1156 page
= allocate_slab(s
,
1157 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1161 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1163 page
->flags
|= 1 << PG_slab
;
1164 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1165 SLAB_STORE_USER
| SLAB_TRACE
))
1166 __SetPageSlubDebug(page
);
1168 start
= page_address(page
);
1170 if (unlikely(s
->flags
& SLAB_POISON
))
1171 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1174 for_each_object(p
, s
, start
, page
->objects
) {
1175 setup_object(s
, page
, last
);
1176 set_freepointer(s
, last
, p
);
1179 setup_object(s
, page
, last
);
1180 set_freepointer(s
, last
, NULL
);
1182 page
->freelist
= start
;
1188 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1190 int order
= compound_order(page
);
1191 int pages
= 1 << order
;
1193 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1196 slab_pad_check(s
, page
);
1197 for_each_object(p
, s
, page_address(page
),
1199 check_object(s
, page
, p
, 0);
1200 __ClearPageSlubDebug(page
);
1203 kmemcheck_free_shadow(page
, compound_order(page
));
1205 mod_zone_page_state(page_zone(page
),
1206 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1207 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1210 __ClearPageSlab(page
);
1211 reset_page_mapcount(page
);
1212 if (current
->reclaim_state
)
1213 current
->reclaim_state
->reclaimed_slab
+= pages
;
1214 __free_pages(page
, order
);
1217 static void rcu_free_slab(struct rcu_head
*h
)
1221 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1222 __free_slab(page
->slab
, page
);
1225 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1227 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1229 * RCU free overloads the RCU head over the LRU
1231 struct rcu_head
*head
= (void *)&page
->lru
;
1233 call_rcu(head
, rcu_free_slab
);
1235 __free_slab(s
, page
);
1238 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1240 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1245 * Per slab locking using the pagelock
1247 static __always_inline
void slab_lock(struct page
*page
)
1249 bit_spin_lock(PG_locked
, &page
->flags
);
1252 static __always_inline
void slab_unlock(struct page
*page
)
1254 __bit_spin_unlock(PG_locked
, &page
->flags
);
1257 static __always_inline
int slab_trylock(struct page
*page
)
1261 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1266 * Management of partially allocated slabs
1268 static void add_partial(struct kmem_cache_node
*n
,
1269 struct page
*page
, int tail
)
1271 spin_lock(&n
->list_lock
);
1274 list_add_tail(&page
->lru
, &n
->partial
);
1276 list_add(&page
->lru
, &n
->partial
);
1277 spin_unlock(&n
->list_lock
);
1280 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1282 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1284 spin_lock(&n
->list_lock
);
1285 list_del(&page
->lru
);
1287 spin_unlock(&n
->list_lock
);
1291 * Lock slab and remove from the partial list.
1293 * Must hold list_lock.
1295 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1298 if (slab_trylock(page
)) {
1299 list_del(&page
->lru
);
1301 __SetPageSlubFrozen(page
);
1308 * Try to allocate a partial slab from a specific node.
1310 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1315 * Racy check. If we mistakenly see no partial slabs then we
1316 * just allocate an empty slab. If we mistakenly try to get a
1317 * partial slab and there is none available then get_partials()
1320 if (!n
|| !n
->nr_partial
)
1323 spin_lock(&n
->list_lock
);
1324 list_for_each_entry(page
, &n
->partial
, lru
)
1325 if (lock_and_freeze_slab(n
, page
))
1329 spin_unlock(&n
->list_lock
);
1334 * Get a page from somewhere. Search in increasing NUMA distances.
1336 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1339 struct zonelist
*zonelist
;
1342 enum zone_type high_zoneidx
= gfp_zone(flags
);
1346 * The defrag ratio allows a configuration of the tradeoffs between
1347 * inter node defragmentation and node local allocations. A lower
1348 * defrag_ratio increases the tendency to do local allocations
1349 * instead of attempting to obtain partial slabs from other nodes.
1351 * If the defrag_ratio is set to 0 then kmalloc() always
1352 * returns node local objects. If the ratio is higher then kmalloc()
1353 * may return off node objects because partial slabs are obtained
1354 * from other nodes and filled up.
1356 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1357 * defrag_ratio = 1000) then every (well almost) allocation will
1358 * first attempt to defrag slab caches on other nodes. This means
1359 * scanning over all nodes to look for partial slabs which may be
1360 * expensive if we do it every time we are trying to find a slab
1361 * with available objects.
1363 if (!s
->remote_node_defrag_ratio
||
1364 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1367 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1368 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1369 struct kmem_cache_node
*n
;
1371 n
= get_node(s
, zone_to_nid(zone
));
1373 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1374 n
->nr_partial
> s
->min_partial
) {
1375 page
= get_partial_node(n
);
1385 * Get a partial page, lock it and return it.
1387 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1390 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1392 page
= get_partial_node(get_node(s
, searchnode
));
1393 if (page
|| (flags
& __GFP_THISNODE
))
1396 return get_any_partial(s
, flags
);
1400 * Move a page back to the lists.
1402 * Must be called with the slab lock held.
1404 * On exit the slab lock will have been dropped.
1406 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1408 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1410 __ClearPageSlubFrozen(page
);
1413 if (page
->freelist
) {
1414 add_partial(n
, page
, tail
);
1415 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1417 stat(s
, DEACTIVATE_FULL
);
1418 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1419 (s
->flags
& SLAB_STORE_USER
))
1424 stat(s
, DEACTIVATE_EMPTY
);
1425 if (n
->nr_partial
< s
->min_partial
) {
1427 * Adding an empty slab to the partial slabs in order
1428 * to avoid page allocator overhead. This slab needs
1429 * to come after the other slabs with objects in
1430 * so that the others get filled first. That way the
1431 * size of the partial list stays small.
1433 * kmem_cache_shrink can reclaim any empty slabs from
1436 add_partial(n
, page
, 1);
1441 discard_slab(s
, page
);
1447 * Remove the cpu slab
1449 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1451 struct page
*page
= c
->page
;
1455 stat(s
, DEACTIVATE_REMOTE_FREES
);
1457 * Merge cpu freelist into slab freelist. Typically we get here
1458 * because both freelists are empty. So this is unlikely
1461 while (unlikely(c
->freelist
)) {
1464 tail
= 0; /* Hot objects. Put the slab first */
1466 /* Retrieve object from cpu_freelist */
1467 object
= c
->freelist
;
1468 c
->freelist
= get_freepointer(s
, c
->freelist
);
1470 /* And put onto the regular freelist */
1471 set_freepointer(s
, object
, page
->freelist
);
1472 page
->freelist
= object
;
1476 unfreeze_slab(s
, page
, tail
);
1479 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1481 stat(s
, CPUSLAB_FLUSH
);
1483 deactivate_slab(s
, c
);
1489 * Called from IPI handler with interrupts disabled.
1491 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1493 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1495 if (likely(c
&& c
->page
))
1499 static void flush_cpu_slab(void *d
)
1501 struct kmem_cache
*s
= d
;
1503 __flush_cpu_slab(s
, smp_processor_id());
1506 static void flush_all(struct kmem_cache
*s
)
1508 on_each_cpu(flush_cpu_slab
, s
, 1);
1512 * Check if the objects in a per cpu structure fit numa
1513 * locality expectations.
1515 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1518 if (node
!= -1 && c
->node
!= node
)
1524 static int count_free(struct page
*page
)
1526 return page
->objects
- page
->inuse
;
1529 static unsigned long count_partial(struct kmem_cache_node
*n
,
1530 int (*get_count
)(struct page
*))
1532 unsigned long flags
;
1533 unsigned long x
= 0;
1536 spin_lock_irqsave(&n
->list_lock
, flags
);
1537 list_for_each_entry(page
, &n
->partial
, lru
)
1538 x
+= get_count(page
);
1539 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1543 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1545 #ifdef CONFIG_SLUB_DEBUG
1546 return atomic_long_read(&n
->total_objects
);
1552 static noinline
void
1553 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1558 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1560 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1561 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1562 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1564 if (oo_order(s
->min
) > get_order(s
->objsize
))
1565 printk(KERN_WARNING
" %s debugging increased min order, use "
1566 "slub_debug=O to disable.\n", s
->name
);
1568 for_each_online_node(node
) {
1569 struct kmem_cache_node
*n
= get_node(s
, node
);
1570 unsigned long nr_slabs
;
1571 unsigned long nr_objs
;
1572 unsigned long nr_free
;
1577 nr_free
= count_partial(n
, count_free
);
1578 nr_slabs
= node_nr_slabs(n
);
1579 nr_objs
= node_nr_objs(n
);
1582 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1583 node
, nr_slabs
, nr_objs
, nr_free
);
1588 * Slow path. The lockless freelist is empty or we need to perform
1591 * Interrupts are disabled.
1593 * Processing is still very fast if new objects have been freed to the
1594 * regular freelist. In that case we simply take over the regular freelist
1595 * as the lockless freelist and zap the regular freelist.
1597 * If that is not working then we fall back to the partial lists. We take the
1598 * first element of the freelist as the object to allocate now and move the
1599 * rest of the freelist to the lockless freelist.
1601 * And if we were unable to get a new slab from the partial slab lists then
1602 * we need to allocate a new slab. This is the slowest path since it involves
1603 * a call to the page allocator and the setup of a new slab.
1605 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1606 unsigned long addr
, struct kmem_cache_cpu
*c
)
1611 /* We handle __GFP_ZERO in the caller */
1612 gfpflags
&= ~__GFP_ZERO
;
1618 if (unlikely(!node_match(c
, node
)))
1621 stat(s
, ALLOC_REFILL
);
1624 object
= c
->page
->freelist
;
1625 if (unlikely(!object
))
1627 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1630 c
->freelist
= get_freepointer(s
, object
);
1631 c
->page
->inuse
= c
->page
->objects
;
1632 c
->page
->freelist
= NULL
;
1633 c
->node
= page_to_nid(c
->page
);
1635 slab_unlock(c
->page
);
1636 stat(s
, ALLOC_SLOWPATH
);
1640 deactivate_slab(s
, c
);
1643 new = get_partial(s
, gfpflags
, node
);
1646 stat(s
, ALLOC_FROM_PARTIAL
);
1650 if (gfpflags
& __GFP_WAIT
)
1653 new = new_slab(s
, gfpflags
, node
);
1655 if (gfpflags
& __GFP_WAIT
)
1656 local_irq_disable();
1659 c
= __this_cpu_ptr(s
->cpu_slab
);
1660 stat(s
, ALLOC_SLAB
);
1664 __SetPageSlubFrozen(new);
1668 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1669 slab_out_of_memory(s
, gfpflags
, node
);
1672 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1676 c
->page
->freelist
= get_freepointer(s
, object
);
1682 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1683 * have the fastpath folded into their functions. So no function call
1684 * overhead for requests that can be satisfied on the fastpath.
1686 * The fastpath works by first checking if the lockless freelist can be used.
1687 * If not then __slab_alloc is called for slow processing.
1689 * Otherwise we can simply pick the next object from the lockless free list.
1691 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1692 gfp_t gfpflags
, int node
, unsigned long addr
)
1695 struct kmem_cache_cpu
*c
;
1696 unsigned long flags
;
1698 gfpflags
&= gfp_allowed_mask
;
1700 lockdep_trace_alloc(gfpflags
);
1701 might_sleep_if(gfpflags
& __GFP_WAIT
);
1703 if (should_failslab(s
->objsize
, gfpflags
))
1706 local_irq_save(flags
);
1707 c
= __this_cpu_ptr(s
->cpu_slab
);
1708 object
= c
->freelist
;
1709 if (unlikely(!object
|| !node_match(c
, node
)))
1711 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1714 c
->freelist
= get_freepointer(s
, object
);
1715 stat(s
, ALLOC_FASTPATH
);
1717 local_irq_restore(flags
);
1719 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1720 memset(object
, 0, s
->objsize
);
1722 kmemcheck_slab_alloc(s
, gfpflags
, object
, s
->objsize
);
1723 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, gfpflags
);
1728 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1730 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1732 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1736 EXPORT_SYMBOL(kmem_cache_alloc
);
1738 #ifdef CONFIG_TRACING
1739 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1741 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1743 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1747 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1749 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1751 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1752 s
->objsize
, s
->size
, gfpflags
, node
);
1756 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1759 #ifdef CONFIG_TRACING
1760 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1764 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1766 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1770 * Slow patch handling. This may still be called frequently since objects
1771 * have a longer lifetime than the cpu slabs in most processing loads.
1773 * So we still attempt to reduce cache line usage. Just take the slab
1774 * lock and free the item. If there is no additional partial page
1775 * handling required then we can return immediately.
1777 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1778 void *x
, unsigned long addr
)
1781 void **object
= (void *)x
;
1783 stat(s
, FREE_SLOWPATH
);
1786 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1790 prior
= page
->freelist
;
1791 set_freepointer(s
, object
, prior
);
1792 page
->freelist
= object
;
1795 if (unlikely(PageSlubFrozen(page
))) {
1796 stat(s
, FREE_FROZEN
);
1800 if (unlikely(!page
->inuse
))
1804 * Objects left in the slab. If it was not on the partial list before
1807 if (unlikely(!prior
)) {
1808 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1809 stat(s
, FREE_ADD_PARTIAL
);
1819 * Slab still on the partial list.
1821 remove_partial(s
, page
);
1822 stat(s
, FREE_REMOVE_PARTIAL
);
1826 discard_slab(s
, page
);
1830 if (!free_debug_processing(s
, page
, x
, addr
))
1836 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1837 * can perform fastpath freeing without additional function calls.
1839 * The fastpath is only possible if we are freeing to the current cpu slab
1840 * of this processor. This typically the case if we have just allocated
1843 * If fastpath is not possible then fall back to __slab_free where we deal
1844 * with all sorts of special processing.
1846 static __always_inline
void slab_free(struct kmem_cache
*s
,
1847 struct page
*page
, void *x
, unsigned long addr
)
1849 void **object
= (void *)x
;
1850 struct kmem_cache_cpu
*c
;
1851 unsigned long flags
;
1853 kmemleak_free_recursive(x
, s
->flags
);
1854 local_irq_save(flags
);
1855 c
= __this_cpu_ptr(s
->cpu_slab
);
1856 kmemcheck_slab_free(s
, object
, s
->objsize
);
1857 debug_check_no_locks_freed(object
, s
->objsize
);
1858 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1859 debug_check_no_obj_freed(object
, s
->objsize
);
1860 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1861 set_freepointer(s
, object
, c
->freelist
);
1862 c
->freelist
= object
;
1863 stat(s
, FREE_FASTPATH
);
1865 __slab_free(s
, page
, x
, addr
);
1867 local_irq_restore(flags
);
1870 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1874 page
= virt_to_head_page(x
);
1876 slab_free(s
, page
, x
, _RET_IP_
);
1878 trace_kmem_cache_free(_RET_IP_
, x
);
1880 EXPORT_SYMBOL(kmem_cache_free
);
1882 /* Figure out on which slab page the object resides */
1883 static struct page
*get_object_page(const void *x
)
1885 struct page
*page
= virt_to_head_page(x
);
1887 if (!PageSlab(page
))
1894 * Object placement in a slab is made very easy because we always start at
1895 * offset 0. If we tune the size of the object to the alignment then we can
1896 * get the required alignment by putting one properly sized object after
1899 * Notice that the allocation order determines the sizes of the per cpu
1900 * caches. Each processor has always one slab available for allocations.
1901 * Increasing the allocation order reduces the number of times that slabs
1902 * must be moved on and off the partial lists and is therefore a factor in
1907 * Mininum / Maximum order of slab pages. This influences locking overhead
1908 * and slab fragmentation. A higher order reduces the number of partial slabs
1909 * and increases the number of allocations possible without having to
1910 * take the list_lock.
1912 static int slub_min_order
;
1913 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1914 static int slub_min_objects
;
1917 * Merge control. If this is set then no merging of slab caches will occur.
1918 * (Could be removed. This was introduced to pacify the merge skeptics.)
1920 static int slub_nomerge
;
1923 * Calculate the order of allocation given an slab object size.
1925 * The order of allocation has significant impact on performance and other
1926 * system components. Generally order 0 allocations should be preferred since
1927 * order 0 does not cause fragmentation in the page allocator. Larger objects
1928 * be problematic to put into order 0 slabs because there may be too much
1929 * unused space left. We go to a higher order if more than 1/16th of the slab
1932 * In order to reach satisfactory performance we must ensure that a minimum
1933 * number of objects is in one slab. Otherwise we may generate too much
1934 * activity on the partial lists which requires taking the list_lock. This is
1935 * less a concern for large slabs though which are rarely used.
1937 * slub_max_order specifies the order where we begin to stop considering the
1938 * number of objects in a slab as critical. If we reach slub_max_order then
1939 * we try to keep the page order as low as possible. So we accept more waste
1940 * of space in favor of a small page order.
1942 * Higher order allocations also allow the placement of more objects in a
1943 * slab and thereby reduce object handling overhead. If the user has
1944 * requested a higher mininum order then we start with that one instead of
1945 * the smallest order which will fit the object.
1947 static inline int slab_order(int size
, int min_objects
,
1948 int max_order
, int fract_leftover
)
1952 int min_order
= slub_min_order
;
1954 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1955 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1957 for (order
= max(min_order
,
1958 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1959 order
<= max_order
; order
++) {
1961 unsigned long slab_size
= PAGE_SIZE
<< order
;
1963 if (slab_size
< min_objects
* size
)
1966 rem
= slab_size
% size
;
1968 if (rem
<= slab_size
/ fract_leftover
)
1976 static inline int calculate_order(int size
)
1984 * Attempt to find best configuration for a slab. This
1985 * works by first attempting to generate a layout with
1986 * the best configuration and backing off gradually.
1988 * First we reduce the acceptable waste in a slab. Then
1989 * we reduce the minimum objects required in a slab.
1991 min_objects
= slub_min_objects
;
1993 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1994 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1995 min_objects
= min(min_objects
, max_objects
);
1997 while (min_objects
> 1) {
1999 while (fraction
>= 4) {
2000 order
= slab_order(size
, min_objects
,
2001 slub_max_order
, fraction
);
2002 if (order
<= slub_max_order
)
2010 * We were unable to place multiple objects in a slab. Now
2011 * lets see if we can place a single object there.
2013 order
= slab_order(size
, 1, slub_max_order
, 1);
2014 if (order
<= slub_max_order
)
2018 * Doh this slab cannot be placed using slub_max_order.
2020 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2021 if (order
< MAX_ORDER
)
2027 * Figure out what the alignment of the objects will be.
2029 static unsigned long calculate_alignment(unsigned long flags
,
2030 unsigned long align
, unsigned long size
)
2033 * If the user wants hardware cache aligned objects then follow that
2034 * suggestion if the object is sufficiently large.
2036 * The hardware cache alignment cannot override the specified
2037 * alignment though. If that is greater then use it.
2039 if (flags
& SLAB_HWCACHE_ALIGN
) {
2040 unsigned long ralign
= cache_line_size();
2041 while (size
<= ralign
/ 2)
2043 align
= max(align
, ralign
);
2046 if (align
< ARCH_SLAB_MINALIGN
)
2047 align
= ARCH_SLAB_MINALIGN
;
2049 return ALIGN(align
, sizeof(void *));
2053 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2056 spin_lock_init(&n
->list_lock
);
2057 INIT_LIST_HEAD(&n
->partial
);
2058 #ifdef CONFIG_SLUB_DEBUG
2059 atomic_long_set(&n
->nr_slabs
, 0);
2060 atomic_long_set(&n
->total_objects
, 0);
2061 INIT_LIST_HEAD(&n
->full
);
2065 static DEFINE_PER_CPU(struct kmem_cache_cpu
, kmalloc_percpu
[KMALLOC_CACHES
]);
2067 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2069 if (s
< kmalloc_caches
+ KMALLOC_CACHES
&& s
>= kmalloc_caches
)
2071 * Boot time creation of the kmalloc array. Use static per cpu data
2072 * since the per cpu allocator is not available yet.
2074 s
->cpu_slab
= per_cpu_var(kmalloc_percpu
) + (s
- kmalloc_caches
);
2076 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2086 * No kmalloc_node yet so do it by hand. We know that this is the first
2087 * slab on the node for this slabcache. There are no concurrent accesses
2090 * Note that this function only works on the kmalloc_node_cache
2091 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2092 * memory on a fresh node that has no slab structures yet.
2094 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2097 struct kmem_cache_node
*n
;
2098 unsigned long flags
;
2100 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2102 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2105 if (page_to_nid(page
) != node
) {
2106 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2108 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2109 "in order to be able to continue\n");
2114 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2116 kmalloc_caches
->node
[node
] = n
;
2117 #ifdef CONFIG_SLUB_DEBUG
2118 init_object(kmalloc_caches
, n
, 1);
2119 init_tracking(kmalloc_caches
, n
);
2121 init_kmem_cache_node(n
, kmalloc_caches
);
2122 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2125 * lockdep requires consistent irq usage for each lock
2126 * so even though there cannot be a race this early in
2127 * the boot sequence, we still disable irqs.
2129 local_irq_save(flags
);
2130 add_partial(n
, page
, 0);
2131 local_irq_restore(flags
);
2134 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2138 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2139 struct kmem_cache_node
*n
= s
->node
[node
];
2140 if (n
&& n
!= &s
->local_node
)
2141 kmem_cache_free(kmalloc_caches
, n
);
2142 s
->node
[node
] = NULL
;
2146 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2151 if (slab_state
>= UP
&& (s
< kmalloc_caches
||
2152 s
> kmalloc_caches
+ KMALLOC_CACHES
))
2153 local_node
= page_to_nid(virt_to_page(s
));
2157 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2158 struct kmem_cache_node
*n
;
2160 if (local_node
== node
)
2163 if (slab_state
== DOWN
) {
2164 early_kmem_cache_node_alloc(gfpflags
, node
);
2167 n
= kmem_cache_alloc_node(kmalloc_caches
,
2171 free_kmem_cache_nodes(s
);
2177 init_kmem_cache_node(n
, s
);
2182 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2186 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2188 init_kmem_cache_node(&s
->local_node
, s
);
2193 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2195 if (min
< MIN_PARTIAL
)
2197 else if (min
> MAX_PARTIAL
)
2199 s
->min_partial
= min
;
2203 * calculate_sizes() determines the order and the distribution of data within
2206 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2208 unsigned long flags
= s
->flags
;
2209 unsigned long size
= s
->objsize
;
2210 unsigned long align
= s
->align
;
2214 * Round up object size to the next word boundary. We can only
2215 * place the free pointer at word boundaries and this determines
2216 * the possible location of the free pointer.
2218 size
= ALIGN(size
, sizeof(void *));
2220 #ifdef CONFIG_SLUB_DEBUG
2222 * Determine if we can poison the object itself. If the user of
2223 * the slab may touch the object after free or before allocation
2224 * then we should never poison the object itself.
2226 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2228 s
->flags
|= __OBJECT_POISON
;
2230 s
->flags
&= ~__OBJECT_POISON
;
2234 * If we are Redzoning then check if there is some space between the
2235 * end of the object and the free pointer. If not then add an
2236 * additional word to have some bytes to store Redzone information.
2238 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2239 size
+= sizeof(void *);
2243 * With that we have determined the number of bytes in actual use
2244 * by the object. This is the potential offset to the free pointer.
2248 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2251 * Relocate free pointer after the object if it is not
2252 * permitted to overwrite the first word of the object on
2255 * This is the case if we do RCU, have a constructor or
2256 * destructor or are poisoning the objects.
2259 size
+= sizeof(void *);
2262 #ifdef CONFIG_SLUB_DEBUG
2263 if (flags
& SLAB_STORE_USER
)
2265 * Need to store information about allocs and frees after
2268 size
+= 2 * sizeof(struct track
);
2270 if (flags
& SLAB_RED_ZONE
)
2272 * Add some empty padding so that we can catch
2273 * overwrites from earlier objects rather than let
2274 * tracking information or the free pointer be
2275 * corrupted if a user writes before the start
2278 size
+= sizeof(void *);
2282 * Determine the alignment based on various parameters that the
2283 * user specified and the dynamic determination of cache line size
2286 align
= calculate_alignment(flags
, align
, s
->objsize
);
2290 * SLUB stores one object immediately after another beginning from
2291 * offset 0. In order to align the objects we have to simply size
2292 * each object to conform to the alignment.
2294 size
= ALIGN(size
, align
);
2296 if (forced_order
>= 0)
2297 order
= forced_order
;
2299 order
= calculate_order(size
);
2306 s
->allocflags
|= __GFP_COMP
;
2308 if (s
->flags
& SLAB_CACHE_DMA
)
2309 s
->allocflags
|= SLUB_DMA
;
2311 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2312 s
->allocflags
|= __GFP_RECLAIMABLE
;
2315 * Determine the number of objects per slab
2317 s
->oo
= oo_make(order
, size
);
2318 s
->min
= oo_make(get_order(size
), size
);
2319 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2322 return !!oo_objects(s
->oo
);
2326 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2327 const char *name
, size_t size
,
2328 size_t align
, unsigned long flags
,
2329 void (*ctor
)(void *))
2331 memset(s
, 0, kmem_size
);
2336 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2338 if (!calculate_sizes(s
, -1))
2340 if (disable_higher_order_debug
) {
2342 * Disable debugging flags that store metadata if the min slab
2345 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2346 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2348 if (!calculate_sizes(s
, -1))
2354 * The larger the object size is, the more pages we want on the partial
2355 * list to avoid pounding the page allocator excessively.
2357 set_min_partial(s
, ilog2(s
->size
));
2360 s
->remote_node_defrag_ratio
= 1000;
2362 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2365 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2368 free_kmem_cache_nodes(s
);
2370 if (flags
& SLAB_PANIC
)
2371 panic("Cannot create slab %s size=%lu realsize=%u "
2372 "order=%u offset=%u flags=%lx\n",
2373 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2379 * Check if a given pointer is valid
2381 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2385 page
= get_object_page(object
);
2387 if (!page
|| s
!= page
->slab
)
2388 /* No slab or wrong slab */
2391 if (!check_valid_pointer(s
, page
, object
))
2395 * We could also check if the object is on the slabs freelist.
2396 * But this would be too expensive and it seems that the main
2397 * purpose of kmem_ptr_valid() is to check if the object belongs
2398 * to a certain slab.
2402 EXPORT_SYMBOL(kmem_ptr_validate
);
2405 * Determine the size of a slab object
2407 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2411 EXPORT_SYMBOL(kmem_cache_size
);
2413 const char *kmem_cache_name(struct kmem_cache
*s
)
2417 EXPORT_SYMBOL(kmem_cache_name
);
2419 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2422 #ifdef CONFIG_SLUB_DEBUG
2423 void *addr
= page_address(page
);
2425 DECLARE_BITMAP(map
, page
->objects
);
2427 bitmap_zero(map
, page
->objects
);
2428 slab_err(s
, page
, "%s", text
);
2430 for_each_free_object(p
, s
, page
->freelist
)
2431 set_bit(slab_index(p
, s
, addr
), map
);
2433 for_each_object(p
, s
, addr
, page
->objects
) {
2435 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2436 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2438 print_tracking(s
, p
);
2446 * Attempt to free all partial slabs on a node.
2448 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2450 unsigned long flags
;
2451 struct page
*page
, *h
;
2453 spin_lock_irqsave(&n
->list_lock
, flags
);
2454 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2456 list_del(&page
->lru
);
2457 discard_slab(s
, page
);
2460 list_slab_objects(s
, page
,
2461 "Objects remaining on kmem_cache_close()");
2464 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2468 * Release all resources used by a slab cache.
2470 static inline int kmem_cache_close(struct kmem_cache
*s
)
2475 free_percpu(s
->cpu_slab
);
2476 /* Attempt to free all objects */
2477 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2478 struct kmem_cache_node
*n
= get_node(s
, node
);
2481 if (n
->nr_partial
|| slabs_node(s
, node
))
2484 free_kmem_cache_nodes(s
);
2489 * Close a cache and release the kmem_cache structure
2490 * (must be used for caches created using kmem_cache_create)
2492 void kmem_cache_destroy(struct kmem_cache
*s
)
2494 down_write(&slub_lock
);
2498 up_write(&slub_lock
);
2499 if (kmem_cache_close(s
)) {
2500 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2501 "still has objects.\n", s
->name
, __func__
);
2504 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2506 sysfs_slab_remove(s
);
2508 up_write(&slub_lock
);
2510 EXPORT_SYMBOL(kmem_cache_destroy
);
2512 /********************************************************************
2514 *******************************************************************/
2516 struct kmem_cache kmalloc_caches
[KMALLOC_CACHES
] __cacheline_aligned
;
2517 EXPORT_SYMBOL(kmalloc_caches
);
2519 static int __init
setup_slub_min_order(char *str
)
2521 get_option(&str
, &slub_min_order
);
2526 __setup("slub_min_order=", setup_slub_min_order
);
2528 static int __init
setup_slub_max_order(char *str
)
2530 get_option(&str
, &slub_max_order
);
2531 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2536 __setup("slub_max_order=", setup_slub_max_order
);
2538 static int __init
setup_slub_min_objects(char *str
)
2540 get_option(&str
, &slub_min_objects
);
2545 __setup("slub_min_objects=", setup_slub_min_objects
);
2547 static int __init
setup_slub_nomerge(char *str
)
2553 __setup("slub_nomerge", setup_slub_nomerge
);
2555 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2556 const char *name
, int size
, gfp_t gfp_flags
)
2558 unsigned int flags
= 0;
2560 if (gfp_flags
& SLUB_DMA
)
2561 flags
= SLAB_CACHE_DMA
;
2564 * This function is called with IRQs disabled during early-boot on
2565 * single CPU so there's no need to take slub_lock here.
2567 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2571 list_add(&s
->list
, &slab_caches
);
2573 if (sysfs_slab_add(s
))
2578 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2581 #ifdef CONFIG_ZONE_DMA
2582 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2584 static void sysfs_add_func(struct work_struct
*w
)
2586 struct kmem_cache
*s
;
2588 down_write(&slub_lock
);
2589 list_for_each_entry(s
, &slab_caches
, list
) {
2590 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2591 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2595 up_write(&slub_lock
);
2598 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2600 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2602 struct kmem_cache
*s
;
2605 unsigned long slabflags
;
2608 s
= kmalloc_caches_dma
[index
];
2612 /* Dynamically create dma cache */
2613 if (flags
& __GFP_WAIT
)
2614 down_write(&slub_lock
);
2616 if (!down_write_trylock(&slub_lock
))
2620 if (kmalloc_caches_dma
[index
])
2623 realsize
= kmalloc_caches
[index
].objsize
;
2624 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2625 (unsigned int)realsize
);
2628 for (i
= 0; i
< KMALLOC_CACHES
; i
++)
2629 if (!kmalloc_caches
[i
].size
)
2632 BUG_ON(i
>= KMALLOC_CACHES
);
2633 s
= kmalloc_caches
+ i
;
2636 * Must defer sysfs creation to a workqueue because we don't know
2637 * what context we are called from. Before sysfs comes up, we don't
2638 * need to do anything because our sysfs initcall will start by
2639 * adding all existing slabs to sysfs.
2641 slabflags
= SLAB_CACHE_DMA
|SLAB_NOTRACK
;
2642 if (slab_state
>= SYSFS
)
2643 slabflags
|= __SYSFS_ADD_DEFERRED
;
2645 if (!text
|| !kmem_cache_open(s
, flags
, text
,
2646 realsize
, ARCH_KMALLOC_MINALIGN
, slabflags
, NULL
)) {
2652 list_add(&s
->list
, &slab_caches
);
2653 kmalloc_caches_dma
[index
] = s
;
2655 if (slab_state
>= SYSFS
)
2656 schedule_work(&sysfs_add_work
);
2659 up_write(&slub_lock
);
2661 return kmalloc_caches_dma
[index
];
2666 * Conversion table for small slabs sizes / 8 to the index in the
2667 * kmalloc array. This is necessary for slabs < 192 since we have non power
2668 * of two cache sizes there. The size of larger slabs can be determined using
2671 static s8 size_index
[24] = {
2698 static inline int size_index_elem(size_t bytes
)
2700 return (bytes
- 1) / 8;
2703 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2709 return ZERO_SIZE_PTR
;
2711 index
= size_index
[size_index_elem(size
)];
2713 index
= fls(size
- 1);
2715 #ifdef CONFIG_ZONE_DMA
2716 if (unlikely((flags
& SLUB_DMA
)))
2717 return dma_kmalloc_cache(index
, flags
);
2720 return &kmalloc_caches
[index
];
2723 void *__kmalloc(size_t size
, gfp_t flags
)
2725 struct kmem_cache
*s
;
2728 if (unlikely(size
> SLUB_MAX_SIZE
))
2729 return kmalloc_large(size
, flags
);
2731 s
= get_slab(size
, flags
);
2733 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2736 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2738 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2742 EXPORT_SYMBOL(__kmalloc
);
2744 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2749 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2750 page
= alloc_pages_node(node
, flags
, get_order(size
));
2752 ptr
= page_address(page
);
2754 kmemleak_alloc(ptr
, size
, 1, flags
);
2759 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2761 struct kmem_cache
*s
;
2764 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2765 ret
= kmalloc_large_node(size
, flags
, node
);
2767 trace_kmalloc_node(_RET_IP_
, ret
,
2768 size
, PAGE_SIZE
<< get_order(size
),
2774 s
= get_slab(size
, flags
);
2776 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2779 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2781 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2785 EXPORT_SYMBOL(__kmalloc_node
);
2788 size_t ksize(const void *object
)
2791 struct kmem_cache
*s
;
2793 if (unlikely(object
== ZERO_SIZE_PTR
))
2796 page
= virt_to_head_page(object
);
2798 if (unlikely(!PageSlab(page
))) {
2799 WARN_ON(!PageCompound(page
));
2800 return PAGE_SIZE
<< compound_order(page
);
2804 #ifdef CONFIG_SLUB_DEBUG
2806 * Debugging requires use of the padding between object
2807 * and whatever may come after it.
2809 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2814 * If we have the need to store the freelist pointer
2815 * back there or track user information then we can
2816 * only use the space before that information.
2818 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2821 * Else we can use all the padding etc for the allocation
2825 EXPORT_SYMBOL(ksize
);
2827 void kfree(const void *x
)
2830 void *object
= (void *)x
;
2832 trace_kfree(_RET_IP_
, x
);
2834 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2837 page
= virt_to_head_page(x
);
2838 if (unlikely(!PageSlab(page
))) {
2839 BUG_ON(!PageCompound(page
));
2844 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2846 EXPORT_SYMBOL(kfree
);
2849 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2850 * the remaining slabs by the number of items in use. The slabs with the
2851 * most items in use come first. New allocations will then fill those up
2852 * and thus they can be removed from the partial lists.
2854 * The slabs with the least items are placed last. This results in them
2855 * being allocated from last increasing the chance that the last objects
2856 * are freed in them.
2858 int kmem_cache_shrink(struct kmem_cache
*s
)
2862 struct kmem_cache_node
*n
;
2865 int objects
= oo_objects(s
->max
);
2866 struct list_head
*slabs_by_inuse
=
2867 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2868 unsigned long flags
;
2870 if (!slabs_by_inuse
)
2874 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2875 n
= get_node(s
, node
);
2880 for (i
= 0; i
< objects
; i
++)
2881 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2883 spin_lock_irqsave(&n
->list_lock
, flags
);
2886 * Build lists indexed by the items in use in each slab.
2888 * Note that concurrent frees may occur while we hold the
2889 * list_lock. page->inuse here is the upper limit.
2891 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2892 if (!page
->inuse
&& slab_trylock(page
)) {
2894 * Must hold slab lock here because slab_free
2895 * may have freed the last object and be
2896 * waiting to release the slab.
2898 list_del(&page
->lru
);
2901 discard_slab(s
, page
);
2903 list_move(&page
->lru
,
2904 slabs_by_inuse
+ page
->inuse
);
2909 * Rebuild the partial list with the slabs filled up most
2910 * first and the least used slabs at the end.
2912 for (i
= objects
- 1; i
>= 0; i
--)
2913 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2915 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2918 kfree(slabs_by_inuse
);
2921 EXPORT_SYMBOL(kmem_cache_shrink
);
2923 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2924 static int slab_mem_going_offline_callback(void *arg
)
2926 struct kmem_cache
*s
;
2928 down_read(&slub_lock
);
2929 list_for_each_entry(s
, &slab_caches
, list
)
2930 kmem_cache_shrink(s
);
2931 up_read(&slub_lock
);
2936 static void slab_mem_offline_callback(void *arg
)
2938 struct kmem_cache_node
*n
;
2939 struct kmem_cache
*s
;
2940 struct memory_notify
*marg
= arg
;
2943 offline_node
= marg
->status_change_nid
;
2946 * If the node still has available memory. we need kmem_cache_node
2949 if (offline_node
< 0)
2952 down_read(&slub_lock
);
2953 list_for_each_entry(s
, &slab_caches
, list
) {
2954 n
= get_node(s
, offline_node
);
2957 * if n->nr_slabs > 0, slabs still exist on the node
2958 * that is going down. We were unable to free them,
2959 * and offline_pages() function shoudn't call this
2960 * callback. So, we must fail.
2962 BUG_ON(slabs_node(s
, offline_node
));
2964 s
->node
[offline_node
] = NULL
;
2965 kmem_cache_free(kmalloc_caches
, n
);
2968 up_read(&slub_lock
);
2971 static int slab_mem_going_online_callback(void *arg
)
2973 struct kmem_cache_node
*n
;
2974 struct kmem_cache
*s
;
2975 struct memory_notify
*marg
= arg
;
2976 int nid
= marg
->status_change_nid
;
2980 * If the node's memory is already available, then kmem_cache_node is
2981 * already created. Nothing to do.
2987 * We are bringing a node online. No memory is available yet. We must
2988 * allocate a kmem_cache_node structure in order to bring the node
2991 down_read(&slub_lock
);
2992 list_for_each_entry(s
, &slab_caches
, list
) {
2994 * XXX: kmem_cache_alloc_node will fallback to other nodes
2995 * since memory is not yet available from the node that
2998 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
3003 init_kmem_cache_node(n
, s
);
3007 up_read(&slub_lock
);
3011 static int slab_memory_callback(struct notifier_block
*self
,
3012 unsigned long action
, void *arg
)
3017 case MEM_GOING_ONLINE
:
3018 ret
= slab_mem_going_online_callback(arg
);
3020 case MEM_GOING_OFFLINE
:
3021 ret
= slab_mem_going_offline_callback(arg
);
3024 case MEM_CANCEL_ONLINE
:
3025 slab_mem_offline_callback(arg
);
3028 case MEM_CANCEL_OFFLINE
:
3032 ret
= notifier_from_errno(ret
);
3038 #endif /* CONFIG_MEMORY_HOTPLUG */
3040 /********************************************************************
3041 * Basic setup of slabs
3042 *******************************************************************/
3044 void __init
kmem_cache_init(void)
3051 * Must first have the slab cache available for the allocations of the
3052 * struct kmem_cache_node's. There is special bootstrap code in
3053 * kmem_cache_open for slab_state == DOWN.
3055 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3056 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3057 kmalloc_caches
[0].refcount
= -1;
3060 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3063 /* Able to allocate the per node structures */
3064 slab_state
= PARTIAL
;
3066 /* Caches that are not of the two-to-the-power-of size */
3067 if (KMALLOC_MIN_SIZE
<= 32) {
3068 create_kmalloc_cache(&kmalloc_caches
[1],
3069 "kmalloc-96", 96, GFP_NOWAIT
);
3072 if (KMALLOC_MIN_SIZE
<= 64) {
3073 create_kmalloc_cache(&kmalloc_caches
[2],
3074 "kmalloc-192", 192, GFP_NOWAIT
);
3078 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3079 create_kmalloc_cache(&kmalloc_caches
[i
],
3080 "kmalloc", 1 << i
, GFP_NOWAIT
);
3086 * Patch up the size_index table if we have strange large alignment
3087 * requirements for the kmalloc array. This is only the case for
3088 * MIPS it seems. The standard arches will not generate any code here.
3090 * Largest permitted alignment is 256 bytes due to the way we
3091 * handle the index determination for the smaller caches.
3093 * Make sure that nothing crazy happens if someone starts tinkering
3094 * around with ARCH_KMALLOC_MINALIGN
3096 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3097 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3099 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3100 int elem
= size_index_elem(i
);
3101 if (elem
>= ARRAY_SIZE(size_index
))
3103 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3106 if (KMALLOC_MIN_SIZE
== 64) {
3108 * The 96 byte size cache is not used if the alignment
3111 for (i
= 64 + 8; i
<= 96; i
+= 8)
3112 size_index
[size_index_elem(i
)] = 7;
3113 } else if (KMALLOC_MIN_SIZE
== 128) {
3115 * The 192 byte sized cache is not used if the alignment
3116 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3119 for (i
= 128 + 8; i
<= 192; i
+= 8)
3120 size_index
[size_index_elem(i
)] = 8;
3125 /* Provide the correct kmalloc names now that the caches are up */
3126 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3127 kmalloc_caches
[i
]. name
=
3128 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3131 register_cpu_notifier(&slab_notifier
);
3134 kmem_size
= offsetof(struct kmem_cache
, node
) +
3135 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3137 kmem_size
= sizeof(struct kmem_cache
);
3141 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3142 " CPUs=%d, Nodes=%d\n",
3143 caches
, cache_line_size(),
3144 slub_min_order
, slub_max_order
, slub_min_objects
,
3145 nr_cpu_ids
, nr_node_ids
);
3148 void __init
kmem_cache_init_late(void)
3153 * Find a mergeable slab cache
3155 static int slab_unmergeable(struct kmem_cache
*s
)
3157 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3164 * We may have set a slab to be unmergeable during bootstrap.
3166 if (s
->refcount
< 0)
3172 static struct kmem_cache
*find_mergeable(size_t size
,
3173 size_t align
, unsigned long flags
, const char *name
,
3174 void (*ctor
)(void *))
3176 struct kmem_cache
*s
;
3178 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3184 size
= ALIGN(size
, sizeof(void *));
3185 align
= calculate_alignment(flags
, align
, size
);
3186 size
= ALIGN(size
, align
);
3187 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3189 list_for_each_entry(s
, &slab_caches
, list
) {
3190 if (slab_unmergeable(s
))
3196 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3199 * Check if alignment is compatible.
3200 * Courtesy of Adrian Drzewiecki
3202 if ((s
->size
& ~(align
- 1)) != s
->size
)
3205 if (s
->size
- size
>= sizeof(void *))
3213 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3214 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3216 struct kmem_cache
*s
;
3221 down_write(&slub_lock
);
3222 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3226 * Adjust the object sizes so that we clear
3227 * the complete object on kzalloc.
3229 s
->objsize
= max(s
->objsize
, (int)size
);
3230 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3231 up_write(&slub_lock
);
3233 if (sysfs_slab_alias(s
, name
)) {
3234 down_write(&slub_lock
);
3236 up_write(&slub_lock
);
3242 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3244 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3245 size
, align
, flags
, ctor
)) {
3246 list_add(&s
->list
, &slab_caches
);
3247 up_write(&slub_lock
);
3248 if (sysfs_slab_add(s
)) {
3249 down_write(&slub_lock
);
3251 up_write(&slub_lock
);
3259 up_write(&slub_lock
);
3262 if (flags
& SLAB_PANIC
)
3263 panic("Cannot create slabcache %s\n", name
);
3268 EXPORT_SYMBOL(kmem_cache_create
);
3272 * Use the cpu notifier to insure that the cpu slabs are flushed when
3275 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3276 unsigned long action
, void *hcpu
)
3278 long cpu
= (long)hcpu
;
3279 struct kmem_cache
*s
;
3280 unsigned long flags
;
3283 case CPU_UP_CANCELED
:
3284 case CPU_UP_CANCELED_FROZEN
:
3286 case CPU_DEAD_FROZEN
:
3287 down_read(&slub_lock
);
3288 list_for_each_entry(s
, &slab_caches
, list
) {
3289 local_irq_save(flags
);
3290 __flush_cpu_slab(s
, cpu
);
3291 local_irq_restore(flags
);
3293 up_read(&slub_lock
);
3301 static struct notifier_block __cpuinitdata slab_notifier
= {
3302 .notifier_call
= slab_cpuup_callback
3307 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3309 struct kmem_cache
*s
;
3312 if (unlikely(size
> SLUB_MAX_SIZE
))
3313 return kmalloc_large(size
, gfpflags
);
3315 s
= get_slab(size
, gfpflags
);
3317 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3320 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3322 /* Honor the call site pointer we recieved. */
3323 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3328 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3329 int node
, unsigned long caller
)
3331 struct kmem_cache
*s
;
3334 if (unlikely(size
> SLUB_MAX_SIZE
))
3335 return kmalloc_large_node(size
, gfpflags
, node
);
3337 s
= get_slab(size
, gfpflags
);
3339 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3342 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3344 /* Honor the call site pointer we recieved. */
3345 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3350 #ifdef CONFIG_SLUB_DEBUG
3351 static int count_inuse(struct page
*page
)
3356 static int count_total(struct page
*page
)
3358 return page
->objects
;
3361 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3365 void *addr
= page_address(page
);
3367 if (!check_slab(s
, page
) ||
3368 !on_freelist(s
, page
, NULL
))
3371 /* Now we know that a valid freelist exists */
3372 bitmap_zero(map
, page
->objects
);
3374 for_each_free_object(p
, s
, page
->freelist
) {
3375 set_bit(slab_index(p
, s
, addr
), map
);
3376 if (!check_object(s
, page
, p
, 0))
3380 for_each_object(p
, s
, addr
, page
->objects
)
3381 if (!test_bit(slab_index(p
, s
, addr
), map
))
3382 if (!check_object(s
, page
, p
, 1))
3387 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3390 if (slab_trylock(page
)) {
3391 validate_slab(s
, page
, map
);
3394 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3397 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3398 if (!PageSlubDebug(page
))
3399 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3400 "on slab 0x%p\n", s
->name
, page
);
3402 if (PageSlubDebug(page
))
3403 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3404 "slab 0x%p\n", s
->name
, page
);
3408 static int validate_slab_node(struct kmem_cache
*s
,
3409 struct kmem_cache_node
*n
, unsigned long *map
)
3411 unsigned long count
= 0;
3413 unsigned long flags
;
3415 spin_lock_irqsave(&n
->list_lock
, flags
);
3417 list_for_each_entry(page
, &n
->partial
, lru
) {
3418 validate_slab_slab(s
, page
, map
);
3421 if (count
!= n
->nr_partial
)
3422 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3423 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3425 if (!(s
->flags
& SLAB_STORE_USER
))
3428 list_for_each_entry(page
, &n
->full
, lru
) {
3429 validate_slab_slab(s
, page
, map
);
3432 if (count
!= atomic_long_read(&n
->nr_slabs
))
3433 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3434 "counter=%ld\n", s
->name
, count
,
3435 atomic_long_read(&n
->nr_slabs
));
3438 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3442 static long validate_slab_cache(struct kmem_cache
*s
)
3445 unsigned long count
= 0;
3446 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3447 sizeof(unsigned long), GFP_KERNEL
);
3453 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3454 struct kmem_cache_node
*n
= get_node(s
, node
);
3456 count
+= validate_slab_node(s
, n
, map
);
3462 #ifdef SLUB_RESILIENCY_TEST
3463 static void resiliency_test(void)
3467 printk(KERN_ERR
"SLUB resiliency testing\n");
3468 printk(KERN_ERR
"-----------------------\n");
3469 printk(KERN_ERR
"A. Corruption after allocation\n");
3471 p
= kzalloc(16, GFP_KERNEL
);
3473 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3474 " 0x12->0x%p\n\n", p
+ 16);
3476 validate_slab_cache(kmalloc_caches
+ 4);
3478 /* Hmmm... The next two are dangerous */
3479 p
= kzalloc(32, GFP_KERNEL
);
3480 p
[32 + sizeof(void *)] = 0x34;
3481 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3482 " 0x34 -> -0x%p\n", p
);
3484 "If allocated object is overwritten then not detectable\n\n");
3486 validate_slab_cache(kmalloc_caches
+ 5);
3487 p
= kzalloc(64, GFP_KERNEL
);
3488 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3490 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3493 "If allocated object is overwritten then not detectable\n\n");
3494 validate_slab_cache(kmalloc_caches
+ 6);
3496 printk(KERN_ERR
"\nB. Corruption after free\n");
3497 p
= kzalloc(128, GFP_KERNEL
);
3500 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3501 validate_slab_cache(kmalloc_caches
+ 7);
3503 p
= kzalloc(256, GFP_KERNEL
);
3506 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3508 validate_slab_cache(kmalloc_caches
+ 8);
3510 p
= kzalloc(512, GFP_KERNEL
);
3513 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3514 validate_slab_cache(kmalloc_caches
+ 9);
3517 static void resiliency_test(void) {};
3521 * Generate lists of code addresses where slabcache objects are allocated
3526 unsigned long count
;
3533 DECLARE_BITMAP(cpus
, NR_CPUS
);
3539 unsigned long count
;
3540 struct location
*loc
;
3543 static void free_loc_track(struct loc_track
*t
)
3546 free_pages((unsigned long)t
->loc
,
3547 get_order(sizeof(struct location
) * t
->max
));
3550 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3555 order
= get_order(sizeof(struct location
) * max
);
3557 l
= (void *)__get_free_pages(flags
, order
);
3562 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3570 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3571 const struct track
*track
)
3573 long start
, end
, pos
;
3575 unsigned long caddr
;
3576 unsigned long age
= jiffies
- track
->when
;
3582 pos
= start
+ (end
- start
+ 1) / 2;
3585 * There is nothing at "end". If we end up there
3586 * we need to add something to before end.
3591 caddr
= t
->loc
[pos
].addr
;
3592 if (track
->addr
== caddr
) {
3598 if (age
< l
->min_time
)
3600 if (age
> l
->max_time
)
3603 if (track
->pid
< l
->min_pid
)
3604 l
->min_pid
= track
->pid
;
3605 if (track
->pid
> l
->max_pid
)
3606 l
->max_pid
= track
->pid
;
3608 cpumask_set_cpu(track
->cpu
,
3609 to_cpumask(l
->cpus
));
3611 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3615 if (track
->addr
< caddr
)
3622 * Not found. Insert new tracking element.
3624 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3630 (t
->count
- pos
) * sizeof(struct location
));
3633 l
->addr
= track
->addr
;
3637 l
->min_pid
= track
->pid
;
3638 l
->max_pid
= track
->pid
;
3639 cpumask_clear(to_cpumask(l
->cpus
));
3640 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3641 nodes_clear(l
->nodes
);
3642 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3646 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3647 struct page
*page
, enum track_item alloc
)
3649 void *addr
= page_address(page
);
3650 DECLARE_BITMAP(map
, page
->objects
);
3653 bitmap_zero(map
, page
->objects
);
3654 for_each_free_object(p
, s
, page
->freelist
)
3655 set_bit(slab_index(p
, s
, addr
), map
);
3657 for_each_object(p
, s
, addr
, page
->objects
)
3658 if (!test_bit(slab_index(p
, s
, addr
), map
))
3659 add_location(t
, s
, get_track(s
, p
, alloc
));
3662 static int list_locations(struct kmem_cache
*s
, char *buf
,
3663 enum track_item alloc
)
3667 struct loc_track t
= { 0, 0, NULL
};
3670 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3672 return sprintf(buf
, "Out of memory\n");
3674 /* Push back cpu slabs */
3677 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3678 struct kmem_cache_node
*n
= get_node(s
, node
);
3679 unsigned long flags
;
3682 if (!atomic_long_read(&n
->nr_slabs
))
3685 spin_lock_irqsave(&n
->list_lock
, flags
);
3686 list_for_each_entry(page
, &n
->partial
, lru
)
3687 process_slab(&t
, s
, page
, alloc
);
3688 list_for_each_entry(page
, &n
->full
, lru
)
3689 process_slab(&t
, s
, page
, alloc
);
3690 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3693 for (i
= 0; i
< t
.count
; i
++) {
3694 struct location
*l
= &t
.loc
[i
];
3696 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3698 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3701 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3703 len
+= sprintf(buf
+ len
, "<not-available>");
3705 if (l
->sum_time
!= l
->min_time
) {
3706 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3708 (long)div_u64(l
->sum_time
, l
->count
),
3711 len
+= sprintf(buf
+ len
, " age=%ld",
3714 if (l
->min_pid
!= l
->max_pid
)
3715 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3716 l
->min_pid
, l
->max_pid
);
3718 len
+= sprintf(buf
+ len
, " pid=%ld",
3721 if (num_online_cpus() > 1 &&
3722 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3723 len
< PAGE_SIZE
- 60) {
3724 len
+= sprintf(buf
+ len
, " cpus=");
3725 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3726 to_cpumask(l
->cpus
));
3729 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3730 len
< PAGE_SIZE
- 60) {
3731 len
+= sprintf(buf
+ len
, " nodes=");
3732 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3736 len
+= sprintf(buf
+ len
, "\n");
3741 len
+= sprintf(buf
, "No data\n");
3745 enum slab_stat_type
{
3746 SL_ALL
, /* All slabs */
3747 SL_PARTIAL
, /* Only partially allocated slabs */
3748 SL_CPU
, /* Only slabs used for cpu caches */
3749 SL_OBJECTS
, /* Determine allocated objects not slabs */
3750 SL_TOTAL
/* Determine object capacity not slabs */
3753 #define SO_ALL (1 << SL_ALL)
3754 #define SO_PARTIAL (1 << SL_PARTIAL)
3755 #define SO_CPU (1 << SL_CPU)
3756 #define SO_OBJECTS (1 << SL_OBJECTS)
3757 #define SO_TOTAL (1 << SL_TOTAL)
3759 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3760 char *buf
, unsigned long flags
)
3762 unsigned long total
= 0;
3765 unsigned long *nodes
;
3766 unsigned long *per_cpu
;
3768 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3771 per_cpu
= nodes
+ nr_node_ids
;
3773 if (flags
& SO_CPU
) {
3776 for_each_possible_cpu(cpu
) {
3777 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3779 if (!c
|| c
->node
< 0)
3783 if (flags
& SO_TOTAL
)
3784 x
= c
->page
->objects
;
3785 else if (flags
& SO_OBJECTS
)
3791 nodes
[c
->node
] += x
;
3797 if (flags
& SO_ALL
) {
3798 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3799 struct kmem_cache_node
*n
= get_node(s
, node
);
3801 if (flags
& SO_TOTAL
)
3802 x
= atomic_long_read(&n
->total_objects
);
3803 else if (flags
& SO_OBJECTS
)
3804 x
= atomic_long_read(&n
->total_objects
) -
3805 count_partial(n
, count_free
);
3808 x
= atomic_long_read(&n
->nr_slabs
);
3813 } else if (flags
& SO_PARTIAL
) {
3814 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3815 struct kmem_cache_node
*n
= get_node(s
, node
);
3817 if (flags
& SO_TOTAL
)
3818 x
= count_partial(n
, count_total
);
3819 else if (flags
& SO_OBJECTS
)
3820 x
= count_partial(n
, count_inuse
);
3827 x
= sprintf(buf
, "%lu", total
);
3829 for_each_node_state(node
, N_NORMAL_MEMORY
)
3831 x
+= sprintf(buf
+ x
, " N%d=%lu",
3835 return x
+ sprintf(buf
+ x
, "\n");
3838 static int any_slab_objects(struct kmem_cache
*s
)
3842 for_each_online_node(node
) {
3843 struct kmem_cache_node
*n
= get_node(s
, node
);
3848 if (atomic_long_read(&n
->total_objects
))
3854 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3855 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3857 struct slab_attribute
{
3858 struct attribute attr
;
3859 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3860 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3863 #define SLAB_ATTR_RO(_name) \
3864 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3866 #define SLAB_ATTR(_name) \
3867 static struct slab_attribute _name##_attr = \
3868 __ATTR(_name, 0644, _name##_show, _name##_store)
3870 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3872 return sprintf(buf
, "%d\n", s
->size
);
3874 SLAB_ATTR_RO(slab_size
);
3876 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3878 return sprintf(buf
, "%d\n", s
->align
);
3880 SLAB_ATTR_RO(align
);
3882 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3884 return sprintf(buf
, "%d\n", s
->objsize
);
3886 SLAB_ATTR_RO(object_size
);
3888 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3890 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3892 SLAB_ATTR_RO(objs_per_slab
);
3894 static ssize_t
order_store(struct kmem_cache
*s
,
3895 const char *buf
, size_t length
)
3897 unsigned long order
;
3900 err
= strict_strtoul(buf
, 10, &order
);
3904 if (order
> slub_max_order
|| order
< slub_min_order
)
3907 calculate_sizes(s
, order
);
3911 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3913 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3917 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3919 return sprintf(buf
, "%lu\n", s
->min_partial
);
3922 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3928 err
= strict_strtoul(buf
, 10, &min
);
3932 set_min_partial(s
, min
);
3935 SLAB_ATTR(min_partial
);
3937 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3940 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3942 return n
+ sprintf(buf
+ n
, "\n");
3948 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3950 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3952 SLAB_ATTR_RO(aliases
);
3954 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3956 return show_slab_objects(s
, buf
, SO_ALL
);
3958 SLAB_ATTR_RO(slabs
);
3960 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3962 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3964 SLAB_ATTR_RO(partial
);
3966 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3968 return show_slab_objects(s
, buf
, SO_CPU
);
3970 SLAB_ATTR_RO(cpu_slabs
);
3972 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3974 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3976 SLAB_ATTR_RO(objects
);
3978 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3980 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3982 SLAB_ATTR_RO(objects_partial
);
3984 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3986 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3988 SLAB_ATTR_RO(total_objects
);
3990 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3992 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3995 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3996 const char *buf
, size_t length
)
3998 s
->flags
&= ~SLAB_DEBUG_FREE
;
4000 s
->flags
|= SLAB_DEBUG_FREE
;
4003 SLAB_ATTR(sanity_checks
);
4005 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4007 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4010 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4013 s
->flags
&= ~SLAB_TRACE
;
4015 s
->flags
|= SLAB_TRACE
;
4020 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4022 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4025 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4026 const char *buf
, size_t length
)
4028 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4030 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4033 SLAB_ATTR(reclaim_account
);
4035 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4037 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4039 SLAB_ATTR_RO(hwcache_align
);
4041 #ifdef CONFIG_ZONE_DMA
4042 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4044 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4046 SLAB_ATTR_RO(cache_dma
);
4049 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4051 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4053 SLAB_ATTR_RO(destroy_by_rcu
);
4055 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4057 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4060 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4061 const char *buf
, size_t length
)
4063 if (any_slab_objects(s
))
4066 s
->flags
&= ~SLAB_RED_ZONE
;
4068 s
->flags
|= SLAB_RED_ZONE
;
4069 calculate_sizes(s
, -1);
4072 SLAB_ATTR(red_zone
);
4074 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4076 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4079 static ssize_t
poison_store(struct kmem_cache
*s
,
4080 const char *buf
, size_t length
)
4082 if (any_slab_objects(s
))
4085 s
->flags
&= ~SLAB_POISON
;
4087 s
->flags
|= SLAB_POISON
;
4088 calculate_sizes(s
, -1);
4093 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4095 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4098 static ssize_t
store_user_store(struct kmem_cache
*s
,
4099 const char *buf
, size_t length
)
4101 if (any_slab_objects(s
))
4104 s
->flags
&= ~SLAB_STORE_USER
;
4106 s
->flags
|= SLAB_STORE_USER
;
4107 calculate_sizes(s
, -1);
4110 SLAB_ATTR(store_user
);
4112 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4117 static ssize_t
validate_store(struct kmem_cache
*s
,
4118 const char *buf
, size_t length
)
4122 if (buf
[0] == '1') {
4123 ret
= validate_slab_cache(s
);
4129 SLAB_ATTR(validate
);
4131 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4136 static ssize_t
shrink_store(struct kmem_cache
*s
,
4137 const char *buf
, size_t length
)
4139 if (buf
[0] == '1') {
4140 int rc
= kmem_cache_shrink(s
);
4150 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4152 if (!(s
->flags
& SLAB_STORE_USER
))
4154 return list_locations(s
, buf
, TRACK_ALLOC
);
4156 SLAB_ATTR_RO(alloc_calls
);
4158 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4160 if (!(s
->flags
& SLAB_STORE_USER
))
4162 return list_locations(s
, buf
, TRACK_FREE
);
4164 SLAB_ATTR_RO(free_calls
);
4167 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4169 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4172 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4173 const char *buf
, size_t length
)
4175 unsigned long ratio
;
4178 err
= strict_strtoul(buf
, 10, &ratio
);
4183 s
->remote_node_defrag_ratio
= ratio
* 10;
4187 SLAB_ATTR(remote_node_defrag_ratio
);
4190 #ifdef CONFIG_SLUB_STATS
4191 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4193 unsigned long sum
= 0;
4196 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4201 for_each_online_cpu(cpu
) {
4202 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4208 len
= sprintf(buf
, "%lu", sum
);
4211 for_each_online_cpu(cpu
) {
4212 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4213 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4217 return len
+ sprintf(buf
+ len
, "\n");
4220 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4224 for_each_online_cpu(cpu
)
4225 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4228 #define STAT_ATTR(si, text) \
4229 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4231 return show_stat(s, buf, si); \
4233 static ssize_t text##_store(struct kmem_cache *s, \
4234 const char *buf, size_t length) \
4236 if (buf[0] != '0') \
4238 clear_stat(s, si); \
4243 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4244 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4245 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4246 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4247 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4248 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4249 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4250 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4251 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4252 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4253 STAT_ATTR(FREE_SLAB
, free_slab
);
4254 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4255 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4256 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4257 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4258 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4259 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4260 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4263 static struct attribute
*slab_attrs
[] = {
4264 &slab_size_attr
.attr
,
4265 &object_size_attr
.attr
,
4266 &objs_per_slab_attr
.attr
,
4268 &min_partial_attr
.attr
,
4270 &objects_partial_attr
.attr
,
4271 &total_objects_attr
.attr
,
4274 &cpu_slabs_attr
.attr
,
4278 &sanity_checks_attr
.attr
,
4280 &hwcache_align_attr
.attr
,
4281 &reclaim_account_attr
.attr
,
4282 &destroy_by_rcu_attr
.attr
,
4283 &red_zone_attr
.attr
,
4285 &store_user_attr
.attr
,
4286 &validate_attr
.attr
,
4288 &alloc_calls_attr
.attr
,
4289 &free_calls_attr
.attr
,
4290 #ifdef CONFIG_ZONE_DMA
4291 &cache_dma_attr
.attr
,
4294 &remote_node_defrag_ratio_attr
.attr
,
4296 #ifdef CONFIG_SLUB_STATS
4297 &alloc_fastpath_attr
.attr
,
4298 &alloc_slowpath_attr
.attr
,
4299 &free_fastpath_attr
.attr
,
4300 &free_slowpath_attr
.attr
,
4301 &free_frozen_attr
.attr
,
4302 &free_add_partial_attr
.attr
,
4303 &free_remove_partial_attr
.attr
,
4304 &alloc_from_partial_attr
.attr
,
4305 &alloc_slab_attr
.attr
,
4306 &alloc_refill_attr
.attr
,
4307 &free_slab_attr
.attr
,
4308 &cpuslab_flush_attr
.attr
,
4309 &deactivate_full_attr
.attr
,
4310 &deactivate_empty_attr
.attr
,
4311 &deactivate_to_head_attr
.attr
,
4312 &deactivate_to_tail_attr
.attr
,
4313 &deactivate_remote_frees_attr
.attr
,
4314 &order_fallback_attr
.attr
,
4319 static struct attribute_group slab_attr_group
= {
4320 .attrs
= slab_attrs
,
4323 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4324 struct attribute
*attr
,
4327 struct slab_attribute
*attribute
;
4328 struct kmem_cache
*s
;
4331 attribute
= to_slab_attr(attr
);
4334 if (!attribute
->show
)
4337 err
= attribute
->show(s
, buf
);
4342 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4343 struct attribute
*attr
,
4344 const char *buf
, size_t len
)
4346 struct slab_attribute
*attribute
;
4347 struct kmem_cache
*s
;
4350 attribute
= to_slab_attr(attr
);
4353 if (!attribute
->store
)
4356 err
= attribute
->store(s
, buf
, len
);
4361 static void kmem_cache_release(struct kobject
*kobj
)
4363 struct kmem_cache
*s
= to_slab(kobj
);
4368 static struct sysfs_ops slab_sysfs_ops
= {
4369 .show
= slab_attr_show
,
4370 .store
= slab_attr_store
,
4373 static struct kobj_type slab_ktype
= {
4374 .sysfs_ops
= &slab_sysfs_ops
,
4375 .release
= kmem_cache_release
4378 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4380 struct kobj_type
*ktype
= get_ktype(kobj
);
4382 if (ktype
== &slab_ktype
)
4387 static struct kset_uevent_ops slab_uevent_ops
= {
4388 .filter
= uevent_filter
,
4391 static struct kset
*slab_kset
;
4393 #define ID_STR_LENGTH 64
4395 /* Create a unique string id for a slab cache:
4397 * Format :[flags-]size
4399 static char *create_unique_id(struct kmem_cache
*s
)
4401 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4408 * First flags affecting slabcache operations. We will only
4409 * get here for aliasable slabs so we do not need to support
4410 * too many flags. The flags here must cover all flags that
4411 * are matched during merging to guarantee that the id is
4414 if (s
->flags
& SLAB_CACHE_DMA
)
4416 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4418 if (s
->flags
& SLAB_DEBUG_FREE
)
4420 if (!(s
->flags
& SLAB_NOTRACK
))
4424 p
+= sprintf(p
, "%07d", s
->size
);
4425 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4429 static int sysfs_slab_add(struct kmem_cache
*s
)
4435 if (slab_state
< SYSFS
)
4436 /* Defer until later */
4439 unmergeable
= slab_unmergeable(s
);
4442 * Slabcache can never be merged so we can use the name proper.
4443 * This is typically the case for debug situations. In that
4444 * case we can catch duplicate names easily.
4446 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4450 * Create a unique name for the slab as a target
4453 name
= create_unique_id(s
);
4456 s
->kobj
.kset
= slab_kset
;
4457 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4459 kobject_put(&s
->kobj
);
4463 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4465 kobject_del(&s
->kobj
);
4466 kobject_put(&s
->kobj
);
4469 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4471 /* Setup first alias */
4472 sysfs_slab_alias(s
, s
->name
);
4478 static void sysfs_slab_remove(struct kmem_cache
*s
)
4480 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4481 kobject_del(&s
->kobj
);
4482 kobject_put(&s
->kobj
);
4486 * Need to buffer aliases during bootup until sysfs becomes
4487 * available lest we lose that information.
4489 struct saved_alias
{
4490 struct kmem_cache
*s
;
4492 struct saved_alias
*next
;
4495 static struct saved_alias
*alias_list
;
4497 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4499 struct saved_alias
*al
;
4501 if (slab_state
== SYSFS
) {
4503 * If we have a leftover link then remove it.
4505 sysfs_remove_link(&slab_kset
->kobj
, name
);
4506 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4509 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4515 al
->next
= alias_list
;
4520 static int __init
slab_sysfs_init(void)
4522 struct kmem_cache
*s
;
4525 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4527 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4533 list_for_each_entry(s
, &slab_caches
, list
) {
4534 err
= sysfs_slab_add(s
);
4536 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4537 " to sysfs\n", s
->name
);
4540 while (alias_list
) {
4541 struct saved_alias
*al
= alias_list
;
4543 alias_list
= alias_list
->next
;
4544 err
= sysfs_slab_alias(al
->s
, al
->name
);
4546 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4547 " %s to sysfs\n", s
->name
);
4555 __initcall(slab_sysfs_init
);
4559 * The /proc/slabinfo ABI
4561 #ifdef CONFIG_SLABINFO
4562 static void print_slabinfo_header(struct seq_file
*m
)
4564 seq_puts(m
, "slabinfo - version: 2.1\n");
4565 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4566 "<objperslab> <pagesperslab>");
4567 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4568 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4572 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4576 down_read(&slub_lock
);
4578 print_slabinfo_header(m
);
4580 return seq_list_start(&slab_caches
, *pos
);
4583 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4585 return seq_list_next(p
, &slab_caches
, pos
);
4588 static void s_stop(struct seq_file
*m
, void *p
)
4590 up_read(&slub_lock
);
4593 static int s_show(struct seq_file
*m
, void *p
)
4595 unsigned long nr_partials
= 0;
4596 unsigned long nr_slabs
= 0;
4597 unsigned long nr_inuse
= 0;
4598 unsigned long nr_objs
= 0;
4599 unsigned long nr_free
= 0;
4600 struct kmem_cache
*s
;
4603 s
= list_entry(p
, struct kmem_cache
, list
);
4605 for_each_online_node(node
) {
4606 struct kmem_cache_node
*n
= get_node(s
, node
);
4611 nr_partials
+= n
->nr_partial
;
4612 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4613 nr_objs
+= atomic_long_read(&n
->total_objects
);
4614 nr_free
+= count_partial(n
, count_free
);
4617 nr_inuse
= nr_objs
- nr_free
;
4619 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4620 nr_objs
, s
->size
, oo_objects(s
->oo
),
4621 (1 << oo_order(s
->oo
)));
4622 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4623 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4629 static const struct seq_operations slabinfo_op
= {
4636 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4638 return seq_open(file
, &slabinfo_op
);
4641 static const struct file_operations proc_slabinfo_operations
= {
4642 .open
= slabinfo_open
,
4644 .llseek
= seq_lseek
,
4645 .release
= seq_release
,
4648 static int __init
slab_proc_init(void)
4650 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4653 module_init(slab_proc_init
);
4654 #endif /* CONFIG_SLABINFO */