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 | \
157 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
158 SLAB_CACHE_DMA | SLAB_NOTRACK)
161 #define OO_MASK ((1 << OO_SHIFT) - 1)
162 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
164 /* Internal SLUB flags */
165 #define __OBJECT_POISON 0x80000000 /* Poison object */
166 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
168 static int kmem_size
= sizeof(struct kmem_cache
);
171 static struct notifier_block slab_notifier
;
175 DOWN
, /* No slab functionality available */
176 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
177 UP
, /* Everything works but does not show up in sysfs */
181 /* A list of all slab caches on the system */
182 static DECLARE_RWSEM(slub_lock
);
183 static LIST_HEAD(slab_caches
);
186 * Tracking user of a slab.
189 unsigned long addr
; /* Called from address */
190 int cpu
; /* Was running on cpu */
191 int pid
; /* Pid context */
192 unsigned long when
; /* When did the operation occur */
195 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
197 #ifdef CONFIG_SLUB_DEBUG
198 static int sysfs_slab_add(struct kmem_cache
*);
199 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
200 static void sysfs_slab_remove(struct kmem_cache
*);
203 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
204 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
206 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
213 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
215 #ifdef CONFIG_SLUB_STATS
216 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
220 /********************************************************************
221 * Core slab cache functions
222 *******************************************************************/
224 int slab_is_available(void)
226 return slab_state
>= UP
;
229 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
232 return s
->node
[node
];
234 return &s
->local_node
;
238 /* Verify that a pointer has an address that is valid within a slab page */
239 static inline int check_valid_pointer(struct kmem_cache
*s
,
240 struct page
*page
, const void *object
)
247 base
= page_address(page
);
248 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
249 (object
- base
) % s
->size
) {
256 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
258 return *(void **)(object
+ s
->offset
);
261 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
263 *(void **)(object
+ s
->offset
) = fp
;
266 /* Loop over all objects in a slab */
267 #define for_each_object(__p, __s, __addr, __objects) \
268 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
272 #define for_each_free_object(__p, __s, __free) \
273 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
275 /* Determine object index from a given position */
276 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
278 return (p
- addr
) / s
->size
;
281 static inline struct kmem_cache_order_objects
oo_make(int order
,
284 struct kmem_cache_order_objects x
= {
285 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
291 static inline int oo_order(struct kmem_cache_order_objects x
)
293 return x
.x
>> OO_SHIFT
;
296 static inline int oo_objects(struct kmem_cache_order_objects x
)
298 return x
.x
& OO_MASK
;
301 #ifdef CONFIG_SLUB_DEBUG
305 #ifdef CONFIG_SLUB_DEBUG_ON
306 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
308 static int slub_debug
;
311 static char *slub_debug_slabs
;
312 static int disable_higher_order_debug
;
317 static void print_section(char *text
, u8
*addr
, unsigned int length
)
325 for (i
= 0; i
< length
; i
++) {
327 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
330 printk(KERN_CONT
" %02x", addr
[i
]);
332 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
334 printk(KERN_CONT
" %s\n", ascii
);
341 printk(KERN_CONT
" ");
345 printk(KERN_CONT
" %s\n", ascii
);
349 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
350 enum track_item alloc
)
355 p
= object
+ s
->offset
+ sizeof(void *);
357 p
= object
+ s
->inuse
;
362 static void set_track(struct kmem_cache
*s
, void *object
,
363 enum track_item alloc
, unsigned long addr
)
365 struct track
*p
= get_track(s
, object
, alloc
);
369 p
->cpu
= smp_processor_id();
370 p
->pid
= current
->pid
;
373 memset(p
, 0, sizeof(struct track
));
376 static void init_tracking(struct kmem_cache
*s
, void *object
)
378 if (!(s
->flags
& SLAB_STORE_USER
))
381 set_track(s
, object
, TRACK_FREE
, 0UL);
382 set_track(s
, object
, TRACK_ALLOC
, 0UL);
385 static void print_track(const char *s
, struct track
*t
)
390 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
391 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
394 static void print_tracking(struct kmem_cache
*s
, void *object
)
396 if (!(s
->flags
& SLAB_STORE_USER
))
399 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
400 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
403 static void print_page_info(struct page
*page
)
405 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
406 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
410 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
416 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
418 printk(KERN_ERR
"========================================"
419 "=====================================\n");
420 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
421 printk(KERN_ERR
"----------------------------------------"
422 "-------------------------------------\n\n");
425 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
431 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
433 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
436 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
438 unsigned int off
; /* Offset of last byte */
439 u8
*addr
= page_address(page
);
441 print_tracking(s
, p
);
443 print_page_info(page
);
445 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
446 p
, p
- addr
, get_freepointer(s
, p
));
449 print_section("Bytes b4", p
- 16, 16);
451 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
453 if (s
->flags
& SLAB_RED_ZONE
)
454 print_section("Redzone", p
+ s
->objsize
,
455 s
->inuse
- s
->objsize
);
458 off
= s
->offset
+ sizeof(void *);
462 if (s
->flags
& SLAB_STORE_USER
)
463 off
+= 2 * sizeof(struct track
);
466 /* Beginning of the filler is the free pointer */
467 print_section("Padding", p
+ off
, s
->size
- off
);
472 static void object_err(struct kmem_cache
*s
, struct page
*page
,
473 u8
*object
, char *reason
)
475 slab_bug(s
, "%s", reason
);
476 print_trailer(s
, page
, object
);
479 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
485 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
487 slab_bug(s
, "%s", buf
);
488 print_page_info(page
);
492 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
496 if (s
->flags
& __OBJECT_POISON
) {
497 memset(p
, POISON_FREE
, s
->objsize
- 1);
498 p
[s
->objsize
- 1] = POISON_END
;
501 if (s
->flags
& SLAB_RED_ZONE
)
502 memset(p
+ s
->objsize
,
503 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
504 s
->inuse
- s
->objsize
);
507 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
510 if (*start
!= (u8
)value
)
518 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
519 void *from
, void *to
)
521 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
522 memset(from
, data
, to
- from
);
525 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
526 u8
*object
, char *what
,
527 u8
*start
, unsigned int value
, unsigned int bytes
)
532 fault
= check_bytes(start
, value
, bytes
);
537 while (end
> fault
&& end
[-1] == value
)
540 slab_bug(s
, "%s overwritten", what
);
541 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
542 fault
, end
- 1, fault
[0], value
);
543 print_trailer(s
, page
, object
);
545 restore_bytes(s
, what
, value
, fault
, end
);
553 * Bytes of the object to be managed.
554 * If the freepointer may overlay the object then the free
555 * pointer is the first word of the object.
557 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
560 * object + s->objsize
561 * Padding to reach word boundary. This is also used for Redzoning.
562 * Padding is extended by another word if Redzoning is enabled and
565 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
566 * 0xcc (RED_ACTIVE) for objects in use.
569 * Meta data starts here.
571 * A. Free pointer (if we cannot overwrite object on free)
572 * B. Tracking data for SLAB_STORE_USER
573 * C. Padding to reach required alignment boundary or at mininum
574 * one word if debugging is on to be able to detect writes
575 * before the word boundary.
577 * Padding is done using 0x5a (POISON_INUSE)
580 * Nothing is used beyond s->size.
582 * If slabcaches are merged then the objsize and inuse boundaries are mostly
583 * ignored. And therefore no slab options that rely on these boundaries
584 * may be used with merged slabcaches.
587 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
589 unsigned long off
= s
->inuse
; /* The end of info */
592 /* Freepointer is placed after the object. */
593 off
+= sizeof(void *);
595 if (s
->flags
& SLAB_STORE_USER
)
596 /* We also have user information there */
597 off
+= 2 * sizeof(struct track
);
602 return check_bytes_and_report(s
, page
, p
, "Object padding",
603 p
+ off
, POISON_INUSE
, s
->size
- off
);
606 /* Check the pad bytes at the end of a slab page */
607 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
615 if (!(s
->flags
& SLAB_POISON
))
618 start
= page_address(page
);
619 length
= (PAGE_SIZE
<< compound_order(page
));
620 end
= start
+ length
;
621 remainder
= length
% s
->size
;
625 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
628 while (end
> fault
&& end
[-1] == POISON_INUSE
)
631 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
632 print_section("Padding", end
- remainder
, remainder
);
634 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
638 static int check_object(struct kmem_cache
*s
, struct page
*page
,
639 void *object
, int active
)
642 u8
*endobject
= object
+ s
->objsize
;
644 if (s
->flags
& SLAB_RED_ZONE
) {
646 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
648 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
649 endobject
, red
, s
->inuse
- s
->objsize
))
652 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
653 check_bytes_and_report(s
, page
, p
, "Alignment padding",
654 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
658 if (s
->flags
& SLAB_POISON
) {
659 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
660 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
661 POISON_FREE
, s
->objsize
- 1) ||
662 !check_bytes_and_report(s
, page
, p
, "Poison",
663 p
+ s
->objsize
- 1, POISON_END
, 1)))
666 * check_pad_bytes cleans up on its own.
668 check_pad_bytes(s
, page
, p
);
671 if (!s
->offset
&& active
)
673 * Object and freepointer overlap. Cannot check
674 * freepointer while object is allocated.
678 /* Check free pointer validity */
679 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
680 object_err(s
, page
, p
, "Freepointer corrupt");
682 * No choice but to zap it and thus lose the remainder
683 * of the free objects in this slab. May cause
684 * another error because the object count is now wrong.
686 set_freepointer(s
, p
, NULL
);
692 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
696 VM_BUG_ON(!irqs_disabled());
698 if (!PageSlab(page
)) {
699 slab_err(s
, page
, "Not a valid slab page");
703 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
704 if (page
->objects
> maxobj
) {
705 slab_err(s
, page
, "objects %u > max %u",
706 s
->name
, page
->objects
, maxobj
);
709 if (page
->inuse
> page
->objects
) {
710 slab_err(s
, page
, "inuse %u > max %u",
711 s
->name
, page
->inuse
, page
->objects
);
714 /* Slab_pad_check fixes things up after itself */
715 slab_pad_check(s
, page
);
720 * Determine if a certain object on a page is on the freelist. Must hold the
721 * slab lock to guarantee that the chains are in a consistent state.
723 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
726 void *fp
= page
->freelist
;
728 unsigned long max_objects
;
730 while (fp
&& nr
<= page
->objects
) {
733 if (!check_valid_pointer(s
, page
, fp
)) {
735 object_err(s
, page
, object
,
736 "Freechain corrupt");
737 set_freepointer(s
, object
, NULL
);
740 slab_err(s
, page
, "Freepointer corrupt");
741 page
->freelist
= NULL
;
742 page
->inuse
= page
->objects
;
743 slab_fix(s
, "Freelist cleared");
749 fp
= get_freepointer(s
, object
);
753 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
754 if (max_objects
> MAX_OBJS_PER_PAGE
)
755 max_objects
= MAX_OBJS_PER_PAGE
;
757 if (page
->objects
!= max_objects
) {
758 slab_err(s
, page
, "Wrong number of objects. Found %d but "
759 "should be %d", page
->objects
, max_objects
);
760 page
->objects
= max_objects
;
761 slab_fix(s
, "Number of objects adjusted.");
763 if (page
->inuse
!= page
->objects
- nr
) {
764 slab_err(s
, page
, "Wrong object count. Counter is %d but "
765 "counted were %d", page
->inuse
, page
->objects
- nr
);
766 page
->inuse
= page
->objects
- nr
;
767 slab_fix(s
, "Object count adjusted.");
769 return search
== NULL
;
772 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
775 if (s
->flags
& SLAB_TRACE
) {
776 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
778 alloc
? "alloc" : "free",
783 print_section("Object", (void *)object
, s
->objsize
);
790 * Tracking of fully allocated slabs for debugging purposes.
792 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
794 spin_lock(&n
->list_lock
);
795 list_add(&page
->lru
, &n
->full
);
796 spin_unlock(&n
->list_lock
);
799 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
801 struct kmem_cache_node
*n
;
803 if (!(s
->flags
& SLAB_STORE_USER
))
806 n
= get_node(s
, page_to_nid(page
));
808 spin_lock(&n
->list_lock
);
809 list_del(&page
->lru
);
810 spin_unlock(&n
->list_lock
);
813 /* Tracking of the number of slabs for debugging purposes */
814 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
816 struct kmem_cache_node
*n
= get_node(s
, node
);
818 return atomic_long_read(&n
->nr_slabs
);
821 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
823 return atomic_long_read(&n
->nr_slabs
);
826 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
828 struct kmem_cache_node
*n
= get_node(s
, node
);
831 * May be called early in order to allocate a slab for the
832 * kmem_cache_node structure. Solve the chicken-egg
833 * dilemma by deferring the increment of the count during
834 * bootstrap (see early_kmem_cache_node_alloc).
836 if (!NUMA_BUILD
|| n
) {
837 atomic_long_inc(&n
->nr_slabs
);
838 atomic_long_add(objects
, &n
->total_objects
);
841 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
843 struct kmem_cache_node
*n
= get_node(s
, node
);
845 atomic_long_dec(&n
->nr_slabs
);
846 atomic_long_sub(objects
, &n
->total_objects
);
849 /* Object debug checks for alloc/free paths */
850 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
853 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
856 init_object(s
, object
, 0);
857 init_tracking(s
, object
);
860 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
861 void *object
, unsigned long addr
)
863 if (!check_slab(s
, page
))
866 if (!on_freelist(s
, page
, object
)) {
867 object_err(s
, page
, object
, "Object already allocated");
871 if (!check_valid_pointer(s
, page
, object
)) {
872 object_err(s
, page
, object
, "Freelist Pointer check fails");
876 if (!check_object(s
, page
, object
, 0))
879 /* Success perform special debug activities for allocs */
880 if (s
->flags
& SLAB_STORE_USER
)
881 set_track(s
, object
, TRACK_ALLOC
, addr
);
882 trace(s
, page
, object
, 1);
883 init_object(s
, object
, 1);
887 if (PageSlab(page
)) {
889 * If this is a slab page then lets do the best we can
890 * to avoid issues in the future. Marking all objects
891 * as used avoids touching the remaining objects.
893 slab_fix(s
, "Marking all objects used");
894 page
->inuse
= page
->objects
;
895 page
->freelist
= NULL
;
900 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
901 void *object
, unsigned long addr
)
903 if (!check_slab(s
, page
))
906 if (!check_valid_pointer(s
, page
, object
)) {
907 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
911 if (on_freelist(s
, page
, object
)) {
912 object_err(s
, page
, object
, "Object already free");
916 if (!check_object(s
, page
, object
, 1))
919 if (unlikely(s
!= page
->slab
)) {
920 if (!PageSlab(page
)) {
921 slab_err(s
, page
, "Attempt to free object(0x%p) "
922 "outside of slab", object
);
923 } else if (!page
->slab
) {
925 "SLUB <none>: no slab for object 0x%p.\n",
929 object_err(s
, page
, object
,
930 "page slab pointer corrupt.");
934 /* Special debug activities for freeing objects */
935 if (!PageSlubFrozen(page
) && !page
->freelist
)
936 remove_full(s
, page
);
937 if (s
->flags
& SLAB_STORE_USER
)
938 set_track(s
, object
, TRACK_FREE
, addr
);
939 trace(s
, page
, object
, 0);
940 init_object(s
, object
, 0);
944 slab_fix(s
, "Object at 0x%p not freed", object
);
948 static int __init
setup_slub_debug(char *str
)
950 slub_debug
= DEBUG_DEFAULT_FLAGS
;
951 if (*str
++ != '=' || !*str
)
953 * No options specified. Switch on full debugging.
959 * No options but restriction on slabs. This means full
960 * debugging for slabs matching a pattern.
964 if (tolower(*str
) == 'o') {
966 * Avoid enabling debugging on caches if its minimum order
967 * would increase as a result.
969 disable_higher_order_debug
= 1;
976 * Switch off all debugging measures.
981 * Determine which debug features should be switched on
983 for (; *str
&& *str
!= ','; str
++) {
984 switch (tolower(*str
)) {
986 slub_debug
|= SLAB_DEBUG_FREE
;
989 slub_debug
|= SLAB_RED_ZONE
;
992 slub_debug
|= SLAB_POISON
;
995 slub_debug
|= SLAB_STORE_USER
;
998 slub_debug
|= SLAB_TRACE
;
1001 slub_debug
|= SLAB_FAILSLAB
;
1004 printk(KERN_ERR
"slub_debug option '%c' "
1005 "unknown. skipped\n", *str
);
1011 slub_debug_slabs
= str
+ 1;
1016 __setup("slub_debug", setup_slub_debug
);
1018 static unsigned long kmem_cache_flags(unsigned long objsize
,
1019 unsigned long flags
, const char *name
,
1020 void (*ctor
)(void *))
1023 * Enable debugging if selected on the kernel commandline.
1025 if (slub_debug
&& (!slub_debug_slabs
||
1026 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1027 flags
|= slub_debug
;
1032 static inline void setup_object_debug(struct kmem_cache
*s
,
1033 struct page
*page
, void *object
) {}
1035 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1036 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1038 static inline int free_debug_processing(struct kmem_cache
*s
,
1039 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1041 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1043 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1044 void *object
, int active
) { return 1; }
1045 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1046 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1047 unsigned long flags
, const char *name
,
1048 void (*ctor
)(void *))
1052 #define slub_debug 0
1054 #define disable_higher_order_debug 0
1056 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1058 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1060 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1062 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1067 * Slab allocation and freeing
1069 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1070 struct kmem_cache_order_objects oo
)
1072 int order
= oo_order(oo
);
1074 flags
|= __GFP_NOTRACK
;
1077 return alloc_pages(flags
, order
);
1079 return alloc_pages_exact_node(node
, flags
, order
);
1082 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1085 struct kmem_cache_order_objects oo
= s
->oo
;
1088 flags
|= s
->allocflags
;
1091 * Let the initial higher-order allocation fail under memory pressure
1092 * so we fall-back to the minimum order allocation.
1094 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1096 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1097 if (unlikely(!page
)) {
1100 * Allocation may have failed due to fragmentation.
1101 * Try a lower order alloc if possible
1103 page
= alloc_slab_page(flags
, node
, oo
);
1107 stat(s
, ORDER_FALLBACK
);
1110 if (kmemcheck_enabled
1111 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1112 int pages
= 1 << oo_order(oo
);
1114 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1117 * Objects from caches that have a constructor don't get
1118 * cleared when they're allocated, so we need to do it here.
1121 kmemcheck_mark_uninitialized_pages(page
, pages
);
1123 kmemcheck_mark_unallocated_pages(page
, pages
);
1126 page
->objects
= oo_objects(oo
);
1127 mod_zone_page_state(page_zone(page
),
1128 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1129 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1135 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1138 setup_object_debug(s
, page
, object
);
1139 if (unlikely(s
->ctor
))
1143 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1150 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1152 page
= allocate_slab(s
,
1153 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1157 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1159 page
->flags
|= 1 << PG_slab
;
1160 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1161 SLAB_STORE_USER
| SLAB_TRACE
))
1162 __SetPageSlubDebug(page
);
1164 start
= page_address(page
);
1166 if (unlikely(s
->flags
& SLAB_POISON
))
1167 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1170 for_each_object(p
, s
, start
, page
->objects
) {
1171 setup_object(s
, page
, last
);
1172 set_freepointer(s
, last
, p
);
1175 setup_object(s
, page
, last
);
1176 set_freepointer(s
, last
, NULL
);
1178 page
->freelist
= start
;
1184 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1186 int order
= compound_order(page
);
1187 int pages
= 1 << order
;
1189 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1192 slab_pad_check(s
, page
);
1193 for_each_object(p
, s
, page_address(page
),
1195 check_object(s
, page
, p
, 0);
1196 __ClearPageSlubDebug(page
);
1199 kmemcheck_free_shadow(page
, compound_order(page
));
1201 mod_zone_page_state(page_zone(page
),
1202 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1203 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1206 __ClearPageSlab(page
);
1207 reset_page_mapcount(page
);
1208 if (current
->reclaim_state
)
1209 current
->reclaim_state
->reclaimed_slab
+= pages
;
1210 __free_pages(page
, order
);
1213 static void rcu_free_slab(struct rcu_head
*h
)
1217 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1218 __free_slab(page
->slab
, page
);
1221 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1223 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1225 * RCU free overloads the RCU head over the LRU
1227 struct rcu_head
*head
= (void *)&page
->lru
;
1229 call_rcu(head
, rcu_free_slab
);
1231 __free_slab(s
, page
);
1234 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1236 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1241 * Per slab locking using the pagelock
1243 static __always_inline
void slab_lock(struct page
*page
)
1245 bit_spin_lock(PG_locked
, &page
->flags
);
1248 static __always_inline
void slab_unlock(struct page
*page
)
1250 __bit_spin_unlock(PG_locked
, &page
->flags
);
1253 static __always_inline
int slab_trylock(struct page
*page
)
1257 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1262 * Management of partially allocated slabs
1264 static void add_partial(struct kmem_cache_node
*n
,
1265 struct page
*page
, int tail
)
1267 spin_lock(&n
->list_lock
);
1270 list_add_tail(&page
->lru
, &n
->partial
);
1272 list_add(&page
->lru
, &n
->partial
);
1273 spin_unlock(&n
->list_lock
);
1276 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1278 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1280 spin_lock(&n
->list_lock
);
1281 list_del(&page
->lru
);
1283 spin_unlock(&n
->list_lock
);
1287 * Lock slab and remove from the partial list.
1289 * Must hold list_lock.
1291 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1294 if (slab_trylock(page
)) {
1295 list_del(&page
->lru
);
1297 __SetPageSlubFrozen(page
);
1304 * Try to allocate a partial slab from a specific node.
1306 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1311 * Racy check. If we mistakenly see no partial slabs then we
1312 * just allocate an empty slab. If we mistakenly try to get a
1313 * partial slab and there is none available then get_partials()
1316 if (!n
|| !n
->nr_partial
)
1319 spin_lock(&n
->list_lock
);
1320 list_for_each_entry(page
, &n
->partial
, lru
)
1321 if (lock_and_freeze_slab(n
, page
))
1325 spin_unlock(&n
->list_lock
);
1330 * Get a page from somewhere. Search in increasing NUMA distances.
1332 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1335 struct zonelist
*zonelist
;
1338 enum zone_type high_zoneidx
= gfp_zone(flags
);
1342 * The defrag ratio allows a configuration of the tradeoffs between
1343 * inter node defragmentation and node local allocations. A lower
1344 * defrag_ratio increases the tendency to do local allocations
1345 * instead of attempting to obtain partial slabs from other nodes.
1347 * If the defrag_ratio is set to 0 then kmalloc() always
1348 * returns node local objects. If the ratio is higher then kmalloc()
1349 * may return off node objects because partial slabs are obtained
1350 * from other nodes and filled up.
1352 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1353 * defrag_ratio = 1000) then every (well almost) allocation will
1354 * first attempt to defrag slab caches on other nodes. This means
1355 * scanning over all nodes to look for partial slabs which may be
1356 * expensive if we do it every time we are trying to find a slab
1357 * with available objects.
1359 if (!s
->remote_node_defrag_ratio
||
1360 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1363 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1364 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1365 struct kmem_cache_node
*n
;
1367 n
= get_node(s
, zone_to_nid(zone
));
1369 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1370 n
->nr_partial
> s
->min_partial
) {
1371 page
= get_partial_node(n
);
1381 * Get a partial page, lock it and return it.
1383 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1386 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1388 page
= get_partial_node(get_node(s
, searchnode
));
1389 if (page
|| (flags
& __GFP_THISNODE
))
1392 return get_any_partial(s
, flags
);
1396 * Move a page back to the lists.
1398 * Must be called with the slab lock held.
1400 * On exit the slab lock will have been dropped.
1402 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1404 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1406 __ClearPageSlubFrozen(page
);
1409 if (page
->freelist
) {
1410 add_partial(n
, page
, tail
);
1411 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1413 stat(s
, DEACTIVATE_FULL
);
1414 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1415 (s
->flags
& SLAB_STORE_USER
))
1420 stat(s
, DEACTIVATE_EMPTY
);
1421 if (n
->nr_partial
< s
->min_partial
) {
1423 * Adding an empty slab to the partial slabs in order
1424 * to avoid page allocator overhead. This slab needs
1425 * to come after the other slabs with objects in
1426 * so that the others get filled first. That way the
1427 * size of the partial list stays small.
1429 * kmem_cache_shrink can reclaim any empty slabs from
1432 add_partial(n
, page
, 1);
1437 discard_slab(s
, page
);
1443 * Remove the cpu slab
1445 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1447 struct page
*page
= c
->page
;
1451 stat(s
, DEACTIVATE_REMOTE_FREES
);
1453 * Merge cpu freelist into slab freelist. Typically we get here
1454 * because both freelists are empty. So this is unlikely
1457 while (unlikely(c
->freelist
)) {
1460 tail
= 0; /* Hot objects. Put the slab first */
1462 /* Retrieve object from cpu_freelist */
1463 object
= c
->freelist
;
1464 c
->freelist
= get_freepointer(s
, c
->freelist
);
1466 /* And put onto the regular freelist */
1467 set_freepointer(s
, object
, page
->freelist
);
1468 page
->freelist
= object
;
1472 unfreeze_slab(s
, page
, tail
);
1475 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1477 stat(s
, CPUSLAB_FLUSH
);
1479 deactivate_slab(s
, c
);
1485 * Called from IPI handler with interrupts disabled.
1487 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1489 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1491 if (likely(c
&& c
->page
))
1495 static void flush_cpu_slab(void *d
)
1497 struct kmem_cache
*s
= d
;
1499 __flush_cpu_slab(s
, smp_processor_id());
1502 static void flush_all(struct kmem_cache
*s
)
1504 on_each_cpu(flush_cpu_slab
, s
, 1);
1508 * Check if the objects in a per cpu structure fit numa
1509 * locality expectations.
1511 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1514 if (node
!= -1 && c
->node
!= node
)
1520 static int count_free(struct page
*page
)
1522 return page
->objects
- page
->inuse
;
1525 static unsigned long count_partial(struct kmem_cache_node
*n
,
1526 int (*get_count
)(struct page
*))
1528 unsigned long flags
;
1529 unsigned long x
= 0;
1532 spin_lock_irqsave(&n
->list_lock
, flags
);
1533 list_for_each_entry(page
, &n
->partial
, lru
)
1534 x
+= get_count(page
);
1535 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1539 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1541 #ifdef CONFIG_SLUB_DEBUG
1542 return atomic_long_read(&n
->total_objects
);
1548 static noinline
void
1549 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1554 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1556 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1557 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1558 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1560 if (oo_order(s
->min
) > get_order(s
->objsize
))
1561 printk(KERN_WARNING
" %s debugging increased min order, use "
1562 "slub_debug=O to disable.\n", s
->name
);
1564 for_each_online_node(node
) {
1565 struct kmem_cache_node
*n
= get_node(s
, node
);
1566 unsigned long nr_slabs
;
1567 unsigned long nr_objs
;
1568 unsigned long nr_free
;
1573 nr_free
= count_partial(n
, count_free
);
1574 nr_slabs
= node_nr_slabs(n
);
1575 nr_objs
= node_nr_objs(n
);
1578 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1579 node
, nr_slabs
, nr_objs
, nr_free
);
1584 * Slow path. The lockless freelist is empty or we need to perform
1587 * Interrupts are disabled.
1589 * Processing is still very fast if new objects have been freed to the
1590 * regular freelist. In that case we simply take over the regular freelist
1591 * as the lockless freelist and zap the regular freelist.
1593 * If that is not working then we fall back to the partial lists. We take the
1594 * first element of the freelist as the object to allocate now and move the
1595 * rest of the freelist to the lockless freelist.
1597 * And if we were unable to get a new slab from the partial slab lists then
1598 * we need to allocate a new slab. This is the slowest path since it involves
1599 * a call to the page allocator and the setup of a new slab.
1601 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1602 unsigned long addr
, struct kmem_cache_cpu
*c
)
1607 /* We handle __GFP_ZERO in the caller */
1608 gfpflags
&= ~__GFP_ZERO
;
1614 if (unlikely(!node_match(c
, node
)))
1617 stat(s
, ALLOC_REFILL
);
1620 object
= c
->page
->freelist
;
1621 if (unlikely(!object
))
1623 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1626 c
->freelist
= get_freepointer(s
, object
);
1627 c
->page
->inuse
= c
->page
->objects
;
1628 c
->page
->freelist
= NULL
;
1629 c
->node
= page_to_nid(c
->page
);
1631 slab_unlock(c
->page
);
1632 stat(s
, ALLOC_SLOWPATH
);
1636 deactivate_slab(s
, c
);
1639 new = get_partial(s
, gfpflags
, node
);
1642 stat(s
, ALLOC_FROM_PARTIAL
);
1646 if (gfpflags
& __GFP_WAIT
)
1649 new = new_slab(s
, gfpflags
, node
);
1651 if (gfpflags
& __GFP_WAIT
)
1652 local_irq_disable();
1655 c
= __this_cpu_ptr(s
->cpu_slab
);
1656 stat(s
, ALLOC_SLAB
);
1660 __SetPageSlubFrozen(new);
1664 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1665 slab_out_of_memory(s
, gfpflags
, node
);
1668 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1672 c
->page
->freelist
= get_freepointer(s
, object
);
1678 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1679 * have the fastpath folded into their functions. So no function call
1680 * overhead for requests that can be satisfied on the fastpath.
1682 * The fastpath works by first checking if the lockless freelist can be used.
1683 * If not then __slab_alloc is called for slow processing.
1685 * Otherwise we can simply pick the next object from the lockless free list.
1687 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1688 gfp_t gfpflags
, int node
, unsigned long addr
)
1691 struct kmem_cache_cpu
*c
;
1692 unsigned long flags
;
1694 gfpflags
&= gfp_allowed_mask
;
1696 lockdep_trace_alloc(gfpflags
);
1697 might_sleep_if(gfpflags
& __GFP_WAIT
);
1699 if (should_failslab(s
->objsize
, gfpflags
, s
->flags
))
1702 local_irq_save(flags
);
1703 c
= __this_cpu_ptr(s
->cpu_slab
);
1704 object
= c
->freelist
;
1705 if (unlikely(!object
|| !node_match(c
, node
)))
1707 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1710 c
->freelist
= get_freepointer(s
, object
);
1711 stat(s
, ALLOC_FASTPATH
);
1713 local_irq_restore(flags
);
1715 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1716 memset(object
, 0, s
->objsize
);
1718 kmemcheck_slab_alloc(s
, gfpflags
, object
, s
->objsize
);
1719 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, gfpflags
);
1724 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1726 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1728 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1732 EXPORT_SYMBOL(kmem_cache_alloc
);
1734 #ifdef CONFIG_TRACING
1735 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1737 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1739 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1743 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1745 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1747 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1748 s
->objsize
, s
->size
, gfpflags
, node
);
1752 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1755 #ifdef CONFIG_TRACING
1756 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1760 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1762 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1766 * Slow patch handling. This may still be called frequently since objects
1767 * have a longer lifetime than the cpu slabs in most processing loads.
1769 * So we still attempt to reduce cache line usage. Just take the slab
1770 * lock and free the item. If there is no additional partial page
1771 * handling required then we can return immediately.
1773 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1774 void *x
, unsigned long addr
)
1777 void **object
= (void *)x
;
1779 stat(s
, FREE_SLOWPATH
);
1782 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1786 prior
= page
->freelist
;
1787 set_freepointer(s
, object
, prior
);
1788 page
->freelist
= object
;
1791 if (unlikely(PageSlubFrozen(page
))) {
1792 stat(s
, FREE_FROZEN
);
1796 if (unlikely(!page
->inuse
))
1800 * Objects left in the slab. If it was not on the partial list before
1803 if (unlikely(!prior
)) {
1804 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1805 stat(s
, FREE_ADD_PARTIAL
);
1815 * Slab still on the partial list.
1817 remove_partial(s
, page
);
1818 stat(s
, FREE_REMOVE_PARTIAL
);
1822 discard_slab(s
, page
);
1826 if (!free_debug_processing(s
, page
, x
, addr
))
1832 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1833 * can perform fastpath freeing without additional function calls.
1835 * The fastpath is only possible if we are freeing to the current cpu slab
1836 * of this processor. This typically the case if we have just allocated
1839 * If fastpath is not possible then fall back to __slab_free where we deal
1840 * with all sorts of special processing.
1842 static __always_inline
void slab_free(struct kmem_cache
*s
,
1843 struct page
*page
, void *x
, unsigned long addr
)
1845 void **object
= (void *)x
;
1846 struct kmem_cache_cpu
*c
;
1847 unsigned long flags
;
1849 kmemleak_free_recursive(x
, s
->flags
);
1850 local_irq_save(flags
);
1851 c
= __this_cpu_ptr(s
->cpu_slab
);
1852 kmemcheck_slab_free(s
, object
, s
->objsize
);
1853 debug_check_no_locks_freed(object
, s
->objsize
);
1854 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1855 debug_check_no_obj_freed(object
, s
->objsize
);
1856 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1857 set_freepointer(s
, object
, c
->freelist
);
1858 c
->freelist
= object
;
1859 stat(s
, FREE_FASTPATH
);
1861 __slab_free(s
, page
, x
, addr
);
1863 local_irq_restore(flags
);
1866 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1870 page
= virt_to_head_page(x
);
1872 slab_free(s
, page
, x
, _RET_IP_
);
1874 trace_kmem_cache_free(_RET_IP_
, x
);
1876 EXPORT_SYMBOL(kmem_cache_free
);
1878 /* Figure out on which slab page the object resides */
1879 static struct page
*get_object_page(const void *x
)
1881 struct page
*page
= virt_to_head_page(x
);
1883 if (!PageSlab(page
))
1890 * Object placement in a slab is made very easy because we always start at
1891 * offset 0. If we tune the size of the object to the alignment then we can
1892 * get the required alignment by putting one properly sized object after
1895 * Notice that the allocation order determines the sizes of the per cpu
1896 * caches. Each processor has always one slab available for allocations.
1897 * Increasing the allocation order reduces the number of times that slabs
1898 * must be moved on and off the partial lists and is therefore a factor in
1903 * Mininum / Maximum order of slab pages. This influences locking overhead
1904 * and slab fragmentation. A higher order reduces the number of partial slabs
1905 * and increases the number of allocations possible without having to
1906 * take the list_lock.
1908 static int slub_min_order
;
1909 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1910 static int slub_min_objects
;
1913 * Merge control. If this is set then no merging of slab caches will occur.
1914 * (Could be removed. This was introduced to pacify the merge skeptics.)
1916 static int slub_nomerge
;
1919 * Calculate the order of allocation given an slab object size.
1921 * The order of allocation has significant impact on performance and other
1922 * system components. Generally order 0 allocations should be preferred since
1923 * order 0 does not cause fragmentation in the page allocator. Larger objects
1924 * be problematic to put into order 0 slabs because there may be too much
1925 * unused space left. We go to a higher order if more than 1/16th of the slab
1928 * In order to reach satisfactory performance we must ensure that a minimum
1929 * number of objects is in one slab. Otherwise we may generate too much
1930 * activity on the partial lists which requires taking the list_lock. This is
1931 * less a concern for large slabs though which are rarely used.
1933 * slub_max_order specifies the order where we begin to stop considering the
1934 * number of objects in a slab as critical. If we reach slub_max_order then
1935 * we try to keep the page order as low as possible. So we accept more waste
1936 * of space in favor of a small page order.
1938 * Higher order allocations also allow the placement of more objects in a
1939 * slab and thereby reduce object handling overhead. If the user has
1940 * requested a higher mininum order then we start with that one instead of
1941 * the smallest order which will fit the object.
1943 static inline int slab_order(int size
, int min_objects
,
1944 int max_order
, int fract_leftover
)
1948 int min_order
= slub_min_order
;
1950 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1951 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1953 for (order
= max(min_order
,
1954 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1955 order
<= max_order
; order
++) {
1957 unsigned long slab_size
= PAGE_SIZE
<< order
;
1959 if (slab_size
< min_objects
* size
)
1962 rem
= slab_size
% size
;
1964 if (rem
<= slab_size
/ fract_leftover
)
1972 static inline int calculate_order(int size
)
1980 * Attempt to find best configuration for a slab. This
1981 * works by first attempting to generate a layout with
1982 * the best configuration and backing off gradually.
1984 * First we reduce the acceptable waste in a slab. Then
1985 * we reduce the minimum objects required in a slab.
1987 min_objects
= slub_min_objects
;
1989 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1990 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1991 min_objects
= min(min_objects
, max_objects
);
1993 while (min_objects
> 1) {
1995 while (fraction
>= 4) {
1996 order
= slab_order(size
, min_objects
,
1997 slub_max_order
, fraction
);
1998 if (order
<= slub_max_order
)
2006 * We were unable to place multiple objects in a slab. Now
2007 * lets see if we can place a single object there.
2009 order
= slab_order(size
, 1, slub_max_order
, 1);
2010 if (order
<= slub_max_order
)
2014 * Doh this slab cannot be placed using slub_max_order.
2016 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2017 if (order
< MAX_ORDER
)
2023 * Figure out what the alignment of the objects will be.
2025 static unsigned long calculate_alignment(unsigned long flags
,
2026 unsigned long align
, unsigned long size
)
2029 * If the user wants hardware cache aligned objects then follow that
2030 * suggestion if the object is sufficiently large.
2032 * The hardware cache alignment cannot override the specified
2033 * alignment though. If that is greater then use it.
2035 if (flags
& SLAB_HWCACHE_ALIGN
) {
2036 unsigned long ralign
= cache_line_size();
2037 while (size
<= ralign
/ 2)
2039 align
= max(align
, ralign
);
2042 if (align
< ARCH_SLAB_MINALIGN
)
2043 align
= ARCH_SLAB_MINALIGN
;
2045 return ALIGN(align
, sizeof(void *));
2049 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2052 spin_lock_init(&n
->list_lock
);
2053 INIT_LIST_HEAD(&n
->partial
);
2054 #ifdef CONFIG_SLUB_DEBUG
2055 atomic_long_set(&n
->nr_slabs
, 0);
2056 atomic_long_set(&n
->total_objects
, 0);
2057 INIT_LIST_HEAD(&n
->full
);
2061 static DEFINE_PER_CPU(struct kmem_cache_cpu
, kmalloc_percpu
[KMALLOC_CACHES
]);
2063 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2065 if (s
< kmalloc_caches
+ KMALLOC_CACHES
&& s
>= kmalloc_caches
)
2067 * Boot time creation of the kmalloc array. Use static per cpu data
2068 * since the per cpu allocator is not available yet.
2070 s
->cpu_slab
= kmalloc_percpu
+ (s
- kmalloc_caches
);
2072 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2082 * No kmalloc_node yet so do it by hand. We know that this is the first
2083 * slab on the node for this slabcache. There are no concurrent accesses
2086 * Note that this function only works on the kmalloc_node_cache
2087 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2088 * memory on a fresh node that has no slab structures yet.
2090 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2093 struct kmem_cache_node
*n
;
2094 unsigned long flags
;
2096 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2098 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2101 if (page_to_nid(page
) != node
) {
2102 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2104 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2105 "in order to be able to continue\n");
2110 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2112 kmalloc_caches
->node
[node
] = n
;
2113 #ifdef CONFIG_SLUB_DEBUG
2114 init_object(kmalloc_caches
, n
, 1);
2115 init_tracking(kmalloc_caches
, n
);
2117 init_kmem_cache_node(n
, kmalloc_caches
);
2118 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2121 * lockdep requires consistent irq usage for each lock
2122 * so even though there cannot be a race this early in
2123 * the boot sequence, we still disable irqs.
2125 local_irq_save(flags
);
2126 add_partial(n
, page
, 0);
2127 local_irq_restore(flags
);
2130 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2134 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2135 struct kmem_cache_node
*n
= s
->node
[node
];
2136 if (n
&& n
!= &s
->local_node
)
2137 kmem_cache_free(kmalloc_caches
, n
);
2138 s
->node
[node
] = NULL
;
2142 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2147 if (slab_state
>= UP
&& (s
< kmalloc_caches
||
2148 s
>= kmalloc_caches
+ KMALLOC_CACHES
))
2149 local_node
= page_to_nid(virt_to_page(s
));
2153 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2154 struct kmem_cache_node
*n
;
2156 if (local_node
== node
)
2159 if (slab_state
== DOWN
) {
2160 early_kmem_cache_node_alloc(gfpflags
, node
);
2163 n
= kmem_cache_alloc_node(kmalloc_caches
,
2167 free_kmem_cache_nodes(s
);
2173 init_kmem_cache_node(n
, s
);
2178 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2182 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2184 init_kmem_cache_node(&s
->local_node
, s
);
2189 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2191 if (min
< MIN_PARTIAL
)
2193 else if (min
> MAX_PARTIAL
)
2195 s
->min_partial
= min
;
2199 * calculate_sizes() determines the order and the distribution of data within
2202 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2204 unsigned long flags
= s
->flags
;
2205 unsigned long size
= s
->objsize
;
2206 unsigned long align
= s
->align
;
2210 * Round up object size to the next word boundary. We can only
2211 * place the free pointer at word boundaries and this determines
2212 * the possible location of the free pointer.
2214 size
= ALIGN(size
, sizeof(void *));
2216 #ifdef CONFIG_SLUB_DEBUG
2218 * Determine if we can poison the object itself. If the user of
2219 * the slab may touch the object after free or before allocation
2220 * then we should never poison the object itself.
2222 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2224 s
->flags
|= __OBJECT_POISON
;
2226 s
->flags
&= ~__OBJECT_POISON
;
2230 * If we are Redzoning then check if there is some space between the
2231 * end of the object and the free pointer. If not then add an
2232 * additional word to have some bytes to store Redzone information.
2234 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2235 size
+= sizeof(void *);
2239 * With that we have determined the number of bytes in actual use
2240 * by the object. This is the potential offset to the free pointer.
2244 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2247 * Relocate free pointer after the object if it is not
2248 * permitted to overwrite the first word of the object on
2251 * This is the case if we do RCU, have a constructor or
2252 * destructor or are poisoning the objects.
2255 size
+= sizeof(void *);
2258 #ifdef CONFIG_SLUB_DEBUG
2259 if (flags
& SLAB_STORE_USER
)
2261 * Need to store information about allocs and frees after
2264 size
+= 2 * sizeof(struct track
);
2266 if (flags
& SLAB_RED_ZONE
)
2268 * Add some empty padding so that we can catch
2269 * overwrites from earlier objects rather than let
2270 * tracking information or the free pointer be
2271 * corrupted if a user writes before the start
2274 size
+= sizeof(void *);
2278 * Determine the alignment based on various parameters that the
2279 * user specified and the dynamic determination of cache line size
2282 align
= calculate_alignment(flags
, align
, s
->objsize
);
2286 * SLUB stores one object immediately after another beginning from
2287 * offset 0. In order to align the objects we have to simply size
2288 * each object to conform to the alignment.
2290 size
= ALIGN(size
, align
);
2292 if (forced_order
>= 0)
2293 order
= forced_order
;
2295 order
= calculate_order(size
);
2302 s
->allocflags
|= __GFP_COMP
;
2304 if (s
->flags
& SLAB_CACHE_DMA
)
2305 s
->allocflags
|= SLUB_DMA
;
2307 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2308 s
->allocflags
|= __GFP_RECLAIMABLE
;
2311 * Determine the number of objects per slab
2313 s
->oo
= oo_make(order
, size
);
2314 s
->min
= oo_make(get_order(size
), size
);
2315 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2318 return !!oo_objects(s
->oo
);
2322 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2323 const char *name
, size_t size
,
2324 size_t align
, unsigned long flags
,
2325 void (*ctor
)(void *))
2327 memset(s
, 0, kmem_size
);
2332 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2334 if (!calculate_sizes(s
, -1))
2336 if (disable_higher_order_debug
) {
2338 * Disable debugging flags that store metadata if the min slab
2341 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2342 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2344 if (!calculate_sizes(s
, -1))
2350 * The larger the object size is, the more pages we want on the partial
2351 * list to avoid pounding the page allocator excessively.
2353 set_min_partial(s
, ilog2(s
->size
));
2356 s
->remote_node_defrag_ratio
= 1000;
2358 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2361 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2364 free_kmem_cache_nodes(s
);
2366 if (flags
& SLAB_PANIC
)
2367 panic("Cannot create slab %s size=%lu realsize=%u "
2368 "order=%u offset=%u flags=%lx\n",
2369 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2375 * Check if a given pointer is valid
2377 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2381 if (!kern_ptr_validate(object
, s
->size
))
2384 page
= get_object_page(object
);
2386 if (!page
|| s
!= page
->slab
)
2387 /* No slab or wrong slab */
2390 if (!check_valid_pointer(s
, page
, object
))
2394 * We could also check if the object is on the slabs freelist.
2395 * But this would be too expensive and it seems that the main
2396 * purpose of kmem_ptr_valid() is to check if the object belongs
2397 * to a certain slab.
2401 EXPORT_SYMBOL(kmem_ptr_validate
);
2404 * Determine the size of a slab object
2406 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2410 EXPORT_SYMBOL(kmem_cache_size
);
2412 const char *kmem_cache_name(struct kmem_cache
*s
)
2416 EXPORT_SYMBOL(kmem_cache_name
);
2418 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2421 #ifdef CONFIG_SLUB_DEBUG
2422 void *addr
= page_address(page
);
2424 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2429 slab_err(s
, page
, "%s", text
);
2431 for_each_free_object(p
, s
, page
->freelist
)
2432 set_bit(slab_index(p
, s
, addr
), map
);
2434 for_each_object(p
, s
, addr
, page
->objects
) {
2436 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2437 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2439 print_tracking(s
, p
);
2448 * Attempt to free all partial slabs on a node.
2450 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2452 unsigned long flags
;
2453 struct page
*page
, *h
;
2455 spin_lock_irqsave(&n
->list_lock
, flags
);
2456 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2458 list_del(&page
->lru
);
2459 discard_slab(s
, page
);
2462 list_slab_objects(s
, page
,
2463 "Objects remaining on kmem_cache_close()");
2466 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2470 * Release all resources used by a slab cache.
2472 static inline int kmem_cache_close(struct kmem_cache
*s
)
2477 free_percpu(s
->cpu_slab
);
2478 /* Attempt to free all objects */
2479 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2480 struct kmem_cache_node
*n
= get_node(s
, node
);
2483 if (n
->nr_partial
|| slabs_node(s
, node
))
2486 free_kmem_cache_nodes(s
);
2491 * Close a cache and release the kmem_cache structure
2492 * (must be used for caches created using kmem_cache_create)
2494 void kmem_cache_destroy(struct kmem_cache
*s
)
2496 down_write(&slub_lock
);
2500 up_write(&slub_lock
);
2501 if (kmem_cache_close(s
)) {
2502 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2503 "still has objects.\n", s
->name
, __func__
);
2506 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2508 sysfs_slab_remove(s
);
2510 up_write(&slub_lock
);
2512 EXPORT_SYMBOL(kmem_cache_destroy
);
2514 /********************************************************************
2516 *******************************************************************/
2518 struct kmem_cache kmalloc_caches
[KMALLOC_CACHES
] __cacheline_aligned
;
2519 EXPORT_SYMBOL(kmalloc_caches
);
2521 static int __init
setup_slub_min_order(char *str
)
2523 get_option(&str
, &slub_min_order
);
2528 __setup("slub_min_order=", setup_slub_min_order
);
2530 static int __init
setup_slub_max_order(char *str
)
2532 get_option(&str
, &slub_max_order
);
2533 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2538 __setup("slub_max_order=", setup_slub_max_order
);
2540 static int __init
setup_slub_min_objects(char *str
)
2542 get_option(&str
, &slub_min_objects
);
2547 __setup("slub_min_objects=", setup_slub_min_objects
);
2549 static int __init
setup_slub_nomerge(char *str
)
2555 __setup("slub_nomerge", setup_slub_nomerge
);
2557 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2558 const char *name
, int size
, gfp_t gfp_flags
)
2560 unsigned int flags
= 0;
2562 if (gfp_flags
& SLUB_DMA
)
2563 flags
= SLAB_CACHE_DMA
;
2566 * This function is called with IRQs disabled during early-boot on
2567 * single CPU so there's no need to take slub_lock here.
2569 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2573 list_add(&s
->list
, &slab_caches
);
2575 if (sysfs_slab_add(s
))
2580 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2583 #ifdef CONFIG_ZONE_DMA
2584 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2586 static void sysfs_add_func(struct work_struct
*w
)
2588 struct kmem_cache
*s
;
2590 down_write(&slub_lock
);
2591 list_for_each_entry(s
, &slab_caches
, list
) {
2592 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2593 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2597 up_write(&slub_lock
);
2600 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2602 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2604 struct kmem_cache
*s
;
2607 unsigned long slabflags
;
2610 s
= kmalloc_caches_dma
[index
];
2614 /* Dynamically create dma cache */
2615 if (flags
& __GFP_WAIT
)
2616 down_write(&slub_lock
);
2618 if (!down_write_trylock(&slub_lock
))
2622 if (kmalloc_caches_dma
[index
])
2625 realsize
= kmalloc_caches
[index
].objsize
;
2626 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2627 (unsigned int)realsize
);
2630 for (i
= 0; i
< KMALLOC_CACHES
; i
++)
2631 if (!kmalloc_caches
[i
].size
)
2634 BUG_ON(i
>= KMALLOC_CACHES
);
2635 s
= kmalloc_caches
+ i
;
2638 * Must defer sysfs creation to a workqueue because we don't know
2639 * what context we are called from. Before sysfs comes up, we don't
2640 * need to do anything because our sysfs initcall will start by
2641 * adding all existing slabs to sysfs.
2643 slabflags
= SLAB_CACHE_DMA
|SLAB_NOTRACK
;
2644 if (slab_state
>= SYSFS
)
2645 slabflags
|= __SYSFS_ADD_DEFERRED
;
2647 if (!text
|| !kmem_cache_open(s
, flags
, text
,
2648 realsize
, ARCH_KMALLOC_MINALIGN
, slabflags
, NULL
)) {
2654 list_add(&s
->list
, &slab_caches
);
2655 kmalloc_caches_dma
[index
] = s
;
2657 if (slab_state
>= SYSFS
)
2658 schedule_work(&sysfs_add_work
);
2661 up_write(&slub_lock
);
2663 return kmalloc_caches_dma
[index
];
2668 * Conversion table for small slabs sizes / 8 to the index in the
2669 * kmalloc array. This is necessary for slabs < 192 since we have non power
2670 * of two cache sizes there. The size of larger slabs can be determined using
2673 static s8 size_index
[24] = {
2700 static inline int size_index_elem(size_t bytes
)
2702 return (bytes
- 1) / 8;
2705 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2711 return ZERO_SIZE_PTR
;
2713 index
= size_index
[size_index_elem(size
)];
2715 index
= fls(size
- 1);
2717 #ifdef CONFIG_ZONE_DMA
2718 if (unlikely((flags
& SLUB_DMA
)))
2719 return dma_kmalloc_cache(index
, flags
);
2722 return &kmalloc_caches
[index
];
2725 void *__kmalloc(size_t size
, gfp_t flags
)
2727 struct kmem_cache
*s
;
2730 if (unlikely(size
> SLUB_MAX_SIZE
))
2731 return kmalloc_large(size
, flags
);
2733 s
= get_slab(size
, flags
);
2735 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2738 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2740 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2744 EXPORT_SYMBOL(__kmalloc
);
2746 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2751 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2752 page
= alloc_pages_node(node
, flags
, get_order(size
));
2754 ptr
= page_address(page
);
2756 kmemleak_alloc(ptr
, size
, 1, flags
);
2761 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2763 struct kmem_cache
*s
;
2766 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2767 ret
= kmalloc_large_node(size
, flags
, node
);
2769 trace_kmalloc_node(_RET_IP_
, ret
,
2770 size
, PAGE_SIZE
<< get_order(size
),
2776 s
= get_slab(size
, flags
);
2778 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2781 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2783 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2787 EXPORT_SYMBOL(__kmalloc_node
);
2790 size_t ksize(const void *object
)
2793 struct kmem_cache
*s
;
2795 if (unlikely(object
== ZERO_SIZE_PTR
))
2798 page
= virt_to_head_page(object
);
2800 if (unlikely(!PageSlab(page
))) {
2801 WARN_ON(!PageCompound(page
));
2802 return PAGE_SIZE
<< compound_order(page
);
2806 #ifdef CONFIG_SLUB_DEBUG
2808 * Debugging requires use of the padding between object
2809 * and whatever may come after it.
2811 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2816 * If we have the need to store the freelist pointer
2817 * back there or track user information then we can
2818 * only use the space before that information.
2820 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2823 * Else we can use all the padding etc for the allocation
2827 EXPORT_SYMBOL(ksize
);
2829 void kfree(const void *x
)
2832 void *object
= (void *)x
;
2834 trace_kfree(_RET_IP_
, x
);
2836 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2839 page
= virt_to_head_page(x
);
2840 if (unlikely(!PageSlab(page
))) {
2841 BUG_ON(!PageCompound(page
));
2846 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2848 EXPORT_SYMBOL(kfree
);
2851 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2852 * the remaining slabs by the number of items in use. The slabs with the
2853 * most items in use come first. New allocations will then fill those up
2854 * and thus they can be removed from the partial lists.
2856 * The slabs with the least items are placed last. This results in them
2857 * being allocated from last increasing the chance that the last objects
2858 * are freed in them.
2860 int kmem_cache_shrink(struct kmem_cache
*s
)
2864 struct kmem_cache_node
*n
;
2867 int objects
= oo_objects(s
->max
);
2868 struct list_head
*slabs_by_inuse
=
2869 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2870 unsigned long flags
;
2872 if (!slabs_by_inuse
)
2876 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2877 n
= get_node(s
, node
);
2882 for (i
= 0; i
< objects
; i
++)
2883 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2885 spin_lock_irqsave(&n
->list_lock
, flags
);
2888 * Build lists indexed by the items in use in each slab.
2890 * Note that concurrent frees may occur while we hold the
2891 * list_lock. page->inuse here is the upper limit.
2893 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2894 if (!page
->inuse
&& slab_trylock(page
)) {
2896 * Must hold slab lock here because slab_free
2897 * may have freed the last object and be
2898 * waiting to release the slab.
2900 list_del(&page
->lru
);
2903 discard_slab(s
, page
);
2905 list_move(&page
->lru
,
2906 slabs_by_inuse
+ page
->inuse
);
2911 * Rebuild the partial list with the slabs filled up most
2912 * first and the least used slabs at the end.
2914 for (i
= objects
- 1; i
>= 0; i
--)
2915 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2917 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2920 kfree(slabs_by_inuse
);
2923 EXPORT_SYMBOL(kmem_cache_shrink
);
2925 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2926 static int slab_mem_going_offline_callback(void *arg
)
2928 struct kmem_cache
*s
;
2930 down_read(&slub_lock
);
2931 list_for_each_entry(s
, &slab_caches
, list
)
2932 kmem_cache_shrink(s
);
2933 up_read(&slub_lock
);
2938 static void slab_mem_offline_callback(void *arg
)
2940 struct kmem_cache_node
*n
;
2941 struct kmem_cache
*s
;
2942 struct memory_notify
*marg
= arg
;
2945 offline_node
= marg
->status_change_nid
;
2948 * If the node still has available memory. we need kmem_cache_node
2951 if (offline_node
< 0)
2954 down_read(&slub_lock
);
2955 list_for_each_entry(s
, &slab_caches
, list
) {
2956 n
= get_node(s
, offline_node
);
2959 * if n->nr_slabs > 0, slabs still exist on the node
2960 * that is going down. We were unable to free them,
2961 * and offline_pages() function shouldn't call this
2962 * callback. So, we must fail.
2964 BUG_ON(slabs_node(s
, offline_node
));
2966 s
->node
[offline_node
] = NULL
;
2967 kmem_cache_free(kmalloc_caches
, n
);
2970 up_read(&slub_lock
);
2973 static int slab_mem_going_online_callback(void *arg
)
2975 struct kmem_cache_node
*n
;
2976 struct kmem_cache
*s
;
2977 struct memory_notify
*marg
= arg
;
2978 int nid
= marg
->status_change_nid
;
2982 * If the node's memory is already available, then kmem_cache_node is
2983 * already created. Nothing to do.
2989 * We are bringing a node online. No memory is available yet. We must
2990 * allocate a kmem_cache_node structure in order to bring the node
2993 down_read(&slub_lock
);
2994 list_for_each_entry(s
, &slab_caches
, list
) {
2996 * XXX: kmem_cache_alloc_node will fallback to other nodes
2997 * since memory is not yet available from the node that
3000 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
3005 init_kmem_cache_node(n
, s
);
3009 up_read(&slub_lock
);
3013 static int slab_memory_callback(struct notifier_block
*self
,
3014 unsigned long action
, void *arg
)
3019 case MEM_GOING_ONLINE
:
3020 ret
= slab_mem_going_online_callback(arg
);
3022 case MEM_GOING_OFFLINE
:
3023 ret
= slab_mem_going_offline_callback(arg
);
3026 case MEM_CANCEL_ONLINE
:
3027 slab_mem_offline_callback(arg
);
3030 case MEM_CANCEL_OFFLINE
:
3034 ret
= notifier_from_errno(ret
);
3040 #endif /* CONFIG_MEMORY_HOTPLUG */
3042 /********************************************************************
3043 * Basic setup of slabs
3044 *******************************************************************/
3046 void __init
kmem_cache_init(void)
3053 * Must first have the slab cache available for the allocations of the
3054 * struct kmem_cache_node's. There is special bootstrap code in
3055 * kmem_cache_open for slab_state == DOWN.
3057 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3058 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3059 kmalloc_caches
[0].refcount
= -1;
3062 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3065 /* Able to allocate the per node structures */
3066 slab_state
= PARTIAL
;
3068 /* Caches that are not of the two-to-the-power-of size */
3069 if (KMALLOC_MIN_SIZE
<= 32) {
3070 create_kmalloc_cache(&kmalloc_caches
[1],
3071 "kmalloc-96", 96, GFP_NOWAIT
);
3074 if (KMALLOC_MIN_SIZE
<= 64) {
3075 create_kmalloc_cache(&kmalloc_caches
[2],
3076 "kmalloc-192", 192, GFP_NOWAIT
);
3080 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3081 create_kmalloc_cache(&kmalloc_caches
[i
],
3082 "kmalloc", 1 << i
, GFP_NOWAIT
);
3088 * Patch up the size_index table if we have strange large alignment
3089 * requirements for the kmalloc array. This is only the case for
3090 * MIPS it seems. The standard arches will not generate any code here.
3092 * Largest permitted alignment is 256 bytes due to the way we
3093 * handle the index determination for the smaller caches.
3095 * Make sure that nothing crazy happens if someone starts tinkering
3096 * around with ARCH_KMALLOC_MINALIGN
3098 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3099 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3101 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3102 int elem
= size_index_elem(i
);
3103 if (elem
>= ARRAY_SIZE(size_index
))
3105 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3108 if (KMALLOC_MIN_SIZE
== 64) {
3110 * The 96 byte size cache is not used if the alignment
3113 for (i
= 64 + 8; i
<= 96; i
+= 8)
3114 size_index
[size_index_elem(i
)] = 7;
3115 } else if (KMALLOC_MIN_SIZE
== 128) {
3117 * The 192 byte sized cache is not used if the alignment
3118 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3121 for (i
= 128 + 8; i
<= 192; i
+= 8)
3122 size_index
[size_index_elem(i
)] = 8;
3127 /* Provide the correct kmalloc names now that the caches are up */
3128 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3129 kmalloc_caches
[i
]. name
=
3130 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3133 register_cpu_notifier(&slab_notifier
);
3136 kmem_size
= offsetof(struct kmem_cache
, node
) +
3137 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3139 kmem_size
= sizeof(struct kmem_cache
);
3143 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3144 " CPUs=%d, Nodes=%d\n",
3145 caches
, cache_line_size(),
3146 slub_min_order
, slub_max_order
, slub_min_objects
,
3147 nr_cpu_ids
, nr_node_ids
);
3150 void __init
kmem_cache_init_late(void)
3155 * Find a mergeable slab cache
3157 static int slab_unmergeable(struct kmem_cache
*s
)
3159 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3166 * We may have set a slab to be unmergeable during bootstrap.
3168 if (s
->refcount
< 0)
3174 static struct kmem_cache
*find_mergeable(size_t size
,
3175 size_t align
, unsigned long flags
, const char *name
,
3176 void (*ctor
)(void *))
3178 struct kmem_cache
*s
;
3180 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3186 size
= ALIGN(size
, sizeof(void *));
3187 align
= calculate_alignment(flags
, align
, size
);
3188 size
= ALIGN(size
, align
);
3189 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3191 list_for_each_entry(s
, &slab_caches
, list
) {
3192 if (slab_unmergeable(s
))
3198 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3201 * Check if alignment is compatible.
3202 * Courtesy of Adrian Drzewiecki
3204 if ((s
->size
& ~(align
- 1)) != s
->size
)
3207 if (s
->size
- size
>= sizeof(void *))
3215 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3216 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3218 struct kmem_cache
*s
;
3223 down_write(&slub_lock
);
3224 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3228 * Adjust the object sizes so that we clear
3229 * the complete object on kzalloc.
3231 s
->objsize
= max(s
->objsize
, (int)size
);
3232 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3233 up_write(&slub_lock
);
3235 if (sysfs_slab_alias(s
, name
)) {
3236 down_write(&slub_lock
);
3238 up_write(&slub_lock
);
3244 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3246 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3247 size
, align
, flags
, ctor
)) {
3248 list_add(&s
->list
, &slab_caches
);
3249 up_write(&slub_lock
);
3250 if (sysfs_slab_add(s
)) {
3251 down_write(&slub_lock
);
3253 up_write(&slub_lock
);
3261 up_write(&slub_lock
);
3264 if (flags
& SLAB_PANIC
)
3265 panic("Cannot create slabcache %s\n", name
);
3270 EXPORT_SYMBOL(kmem_cache_create
);
3274 * Use the cpu notifier to insure that the cpu slabs are flushed when
3277 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3278 unsigned long action
, void *hcpu
)
3280 long cpu
= (long)hcpu
;
3281 struct kmem_cache
*s
;
3282 unsigned long flags
;
3285 case CPU_UP_CANCELED
:
3286 case CPU_UP_CANCELED_FROZEN
:
3288 case CPU_DEAD_FROZEN
:
3289 down_read(&slub_lock
);
3290 list_for_each_entry(s
, &slab_caches
, list
) {
3291 local_irq_save(flags
);
3292 __flush_cpu_slab(s
, cpu
);
3293 local_irq_restore(flags
);
3295 up_read(&slub_lock
);
3303 static struct notifier_block __cpuinitdata slab_notifier
= {
3304 .notifier_call
= slab_cpuup_callback
3309 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3311 struct kmem_cache
*s
;
3314 if (unlikely(size
> SLUB_MAX_SIZE
))
3315 return kmalloc_large(size
, gfpflags
);
3317 s
= get_slab(size
, gfpflags
);
3319 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3322 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3324 /* Honor the call site pointer we recieved. */
3325 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3330 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3331 int node
, unsigned long caller
)
3333 struct kmem_cache
*s
;
3336 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3337 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3339 trace_kmalloc_node(caller
, ret
,
3340 size
, PAGE_SIZE
<< get_order(size
),
3346 s
= get_slab(size
, gfpflags
);
3348 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3351 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3353 /* Honor the call site pointer we recieved. */
3354 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3359 #ifdef CONFIG_SLUB_DEBUG
3360 static int count_inuse(struct page
*page
)
3365 static int count_total(struct page
*page
)
3367 return page
->objects
;
3370 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3374 void *addr
= page_address(page
);
3376 if (!check_slab(s
, page
) ||
3377 !on_freelist(s
, page
, NULL
))
3380 /* Now we know that a valid freelist exists */
3381 bitmap_zero(map
, page
->objects
);
3383 for_each_free_object(p
, s
, page
->freelist
) {
3384 set_bit(slab_index(p
, s
, addr
), map
);
3385 if (!check_object(s
, page
, p
, 0))
3389 for_each_object(p
, s
, addr
, page
->objects
)
3390 if (!test_bit(slab_index(p
, s
, addr
), map
))
3391 if (!check_object(s
, page
, p
, 1))
3396 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3399 if (slab_trylock(page
)) {
3400 validate_slab(s
, page
, map
);
3403 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3406 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3407 if (!PageSlubDebug(page
))
3408 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3409 "on slab 0x%p\n", s
->name
, page
);
3411 if (PageSlubDebug(page
))
3412 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3413 "slab 0x%p\n", s
->name
, page
);
3417 static int validate_slab_node(struct kmem_cache
*s
,
3418 struct kmem_cache_node
*n
, unsigned long *map
)
3420 unsigned long count
= 0;
3422 unsigned long flags
;
3424 spin_lock_irqsave(&n
->list_lock
, flags
);
3426 list_for_each_entry(page
, &n
->partial
, lru
) {
3427 validate_slab_slab(s
, page
, map
);
3430 if (count
!= n
->nr_partial
)
3431 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3432 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3434 if (!(s
->flags
& SLAB_STORE_USER
))
3437 list_for_each_entry(page
, &n
->full
, lru
) {
3438 validate_slab_slab(s
, page
, map
);
3441 if (count
!= atomic_long_read(&n
->nr_slabs
))
3442 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3443 "counter=%ld\n", s
->name
, count
,
3444 atomic_long_read(&n
->nr_slabs
));
3447 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3451 static long validate_slab_cache(struct kmem_cache
*s
)
3454 unsigned long count
= 0;
3455 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3456 sizeof(unsigned long), GFP_KERNEL
);
3462 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3463 struct kmem_cache_node
*n
= get_node(s
, node
);
3465 count
+= validate_slab_node(s
, n
, map
);
3471 #ifdef SLUB_RESILIENCY_TEST
3472 static void resiliency_test(void)
3476 printk(KERN_ERR
"SLUB resiliency testing\n");
3477 printk(KERN_ERR
"-----------------------\n");
3478 printk(KERN_ERR
"A. Corruption after allocation\n");
3480 p
= kzalloc(16, GFP_KERNEL
);
3482 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3483 " 0x12->0x%p\n\n", p
+ 16);
3485 validate_slab_cache(kmalloc_caches
+ 4);
3487 /* Hmmm... The next two are dangerous */
3488 p
= kzalloc(32, GFP_KERNEL
);
3489 p
[32 + sizeof(void *)] = 0x34;
3490 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3491 " 0x34 -> -0x%p\n", p
);
3493 "If allocated object is overwritten then not detectable\n\n");
3495 validate_slab_cache(kmalloc_caches
+ 5);
3496 p
= kzalloc(64, GFP_KERNEL
);
3497 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3499 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3502 "If allocated object is overwritten then not detectable\n\n");
3503 validate_slab_cache(kmalloc_caches
+ 6);
3505 printk(KERN_ERR
"\nB. Corruption after free\n");
3506 p
= kzalloc(128, GFP_KERNEL
);
3509 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3510 validate_slab_cache(kmalloc_caches
+ 7);
3512 p
= kzalloc(256, GFP_KERNEL
);
3515 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3517 validate_slab_cache(kmalloc_caches
+ 8);
3519 p
= kzalloc(512, GFP_KERNEL
);
3522 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3523 validate_slab_cache(kmalloc_caches
+ 9);
3526 static void resiliency_test(void) {};
3530 * Generate lists of code addresses where slabcache objects are allocated
3535 unsigned long count
;
3542 DECLARE_BITMAP(cpus
, NR_CPUS
);
3548 unsigned long count
;
3549 struct location
*loc
;
3552 static void free_loc_track(struct loc_track
*t
)
3555 free_pages((unsigned long)t
->loc
,
3556 get_order(sizeof(struct location
) * t
->max
));
3559 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3564 order
= get_order(sizeof(struct location
) * max
);
3566 l
= (void *)__get_free_pages(flags
, order
);
3571 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3579 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3580 const struct track
*track
)
3582 long start
, end
, pos
;
3584 unsigned long caddr
;
3585 unsigned long age
= jiffies
- track
->when
;
3591 pos
= start
+ (end
- start
+ 1) / 2;
3594 * There is nothing at "end". If we end up there
3595 * we need to add something to before end.
3600 caddr
= t
->loc
[pos
].addr
;
3601 if (track
->addr
== caddr
) {
3607 if (age
< l
->min_time
)
3609 if (age
> l
->max_time
)
3612 if (track
->pid
< l
->min_pid
)
3613 l
->min_pid
= track
->pid
;
3614 if (track
->pid
> l
->max_pid
)
3615 l
->max_pid
= track
->pid
;
3617 cpumask_set_cpu(track
->cpu
,
3618 to_cpumask(l
->cpus
));
3620 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3624 if (track
->addr
< caddr
)
3631 * Not found. Insert new tracking element.
3633 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3639 (t
->count
- pos
) * sizeof(struct location
));
3642 l
->addr
= track
->addr
;
3646 l
->min_pid
= track
->pid
;
3647 l
->max_pid
= track
->pid
;
3648 cpumask_clear(to_cpumask(l
->cpus
));
3649 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3650 nodes_clear(l
->nodes
);
3651 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3655 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3656 struct page
*page
, enum track_item alloc
,
3659 void *addr
= page_address(page
);
3662 bitmap_zero(map
, page
->objects
);
3663 for_each_free_object(p
, s
, page
->freelist
)
3664 set_bit(slab_index(p
, s
, addr
), map
);
3666 for_each_object(p
, s
, addr
, page
->objects
)
3667 if (!test_bit(slab_index(p
, s
, addr
), map
))
3668 add_location(t
, s
, get_track(s
, p
, alloc
));
3671 static int list_locations(struct kmem_cache
*s
, char *buf
,
3672 enum track_item alloc
)
3676 struct loc_track t
= { 0, 0, NULL
};
3678 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3679 sizeof(unsigned long), GFP_KERNEL
);
3681 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3684 return sprintf(buf
, "Out of memory\n");
3686 /* Push back cpu slabs */
3689 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3690 struct kmem_cache_node
*n
= get_node(s
, node
);
3691 unsigned long flags
;
3694 if (!atomic_long_read(&n
->nr_slabs
))
3697 spin_lock_irqsave(&n
->list_lock
, flags
);
3698 list_for_each_entry(page
, &n
->partial
, lru
)
3699 process_slab(&t
, s
, page
, alloc
, map
);
3700 list_for_each_entry(page
, &n
->full
, lru
)
3701 process_slab(&t
, s
, page
, alloc
, map
);
3702 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3705 for (i
= 0; i
< t
.count
; i
++) {
3706 struct location
*l
= &t
.loc
[i
];
3708 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3710 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3713 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3715 len
+= sprintf(buf
+ len
, "<not-available>");
3717 if (l
->sum_time
!= l
->min_time
) {
3718 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3720 (long)div_u64(l
->sum_time
, l
->count
),
3723 len
+= sprintf(buf
+ len
, " age=%ld",
3726 if (l
->min_pid
!= l
->max_pid
)
3727 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3728 l
->min_pid
, l
->max_pid
);
3730 len
+= sprintf(buf
+ len
, " pid=%ld",
3733 if (num_online_cpus() > 1 &&
3734 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3735 len
< PAGE_SIZE
- 60) {
3736 len
+= sprintf(buf
+ len
, " cpus=");
3737 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3738 to_cpumask(l
->cpus
));
3741 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3742 len
< PAGE_SIZE
- 60) {
3743 len
+= sprintf(buf
+ len
, " nodes=");
3744 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3748 len
+= sprintf(buf
+ len
, "\n");
3754 len
+= sprintf(buf
, "No data\n");
3758 enum slab_stat_type
{
3759 SL_ALL
, /* All slabs */
3760 SL_PARTIAL
, /* Only partially allocated slabs */
3761 SL_CPU
, /* Only slabs used for cpu caches */
3762 SL_OBJECTS
, /* Determine allocated objects not slabs */
3763 SL_TOTAL
/* Determine object capacity not slabs */
3766 #define SO_ALL (1 << SL_ALL)
3767 #define SO_PARTIAL (1 << SL_PARTIAL)
3768 #define SO_CPU (1 << SL_CPU)
3769 #define SO_OBJECTS (1 << SL_OBJECTS)
3770 #define SO_TOTAL (1 << SL_TOTAL)
3772 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3773 char *buf
, unsigned long flags
)
3775 unsigned long total
= 0;
3778 unsigned long *nodes
;
3779 unsigned long *per_cpu
;
3781 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3784 per_cpu
= nodes
+ nr_node_ids
;
3786 if (flags
& SO_CPU
) {
3789 for_each_possible_cpu(cpu
) {
3790 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3792 if (!c
|| c
->node
< 0)
3796 if (flags
& SO_TOTAL
)
3797 x
= c
->page
->objects
;
3798 else if (flags
& SO_OBJECTS
)
3804 nodes
[c
->node
] += x
;
3810 if (flags
& SO_ALL
) {
3811 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3812 struct kmem_cache_node
*n
= get_node(s
, node
);
3814 if (flags
& SO_TOTAL
)
3815 x
= atomic_long_read(&n
->total_objects
);
3816 else if (flags
& SO_OBJECTS
)
3817 x
= atomic_long_read(&n
->total_objects
) -
3818 count_partial(n
, count_free
);
3821 x
= atomic_long_read(&n
->nr_slabs
);
3826 } else if (flags
& SO_PARTIAL
) {
3827 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3828 struct kmem_cache_node
*n
= get_node(s
, node
);
3830 if (flags
& SO_TOTAL
)
3831 x
= count_partial(n
, count_total
);
3832 else if (flags
& SO_OBJECTS
)
3833 x
= count_partial(n
, count_inuse
);
3840 x
= sprintf(buf
, "%lu", total
);
3842 for_each_node_state(node
, N_NORMAL_MEMORY
)
3844 x
+= sprintf(buf
+ x
, " N%d=%lu",
3848 return x
+ sprintf(buf
+ x
, "\n");
3851 static int any_slab_objects(struct kmem_cache
*s
)
3855 for_each_online_node(node
) {
3856 struct kmem_cache_node
*n
= get_node(s
, node
);
3861 if (atomic_long_read(&n
->total_objects
))
3867 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3868 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3870 struct slab_attribute
{
3871 struct attribute attr
;
3872 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3873 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3876 #define SLAB_ATTR_RO(_name) \
3877 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3879 #define SLAB_ATTR(_name) \
3880 static struct slab_attribute _name##_attr = \
3881 __ATTR(_name, 0644, _name##_show, _name##_store)
3883 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3885 return sprintf(buf
, "%d\n", s
->size
);
3887 SLAB_ATTR_RO(slab_size
);
3889 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3891 return sprintf(buf
, "%d\n", s
->align
);
3893 SLAB_ATTR_RO(align
);
3895 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3897 return sprintf(buf
, "%d\n", s
->objsize
);
3899 SLAB_ATTR_RO(object_size
);
3901 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3903 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3905 SLAB_ATTR_RO(objs_per_slab
);
3907 static ssize_t
order_store(struct kmem_cache
*s
,
3908 const char *buf
, size_t length
)
3910 unsigned long order
;
3913 err
= strict_strtoul(buf
, 10, &order
);
3917 if (order
> slub_max_order
|| order
< slub_min_order
)
3920 calculate_sizes(s
, order
);
3924 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3926 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3930 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3932 return sprintf(buf
, "%lu\n", s
->min_partial
);
3935 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3941 err
= strict_strtoul(buf
, 10, &min
);
3945 set_min_partial(s
, min
);
3948 SLAB_ATTR(min_partial
);
3950 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3953 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3955 return n
+ sprintf(buf
+ n
, "\n");
3961 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3963 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3965 SLAB_ATTR_RO(aliases
);
3967 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3969 return show_slab_objects(s
, buf
, SO_ALL
);
3971 SLAB_ATTR_RO(slabs
);
3973 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3975 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3977 SLAB_ATTR_RO(partial
);
3979 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3981 return show_slab_objects(s
, buf
, SO_CPU
);
3983 SLAB_ATTR_RO(cpu_slabs
);
3985 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3987 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3989 SLAB_ATTR_RO(objects
);
3991 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3993 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3995 SLAB_ATTR_RO(objects_partial
);
3997 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3999 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4001 SLAB_ATTR_RO(total_objects
);
4003 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4005 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4008 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4009 const char *buf
, size_t length
)
4011 s
->flags
&= ~SLAB_DEBUG_FREE
;
4013 s
->flags
|= SLAB_DEBUG_FREE
;
4016 SLAB_ATTR(sanity_checks
);
4018 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4020 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4023 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4026 s
->flags
&= ~SLAB_TRACE
;
4028 s
->flags
|= SLAB_TRACE
;
4033 #ifdef CONFIG_FAILSLAB
4034 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4036 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4039 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4042 s
->flags
&= ~SLAB_FAILSLAB
;
4044 s
->flags
|= SLAB_FAILSLAB
;
4047 SLAB_ATTR(failslab
);
4050 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4052 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4055 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4056 const char *buf
, size_t length
)
4058 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4060 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4063 SLAB_ATTR(reclaim_account
);
4065 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4067 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4069 SLAB_ATTR_RO(hwcache_align
);
4071 #ifdef CONFIG_ZONE_DMA
4072 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4074 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4076 SLAB_ATTR_RO(cache_dma
);
4079 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4081 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4083 SLAB_ATTR_RO(destroy_by_rcu
);
4085 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4087 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4090 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4091 const char *buf
, size_t length
)
4093 if (any_slab_objects(s
))
4096 s
->flags
&= ~SLAB_RED_ZONE
;
4098 s
->flags
|= SLAB_RED_ZONE
;
4099 calculate_sizes(s
, -1);
4102 SLAB_ATTR(red_zone
);
4104 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4106 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4109 static ssize_t
poison_store(struct kmem_cache
*s
,
4110 const char *buf
, size_t length
)
4112 if (any_slab_objects(s
))
4115 s
->flags
&= ~SLAB_POISON
;
4117 s
->flags
|= SLAB_POISON
;
4118 calculate_sizes(s
, -1);
4123 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4125 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4128 static ssize_t
store_user_store(struct kmem_cache
*s
,
4129 const char *buf
, size_t length
)
4131 if (any_slab_objects(s
))
4134 s
->flags
&= ~SLAB_STORE_USER
;
4136 s
->flags
|= SLAB_STORE_USER
;
4137 calculate_sizes(s
, -1);
4140 SLAB_ATTR(store_user
);
4142 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4147 static ssize_t
validate_store(struct kmem_cache
*s
,
4148 const char *buf
, size_t length
)
4152 if (buf
[0] == '1') {
4153 ret
= validate_slab_cache(s
);
4159 SLAB_ATTR(validate
);
4161 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4166 static ssize_t
shrink_store(struct kmem_cache
*s
,
4167 const char *buf
, size_t length
)
4169 if (buf
[0] == '1') {
4170 int rc
= kmem_cache_shrink(s
);
4180 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4182 if (!(s
->flags
& SLAB_STORE_USER
))
4184 return list_locations(s
, buf
, TRACK_ALLOC
);
4186 SLAB_ATTR_RO(alloc_calls
);
4188 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4190 if (!(s
->flags
& SLAB_STORE_USER
))
4192 return list_locations(s
, buf
, TRACK_FREE
);
4194 SLAB_ATTR_RO(free_calls
);
4197 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4199 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4202 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4203 const char *buf
, size_t length
)
4205 unsigned long ratio
;
4208 err
= strict_strtoul(buf
, 10, &ratio
);
4213 s
->remote_node_defrag_ratio
= ratio
* 10;
4217 SLAB_ATTR(remote_node_defrag_ratio
);
4220 #ifdef CONFIG_SLUB_STATS
4221 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4223 unsigned long sum
= 0;
4226 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4231 for_each_online_cpu(cpu
) {
4232 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4238 len
= sprintf(buf
, "%lu", sum
);
4241 for_each_online_cpu(cpu
) {
4242 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4243 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4247 return len
+ sprintf(buf
+ len
, "\n");
4250 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4254 for_each_online_cpu(cpu
)
4255 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4258 #define STAT_ATTR(si, text) \
4259 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4261 return show_stat(s, buf, si); \
4263 static ssize_t text##_store(struct kmem_cache *s, \
4264 const char *buf, size_t length) \
4266 if (buf[0] != '0') \
4268 clear_stat(s, si); \
4273 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4274 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4275 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4276 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4277 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4278 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4279 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4280 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4281 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4282 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4283 STAT_ATTR(FREE_SLAB
, free_slab
);
4284 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4285 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4286 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4287 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4288 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4289 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4290 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4293 static struct attribute
*slab_attrs
[] = {
4294 &slab_size_attr
.attr
,
4295 &object_size_attr
.attr
,
4296 &objs_per_slab_attr
.attr
,
4298 &min_partial_attr
.attr
,
4300 &objects_partial_attr
.attr
,
4301 &total_objects_attr
.attr
,
4304 &cpu_slabs_attr
.attr
,
4308 &sanity_checks_attr
.attr
,
4310 &hwcache_align_attr
.attr
,
4311 &reclaim_account_attr
.attr
,
4312 &destroy_by_rcu_attr
.attr
,
4313 &red_zone_attr
.attr
,
4315 &store_user_attr
.attr
,
4316 &validate_attr
.attr
,
4318 &alloc_calls_attr
.attr
,
4319 &free_calls_attr
.attr
,
4320 #ifdef CONFIG_ZONE_DMA
4321 &cache_dma_attr
.attr
,
4324 &remote_node_defrag_ratio_attr
.attr
,
4326 #ifdef CONFIG_SLUB_STATS
4327 &alloc_fastpath_attr
.attr
,
4328 &alloc_slowpath_attr
.attr
,
4329 &free_fastpath_attr
.attr
,
4330 &free_slowpath_attr
.attr
,
4331 &free_frozen_attr
.attr
,
4332 &free_add_partial_attr
.attr
,
4333 &free_remove_partial_attr
.attr
,
4334 &alloc_from_partial_attr
.attr
,
4335 &alloc_slab_attr
.attr
,
4336 &alloc_refill_attr
.attr
,
4337 &free_slab_attr
.attr
,
4338 &cpuslab_flush_attr
.attr
,
4339 &deactivate_full_attr
.attr
,
4340 &deactivate_empty_attr
.attr
,
4341 &deactivate_to_head_attr
.attr
,
4342 &deactivate_to_tail_attr
.attr
,
4343 &deactivate_remote_frees_attr
.attr
,
4344 &order_fallback_attr
.attr
,
4346 #ifdef CONFIG_FAILSLAB
4347 &failslab_attr
.attr
,
4353 static struct attribute_group slab_attr_group
= {
4354 .attrs
= slab_attrs
,
4357 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4358 struct attribute
*attr
,
4361 struct slab_attribute
*attribute
;
4362 struct kmem_cache
*s
;
4365 attribute
= to_slab_attr(attr
);
4368 if (!attribute
->show
)
4371 err
= attribute
->show(s
, buf
);
4376 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4377 struct attribute
*attr
,
4378 const char *buf
, size_t len
)
4380 struct slab_attribute
*attribute
;
4381 struct kmem_cache
*s
;
4384 attribute
= to_slab_attr(attr
);
4387 if (!attribute
->store
)
4390 err
= attribute
->store(s
, buf
, len
);
4395 static void kmem_cache_release(struct kobject
*kobj
)
4397 struct kmem_cache
*s
= to_slab(kobj
);
4402 static const struct sysfs_ops slab_sysfs_ops
= {
4403 .show
= slab_attr_show
,
4404 .store
= slab_attr_store
,
4407 static struct kobj_type slab_ktype
= {
4408 .sysfs_ops
= &slab_sysfs_ops
,
4409 .release
= kmem_cache_release
4412 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4414 struct kobj_type
*ktype
= get_ktype(kobj
);
4416 if (ktype
== &slab_ktype
)
4421 static const struct kset_uevent_ops slab_uevent_ops
= {
4422 .filter
= uevent_filter
,
4425 static struct kset
*slab_kset
;
4427 #define ID_STR_LENGTH 64
4429 /* Create a unique string id for a slab cache:
4431 * Format :[flags-]size
4433 static char *create_unique_id(struct kmem_cache
*s
)
4435 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4442 * First flags affecting slabcache operations. We will only
4443 * get here for aliasable slabs so we do not need to support
4444 * too many flags. The flags here must cover all flags that
4445 * are matched during merging to guarantee that the id is
4448 if (s
->flags
& SLAB_CACHE_DMA
)
4450 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4452 if (s
->flags
& SLAB_DEBUG_FREE
)
4454 if (!(s
->flags
& SLAB_NOTRACK
))
4458 p
+= sprintf(p
, "%07d", s
->size
);
4459 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4463 static int sysfs_slab_add(struct kmem_cache
*s
)
4469 if (slab_state
< SYSFS
)
4470 /* Defer until later */
4473 unmergeable
= slab_unmergeable(s
);
4476 * Slabcache can never be merged so we can use the name proper.
4477 * This is typically the case for debug situations. In that
4478 * case we can catch duplicate names easily.
4480 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4484 * Create a unique name for the slab as a target
4487 name
= create_unique_id(s
);
4490 s
->kobj
.kset
= slab_kset
;
4491 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4493 kobject_put(&s
->kobj
);
4497 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4499 kobject_del(&s
->kobj
);
4500 kobject_put(&s
->kobj
);
4503 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4505 /* Setup first alias */
4506 sysfs_slab_alias(s
, s
->name
);
4512 static void sysfs_slab_remove(struct kmem_cache
*s
)
4514 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4515 kobject_del(&s
->kobj
);
4516 kobject_put(&s
->kobj
);
4520 * Need to buffer aliases during bootup until sysfs becomes
4521 * available lest we lose that information.
4523 struct saved_alias
{
4524 struct kmem_cache
*s
;
4526 struct saved_alias
*next
;
4529 static struct saved_alias
*alias_list
;
4531 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4533 struct saved_alias
*al
;
4535 if (slab_state
== SYSFS
) {
4537 * If we have a leftover link then remove it.
4539 sysfs_remove_link(&slab_kset
->kobj
, name
);
4540 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4543 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4549 al
->next
= alias_list
;
4554 static int __init
slab_sysfs_init(void)
4556 struct kmem_cache
*s
;
4559 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4561 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4567 list_for_each_entry(s
, &slab_caches
, list
) {
4568 err
= sysfs_slab_add(s
);
4570 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4571 " to sysfs\n", s
->name
);
4574 while (alias_list
) {
4575 struct saved_alias
*al
= alias_list
;
4577 alias_list
= alias_list
->next
;
4578 err
= sysfs_slab_alias(al
->s
, al
->name
);
4580 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4581 " %s to sysfs\n", s
->name
);
4589 __initcall(slab_sysfs_init
);
4593 * The /proc/slabinfo ABI
4595 #ifdef CONFIG_SLABINFO
4596 static void print_slabinfo_header(struct seq_file
*m
)
4598 seq_puts(m
, "slabinfo - version: 2.1\n");
4599 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4600 "<objperslab> <pagesperslab>");
4601 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4602 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4606 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4610 down_read(&slub_lock
);
4612 print_slabinfo_header(m
);
4614 return seq_list_start(&slab_caches
, *pos
);
4617 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4619 return seq_list_next(p
, &slab_caches
, pos
);
4622 static void s_stop(struct seq_file
*m
, void *p
)
4624 up_read(&slub_lock
);
4627 static int s_show(struct seq_file
*m
, void *p
)
4629 unsigned long nr_partials
= 0;
4630 unsigned long nr_slabs
= 0;
4631 unsigned long nr_inuse
= 0;
4632 unsigned long nr_objs
= 0;
4633 unsigned long nr_free
= 0;
4634 struct kmem_cache
*s
;
4637 s
= list_entry(p
, struct kmem_cache
, list
);
4639 for_each_online_node(node
) {
4640 struct kmem_cache_node
*n
= get_node(s
, node
);
4645 nr_partials
+= n
->nr_partial
;
4646 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4647 nr_objs
+= atomic_long_read(&n
->total_objects
);
4648 nr_free
+= count_partial(n
, count_free
);
4651 nr_inuse
= nr_objs
- nr_free
;
4653 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4654 nr_objs
, s
->size
, oo_objects(s
->oo
),
4655 (1 << oo_order(s
->oo
)));
4656 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4657 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4663 static const struct seq_operations slabinfo_op
= {
4670 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4672 return seq_open(file
, &slabinfo_op
);
4675 static const struct file_operations proc_slabinfo_operations
= {
4676 .open
= slabinfo_open
,
4678 .llseek
= seq_lseek
,
4679 .release
= seq_release
,
4682 static int __init
slab_proc_init(void)
4684 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4687 module_init(slab_proc_init
);
4688 #endif /* CONFIG_SLABINFO */