2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemcheck.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/mempolicy.h>
24 #include <linux/ctype.h>
25 #include <linux/debugobjects.h>
26 #include <linux/kallsyms.h>
27 #include <linux/memory.h>
28 #include <linux/math64.h>
29 #include <linux/fault-inject.h>
36 * The slab_lock protects operations on the object of a particular
37 * slab and its metadata in the page struct. If the slab lock
38 * has been taken then no allocations nor frees can be performed
39 * on the objects in the slab nor can the slab be added or removed
40 * from the partial or full lists since this would mean modifying
41 * the page_struct of the slab.
43 * The list_lock protects the partial and full list on each node and
44 * the partial slab counter. If taken then no new slabs may be added or
45 * removed from the lists nor make the number of partial slabs be modified.
46 * (Note that the total number of slabs is an atomic value that may be
47 * modified without taking the list lock).
49 * The list_lock is a centralized lock and thus we avoid taking it as
50 * much as possible. As long as SLUB does not have to handle partial
51 * slabs, operations can continue without any centralized lock. F.e.
52 * allocating a long series of objects that fill up slabs does not require
55 * The lock order is sometimes inverted when we are trying to get a slab
56 * off a list. We take the list_lock and then look for a page on the list
57 * to use. While we do that objects in the slabs may be freed. We can
58 * only operate on the slab if we have also taken the slab_lock. So we use
59 * a slab_trylock() on the slab. If trylock was successful then no frees
60 * can occur anymore and we can use the slab for allocations etc. If the
61 * slab_trylock() does not succeed then frees are in progress in the slab and
62 * we must stay away from it for a while since we may cause a bouncing
63 * cacheline if we try to acquire the lock. So go onto the next slab.
64 * If all pages are busy then we may allocate a new slab instead of reusing
65 * a partial slab. A new slab has noone operating on it and thus there is
66 * no danger of cacheline contention.
68 * Interrupts are disabled during allocation and deallocation in order to
69 * make the slab allocator safe to use in the context of an irq. In addition
70 * interrupts are disabled to ensure that the processor does not change
71 * while handling per_cpu slabs, due to kernel preemption.
73 * SLUB assigns one slab for allocation to each processor.
74 * Allocations only occur from these slabs called cpu slabs.
76 * Slabs with free elements are kept on a partial list and during regular
77 * operations no list for full slabs is used. If an object in a full slab is
78 * freed then the slab will show up again on the partial lists.
79 * We track full slabs for debugging purposes though because otherwise we
80 * cannot scan all objects.
82 * Slabs are freed when they become empty. Teardown and setup is
83 * minimal so we rely on the page allocators per cpu caches for
84 * fast frees and allocs.
86 * Overloading of page flags that are otherwise used for LRU management.
88 * PageActive The slab is frozen and exempt from list processing.
89 * This means that the slab is dedicated to a purpose
90 * such as satisfying allocations for a specific
91 * processor. Objects may be freed in the slab while
92 * it is frozen but slab_free will then skip the usual
93 * list operations. It is up to the processor holding
94 * the slab to integrate the slab into the slab lists
95 * when the slab is no longer needed.
97 * One use of this flag is to mark slabs that are
98 * used for allocations. Then such a slab becomes a cpu
99 * slab. The cpu slab may be equipped with an additional
100 * freelist that allows lockless access to
101 * free objects in addition to the regular freelist
102 * that requires the slab lock.
104 * PageError Slab requires special handling due to debug
105 * options set. This moves slab handling out of
106 * the fast path and disables lockless freelists.
109 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
110 SLAB_TRACE | SLAB_DEBUG_FREE)
112 static inline int kmem_cache_debug(struct kmem_cache
*s
)
114 #ifdef CONFIG_SLUB_DEBUG
115 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
122 * Issues still to be resolved:
124 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
126 * - Variable sizing of the per node arrays
129 /* Enable to test recovery from slab corruption on boot */
130 #undef SLUB_RESILIENCY_TEST
133 * Mininum number of partial slabs. These will be left on the partial
134 * lists even if they are empty. kmem_cache_shrink may reclaim them.
136 #define MIN_PARTIAL 5
139 * Maximum number of desirable partial slabs.
140 * The existence of more partial slabs makes kmem_cache_shrink
141 * sort the partial list by the number of objects in the.
143 #define MAX_PARTIAL 10
145 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
146 SLAB_POISON | SLAB_STORE_USER)
149 * Debugging flags that require metadata to be stored in the slab. These get
150 * disabled when slub_debug=O is used and a cache's min order increases with
153 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
156 * Set of flags that will prevent slab merging
158 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
159 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
162 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
163 SLAB_CACHE_DMA | SLAB_NOTRACK)
166 #define OO_MASK ((1 << OO_SHIFT) - 1)
167 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
169 /* Internal SLUB flags */
170 #define __OBJECT_POISON 0x80000000UL /* Poison object */
172 static int kmem_size
= sizeof(struct kmem_cache
);
175 static struct notifier_block slab_notifier
;
179 DOWN
, /* No slab functionality available */
180 PARTIAL
, /* Kmem_cache_node works */
181 UP
, /* Everything works but does not show up in sysfs */
185 /* A list of all slab caches on the system */
186 static DECLARE_RWSEM(slub_lock
);
187 static LIST_HEAD(slab_caches
);
190 * Tracking user of a slab.
193 unsigned long addr
; /* Called from address */
194 int cpu
; /* Was running on cpu */
195 int pid
; /* Pid context */
196 unsigned long when
; /* When did the operation occur */
199 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
201 #ifdef CONFIG_SLUB_DEBUG
202 static int sysfs_slab_add(struct kmem_cache
*);
203 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
204 static void sysfs_slab_remove(struct kmem_cache
*);
207 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
208 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
210 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
218 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
220 #ifdef CONFIG_SLUB_STATS
221 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
225 /********************************************************************
226 * Core slab cache functions
227 *******************************************************************/
229 int slab_is_available(void)
231 return slab_state
>= UP
;
234 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
236 return s
->node
[node
];
239 /* Verify that a pointer has an address that is valid within a slab page */
240 static inline int check_valid_pointer(struct kmem_cache
*s
,
241 struct page
*page
, const void *object
)
248 base
= page_address(page
);
249 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
250 (object
- base
) % s
->size
) {
257 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
259 return *(void **)(object
+ s
->offset
);
262 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
264 *(void **)(object
+ s
->offset
) = fp
;
267 /* Loop over all objects in a slab */
268 #define for_each_object(__p, __s, __addr, __objects) \
269 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
273 #define for_each_free_object(__p, __s, __free) \
274 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
276 /* Determine object index from a given position */
277 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
279 return (p
- addr
) / s
->size
;
282 static inline struct kmem_cache_order_objects
oo_make(int order
,
285 struct kmem_cache_order_objects x
= {
286 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
292 static inline int oo_order(struct kmem_cache_order_objects x
)
294 return x
.x
>> OO_SHIFT
;
297 static inline int oo_objects(struct kmem_cache_order_objects x
)
299 return x
.x
& OO_MASK
;
302 #ifdef CONFIG_SLUB_DEBUG
306 #ifdef CONFIG_SLUB_DEBUG_ON
307 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
309 static int slub_debug
;
312 static char *slub_debug_slabs
;
313 static int disable_higher_order_debug
;
318 static void print_section(char *text
, u8
*addr
, unsigned int length
)
326 for (i
= 0; i
< length
; i
++) {
328 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
331 printk(KERN_CONT
" %02x", addr
[i
]);
333 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
335 printk(KERN_CONT
" %s\n", ascii
);
342 printk(KERN_CONT
" ");
346 printk(KERN_CONT
" %s\n", ascii
);
350 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
351 enum track_item alloc
)
356 p
= object
+ s
->offset
+ sizeof(void *);
358 p
= object
+ s
->inuse
;
363 static void set_track(struct kmem_cache
*s
, void *object
,
364 enum track_item alloc
, unsigned long addr
)
366 struct track
*p
= get_track(s
, object
, alloc
);
370 p
->cpu
= smp_processor_id();
371 p
->pid
= current
->pid
;
374 memset(p
, 0, sizeof(struct track
));
377 static void init_tracking(struct kmem_cache
*s
, void *object
)
379 if (!(s
->flags
& SLAB_STORE_USER
))
382 set_track(s
, object
, TRACK_FREE
, 0UL);
383 set_track(s
, object
, TRACK_ALLOC
, 0UL);
386 static void print_track(const char *s
, struct track
*t
)
391 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
392 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
395 static void print_tracking(struct kmem_cache
*s
, void *object
)
397 if (!(s
->flags
& SLAB_STORE_USER
))
400 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
401 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
404 static void print_page_info(struct page
*page
)
406 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
407 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
411 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
417 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
419 printk(KERN_ERR
"========================================"
420 "=====================================\n");
421 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
422 printk(KERN_ERR
"----------------------------------------"
423 "-------------------------------------\n\n");
426 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
432 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
434 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
437 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
439 unsigned int off
; /* Offset of last byte */
440 u8
*addr
= page_address(page
);
442 print_tracking(s
, p
);
444 print_page_info(page
);
446 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
447 p
, p
- addr
, get_freepointer(s
, p
));
450 print_section("Bytes b4", p
- 16, 16);
452 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
454 if (s
->flags
& SLAB_RED_ZONE
)
455 print_section("Redzone", p
+ s
->objsize
,
456 s
->inuse
- s
->objsize
);
459 off
= s
->offset
+ sizeof(void *);
463 if (s
->flags
& SLAB_STORE_USER
)
464 off
+= 2 * sizeof(struct track
);
467 /* Beginning of the filler is the free pointer */
468 print_section("Padding", p
+ off
, s
->size
- off
);
473 static void object_err(struct kmem_cache
*s
, struct page
*page
,
474 u8
*object
, char *reason
)
476 slab_bug(s
, "%s", reason
);
477 print_trailer(s
, page
, object
);
480 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
486 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
488 slab_bug(s
, "%s", buf
);
489 print_page_info(page
);
493 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
497 if (s
->flags
& __OBJECT_POISON
) {
498 memset(p
, POISON_FREE
, s
->objsize
- 1);
499 p
[s
->objsize
- 1] = POISON_END
;
502 if (s
->flags
& SLAB_RED_ZONE
)
503 memset(p
+ s
->objsize
, val
, s
->inuse
- s
->objsize
);
506 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
509 if (*start
!= (u8
)value
)
517 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
518 void *from
, void *to
)
520 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
521 memset(from
, data
, to
- from
);
524 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
525 u8
*object
, char *what
,
526 u8
*start
, unsigned int value
, unsigned int bytes
)
531 fault
= check_bytes(start
, value
, bytes
);
536 while (end
> fault
&& end
[-1] == value
)
539 slab_bug(s
, "%s overwritten", what
);
540 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
541 fault
, end
- 1, fault
[0], value
);
542 print_trailer(s
, page
, object
);
544 restore_bytes(s
, what
, value
, fault
, end
);
552 * Bytes of the object to be managed.
553 * If the freepointer may overlay the object then the free
554 * pointer is the first word of the object.
556 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
559 * object + s->objsize
560 * Padding to reach word boundary. This is also used for Redzoning.
561 * Padding is extended by another word if Redzoning is enabled and
564 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
565 * 0xcc (RED_ACTIVE) for objects in use.
568 * Meta data starts here.
570 * A. Free pointer (if we cannot overwrite object on free)
571 * B. Tracking data for SLAB_STORE_USER
572 * C. Padding to reach required alignment boundary or at mininum
573 * one word if debugging is on to be able to detect writes
574 * before the word boundary.
576 * Padding is done using 0x5a (POISON_INUSE)
579 * Nothing is used beyond s->size.
581 * If slabcaches are merged then the objsize and inuse boundaries are mostly
582 * ignored. And therefore no slab options that rely on these boundaries
583 * may be used with merged slabcaches.
586 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
588 unsigned long off
= s
->inuse
; /* The end of info */
591 /* Freepointer is placed after the object. */
592 off
+= sizeof(void *);
594 if (s
->flags
& SLAB_STORE_USER
)
595 /* We also have user information there */
596 off
+= 2 * sizeof(struct track
);
601 return check_bytes_and_report(s
, page
, p
, "Object padding",
602 p
+ off
, POISON_INUSE
, s
->size
- off
);
605 /* Check the pad bytes at the end of a slab page */
606 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
614 if (!(s
->flags
& SLAB_POISON
))
617 start
= page_address(page
);
618 length
= (PAGE_SIZE
<< compound_order(page
));
619 end
= start
+ length
;
620 remainder
= length
% s
->size
;
624 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
627 while (end
> fault
&& end
[-1] == POISON_INUSE
)
630 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
631 print_section("Padding", end
- remainder
, remainder
);
633 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
637 static int check_object(struct kmem_cache
*s
, struct page
*page
,
638 void *object
, u8 val
)
641 u8
*endobject
= object
+ s
->objsize
;
643 if (s
->flags
& SLAB_RED_ZONE
) {
644 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
645 endobject
, val
, s
->inuse
- s
->objsize
))
648 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
649 check_bytes_and_report(s
, page
, p
, "Alignment padding",
650 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
654 if (s
->flags
& SLAB_POISON
) {
655 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
656 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
657 POISON_FREE
, s
->objsize
- 1) ||
658 !check_bytes_and_report(s
, page
, p
, "Poison",
659 p
+ s
->objsize
- 1, POISON_END
, 1)))
662 * check_pad_bytes cleans up on its own.
664 check_pad_bytes(s
, page
, p
);
667 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
669 * Object and freepointer overlap. Cannot check
670 * freepointer while object is allocated.
674 /* Check free pointer validity */
675 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
676 object_err(s
, page
, p
, "Freepointer corrupt");
678 * No choice but to zap it and thus lose the remainder
679 * of the free objects in this slab. May cause
680 * another error because the object count is now wrong.
682 set_freepointer(s
, p
, NULL
);
688 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
692 VM_BUG_ON(!irqs_disabled());
694 if (!PageSlab(page
)) {
695 slab_err(s
, page
, "Not a valid slab page");
699 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
700 if (page
->objects
> maxobj
) {
701 slab_err(s
, page
, "objects %u > max %u",
702 s
->name
, page
->objects
, maxobj
);
705 if (page
->inuse
> page
->objects
) {
706 slab_err(s
, page
, "inuse %u > max %u",
707 s
->name
, page
->inuse
, page
->objects
);
710 /* Slab_pad_check fixes things up after itself */
711 slab_pad_check(s
, page
);
716 * Determine if a certain object on a page is on the freelist. Must hold the
717 * slab lock to guarantee that the chains are in a consistent state.
719 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
722 void *fp
= page
->freelist
;
724 unsigned long max_objects
;
726 while (fp
&& nr
<= page
->objects
) {
729 if (!check_valid_pointer(s
, page
, fp
)) {
731 object_err(s
, page
, object
,
732 "Freechain corrupt");
733 set_freepointer(s
, object
, NULL
);
736 slab_err(s
, page
, "Freepointer corrupt");
737 page
->freelist
= NULL
;
738 page
->inuse
= page
->objects
;
739 slab_fix(s
, "Freelist cleared");
745 fp
= get_freepointer(s
, object
);
749 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
750 if (max_objects
> MAX_OBJS_PER_PAGE
)
751 max_objects
= MAX_OBJS_PER_PAGE
;
753 if (page
->objects
!= max_objects
) {
754 slab_err(s
, page
, "Wrong number of objects. Found %d but "
755 "should be %d", page
->objects
, max_objects
);
756 page
->objects
= max_objects
;
757 slab_fix(s
, "Number of objects adjusted.");
759 if (page
->inuse
!= page
->objects
- nr
) {
760 slab_err(s
, page
, "Wrong object count. Counter is %d but "
761 "counted were %d", page
->inuse
, page
->objects
- nr
);
762 page
->inuse
= page
->objects
- nr
;
763 slab_fix(s
, "Object count adjusted.");
765 return search
== NULL
;
768 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
771 if (s
->flags
& SLAB_TRACE
) {
772 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
774 alloc
? "alloc" : "free",
779 print_section("Object", (void *)object
, s
->objsize
);
786 * Hooks for other subsystems that check memory allocations. In a typical
787 * production configuration these hooks all should produce no code at all.
789 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
791 flags
&= gfp_allowed_mask
;
792 lockdep_trace_alloc(flags
);
793 might_sleep_if(flags
& __GFP_WAIT
);
795 return should_failslab(s
->objsize
, flags
, s
->flags
);
798 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
800 flags
&= gfp_allowed_mask
;
801 kmemcheck_slab_alloc(s
, flags
, object
, s
->objsize
);
802 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
805 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
807 kmemleak_free_recursive(x
, s
->flags
);
810 static inline void slab_free_hook_irq(struct kmem_cache
*s
, void *object
)
812 kmemcheck_slab_free(s
, object
, s
->objsize
);
813 debug_check_no_locks_freed(object
, s
->objsize
);
814 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
815 debug_check_no_obj_freed(object
, s
->objsize
);
819 * Tracking of fully allocated slabs for debugging purposes.
821 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
823 spin_lock(&n
->list_lock
);
824 list_add(&page
->lru
, &n
->full
);
825 spin_unlock(&n
->list_lock
);
828 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
830 struct kmem_cache_node
*n
;
832 if (!(s
->flags
& SLAB_STORE_USER
))
835 n
= get_node(s
, page_to_nid(page
));
837 spin_lock(&n
->list_lock
);
838 list_del(&page
->lru
);
839 spin_unlock(&n
->list_lock
);
842 /* Tracking of the number of slabs for debugging purposes */
843 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
845 struct kmem_cache_node
*n
= get_node(s
, node
);
847 return atomic_long_read(&n
->nr_slabs
);
850 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
852 return atomic_long_read(&n
->nr_slabs
);
855 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
857 struct kmem_cache_node
*n
= get_node(s
, node
);
860 * May be called early in order to allocate a slab for the
861 * kmem_cache_node structure. Solve the chicken-egg
862 * dilemma by deferring the increment of the count during
863 * bootstrap (see early_kmem_cache_node_alloc).
866 atomic_long_inc(&n
->nr_slabs
);
867 atomic_long_add(objects
, &n
->total_objects
);
870 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
872 struct kmem_cache_node
*n
= get_node(s
, node
);
874 atomic_long_dec(&n
->nr_slabs
);
875 atomic_long_sub(objects
, &n
->total_objects
);
878 /* Object debug checks for alloc/free paths */
879 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
882 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
885 init_object(s
, object
, SLUB_RED_INACTIVE
);
886 init_tracking(s
, object
);
889 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
890 void *object
, unsigned long addr
)
892 if (!check_slab(s
, page
))
895 if (!on_freelist(s
, page
, object
)) {
896 object_err(s
, page
, object
, "Object already allocated");
900 if (!check_valid_pointer(s
, page
, object
)) {
901 object_err(s
, page
, object
, "Freelist Pointer check fails");
905 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
908 /* Success perform special debug activities for allocs */
909 if (s
->flags
& SLAB_STORE_USER
)
910 set_track(s
, object
, TRACK_ALLOC
, addr
);
911 trace(s
, page
, object
, 1);
912 init_object(s
, object
, SLUB_RED_ACTIVE
);
916 if (PageSlab(page
)) {
918 * If this is a slab page then lets do the best we can
919 * to avoid issues in the future. Marking all objects
920 * as used avoids touching the remaining objects.
922 slab_fix(s
, "Marking all objects used");
923 page
->inuse
= page
->objects
;
924 page
->freelist
= NULL
;
929 static noinline
int free_debug_processing(struct kmem_cache
*s
,
930 struct page
*page
, void *object
, unsigned long addr
)
932 if (!check_slab(s
, page
))
935 if (!check_valid_pointer(s
, page
, object
)) {
936 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
940 if (on_freelist(s
, page
, object
)) {
941 object_err(s
, page
, object
, "Object already free");
945 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
948 if (unlikely(s
!= page
->slab
)) {
949 if (!PageSlab(page
)) {
950 slab_err(s
, page
, "Attempt to free object(0x%p) "
951 "outside of slab", object
);
952 } else if (!page
->slab
) {
954 "SLUB <none>: no slab for object 0x%p.\n",
958 object_err(s
, page
, object
,
959 "page slab pointer corrupt.");
963 /* Special debug activities for freeing objects */
964 if (!PageSlubFrozen(page
) && !page
->freelist
)
965 remove_full(s
, page
);
966 if (s
->flags
& SLAB_STORE_USER
)
967 set_track(s
, object
, TRACK_FREE
, addr
);
968 trace(s
, page
, object
, 0);
969 init_object(s
, object
, SLUB_RED_INACTIVE
);
973 slab_fix(s
, "Object at 0x%p not freed", object
);
977 static int __init
setup_slub_debug(char *str
)
979 slub_debug
= DEBUG_DEFAULT_FLAGS
;
980 if (*str
++ != '=' || !*str
)
982 * No options specified. Switch on full debugging.
988 * No options but restriction on slabs. This means full
989 * debugging for slabs matching a pattern.
993 if (tolower(*str
) == 'o') {
995 * Avoid enabling debugging on caches if its minimum order
996 * would increase as a result.
998 disable_higher_order_debug
= 1;
1005 * Switch off all debugging measures.
1010 * Determine which debug features should be switched on
1012 for (; *str
&& *str
!= ','; str
++) {
1013 switch (tolower(*str
)) {
1015 slub_debug
|= SLAB_DEBUG_FREE
;
1018 slub_debug
|= SLAB_RED_ZONE
;
1021 slub_debug
|= SLAB_POISON
;
1024 slub_debug
|= SLAB_STORE_USER
;
1027 slub_debug
|= SLAB_TRACE
;
1030 slub_debug
|= SLAB_FAILSLAB
;
1033 printk(KERN_ERR
"slub_debug option '%c' "
1034 "unknown. skipped\n", *str
);
1040 slub_debug_slabs
= str
+ 1;
1045 __setup("slub_debug", setup_slub_debug
);
1047 static unsigned long kmem_cache_flags(unsigned long objsize
,
1048 unsigned long flags
, const char *name
,
1049 void (*ctor
)(void *))
1052 * Enable debugging if selected on the kernel commandline.
1054 if (slub_debug
&& (!slub_debug_slabs
||
1055 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1056 flags
|= slub_debug
;
1061 static inline void setup_object_debug(struct kmem_cache
*s
,
1062 struct page
*page
, void *object
) {}
1064 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1065 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1067 static inline int free_debug_processing(struct kmem_cache
*s
,
1068 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1070 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1072 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1073 void *object
, u8 val
) { return 1; }
1074 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1075 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1076 unsigned long flags
, const char *name
,
1077 void (*ctor
)(void *))
1081 #define slub_debug 0
1083 #define disable_higher_order_debug 0
1085 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1087 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1089 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1091 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1094 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1097 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1100 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1102 static inline void slab_free_hook_irq(struct kmem_cache
*s
,
1108 * Slab allocation and freeing
1110 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1111 struct kmem_cache_order_objects oo
)
1113 int order
= oo_order(oo
);
1115 flags
|= __GFP_NOTRACK
;
1117 if (node
== NUMA_NO_NODE
)
1118 return alloc_pages(flags
, order
);
1120 return alloc_pages_exact_node(node
, flags
, order
);
1123 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1126 struct kmem_cache_order_objects oo
= s
->oo
;
1129 flags
|= s
->allocflags
;
1132 * Let the initial higher-order allocation fail under memory pressure
1133 * so we fall-back to the minimum order allocation.
1135 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1137 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1138 if (unlikely(!page
)) {
1141 * Allocation may have failed due to fragmentation.
1142 * Try a lower order alloc if possible
1144 page
= alloc_slab_page(flags
, node
, oo
);
1148 stat(s
, ORDER_FALLBACK
);
1151 if (kmemcheck_enabled
1152 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1153 int pages
= 1 << oo_order(oo
);
1155 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1158 * Objects from caches that have a constructor don't get
1159 * cleared when they're allocated, so we need to do it here.
1162 kmemcheck_mark_uninitialized_pages(page
, pages
);
1164 kmemcheck_mark_unallocated_pages(page
, pages
);
1167 page
->objects
= oo_objects(oo
);
1168 mod_zone_page_state(page_zone(page
),
1169 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1170 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1176 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1179 setup_object_debug(s
, page
, object
);
1180 if (unlikely(s
->ctor
))
1184 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1191 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1193 page
= allocate_slab(s
,
1194 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1198 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1200 page
->flags
|= 1 << PG_slab
;
1202 start
= page_address(page
);
1204 if (unlikely(s
->flags
& SLAB_POISON
))
1205 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1208 for_each_object(p
, s
, start
, page
->objects
) {
1209 setup_object(s
, page
, last
);
1210 set_freepointer(s
, last
, p
);
1213 setup_object(s
, page
, last
);
1214 set_freepointer(s
, last
, NULL
);
1216 page
->freelist
= start
;
1222 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1224 int order
= compound_order(page
);
1225 int pages
= 1 << order
;
1227 if (kmem_cache_debug(s
)) {
1230 slab_pad_check(s
, page
);
1231 for_each_object(p
, s
, page_address(page
),
1233 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1236 kmemcheck_free_shadow(page
, compound_order(page
));
1238 mod_zone_page_state(page_zone(page
),
1239 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1240 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1243 __ClearPageSlab(page
);
1244 reset_page_mapcount(page
);
1245 if (current
->reclaim_state
)
1246 current
->reclaim_state
->reclaimed_slab
+= pages
;
1247 __free_pages(page
, order
);
1250 static void rcu_free_slab(struct rcu_head
*h
)
1254 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1255 __free_slab(page
->slab
, page
);
1258 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1260 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1262 * RCU free overloads the RCU head over the LRU
1264 struct rcu_head
*head
= (void *)&page
->lru
;
1266 call_rcu(head
, rcu_free_slab
);
1268 __free_slab(s
, page
);
1271 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1273 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1278 * Per slab locking using the pagelock
1280 static __always_inline
void slab_lock(struct page
*page
)
1282 bit_spin_lock(PG_locked
, &page
->flags
);
1285 static __always_inline
void slab_unlock(struct page
*page
)
1287 __bit_spin_unlock(PG_locked
, &page
->flags
);
1290 static __always_inline
int slab_trylock(struct page
*page
)
1294 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1299 * Management of partially allocated slabs
1301 static void add_partial(struct kmem_cache_node
*n
,
1302 struct page
*page
, int tail
)
1304 spin_lock(&n
->list_lock
);
1307 list_add_tail(&page
->lru
, &n
->partial
);
1309 list_add(&page
->lru
, &n
->partial
);
1310 spin_unlock(&n
->list_lock
);
1313 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1315 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1317 spin_lock(&n
->list_lock
);
1318 list_del(&page
->lru
);
1320 spin_unlock(&n
->list_lock
);
1324 * Lock slab and remove from the partial list.
1326 * Must hold list_lock.
1328 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1331 if (slab_trylock(page
)) {
1332 list_del(&page
->lru
);
1334 __SetPageSlubFrozen(page
);
1341 * Try to allocate a partial slab from a specific node.
1343 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1348 * Racy check. If we mistakenly see no partial slabs then we
1349 * just allocate an empty slab. If we mistakenly try to get a
1350 * partial slab and there is none available then get_partials()
1353 if (!n
|| !n
->nr_partial
)
1356 spin_lock(&n
->list_lock
);
1357 list_for_each_entry(page
, &n
->partial
, lru
)
1358 if (lock_and_freeze_slab(n
, page
))
1362 spin_unlock(&n
->list_lock
);
1367 * Get a page from somewhere. Search in increasing NUMA distances.
1369 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1372 struct zonelist
*zonelist
;
1375 enum zone_type high_zoneidx
= gfp_zone(flags
);
1379 * The defrag ratio allows a configuration of the tradeoffs between
1380 * inter node defragmentation and node local allocations. A lower
1381 * defrag_ratio increases the tendency to do local allocations
1382 * instead of attempting to obtain partial slabs from other nodes.
1384 * If the defrag_ratio is set to 0 then kmalloc() always
1385 * returns node local objects. If the ratio is higher then kmalloc()
1386 * may return off node objects because partial slabs are obtained
1387 * from other nodes and filled up.
1389 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1390 * defrag_ratio = 1000) then every (well almost) allocation will
1391 * first attempt to defrag slab caches on other nodes. This means
1392 * scanning over all nodes to look for partial slabs which may be
1393 * expensive if we do it every time we are trying to find a slab
1394 * with available objects.
1396 if (!s
->remote_node_defrag_ratio
||
1397 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1401 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1402 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1403 struct kmem_cache_node
*n
;
1405 n
= get_node(s
, zone_to_nid(zone
));
1407 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1408 n
->nr_partial
> s
->min_partial
) {
1409 page
= get_partial_node(n
);
1422 * Get a partial page, lock it and return it.
1424 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1427 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1429 page
= get_partial_node(get_node(s
, searchnode
));
1430 if (page
|| node
!= -1)
1433 return get_any_partial(s
, flags
);
1437 * Move a page back to the lists.
1439 * Must be called with the slab lock held.
1441 * On exit the slab lock will have been dropped.
1443 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1445 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1447 __ClearPageSlubFrozen(page
);
1450 if (page
->freelist
) {
1451 add_partial(n
, page
, tail
);
1452 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1454 stat(s
, DEACTIVATE_FULL
);
1455 if (kmem_cache_debug(s
) && (s
->flags
& SLAB_STORE_USER
))
1460 stat(s
, DEACTIVATE_EMPTY
);
1461 if (n
->nr_partial
< s
->min_partial
) {
1463 * Adding an empty slab to the partial slabs in order
1464 * to avoid page allocator overhead. This slab needs
1465 * to come after the other slabs with objects in
1466 * so that the others get filled first. That way the
1467 * size of the partial list stays small.
1469 * kmem_cache_shrink can reclaim any empty slabs from
1472 add_partial(n
, page
, 1);
1477 discard_slab(s
, page
);
1483 * Remove the cpu slab
1485 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1487 struct page
*page
= c
->page
;
1491 stat(s
, DEACTIVATE_REMOTE_FREES
);
1493 * Merge cpu freelist into slab freelist. Typically we get here
1494 * because both freelists are empty. So this is unlikely
1497 while (unlikely(c
->freelist
)) {
1500 tail
= 0; /* Hot objects. Put the slab first */
1502 /* Retrieve object from cpu_freelist */
1503 object
= c
->freelist
;
1504 c
->freelist
= get_freepointer(s
, c
->freelist
);
1506 /* And put onto the regular freelist */
1507 set_freepointer(s
, object
, page
->freelist
);
1508 page
->freelist
= object
;
1512 unfreeze_slab(s
, page
, tail
);
1515 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1517 stat(s
, CPUSLAB_FLUSH
);
1519 deactivate_slab(s
, c
);
1525 * Called from IPI handler with interrupts disabled.
1527 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1529 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1531 if (likely(c
&& c
->page
))
1535 static void flush_cpu_slab(void *d
)
1537 struct kmem_cache
*s
= d
;
1539 __flush_cpu_slab(s
, smp_processor_id());
1542 static void flush_all(struct kmem_cache
*s
)
1544 on_each_cpu(flush_cpu_slab
, s
, 1);
1548 * Check if the objects in a per cpu structure fit numa
1549 * locality expectations.
1551 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1554 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
1560 static int count_free(struct page
*page
)
1562 return page
->objects
- page
->inuse
;
1565 static unsigned long count_partial(struct kmem_cache_node
*n
,
1566 int (*get_count
)(struct page
*))
1568 unsigned long flags
;
1569 unsigned long x
= 0;
1572 spin_lock_irqsave(&n
->list_lock
, flags
);
1573 list_for_each_entry(page
, &n
->partial
, lru
)
1574 x
+= get_count(page
);
1575 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1579 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1581 #ifdef CONFIG_SLUB_DEBUG
1582 return atomic_long_read(&n
->total_objects
);
1588 static noinline
void
1589 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1594 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1596 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1597 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1598 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1600 if (oo_order(s
->min
) > get_order(s
->objsize
))
1601 printk(KERN_WARNING
" %s debugging increased min order, use "
1602 "slub_debug=O to disable.\n", s
->name
);
1604 for_each_online_node(node
) {
1605 struct kmem_cache_node
*n
= get_node(s
, node
);
1606 unsigned long nr_slabs
;
1607 unsigned long nr_objs
;
1608 unsigned long nr_free
;
1613 nr_free
= count_partial(n
, count_free
);
1614 nr_slabs
= node_nr_slabs(n
);
1615 nr_objs
= node_nr_objs(n
);
1618 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1619 node
, nr_slabs
, nr_objs
, nr_free
);
1624 * Slow path. The lockless freelist is empty or we need to perform
1627 * Interrupts are disabled.
1629 * Processing is still very fast if new objects have been freed to the
1630 * regular freelist. In that case we simply take over the regular freelist
1631 * as the lockless freelist and zap the regular freelist.
1633 * If that is not working then we fall back to the partial lists. We take the
1634 * first element of the freelist as the object to allocate now and move the
1635 * rest of the freelist to the lockless freelist.
1637 * And if we were unable to get a new slab from the partial slab lists then
1638 * we need to allocate a new slab. This is the slowest path since it involves
1639 * a call to the page allocator and the setup of a new slab.
1641 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1642 unsigned long addr
, struct kmem_cache_cpu
*c
)
1647 /* We handle __GFP_ZERO in the caller */
1648 gfpflags
&= ~__GFP_ZERO
;
1654 if (unlikely(!node_match(c
, node
)))
1657 stat(s
, ALLOC_REFILL
);
1660 object
= c
->page
->freelist
;
1661 if (unlikely(!object
))
1663 if (kmem_cache_debug(s
))
1666 c
->freelist
= get_freepointer(s
, object
);
1667 c
->page
->inuse
= c
->page
->objects
;
1668 c
->page
->freelist
= NULL
;
1669 c
->node
= page_to_nid(c
->page
);
1671 slab_unlock(c
->page
);
1672 stat(s
, ALLOC_SLOWPATH
);
1676 deactivate_slab(s
, c
);
1679 new = get_partial(s
, gfpflags
, node
);
1682 stat(s
, ALLOC_FROM_PARTIAL
);
1686 gfpflags
&= gfp_allowed_mask
;
1687 if (gfpflags
& __GFP_WAIT
)
1690 new = new_slab(s
, gfpflags
, node
);
1692 if (gfpflags
& __GFP_WAIT
)
1693 local_irq_disable();
1696 c
= __this_cpu_ptr(s
->cpu_slab
);
1697 stat(s
, ALLOC_SLAB
);
1701 __SetPageSlubFrozen(new);
1705 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1706 slab_out_of_memory(s
, gfpflags
, node
);
1709 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1713 c
->page
->freelist
= get_freepointer(s
, object
);
1719 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1720 * have the fastpath folded into their functions. So no function call
1721 * overhead for requests that can be satisfied on the fastpath.
1723 * The fastpath works by first checking if the lockless freelist can be used.
1724 * If not then __slab_alloc is called for slow processing.
1726 * Otherwise we can simply pick the next object from the lockless free list.
1728 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1729 gfp_t gfpflags
, int node
, unsigned long addr
)
1732 struct kmem_cache_cpu
*c
;
1733 unsigned long flags
;
1735 if (slab_pre_alloc_hook(s
, gfpflags
))
1738 local_irq_save(flags
);
1739 c
= __this_cpu_ptr(s
->cpu_slab
);
1740 object
= c
->freelist
;
1741 if (unlikely(!object
|| !node_match(c
, node
)))
1743 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1746 c
->freelist
= get_freepointer(s
, object
);
1747 stat(s
, ALLOC_FASTPATH
);
1749 local_irq_restore(flags
);
1751 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1752 memset(object
, 0, s
->objsize
);
1754 slab_post_alloc_hook(s
, gfpflags
, object
);
1759 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1761 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1763 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1767 EXPORT_SYMBOL(kmem_cache_alloc
);
1769 #ifdef CONFIG_TRACING
1770 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1772 return slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1774 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1778 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1780 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1782 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1783 s
->objsize
, s
->size
, gfpflags
, node
);
1787 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1790 #ifdef CONFIG_TRACING
1791 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1795 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1797 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1801 * Slow patch handling. This may still be called frequently since objects
1802 * have a longer lifetime than the cpu slabs in most processing loads.
1804 * So we still attempt to reduce cache line usage. Just take the slab
1805 * lock and free the item. If there is no additional partial page
1806 * handling required then we can return immediately.
1808 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1809 void *x
, unsigned long addr
)
1812 void **object
= (void *)x
;
1814 stat(s
, FREE_SLOWPATH
);
1817 if (kmem_cache_debug(s
))
1821 prior
= page
->freelist
;
1822 set_freepointer(s
, object
, prior
);
1823 page
->freelist
= object
;
1826 if (unlikely(PageSlubFrozen(page
))) {
1827 stat(s
, FREE_FROZEN
);
1831 if (unlikely(!page
->inuse
))
1835 * Objects left in the slab. If it was not on the partial list before
1838 if (unlikely(!prior
)) {
1839 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1840 stat(s
, FREE_ADD_PARTIAL
);
1850 * Slab still on the partial list.
1852 remove_partial(s
, page
);
1853 stat(s
, FREE_REMOVE_PARTIAL
);
1857 discard_slab(s
, page
);
1861 if (!free_debug_processing(s
, page
, x
, addr
))
1867 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1868 * can perform fastpath freeing without additional function calls.
1870 * The fastpath is only possible if we are freeing to the current cpu slab
1871 * of this processor. This typically the case if we have just allocated
1874 * If fastpath is not possible then fall back to __slab_free where we deal
1875 * with all sorts of special processing.
1877 static __always_inline
void slab_free(struct kmem_cache
*s
,
1878 struct page
*page
, void *x
, unsigned long addr
)
1880 void **object
= (void *)x
;
1881 struct kmem_cache_cpu
*c
;
1882 unsigned long flags
;
1884 slab_free_hook(s
, x
);
1886 local_irq_save(flags
);
1887 c
= __this_cpu_ptr(s
->cpu_slab
);
1889 slab_free_hook_irq(s
, x
);
1891 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1892 set_freepointer(s
, object
, c
->freelist
);
1893 c
->freelist
= object
;
1894 stat(s
, FREE_FASTPATH
);
1896 __slab_free(s
, page
, x
, addr
);
1898 local_irq_restore(flags
);
1901 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1905 page
= virt_to_head_page(x
);
1907 slab_free(s
, page
, x
, _RET_IP_
);
1909 trace_kmem_cache_free(_RET_IP_
, x
);
1911 EXPORT_SYMBOL(kmem_cache_free
);
1913 /* Figure out on which slab page the object resides */
1914 static struct page
*get_object_page(const void *x
)
1916 struct page
*page
= virt_to_head_page(x
);
1918 if (!PageSlab(page
))
1925 * Object placement in a slab is made very easy because we always start at
1926 * offset 0. If we tune the size of the object to the alignment then we can
1927 * get the required alignment by putting one properly sized object after
1930 * Notice that the allocation order determines the sizes of the per cpu
1931 * caches. Each processor has always one slab available for allocations.
1932 * Increasing the allocation order reduces the number of times that slabs
1933 * must be moved on and off the partial lists and is therefore a factor in
1938 * Mininum / Maximum order of slab pages. This influences locking overhead
1939 * and slab fragmentation. A higher order reduces the number of partial slabs
1940 * and increases the number of allocations possible without having to
1941 * take the list_lock.
1943 static int slub_min_order
;
1944 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1945 static int slub_min_objects
;
1948 * Merge control. If this is set then no merging of slab caches will occur.
1949 * (Could be removed. This was introduced to pacify the merge skeptics.)
1951 static int slub_nomerge
;
1954 * Calculate the order of allocation given an slab object size.
1956 * The order of allocation has significant impact on performance and other
1957 * system components. Generally order 0 allocations should be preferred since
1958 * order 0 does not cause fragmentation in the page allocator. Larger objects
1959 * be problematic to put into order 0 slabs because there may be too much
1960 * unused space left. We go to a higher order if more than 1/16th of the slab
1963 * In order to reach satisfactory performance we must ensure that a minimum
1964 * number of objects is in one slab. Otherwise we may generate too much
1965 * activity on the partial lists which requires taking the list_lock. This is
1966 * less a concern for large slabs though which are rarely used.
1968 * slub_max_order specifies the order where we begin to stop considering the
1969 * number of objects in a slab as critical. If we reach slub_max_order then
1970 * we try to keep the page order as low as possible. So we accept more waste
1971 * of space in favor of a small page order.
1973 * Higher order allocations also allow the placement of more objects in a
1974 * slab and thereby reduce object handling overhead. If the user has
1975 * requested a higher mininum order then we start with that one instead of
1976 * the smallest order which will fit the object.
1978 static inline int slab_order(int size
, int min_objects
,
1979 int max_order
, int fract_leftover
)
1983 int min_order
= slub_min_order
;
1985 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1986 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1988 for (order
= max(min_order
,
1989 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1990 order
<= max_order
; order
++) {
1992 unsigned long slab_size
= PAGE_SIZE
<< order
;
1994 if (slab_size
< min_objects
* size
)
1997 rem
= slab_size
% size
;
1999 if (rem
<= slab_size
/ fract_leftover
)
2007 static inline int calculate_order(int size
)
2015 * Attempt to find best configuration for a slab. This
2016 * works by first attempting to generate a layout with
2017 * the best configuration and backing off gradually.
2019 * First we reduce the acceptable waste in a slab. Then
2020 * we reduce the minimum objects required in a slab.
2022 min_objects
= slub_min_objects
;
2024 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2025 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
2026 min_objects
= min(min_objects
, max_objects
);
2028 while (min_objects
> 1) {
2030 while (fraction
>= 4) {
2031 order
= slab_order(size
, min_objects
,
2032 slub_max_order
, fraction
);
2033 if (order
<= slub_max_order
)
2041 * We were unable to place multiple objects in a slab. Now
2042 * lets see if we can place a single object there.
2044 order
= slab_order(size
, 1, slub_max_order
, 1);
2045 if (order
<= slub_max_order
)
2049 * Doh this slab cannot be placed using slub_max_order.
2051 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2052 if (order
< MAX_ORDER
)
2058 * Figure out what the alignment of the objects will be.
2060 static unsigned long calculate_alignment(unsigned long flags
,
2061 unsigned long align
, unsigned long size
)
2064 * If the user wants hardware cache aligned objects then follow that
2065 * suggestion if the object is sufficiently large.
2067 * The hardware cache alignment cannot override the specified
2068 * alignment though. If that is greater then use it.
2070 if (flags
& SLAB_HWCACHE_ALIGN
) {
2071 unsigned long ralign
= cache_line_size();
2072 while (size
<= ralign
/ 2)
2074 align
= max(align
, ralign
);
2077 if (align
< ARCH_SLAB_MINALIGN
)
2078 align
= ARCH_SLAB_MINALIGN
;
2080 return ALIGN(align
, sizeof(void *));
2084 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2087 spin_lock_init(&n
->list_lock
);
2088 INIT_LIST_HEAD(&n
->partial
);
2089 #ifdef CONFIG_SLUB_DEBUG
2090 atomic_long_set(&n
->nr_slabs
, 0);
2091 atomic_long_set(&n
->total_objects
, 0);
2092 INIT_LIST_HEAD(&n
->full
);
2096 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2098 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2099 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2101 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2103 return s
->cpu_slab
!= NULL
;
2106 static struct kmem_cache
*kmem_cache_node
;
2109 * No kmalloc_node yet so do it by hand. We know that this is the first
2110 * slab on the node for this slabcache. There are no concurrent accesses
2113 * Note that this function only works on the kmalloc_node_cache
2114 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2115 * memory on a fresh node that has no slab structures yet.
2117 static void early_kmem_cache_node_alloc(int node
)
2120 struct kmem_cache_node
*n
;
2121 unsigned long flags
;
2123 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2125 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2128 if (page_to_nid(page
) != node
) {
2129 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2131 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2132 "in order to be able to continue\n");
2137 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2139 kmem_cache_node
->node
[node
] = n
;
2140 #ifdef CONFIG_SLUB_DEBUG
2141 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2142 init_tracking(kmem_cache_node
, n
);
2144 init_kmem_cache_node(n
, kmem_cache_node
);
2145 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2148 * lockdep requires consistent irq usage for each lock
2149 * so even though there cannot be a race this early in
2150 * the boot sequence, we still disable irqs.
2152 local_irq_save(flags
);
2153 add_partial(n
, page
, 0);
2154 local_irq_restore(flags
);
2157 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2161 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2162 struct kmem_cache_node
*n
= s
->node
[node
];
2165 kmem_cache_free(kmem_cache_node
, n
);
2167 s
->node
[node
] = NULL
;
2171 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2175 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2176 struct kmem_cache_node
*n
;
2178 if (slab_state
== DOWN
) {
2179 early_kmem_cache_node_alloc(node
);
2182 n
= kmem_cache_alloc_node(kmem_cache_node
,
2186 free_kmem_cache_nodes(s
);
2191 init_kmem_cache_node(n
, s
);
2196 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2198 if (min
< MIN_PARTIAL
)
2200 else if (min
> MAX_PARTIAL
)
2202 s
->min_partial
= min
;
2206 * calculate_sizes() determines the order and the distribution of data within
2209 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2211 unsigned long flags
= s
->flags
;
2212 unsigned long size
= s
->objsize
;
2213 unsigned long align
= s
->align
;
2217 * Round up object size to the next word boundary. We can only
2218 * place the free pointer at word boundaries and this determines
2219 * the possible location of the free pointer.
2221 size
= ALIGN(size
, sizeof(void *));
2223 #ifdef CONFIG_SLUB_DEBUG
2225 * Determine if we can poison the object itself. If the user of
2226 * the slab may touch the object after free or before allocation
2227 * then we should never poison the object itself.
2229 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2231 s
->flags
|= __OBJECT_POISON
;
2233 s
->flags
&= ~__OBJECT_POISON
;
2237 * If we are Redzoning then check if there is some space between the
2238 * end of the object and the free pointer. If not then add an
2239 * additional word to have some bytes to store Redzone information.
2241 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2242 size
+= sizeof(void *);
2246 * With that we have determined the number of bytes in actual use
2247 * by the object. This is the potential offset to the free pointer.
2251 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2254 * Relocate free pointer after the object if it is not
2255 * permitted to overwrite the first word of the object on
2258 * This is the case if we do RCU, have a constructor or
2259 * destructor or are poisoning the objects.
2262 size
+= sizeof(void *);
2265 #ifdef CONFIG_SLUB_DEBUG
2266 if (flags
& SLAB_STORE_USER
)
2268 * Need to store information about allocs and frees after
2271 size
+= 2 * sizeof(struct track
);
2273 if (flags
& SLAB_RED_ZONE
)
2275 * Add some empty padding so that we can catch
2276 * overwrites from earlier objects rather than let
2277 * tracking information or the free pointer be
2278 * corrupted if a user writes before the start
2281 size
+= sizeof(void *);
2285 * Determine the alignment based on various parameters that the
2286 * user specified and the dynamic determination of cache line size
2289 align
= calculate_alignment(flags
, align
, s
->objsize
);
2293 * SLUB stores one object immediately after another beginning from
2294 * offset 0. In order to align the objects we have to simply size
2295 * each object to conform to the alignment.
2297 size
= ALIGN(size
, align
);
2299 if (forced_order
>= 0)
2300 order
= forced_order
;
2302 order
= calculate_order(size
);
2309 s
->allocflags
|= __GFP_COMP
;
2311 if (s
->flags
& SLAB_CACHE_DMA
)
2312 s
->allocflags
|= SLUB_DMA
;
2314 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2315 s
->allocflags
|= __GFP_RECLAIMABLE
;
2318 * Determine the number of objects per slab
2320 s
->oo
= oo_make(order
, size
);
2321 s
->min
= oo_make(get_order(size
), size
);
2322 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2325 return !!oo_objects(s
->oo
);
2329 static int kmem_cache_open(struct kmem_cache
*s
,
2330 const char *name
, size_t size
,
2331 size_t align
, unsigned long flags
,
2332 void (*ctor
)(void *))
2334 memset(s
, 0, kmem_size
);
2339 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2341 if (!calculate_sizes(s
, -1))
2343 if (disable_higher_order_debug
) {
2345 * Disable debugging flags that store metadata if the min slab
2348 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2349 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2351 if (!calculate_sizes(s
, -1))
2357 * The larger the object size is, the more pages we want on the partial
2358 * list to avoid pounding the page allocator excessively.
2360 set_min_partial(s
, ilog2(s
->size
));
2363 s
->remote_node_defrag_ratio
= 1000;
2365 if (!init_kmem_cache_nodes(s
))
2368 if (alloc_kmem_cache_cpus(s
))
2371 free_kmem_cache_nodes(s
);
2373 if (flags
& SLAB_PANIC
)
2374 panic("Cannot create slab %s size=%lu realsize=%u "
2375 "order=%u offset=%u flags=%lx\n",
2376 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2382 * Check if a given pointer is valid
2384 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2388 if (!kern_ptr_validate(object
, s
->size
))
2391 page
= get_object_page(object
);
2393 if (!page
|| s
!= page
->slab
)
2394 /* No slab or wrong slab */
2397 if (!check_valid_pointer(s
, page
, object
))
2401 * We could also check if the object is on the slabs freelist.
2402 * But this would be too expensive and it seems that the main
2403 * purpose of kmem_ptr_valid() is to check if the object belongs
2404 * to a certain slab.
2408 EXPORT_SYMBOL(kmem_ptr_validate
);
2411 * Determine the size of a slab object
2413 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2417 EXPORT_SYMBOL(kmem_cache_size
);
2419 const char *kmem_cache_name(struct kmem_cache
*s
)
2423 EXPORT_SYMBOL(kmem_cache_name
);
2425 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2428 #ifdef CONFIG_SLUB_DEBUG
2429 void *addr
= page_address(page
);
2431 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2436 slab_err(s
, page
, "%s", text
);
2438 for_each_free_object(p
, s
, page
->freelist
)
2439 set_bit(slab_index(p
, s
, addr
), map
);
2441 for_each_object(p
, s
, addr
, page
->objects
) {
2443 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2444 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2446 print_tracking(s
, p
);
2455 * Attempt to free all partial slabs on a node.
2457 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2459 unsigned long flags
;
2460 struct page
*page
, *h
;
2462 spin_lock_irqsave(&n
->list_lock
, flags
);
2463 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2465 list_del(&page
->lru
);
2466 discard_slab(s
, page
);
2469 list_slab_objects(s
, page
,
2470 "Objects remaining on kmem_cache_close()");
2473 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2477 * Release all resources used by a slab cache.
2479 static inline int kmem_cache_close(struct kmem_cache
*s
)
2484 free_percpu(s
->cpu_slab
);
2485 /* Attempt to free all objects */
2486 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2487 struct kmem_cache_node
*n
= get_node(s
, node
);
2490 if (n
->nr_partial
|| slabs_node(s
, node
))
2493 free_kmem_cache_nodes(s
);
2498 * Close a cache and release the kmem_cache structure
2499 * (must be used for caches created using kmem_cache_create)
2501 void kmem_cache_destroy(struct kmem_cache
*s
)
2503 down_write(&slub_lock
);
2507 if (kmem_cache_close(s
)) {
2508 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2509 "still has objects.\n", s
->name
, __func__
);
2512 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2514 sysfs_slab_remove(s
);
2516 up_write(&slub_lock
);
2518 EXPORT_SYMBOL(kmem_cache_destroy
);
2520 /********************************************************************
2522 *******************************************************************/
2524 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
2525 EXPORT_SYMBOL(kmalloc_caches
);
2527 static struct kmem_cache
*kmem_cache
;
2529 #ifdef CONFIG_ZONE_DMA
2530 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
2533 static int __init
setup_slub_min_order(char *str
)
2535 get_option(&str
, &slub_min_order
);
2540 __setup("slub_min_order=", setup_slub_min_order
);
2542 static int __init
setup_slub_max_order(char *str
)
2544 get_option(&str
, &slub_max_order
);
2545 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2550 __setup("slub_max_order=", setup_slub_max_order
);
2552 static int __init
setup_slub_min_objects(char *str
)
2554 get_option(&str
, &slub_min_objects
);
2559 __setup("slub_min_objects=", setup_slub_min_objects
);
2561 static int __init
setup_slub_nomerge(char *str
)
2567 __setup("slub_nomerge", setup_slub_nomerge
);
2569 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
2570 int size
, unsigned int flags
)
2572 struct kmem_cache
*s
;
2574 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
2577 * This function is called with IRQs disabled during early-boot on
2578 * single CPU so there's no need to take slub_lock here.
2580 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2584 list_add(&s
->list
, &slab_caches
);
2588 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2593 * Conversion table for small slabs sizes / 8 to the index in the
2594 * kmalloc array. This is necessary for slabs < 192 since we have non power
2595 * of two cache sizes there. The size of larger slabs can be determined using
2598 static s8 size_index
[24] = {
2625 static inline int size_index_elem(size_t bytes
)
2627 return (bytes
- 1) / 8;
2630 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2636 return ZERO_SIZE_PTR
;
2638 index
= size_index
[size_index_elem(size
)];
2640 index
= fls(size
- 1);
2642 #ifdef CONFIG_ZONE_DMA
2643 if (unlikely((flags
& SLUB_DMA
)))
2644 return kmalloc_dma_caches
[index
];
2647 return kmalloc_caches
[index
];
2650 void *__kmalloc(size_t size
, gfp_t flags
)
2652 struct kmem_cache
*s
;
2655 if (unlikely(size
> SLUB_MAX_SIZE
))
2656 return kmalloc_large(size
, flags
);
2658 s
= get_slab(size
, flags
);
2660 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2663 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
2665 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2669 EXPORT_SYMBOL(__kmalloc
);
2671 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2676 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2677 page
= alloc_pages_node(node
, flags
, get_order(size
));
2679 ptr
= page_address(page
);
2681 kmemleak_alloc(ptr
, size
, 1, flags
);
2686 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2688 struct kmem_cache
*s
;
2691 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2692 ret
= kmalloc_large_node(size
, flags
, node
);
2694 trace_kmalloc_node(_RET_IP_
, ret
,
2695 size
, PAGE_SIZE
<< get_order(size
),
2701 s
= get_slab(size
, flags
);
2703 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2706 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2708 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2712 EXPORT_SYMBOL(__kmalloc_node
);
2715 size_t ksize(const void *object
)
2718 struct kmem_cache
*s
;
2720 if (unlikely(object
== ZERO_SIZE_PTR
))
2723 page
= virt_to_head_page(object
);
2725 if (unlikely(!PageSlab(page
))) {
2726 WARN_ON(!PageCompound(page
));
2727 return PAGE_SIZE
<< compound_order(page
);
2731 #ifdef CONFIG_SLUB_DEBUG
2733 * Debugging requires use of the padding between object
2734 * and whatever may come after it.
2736 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2741 * If we have the need to store the freelist pointer
2742 * back there or track user information then we can
2743 * only use the space before that information.
2745 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2748 * Else we can use all the padding etc for the allocation
2752 EXPORT_SYMBOL(ksize
);
2754 void kfree(const void *x
)
2757 void *object
= (void *)x
;
2759 trace_kfree(_RET_IP_
, x
);
2761 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2764 page
= virt_to_head_page(x
);
2765 if (unlikely(!PageSlab(page
))) {
2766 BUG_ON(!PageCompound(page
));
2771 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2773 EXPORT_SYMBOL(kfree
);
2776 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2777 * the remaining slabs by the number of items in use. The slabs with the
2778 * most items in use come first. New allocations will then fill those up
2779 * and thus they can be removed from the partial lists.
2781 * The slabs with the least items are placed last. This results in them
2782 * being allocated from last increasing the chance that the last objects
2783 * are freed in them.
2785 int kmem_cache_shrink(struct kmem_cache
*s
)
2789 struct kmem_cache_node
*n
;
2792 int objects
= oo_objects(s
->max
);
2793 struct list_head
*slabs_by_inuse
=
2794 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2795 unsigned long flags
;
2797 if (!slabs_by_inuse
)
2801 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2802 n
= get_node(s
, node
);
2807 for (i
= 0; i
< objects
; i
++)
2808 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2810 spin_lock_irqsave(&n
->list_lock
, flags
);
2813 * Build lists indexed by the items in use in each slab.
2815 * Note that concurrent frees may occur while we hold the
2816 * list_lock. page->inuse here is the upper limit.
2818 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2819 if (!page
->inuse
&& slab_trylock(page
)) {
2821 * Must hold slab lock here because slab_free
2822 * may have freed the last object and be
2823 * waiting to release the slab.
2825 list_del(&page
->lru
);
2828 discard_slab(s
, page
);
2830 list_move(&page
->lru
,
2831 slabs_by_inuse
+ page
->inuse
);
2836 * Rebuild the partial list with the slabs filled up most
2837 * first and the least used slabs at the end.
2839 for (i
= objects
- 1; i
>= 0; i
--)
2840 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2842 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2845 kfree(slabs_by_inuse
);
2848 EXPORT_SYMBOL(kmem_cache_shrink
);
2850 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2851 static int slab_mem_going_offline_callback(void *arg
)
2853 struct kmem_cache
*s
;
2855 down_read(&slub_lock
);
2856 list_for_each_entry(s
, &slab_caches
, list
)
2857 kmem_cache_shrink(s
);
2858 up_read(&slub_lock
);
2863 static void slab_mem_offline_callback(void *arg
)
2865 struct kmem_cache_node
*n
;
2866 struct kmem_cache
*s
;
2867 struct memory_notify
*marg
= arg
;
2870 offline_node
= marg
->status_change_nid
;
2873 * If the node still has available memory. we need kmem_cache_node
2876 if (offline_node
< 0)
2879 down_read(&slub_lock
);
2880 list_for_each_entry(s
, &slab_caches
, list
) {
2881 n
= get_node(s
, offline_node
);
2884 * if n->nr_slabs > 0, slabs still exist on the node
2885 * that is going down. We were unable to free them,
2886 * and offline_pages() function shouldn't call this
2887 * callback. So, we must fail.
2889 BUG_ON(slabs_node(s
, offline_node
));
2891 s
->node
[offline_node
] = NULL
;
2892 kmem_cache_free(kmem_cache_node
, n
);
2895 up_read(&slub_lock
);
2898 static int slab_mem_going_online_callback(void *arg
)
2900 struct kmem_cache_node
*n
;
2901 struct kmem_cache
*s
;
2902 struct memory_notify
*marg
= arg
;
2903 int nid
= marg
->status_change_nid
;
2907 * If the node's memory is already available, then kmem_cache_node is
2908 * already created. Nothing to do.
2914 * We are bringing a node online. No memory is available yet. We must
2915 * allocate a kmem_cache_node structure in order to bring the node
2918 down_read(&slub_lock
);
2919 list_for_each_entry(s
, &slab_caches
, list
) {
2921 * XXX: kmem_cache_alloc_node will fallback to other nodes
2922 * since memory is not yet available from the node that
2925 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
2930 init_kmem_cache_node(n
, s
);
2934 up_read(&slub_lock
);
2938 static int slab_memory_callback(struct notifier_block
*self
,
2939 unsigned long action
, void *arg
)
2944 case MEM_GOING_ONLINE
:
2945 ret
= slab_mem_going_online_callback(arg
);
2947 case MEM_GOING_OFFLINE
:
2948 ret
= slab_mem_going_offline_callback(arg
);
2951 case MEM_CANCEL_ONLINE
:
2952 slab_mem_offline_callback(arg
);
2955 case MEM_CANCEL_OFFLINE
:
2959 ret
= notifier_from_errno(ret
);
2965 #endif /* CONFIG_MEMORY_HOTPLUG */
2967 /********************************************************************
2968 * Basic setup of slabs
2969 *******************************************************************/
2972 * Used for early kmem_cache structures that were allocated using
2973 * the page allocator
2976 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
2980 list_add(&s
->list
, &slab_caches
);
2983 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2984 struct kmem_cache_node
*n
= get_node(s
, node
);
2988 list_for_each_entry(p
, &n
->partial
, lru
)
2991 #ifdef CONFIG_SLAB_DEBUG
2992 list_for_each_entry(p
, &n
->full
, lru
)
2999 void __init
kmem_cache_init(void)
3003 struct kmem_cache
*temp_kmem_cache
;
3005 struct kmem_cache
*temp_kmem_cache_node
;
3006 unsigned long kmalloc_size
;
3008 kmem_size
= offsetof(struct kmem_cache
, node
) +
3009 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3011 /* Allocate two kmem_caches from the page allocator */
3012 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3013 order
= get_order(2 * kmalloc_size
);
3014 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3017 * Must first have the slab cache available for the allocations of the
3018 * struct kmem_cache_node's. There is special bootstrap code in
3019 * kmem_cache_open for slab_state == DOWN.
3021 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3023 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3024 sizeof(struct kmem_cache_node
),
3025 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3027 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3029 /* Able to allocate the per node structures */
3030 slab_state
= PARTIAL
;
3032 temp_kmem_cache
= kmem_cache
;
3033 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3034 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3035 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3036 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3039 * Allocate kmem_cache_node properly from the kmem_cache slab.
3040 * kmem_cache_node is separately allocated so no need to
3041 * update any list pointers.
3043 temp_kmem_cache_node
= kmem_cache_node
;
3045 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3046 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3048 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3051 kmem_cache_bootstrap_fixup(kmem_cache
);
3053 /* Free temporary boot structure */
3054 free_pages((unsigned long)temp_kmem_cache
, order
);
3056 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3059 * Patch up the size_index table if we have strange large alignment
3060 * requirements for the kmalloc array. This is only the case for
3061 * MIPS it seems. The standard arches will not generate any code here.
3063 * Largest permitted alignment is 256 bytes due to the way we
3064 * handle the index determination for the smaller caches.
3066 * Make sure that nothing crazy happens if someone starts tinkering
3067 * around with ARCH_KMALLOC_MINALIGN
3069 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3070 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3072 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3073 int elem
= size_index_elem(i
);
3074 if (elem
>= ARRAY_SIZE(size_index
))
3076 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3079 if (KMALLOC_MIN_SIZE
== 64) {
3081 * The 96 byte size cache is not used if the alignment
3084 for (i
= 64 + 8; i
<= 96; i
+= 8)
3085 size_index
[size_index_elem(i
)] = 7;
3086 } else if (KMALLOC_MIN_SIZE
== 128) {
3088 * The 192 byte sized cache is not used if the alignment
3089 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3092 for (i
= 128 + 8; i
<= 192; i
+= 8)
3093 size_index
[size_index_elem(i
)] = 8;
3096 /* Caches that are not of the two-to-the-power-of size */
3097 if (KMALLOC_MIN_SIZE
<= 32) {
3098 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3102 if (KMALLOC_MIN_SIZE
<= 64) {
3103 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3107 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3108 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3114 /* Provide the correct kmalloc names now that the caches are up */
3115 if (KMALLOC_MIN_SIZE
<= 32) {
3116 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3117 BUG_ON(!kmalloc_caches
[1]->name
);
3120 if (KMALLOC_MIN_SIZE
<= 64) {
3121 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3122 BUG_ON(!kmalloc_caches
[2]->name
);
3125 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3126 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3129 kmalloc_caches
[i
]->name
= s
;
3133 register_cpu_notifier(&slab_notifier
);
3136 #ifdef CONFIG_ZONE_DMA
3137 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3138 struct kmem_cache
*s
= kmalloc_caches
[i
];
3141 char *name
= kasprintf(GFP_NOWAIT
,
3142 "dma-kmalloc-%d", s
->objsize
);
3145 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3146 s
->objsize
, SLAB_CACHE_DMA
);
3151 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3152 " CPUs=%d, Nodes=%d\n",
3153 caches
, cache_line_size(),
3154 slub_min_order
, slub_max_order
, slub_min_objects
,
3155 nr_cpu_ids
, nr_node_ids
);
3158 void __init
kmem_cache_init_late(void)
3163 * Find a mergeable slab cache
3165 static int slab_unmergeable(struct kmem_cache
*s
)
3167 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3174 * We may have set a slab to be unmergeable during bootstrap.
3176 if (s
->refcount
< 0)
3182 static struct kmem_cache
*find_mergeable(size_t size
,
3183 size_t align
, unsigned long flags
, const char *name
,
3184 void (*ctor
)(void *))
3186 struct kmem_cache
*s
;
3188 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3194 size
= ALIGN(size
, sizeof(void *));
3195 align
= calculate_alignment(flags
, align
, size
);
3196 size
= ALIGN(size
, align
);
3197 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3199 list_for_each_entry(s
, &slab_caches
, list
) {
3200 if (slab_unmergeable(s
))
3206 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3209 * Check if alignment is compatible.
3210 * Courtesy of Adrian Drzewiecki
3212 if ((s
->size
& ~(align
- 1)) != s
->size
)
3215 if (s
->size
- size
>= sizeof(void *))
3223 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3224 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3226 struct kmem_cache
*s
;
3232 down_write(&slub_lock
);
3233 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3237 * Adjust the object sizes so that we clear
3238 * the complete object on kzalloc.
3240 s
->objsize
= max(s
->objsize
, (int)size
);
3241 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3243 if (sysfs_slab_alias(s
, name
)) {
3247 up_write(&slub_lock
);
3251 n
= kstrdup(name
, GFP_KERNEL
);
3255 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3257 if (kmem_cache_open(s
, n
,
3258 size
, align
, flags
, ctor
)) {
3259 list_add(&s
->list
, &slab_caches
);
3260 if (sysfs_slab_add(s
)) {
3266 up_write(&slub_lock
);
3272 up_write(&slub_lock
);
3275 if (flags
& SLAB_PANIC
)
3276 panic("Cannot create slabcache %s\n", name
);
3281 EXPORT_SYMBOL(kmem_cache_create
);
3285 * Use the cpu notifier to insure that the cpu slabs are flushed when
3288 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3289 unsigned long action
, void *hcpu
)
3291 long cpu
= (long)hcpu
;
3292 struct kmem_cache
*s
;
3293 unsigned long flags
;
3296 case CPU_UP_CANCELED
:
3297 case CPU_UP_CANCELED_FROZEN
:
3299 case CPU_DEAD_FROZEN
:
3300 down_read(&slub_lock
);
3301 list_for_each_entry(s
, &slab_caches
, list
) {
3302 local_irq_save(flags
);
3303 __flush_cpu_slab(s
, cpu
);
3304 local_irq_restore(flags
);
3306 up_read(&slub_lock
);
3314 static struct notifier_block __cpuinitdata slab_notifier
= {
3315 .notifier_call
= slab_cpuup_callback
3320 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3322 struct kmem_cache
*s
;
3325 if (unlikely(size
> SLUB_MAX_SIZE
))
3326 return kmalloc_large(size
, gfpflags
);
3328 s
= get_slab(size
, gfpflags
);
3330 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3333 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
3335 /* Honor the call site pointer we recieved. */
3336 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3341 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3342 int node
, unsigned long caller
)
3344 struct kmem_cache
*s
;
3347 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3348 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3350 trace_kmalloc_node(caller
, ret
,
3351 size
, PAGE_SIZE
<< get_order(size
),
3357 s
= get_slab(size
, gfpflags
);
3359 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3362 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3364 /* Honor the call site pointer we recieved. */
3365 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3370 #ifdef CONFIG_SLUB_DEBUG
3371 static int count_inuse(struct page
*page
)
3376 static int count_total(struct page
*page
)
3378 return page
->objects
;
3381 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3385 void *addr
= page_address(page
);
3387 if (!check_slab(s
, page
) ||
3388 !on_freelist(s
, page
, NULL
))
3391 /* Now we know that a valid freelist exists */
3392 bitmap_zero(map
, page
->objects
);
3394 for_each_free_object(p
, s
, page
->freelist
) {
3395 set_bit(slab_index(p
, s
, addr
), map
);
3396 if (!check_object(s
, page
, p
, 0))
3400 for_each_object(p
, s
, addr
, page
->objects
)
3401 if (!test_bit(slab_index(p
, s
, addr
), map
))
3402 if (!check_object(s
, page
, p
, 1))
3407 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3410 if (slab_trylock(page
)) {
3411 validate_slab(s
, page
, map
);
3414 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3418 static int validate_slab_node(struct kmem_cache
*s
,
3419 struct kmem_cache_node
*n
, unsigned long *map
)
3421 unsigned long count
= 0;
3423 unsigned long flags
;
3425 spin_lock_irqsave(&n
->list_lock
, flags
);
3427 list_for_each_entry(page
, &n
->partial
, lru
) {
3428 validate_slab_slab(s
, page
, map
);
3431 if (count
!= n
->nr_partial
)
3432 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3433 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3435 if (!(s
->flags
& SLAB_STORE_USER
))
3438 list_for_each_entry(page
, &n
->full
, lru
) {
3439 validate_slab_slab(s
, page
, map
);
3442 if (count
!= atomic_long_read(&n
->nr_slabs
))
3443 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3444 "counter=%ld\n", s
->name
, count
,
3445 atomic_long_read(&n
->nr_slabs
));
3448 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3452 static long validate_slab_cache(struct kmem_cache
*s
)
3455 unsigned long count
= 0;
3456 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3457 sizeof(unsigned long), GFP_KERNEL
);
3463 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3464 struct kmem_cache_node
*n
= get_node(s
, node
);
3466 count
+= validate_slab_node(s
, n
, map
);
3472 #ifdef SLUB_RESILIENCY_TEST
3473 static void resiliency_test(void)
3477 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
3479 printk(KERN_ERR
"SLUB resiliency testing\n");
3480 printk(KERN_ERR
"-----------------------\n");
3481 printk(KERN_ERR
"A. Corruption after allocation\n");
3483 p
= kzalloc(16, GFP_KERNEL
);
3485 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3486 " 0x12->0x%p\n\n", p
+ 16);
3488 validate_slab_cache(kmalloc_caches
[4]);
3490 /* Hmmm... The next two are dangerous */
3491 p
= kzalloc(32, GFP_KERNEL
);
3492 p
[32 + sizeof(void *)] = 0x34;
3493 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3494 " 0x34 -> -0x%p\n", p
);
3496 "If allocated object is overwritten then not detectable\n\n");
3498 validate_slab_cache(kmalloc_caches
[5]);
3499 p
= kzalloc(64, GFP_KERNEL
);
3500 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3502 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3505 "If allocated object is overwritten then not detectable\n\n");
3506 validate_slab_cache(kmalloc_caches
[6]);
3508 printk(KERN_ERR
"\nB. Corruption after free\n");
3509 p
= kzalloc(128, GFP_KERNEL
);
3512 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3513 validate_slab_cache(kmalloc_caches
[7]);
3515 p
= kzalloc(256, GFP_KERNEL
);
3518 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3520 validate_slab_cache(kmalloc_caches
[8]);
3522 p
= kzalloc(512, GFP_KERNEL
);
3525 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3526 validate_slab_cache(kmalloc_caches
[9]);
3529 static void resiliency_test(void) {};
3533 * Generate lists of code addresses where slabcache objects are allocated
3538 unsigned long count
;
3545 DECLARE_BITMAP(cpus
, NR_CPUS
);
3551 unsigned long count
;
3552 struct location
*loc
;
3555 static void free_loc_track(struct loc_track
*t
)
3558 free_pages((unsigned long)t
->loc
,
3559 get_order(sizeof(struct location
) * t
->max
));
3562 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3567 order
= get_order(sizeof(struct location
) * max
);
3569 l
= (void *)__get_free_pages(flags
, order
);
3574 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3582 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3583 const struct track
*track
)
3585 long start
, end
, pos
;
3587 unsigned long caddr
;
3588 unsigned long age
= jiffies
- track
->when
;
3594 pos
= start
+ (end
- start
+ 1) / 2;
3597 * There is nothing at "end". If we end up there
3598 * we need to add something to before end.
3603 caddr
= t
->loc
[pos
].addr
;
3604 if (track
->addr
== caddr
) {
3610 if (age
< l
->min_time
)
3612 if (age
> l
->max_time
)
3615 if (track
->pid
< l
->min_pid
)
3616 l
->min_pid
= track
->pid
;
3617 if (track
->pid
> l
->max_pid
)
3618 l
->max_pid
= track
->pid
;
3620 cpumask_set_cpu(track
->cpu
,
3621 to_cpumask(l
->cpus
));
3623 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3627 if (track
->addr
< caddr
)
3634 * Not found. Insert new tracking element.
3636 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3642 (t
->count
- pos
) * sizeof(struct location
));
3645 l
->addr
= track
->addr
;
3649 l
->min_pid
= track
->pid
;
3650 l
->max_pid
= track
->pid
;
3651 cpumask_clear(to_cpumask(l
->cpus
));
3652 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3653 nodes_clear(l
->nodes
);
3654 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3658 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3659 struct page
*page
, enum track_item alloc
,
3662 void *addr
= page_address(page
);
3665 bitmap_zero(map
, page
->objects
);
3666 for_each_free_object(p
, s
, page
->freelist
)
3667 set_bit(slab_index(p
, s
, addr
), map
);
3669 for_each_object(p
, s
, addr
, page
->objects
)
3670 if (!test_bit(slab_index(p
, s
, addr
), map
))
3671 add_location(t
, s
, get_track(s
, p
, alloc
));
3674 static int list_locations(struct kmem_cache
*s
, char *buf
,
3675 enum track_item alloc
)
3679 struct loc_track t
= { 0, 0, NULL
};
3681 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3682 sizeof(unsigned long), GFP_KERNEL
);
3684 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3687 return sprintf(buf
, "Out of memory\n");
3689 /* Push back cpu slabs */
3692 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3693 struct kmem_cache_node
*n
= get_node(s
, node
);
3694 unsigned long flags
;
3697 if (!atomic_long_read(&n
->nr_slabs
))
3700 spin_lock_irqsave(&n
->list_lock
, flags
);
3701 list_for_each_entry(page
, &n
->partial
, lru
)
3702 process_slab(&t
, s
, page
, alloc
, map
);
3703 list_for_each_entry(page
, &n
->full
, lru
)
3704 process_slab(&t
, s
, page
, alloc
, map
);
3705 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3708 for (i
= 0; i
< t
.count
; i
++) {
3709 struct location
*l
= &t
.loc
[i
];
3711 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3713 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3716 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3718 len
+= sprintf(buf
+ len
, "<not-available>");
3720 if (l
->sum_time
!= l
->min_time
) {
3721 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3723 (long)div_u64(l
->sum_time
, l
->count
),
3726 len
+= sprintf(buf
+ len
, " age=%ld",
3729 if (l
->min_pid
!= l
->max_pid
)
3730 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3731 l
->min_pid
, l
->max_pid
);
3733 len
+= sprintf(buf
+ len
, " pid=%ld",
3736 if (num_online_cpus() > 1 &&
3737 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3738 len
< PAGE_SIZE
- 60) {
3739 len
+= sprintf(buf
+ len
, " cpus=");
3740 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3741 to_cpumask(l
->cpus
));
3744 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3745 len
< PAGE_SIZE
- 60) {
3746 len
+= sprintf(buf
+ len
, " nodes=");
3747 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3751 len
+= sprintf(buf
+ len
, "\n");
3757 len
+= sprintf(buf
, "No data\n");
3761 enum slab_stat_type
{
3762 SL_ALL
, /* All slabs */
3763 SL_PARTIAL
, /* Only partially allocated slabs */
3764 SL_CPU
, /* Only slabs used for cpu caches */
3765 SL_OBJECTS
, /* Determine allocated objects not slabs */
3766 SL_TOTAL
/* Determine object capacity not slabs */
3769 #define SO_ALL (1 << SL_ALL)
3770 #define SO_PARTIAL (1 << SL_PARTIAL)
3771 #define SO_CPU (1 << SL_CPU)
3772 #define SO_OBJECTS (1 << SL_OBJECTS)
3773 #define SO_TOTAL (1 << SL_TOTAL)
3775 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3776 char *buf
, unsigned long flags
)
3778 unsigned long total
= 0;
3781 unsigned long *nodes
;
3782 unsigned long *per_cpu
;
3784 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3787 per_cpu
= nodes
+ nr_node_ids
;
3789 if (flags
& SO_CPU
) {
3792 for_each_possible_cpu(cpu
) {
3793 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3795 if (!c
|| c
->node
< 0)
3799 if (flags
& SO_TOTAL
)
3800 x
= c
->page
->objects
;
3801 else if (flags
& SO_OBJECTS
)
3807 nodes
[c
->node
] += x
;
3813 if (flags
& SO_ALL
) {
3814 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3815 struct kmem_cache_node
*n
= get_node(s
, node
);
3817 if (flags
& SO_TOTAL
)
3818 x
= atomic_long_read(&n
->total_objects
);
3819 else if (flags
& SO_OBJECTS
)
3820 x
= atomic_long_read(&n
->total_objects
) -
3821 count_partial(n
, count_free
);
3824 x
= atomic_long_read(&n
->nr_slabs
);
3829 } else if (flags
& SO_PARTIAL
) {
3830 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3831 struct kmem_cache_node
*n
= get_node(s
, node
);
3833 if (flags
& SO_TOTAL
)
3834 x
= count_partial(n
, count_total
);
3835 else if (flags
& SO_OBJECTS
)
3836 x
= count_partial(n
, count_inuse
);
3843 x
= sprintf(buf
, "%lu", total
);
3845 for_each_node_state(node
, N_NORMAL_MEMORY
)
3847 x
+= sprintf(buf
+ x
, " N%d=%lu",
3851 return x
+ sprintf(buf
+ x
, "\n");
3854 static int any_slab_objects(struct kmem_cache
*s
)
3858 for_each_online_node(node
) {
3859 struct kmem_cache_node
*n
= get_node(s
, node
);
3864 if (atomic_long_read(&n
->total_objects
))
3870 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3871 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3873 struct slab_attribute
{
3874 struct attribute attr
;
3875 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3876 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3879 #define SLAB_ATTR_RO(_name) \
3880 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3882 #define SLAB_ATTR(_name) \
3883 static struct slab_attribute _name##_attr = \
3884 __ATTR(_name, 0644, _name##_show, _name##_store)
3886 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3888 return sprintf(buf
, "%d\n", s
->size
);
3890 SLAB_ATTR_RO(slab_size
);
3892 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3894 return sprintf(buf
, "%d\n", s
->align
);
3896 SLAB_ATTR_RO(align
);
3898 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3900 return sprintf(buf
, "%d\n", s
->objsize
);
3902 SLAB_ATTR_RO(object_size
);
3904 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3906 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3908 SLAB_ATTR_RO(objs_per_slab
);
3910 static ssize_t
order_store(struct kmem_cache
*s
,
3911 const char *buf
, size_t length
)
3913 unsigned long order
;
3916 err
= strict_strtoul(buf
, 10, &order
);
3920 if (order
> slub_max_order
|| order
< slub_min_order
)
3923 calculate_sizes(s
, order
);
3927 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3929 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3933 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3935 return sprintf(buf
, "%lu\n", s
->min_partial
);
3938 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3944 err
= strict_strtoul(buf
, 10, &min
);
3948 set_min_partial(s
, min
);
3951 SLAB_ATTR(min_partial
);
3953 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3956 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3958 return n
+ sprintf(buf
+ n
, "\n");
3964 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3966 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3968 SLAB_ATTR_RO(aliases
);
3970 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3972 return show_slab_objects(s
, buf
, SO_ALL
);
3974 SLAB_ATTR_RO(slabs
);
3976 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3978 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3980 SLAB_ATTR_RO(partial
);
3982 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3984 return show_slab_objects(s
, buf
, SO_CPU
);
3986 SLAB_ATTR_RO(cpu_slabs
);
3988 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3990 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3992 SLAB_ATTR_RO(objects
);
3994 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3996 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3998 SLAB_ATTR_RO(objects_partial
);
4000 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4002 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4004 SLAB_ATTR_RO(total_objects
);
4006 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4008 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4011 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4012 const char *buf
, size_t length
)
4014 s
->flags
&= ~SLAB_DEBUG_FREE
;
4016 s
->flags
|= SLAB_DEBUG_FREE
;
4019 SLAB_ATTR(sanity_checks
);
4021 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4023 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4026 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4029 s
->flags
&= ~SLAB_TRACE
;
4031 s
->flags
|= SLAB_TRACE
;
4036 #ifdef CONFIG_FAILSLAB
4037 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4039 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4042 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4045 s
->flags
&= ~SLAB_FAILSLAB
;
4047 s
->flags
|= SLAB_FAILSLAB
;
4050 SLAB_ATTR(failslab
);
4053 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4055 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4058 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4059 const char *buf
, size_t length
)
4061 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4063 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4066 SLAB_ATTR(reclaim_account
);
4068 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4070 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4072 SLAB_ATTR_RO(hwcache_align
);
4074 #ifdef CONFIG_ZONE_DMA
4075 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4077 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4079 SLAB_ATTR_RO(cache_dma
);
4082 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4084 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4086 SLAB_ATTR_RO(destroy_by_rcu
);
4088 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4090 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4093 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4094 const char *buf
, size_t length
)
4096 if (any_slab_objects(s
))
4099 s
->flags
&= ~SLAB_RED_ZONE
;
4101 s
->flags
|= SLAB_RED_ZONE
;
4102 calculate_sizes(s
, -1);
4105 SLAB_ATTR(red_zone
);
4107 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4109 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4112 static ssize_t
poison_store(struct kmem_cache
*s
,
4113 const char *buf
, size_t length
)
4115 if (any_slab_objects(s
))
4118 s
->flags
&= ~SLAB_POISON
;
4120 s
->flags
|= SLAB_POISON
;
4121 calculate_sizes(s
, -1);
4126 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4128 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4131 static ssize_t
store_user_store(struct kmem_cache
*s
,
4132 const char *buf
, size_t length
)
4134 if (any_slab_objects(s
))
4137 s
->flags
&= ~SLAB_STORE_USER
;
4139 s
->flags
|= SLAB_STORE_USER
;
4140 calculate_sizes(s
, -1);
4143 SLAB_ATTR(store_user
);
4145 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4150 static ssize_t
validate_store(struct kmem_cache
*s
,
4151 const char *buf
, size_t length
)
4155 if (buf
[0] == '1') {
4156 ret
= validate_slab_cache(s
);
4162 SLAB_ATTR(validate
);
4164 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4169 static ssize_t
shrink_store(struct kmem_cache
*s
,
4170 const char *buf
, size_t length
)
4172 if (buf
[0] == '1') {
4173 int rc
= kmem_cache_shrink(s
);
4183 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4185 if (!(s
->flags
& SLAB_STORE_USER
))
4187 return list_locations(s
, buf
, TRACK_ALLOC
);
4189 SLAB_ATTR_RO(alloc_calls
);
4191 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4193 if (!(s
->flags
& SLAB_STORE_USER
))
4195 return list_locations(s
, buf
, TRACK_FREE
);
4197 SLAB_ATTR_RO(free_calls
);
4200 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4202 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4205 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4206 const char *buf
, size_t length
)
4208 unsigned long ratio
;
4211 err
= strict_strtoul(buf
, 10, &ratio
);
4216 s
->remote_node_defrag_ratio
= ratio
* 10;
4220 SLAB_ATTR(remote_node_defrag_ratio
);
4223 #ifdef CONFIG_SLUB_STATS
4224 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4226 unsigned long sum
= 0;
4229 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4234 for_each_online_cpu(cpu
) {
4235 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4241 len
= sprintf(buf
, "%lu", sum
);
4244 for_each_online_cpu(cpu
) {
4245 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4246 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4250 return len
+ sprintf(buf
+ len
, "\n");
4253 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4257 for_each_online_cpu(cpu
)
4258 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4261 #define STAT_ATTR(si, text) \
4262 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4264 return show_stat(s, buf, si); \
4266 static ssize_t text##_store(struct kmem_cache *s, \
4267 const char *buf, size_t length) \
4269 if (buf[0] != '0') \
4271 clear_stat(s, si); \
4276 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4277 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4278 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4279 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4280 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4281 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4282 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4283 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4284 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4285 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4286 STAT_ATTR(FREE_SLAB
, free_slab
);
4287 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4288 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4289 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4290 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4291 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4292 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4293 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4296 static struct attribute
*slab_attrs
[] = {
4297 &slab_size_attr
.attr
,
4298 &object_size_attr
.attr
,
4299 &objs_per_slab_attr
.attr
,
4301 &min_partial_attr
.attr
,
4303 &objects_partial_attr
.attr
,
4304 &total_objects_attr
.attr
,
4307 &cpu_slabs_attr
.attr
,
4311 &sanity_checks_attr
.attr
,
4313 &hwcache_align_attr
.attr
,
4314 &reclaim_account_attr
.attr
,
4315 &destroy_by_rcu_attr
.attr
,
4316 &red_zone_attr
.attr
,
4318 &store_user_attr
.attr
,
4319 &validate_attr
.attr
,
4321 &alloc_calls_attr
.attr
,
4322 &free_calls_attr
.attr
,
4323 #ifdef CONFIG_ZONE_DMA
4324 &cache_dma_attr
.attr
,
4327 &remote_node_defrag_ratio_attr
.attr
,
4329 #ifdef CONFIG_SLUB_STATS
4330 &alloc_fastpath_attr
.attr
,
4331 &alloc_slowpath_attr
.attr
,
4332 &free_fastpath_attr
.attr
,
4333 &free_slowpath_attr
.attr
,
4334 &free_frozen_attr
.attr
,
4335 &free_add_partial_attr
.attr
,
4336 &free_remove_partial_attr
.attr
,
4337 &alloc_from_partial_attr
.attr
,
4338 &alloc_slab_attr
.attr
,
4339 &alloc_refill_attr
.attr
,
4340 &free_slab_attr
.attr
,
4341 &cpuslab_flush_attr
.attr
,
4342 &deactivate_full_attr
.attr
,
4343 &deactivate_empty_attr
.attr
,
4344 &deactivate_to_head_attr
.attr
,
4345 &deactivate_to_tail_attr
.attr
,
4346 &deactivate_remote_frees_attr
.attr
,
4347 &order_fallback_attr
.attr
,
4349 #ifdef CONFIG_FAILSLAB
4350 &failslab_attr
.attr
,
4356 static struct attribute_group slab_attr_group
= {
4357 .attrs
= slab_attrs
,
4360 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4361 struct attribute
*attr
,
4364 struct slab_attribute
*attribute
;
4365 struct kmem_cache
*s
;
4368 attribute
= to_slab_attr(attr
);
4371 if (!attribute
->show
)
4374 err
= attribute
->show(s
, buf
);
4379 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4380 struct attribute
*attr
,
4381 const char *buf
, size_t len
)
4383 struct slab_attribute
*attribute
;
4384 struct kmem_cache
*s
;
4387 attribute
= to_slab_attr(attr
);
4390 if (!attribute
->store
)
4393 err
= attribute
->store(s
, buf
, len
);
4398 static void kmem_cache_release(struct kobject
*kobj
)
4400 struct kmem_cache
*s
= to_slab(kobj
);
4406 static const struct sysfs_ops slab_sysfs_ops
= {
4407 .show
= slab_attr_show
,
4408 .store
= slab_attr_store
,
4411 static struct kobj_type slab_ktype
= {
4412 .sysfs_ops
= &slab_sysfs_ops
,
4413 .release
= kmem_cache_release
4416 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4418 struct kobj_type
*ktype
= get_ktype(kobj
);
4420 if (ktype
== &slab_ktype
)
4425 static const struct kset_uevent_ops slab_uevent_ops
= {
4426 .filter
= uevent_filter
,
4429 static struct kset
*slab_kset
;
4431 #define ID_STR_LENGTH 64
4433 /* Create a unique string id for a slab cache:
4435 * Format :[flags-]size
4437 static char *create_unique_id(struct kmem_cache
*s
)
4439 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4446 * First flags affecting slabcache operations. We will only
4447 * get here for aliasable slabs so we do not need to support
4448 * too many flags. The flags here must cover all flags that
4449 * are matched during merging to guarantee that the id is
4452 if (s
->flags
& SLAB_CACHE_DMA
)
4454 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4456 if (s
->flags
& SLAB_DEBUG_FREE
)
4458 if (!(s
->flags
& SLAB_NOTRACK
))
4462 p
+= sprintf(p
, "%07d", s
->size
);
4463 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4467 static int sysfs_slab_add(struct kmem_cache
*s
)
4473 if (slab_state
< SYSFS
)
4474 /* Defer until later */
4477 unmergeable
= slab_unmergeable(s
);
4480 * Slabcache can never be merged so we can use the name proper.
4481 * This is typically the case for debug situations. In that
4482 * case we can catch duplicate names easily.
4484 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4488 * Create a unique name for the slab as a target
4491 name
= create_unique_id(s
);
4494 s
->kobj
.kset
= slab_kset
;
4495 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4497 kobject_put(&s
->kobj
);
4501 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4503 kobject_del(&s
->kobj
);
4504 kobject_put(&s
->kobj
);
4507 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4509 /* Setup first alias */
4510 sysfs_slab_alias(s
, s
->name
);
4516 static void sysfs_slab_remove(struct kmem_cache
*s
)
4518 if (slab_state
< SYSFS
)
4520 * Sysfs has not been setup yet so no need to remove the
4525 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4526 kobject_del(&s
->kobj
);
4527 kobject_put(&s
->kobj
);
4531 * Need to buffer aliases during bootup until sysfs becomes
4532 * available lest we lose that information.
4534 struct saved_alias
{
4535 struct kmem_cache
*s
;
4537 struct saved_alias
*next
;
4540 static struct saved_alias
*alias_list
;
4542 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4544 struct saved_alias
*al
;
4546 if (slab_state
== SYSFS
) {
4548 * If we have a leftover link then remove it.
4550 sysfs_remove_link(&slab_kset
->kobj
, name
);
4551 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4554 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4560 al
->next
= alias_list
;
4565 static int __init
slab_sysfs_init(void)
4567 struct kmem_cache
*s
;
4570 down_write(&slub_lock
);
4572 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4574 up_write(&slub_lock
);
4575 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4581 list_for_each_entry(s
, &slab_caches
, list
) {
4582 err
= sysfs_slab_add(s
);
4584 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4585 " to sysfs\n", s
->name
);
4588 while (alias_list
) {
4589 struct saved_alias
*al
= alias_list
;
4591 alias_list
= alias_list
->next
;
4592 err
= sysfs_slab_alias(al
->s
, al
->name
);
4594 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4595 " %s to sysfs\n", s
->name
);
4599 up_write(&slub_lock
);
4604 __initcall(slab_sysfs_init
);
4608 * The /proc/slabinfo ABI
4610 #ifdef CONFIG_SLABINFO
4611 static void print_slabinfo_header(struct seq_file
*m
)
4613 seq_puts(m
, "slabinfo - version: 2.1\n");
4614 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4615 "<objperslab> <pagesperslab>");
4616 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4617 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4621 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4625 down_read(&slub_lock
);
4627 print_slabinfo_header(m
);
4629 return seq_list_start(&slab_caches
, *pos
);
4632 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4634 return seq_list_next(p
, &slab_caches
, pos
);
4637 static void s_stop(struct seq_file
*m
, void *p
)
4639 up_read(&slub_lock
);
4642 static int s_show(struct seq_file
*m
, void *p
)
4644 unsigned long nr_partials
= 0;
4645 unsigned long nr_slabs
= 0;
4646 unsigned long nr_inuse
= 0;
4647 unsigned long nr_objs
= 0;
4648 unsigned long nr_free
= 0;
4649 struct kmem_cache
*s
;
4652 s
= list_entry(p
, struct kmem_cache
, list
);
4654 for_each_online_node(node
) {
4655 struct kmem_cache_node
*n
= get_node(s
, node
);
4660 nr_partials
+= n
->nr_partial
;
4661 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4662 nr_objs
+= atomic_long_read(&n
->total_objects
);
4663 nr_free
+= count_partial(n
, count_free
);
4666 nr_inuse
= nr_objs
- nr_free
;
4668 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4669 nr_objs
, s
->size
, oo_objects(s
->oo
),
4670 (1 << oo_order(s
->oo
)));
4671 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4672 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4678 static const struct seq_operations slabinfo_op
= {
4685 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4687 return seq_open(file
, &slabinfo_op
);
4690 static const struct file_operations proc_slabinfo_operations
= {
4691 .open
= slabinfo_open
,
4693 .llseek
= seq_lseek
,
4694 .release
= seq_release
,
4697 static int __init
slab_proc_init(void)
4699 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4702 module_init(slab_proc_init
);
4703 #endif /* CONFIG_SLABINFO */