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
, int active
)
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
,
504 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
505 s
->inuse
- s
->objsize
);
508 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
511 if (*start
!= (u8
)value
)
519 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
520 void *from
, void *to
)
522 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
523 memset(from
, data
, to
- from
);
526 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
527 u8
*object
, char *what
,
528 u8
*start
, unsigned int value
, unsigned int bytes
)
533 fault
= check_bytes(start
, value
, bytes
);
538 while (end
> fault
&& end
[-1] == value
)
541 slab_bug(s
, "%s overwritten", what
);
542 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
543 fault
, end
- 1, fault
[0], value
);
544 print_trailer(s
, page
, object
);
546 restore_bytes(s
, what
, value
, fault
, end
);
554 * Bytes of the object to be managed.
555 * If the freepointer may overlay the object then the free
556 * pointer is the first word of the object.
558 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
561 * object + s->objsize
562 * Padding to reach word boundary. This is also used for Redzoning.
563 * Padding is extended by another word if Redzoning is enabled and
566 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
567 * 0xcc (RED_ACTIVE) for objects in use.
570 * Meta data starts here.
572 * A. Free pointer (if we cannot overwrite object on free)
573 * B. Tracking data for SLAB_STORE_USER
574 * C. Padding to reach required alignment boundary or at mininum
575 * one word if debugging is on to be able to detect writes
576 * before the word boundary.
578 * Padding is done using 0x5a (POISON_INUSE)
581 * Nothing is used beyond s->size.
583 * If slabcaches are merged then the objsize and inuse boundaries are mostly
584 * ignored. And therefore no slab options that rely on these boundaries
585 * may be used with merged slabcaches.
588 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
590 unsigned long off
= s
->inuse
; /* The end of info */
593 /* Freepointer is placed after the object. */
594 off
+= sizeof(void *);
596 if (s
->flags
& SLAB_STORE_USER
)
597 /* We also have user information there */
598 off
+= 2 * sizeof(struct track
);
603 return check_bytes_and_report(s
, page
, p
, "Object padding",
604 p
+ off
, POISON_INUSE
, s
->size
- off
);
607 /* Check the pad bytes at the end of a slab page */
608 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
616 if (!(s
->flags
& SLAB_POISON
))
619 start
= page_address(page
);
620 length
= (PAGE_SIZE
<< compound_order(page
));
621 end
= start
+ length
;
622 remainder
= length
% s
->size
;
626 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
629 while (end
> fault
&& end
[-1] == POISON_INUSE
)
632 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
633 print_section("Padding", end
- remainder
, remainder
);
635 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
639 static int check_object(struct kmem_cache
*s
, struct page
*page
,
640 void *object
, int active
)
643 u8
*endobject
= object
+ s
->objsize
;
645 if (s
->flags
& SLAB_RED_ZONE
) {
647 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
649 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
650 endobject
, red
, s
->inuse
- s
->objsize
))
653 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
654 check_bytes_and_report(s
, page
, p
, "Alignment padding",
655 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
659 if (s
->flags
& SLAB_POISON
) {
660 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
661 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
662 POISON_FREE
, s
->objsize
- 1) ||
663 !check_bytes_and_report(s
, page
, p
, "Poison",
664 p
+ s
->objsize
- 1, POISON_END
, 1)))
667 * check_pad_bytes cleans up on its own.
669 check_pad_bytes(s
, page
, p
);
672 if (!s
->offset
&& active
)
674 * Object and freepointer overlap. Cannot check
675 * freepointer while object is allocated.
679 /* Check free pointer validity */
680 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
681 object_err(s
, page
, p
, "Freepointer corrupt");
683 * No choice but to zap it and thus lose the remainder
684 * of the free objects in this slab. May cause
685 * another error because the object count is now wrong.
687 set_freepointer(s
, p
, NULL
);
693 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
697 VM_BUG_ON(!irqs_disabled());
699 if (!PageSlab(page
)) {
700 slab_err(s
, page
, "Not a valid slab page");
704 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
705 if (page
->objects
> maxobj
) {
706 slab_err(s
, page
, "objects %u > max %u",
707 s
->name
, page
->objects
, maxobj
);
710 if (page
->inuse
> page
->objects
) {
711 slab_err(s
, page
, "inuse %u > max %u",
712 s
->name
, page
->inuse
, page
->objects
);
715 /* Slab_pad_check fixes things up after itself */
716 slab_pad_check(s
, page
);
721 * Determine if a certain object on a page is on the freelist. Must hold the
722 * slab lock to guarantee that the chains are in a consistent state.
724 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
727 void *fp
= page
->freelist
;
729 unsigned long max_objects
;
731 while (fp
&& nr
<= page
->objects
) {
734 if (!check_valid_pointer(s
, page
, fp
)) {
736 object_err(s
, page
, object
,
737 "Freechain corrupt");
738 set_freepointer(s
, object
, NULL
);
741 slab_err(s
, page
, "Freepointer corrupt");
742 page
->freelist
= NULL
;
743 page
->inuse
= page
->objects
;
744 slab_fix(s
, "Freelist cleared");
750 fp
= get_freepointer(s
, object
);
754 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
755 if (max_objects
> MAX_OBJS_PER_PAGE
)
756 max_objects
= MAX_OBJS_PER_PAGE
;
758 if (page
->objects
!= max_objects
) {
759 slab_err(s
, page
, "Wrong number of objects. Found %d but "
760 "should be %d", page
->objects
, max_objects
);
761 page
->objects
= max_objects
;
762 slab_fix(s
, "Number of objects adjusted.");
764 if (page
->inuse
!= page
->objects
- nr
) {
765 slab_err(s
, page
, "Wrong object count. Counter is %d but "
766 "counted were %d", page
->inuse
, page
->objects
- nr
);
767 page
->inuse
= page
->objects
- nr
;
768 slab_fix(s
, "Object count adjusted.");
770 return search
== NULL
;
773 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
776 if (s
->flags
& SLAB_TRACE
) {
777 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
779 alloc
? "alloc" : "free",
784 print_section("Object", (void *)object
, s
->objsize
);
791 * Hooks for other subsystems that check memory allocations. In a typical
792 * production configuration these hooks all should produce no code at all.
794 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
796 flags
&= gfp_allowed_mask
;
797 lockdep_trace_alloc(flags
);
798 might_sleep_if(flags
& __GFP_WAIT
);
800 return should_failslab(s
->objsize
, flags
, s
->flags
);
803 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
805 flags
&= gfp_allowed_mask
;
806 kmemcheck_slab_alloc(s
, flags
, object
, s
->objsize
);
807 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
810 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
812 kmemleak_free_recursive(x
, s
->flags
);
815 static inline void slab_free_hook_irq(struct kmem_cache
*s
, void *object
)
817 kmemcheck_slab_free(s
, object
, s
->objsize
);
818 debug_check_no_locks_freed(object
, s
->objsize
);
819 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
820 debug_check_no_obj_freed(object
, s
->objsize
);
824 * Tracking of fully allocated slabs for debugging purposes.
826 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
828 spin_lock(&n
->list_lock
);
829 list_add(&page
->lru
, &n
->full
);
830 spin_unlock(&n
->list_lock
);
833 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
835 struct kmem_cache_node
*n
;
837 if (!(s
->flags
& SLAB_STORE_USER
))
840 n
= get_node(s
, page_to_nid(page
));
842 spin_lock(&n
->list_lock
);
843 list_del(&page
->lru
);
844 spin_unlock(&n
->list_lock
);
847 /* Tracking of the number of slabs for debugging purposes */
848 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
850 struct kmem_cache_node
*n
= get_node(s
, node
);
852 return atomic_long_read(&n
->nr_slabs
);
855 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
857 return atomic_long_read(&n
->nr_slabs
);
860 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
862 struct kmem_cache_node
*n
= get_node(s
, node
);
865 * May be called early in order to allocate a slab for the
866 * kmem_cache_node structure. Solve the chicken-egg
867 * dilemma by deferring the increment of the count during
868 * bootstrap (see early_kmem_cache_node_alloc).
871 atomic_long_inc(&n
->nr_slabs
);
872 atomic_long_add(objects
, &n
->total_objects
);
875 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
877 struct kmem_cache_node
*n
= get_node(s
, node
);
879 atomic_long_dec(&n
->nr_slabs
);
880 atomic_long_sub(objects
, &n
->total_objects
);
883 /* Object debug checks for alloc/free paths */
884 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
887 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
890 init_object(s
, object
, 0);
891 init_tracking(s
, object
);
894 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
895 void *object
, unsigned long addr
)
897 if (!check_slab(s
, page
))
900 if (!on_freelist(s
, page
, object
)) {
901 object_err(s
, page
, object
, "Object already allocated");
905 if (!check_valid_pointer(s
, page
, object
)) {
906 object_err(s
, page
, object
, "Freelist Pointer check fails");
910 if (!check_object(s
, page
, object
, 0))
913 /* Success perform special debug activities for allocs */
914 if (s
->flags
& SLAB_STORE_USER
)
915 set_track(s
, object
, TRACK_ALLOC
, addr
);
916 trace(s
, page
, object
, 1);
917 init_object(s
, object
, 1);
921 if (PageSlab(page
)) {
923 * If this is a slab page then lets do the best we can
924 * to avoid issues in the future. Marking all objects
925 * as used avoids touching the remaining objects.
927 slab_fix(s
, "Marking all objects used");
928 page
->inuse
= page
->objects
;
929 page
->freelist
= NULL
;
934 static noinline
int free_debug_processing(struct kmem_cache
*s
,
935 struct page
*page
, void *object
, unsigned long addr
)
937 if (!check_slab(s
, page
))
940 if (!check_valid_pointer(s
, page
, object
)) {
941 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
945 if (on_freelist(s
, page
, object
)) {
946 object_err(s
, page
, object
, "Object already free");
950 if (!check_object(s
, page
, object
, 1))
953 if (unlikely(s
!= page
->slab
)) {
954 if (!PageSlab(page
)) {
955 slab_err(s
, page
, "Attempt to free object(0x%p) "
956 "outside of slab", object
);
957 } else if (!page
->slab
) {
959 "SLUB <none>: no slab for object 0x%p.\n",
963 object_err(s
, page
, object
,
964 "page slab pointer corrupt.");
968 /* Special debug activities for freeing objects */
969 if (!PageSlubFrozen(page
) && !page
->freelist
)
970 remove_full(s
, page
);
971 if (s
->flags
& SLAB_STORE_USER
)
972 set_track(s
, object
, TRACK_FREE
, addr
);
973 trace(s
, page
, object
, 0);
974 init_object(s
, object
, 0);
978 slab_fix(s
, "Object at 0x%p not freed", object
);
982 static int __init
setup_slub_debug(char *str
)
984 slub_debug
= DEBUG_DEFAULT_FLAGS
;
985 if (*str
++ != '=' || !*str
)
987 * No options specified. Switch on full debugging.
993 * No options but restriction on slabs. This means full
994 * debugging for slabs matching a pattern.
998 if (tolower(*str
) == 'o') {
1000 * Avoid enabling debugging on caches if its minimum order
1001 * would increase as a result.
1003 disable_higher_order_debug
= 1;
1010 * Switch off all debugging measures.
1015 * Determine which debug features should be switched on
1017 for (; *str
&& *str
!= ','; str
++) {
1018 switch (tolower(*str
)) {
1020 slub_debug
|= SLAB_DEBUG_FREE
;
1023 slub_debug
|= SLAB_RED_ZONE
;
1026 slub_debug
|= SLAB_POISON
;
1029 slub_debug
|= SLAB_STORE_USER
;
1032 slub_debug
|= SLAB_TRACE
;
1035 slub_debug
|= SLAB_FAILSLAB
;
1038 printk(KERN_ERR
"slub_debug option '%c' "
1039 "unknown. skipped\n", *str
);
1045 slub_debug_slabs
= str
+ 1;
1050 __setup("slub_debug", setup_slub_debug
);
1052 static unsigned long kmem_cache_flags(unsigned long objsize
,
1053 unsigned long flags
, const char *name
,
1054 void (*ctor
)(void *))
1057 * Enable debugging if selected on the kernel commandline.
1059 if (slub_debug
&& (!slub_debug_slabs
||
1060 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1061 flags
|= slub_debug
;
1066 static inline void setup_object_debug(struct kmem_cache
*s
,
1067 struct page
*page
, void *object
) {}
1069 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1070 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1072 static inline int free_debug_processing(struct kmem_cache
*s
,
1073 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1075 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1077 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1078 void *object
, int active
) { return 1; }
1079 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1080 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1081 unsigned long flags
, const char *name
,
1082 void (*ctor
)(void *))
1086 #define slub_debug 0
1088 #define disable_higher_order_debug 0
1090 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1092 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1094 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1096 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1099 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1102 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1105 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1107 static inline void slab_free_hook_irq(struct kmem_cache
*s
,
1113 * Slab allocation and freeing
1115 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1116 struct kmem_cache_order_objects oo
)
1118 int order
= oo_order(oo
);
1120 flags
|= __GFP_NOTRACK
;
1122 if (node
== NUMA_NO_NODE
)
1123 return alloc_pages(flags
, order
);
1125 return alloc_pages_exact_node(node
, flags
, order
);
1128 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1131 struct kmem_cache_order_objects oo
= s
->oo
;
1134 flags
|= s
->allocflags
;
1137 * Let the initial higher-order allocation fail under memory pressure
1138 * so we fall-back to the minimum order allocation.
1140 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1142 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1143 if (unlikely(!page
)) {
1146 * Allocation may have failed due to fragmentation.
1147 * Try a lower order alloc if possible
1149 page
= alloc_slab_page(flags
, node
, oo
);
1153 stat(s
, ORDER_FALLBACK
);
1156 if (kmemcheck_enabled
1157 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1158 int pages
= 1 << oo_order(oo
);
1160 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1163 * Objects from caches that have a constructor don't get
1164 * cleared when they're allocated, so we need to do it here.
1167 kmemcheck_mark_uninitialized_pages(page
, pages
);
1169 kmemcheck_mark_unallocated_pages(page
, pages
);
1172 page
->objects
= oo_objects(oo
);
1173 mod_zone_page_state(page_zone(page
),
1174 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1175 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1181 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1184 setup_object_debug(s
, page
, object
);
1185 if (unlikely(s
->ctor
))
1189 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1196 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1198 page
= allocate_slab(s
,
1199 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1203 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1205 page
->flags
|= 1 << PG_slab
;
1207 start
= page_address(page
);
1209 if (unlikely(s
->flags
& SLAB_POISON
))
1210 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1213 for_each_object(p
, s
, start
, page
->objects
) {
1214 setup_object(s
, page
, last
);
1215 set_freepointer(s
, last
, p
);
1218 setup_object(s
, page
, last
);
1219 set_freepointer(s
, last
, NULL
);
1221 page
->freelist
= start
;
1227 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1229 int order
= compound_order(page
);
1230 int pages
= 1 << order
;
1232 if (kmem_cache_debug(s
)) {
1235 slab_pad_check(s
, page
);
1236 for_each_object(p
, s
, page_address(page
),
1238 check_object(s
, page
, p
, 0);
1241 kmemcheck_free_shadow(page
, compound_order(page
));
1243 mod_zone_page_state(page_zone(page
),
1244 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1245 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1248 __ClearPageSlab(page
);
1249 reset_page_mapcount(page
);
1250 if (current
->reclaim_state
)
1251 current
->reclaim_state
->reclaimed_slab
+= pages
;
1252 __free_pages(page
, order
);
1255 static void rcu_free_slab(struct rcu_head
*h
)
1259 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1260 __free_slab(page
->slab
, page
);
1263 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1265 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1267 * RCU free overloads the RCU head over the LRU
1269 struct rcu_head
*head
= (void *)&page
->lru
;
1271 call_rcu(head
, rcu_free_slab
);
1273 __free_slab(s
, page
);
1276 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1278 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1283 * Per slab locking using the pagelock
1285 static __always_inline
void slab_lock(struct page
*page
)
1287 bit_spin_lock(PG_locked
, &page
->flags
);
1290 static __always_inline
void slab_unlock(struct page
*page
)
1292 __bit_spin_unlock(PG_locked
, &page
->flags
);
1295 static __always_inline
int slab_trylock(struct page
*page
)
1299 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1304 * Management of partially allocated slabs
1306 static void add_partial(struct kmem_cache_node
*n
,
1307 struct page
*page
, int tail
)
1309 spin_lock(&n
->list_lock
);
1312 list_add_tail(&page
->lru
, &n
->partial
);
1314 list_add(&page
->lru
, &n
->partial
);
1315 spin_unlock(&n
->list_lock
);
1318 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1320 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1322 spin_lock(&n
->list_lock
);
1323 list_del(&page
->lru
);
1325 spin_unlock(&n
->list_lock
);
1329 * Lock slab and remove from the partial list.
1331 * Must hold list_lock.
1333 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1336 if (slab_trylock(page
)) {
1337 list_del(&page
->lru
);
1339 __SetPageSlubFrozen(page
);
1346 * Try to allocate a partial slab from a specific node.
1348 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1353 * Racy check. If we mistakenly see no partial slabs then we
1354 * just allocate an empty slab. If we mistakenly try to get a
1355 * partial slab and there is none available then get_partials()
1358 if (!n
|| !n
->nr_partial
)
1361 spin_lock(&n
->list_lock
);
1362 list_for_each_entry(page
, &n
->partial
, lru
)
1363 if (lock_and_freeze_slab(n
, page
))
1367 spin_unlock(&n
->list_lock
);
1372 * Get a page from somewhere. Search in increasing NUMA distances.
1374 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1377 struct zonelist
*zonelist
;
1380 enum zone_type high_zoneidx
= gfp_zone(flags
);
1384 * The defrag ratio allows a configuration of the tradeoffs between
1385 * inter node defragmentation and node local allocations. A lower
1386 * defrag_ratio increases the tendency to do local allocations
1387 * instead of attempting to obtain partial slabs from other nodes.
1389 * If the defrag_ratio is set to 0 then kmalloc() always
1390 * returns node local objects. If the ratio is higher then kmalloc()
1391 * may return off node objects because partial slabs are obtained
1392 * from other nodes and filled up.
1394 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1395 * defrag_ratio = 1000) then every (well almost) allocation will
1396 * first attempt to defrag slab caches on other nodes. This means
1397 * scanning over all nodes to look for partial slabs which may be
1398 * expensive if we do it every time we are trying to find a slab
1399 * with available objects.
1401 if (!s
->remote_node_defrag_ratio
||
1402 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1406 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1407 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1408 struct kmem_cache_node
*n
;
1410 n
= get_node(s
, zone_to_nid(zone
));
1412 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1413 n
->nr_partial
> s
->min_partial
) {
1414 page
= get_partial_node(n
);
1427 * Get a partial page, lock it and return it.
1429 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1432 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1434 page
= get_partial_node(get_node(s
, searchnode
));
1435 if (page
|| node
!= -1)
1438 return get_any_partial(s
, flags
);
1442 * Move a page back to the lists.
1444 * Must be called with the slab lock held.
1446 * On exit the slab lock will have been dropped.
1448 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1450 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1452 __ClearPageSlubFrozen(page
);
1455 if (page
->freelist
) {
1456 add_partial(n
, page
, tail
);
1457 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1459 stat(s
, DEACTIVATE_FULL
);
1460 if (kmem_cache_debug(s
) && (s
->flags
& SLAB_STORE_USER
))
1465 stat(s
, DEACTIVATE_EMPTY
);
1466 if (n
->nr_partial
< s
->min_partial
) {
1468 * Adding an empty slab to the partial slabs in order
1469 * to avoid page allocator overhead. This slab needs
1470 * to come after the other slabs with objects in
1471 * so that the others get filled first. That way the
1472 * size of the partial list stays small.
1474 * kmem_cache_shrink can reclaim any empty slabs from
1477 add_partial(n
, page
, 1);
1482 discard_slab(s
, page
);
1488 * Remove the cpu slab
1490 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1492 struct page
*page
= c
->page
;
1496 stat(s
, DEACTIVATE_REMOTE_FREES
);
1498 * Merge cpu freelist into slab freelist. Typically we get here
1499 * because both freelists are empty. So this is unlikely
1502 while (unlikely(c
->freelist
)) {
1505 tail
= 0; /* Hot objects. Put the slab first */
1507 /* Retrieve object from cpu_freelist */
1508 object
= c
->freelist
;
1509 c
->freelist
= get_freepointer(s
, c
->freelist
);
1511 /* And put onto the regular freelist */
1512 set_freepointer(s
, object
, page
->freelist
);
1513 page
->freelist
= object
;
1517 unfreeze_slab(s
, page
, tail
);
1520 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1522 stat(s
, CPUSLAB_FLUSH
);
1524 deactivate_slab(s
, c
);
1530 * Called from IPI handler with interrupts disabled.
1532 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1534 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1536 if (likely(c
&& c
->page
))
1540 static void flush_cpu_slab(void *d
)
1542 struct kmem_cache
*s
= d
;
1544 __flush_cpu_slab(s
, smp_processor_id());
1547 static void flush_all(struct kmem_cache
*s
)
1549 on_each_cpu(flush_cpu_slab
, s
, 1);
1553 * Check if the objects in a per cpu structure fit numa
1554 * locality expectations.
1556 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1559 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
1565 static int count_free(struct page
*page
)
1567 return page
->objects
- page
->inuse
;
1570 static unsigned long count_partial(struct kmem_cache_node
*n
,
1571 int (*get_count
)(struct page
*))
1573 unsigned long flags
;
1574 unsigned long x
= 0;
1577 spin_lock_irqsave(&n
->list_lock
, flags
);
1578 list_for_each_entry(page
, &n
->partial
, lru
)
1579 x
+= get_count(page
);
1580 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1584 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1586 #ifdef CONFIG_SLUB_DEBUG
1587 return atomic_long_read(&n
->total_objects
);
1593 static noinline
void
1594 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1599 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1601 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1602 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1603 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1605 if (oo_order(s
->min
) > get_order(s
->objsize
))
1606 printk(KERN_WARNING
" %s debugging increased min order, use "
1607 "slub_debug=O to disable.\n", s
->name
);
1609 for_each_online_node(node
) {
1610 struct kmem_cache_node
*n
= get_node(s
, node
);
1611 unsigned long nr_slabs
;
1612 unsigned long nr_objs
;
1613 unsigned long nr_free
;
1618 nr_free
= count_partial(n
, count_free
);
1619 nr_slabs
= node_nr_slabs(n
);
1620 nr_objs
= node_nr_objs(n
);
1623 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1624 node
, nr_slabs
, nr_objs
, nr_free
);
1629 * Slow path. The lockless freelist is empty or we need to perform
1632 * Interrupts are disabled.
1634 * Processing is still very fast if new objects have been freed to the
1635 * regular freelist. In that case we simply take over the regular freelist
1636 * as the lockless freelist and zap the regular freelist.
1638 * If that is not working then we fall back to the partial lists. We take the
1639 * first element of the freelist as the object to allocate now and move the
1640 * rest of the freelist to the lockless freelist.
1642 * And if we were unable to get a new slab from the partial slab lists then
1643 * we need to allocate a new slab. This is the slowest path since it involves
1644 * a call to the page allocator and the setup of a new slab.
1646 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1647 unsigned long addr
, struct kmem_cache_cpu
*c
)
1652 /* We handle __GFP_ZERO in the caller */
1653 gfpflags
&= ~__GFP_ZERO
;
1659 if (unlikely(!node_match(c
, node
)))
1662 stat(s
, ALLOC_REFILL
);
1665 object
= c
->page
->freelist
;
1666 if (unlikely(!object
))
1668 if (kmem_cache_debug(s
))
1671 c
->freelist
= get_freepointer(s
, object
);
1672 c
->page
->inuse
= c
->page
->objects
;
1673 c
->page
->freelist
= NULL
;
1674 c
->node
= page_to_nid(c
->page
);
1676 slab_unlock(c
->page
);
1677 stat(s
, ALLOC_SLOWPATH
);
1681 deactivate_slab(s
, c
);
1684 new = get_partial(s
, gfpflags
, node
);
1687 stat(s
, ALLOC_FROM_PARTIAL
);
1691 gfpflags
&= gfp_allowed_mask
;
1692 if (gfpflags
& __GFP_WAIT
)
1695 new = new_slab(s
, gfpflags
, node
);
1697 if (gfpflags
& __GFP_WAIT
)
1698 local_irq_disable();
1701 c
= __this_cpu_ptr(s
->cpu_slab
);
1702 stat(s
, ALLOC_SLAB
);
1706 __SetPageSlubFrozen(new);
1710 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1711 slab_out_of_memory(s
, gfpflags
, node
);
1714 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1718 c
->page
->freelist
= get_freepointer(s
, object
);
1724 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1725 * have the fastpath folded into their functions. So no function call
1726 * overhead for requests that can be satisfied on the fastpath.
1728 * The fastpath works by first checking if the lockless freelist can be used.
1729 * If not then __slab_alloc is called for slow processing.
1731 * Otherwise we can simply pick the next object from the lockless free list.
1733 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1734 gfp_t gfpflags
, int node
, unsigned long addr
)
1737 struct kmem_cache_cpu
*c
;
1738 unsigned long flags
;
1740 if (slab_pre_alloc_hook(s
, gfpflags
))
1743 local_irq_save(flags
);
1744 c
= __this_cpu_ptr(s
->cpu_slab
);
1745 object
= c
->freelist
;
1746 if (unlikely(!object
|| !node_match(c
, node
)))
1748 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1751 c
->freelist
= get_freepointer(s
, object
);
1752 stat(s
, ALLOC_FASTPATH
);
1754 local_irq_restore(flags
);
1756 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1757 memset(object
, 0, s
->objsize
);
1759 slab_post_alloc_hook(s
, gfpflags
, object
);
1764 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1766 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1768 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1772 EXPORT_SYMBOL(kmem_cache_alloc
);
1774 #ifdef CONFIG_TRACING
1775 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1777 return slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1779 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1783 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1785 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1787 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1788 s
->objsize
, s
->size
, gfpflags
, node
);
1792 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1795 #ifdef CONFIG_TRACING
1796 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1800 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1802 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1806 * Slow patch handling. This may still be called frequently since objects
1807 * have a longer lifetime than the cpu slabs in most processing loads.
1809 * So we still attempt to reduce cache line usage. Just take the slab
1810 * lock and free the item. If there is no additional partial page
1811 * handling required then we can return immediately.
1813 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1814 void *x
, unsigned long addr
)
1817 void **object
= (void *)x
;
1819 stat(s
, FREE_SLOWPATH
);
1822 if (kmem_cache_debug(s
))
1826 prior
= page
->freelist
;
1827 set_freepointer(s
, object
, prior
);
1828 page
->freelist
= object
;
1831 if (unlikely(PageSlubFrozen(page
))) {
1832 stat(s
, FREE_FROZEN
);
1836 if (unlikely(!page
->inuse
))
1840 * Objects left in the slab. If it was not on the partial list before
1843 if (unlikely(!prior
)) {
1844 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1845 stat(s
, FREE_ADD_PARTIAL
);
1855 * Slab still on the partial list.
1857 remove_partial(s
, page
);
1858 stat(s
, FREE_REMOVE_PARTIAL
);
1862 discard_slab(s
, page
);
1866 if (!free_debug_processing(s
, page
, x
, addr
))
1872 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1873 * can perform fastpath freeing without additional function calls.
1875 * The fastpath is only possible if we are freeing to the current cpu slab
1876 * of this processor. This typically the case if we have just allocated
1879 * If fastpath is not possible then fall back to __slab_free where we deal
1880 * with all sorts of special processing.
1882 static __always_inline
void slab_free(struct kmem_cache
*s
,
1883 struct page
*page
, void *x
, unsigned long addr
)
1885 void **object
= (void *)x
;
1886 struct kmem_cache_cpu
*c
;
1887 unsigned long flags
;
1889 slab_free_hook(s
, x
);
1891 local_irq_save(flags
);
1892 c
= __this_cpu_ptr(s
->cpu_slab
);
1894 slab_free_hook_irq(s
, x
);
1896 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1897 set_freepointer(s
, object
, c
->freelist
);
1898 c
->freelist
= object
;
1899 stat(s
, FREE_FASTPATH
);
1901 __slab_free(s
, page
, x
, addr
);
1903 local_irq_restore(flags
);
1906 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1910 page
= virt_to_head_page(x
);
1912 slab_free(s
, page
, x
, _RET_IP_
);
1914 trace_kmem_cache_free(_RET_IP_
, x
);
1916 EXPORT_SYMBOL(kmem_cache_free
);
1918 /* Figure out on which slab page the object resides */
1919 static struct page
*get_object_page(const void *x
)
1921 struct page
*page
= virt_to_head_page(x
);
1923 if (!PageSlab(page
))
1930 * Object placement in a slab is made very easy because we always start at
1931 * offset 0. If we tune the size of the object to the alignment then we can
1932 * get the required alignment by putting one properly sized object after
1935 * Notice that the allocation order determines the sizes of the per cpu
1936 * caches. Each processor has always one slab available for allocations.
1937 * Increasing the allocation order reduces the number of times that slabs
1938 * must be moved on and off the partial lists and is therefore a factor in
1943 * Mininum / Maximum order of slab pages. This influences locking overhead
1944 * and slab fragmentation. A higher order reduces the number of partial slabs
1945 * and increases the number of allocations possible without having to
1946 * take the list_lock.
1948 static int slub_min_order
;
1949 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1950 static int slub_min_objects
;
1953 * Merge control. If this is set then no merging of slab caches will occur.
1954 * (Could be removed. This was introduced to pacify the merge skeptics.)
1956 static int slub_nomerge
;
1959 * Calculate the order of allocation given an slab object size.
1961 * The order of allocation has significant impact on performance and other
1962 * system components. Generally order 0 allocations should be preferred since
1963 * order 0 does not cause fragmentation in the page allocator. Larger objects
1964 * be problematic to put into order 0 slabs because there may be too much
1965 * unused space left. We go to a higher order if more than 1/16th of the slab
1968 * In order to reach satisfactory performance we must ensure that a minimum
1969 * number of objects is in one slab. Otherwise we may generate too much
1970 * activity on the partial lists which requires taking the list_lock. This is
1971 * less a concern for large slabs though which are rarely used.
1973 * slub_max_order specifies the order where we begin to stop considering the
1974 * number of objects in a slab as critical. If we reach slub_max_order then
1975 * we try to keep the page order as low as possible. So we accept more waste
1976 * of space in favor of a small page order.
1978 * Higher order allocations also allow the placement of more objects in a
1979 * slab and thereby reduce object handling overhead. If the user has
1980 * requested a higher mininum order then we start with that one instead of
1981 * the smallest order which will fit the object.
1983 static inline int slab_order(int size
, int min_objects
,
1984 int max_order
, int fract_leftover
)
1988 int min_order
= slub_min_order
;
1990 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1991 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1993 for (order
= max(min_order
,
1994 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1995 order
<= max_order
; order
++) {
1997 unsigned long slab_size
= PAGE_SIZE
<< order
;
1999 if (slab_size
< min_objects
* size
)
2002 rem
= slab_size
% size
;
2004 if (rem
<= slab_size
/ fract_leftover
)
2012 static inline int calculate_order(int size
)
2020 * Attempt to find best configuration for a slab. This
2021 * works by first attempting to generate a layout with
2022 * the best configuration and backing off gradually.
2024 * First we reduce the acceptable waste in a slab. Then
2025 * we reduce the minimum objects required in a slab.
2027 min_objects
= slub_min_objects
;
2029 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2030 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
2031 min_objects
= min(min_objects
, max_objects
);
2033 while (min_objects
> 1) {
2035 while (fraction
>= 4) {
2036 order
= slab_order(size
, min_objects
,
2037 slub_max_order
, fraction
);
2038 if (order
<= slub_max_order
)
2046 * We were unable to place multiple objects in a slab. Now
2047 * lets see if we can place a single object there.
2049 order
= slab_order(size
, 1, slub_max_order
, 1);
2050 if (order
<= slub_max_order
)
2054 * Doh this slab cannot be placed using slub_max_order.
2056 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2057 if (order
< MAX_ORDER
)
2063 * Figure out what the alignment of the objects will be.
2065 static unsigned long calculate_alignment(unsigned long flags
,
2066 unsigned long align
, unsigned long size
)
2069 * If the user wants hardware cache aligned objects then follow that
2070 * suggestion if the object is sufficiently large.
2072 * The hardware cache alignment cannot override the specified
2073 * alignment though. If that is greater then use it.
2075 if (flags
& SLAB_HWCACHE_ALIGN
) {
2076 unsigned long ralign
= cache_line_size();
2077 while (size
<= ralign
/ 2)
2079 align
= max(align
, ralign
);
2082 if (align
< ARCH_SLAB_MINALIGN
)
2083 align
= ARCH_SLAB_MINALIGN
;
2085 return ALIGN(align
, sizeof(void *));
2089 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2092 spin_lock_init(&n
->list_lock
);
2093 INIT_LIST_HEAD(&n
->partial
);
2094 #ifdef CONFIG_SLUB_DEBUG
2095 atomic_long_set(&n
->nr_slabs
, 0);
2096 atomic_long_set(&n
->total_objects
, 0);
2097 INIT_LIST_HEAD(&n
->full
);
2101 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2103 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2104 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2106 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2108 return s
->cpu_slab
!= NULL
;
2111 static struct kmem_cache
*kmem_cache_node
;
2114 * No kmalloc_node yet so do it by hand. We know that this is the first
2115 * slab on the node for this slabcache. There are no concurrent accesses
2118 * Note that this function only works on the kmalloc_node_cache
2119 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2120 * memory on a fresh node that has no slab structures yet.
2122 static void early_kmem_cache_node_alloc(int node
)
2125 struct kmem_cache_node
*n
;
2126 unsigned long flags
;
2128 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2130 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2133 if (page_to_nid(page
) != node
) {
2134 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2136 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2137 "in order to be able to continue\n");
2142 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2144 kmem_cache_node
->node
[node
] = n
;
2145 #ifdef CONFIG_SLUB_DEBUG
2146 init_object(kmem_cache_node
, n
, 1);
2147 init_tracking(kmem_cache_node
, n
);
2149 init_kmem_cache_node(n
, kmem_cache_node
);
2150 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2153 * lockdep requires consistent irq usage for each lock
2154 * so even though there cannot be a race this early in
2155 * the boot sequence, we still disable irqs.
2157 local_irq_save(flags
);
2158 add_partial(n
, page
, 0);
2159 local_irq_restore(flags
);
2162 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2166 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2167 struct kmem_cache_node
*n
= s
->node
[node
];
2170 kmem_cache_free(kmem_cache_node
, n
);
2172 s
->node
[node
] = NULL
;
2176 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2180 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2181 struct kmem_cache_node
*n
;
2183 if (slab_state
== DOWN
) {
2184 early_kmem_cache_node_alloc(node
);
2187 n
= kmem_cache_alloc_node(kmem_cache_node
,
2191 free_kmem_cache_nodes(s
);
2196 init_kmem_cache_node(n
, s
);
2201 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2203 if (min
< MIN_PARTIAL
)
2205 else if (min
> MAX_PARTIAL
)
2207 s
->min_partial
= min
;
2211 * calculate_sizes() determines the order and the distribution of data within
2214 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2216 unsigned long flags
= s
->flags
;
2217 unsigned long size
= s
->objsize
;
2218 unsigned long align
= s
->align
;
2222 * Round up object size to the next word boundary. We can only
2223 * place the free pointer at word boundaries and this determines
2224 * the possible location of the free pointer.
2226 size
= ALIGN(size
, sizeof(void *));
2228 #ifdef CONFIG_SLUB_DEBUG
2230 * Determine if we can poison the object itself. If the user of
2231 * the slab may touch the object after free or before allocation
2232 * then we should never poison the object itself.
2234 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2236 s
->flags
|= __OBJECT_POISON
;
2238 s
->flags
&= ~__OBJECT_POISON
;
2242 * If we are Redzoning then check if there is some space between the
2243 * end of the object and the free pointer. If not then add an
2244 * additional word to have some bytes to store Redzone information.
2246 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2247 size
+= sizeof(void *);
2251 * With that we have determined the number of bytes in actual use
2252 * by the object. This is the potential offset to the free pointer.
2256 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2259 * Relocate free pointer after the object if it is not
2260 * permitted to overwrite the first word of the object on
2263 * This is the case if we do RCU, have a constructor or
2264 * destructor or are poisoning the objects.
2267 size
+= sizeof(void *);
2270 #ifdef CONFIG_SLUB_DEBUG
2271 if (flags
& SLAB_STORE_USER
)
2273 * Need to store information about allocs and frees after
2276 size
+= 2 * sizeof(struct track
);
2278 if (flags
& SLAB_RED_ZONE
)
2280 * Add some empty padding so that we can catch
2281 * overwrites from earlier objects rather than let
2282 * tracking information or the free pointer be
2283 * corrupted if a user writes before the start
2286 size
+= sizeof(void *);
2290 * Determine the alignment based on various parameters that the
2291 * user specified and the dynamic determination of cache line size
2294 align
= calculate_alignment(flags
, align
, s
->objsize
);
2298 * SLUB stores one object immediately after another beginning from
2299 * offset 0. In order to align the objects we have to simply size
2300 * each object to conform to the alignment.
2302 size
= ALIGN(size
, align
);
2304 if (forced_order
>= 0)
2305 order
= forced_order
;
2307 order
= calculate_order(size
);
2314 s
->allocflags
|= __GFP_COMP
;
2316 if (s
->flags
& SLAB_CACHE_DMA
)
2317 s
->allocflags
|= SLUB_DMA
;
2319 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2320 s
->allocflags
|= __GFP_RECLAIMABLE
;
2323 * Determine the number of objects per slab
2325 s
->oo
= oo_make(order
, size
);
2326 s
->min
= oo_make(get_order(size
), size
);
2327 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2330 return !!oo_objects(s
->oo
);
2334 static int kmem_cache_open(struct kmem_cache
*s
,
2335 const char *name
, size_t size
,
2336 size_t align
, unsigned long flags
,
2337 void (*ctor
)(void *))
2339 memset(s
, 0, kmem_size
);
2344 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2346 if (!calculate_sizes(s
, -1))
2348 if (disable_higher_order_debug
) {
2350 * Disable debugging flags that store metadata if the min slab
2353 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2354 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2356 if (!calculate_sizes(s
, -1))
2362 * The larger the object size is, the more pages we want on the partial
2363 * list to avoid pounding the page allocator excessively.
2365 set_min_partial(s
, ilog2(s
->size
));
2368 s
->remote_node_defrag_ratio
= 1000;
2370 if (!init_kmem_cache_nodes(s
))
2373 if (alloc_kmem_cache_cpus(s
))
2376 free_kmem_cache_nodes(s
);
2378 if (flags
& SLAB_PANIC
)
2379 panic("Cannot create slab %s size=%lu realsize=%u "
2380 "order=%u offset=%u flags=%lx\n",
2381 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2387 * Check if a given pointer is valid
2389 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2393 if (!kern_ptr_validate(object
, s
->size
))
2396 page
= get_object_page(object
);
2398 if (!page
|| s
!= page
->slab
)
2399 /* No slab or wrong slab */
2402 if (!check_valid_pointer(s
, page
, object
))
2406 * We could also check if the object is on the slabs freelist.
2407 * But this would be too expensive and it seems that the main
2408 * purpose of kmem_ptr_valid() is to check if the object belongs
2409 * to a certain slab.
2413 EXPORT_SYMBOL(kmem_ptr_validate
);
2416 * Determine the size of a slab object
2418 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2422 EXPORT_SYMBOL(kmem_cache_size
);
2424 const char *kmem_cache_name(struct kmem_cache
*s
)
2428 EXPORT_SYMBOL(kmem_cache_name
);
2430 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2433 #ifdef CONFIG_SLUB_DEBUG
2434 void *addr
= page_address(page
);
2436 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2441 slab_err(s
, page
, "%s", text
);
2443 for_each_free_object(p
, s
, page
->freelist
)
2444 set_bit(slab_index(p
, s
, addr
), map
);
2446 for_each_object(p
, s
, addr
, page
->objects
) {
2448 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2449 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2451 print_tracking(s
, p
);
2460 * Attempt to free all partial slabs on a node.
2462 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2464 unsigned long flags
;
2465 struct page
*page
, *h
;
2467 spin_lock_irqsave(&n
->list_lock
, flags
);
2468 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2470 list_del(&page
->lru
);
2471 discard_slab(s
, page
);
2474 list_slab_objects(s
, page
,
2475 "Objects remaining on kmem_cache_close()");
2478 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2482 * Release all resources used by a slab cache.
2484 static inline int kmem_cache_close(struct kmem_cache
*s
)
2489 free_percpu(s
->cpu_slab
);
2490 /* Attempt to free all objects */
2491 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2492 struct kmem_cache_node
*n
= get_node(s
, node
);
2495 if (n
->nr_partial
|| slabs_node(s
, node
))
2498 free_kmem_cache_nodes(s
);
2503 * Close a cache and release the kmem_cache structure
2504 * (must be used for caches created using kmem_cache_create)
2506 void kmem_cache_destroy(struct kmem_cache
*s
)
2508 down_write(&slub_lock
);
2512 if (kmem_cache_close(s
)) {
2513 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2514 "still has objects.\n", s
->name
, __func__
);
2517 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2519 sysfs_slab_remove(s
);
2521 up_write(&slub_lock
);
2523 EXPORT_SYMBOL(kmem_cache_destroy
);
2525 /********************************************************************
2527 *******************************************************************/
2529 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
2530 EXPORT_SYMBOL(kmalloc_caches
);
2532 static struct kmem_cache
*kmem_cache
;
2534 #ifdef CONFIG_ZONE_DMA
2535 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
2538 static int __init
setup_slub_min_order(char *str
)
2540 get_option(&str
, &slub_min_order
);
2545 __setup("slub_min_order=", setup_slub_min_order
);
2547 static int __init
setup_slub_max_order(char *str
)
2549 get_option(&str
, &slub_max_order
);
2550 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2555 __setup("slub_max_order=", setup_slub_max_order
);
2557 static int __init
setup_slub_min_objects(char *str
)
2559 get_option(&str
, &slub_min_objects
);
2564 __setup("slub_min_objects=", setup_slub_min_objects
);
2566 static int __init
setup_slub_nomerge(char *str
)
2572 __setup("slub_nomerge", setup_slub_nomerge
);
2574 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
2575 int size
, unsigned int flags
)
2577 struct kmem_cache
*s
;
2579 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
2582 * This function is called with IRQs disabled during early-boot on
2583 * single CPU so there's no need to take slub_lock here.
2585 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2589 list_add(&s
->list
, &slab_caches
);
2593 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2598 * Conversion table for small slabs sizes / 8 to the index in the
2599 * kmalloc array. This is necessary for slabs < 192 since we have non power
2600 * of two cache sizes there. The size of larger slabs can be determined using
2603 static s8 size_index
[24] = {
2630 static inline int size_index_elem(size_t bytes
)
2632 return (bytes
- 1) / 8;
2635 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2641 return ZERO_SIZE_PTR
;
2643 index
= size_index
[size_index_elem(size
)];
2645 index
= fls(size
- 1);
2647 #ifdef CONFIG_ZONE_DMA
2648 if (unlikely((flags
& SLUB_DMA
)))
2649 return kmalloc_dma_caches
[index
];
2652 return kmalloc_caches
[index
];
2655 void *__kmalloc(size_t size
, gfp_t flags
)
2657 struct kmem_cache
*s
;
2660 if (unlikely(size
> SLUB_MAX_SIZE
))
2661 return kmalloc_large(size
, flags
);
2663 s
= get_slab(size
, flags
);
2665 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2668 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
2670 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2674 EXPORT_SYMBOL(__kmalloc
);
2676 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2681 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2682 page
= alloc_pages_node(node
, flags
, get_order(size
));
2684 ptr
= page_address(page
);
2686 kmemleak_alloc(ptr
, size
, 1, flags
);
2691 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2693 struct kmem_cache
*s
;
2696 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2697 ret
= kmalloc_large_node(size
, flags
, node
);
2699 trace_kmalloc_node(_RET_IP_
, ret
,
2700 size
, PAGE_SIZE
<< get_order(size
),
2706 s
= get_slab(size
, flags
);
2708 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2711 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2713 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2717 EXPORT_SYMBOL(__kmalloc_node
);
2720 size_t ksize(const void *object
)
2723 struct kmem_cache
*s
;
2725 if (unlikely(object
== ZERO_SIZE_PTR
))
2728 page
= virt_to_head_page(object
);
2730 if (unlikely(!PageSlab(page
))) {
2731 WARN_ON(!PageCompound(page
));
2732 return PAGE_SIZE
<< compound_order(page
);
2736 #ifdef CONFIG_SLUB_DEBUG
2738 * Debugging requires use of the padding between object
2739 * and whatever may come after it.
2741 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2746 * If we have the need to store the freelist pointer
2747 * back there or track user information then we can
2748 * only use the space before that information.
2750 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2753 * Else we can use all the padding etc for the allocation
2757 EXPORT_SYMBOL(ksize
);
2759 void kfree(const void *x
)
2762 void *object
= (void *)x
;
2764 trace_kfree(_RET_IP_
, x
);
2766 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2769 page
= virt_to_head_page(x
);
2770 if (unlikely(!PageSlab(page
))) {
2771 BUG_ON(!PageCompound(page
));
2776 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2778 EXPORT_SYMBOL(kfree
);
2781 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2782 * the remaining slabs by the number of items in use. The slabs with the
2783 * most items in use come first. New allocations will then fill those up
2784 * and thus they can be removed from the partial lists.
2786 * The slabs with the least items are placed last. This results in them
2787 * being allocated from last increasing the chance that the last objects
2788 * are freed in them.
2790 int kmem_cache_shrink(struct kmem_cache
*s
)
2794 struct kmem_cache_node
*n
;
2797 int objects
= oo_objects(s
->max
);
2798 struct list_head
*slabs_by_inuse
=
2799 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2800 unsigned long flags
;
2802 if (!slabs_by_inuse
)
2806 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2807 n
= get_node(s
, node
);
2812 for (i
= 0; i
< objects
; i
++)
2813 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2815 spin_lock_irqsave(&n
->list_lock
, flags
);
2818 * Build lists indexed by the items in use in each slab.
2820 * Note that concurrent frees may occur while we hold the
2821 * list_lock. page->inuse here is the upper limit.
2823 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2824 if (!page
->inuse
&& slab_trylock(page
)) {
2826 * Must hold slab lock here because slab_free
2827 * may have freed the last object and be
2828 * waiting to release the slab.
2830 list_del(&page
->lru
);
2833 discard_slab(s
, page
);
2835 list_move(&page
->lru
,
2836 slabs_by_inuse
+ page
->inuse
);
2841 * Rebuild the partial list with the slabs filled up most
2842 * first and the least used slabs at the end.
2844 for (i
= objects
- 1; i
>= 0; i
--)
2845 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2847 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2850 kfree(slabs_by_inuse
);
2853 EXPORT_SYMBOL(kmem_cache_shrink
);
2855 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2856 static int slab_mem_going_offline_callback(void *arg
)
2858 struct kmem_cache
*s
;
2860 down_read(&slub_lock
);
2861 list_for_each_entry(s
, &slab_caches
, list
)
2862 kmem_cache_shrink(s
);
2863 up_read(&slub_lock
);
2868 static void slab_mem_offline_callback(void *arg
)
2870 struct kmem_cache_node
*n
;
2871 struct kmem_cache
*s
;
2872 struct memory_notify
*marg
= arg
;
2875 offline_node
= marg
->status_change_nid
;
2878 * If the node still has available memory. we need kmem_cache_node
2881 if (offline_node
< 0)
2884 down_read(&slub_lock
);
2885 list_for_each_entry(s
, &slab_caches
, list
) {
2886 n
= get_node(s
, offline_node
);
2889 * if n->nr_slabs > 0, slabs still exist on the node
2890 * that is going down. We were unable to free them,
2891 * and offline_pages() function shouldn't call this
2892 * callback. So, we must fail.
2894 BUG_ON(slabs_node(s
, offline_node
));
2896 s
->node
[offline_node
] = NULL
;
2897 kmem_cache_free(kmem_cache_node
, n
);
2900 up_read(&slub_lock
);
2903 static int slab_mem_going_online_callback(void *arg
)
2905 struct kmem_cache_node
*n
;
2906 struct kmem_cache
*s
;
2907 struct memory_notify
*marg
= arg
;
2908 int nid
= marg
->status_change_nid
;
2912 * If the node's memory is already available, then kmem_cache_node is
2913 * already created. Nothing to do.
2919 * We are bringing a node online. No memory is available yet. We must
2920 * allocate a kmem_cache_node structure in order to bring the node
2923 down_read(&slub_lock
);
2924 list_for_each_entry(s
, &slab_caches
, list
) {
2926 * XXX: kmem_cache_alloc_node will fallback to other nodes
2927 * since memory is not yet available from the node that
2930 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
2935 init_kmem_cache_node(n
, s
);
2939 up_read(&slub_lock
);
2943 static int slab_memory_callback(struct notifier_block
*self
,
2944 unsigned long action
, void *arg
)
2949 case MEM_GOING_ONLINE
:
2950 ret
= slab_mem_going_online_callback(arg
);
2952 case MEM_GOING_OFFLINE
:
2953 ret
= slab_mem_going_offline_callback(arg
);
2956 case MEM_CANCEL_ONLINE
:
2957 slab_mem_offline_callback(arg
);
2960 case MEM_CANCEL_OFFLINE
:
2964 ret
= notifier_from_errno(ret
);
2970 #endif /* CONFIG_MEMORY_HOTPLUG */
2972 /********************************************************************
2973 * Basic setup of slabs
2974 *******************************************************************/
2977 * Used for early kmem_cache structures that were allocated using
2978 * the page allocator
2981 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
2985 list_add(&s
->list
, &slab_caches
);
2988 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2989 struct kmem_cache_node
*n
= get_node(s
, node
);
2993 list_for_each_entry(p
, &n
->partial
, lru
)
2996 #ifdef CONFIG_SLAB_DEBUG
2997 list_for_each_entry(p
, &n
->full
, lru
)
3004 void __init
kmem_cache_init(void)
3008 struct kmem_cache
*temp_kmem_cache
;
3010 struct kmem_cache
*temp_kmem_cache_node
;
3011 unsigned long kmalloc_size
;
3013 kmem_size
= offsetof(struct kmem_cache
, node
) +
3014 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3016 /* Allocate two kmem_caches from the page allocator */
3017 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3018 order
= get_order(2 * kmalloc_size
);
3019 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3022 * Must first have the slab cache available for the allocations of the
3023 * struct kmem_cache_node's. There is special bootstrap code in
3024 * kmem_cache_open for slab_state == DOWN.
3026 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3028 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3029 sizeof(struct kmem_cache_node
),
3030 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3032 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3034 /* Able to allocate the per node structures */
3035 slab_state
= PARTIAL
;
3037 temp_kmem_cache
= kmem_cache
;
3038 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3039 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3040 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3041 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3044 * Allocate kmem_cache_node properly from the kmem_cache slab.
3045 * kmem_cache_node is separately allocated so no need to
3046 * update any list pointers.
3048 temp_kmem_cache_node
= kmem_cache_node
;
3050 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3051 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3053 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3056 kmem_cache_bootstrap_fixup(kmem_cache
);
3058 /* Free temporary boot structure */
3059 free_pages((unsigned long)temp_kmem_cache
, order
);
3061 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3064 * Patch up the size_index table if we have strange large alignment
3065 * requirements for the kmalloc array. This is only the case for
3066 * MIPS it seems. The standard arches will not generate any code here.
3068 * Largest permitted alignment is 256 bytes due to the way we
3069 * handle the index determination for the smaller caches.
3071 * Make sure that nothing crazy happens if someone starts tinkering
3072 * around with ARCH_KMALLOC_MINALIGN
3074 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3075 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3077 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3078 int elem
= size_index_elem(i
);
3079 if (elem
>= ARRAY_SIZE(size_index
))
3081 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3084 if (KMALLOC_MIN_SIZE
== 64) {
3086 * The 96 byte size cache is not used if the alignment
3089 for (i
= 64 + 8; i
<= 96; i
+= 8)
3090 size_index
[size_index_elem(i
)] = 7;
3091 } else if (KMALLOC_MIN_SIZE
== 128) {
3093 * The 192 byte sized cache is not used if the alignment
3094 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3097 for (i
= 128 + 8; i
<= 192; i
+= 8)
3098 size_index
[size_index_elem(i
)] = 8;
3101 /* Caches that are not of the two-to-the-power-of size */
3102 if (KMALLOC_MIN_SIZE
<= 32) {
3103 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3107 if (KMALLOC_MIN_SIZE
<= 64) {
3108 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3112 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3113 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3119 /* Provide the correct kmalloc names now that the caches are up */
3120 if (KMALLOC_MIN_SIZE
<= 32) {
3121 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3122 BUG_ON(!kmalloc_caches
[1]->name
);
3125 if (KMALLOC_MIN_SIZE
<= 64) {
3126 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3127 BUG_ON(!kmalloc_caches
[2]->name
);
3130 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3131 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3134 kmalloc_caches
[i
]->name
= s
;
3138 register_cpu_notifier(&slab_notifier
);
3141 #ifdef CONFIG_ZONE_DMA
3142 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3143 struct kmem_cache
*s
= kmalloc_caches
[i
];
3146 char *name
= kasprintf(GFP_NOWAIT
,
3147 "dma-kmalloc-%d", s
->objsize
);
3150 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3151 s
->objsize
, SLAB_CACHE_DMA
);
3156 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3157 " CPUs=%d, Nodes=%d\n",
3158 caches
, cache_line_size(),
3159 slub_min_order
, slub_max_order
, slub_min_objects
,
3160 nr_cpu_ids
, nr_node_ids
);
3163 void __init
kmem_cache_init_late(void)
3168 * Find a mergeable slab cache
3170 static int slab_unmergeable(struct kmem_cache
*s
)
3172 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3179 * We may have set a slab to be unmergeable during bootstrap.
3181 if (s
->refcount
< 0)
3187 static struct kmem_cache
*find_mergeable(size_t size
,
3188 size_t align
, unsigned long flags
, const char *name
,
3189 void (*ctor
)(void *))
3191 struct kmem_cache
*s
;
3193 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3199 size
= ALIGN(size
, sizeof(void *));
3200 align
= calculate_alignment(flags
, align
, size
);
3201 size
= ALIGN(size
, align
);
3202 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3204 list_for_each_entry(s
, &slab_caches
, list
) {
3205 if (slab_unmergeable(s
))
3211 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3214 * Check if alignment is compatible.
3215 * Courtesy of Adrian Drzewiecki
3217 if ((s
->size
& ~(align
- 1)) != s
->size
)
3220 if (s
->size
- size
>= sizeof(void *))
3228 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3229 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3231 struct kmem_cache
*s
;
3237 down_write(&slub_lock
);
3238 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3242 * Adjust the object sizes so that we clear
3243 * the complete object on kzalloc.
3245 s
->objsize
= max(s
->objsize
, (int)size
);
3246 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3248 if (sysfs_slab_alias(s
, name
)) {
3252 up_write(&slub_lock
);
3256 n
= kstrdup(name
, GFP_KERNEL
);
3260 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3262 if (kmem_cache_open(s
, n
,
3263 size
, align
, flags
, ctor
)) {
3264 list_add(&s
->list
, &slab_caches
);
3265 if (sysfs_slab_add(s
)) {
3271 up_write(&slub_lock
);
3277 up_write(&slub_lock
);
3280 if (flags
& SLAB_PANIC
)
3281 panic("Cannot create slabcache %s\n", name
);
3286 EXPORT_SYMBOL(kmem_cache_create
);
3290 * Use the cpu notifier to insure that the cpu slabs are flushed when
3293 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3294 unsigned long action
, void *hcpu
)
3296 long cpu
= (long)hcpu
;
3297 struct kmem_cache
*s
;
3298 unsigned long flags
;
3301 case CPU_UP_CANCELED
:
3302 case CPU_UP_CANCELED_FROZEN
:
3304 case CPU_DEAD_FROZEN
:
3305 down_read(&slub_lock
);
3306 list_for_each_entry(s
, &slab_caches
, list
) {
3307 local_irq_save(flags
);
3308 __flush_cpu_slab(s
, cpu
);
3309 local_irq_restore(flags
);
3311 up_read(&slub_lock
);
3319 static struct notifier_block __cpuinitdata slab_notifier
= {
3320 .notifier_call
= slab_cpuup_callback
3325 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3327 struct kmem_cache
*s
;
3330 if (unlikely(size
> SLUB_MAX_SIZE
))
3331 return kmalloc_large(size
, gfpflags
);
3333 s
= get_slab(size
, gfpflags
);
3335 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3338 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
3340 /* Honor the call site pointer we recieved. */
3341 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3346 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3347 int node
, unsigned long caller
)
3349 struct kmem_cache
*s
;
3352 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3353 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3355 trace_kmalloc_node(caller
, ret
,
3356 size
, PAGE_SIZE
<< get_order(size
),
3362 s
= get_slab(size
, gfpflags
);
3364 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3367 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3369 /* Honor the call site pointer we recieved. */
3370 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3375 #ifdef CONFIG_SLUB_DEBUG
3376 static int count_inuse(struct page
*page
)
3381 static int count_total(struct page
*page
)
3383 return page
->objects
;
3386 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3390 void *addr
= page_address(page
);
3392 if (!check_slab(s
, page
) ||
3393 !on_freelist(s
, page
, NULL
))
3396 /* Now we know that a valid freelist exists */
3397 bitmap_zero(map
, page
->objects
);
3399 for_each_free_object(p
, s
, page
->freelist
) {
3400 set_bit(slab_index(p
, s
, addr
), map
);
3401 if (!check_object(s
, page
, p
, 0))
3405 for_each_object(p
, s
, addr
, page
->objects
)
3406 if (!test_bit(slab_index(p
, s
, addr
), map
))
3407 if (!check_object(s
, page
, p
, 1))
3412 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3415 if (slab_trylock(page
)) {
3416 validate_slab(s
, page
, map
);
3419 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3423 static int validate_slab_node(struct kmem_cache
*s
,
3424 struct kmem_cache_node
*n
, unsigned long *map
)
3426 unsigned long count
= 0;
3428 unsigned long flags
;
3430 spin_lock_irqsave(&n
->list_lock
, flags
);
3432 list_for_each_entry(page
, &n
->partial
, lru
) {
3433 validate_slab_slab(s
, page
, map
);
3436 if (count
!= n
->nr_partial
)
3437 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3438 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3440 if (!(s
->flags
& SLAB_STORE_USER
))
3443 list_for_each_entry(page
, &n
->full
, lru
) {
3444 validate_slab_slab(s
, page
, map
);
3447 if (count
!= atomic_long_read(&n
->nr_slabs
))
3448 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3449 "counter=%ld\n", s
->name
, count
,
3450 atomic_long_read(&n
->nr_slabs
));
3453 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3457 static long validate_slab_cache(struct kmem_cache
*s
)
3460 unsigned long count
= 0;
3461 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3462 sizeof(unsigned long), GFP_KERNEL
);
3468 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3469 struct kmem_cache_node
*n
= get_node(s
, node
);
3471 count
+= validate_slab_node(s
, n
, map
);
3477 #ifdef SLUB_RESILIENCY_TEST
3478 static void resiliency_test(void)
3482 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
3484 printk(KERN_ERR
"SLUB resiliency testing\n");
3485 printk(KERN_ERR
"-----------------------\n");
3486 printk(KERN_ERR
"A. Corruption after allocation\n");
3488 p
= kzalloc(16, GFP_KERNEL
);
3490 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3491 " 0x12->0x%p\n\n", p
+ 16);
3493 validate_slab_cache(kmalloc_caches
[4]);
3495 /* Hmmm... The next two are dangerous */
3496 p
= kzalloc(32, GFP_KERNEL
);
3497 p
[32 + sizeof(void *)] = 0x34;
3498 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3499 " 0x34 -> -0x%p\n", p
);
3501 "If allocated object is overwritten then not detectable\n\n");
3503 validate_slab_cache(kmalloc_caches
[5]);
3504 p
= kzalloc(64, GFP_KERNEL
);
3505 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3507 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3510 "If allocated object is overwritten then not detectable\n\n");
3511 validate_slab_cache(kmalloc_caches
[6]);
3513 printk(KERN_ERR
"\nB. Corruption after free\n");
3514 p
= kzalloc(128, GFP_KERNEL
);
3517 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3518 validate_slab_cache(kmalloc_caches
[7]);
3520 p
= kzalloc(256, GFP_KERNEL
);
3523 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3525 validate_slab_cache(kmalloc_caches
[8]);
3527 p
= kzalloc(512, GFP_KERNEL
);
3530 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3531 validate_slab_cache(kmalloc_caches
[9]);
3534 static void resiliency_test(void) {};
3538 * Generate lists of code addresses where slabcache objects are allocated
3543 unsigned long count
;
3550 DECLARE_BITMAP(cpus
, NR_CPUS
);
3556 unsigned long count
;
3557 struct location
*loc
;
3560 static void free_loc_track(struct loc_track
*t
)
3563 free_pages((unsigned long)t
->loc
,
3564 get_order(sizeof(struct location
) * t
->max
));
3567 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3572 order
= get_order(sizeof(struct location
) * max
);
3574 l
= (void *)__get_free_pages(flags
, order
);
3579 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3587 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3588 const struct track
*track
)
3590 long start
, end
, pos
;
3592 unsigned long caddr
;
3593 unsigned long age
= jiffies
- track
->when
;
3599 pos
= start
+ (end
- start
+ 1) / 2;
3602 * There is nothing at "end". If we end up there
3603 * we need to add something to before end.
3608 caddr
= t
->loc
[pos
].addr
;
3609 if (track
->addr
== caddr
) {
3615 if (age
< l
->min_time
)
3617 if (age
> l
->max_time
)
3620 if (track
->pid
< l
->min_pid
)
3621 l
->min_pid
= track
->pid
;
3622 if (track
->pid
> l
->max_pid
)
3623 l
->max_pid
= track
->pid
;
3625 cpumask_set_cpu(track
->cpu
,
3626 to_cpumask(l
->cpus
));
3628 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3632 if (track
->addr
< caddr
)
3639 * Not found. Insert new tracking element.
3641 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3647 (t
->count
- pos
) * sizeof(struct location
));
3650 l
->addr
= track
->addr
;
3654 l
->min_pid
= track
->pid
;
3655 l
->max_pid
= track
->pid
;
3656 cpumask_clear(to_cpumask(l
->cpus
));
3657 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3658 nodes_clear(l
->nodes
);
3659 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3663 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3664 struct page
*page
, enum track_item alloc
,
3667 void *addr
= page_address(page
);
3670 bitmap_zero(map
, page
->objects
);
3671 for_each_free_object(p
, s
, page
->freelist
)
3672 set_bit(slab_index(p
, s
, addr
), map
);
3674 for_each_object(p
, s
, addr
, page
->objects
)
3675 if (!test_bit(slab_index(p
, s
, addr
), map
))
3676 add_location(t
, s
, get_track(s
, p
, alloc
));
3679 static int list_locations(struct kmem_cache
*s
, char *buf
,
3680 enum track_item alloc
)
3684 struct loc_track t
= { 0, 0, NULL
};
3686 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3687 sizeof(unsigned long), GFP_KERNEL
);
3689 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3692 return sprintf(buf
, "Out of memory\n");
3694 /* Push back cpu slabs */
3697 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3698 struct kmem_cache_node
*n
= get_node(s
, node
);
3699 unsigned long flags
;
3702 if (!atomic_long_read(&n
->nr_slabs
))
3705 spin_lock_irqsave(&n
->list_lock
, flags
);
3706 list_for_each_entry(page
, &n
->partial
, lru
)
3707 process_slab(&t
, s
, page
, alloc
, map
);
3708 list_for_each_entry(page
, &n
->full
, lru
)
3709 process_slab(&t
, s
, page
, alloc
, map
);
3710 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3713 for (i
= 0; i
< t
.count
; i
++) {
3714 struct location
*l
= &t
.loc
[i
];
3716 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3718 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3721 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3723 len
+= sprintf(buf
+ len
, "<not-available>");
3725 if (l
->sum_time
!= l
->min_time
) {
3726 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3728 (long)div_u64(l
->sum_time
, l
->count
),
3731 len
+= sprintf(buf
+ len
, " age=%ld",
3734 if (l
->min_pid
!= l
->max_pid
)
3735 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3736 l
->min_pid
, l
->max_pid
);
3738 len
+= sprintf(buf
+ len
, " pid=%ld",
3741 if (num_online_cpus() > 1 &&
3742 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3743 len
< PAGE_SIZE
- 60) {
3744 len
+= sprintf(buf
+ len
, " cpus=");
3745 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3746 to_cpumask(l
->cpus
));
3749 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3750 len
< PAGE_SIZE
- 60) {
3751 len
+= sprintf(buf
+ len
, " nodes=");
3752 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3756 len
+= sprintf(buf
+ len
, "\n");
3762 len
+= sprintf(buf
, "No data\n");
3766 enum slab_stat_type
{
3767 SL_ALL
, /* All slabs */
3768 SL_PARTIAL
, /* Only partially allocated slabs */
3769 SL_CPU
, /* Only slabs used for cpu caches */
3770 SL_OBJECTS
, /* Determine allocated objects not slabs */
3771 SL_TOTAL
/* Determine object capacity not slabs */
3774 #define SO_ALL (1 << SL_ALL)
3775 #define SO_PARTIAL (1 << SL_PARTIAL)
3776 #define SO_CPU (1 << SL_CPU)
3777 #define SO_OBJECTS (1 << SL_OBJECTS)
3778 #define SO_TOTAL (1 << SL_TOTAL)
3780 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3781 char *buf
, unsigned long flags
)
3783 unsigned long total
= 0;
3786 unsigned long *nodes
;
3787 unsigned long *per_cpu
;
3789 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3792 per_cpu
= nodes
+ nr_node_ids
;
3794 if (flags
& SO_CPU
) {
3797 for_each_possible_cpu(cpu
) {
3798 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3800 if (!c
|| c
->node
< 0)
3804 if (flags
& SO_TOTAL
)
3805 x
= c
->page
->objects
;
3806 else if (flags
& SO_OBJECTS
)
3812 nodes
[c
->node
] += x
;
3818 if (flags
& SO_ALL
) {
3819 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3820 struct kmem_cache_node
*n
= get_node(s
, node
);
3822 if (flags
& SO_TOTAL
)
3823 x
= atomic_long_read(&n
->total_objects
);
3824 else if (flags
& SO_OBJECTS
)
3825 x
= atomic_long_read(&n
->total_objects
) -
3826 count_partial(n
, count_free
);
3829 x
= atomic_long_read(&n
->nr_slabs
);
3834 } else if (flags
& SO_PARTIAL
) {
3835 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3836 struct kmem_cache_node
*n
= get_node(s
, node
);
3838 if (flags
& SO_TOTAL
)
3839 x
= count_partial(n
, count_total
);
3840 else if (flags
& SO_OBJECTS
)
3841 x
= count_partial(n
, count_inuse
);
3848 x
= sprintf(buf
, "%lu", total
);
3850 for_each_node_state(node
, N_NORMAL_MEMORY
)
3852 x
+= sprintf(buf
+ x
, " N%d=%lu",
3856 return x
+ sprintf(buf
+ x
, "\n");
3859 static int any_slab_objects(struct kmem_cache
*s
)
3863 for_each_online_node(node
) {
3864 struct kmem_cache_node
*n
= get_node(s
, node
);
3869 if (atomic_long_read(&n
->total_objects
))
3875 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3876 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3878 struct slab_attribute
{
3879 struct attribute attr
;
3880 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3881 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3884 #define SLAB_ATTR_RO(_name) \
3885 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3887 #define SLAB_ATTR(_name) \
3888 static struct slab_attribute _name##_attr = \
3889 __ATTR(_name, 0644, _name##_show, _name##_store)
3891 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3893 return sprintf(buf
, "%d\n", s
->size
);
3895 SLAB_ATTR_RO(slab_size
);
3897 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3899 return sprintf(buf
, "%d\n", s
->align
);
3901 SLAB_ATTR_RO(align
);
3903 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3905 return sprintf(buf
, "%d\n", s
->objsize
);
3907 SLAB_ATTR_RO(object_size
);
3909 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3911 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3913 SLAB_ATTR_RO(objs_per_slab
);
3915 static ssize_t
order_store(struct kmem_cache
*s
,
3916 const char *buf
, size_t length
)
3918 unsigned long order
;
3921 err
= strict_strtoul(buf
, 10, &order
);
3925 if (order
> slub_max_order
|| order
< slub_min_order
)
3928 calculate_sizes(s
, order
);
3932 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3934 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3938 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3940 return sprintf(buf
, "%lu\n", s
->min_partial
);
3943 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3949 err
= strict_strtoul(buf
, 10, &min
);
3953 set_min_partial(s
, min
);
3956 SLAB_ATTR(min_partial
);
3958 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3961 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3963 return n
+ sprintf(buf
+ n
, "\n");
3969 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3971 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3973 SLAB_ATTR_RO(aliases
);
3975 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3977 return show_slab_objects(s
, buf
, SO_ALL
);
3979 SLAB_ATTR_RO(slabs
);
3981 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3983 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3985 SLAB_ATTR_RO(partial
);
3987 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3989 return show_slab_objects(s
, buf
, SO_CPU
);
3991 SLAB_ATTR_RO(cpu_slabs
);
3993 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3995 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3997 SLAB_ATTR_RO(objects
);
3999 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4001 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4003 SLAB_ATTR_RO(objects_partial
);
4005 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4007 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4009 SLAB_ATTR_RO(total_objects
);
4011 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4013 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4016 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4017 const char *buf
, size_t length
)
4019 s
->flags
&= ~SLAB_DEBUG_FREE
;
4021 s
->flags
|= SLAB_DEBUG_FREE
;
4024 SLAB_ATTR(sanity_checks
);
4026 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4028 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4031 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4034 s
->flags
&= ~SLAB_TRACE
;
4036 s
->flags
|= SLAB_TRACE
;
4041 #ifdef CONFIG_FAILSLAB
4042 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4044 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4047 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4050 s
->flags
&= ~SLAB_FAILSLAB
;
4052 s
->flags
|= SLAB_FAILSLAB
;
4055 SLAB_ATTR(failslab
);
4058 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4060 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4063 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4064 const char *buf
, size_t length
)
4066 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4068 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4071 SLAB_ATTR(reclaim_account
);
4073 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4075 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4077 SLAB_ATTR_RO(hwcache_align
);
4079 #ifdef CONFIG_ZONE_DMA
4080 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4082 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4084 SLAB_ATTR_RO(cache_dma
);
4087 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4089 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4091 SLAB_ATTR_RO(destroy_by_rcu
);
4093 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4095 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4098 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4099 const char *buf
, size_t length
)
4101 if (any_slab_objects(s
))
4104 s
->flags
&= ~SLAB_RED_ZONE
;
4106 s
->flags
|= SLAB_RED_ZONE
;
4107 calculate_sizes(s
, -1);
4110 SLAB_ATTR(red_zone
);
4112 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4114 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4117 static ssize_t
poison_store(struct kmem_cache
*s
,
4118 const char *buf
, size_t length
)
4120 if (any_slab_objects(s
))
4123 s
->flags
&= ~SLAB_POISON
;
4125 s
->flags
|= SLAB_POISON
;
4126 calculate_sizes(s
, -1);
4131 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4133 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4136 static ssize_t
store_user_store(struct kmem_cache
*s
,
4137 const char *buf
, size_t length
)
4139 if (any_slab_objects(s
))
4142 s
->flags
&= ~SLAB_STORE_USER
;
4144 s
->flags
|= SLAB_STORE_USER
;
4145 calculate_sizes(s
, -1);
4148 SLAB_ATTR(store_user
);
4150 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4155 static ssize_t
validate_store(struct kmem_cache
*s
,
4156 const char *buf
, size_t length
)
4160 if (buf
[0] == '1') {
4161 ret
= validate_slab_cache(s
);
4167 SLAB_ATTR(validate
);
4169 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4174 static ssize_t
shrink_store(struct kmem_cache
*s
,
4175 const char *buf
, size_t length
)
4177 if (buf
[0] == '1') {
4178 int rc
= kmem_cache_shrink(s
);
4188 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4190 if (!(s
->flags
& SLAB_STORE_USER
))
4192 return list_locations(s
, buf
, TRACK_ALLOC
);
4194 SLAB_ATTR_RO(alloc_calls
);
4196 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4198 if (!(s
->flags
& SLAB_STORE_USER
))
4200 return list_locations(s
, buf
, TRACK_FREE
);
4202 SLAB_ATTR_RO(free_calls
);
4205 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4207 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4210 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4211 const char *buf
, size_t length
)
4213 unsigned long ratio
;
4216 err
= strict_strtoul(buf
, 10, &ratio
);
4221 s
->remote_node_defrag_ratio
= ratio
* 10;
4225 SLAB_ATTR(remote_node_defrag_ratio
);
4228 #ifdef CONFIG_SLUB_STATS
4229 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4231 unsigned long sum
= 0;
4234 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4239 for_each_online_cpu(cpu
) {
4240 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4246 len
= sprintf(buf
, "%lu", sum
);
4249 for_each_online_cpu(cpu
) {
4250 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4251 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4255 return len
+ sprintf(buf
+ len
, "\n");
4258 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4262 for_each_online_cpu(cpu
)
4263 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4266 #define STAT_ATTR(si, text) \
4267 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4269 return show_stat(s, buf, si); \
4271 static ssize_t text##_store(struct kmem_cache *s, \
4272 const char *buf, size_t length) \
4274 if (buf[0] != '0') \
4276 clear_stat(s, si); \
4281 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4282 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4283 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4284 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4285 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4286 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4287 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4288 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4289 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4290 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4291 STAT_ATTR(FREE_SLAB
, free_slab
);
4292 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4293 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4294 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4295 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4296 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4297 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4298 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4301 static struct attribute
*slab_attrs
[] = {
4302 &slab_size_attr
.attr
,
4303 &object_size_attr
.attr
,
4304 &objs_per_slab_attr
.attr
,
4306 &min_partial_attr
.attr
,
4308 &objects_partial_attr
.attr
,
4309 &total_objects_attr
.attr
,
4312 &cpu_slabs_attr
.attr
,
4316 &sanity_checks_attr
.attr
,
4318 &hwcache_align_attr
.attr
,
4319 &reclaim_account_attr
.attr
,
4320 &destroy_by_rcu_attr
.attr
,
4321 &red_zone_attr
.attr
,
4323 &store_user_attr
.attr
,
4324 &validate_attr
.attr
,
4326 &alloc_calls_attr
.attr
,
4327 &free_calls_attr
.attr
,
4328 #ifdef CONFIG_ZONE_DMA
4329 &cache_dma_attr
.attr
,
4332 &remote_node_defrag_ratio_attr
.attr
,
4334 #ifdef CONFIG_SLUB_STATS
4335 &alloc_fastpath_attr
.attr
,
4336 &alloc_slowpath_attr
.attr
,
4337 &free_fastpath_attr
.attr
,
4338 &free_slowpath_attr
.attr
,
4339 &free_frozen_attr
.attr
,
4340 &free_add_partial_attr
.attr
,
4341 &free_remove_partial_attr
.attr
,
4342 &alloc_from_partial_attr
.attr
,
4343 &alloc_slab_attr
.attr
,
4344 &alloc_refill_attr
.attr
,
4345 &free_slab_attr
.attr
,
4346 &cpuslab_flush_attr
.attr
,
4347 &deactivate_full_attr
.attr
,
4348 &deactivate_empty_attr
.attr
,
4349 &deactivate_to_head_attr
.attr
,
4350 &deactivate_to_tail_attr
.attr
,
4351 &deactivate_remote_frees_attr
.attr
,
4352 &order_fallback_attr
.attr
,
4354 #ifdef CONFIG_FAILSLAB
4355 &failslab_attr
.attr
,
4361 static struct attribute_group slab_attr_group
= {
4362 .attrs
= slab_attrs
,
4365 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4366 struct attribute
*attr
,
4369 struct slab_attribute
*attribute
;
4370 struct kmem_cache
*s
;
4373 attribute
= to_slab_attr(attr
);
4376 if (!attribute
->show
)
4379 err
= attribute
->show(s
, buf
);
4384 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4385 struct attribute
*attr
,
4386 const char *buf
, size_t len
)
4388 struct slab_attribute
*attribute
;
4389 struct kmem_cache
*s
;
4392 attribute
= to_slab_attr(attr
);
4395 if (!attribute
->store
)
4398 err
= attribute
->store(s
, buf
, len
);
4403 static void kmem_cache_release(struct kobject
*kobj
)
4405 struct kmem_cache
*s
= to_slab(kobj
);
4411 static const struct sysfs_ops slab_sysfs_ops
= {
4412 .show
= slab_attr_show
,
4413 .store
= slab_attr_store
,
4416 static struct kobj_type slab_ktype
= {
4417 .sysfs_ops
= &slab_sysfs_ops
,
4418 .release
= kmem_cache_release
4421 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4423 struct kobj_type
*ktype
= get_ktype(kobj
);
4425 if (ktype
== &slab_ktype
)
4430 static const struct kset_uevent_ops slab_uevent_ops
= {
4431 .filter
= uevent_filter
,
4434 static struct kset
*slab_kset
;
4436 #define ID_STR_LENGTH 64
4438 /* Create a unique string id for a slab cache:
4440 * Format :[flags-]size
4442 static char *create_unique_id(struct kmem_cache
*s
)
4444 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4451 * First flags affecting slabcache operations. We will only
4452 * get here for aliasable slabs so we do not need to support
4453 * too many flags. The flags here must cover all flags that
4454 * are matched during merging to guarantee that the id is
4457 if (s
->flags
& SLAB_CACHE_DMA
)
4459 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4461 if (s
->flags
& SLAB_DEBUG_FREE
)
4463 if (!(s
->flags
& SLAB_NOTRACK
))
4467 p
+= sprintf(p
, "%07d", s
->size
);
4468 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4472 static int sysfs_slab_add(struct kmem_cache
*s
)
4478 if (slab_state
< SYSFS
)
4479 /* Defer until later */
4482 unmergeable
= slab_unmergeable(s
);
4485 * Slabcache can never be merged so we can use the name proper.
4486 * This is typically the case for debug situations. In that
4487 * case we can catch duplicate names easily.
4489 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4493 * Create a unique name for the slab as a target
4496 name
= create_unique_id(s
);
4499 s
->kobj
.kset
= slab_kset
;
4500 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4502 kobject_put(&s
->kobj
);
4506 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4508 kobject_del(&s
->kobj
);
4509 kobject_put(&s
->kobj
);
4512 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4514 /* Setup first alias */
4515 sysfs_slab_alias(s
, s
->name
);
4521 static void sysfs_slab_remove(struct kmem_cache
*s
)
4523 if (slab_state
< SYSFS
)
4525 * Sysfs has not been setup yet so no need to remove the
4530 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4531 kobject_del(&s
->kobj
);
4532 kobject_put(&s
->kobj
);
4536 * Need to buffer aliases during bootup until sysfs becomes
4537 * available lest we lose that information.
4539 struct saved_alias
{
4540 struct kmem_cache
*s
;
4542 struct saved_alias
*next
;
4545 static struct saved_alias
*alias_list
;
4547 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4549 struct saved_alias
*al
;
4551 if (slab_state
== SYSFS
) {
4553 * If we have a leftover link then remove it.
4555 sysfs_remove_link(&slab_kset
->kobj
, name
);
4556 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4559 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4565 al
->next
= alias_list
;
4570 static int __init
slab_sysfs_init(void)
4572 struct kmem_cache
*s
;
4575 down_write(&slub_lock
);
4577 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4579 up_write(&slub_lock
);
4580 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4586 list_for_each_entry(s
, &slab_caches
, list
) {
4587 err
= sysfs_slab_add(s
);
4589 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4590 " to sysfs\n", s
->name
);
4593 while (alias_list
) {
4594 struct saved_alias
*al
= alias_list
;
4596 alias_list
= alias_list
->next
;
4597 err
= sysfs_slab_alias(al
->s
, al
->name
);
4599 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4600 " %s to sysfs\n", s
->name
);
4604 up_write(&slub_lock
);
4609 __initcall(slab_sysfs_init
);
4613 * The /proc/slabinfo ABI
4615 #ifdef CONFIG_SLABINFO
4616 static void print_slabinfo_header(struct seq_file
*m
)
4618 seq_puts(m
, "slabinfo - version: 2.1\n");
4619 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4620 "<objperslab> <pagesperslab>");
4621 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4622 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4626 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4630 down_read(&slub_lock
);
4632 print_slabinfo_header(m
);
4634 return seq_list_start(&slab_caches
, *pos
);
4637 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4639 return seq_list_next(p
, &slab_caches
, pos
);
4642 static void s_stop(struct seq_file
*m
, void *p
)
4644 up_read(&slub_lock
);
4647 static int s_show(struct seq_file
*m
, void *p
)
4649 unsigned long nr_partials
= 0;
4650 unsigned long nr_slabs
= 0;
4651 unsigned long nr_inuse
= 0;
4652 unsigned long nr_objs
= 0;
4653 unsigned long nr_free
= 0;
4654 struct kmem_cache
*s
;
4657 s
= list_entry(p
, struct kmem_cache
, list
);
4659 for_each_online_node(node
) {
4660 struct kmem_cache_node
*n
= get_node(s
, node
);
4665 nr_partials
+= n
->nr_partial
;
4666 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4667 nr_objs
+= atomic_long_read(&n
->total_objects
);
4668 nr_free
+= count_partial(n
, count_free
);
4671 nr_inuse
= nr_objs
- nr_free
;
4673 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4674 nr_objs
, s
->size
, oo_objects(s
->oo
),
4675 (1 << oo_order(s
->oo
)));
4676 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4677 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4683 static const struct seq_operations slabinfo_op
= {
4690 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4692 return seq_open(file
, &slabinfo_op
);
4695 static const struct file_operations proc_slabinfo_operations
= {
4696 .open
= slabinfo_open
,
4698 .llseek
= seq_lseek
,
4699 .release
= seq_release
,
4702 static int __init
slab_proc_init(void)
4704 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4707 module_init(slab_proc_init
);
4708 #endif /* CONFIG_SLABINFO */