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>
31 #include <trace/events/kmem.h>
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache
*s
)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
135 * Mininum number of partial slabs. These will be left on the partial
136 * lists even if they are empty. kmem_cache_shrink may reclaim them.
138 #define MIN_PARTIAL 5
141 * Maximum number of desirable partial slabs.
142 * The existence of more partial slabs makes kmem_cache_shrink
143 * sort the partial list by the number of objects in the.
145 #define MAX_PARTIAL 10
147 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
148 SLAB_POISON | SLAB_STORE_USER)
151 * Debugging flags that require metadata to be stored in the slab. These get
152 * disabled when slub_debug=O is used and a cache's min order increases with
155 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
158 * Set of flags that will prevent slab merging
160 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
161 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
164 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
165 SLAB_CACHE_DMA | SLAB_NOTRACK)
168 #define OO_MASK ((1 << OO_SHIFT) - 1)
169 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
171 /* Internal SLUB flags */
172 #define __OBJECT_POISON 0x80000000UL /* Poison object */
174 static int kmem_size
= sizeof(struct kmem_cache
);
177 static struct notifier_block slab_notifier
;
181 DOWN
, /* No slab functionality available */
182 PARTIAL
, /* Kmem_cache_node works */
183 UP
, /* Everything works but does not show up in sysfs */
187 /* A list of all slab caches on the system */
188 static DECLARE_RWSEM(slub_lock
);
189 static LIST_HEAD(slab_caches
);
192 * Tracking user of a slab.
195 unsigned long addr
; /* Called from address */
196 int cpu
; /* Was running on cpu */
197 int pid
; /* Pid context */
198 unsigned long when
; /* When did the operation occur */
201 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
204 static int sysfs_slab_add(struct kmem_cache
*);
205 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
206 static void sysfs_slab_remove(struct kmem_cache
*);
209 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
210 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
212 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
220 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
222 #ifdef CONFIG_SLUB_STATS
223 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
227 /********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
231 int slab_is_available(void)
233 return slab_state
>= UP
;
236 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
238 return s
->node
[node
];
241 /* Verify that a pointer has an address that is valid within a slab page */
242 static inline int check_valid_pointer(struct kmem_cache
*s
,
243 struct page
*page
, const void *object
)
250 base
= page_address(page
);
251 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
252 (object
- base
) % s
->size
) {
259 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
261 return *(void **)(object
+ s
->offset
);
264 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
266 *(void **)(object
+ s
->offset
) = fp
;
269 /* Loop over all objects in a slab */
270 #define for_each_object(__p, __s, __addr, __objects) \
271 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
275 #define for_each_free_object(__p, __s, __free) \
276 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
278 /* Determine object index from a given position */
279 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
281 return (p
- addr
) / s
->size
;
284 static inline size_t slab_ksize(const struct kmem_cache
*s
)
286 #ifdef CONFIG_SLUB_DEBUG
288 * Debugging requires use of the padding between object
289 * and whatever may come after it.
291 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
296 * If we have the need to store the freelist pointer
297 * back there or track user information then we can
298 * only use the space before that information.
300 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
303 * Else we can use all the padding etc for the allocation
308 static inline int order_objects(int order
, unsigned long size
, int reserved
)
310 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
313 static inline struct kmem_cache_order_objects
oo_make(int order
,
314 unsigned long size
, int reserved
)
316 struct kmem_cache_order_objects x
= {
317 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
323 static inline int oo_order(struct kmem_cache_order_objects x
)
325 return x
.x
>> OO_SHIFT
;
328 static inline int oo_objects(struct kmem_cache_order_objects x
)
330 return x
.x
& OO_MASK
;
333 #ifdef CONFIG_SLUB_DEBUG
337 #ifdef CONFIG_SLUB_DEBUG_ON
338 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
340 static int slub_debug
;
343 static char *slub_debug_slabs
;
344 static int disable_higher_order_debug
;
349 static void print_section(char *text
, u8
*addr
, unsigned int length
)
357 for (i
= 0; i
< length
; i
++) {
359 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
362 printk(KERN_CONT
" %02x", addr
[i
]);
364 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
366 printk(KERN_CONT
" %s\n", ascii
);
373 printk(KERN_CONT
" ");
377 printk(KERN_CONT
" %s\n", ascii
);
381 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
382 enum track_item alloc
)
387 p
= object
+ s
->offset
+ sizeof(void *);
389 p
= object
+ s
->inuse
;
394 static void set_track(struct kmem_cache
*s
, void *object
,
395 enum track_item alloc
, unsigned long addr
)
397 struct track
*p
= get_track(s
, object
, alloc
);
401 p
->cpu
= smp_processor_id();
402 p
->pid
= current
->pid
;
405 memset(p
, 0, sizeof(struct track
));
408 static void init_tracking(struct kmem_cache
*s
, void *object
)
410 if (!(s
->flags
& SLAB_STORE_USER
))
413 set_track(s
, object
, TRACK_FREE
, 0UL);
414 set_track(s
, object
, TRACK_ALLOC
, 0UL);
417 static void print_track(const char *s
, struct track
*t
)
422 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
423 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
426 static void print_tracking(struct kmem_cache
*s
, void *object
)
428 if (!(s
->flags
& SLAB_STORE_USER
))
431 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
432 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
435 static void print_page_info(struct page
*page
)
437 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
438 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
442 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
448 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
450 printk(KERN_ERR
"========================================"
451 "=====================================\n");
452 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
453 printk(KERN_ERR
"----------------------------------------"
454 "-------------------------------------\n\n");
457 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
463 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
465 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
468 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
470 unsigned int off
; /* Offset of last byte */
471 u8
*addr
= page_address(page
);
473 print_tracking(s
, p
);
475 print_page_info(page
);
477 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
478 p
, p
- addr
, get_freepointer(s
, p
));
481 print_section("Bytes b4", p
- 16, 16);
483 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
485 if (s
->flags
& SLAB_RED_ZONE
)
486 print_section("Redzone", p
+ s
->objsize
,
487 s
->inuse
- s
->objsize
);
490 off
= s
->offset
+ sizeof(void *);
494 if (s
->flags
& SLAB_STORE_USER
)
495 off
+= 2 * sizeof(struct track
);
498 /* Beginning of the filler is the free pointer */
499 print_section("Padding", p
+ off
, s
->size
- off
);
504 static void object_err(struct kmem_cache
*s
, struct page
*page
,
505 u8
*object
, char *reason
)
507 slab_bug(s
, "%s", reason
);
508 print_trailer(s
, page
, object
);
511 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
517 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
519 slab_bug(s
, "%s", buf
);
520 print_page_info(page
);
524 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
528 if (s
->flags
& __OBJECT_POISON
) {
529 memset(p
, POISON_FREE
, s
->objsize
- 1);
530 p
[s
->objsize
- 1] = POISON_END
;
533 if (s
->flags
& SLAB_RED_ZONE
)
534 memset(p
+ s
->objsize
, val
, s
->inuse
- s
->objsize
);
537 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
540 if (*start
!= (u8
)value
)
548 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
549 void *from
, void *to
)
551 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
552 memset(from
, data
, to
- from
);
555 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
556 u8
*object
, char *what
,
557 u8
*start
, unsigned int value
, unsigned int bytes
)
562 fault
= check_bytes(start
, value
, bytes
);
567 while (end
> fault
&& end
[-1] == value
)
570 slab_bug(s
, "%s overwritten", what
);
571 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
572 fault
, end
- 1, fault
[0], value
);
573 print_trailer(s
, page
, object
);
575 restore_bytes(s
, what
, value
, fault
, end
);
583 * Bytes of the object to be managed.
584 * If the freepointer may overlay the object then the free
585 * pointer is the first word of the object.
587 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
590 * object + s->objsize
591 * Padding to reach word boundary. This is also used for Redzoning.
592 * Padding is extended by another word if Redzoning is enabled and
595 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
596 * 0xcc (RED_ACTIVE) for objects in use.
599 * Meta data starts here.
601 * A. Free pointer (if we cannot overwrite object on free)
602 * B. Tracking data for SLAB_STORE_USER
603 * C. Padding to reach required alignment boundary or at mininum
604 * one word if debugging is on to be able to detect writes
605 * before the word boundary.
607 * Padding is done using 0x5a (POISON_INUSE)
610 * Nothing is used beyond s->size.
612 * If slabcaches are merged then the objsize and inuse boundaries are mostly
613 * ignored. And therefore no slab options that rely on these boundaries
614 * may be used with merged slabcaches.
617 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
619 unsigned long off
= s
->inuse
; /* The end of info */
622 /* Freepointer is placed after the object. */
623 off
+= sizeof(void *);
625 if (s
->flags
& SLAB_STORE_USER
)
626 /* We also have user information there */
627 off
+= 2 * sizeof(struct track
);
632 return check_bytes_and_report(s
, page
, p
, "Object padding",
633 p
+ off
, POISON_INUSE
, s
->size
- off
);
636 /* Check the pad bytes at the end of a slab page */
637 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
645 if (!(s
->flags
& SLAB_POISON
))
648 start
= page_address(page
);
649 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
650 end
= start
+ length
;
651 remainder
= length
% s
->size
;
655 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
658 while (end
> fault
&& end
[-1] == POISON_INUSE
)
661 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
662 print_section("Padding", end
- remainder
, remainder
);
664 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
668 static int check_object(struct kmem_cache
*s
, struct page
*page
,
669 void *object
, u8 val
)
672 u8
*endobject
= object
+ s
->objsize
;
674 if (s
->flags
& SLAB_RED_ZONE
) {
675 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
676 endobject
, val
, s
->inuse
- s
->objsize
))
679 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
680 check_bytes_and_report(s
, page
, p
, "Alignment padding",
681 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
685 if (s
->flags
& SLAB_POISON
) {
686 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
687 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
688 POISON_FREE
, s
->objsize
- 1) ||
689 !check_bytes_and_report(s
, page
, p
, "Poison",
690 p
+ s
->objsize
- 1, POISON_END
, 1)))
693 * check_pad_bytes cleans up on its own.
695 check_pad_bytes(s
, page
, p
);
698 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
700 * Object and freepointer overlap. Cannot check
701 * freepointer while object is allocated.
705 /* Check free pointer validity */
706 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
707 object_err(s
, page
, p
, "Freepointer corrupt");
709 * No choice but to zap it and thus lose the remainder
710 * of the free objects in this slab. May cause
711 * another error because the object count is now wrong.
713 set_freepointer(s
, p
, NULL
);
719 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
723 VM_BUG_ON(!irqs_disabled());
725 if (!PageSlab(page
)) {
726 slab_err(s
, page
, "Not a valid slab page");
730 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
731 if (page
->objects
> maxobj
) {
732 slab_err(s
, page
, "objects %u > max %u",
733 s
->name
, page
->objects
, maxobj
);
736 if (page
->inuse
> page
->objects
) {
737 slab_err(s
, page
, "inuse %u > max %u",
738 s
->name
, page
->inuse
, page
->objects
);
741 /* Slab_pad_check fixes things up after itself */
742 slab_pad_check(s
, page
);
747 * Determine if a certain object on a page is on the freelist. Must hold the
748 * slab lock to guarantee that the chains are in a consistent state.
750 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
753 void *fp
= page
->freelist
;
755 unsigned long max_objects
;
757 while (fp
&& nr
<= page
->objects
) {
760 if (!check_valid_pointer(s
, page
, fp
)) {
762 object_err(s
, page
, object
,
763 "Freechain corrupt");
764 set_freepointer(s
, object
, NULL
);
767 slab_err(s
, page
, "Freepointer corrupt");
768 page
->freelist
= NULL
;
769 page
->inuse
= page
->objects
;
770 slab_fix(s
, "Freelist cleared");
776 fp
= get_freepointer(s
, object
);
780 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
781 if (max_objects
> MAX_OBJS_PER_PAGE
)
782 max_objects
= MAX_OBJS_PER_PAGE
;
784 if (page
->objects
!= max_objects
) {
785 slab_err(s
, page
, "Wrong number of objects. Found %d but "
786 "should be %d", page
->objects
, max_objects
);
787 page
->objects
= max_objects
;
788 slab_fix(s
, "Number of objects adjusted.");
790 if (page
->inuse
!= page
->objects
- nr
) {
791 slab_err(s
, page
, "Wrong object count. Counter is %d but "
792 "counted were %d", page
->inuse
, page
->objects
- nr
);
793 page
->inuse
= page
->objects
- nr
;
794 slab_fix(s
, "Object count adjusted.");
796 return search
== NULL
;
799 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
802 if (s
->flags
& SLAB_TRACE
) {
803 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
805 alloc
? "alloc" : "free",
810 print_section("Object", (void *)object
, s
->objsize
);
817 * Hooks for other subsystems that check memory allocations. In a typical
818 * production configuration these hooks all should produce no code at all.
820 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
822 flags
&= gfp_allowed_mask
;
823 lockdep_trace_alloc(flags
);
824 might_sleep_if(flags
& __GFP_WAIT
);
826 return should_failslab(s
->objsize
, flags
, s
->flags
);
829 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
831 flags
&= gfp_allowed_mask
;
832 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
833 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
836 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
838 kmemleak_free_recursive(x
, s
->flags
);
841 * Trouble is that we may no longer disable interupts in the fast path
842 * So in order to make the debug calls that expect irqs to be
843 * disabled we need to disable interrupts temporarily.
845 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
849 local_irq_save(flags
);
850 kmemcheck_slab_free(s
, x
, s
->objsize
);
851 debug_check_no_locks_freed(x
, s
->objsize
);
852 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
853 debug_check_no_obj_freed(x
, s
->objsize
);
854 local_irq_restore(flags
);
860 * Tracking of fully allocated slabs for debugging purposes.
862 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
864 spin_lock(&n
->list_lock
);
865 list_add(&page
->lru
, &n
->full
);
866 spin_unlock(&n
->list_lock
);
869 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
871 struct kmem_cache_node
*n
;
873 if (!(s
->flags
& SLAB_STORE_USER
))
876 n
= get_node(s
, page_to_nid(page
));
878 spin_lock(&n
->list_lock
);
879 list_del(&page
->lru
);
880 spin_unlock(&n
->list_lock
);
883 /* Tracking of the number of slabs for debugging purposes */
884 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
886 struct kmem_cache_node
*n
= get_node(s
, node
);
888 return atomic_long_read(&n
->nr_slabs
);
891 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
893 return atomic_long_read(&n
->nr_slabs
);
896 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
898 struct kmem_cache_node
*n
= get_node(s
, node
);
901 * May be called early in order to allocate a slab for the
902 * kmem_cache_node structure. Solve the chicken-egg
903 * dilemma by deferring the increment of the count during
904 * bootstrap (see early_kmem_cache_node_alloc).
907 atomic_long_inc(&n
->nr_slabs
);
908 atomic_long_add(objects
, &n
->total_objects
);
911 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
913 struct kmem_cache_node
*n
= get_node(s
, node
);
915 atomic_long_dec(&n
->nr_slabs
);
916 atomic_long_sub(objects
, &n
->total_objects
);
919 /* Object debug checks for alloc/free paths */
920 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
923 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
926 init_object(s
, object
, SLUB_RED_INACTIVE
);
927 init_tracking(s
, object
);
930 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
931 void *object
, unsigned long addr
)
933 if (!check_slab(s
, page
))
936 if (!on_freelist(s
, page
, object
)) {
937 object_err(s
, page
, object
, "Object already allocated");
941 if (!check_valid_pointer(s
, page
, object
)) {
942 object_err(s
, page
, object
, "Freelist Pointer check fails");
946 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
949 /* Success perform special debug activities for allocs */
950 if (s
->flags
& SLAB_STORE_USER
)
951 set_track(s
, object
, TRACK_ALLOC
, addr
);
952 trace(s
, page
, object
, 1);
953 init_object(s
, object
, SLUB_RED_ACTIVE
);
957 if (PageSlab(page
)) {
959 * If this is a slab page then lets do the best we can
960 * to avoid issues in the future. Marking all objects
961 * as used avoids touching the remaining objects.
963 slab_fix(s
, "Marking all objects used");
964 page
->inuse
= page
->objects
;
965 page
->freelist
= NULL
;
970 static noinline
int free_debug_processing(struct kmem_cache
*s
,
971 struct page
*page
, void *object
, unsigned long addr
)
973 if (!check_slab(s
, page
))
976 if (!check_valid_pointer(s
, page
, object
)) {
977 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
981 if (on_freelist(s
, page
, object
)) {
982 object_err(s
, page
, object
, "Object already free");
986 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
989 if (unlikely(s
!= page
->slab
)) {
990 if (!PageSlab(page
)) {
991 slab_err(s
, page
, "Attempt to free object(0x%p) "
992 "outside of slab", object
);
993 } else if (!page
->slab
) {
995 "SLUB <none>: no slab for object 0x%p.\n",
999 object_err(s
, page
, object
,
1000 "page slab pointer corrupt.");
1004 /* Special debug activities for freeing objects */
1005 if (!PageSlubFrozen(page
) && !page
->freelist
)
1006 remove_full(s
, page
);
1007 if (s
->flags
& SLAB_STORE_USER
)
1008 set_track(s
, object
, TRACK_FREE
, addr
);
1009 trace(s
, page
, object
, 0);
1010 init_object(s
, object
, SLUB_RED_INACTIVE
);
1014 slab_fix(s
, "Object at 0x%p not freed", object
);
1018 static int __init
setup_slub_debug(char *str
)
1020 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1021 if (*str
++ != '=' || !*str
)
1023 * No options specified. Switch on full debugging.
1029 * No options but restriction on slabs. This means full
1030 * debugging for slabs matching a pattern.
1034 if (tolower(*str
) == 'o') {
1036 * Avoid enabling debugging on caches if its minimum order
1037 * would increase as a result.
1039 disable_higher_order_debug
= 1;
1046 * Switch off all debugging measures.
1051 * Determine which debug features should be switched on
1053 for (; *str
&& *str
!= ','; str
++) {
1054 switch (tolower(*str
)) {
1056 slub_debug
|= SLAB_DEBUG_FREE
;
1059 slub_debug
|= SLAB_RED_ZONE
;
1062 slub_debug
|= SLAB_POISON
;
1065 slub_debug
|= SLAB_STORE_USER
;
1068 slub_debug
|= SLAB_TRACE
;
1071 slub_debug
|= SLAB_FAILSLAB
;
1074 printk(KERN_ERR
"slub_debug option '%c' "
1075 "unknown. skipped\n", *str
);
1081 slub_debug_slabs
= str
+ 1;
1086 __setup("slub_debug", setup_slub_debug
);
1088 static unsigned long kmem_cache_flags(unsigned long objsize
,
1089 unsigned long flags
, const char *name
,
1090 void (*ctor
)(void *))
1093 * Enable debugging if selected on the kernel commandline.
1095 if (slub_debug
&& (!slub_debug_slabs
||
1096 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1097 flags
|= slub_debug
;
1102 static inline void setup_object_debug(struct kmem_cache
*s
,
1103 struct page
*page
, void *object
) {}
1105 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1106 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1108 static inline int free_debug_processing(struct kmem_cache
*s
,
1109 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1111 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1113 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1114 void *object
, u8 val
) { return 1; }
1115 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1116 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1117 unsigned long flags
, const char *name
,
1118 void (*ctor
)(void *))
1122 #define slub_debug 0
1124 #define disable_higher_order_debug 0
1126 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1128 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1130 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1132 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1135 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1138 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1141 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1143 #endif /* CONFIG_SLUB_DEBUG */
1146 * Slab allocation and freeing
1148 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1149 struct kmem_cache_order_objects oo
)
1151 int order
= oo_order(oo
);
1153 flags
|= __GFP_NOTRACK
;
1155 if (node
== NUMA_NO_NODE
)
1156 return alloc_pages(flags
, order
);
1158 return alloc_pages_exact_node(node
, flags
, order
);
1161 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1164 struct kmem_cache_order_objects oo
= s
->oo
;
1167 flags
|= s
->allocflags
;
1170 * Let the initial higher-order allocation fail under memory pressure
1171 * so we fall-back to the minimum order allocation.
1173 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1175 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1176 if (unlikely(!page
)) {
1179 * Allocation may have failed due to fragmentation.
1180 * Try a lower order alloc if possible
1182 page
= alloc_slab_page(flags
, node
, oo
);
1186 stat(s
, ORDER_FALLBACK
);
1189 if (kmemcheck_enabled
1190 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1191 int pages
= 1 << oo_order(oo
);
1193 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1196 * Objects from caches that have a constructor don't get
1197 * cleared when they're allocated, so we need to do it here.
1200 kmemcheck_mark_uninitialized_pages(page
, pages
);
1202 kmemcheck_mark_unallocated_pages(page
, pages
);
1205 page
->objects
= oo_objects(oo
);
1206 mod_zone_page_state(page_zone(page
),
1207 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1208 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1214 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1217 setup_object_debug(s
, page
, object
);
1218 if (unlikely(s
->ctor
))
1222 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1229 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1231 page
= allocate_slab(s
,
1232 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1236 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1238 page
->flags
|= 1 << PG_slab
;
1240 start
= page_address(page
);
1242 if (unlikely(s
->flags
& SLAB_POISON
))
1243 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1246 for_each_object(p
, s
, start
, page
->objects
) {
1247 setup_object(s
, page
, last
);
1248 set_freepointer(s
, last
, p
);
1251 setup_object(s
, page
, last
);
1252 set_freepointer(s
, last
, NULL
);
1254 page
->freelist
= start
;
1260 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1262 int order
= compound_order(page
);
1263 int pages
= 1 << order
;
1265 if (kmem_cache_debug(s
)) {
1268 slab_pad_check(s
, page
);
1269 for_each_object(p
, s
, page_address(page
),
1271 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1274 kmemcheck_free_shadow(page
, compound_order(page
));
1276 mod_zone_page_state(page_zone(page
),
1277 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1278 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1281 __ClearPageSlab(page
);
1282 reset_page_mapcount(page
);
1283 if (current
->reclaim_state
)
1284 current
->reclaim_state
->reclaimed_slab
+= pages
;
1285 __free_pages(page
, order
);
1288 #define need_reserve_slab_rcu \
1289 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1291 static void rcu_free_slab(struct rcu_head
*h
)
1295 if (need_reserve_slab_rcu
)
1296 page
= virt_to_head_page(h
);
1298 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1300 __free_slab(page
->slab
, page
);
1303 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1305 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1306 struct rcu_head
*head
;
1308 if (need_reserve_slab_rcu
) {
1309 int order
= compound_order(page
);
1310 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1312 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1313 head
= page_address(page
) + offset
;
1316 * RCU free overloads the RCU head over the LRU
1318 head
= (void *)&page
->lru
;
1321 call_rcu(head
, rcu_free_slab
);
1323 __free_slab(s
, page
);
1326 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1328 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1333 * Per slab locking using the pagelock
1335 static __always_inline
void slab_lock(struct page
*page
)
1337 bit_spin_lock(PG_locked
, &page
->flags
);
1340 static __always_inline
void slab_unlock(struct page
*page
)
1342 __bit_spin_unlock(PG_locked
, &page
->flags
);
1345 static __always_inline
int slab_trylock(struct page
*page
)
1349 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1354 * Management of partially allocated slabs
1356 static void add_partial(struct kmem_cache_node
*n
,
1357 struct page
*page
, int tail
)
1359 spin_lock(&n
->list_lock
);
1362 list_add_tail(&page
->lru
, &n
->partial
);
1364 list_add(&page
->lru
, &n
->partial
);
1365 spin_unlock(&n
->list_lock
);
1368 static inline void __remove_partial(struct kmem_cache_node
*n
,
1371 list_del(&page
->lru
);
1375 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1377 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1379 spin_lock(&n
->list_lock
);
1380 __remove_partial(n
, page
);
1381 spin_unlock(&n
->list_lock
);
1385 * Lock slab and remove from the partial list.
1387 * Must hold list_lock.
1389 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1392 if (slab_trylock(page
)) {
1393 __remove_partial(n
, page
);
1394 __SetPageSlubFrozen(page
);
1401 * Try to allocate a partial slab from a specific node.
1403 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1408 * Racy check. If we mistakenly see no partial slabs then we
1409 * just allocate an empty slab. If we mistakenly try to get a
1410 * partial slab and there is none available then get_partials()
1413 if (!n
|| !n
->nr_partial
)
1416 spin_lock(&n
->list_lock
);
1417 list_for_each_entry(page
, &n
->partial
, lru
)
1418 if (lock_and_freeze_slab(n
, page
))
1422 spin_unlock(&n
->list_lock
);
1427 * Get a page from somewhere. Search in increasing NUMA distances.
1429 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1432 struct zonelist
*zonelist
;
1435 enum zone_type high_zoneidx
= gfp_zone(flags
);
1439 * The defrag ratio allows a configuration of the tradeoffs between
1440 * inter node defragmentation and node local allocations. A lower
1441 * defrag_ratio increases the tendency to do local allocations
1442 * instead of attempting to obtain partial slabs from other nodes.
1444 * If the defrag_ratio is set to 0 then kmalloc() always
1445 * returns node local objects. If the ratio is higher then kmalloc()
1446 * may return off node objects because partial slabs are obtained
1447 * from other nodes and filled up.
1449 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1450 * defrag_ratio = 1000) then every (well almost) allocation will
1451 * first attempt to defrag slab caches on other nodes. This means
1452 * scanning over all nodes to look for partial slabs which may be
1453 * expensive if we do it every time we are trying to find a slab
1454 * with available objects.
1456 if (!s
->remote_node_defrag_ratio
||
1457 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1461 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1462 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1463 struct kmem_cache_node
*n
;
1465 n
= get_node(s
, zone_to_nid(zone
));
1467 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1468 n
->nr_partial
> s
->min_partial
) {
1469 page
= get_partial_node(n
);
1482 * Get a partial page, lock it and return it.
1484 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1487 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1489 page
= get_partial_node(get_node(s
, searchnode
));
1490 if (page
|| node
!= -1)
1493 return get_any_partial(s
, flags
);
1497 * Move a page back to the lists.
1499 * Must be called with the slab lock held.
1501 * On exit the slab lock will have been dropped.
1503 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1506 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1508 __ClearPageSlubFrozen(page
);
1511 if (page
->freelist
) {
1512 add_partial(n
, page
, tail
);
1513 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1515 stat(s
, DEACTIVATE_FULL
);
1516 if (kmem_cache_debug(s
) && (s
->flags
& SLAB_STORE_USER
))
1521 stat(s
, DEACTIVATE_EMPTY
);
1522 if (n
->nr_partial
< s
->min_partial
) {
1524 * Adding an empty slab to the partial slabs in order
1525 * to avoid page allocator overhead. This slab needs
1526 * to come after the other slabs with objects in
1527 * so that the others get filled first. That way the
1528 * size of the partial list stays small.
1530 * kmem_cache_shrink can reclaim any empty slabs from
1533 add_partial(n
, page
, 1);
1538 discard_slab(s
, page
);
1543 #ifdef CONFIG_CMPXCHG_LOCAL
1544 #ifdef CONFIG_PREEMPT
1546 * Calculate the next globally unique transaction for disambiguiation
1547 * during cmpxchg. The transactions start with the cpu number and are then
1548 * incremented by CONFIG_NR_CPUS.
1550 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1553 * No preemption supported therefore also no need to check for
1559 static inline unsigned long next_tid(unsigned long tid
)
1561 return tid
+ TID_STEP
;
1564 static inline unsigned int tid_to_cpu(unsigned long tid
)
1566 return tid
% TID_STEP
;
1569 static inline unsigned long tid_to_event(unsigned long tid
)
1571 return tid
/ TID_STEP
;
1574 static inline unsigned int init_tid(int cpu
)
1579 static inline void note_cmpxchg_failure(const char *n
,
1580 const struct kmem_cache
*s
, unsigned long tid
)
1582 #ifdef SLUB_DEBUG_CMPXCHG
1583 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1585 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1587 #ifdef CONFIG_PREEMPT
1588 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1589 printk("due to cpu change %d -> %d\n",
1590 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1593 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1594 printk("due to cpu running other code. Event %ld->%ld\n",
1595 tid_to_event(tid
), tid_to_event(actual_tid
));
1597 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1598 actual_tid
, tid
, next_tid(tid
));
1604 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1606 #if defined(CONFIG_CMPXCHG_LOCAL) && defined(CONFIG_PREEMPT)
1609 for_each_possible_cpu(cpu
)
1610 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1615 * Remove the cpu slab
1617 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1620 struct page
*page
= c
->page
;
1624 stat(s
, DEACTIVATE_REMOTE_FREES
);
1626 * Merge cpu freelist into slab freelist. Typically we get here
1627 * because both freelists are empty. So this is unlikely
1630 while (unlikely(c
->freelist
)) {
1633 tail
= 0; /* Hot objects. Put the slab first */
1635 /* Retrieve object from cpu_freelist */
1636 object
= c
->freelist
;
1637 c
->freelist
= get_freepointer(s
, c
->freelist
);
1639 /* And put onto the regular freelist */
1640 set_freepointer(s
, object
, page
->freelist
);
1641 page
->freelist
= object
;
1645 #ifdef CONFIG_CMPXCHG_LOCAL
1646 c
->tid
= next_tid(c
->tid
);
1648 unfreeze_slab(s
, page
, tail
);
1651 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1653 stat(s
, CPUSLAB_FLUSH
);
1655 deactivate_slab(s
, c
);
1661 * Called from IPI handler with interrupts disabled.
1663 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1665 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1667 if (likely(c
&& c
->page
))
1671 static void flush_cpu_slab(void *d
)
1673 struct kmem_cache
*s
= d
;
1675 __flush_cpu_slab(s
, smp_processor_id());
1678 static void flush_all(struct kmem_cache
*s
)
1680 on_each_cpu(flush_cpu_slab
, s
, 1);
1684 * Check if the objects in a per cpu structure fit numa
1685 * locality expectations.
1687 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1690 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
1696 static int count_free(struct page
*page
)
1698 return page
->objects
- page
->inuse
;
1701 static unsigned long count_partial(struct kmem_cache_node
*n
,
1702 int (*get_count
)(struct page
*))
1704 unsigned long flags
;
1705 unsigned long x
= 0;
1708 spin_lock_irqsave(&n
->list_lock
, flags
);
1709 list_for_each_entry(page
, &n
->partial
, lru
)
1710 x
+= get_count(page
);
1711 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1715 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1717 #ifdef CONFIG_SLUB_DEBUG
1718 return atomic_long_read(&n
->total_objects
);
1724 static noinline
void
1725 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1730 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1732 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1733 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1734 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1736 if (oo_order(s
->min
) > get_order(s
->objsize
))
1737 printk(KERN_WARNING
" %s debugging increased min order, use "
1738 "slub_debug=O to disable.\n", s
->name
);
1740 for_each_online_node(node
) {
1741 struct kmem_cache_node
*n
= get_node(s
, node
);
1742 unsigned long nr_slabs
;
1743 unsigned long nr_objs
;
1744 unsigned long nr_free
;
1749 nr_free
= count_partial(n
, count_free
);
1750 nr_slabs
= node_nr_slabs(n
);
1751 nr_objs
= node_nr_objs(n
);
1754 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1755 node
, nr_slabs
, nr_objs
, nr_free
);
1760 * Slow path. The lockless freelist is empty or we need to perform
1763 * Interrupts are disabled.
1765 * Processing is still very fast if new objects have been freed to the
1766 * regular freelist. In that case we simply take over the regular freelist
1767 * as the lockless freelist and zap the regular freelist.
1769 * If that is not working then we fall back to the partial lists. We take the
1770 * first element of the freelist as the object to allocate now and move the
1771 * rest of the freelist to the lockless freelist.
1773 * And if we were unable to get a new slab from the partial slab lists then
1774 * we need to allocate a new slab. This is the slowest path since it involves
1775 * a call to the page allocator and the setup of a new slab.
1777 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1778 unsigned long addr
, struct kmem_cache_cpu
*c
)
1782 #ifdef CONFIG_CMPXCHG_LOCAL
1783 unsigned long flags
;
1785 local_irq_save(flags
);
1786 #ifdef CONFIG_PREEMPT
1788 * We may have been preempted and rescheduled on a different
1789 * cpu before disabling interrupts. Need to reload cpu area
1792 c
= this_cpu_ptr(s
->cpu_slab
);
1796 /* We handle __GFP_ZERO in the caller */
1797 gfpflags
&= ~__GFP_ZERO
;
1803 if (unlikely(!node_match(c
, node
)))
1806 stat(s
, ALLOC_REFILL
);
1809 object
= c
->page
->freelist
;
1810 if (unlikely(!object
))
1812 if (kmem_cache_debug(s
))
1815 c
->freelist
= get_freepointer(s
, object
);
1816 c
->page
->inuse
= c
->page
->objects
;
1817 c
->page
->freelist
= NULL
;
1818 c
->node
= page_to_nid(c
->page
);
1820 slab_unlock(c
->page
);
1821 #ifdef CONFIG_CMPXCHG_LOCAL
1822 c
->tid
= next_tid(c
->tid
);
1823 local_irq_restore(flags
);
1825 stat(s
, ALLOC_SLOWPATH
);
1829 deactivate_slab(s
, c
);
1832 new = get_partial(s
, gfpflags
, node
);
1835 stat(s
, ALLOC_FROM_PARTIAL
);
1839 gfpflags
&= gfp_allowed_mask
;
1840 if (gfpflags
& __GFP_WAIT
)
1843 new = new_slab(s
, gfpflags
, node
);
1845 if (gfpflags
& __GFP_WAIT
)
1846 local_irq_disable();
1849 c
= __this_cpu_ptr(s
->cpu_slab
);
1850 stat(s
, ALLOC_SLAB
);
1854 __SetPageSlubFrozen(new);
1858 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1859 slab_out_of_memory(s
, gfpflags
, node
);
1862 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1866 c
->page
->freelist
= get_freepointer(s
, object
);
1867 c
->node
= NUMA_NO_NODE
;
1872 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1873 * have the fastpath folded into their functions. So no function call
1874 * overhead for requests that can be satisfied on the fastpath.
1876 * The fastpath works by first checking if the lockless freelist can be used.
1877 * If not then __slab_alloc is called for slow processing.
1879 * Otherwise we can simply pick the next object from the lockless free list.
1881 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1882 gfp_t gfpflags
, int node
, unsigned long addr
)
1885 struct kmem_cache_cpu
*c
;
1886 #ifdef CONFIG_CMPXCHG_LOCAL
1889 unsigned long flags
;
1892 if (slab_pre_alloc_hook(s
, gfpflags
))
1895 #ifndef CONFIG_CMPXCHG_LOCAL
1896 local_irq_save(flags
);
1902 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1903 * enabled. We may switch back and forth between cpus while
1904 * reading from one cpu area. That does not matter as long
1905 * as we end up on the original cpu again when doing the cmpxchg.
1907 c
= __this_cpu_ptr(s
->cpu_slab
);
1909 #ifdef CONFIG_CMPXCHG_LOCAL
1911 * The transaction ids are globally unique per cpu and per operation on
1912 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1913 * occurs on the right processor and that there was no operation on the
1914 * linked list in between.
1920 object
= c
->freelist
;
1921 if (unlikely(!object
|| !node_match(c
, node
)))
1923 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1926 #ifdef CONFIG_CMPXCHG_LOCAL
1928 * The cmpxchg will only match if there was no additonal
1929 * operation and if we are on the right processor.
1931 * The cmpxchg does the following atomically (without lock semantics!)
1932 * 1. Relocate first pointer to the current per cpu area.
1933 * 2. Verify that tid and freelist have not been changed
1934 * 3. If they were not changed replace tid and freelist
1936 * Since this is without lock semantics the protection is only against
1937 * code executing on this cpu *not* from access by other cpus.
1939 if (unlikely(!this_cpu_cmpxchg_double(
1940 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
1942 get_freepointer(s
, object
), next_tid(tid
)))) {
1944 note_cmpxchg_failure("slab_alloc", s
, tid
);
1948 c
->freelist
= get_freepointer(s
, object
);
1950 stat(s
, ALLOC_FASTPATH
);
1953 #ifndef CONFIG_CMPXCHG_LOCAL
1954 local_irq_restore(flags
);
1957 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1958 memset(object
, 0, s
->objsize
);
1960 slab_post_alloc_hook(s
, gfpflags
, object
);
1965 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1967 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1969 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1973 EXPORT_SYMBOL(kmem_cache_alloc
);
1975 #ifdef CONFIG_TRACING
1976 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
1978 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1979 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
1982 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
1984 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
1986 void *ret
= kmalloc_order(size
, flags
, order
);
1987 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
1990 EXPORT_SYMBOL(kmalloc_order_trace
);
1994 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1996 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1998 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1999 s
->objsize
, s
->size
, gfpflags
, node
);
2003 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2005 #ifdef CONFIG_TRACING
2006 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2008 int node
, size_t size
)
2010 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2012 trace_kmalloc_node(_RET_IP_
, ret
,
2013 size
, s
->size
, gfpflags
, node
);
2016 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2021 * Slow patch handling. This may still be called frequently since objects
2022 * have a longer lifetime than the cpu slabs in most processing loads.
2024 * So we still attempt to reduce cache line usage. Just take the slab
2025 * lock and free the item. If there is no additional partial page
2026 * handling required then we can return immediately.
2028 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2029 void *x
, unsigned long addr
)
2032 void **object
= (void *)x
;
2033 #ifdef CONFIG_CMPXCHG_LOCAL
2034 unsigned long flags
;
2036 local_irq_save(flags
);
2039 stat(s
, FREE_SLOWPATH
);
2041 if (kmem_cache_debug(s
))
2045 prior
= page
->freelist
;
2046 set_freepointer(s
, object
, prior
);
2047 page
->freelist
= object
;
2050 if (unlikely(PageSlubFrozen(page
))) {
2051 stat(s
, FREE_FROZEN
);
2055 if (unlikely(!page
->inuse
))
2059 * Objects left in the slab. If it was not on the partial list before
2062 if (unlikely(!prior
)) {
2063 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
2064 stat(s
, FREE_ADD_PARTIAL
);
2069 #ifdef CONFIG_CMPXCHG_LOCAL
2070 local_irq_restore(flags
);
2077 * Slab still on the partial list.
2079 remove_partial(s
, page
);
2080 stat(s
, FREE_REMOVE_PARTIAL
);
2083 #ifdef CONFIG_CMPXCHG_LOCAL
2084 local_irq_restore(flags
);
2087 discard_slab(s
, page
);
2091 if (!free_debug_processing(s
, page
, x
, addr
))
2097 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2098 * can perform fastpath freeing without additional function calls.
2100 * The fastpath is only possible if we are freeing to the current cpu slab
2101 * of this processor. This typically the case if we have just allocated
2104 * If fastpath is not possible then fall back to __slab_free where we deal
2105 * with all sorts of special processing.
2107 static __always_inline
void slab_free(struct kmem_cache
*s
,
2108 struct page
*page
, void *x
, unsigned long addr
)
2110 void **object
= (void *)x
;
2111 struct kmem_cache_cpu
*c
;
2112 #ifdef CONFIG_CMPXCHG_LOCAL
2115 unsigned long flags
;
2118 slab_free_hook(s
, x
);
2120 #ifndef CONFIG_CMPXCHG_LOCAL
2121 local_irq_save(flags
);
2128 * Determine the currently cpus per cpu slab.
2129 * The cpu may change afterward. However that does not matter since
2130 * data is retrieved via this pointer. If we are on the same cpu
2131 * during the cmpxchg then the free will succedd.
2133 c
= __this_cpu_ptr(s
->cpu_slab
);
2135 #ifdef CONFIG_CMPXCHG_LOCAL
2140 if (likely(page
== c
->page
&& c
->node
!= NUMA_NO_NODE
)) {
2141 set_freepointer(s
, object
, c
->freelist
);
2143 #ifdef CONFIG_CMPXCHG_LOCAL
2144 if (unlikely(!this_cpu_cmpxchg_double(
2145 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2147 object
, next_tid(tid
)))) {
2149 note_cmpxchg_failure("slab_free", s
, tid
);
2153 c
->freelist
= object
;
2155 stat(s
, FREE_FASTPATH
);
2157 __slab_free(s
, page
, x
, addr
);
2159 #ifndef CONFIG_CMPXCHG_LOCAL
2160 local_irq_restore(flags
);
2164 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2168 page
= virt_to_head_page(x
);
2170 slab_free(s
, page
, x
, _RET_IP_
);
2172 trace_kmem_cache_free(_RET_IP_
, x
);
2174 EXPORT_SYMBOL(kmem_cache_free
);
2177 * Object placement in a slab is made very easy because we always start at
2178 * offset 0. If we tune the size of the object to the alignment then we can
2179 * get the required alignment by putting one properly sized object after
2182 * Notice that the allocation order determines the sizes of the per cpu
2183 * caches. Each processor has always one slab available for allocations.
2184 * Increasing the allocation order reduces the number of times that slabs
2185 * must be moved on and off the partial lists and is therefore a factor in
2190 * Mininum / Maximum order of slab pages. This influences locking overhead
2191 * and slab fragmentation. A higher order reduces the number of partial slabs
2192 * and increases the number of allocations possible without having to
2193 * take the list_lock.
2195 static int slub_min_order
;
2196 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2197 static int slub_min_objects
;
2200 * Merge control. If this is set then no merging of slab caches will occur.
2201 * (Could be removed. This was introduced to pacify the merge skeptics.)
2203 static int slub_nomerge
;
2206 * Calculate the order of allocation given an slab object size.
2208 * The order of allocation has significant impact on performance and other
2209 * system components. Generally order 0 allocations should be preferred since
2210 * order 0 does not cause fragmentation in the page allocator. Larger objects
2211 * be problematic to put into order 0 slabs because there may be too much
2212 * unused space left. We go to a higher order if more than 1/16th of the slab
2215 * In order to reach satisfactory performance we must ensure that a minimum
2216 * number of objects is in one slab. Otherwise we may generate too much
2217 * activity on the partial lists which requires taking the list_lock. This is
2218 * less a concern for large slabs though which are rarely used.
2220 * slub_max_order specifies the order where we begin to stop considering the
2221 * number of objects in a slab as critical. If we reach slub_max_order then
2222 * we try to keep the page order as low as possible. So we accept more waste
2223 * of space in favor of a small page order.
2225 * Higher order allocations also allow the placement of more objects in a
2226 * slab and thereby reduce object handling overhead. If the user has
2227 * requested a higher mininum order then we start with that one instead of
2228 * the smallest order which will fit the object.
2230 static inline int slab_order(int size
, int min_objects
,
2231 int max_order
, int fract_leftover
, int reserved
)
2235 int min_order
= slub_min_order
;
2237 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2238 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2240 for (order
= max(min_order
,
2241 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2242 order
<= max_order
; order
++) {
2244 unsigned long slab_size
= PAGE_SIZE
<< order
;
2246 if (slab_size
< min_objects
* size
+ reserved
)
2249 rem
= (slab_size
- reserved
) % size
;
2251 if (rem
<= slab_size
/ fract_leftover
)
2259 static inline int calculate_order(int size
, int reserved
)
2267 * Attempt to find best configuration for a slab. This
2268 * works by first attempting to generate a layout with
2269 * the best configuration and backing off gradually.
2271 * First we reduce the acceptable waste in a slab. Then
2272 * we reduce the minimum objects required in a slab.
2274 min_objects
= slub_min_objects
;
2276 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2277 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2278 min_objects
= min(min_objects
, max_objects
);
2280 while (min_objects
> 1) {
2282 while (fraction
>= 4) {
2283 order
= slab_order(size
, min_objects
,
2284 slub_max_order
, fraction
, reserved
);
2285 if (order
<= slub_max_order
)
2293 * We were unable to place multiple objects in a slab. Now
2294 * lets see if we can place a single object there.
2296 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2297 if (order
<= slub_max_order
)
2301 * Doh this slab cannot be placed using slub_max_order.
2303 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2304 if (order
< MAX_ORDER
)
2310 * Figure out what the alignment of the objects will be.
2312 static unsigned long calculate_alignment(unsigned long flags
,
2313 unsigned long align
, unsigned long size
)
2316 * If the user wants hardware cache aligned objects then follow that
2317 * suggestion if the object is sufficiently large.
2319 * The hardware cache alignment cannot override the specified
2320 * alignment though. If that is greater then use it.
2322 if (flags
& SLAB_HWCACHE_ALIGN
) {
2323 unsigned long ralign
= cache_line_size();
2324 while (size
<= ralign
/ 2)
2326 align
= max(align
, ralign
);
2329 if (align
< ARCH_SLAB_MINALIGN
)
2330 align
= ARCH_SLAB_MINALIGN
;
2332 return ALIGN(align
, sizeof(void *));
2336 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2339 spin_lock_init(&n
->list_lock
);
2340 INIT_LIST_HEAD(&n
->partial
);
2341 #ifdef CONFIG_SLUB_DEBUG
2342 atomic_long_set(&n
->nr_slabs
, 0);
2343 atomic_long_set(&n
->total_objects
, 0);
2344 INIT_LIST_HEAD(&n
->full
);
2348 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2350 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2351 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2353 #ifdef CONFIG_CMPXCHG_LOCAL
2355 * Must align to double word boundary for the double cmpxchg instructions
2358 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
), 2 * sizeof(void *));
2360 /* Regular alignment is sufficient */
2361 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2367 init_kmem_cache_cpus(s
);
2372 static struct kmem_cache
*kmem_cache_node
;
2375 * No kmalloc_node yet so do it by hand. We know that this is the first
2376 * slab on the node for this slabcache. There are no concurrent accesses
2379 * Note that this function only works on the kmalloc_node_cache
2380 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2381 * memory on a fresh node that has no slab structures yet.
2383 static void early_kmem_cache_node_alloc(int node
)
2386 struct kmem_cache_node
*n
;
2387 unsigned long flags
;
2389 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2391 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2394 if (page_to_nid(page
) != node
) {
2395 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2397 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2398 "in order to be able to continue\n");
2403 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2405 kmem_cache_node
->node
[node
] = n
;
2406 #ifdef CONFIG_SLUB_DEBUG
2407 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2408 init_tracking(kmem_cache_node
, n
);
2410 init_kmem_cache_node(n
, kmem_cache_node
);
2411 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2414 * lockdep requires consistent irq usage for each lock
2415 * so even though there cannot be a race this early in
2416 * the boot sequence, we still disable irqs.
2418 local_irq_save(flags
);
2419 add_partial(n
, page
, 0);
2420 local_irq_restore(flags
);
2423 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2427 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2428 struct kmem_cache_node
*n
= s
->node
[node
];
2431 kmem_cache_free(kmem_cache_node
, n
);
2433 s
->node
[node
] = NULL
;
2437 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2441 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2442 struct kmem_cache_node
*n
;
2444 if (slab_state
== DOWN
) {
2445 early_kmem_cache_node_alloc(node
);
2448 n
= kmem_cache_alloc_node(kmem_cache_node
,
2452 free_kmem_cache_nodes(s
);
2457 init_kmem_cache_node(n
, s
);
2462 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2464 if (min
< MIN_PARTIAL
)
2466 else if (min
> MAX_PARTIAL
)
2468 s
->min_partial
= min
;
2472 * calculate_sizes() determines the order and the distribution of data within
2475 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2477 unsigned long flags
= s
->flags
;
2478 unsigned long size
= s
->objsize
;
2479 unsigned long align
= s
->align
;
2483 * Round up object size to the next word boundary. We can only
2484 * place the free pointer at word boundaries and this determines
2485 * the possible location of the free pointer.
2487 size
= ALIGN(size
, sizeof(void *));
2489 #ifdef CONFIG_SLUB_DEBUG
2491 * Determine if we can poison the object itself. If the user of
2492 * the slab may touch the object after free or before allocation
2493 * then we should never poison the object itself.
2495 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2497 s
->flags
|= __OBJECT_POISON
;
2499 s
->flags
&= ~__OBJECT_POISON
;
2503 * If we are Redzoning then check if there is some space between the
2504 * end of the object and the free pointer. If not then add an
2505 * additional word to have some bytes to store Redzone information.
2507 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2508 size
+= sizeof(void *);
2512 * With that we have determined the number of bytes in actual use
2513 * by the object. This is the potential offset to the free pointer.
2517 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2520 * Relocate free pointer after the object if it is not
2521 * permitted to overwrite the first word of the object on
2524 * This is the case if we do RCU, have a constructor or
2525 * destructor or are poisoning the objects.
2528 size
+= sizeof(void *);
2531 #ifdef CONFIG_SLUB_DEBUG
2532 if (flags
& SLAB_STORE_USER
)
2534 * Need to store information about allocs and frees after
2537 size
+= 2 * sizeof(struct track
);
2539 if (flags
& SLAB_RED_ZONE
)
2541 * Add some empty padding so that we can catch
2542 * overwrites from earlier objects rather than let
2543 * tracking information or the free pointer be
2544 * corrupted if a user writes before the start
2547 size
+= sizeof(void *);
2551 * Determine the alignment based on various parameters that the
2552 * user specified and the dynamic determination of cache line size
2555 align
= calculate_alignment(flags
, align
, s
->objsize
);
2559 * SLUB stores one object immediately after another beginning from
2560 * offset 0. In order to align the objects we have to simply size
2561 * each object to conform to the alignment.
2563 size
= ALIGN(size
, align
);
2565 if (forced_order
>= 0)
2566 order
= forced_order
;
2568 order
= calculate_order(size
, s
->reserved
);
2575 s
->allocflags
|= __GFP_COMP
;
2577 if (s
->flags
& SLAB_CACHE_DMA
)
2578 s
->allocflags
|= SLUB_DMA
;
2580 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2581 s
->allocflags
|= __GFP_RECLAIMABLE
;
2584 * Determine the number of objects per slab
2586 s
->oo
= oo_make(order
, size
, s
->reserved
);
2587 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
2588 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2591 return !!oo_objects(s
->oo
);
2595 static int kmem_cache_open(struct kmem_cache
*s
,
2596 const char *name
, size_t size
,
2597 size_t align
, unsigned long flags
,
2598 void (*ctor
)(void *))
2600 memset(s
, 0, kmem_size
);
2605 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2608 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
2609 s
->reserved
= sizeof(struct rcu_head
);
2611 if (!calculate_sizes(s
, -1))
2613 if (disable_higher_order_debug
) {
2615 * Disable debugging flags that store metadata if the min slab
2618 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2619 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2621 if (!calculate_sizes(s
, -1))
2627 * The larger the object size is, the more pages we want on the partial
2628 * list to avoid pounding the page allocator excessively.
2630 set_min_partial(s
, ilog2(s
->size
));
2633 s
->remote_node_defrag_ratio
= 1000;
2635 if (!init_kmem_cache_nodes(s
))
2638 if (alloc_kmem_cache_cpus(s
))
2641 free_kmem_cache_nodes(s
);
2643 if (flags
& SLAB_PANIC
)
2644 panic("Cannot create slab %s size=%lu realsize=%u "
2645 "order=%u offset=%u flags=%lx\n",
2646 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2652 * Determine the size of a slab object
2654 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2658 EXPORT_SYMBOL(kmem_cache_size
);
2660 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2663 #ifdef CONFIG_SLUB_DEBUG
2664 void *addr
= page_address(page
);
2666 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
2667 sizeof(long), GFP_ATOMIC
);
2670 slab_err(s
, page
, "%s", text
);
2672 for_each_free_object(p
, s
, page
->freelist
)
2673 set_bit(slab_index(p
, s
, addr
), map
);
2675 for_each_object(p
, s
, addr
, page
->objects
) {
2677 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2678 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2680 print_tracking(s
, p
);
2689 * Attempt to free all partial slabs on a node.
2691 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2693 unsigned long flags
;
2694 struct page
*page
, *h
;
2696 spin_lock_irqsave(&n
->list_lock
, flags
);
2697 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2699 __remove_partial(n
, page
);
2700 discard_slab(s
, page
);
2702 list_slab_objects(s
, page
,
2703 "Objects remaining on kmem_cache_close()");
2706 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2710 * Release all resources used by a slab cache.
2712 static inline int kmem_cache_close(struct kmem_cache
*s
)
2717 free_percpu(s
->cpu_slab
);
2718 /* Attempt to free all objects */
2719 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2720 struct kmem_cache_node
*n
= get_node(s
, node
);
2723 if (n
->nr_partial
|| slabs_node(s
, node
))
2726 free_kmem_cache_nodes(s
);
2731 * Close a cache and release the kmem_cache structure
2732 * (must be used for caches created using kmem_cache_create)
2734 void kmem_cache_destroy(struct kmem_cache
*s
)
2736 down_write(&slub_lock
);
2740 if (kmem_cache_close(s
)) {
2741 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2742 "still has objects.\n", s
->name
, __func__
);
2745 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2747 sysfs_slab_remove(s
);
2749 up_write(&slub_lock
);
2751 EXPORT_SYMBOL(kmem_cache_destroy
);
2753 /********************************************************************
2755 *******************************************************************/
2757 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
2758 EXPORT_SYMBOL(kmalloc_caches
);
2760 static struct kmem_cache
*kmem_cache
;
2762 #ifdef CONFIG_ZONE_DMA
2763 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
2766 static int __init
setup_slub_min_order(char *str
)
2768 get_option(&str
, &slub_min_order
);
2773 __setup("slub_min_order=", setup_slub_min_order
);
2775 static int __init
setup_slub_max_order(char *str
)
2777 get_option(&str
, &slub_max_order
);
2778 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2783 __setup("slub_max_order=", setup_slub_max_order
);
2785 static int __init
setup_slub_min_objects(char *str
)
2787 get_option(&str
, &slub_min_objects
);
2792 __setup("slub_min_objects=", setup_slub_min_objects
);
2794 static int __init
setup_slub_nomerge(char *str
)
2800 __setup("slub_nomerge", setup_slub_nomerge
);
2802 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
2803 int size
, unsigned int flags
)
2805 struct kmem_cache
*s
;
2807 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
2810 * This function is called with IRQs disabled during early-boot on
2811 * single CPU so there's no need to take slub_lock here.
2813 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2817 list_add(&s
->list
, &slab_caches
);
2821 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2826 * Conversion table for small slabs sizes / 8 to the index in the
2827 * kmalloc array. This is necessary for slabs < 192 since we have non power
2828 * of two cache sizes there. The size of larger slabs can be determined using
2831 static s8 size_index
[24] = {
2858 static inline int size_index_elem(size_t bytes
)
2860 return (bytes
- 1) / 8;
2863 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2869 return ZERO_SIZE_PTR
;
2871 index
= size_index
[size_index_elem(size
)];
2873 index
= fls(size
- 1);
2875 #ifdef CONFIG_ZONE_DMA
2876 if (unlikely((flags
& SLUB_DMA
)))
2877 return kmalloc_dma_caches
[index
];
2880 return kmalloc_caches
[index
];
2883 void *__kmalloc(size_t size
, gfp_t flags
)
2885 struct kmem_cache
*s
;
2888 if (unlikely(size
> SLUB_MAX_SIZE
))
2889 return kmalloc_large(size
, flags
);
2891 s
= get_slab(size
, flags
);
2893 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2896 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
2898 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2902 EXPORT_SYMBOL(__kmalloc
);
2905 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2910 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2911 page
= alloc_pages_node(node
, flags
, get_order(size
));
2913 ptr
= page_address(page
);
2915 kmemleak_alloc(ptr
, size
, 1, flags
);
2919 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2921 struct kmem_cache
*s
;
2924 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2925 ret
= kmalloc_large_node(size
, flags
, node
);
2927 trace_kmalloc_node(_RET_IP_
, ret
,
2928 size
, PAGE_SIZE
<< get_order(size
),
2934 s
= get_slab(size
, flags
);
2936 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2939 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2941 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2945 EXPORT_SYMBOL(__kmalloc_node
);
2948 size_t ksize(const void *object
)
2952 if (unlikely(object
== ZERO_SIZE_PTR
))
2955 page
= virt_to_head_page(object
);
2957 if (unlikely(!PageSlab(page
))) {
2958 WARN_ON(!PageCompound(page
));
2959 return PAGE_SIZE
<< compound_order(page
);
2962 return slab_ksize(page
->slab
);
2964 EXPORT_SYMBOL(ksize
);
2966 void kfree(const void *x
)
2969 void *object
= (void *)x
;
2971 trace_kfree(_RET_IP_
, x
);
2973 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2976 page
= virt_to_head_page(x
);
2977 if (unlikely(!PageSlab(page
))) {
2978 BUG_ON(!PageCompound(page
));
2983 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2985 EXPORT_SYMBOL(kfree
);
2988 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2989 * the remaining slabs by the number of items in use. The slabs with the
2990 * most items in use come first. New allocations will then fill those up
2991 * and thus they can be removed from the partial lists.
2993 * The slabs with the least items are placed last. This results in them
2994 * being allocated from last increasing the chance that the last objects
2995 * are freed in them.
2997 int kmem_cache_shrink(struct kmem_cache
*s
)
3001 struct kmem_cache_node
*n
;
3004 int objects
= oo_objects(s
->max
);
3005 struct list_head
*slabs_by_inuse
=
3006 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3007 unsigned long flags
;
3009 if (!slabs_by_inuse
)
3013 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3014 n
= get_node(s
, node
);
3019 for (i
= 0; i
< objects
; i
++)
3020 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3022 spin_lock_irqsave(&n
->list_lock
, flags
);
3025 * Build lists indexed by the items in use in each slab.
3027 * Note that concurrent frees may occur while we hold the
3028 * list_lock. page->inuse here is the upper limit.
3030 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3031 if (!page
->inuse
&& slab_trylock(page
)) {
3033 * Must hold slab lock here because slab_free
3034 * may have freed the last object and be
3035 * waiting to release the slab.
3037 __remove_partial(n
, page
);
3039 discard_slab(s
, page
);
3041 list_move(&page
->lru
,
3042 slabs_by_inuse
+ page
->inuse
);
3047 * Rebuild the partial list with the slabs filled up most
3048 * first and the least used slabs at the end.
3050 for (i
= objects
- 1; i
>= 0; i
--)
3051 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3053 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3056 kfree(slabs_by_inuse
);
3059 EXPORT_SYMBOL(kmem_cache_shrink
);
3061 #if defined(CONFIG_MEMORY_HOTPLUG)
3062 static int slab_mem_going_offline_callback(void *arg
)
3064 struct kmem_cache
*s
;
3066 down_read(&slub_lock
);
3067 list_for_each_entry(s
, &slab_caches
, list
)
3068 kmem_cache_shrink(s
);
3069 up_read(&slub_lock
);
3074 static void slab_mem_offline_callback(void *arg
)
3076 struct kmem_cache_node
*n
;
3077 struct kmem_cache
*s
;
3078 struct memory_notify
*marg
= arg
;
3081 offline_node
= marg
->status_change_nid
;
3084 * If the node still has available memory. we need kmem_cache_node
3087 if (offline_node
< 0)
3090 down_read(&slub_lock
);
3091 list_for_each_entry(s
, &slab_caches
, list
) {
3092 n
= get_node(s
, offline_node
);
3095 * if n->nr_slabs > 0, slabs still exist on the node
3096 * that is going down. We were unable to free them,
3097 * and offline_pages() function shouldn't call this
3098 * callback. So, we must fail.
3100 BUG_ON(slabs_node(s
, offline_node
));
3102 s
->node
[offline_node
] = NULL
;
3103 kmem_cache_free(kmem_cache_node
, n
);
3106 up_read(&slub_lock
);
3109 static int slab_mem_going_online_callback(void *arg
)
3111 struct kmem_cache_node
*n
;
3112 struct kmem_cache
*s
;
3113 struct memory_notify
*marg
= arg
;
3114 int nid
= marg
->status_change_nid
;
3118 * If the node's memory is already available, then kmem_cache_node is
3119 * already created. Nothing to do.
3125 * We are bringing a node online. No memory is available yet. We must
3126 * allocate a kmem_cache_node structure in order to bring the node
3129 down_read(&slub_lock
);
3130 list_for_each_entry(s
, &slab_caches
, list
) {
3132 * XXX: kmem_cache_alloc_node will fallback to other nodes
3133 * since memory is not yet available from the node that
3136 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3141 init_kmem_cache_node(n
, s
);
3145 up_read(&slub_lock
);
3149 static int slab_memory_callback(struct notifier_block
*self
,
3150 unsigned long action
, void *arg
)
3155 case MEM_GOING_ONLINE
:
3156 ret
= slab_mem_going_online_callback(arg
);
3158 case MEM_GOING_OFFLINE
:
3159 ret
= slab_mem_going_offline_callback(arg
);
3162 case MEM_CANCEL_ONLINE
:
3163 slab_mem_offline_callback(arg
);
3166 case MEM_CANCEL_OFFLINE
:
3170 ret
= notifier_from_errno(ret
);
3176 #endif /* CONFIG_MEMORY_HOTPLUG */
3178 /********************************************************************
3179 * Basic setup of slabs
3180 *******************************************************************/
3183 * Used for early kmem_cache structures that were allocated using
3184 * the page allocator
3187 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3191 list_add(&s
->list
, &slab_caches
);
3194 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3195 struct kmem_cache_node
*n
= get_node(s
, node
);
3199 list_for_each_entry(p
, &n
->partial
, lru
)
3202 #ifdef CONFIG_SLAB_DEBUG
3203 list_for_each_entry(p
, &n
->full
, lru
)
3210 void __init
kmem_cache_init(void)
3214 struct kmem_cache
*temp_kmem_cache
;
3216 struct kmem_cache
*temp_kmem_cache_node
;
3217 unsigned long kmalloc_size
;
3219 kmem_size
= offsetof(struct kmem_cache
, node
) +
3220 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3222 /* Allocate two kmem_caches from the page allocator */
3223 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3224 order
= get_order(2 * kmalloc_size
);
3225 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3228 * Must first have the slab cache available for the allocations of the
3229 * struct kmem_cache_node's. There is special bootstrap code in
3230 * kmem_cache_open for slab_state == DOWN.
3232 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3234 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3235 sizeof(struct kmem_cache_node
),
3236 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3238 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3240 /* Able to allocate the per node structures */
3241 slab_state
= PARTIAL
;
3243 temp_kmem_cache
= kmem_cache
;
3244 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3245 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3246 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3247 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3250 * Allocate kmem_cache_node properly from the kmem_cache slab.
3251 * kmem_cache_node is separately allocated so no need to
3252 * update any list pointers.
3254 temp_kmem_cache_node
= kmem_cache_node
;
3256 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3257 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3259 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3262 kmem_cache_bootstrap_fixup(kmem_cache
);
3264 /* Free temporary boot structure */
3265 free_pages((unsigned long)temp_kmem_cache
, order
);
3267 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3270 * Patch up the size_index table if we have strange large alignment
3271 * requirements for the kmalloc array. This is only the case for
3272 * MIPS it seems. The standard arches will not generate any code here.
3274 * Largest permitted alignment is 256 bytes due to the way we
3275 * handle the index determination for the smaller caches.
3277 * Make sure that nothing crazy happens if someone starts tinkering
3278 * around with ARCH_KMALLOC_MINALIGN
3280 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3281 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3283 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3284 int elem
= size_index_elem(i
);
3285 if (elem
>= ARRAY_SIZE(size_index
))
3287 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3290 if (KMALLOC_MIN_SIZE
== 64) {
3292 * The 96 byte size cache is not used if the alignment
3295 for (i
= 64 + 8; i
<= 96; i
+= 8)
3296 size_index
[size_index_elem(i
)] = 7;
3297 } else if (KMALLOC_MIN_SIZE
== 128) {
3299 * The 192 byte sized cache is not used if the alignment
3300 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3303 for (i
= 128 + 8; i
<= 192; i
+= 8)
3304 size_index
[size_index_elem(i
)] = 8;
3307 /* Caches that are not of the two-to-the-power-of size */
3308 if (KMALLOC_MIN_SIZE
<= 32) {
3309 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3313 if (KMALLOC_MIN_SIZE
<= 64) {
3314 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3318 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3319 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3325 /* Provide the correct kmalloc names now that the caches are up */
3326 if (KMALLOC_MIN_SIZE
<= 32) {
3327 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3328 BUG_ON(!kmalloc_caches
[1]->name
);
3331 if (KMALLOC_MIN_SIZE
<= 64) {
3332 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3333 BUG_ON(!kmalloc_caches
[2]->name
);
3336 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3337 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3340 kmalloc_caches
[i
]->name
= s
;
3344 register_cpu_notifier(&slab_notifier
);
3347 #ifdef CONFIG_ZONE_DMA
3348 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3349 struct kmem_cache
*s
= kmalloc_caches
[i
];
3352 char *name
= kasprintf(GFP_NOWAIT
,
3353 "dma-kmalloc-%d", s
->objsize
);
3356 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3357 s
->objsize
, SLAB_CACHE_DMA
);
3362 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3363 " CPUs=%d, Nodes=%d\n",
3364 caches
, cache_line_size(),
3365 slub_min_order
, slub_max_order
, slub_min_objects
,
3366 nr_cpu_ids
, nr_node_ids
);
3369 void __init
kmem_cache_init_late(void)
3374 * Find a mergeable slab cache
3376 static int slab_unmergeable(struct kmem_cache
*s
)
3378 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3385 * We may have set a slab to be unmergeable during bootstrap.
3387 if (s
->refcount
< 0)
3393 static struct kmem_cache
*find_mergeable(size_t size
,
3394 size_t align
, unsigned long flags
, const char *name
,
3395 void (*ctor
)(void *))
3397 struct kmem_cache
*s
;
3399 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3405 size
= ALIGN(size
, sizeof(void *));
3406 align
= calculate_alignment(flags
, align
, size
);
3407 size
= ALIGN(size
, align
);
3408 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3410 list_for_each_entry(s
, &slab_caches
, list
) {
3411 if (slab_unmergeable(s
))
3417 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3420 * Check if alignment is compatible.
3421 * Courtesy of Adrian Drzewiecki
3423 if ((s
->size
& ~(align
- 1)) != s
->size
)
3426 if (s
->size
- size
>= sizeof(void *))
3434 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3435 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3437 struct kmem_cache
*s
;
3443 down_write(&slub_lock
);
3444 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3448 * Adjust the object sizes so that we clear
3449 * the complete object on kzalloc.
3451 s
->objsize
= max(s
->objsize
, (int)size
);
3452 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3454 if (sysfs_slab_alias(s
, name
)) {
3458 up_write(&slub_lock
);
3462 n
= kstrdup(name
, GFP_KERNEL
);
3466 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3468 if (kmem_cache_open(s
, n
,
3469 size
, align
, flags
, ctor
)) {
3470 list_add(&s
->list
, &slab_caches
);
3471 if (sysfs_slab_add(s
)) {
3477 up_write(&slub_lock
);
3484 up_write(&slub_lock
);
3486 if (flags
& SLAB_PANIC
)
3487 panic("Cannot create slabcache %s\n", name
);
3492 EXPORT_SYMBOL(kmem_cache_create
);
3496 * Use the cpu notifier to insure that the cpu slabs are flushed when
3499 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3500 unsigned long action
, void *hcpu
)
3502 long cpu
= (long)hcpu
;
3503 struct kmem_cache
*s
;
3504 unsigned long flags
;
3507 case CPU_UP_CANCELED
:
3508 case CPU_UP_CANCELED_FROZEN
:
3510 case CPU_DEAD_FROZEN
:
3511 down_read(&slub_lock
);
3512 list_for_each_entry(s
, &slab_caches
, list
) {
3513 local_irq_save(flags
);
3514 __flush_cpu_slab(s
, cpu
);
3515 local_irq_restore(flags
);
3517 up_read(&slub_lock
);
3525 static struct notifier_block __cpuinitdata slab_notifier
= {
3526 .notifier_call
= slab_cpuup_callback
3531 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3533 struct kmem_cache
*s
;
3536 if (unlikely(size
> SLUB_MAX_SIZE
))
3537 return kmalloc_large(size
, gfpflags
);
3539 s
= get_slab(size
, gfpflags
);
3541 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3544 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
3546 /* Honor the call site pointer we recieved. */
3547 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3553 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3554 int node
, unsigned long caller
)
3556 struct kmem_cache
*s
;
3559 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3560 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3562 trace_kmalloc_node(caller
, ret
,
3563 size
, PAGE_SIZE
<< get_order(size
),
3569 s
= get_slab(size
, gfpflags
);
3571 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3574 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3576 /* Honor the call site pointer we recieved. */
3577 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3584 static int count_inuse(struct page
*page
)
3589 static int count_total(struct page
*page
)
3591 return page
->objects
;
3595 #ifdef CONFIG_SLUB_DEBUG
3596 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3600 void *addr
= page_address(page
);
3602 if (!check_slab(s
, page
) ||
3603 !on_freelist(s
, page
, NULL
))
3606 /* Now we know that a valid freelist exists */
3607 bitmap_zero(map
, page
->objects
);
3609 for_each_free_object(p
, s
, page
->freelist
) {
3610 set_bit(slab_index(p
, s
, addr
), map
);
3611 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
3615 for_each_object(p
, s
, addr
, page
->objects
)
3616 if (!test_bit(slab_index(p
, s
, addr
), map
))
3617 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
3622 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3625 if (slab_trylock(page
)) {
3626 validate_slab(s
, page
, map
);
3629 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3633 static int validate_slab_node(struct kmem_cache
*s
,
3634 struct kmem_cache_node
*n
, unsigned long *map
)
3636 unsigned long count
= 0;
3638 unsigned long flags
;
3640 spin_lock_irqsave(&n
->list_lock
, flags
);
3642 list_for_each_entry(page
, &n
->partial
, lru
) {
3643 validate_slab_slab(s
, page
, map
);
3646 if (count
!= n
->nr_partial
)
3647 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3648 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3650 if (!(s
->flags
& SLAB_STORE_USER
))
3653 list_for_each_entry(page
, &n
->full
, lru
) {
3654 validate_slab_slab(s
, page
, map
);
3657 if (count
!= atomic_long_read(&n
->nr_slabs
))
3658 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3659 "counter=%ld\n", s
->name
, count
,
3660 atomic_long_read(&n
->nr_slabs
));
3663 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3667 static long validate_slab_cache(struct kmem_cache
*s
)
3670 unsigned long count
= 0;
3671 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3672 sizeof(unsigned long), GFP_KERNEL
);
3678 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3679 struct kmem_cache_node
*n
= get_node(s
, node
);
3681 count
+= validate_slab_node(s
, n
, map
);
3687 * Generate lists of code addresses where slabcache objects are allocated
3692 unsigned long count
;
3699 DECLARE_BITMAP(cpus
, NR_CPUS
);
3705 unsigned long count
;
3706 struct location
*loc
;
3709 static void free_loc_track(struct loc_track
*t
)
3712 free_pages((unsigned long)t
->loc
,
3713 get_order(sizeof(struct location
) * t
->max
));
3716 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3721 order
= get_order(sizeof(struct location
) * max
);
3723 l
= (void *)__get_free_pages(flags
, order
);
3728 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3736 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3737 const struct track
*track
)
3739 long start
, end
, pos
;
3741 unsigned long caddr
;
3742 unsigned long age
= jiffies
- track
->when
;
3748 pos
= start
+ (end
- start
+ 1) / 2;
3751 * There is nothing at "end". If we end up there
3752 * we need to add something to before end.
3757 caddr
= t
->loc
[pos
].addr
;
3758 if (track
->addr
== caddr
) {
3764 if (age
< l
->min_time
)
3766 if (age
> l
->max_time
)
3769 if (track
->pid
< l
->min_pid
)
3770 l
->min_pid
= track
->pid
;
3771 if (track
->pid
> l
->max_pid
)
3772 l
->max_pid
= track
->pid
;
3774 cpumask_set_cpu(track
->cpu
,
3775 to_cpumask(l
->cpus
));
3777 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3781 if (track
->addr
< caddr
)
3788 * Not found. Insert new tracking element.
3790 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3796 (t
->count
- pos
) * sizeof(struct location
));
3799 l
->addr
= track
->addr
;
3803 l
->min_pid
= track
->pid
;
3804 l
->max_pid
= track
->pid
;
3805 cpumask_clear(to_cpumask(l
->cpus
));
3806 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3807 nodes_clear(l
->nodes
);
3808 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3812 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3813 struct page
*page
, enum track_item alloc
,
3816 void *addr
= page_address(page
);
3819 bitmap_zero(map
, page
->objects
);
3820 for_each_free_object(p
, s
, page
->freelist
)
3821 set_bit(slab_index(p
, s
, addr
), map
);
3823 for_each_object(p
, s
, addr
, page
->objects
)
3824 if (!test_bit(slab_index(p
, s
, addr
), map
))
3825 add_location(t
, s
, get_track(s
, p
, alloc
));
3828 static int list_locations(struct kmem_cache
*s
, char *buf
,
3829 enum track_item alloc
)
3833 struct loc_track t
= { 0, 0, NULL
};
3835 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3836 sizeof(unsigned long), GFP_KERNEL
);
3838 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3841 return sprintf(buf
, "Out of memory\n");
3843 /* Push back cpu slabs */
3846 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3847 struct kmem_cache_node
*n
= get_node(s
, node
);
3848 unsigned long flags
;
3851 if (!atomic_long_read(&n
->nr_slabs
))
3854 spin_lock_irqsave(&n
->list_lock
, flags
);
3855 list_for_each_entry(page
, &n
->partial
, lru
)
3856 process_slab(&t
, s
, page
, alloc
, map
);
3857 list_for_each_entry(page
, &n
->full
, lru
)
3858 process_slab(&t
, s
, page
, alloc
, map
);
3859 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3862 for (i
= 0; i
< t
.count
; i
++) {
3863 struct location
*l
= &t
.loc
[i
];
3865 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3867 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3870 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
3872 len
+= sprintf(buf
+ len
, "<not-available>");
3874 if (l
->sum_time
!= l
->min_time
) {
3875 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3877 (long)div_u64(l
->sum_time
, l
->count
),
3880 len
+= sprintf(buf
+ len
, " age=%ld",
3883 if (l
->min_pid
!= l
->max_pid
)
3884 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3885 l
->min_pid
, l
->max_pid
);
3887 len
+= sprintf(buf
+ len
, " pid=%ld",
3890 if (num_online_cpus() > 1 &&
3891 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3892 len
< PAGE_SIZE
- 60) {
3893 len
+= sprintf(buf
+ len
, " cpus=");
3894 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3895 to_cpumask(l
->cpus
));
3898 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3899 len
< PAGE_SIZE
- 60) {
3900 len
+= sprintf(buf
+ len
, " nodes=");
3901 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3905 len
+= sprintf(buf
+ len
, "\n");
3911 len
+= sprintf(buf
, "No data\n");
3916 #ifdef SLUB_RESILIENCY_TEST
3917 static void resiliency_test(void)
3921 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
3923 printk(KERN_ERR
"SLUB resiliency testing\n");
3924 printk(KERN_ERR
"-----------------------\n");
3925 printk(KERN_ERR
"A. Corruption after allocation\n");
3927 p
= kzalloc(16, GFP_KERNEL
);
3929 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3930 " 0x12->0x%p\n\n", p
+ 16);
3932 validate_slab_cache(kmalloc_caches
[4]);
3934 /* Hmmm... The next two are dangerous */
3935 p
= kzalloc(32, GFP_KERNEL
);
3936 p
[32 + sizeof(void *)] = 0x34;
3937 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3938 " 0x34 -> -0x%p\n", p
);
3940 "If allocated object is overwritten then not detectable\n\n");
3942 validate_slab_cache(kmalloc_caches
[5]);
3943 p
= kzalloc(64, GFP_KERNEL
);
3944 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3946 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3949 "If allocated object is overwritten then not detectable\n\n");
3950 validate_slab_cache(kmalloc_caches
[6]);
3952 printk(KERN_ERR
"\nB. Corruption after free\n");
3953 p
= kzalloc(128, GFP_KERNEL
);
3956 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3957 validate_slab_cache(kmalloc_caches
[7]);
3959 p
= kzalloc(256, GFP_KERNEL
);
3962 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3964 validate_slab_cache(kmalloc_caches
[8]);
3966 p
= kzalloc(512, GFP_KERNEL
);
3969 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3970 validate_slab_cache(kmalloc_caches
[9]);
3974 static void resiliency_test(void) {};
3979 enum slab_stat_type
{
3980 SL_ALL
, /* All slabs */
3981 SL_PARTIAL
, /* Only partially allocated slabs */
3982 SL_CPU
, /* Only slabs used for cpu caches */
3983 SL_OBJECTS
, /* Determine allocated objects not slabs */
3984 SL_TOTAL
/* Determine object capacity not slabs */
3987 #define SO_ALL (1 << SL_ALL)
3988 #define SO_PARTIAL (1 << SL_PARTIAL)
3989 #define SO_CPU (1 << SL_CPU)
3990 #define SO_OBJECTS (1 << SL_OBJECTS)
3991 #define SO_TOTAL (1 << SL_TOTAL)
3993 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3994 char *buf
, unsigned long flags
)
3996 unsigned long total
= 0;
3999 unsigned long *nodes
;
4000 unsigned long *per_cpu
;
4002 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4005 per_cpu
= nodes
+ nr_node_ids
;
4007 if (flags
& SO_CPU
) {
4010 for_each_possible_cpu(cpu
) {
4011 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4013 if (!c
|| c
->node
< 0)
4017 if (flags
& SO_TOTAL
)
4018 x
= c
->page
->objects
;
4019 else if (flags
& SO_OBJECTS
)
4025 nodes
[c
->node
] += x
;
4031 lock_memory_hotplug();
4032 #ifdef CONFIG_SLUB_DEBUG
4033 if (flags
& SO_ALL
) {
4034 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4035 struct kmem_cache_node
*n
= get_node(s
, node
);
4037 if (flags
& SO_TOTAL
)
4038 x
= atomic_long_read(&n
->total_objects
);
4039 else if (flags
& SO_OBJECTS
)
4040 x
= atomic_long_read(&n
->total_objects
) -
4041 count_partial(n
, count_free
);
4044 x
= atomic_long_read(&n
->nr_slabs
);
4051 if (flags
& SO_PARTIAL
) {
4052 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4053 struct kmem_cache_node
*n
= get_node(s
, node
);
4055 if (flags
& SO_TOTAL
)
4056 x
= count_partial(n
, count_total
);
4057 else if (flags
& SO_OBJECTS
)
4058 x
= count_partial(n
, count_inuse
);
4065 x
= sprintf(buf
, "%lu", total
);
4067 for_each_node_state(node
, N_NORMAL_MEMORY
)
4069 x
+= sprintf(buf
+ x
, " N%d=%lu",
4072 unlock_memory_hotplug();
4074 return x
+ sprintf(buf
+ x
, "\n");
4077 #ifdef CONFIG_SLUB_DEBUG
4078 static int any_slab_objects(struct kmem_cache
*s
)
4082 for_each_online_node(node
) {
4083 struct kmem_cache_node
*n
= get_node(s
, node
);
4088 if (atomic_long_read(&n
->total_objects
))
4095 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4096 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
4098 struct slab_attribute
{
4099 struct attribute attr
;
4100 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4101 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4104 #define SLAB_ATTR_RO(_name) \
4105 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4107 #define SLAB_ATTR(_name) \
4108 static struct slab_attribute _name##_attr = \
4109 __ATTR(_name, 0644, _name##_show, _name##_store)
4111 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4113 return sprintf(buf
, "%d\n", s
->size
);
4115 SLAB_ATTR_RO(slab_size
);
4117 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4119 return sprintf(buf
, "%d\n", s
->align
);
4121 SLAB_ATTR_RO(align
);
4123 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4125 return sprintf(buf
, "%d\n", s
->objsize
);
4127 SLAB_ATTR_RO(object_size
);
4129 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4131 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4133 SLAB_ATTR_RO(objs_per_slab
);
4135 static ssize_t
order_store(struct kmem_cache
*s
,
4136 const char *buf
, size_t length
)
4138 unsigned long order
;
4141 err
= strict_strtoul(buf
, 10, &order
);
4145 if (order
> slub_max_order
|| order
< slub_min_order
)
4148 calculate_sizes(s
, order
);
4152 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4154 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4158 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4160 return sprintf(buf
, "%lu\n", s
->min_partial
);
4163 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4169 err
= strict_strtoul(buf
, 10, &min
);
4173 set_min_partial(s
, min
);
4176 SLAB_ATTR(min_partial
);
4178 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4182 return sprintf(buf
, "%pS\n", s
->ctor
);
4186 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4188 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4190 SLAB_ATTR_RO(aliases
);
4192 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4194 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4196 SLAB_ATTR_RO(partial
);
4198 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4200 return show_slab_objects(s
, buf
, SO_CPU
);
4202 SLAB_ATTR_RO(cpu_slabs
);
4204 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4206 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4208 SLAB_ATTR_RO(objects
);
4210 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4212 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4214 SLAB_ATTR_RO(objects_partial
);
4216 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4218 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4221 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4222 const char *buf
, size_t length
)
4224 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4226 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4229 SLAB_ATTR(reclaim_account
);
4231 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4233 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4235 SLAB_ATTR_RO(hwcache_align
);
4237 #ifdef CONFIG_ZONE_DMA
4238 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4240 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4242 SLAB_ATTR_RO(cache_dma
);
4245 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4247 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4249 SLAB_ATTR_RO(destroy_by_rcu
);
4251 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4253 return sprintf(buf
, "%d\n", s
->reserved
);
4255 SLAB_ATTR_RO(reserved
);
4257 #ifdef CONFIG_SLUB_DEBUG
4258 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4260 return show_slab_objects(s
, buf
, SO_ALL
);
4262 SLAB_ATTR_RO(slabs
);
4264 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4266 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4268 SLAB_ATTR_RO(total_objects
);
4270 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4272 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4275 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4276 const char *buf
, size_t length
)
4278 s
->flags
&= ~SLAB_DEBUG_FREE
;
4280 s
->flags
|= SLAB_DEBUG_FREE
;
4283 SLAB_ATTR(sanity_checks
);
4285 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4287 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4290 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4293 s
->flags
&= ~SLAB_TRACE
;
4295 s
->flags
|= SLAB_TRACE
;
4300 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4302 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4305 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4306 const char *buf
, size_t length
)
4308 if (any_slab_objects(s
))
4311 s
->flags
&= ~SLAB_RED_ZONE
;
4313 s
->flags
|= SLAB_RED_ZONE
;
4314 calculate_sizes(s
, -1);
4317 SLAB_ATTR(red_zone
);
4319 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4321 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4324 static ssize_t
poison_store(struct kmem_cache
*s
,
4325 const char *buf
, size_t length
)
4327 if (any_slab_objects(s
))
4330 s
->flags
&= ~SLAB_POISON
;
4332 s
->flags
|= SLAB_POISON
;
4333 calculate_sizes(s
, -1);
4338 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4340 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4343 static ssize_t
store_user_store(struct kmem_cache
*s
,
4344 const char *buf
, size_t length
)
4346 if (any_slab_objects(s
))
4349 s
->flags
&= ~SLAB_STORE_USER
;
4351 s
->flags
|= SLAB_STORE_USER
;
4352 calculate_sizes(s
, -1);
4355 SLAB_ATTR(store_user
);
4357 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4362 static ssize_t
validate_store(struct kmem_cache
*s
,
4363 const char *buf
, size_t length
)
4367 if (buf
[0] == '1') {
4368 ret
= validate_slab_cache(s
);
4374 SLAB_ATTR(validate
);
4376 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4378 if (!(s
->flags
& SLAB_STORE_USER
))
4380 return list_locations(s
, buf
, TRACK_ALLOC
);
4382 SLAB_ATTR_RO(alloc_calls
);
4384 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4386 if (!(s
->flags
& SLAB_STORE_USER
))
4388 return list_locations(s
, buf
, TRACK_FREE
);
4390 SLAB_ATTR_RO(free_calls
);
4391 #endif /* CONFIG_SLUB_DEBUG */
4393 #ifdef CONFIG_FAILSLAB
4394 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4396 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4399 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4402 s
->flags
&= ~SLAB_FAILSLAB
;
4404 s
->flags
|= SLAB_FAILSLAB
;
4407 SLAB_ATTR(failslab
);
4410 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4415 static ssize_t
shrink_store(struct kmem_cache
*s
,
4416 const char *buf
, size_t length
)
4418 if (buf
[0] == '1') {
4419 int rc
= kmem_cache_shrink(s
);
4430 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4432 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4435 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4436 const char *buf
, size_t length
)
4438 unsigned long ratio
;
4441 err
= strict_strtoul(buf
, 10, &ratio
);
4446 s
->remote_node_defrag_ratio
= ratio
* 10;
4450 SLAB_ATTR(remote_node_defrag_ratio
);
4453 #ifdef CONFIG_SLUB_STATS
4454 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4456 unsigned long sum
= 0;
4459 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4464 for_each_online_cpu(cpu
) {
4465 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4471 len
= sprintf(buf
, "%lu", sum
);
4474 for_each_online_cpu(cpu
) {
4475 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4476 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4480 return len
+ sprintf(buf
+ len
, "\n");
4483 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4487 for_each_online_cpu(cpu
)
4488 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4491 #define STAT_ATTR(si, text) \
4492 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4494 return show_stat(s, buf, si); \
4496 static ssize_t text##_store(struct kmem_cache *s, \
4497 const char *buf, size_t length) \
4499 if (buf[0] != '0') \
4501 clear_stat(s, si); \
4506 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4507 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4508 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4509 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4510 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4511 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4512 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4513 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4514 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4515 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4516 STAT_ATTR(FREE_SLAB
, free_slab
);
4517 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4518 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4519 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4520 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4521 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4522 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4523 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4526 static struct attribute
*slab_attrs
[] = {
4527 &slab_size_attr
.attr
,
4528 &object_size_attr
.attr
,
4529 &objs_per_slab_attr
.attr
,
4531 &min_partial_attr
.attr
,
4533 &objects_partial_attr
.attr
,
4535 &cpu_slabs_attr
.attr
,
4539 &hwcache_align_attr
.attr
,
4540 &reclaim_account_attr
.attr
,
4541 &destroy_by_rcu_attr
.attr
,
4543 &reserved_attr
.attr
,
4544 #ifdef CONFIG_SLUB_DEBUG
4545 &total_objects_attr
.attr
,
4547 &sanity_checks_attr
.attr
,
4549 &red_zone_attr
.attr
,
4551 &store_user_attr
.attr
,
4552 &validate_attr
.attr
,
4553 &alloc_calls_attr
.attr
,
4554 &free_calls_attr
.attr
,
4556 #ifdef CONFIG_ZONE_DMA
4557 &cache_dma_attr
.attr
,
4560 &remote_node_defrag_ratio_attr
.attr
,
4562 #ifdef CONFIG_SLUB_STATS
4563 &alloc_fastpath_attr
.attr
,
4564 &alloc_slowpath_attr
.attr
,
4565 &free_fastpath_attr
.attr
,
4566 &free_slowpath_attr
.attr
,
4567 &free_frozen_attr
.attr
,
4568 &free_add_partial_attr
.attr
,
4569 &free_remove_partial_attr
.attr
,
4570 &alloc_from_partial_attr
.attr
,
4571 &alloc_slab_attr
.attr
,
4572 &alloc_refill_attr
.attr
,
4573 &free_slab_attr
.attr
,
4574 &cpuslab_flush_attr
.attr
,
4575 &deactivate_full_attr
.attr
,
4576 &deactivate_empty_attr
.attr
,
4577 &deactivate_to_head_attr
.attr
,
4578 &deactivate_to_tail_attr
.attr
,
4579 &deactivate_remote_frees_attr
.attr
,
4580 &order_fallback_attr
.attr
,
4582 #ifdef CONFIG_FAILSLAB
4583 &failslab_attr
.attr
,
4589 static struct attribute_group slab_attr_group
= {
4590 .attrs
= slab_attrs
,
4593 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4594 struct attribute
*attr
,
4597 struct slab_attribute
*attribute
;
4598 struct kmem_cache
*s
;
4601 attribute
= to_slab_attr(attr
);
4604 if (!attribute
->show
)
4607 err
= attribute
->show(s
, buf
);
4612 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4613 struct attribute
*attr
,
4614 const char *buf
, size_t len
)
4616 struct slab_attribute
*attribute
;
4617 struct kmem_cache
*s
;
4620 attribute
= to_slab_attr(attr
);
4623 if (!attribute
->store
)
4626 err
= attribute
->store(s
, buf
, len
);
4631 static void kmem_cache_release(struct kobject
*kobj
)
4633 struct kmem_cache
*s
= to_slab(kobj
);
4639 static const struct sysfs_ops slab_sysfs_ops
= {
4640 .show
= slab_attr_show
,
4641 .store
= slab_attr_store
,
4644 static struct kobj_type slab_ktype
= {
4645 .sysfs_ops
= &slab_sysfs_ops
,
4646 .release
= kmem_cache_release
4649 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4651 struct kobj_type
*ktype
= get_ktype(kobj
);
4653 if (ktype
== &slab_ktype
)
4658 static const struct kset_uevent_ops slab_uevent_ops
= {
4659 .filter
= uevent_filter
,
4662 static struct kset
*slab_kset
;
4664 #define ID_STR_LENGTH 64
4666 /* Create a unique string id for a slab cache:
4668 * Format :[flags-]size
4670 static char *create_unique_id(struct kmem_cache
*s
)
4672 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4679 * First flags affecting slabcache operations. We will only
4680 * get here for aliasable slabs so we do not need to support
4681 * too many flags. The flags here must cover all flags that
4682 * are matched during merging to guarantee that the id is
4685 if (s
->flags
& SLAB_CACHE_DMA
)
4687 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4689 if (s
->flags
& SLAB_DEBUG_FREE
)
4691 if (!(s
->flags
& SLAB_NOTRACK
))
4695 p
+= sprintf(p
, "%07d", s
->size
);
4696 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4700 static int sysfs_slab_add(struct kmem_cache
*s
)
4706 if (slab_state
< SYSFS
)
4707 /* Defer until later */
4710 unmergeable
= slab_unmergeable(s
);
4713 * Slabcache can never be merged so we can use the name proper.
4714 * This is typically the case for debug situations. In that
4715 * case we can catch duplicate names easily.
4717 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4721 * Create a unique name for the slab as a target
4724 name
= create_unique_id(s
);
4727 s
->kobj
.kset
= slab_kset
;
4728 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4730 kobject_put(&s
->kobj
);
4734 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4736 kobject_del(&s
->kobj
);
4737 kobject_put(&s
->kobj
);
4740 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4742 /* Setup first alias */
4743 sysfs_slab_alias(s
, s
->name
);
4749 static void sysfs_slab_remove(struct kmem_cache
*s
)
4751 if (slab_state
< SYSFS
)
4753 * Sysfs has not been setup yet so no need to remove the
4758 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4759 kobject_del(&s
->kobj
);
4760 kobject_put(&s
->kobj
);
4764 * Need to buffer aliases during bootup until sysfs becomes
4765 * available lest we lose that information.
4767 struct saved_alias
{
4768 struct kmem_cache
*s
;
4770 struct saved_alias
*next
;
4773 static struct saved_alias
*alias_list
;
4775 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4777 struct saved_alias
*al
;
4779 if (slab_state
== SYSFS
) {
4781 * If we have a leftover link then remove it.
4783 sysfs_remove_link(&slab_kset
->kobj
, name
);
4784 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4787 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4793 al
->next
= alias_list
;
4798 static int __init
slab_sysfs_init(void)
4800 struct kmem_cache
*s
;
4803 down_write(&slub_lock
);
4805 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4807 up_write(&slub_lock
);
4808 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4814 list_for_each_entry(s
, &slab_caches
, list
) {
4815 err
= sysfs_slab_add(s
);
4817 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4818 " to sysfs\n", s
->name
);
4821 while (alias_list
) {
4822 struct saved_alias
*al
= alias_list
;
4824 alias_list
= alias_list
->next
;
4825 err
= sysfs_slab_alias(al
->s
, al
->name
);
4827 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4828 " %s to sysfs\n", s
->name
);
4832 up_write(&slub_lock
);
4837 __initcall(slab_sysfs_init
);
4838 #endif /* CONFIG_SYSFS */
4841 * The /proc/slabinfo ABI
4843 #ifdef CONFIG_SLABINFO
4844 static void print_slabinfo_header(struct seq_file
*m
)
4846 seq_puts(m
, "slabinfo - version: 2.1\n");
4847 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4848 "<objperslab> <pagesperslab>");
4849 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4850 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4854 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4858 down_read(&slub_lock
);
4860 print_slabinfo_header(m
);
4862 return seq_list_start(&slab_caches
, *pos
);
4865 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4867 return seq_list_next(p
, &slab_caches
, pos
);
4870 static void s_stop(struct seq_file
*m
, void *p
)
4872 up_read(&slub_lock
);
4875 static int s_show(struct seq_file
*m
, void *p
)
4877 unsigned long nr_partials
= 0;
4878 unsigned long nr_slabs
= 0;
4879 unsigned long nr_inuse
= 0;
4880 unsigned long nr_objs
= 0;
4881 unsigned long nr_free
= 0;
4882 struct kmem_cache
*s
;
4885 s
= list_entry(p
, struct kmem_cache
, list
);
4887 for_each_online_node(node
) {
4888 struct kmem_cache_node
*n
= get_node(s
, node
);
4893 nr_partials
+= n
->nr_partial
;
4894 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4895 nr_objs
+= atomic_long_read(&n
->total_objects
);
4896 nr_free
+= count_partial(n
, count_free
);
4899 nr_inuse
= nr_objs
- nr_free
;
4901 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4902 nr_objs
, s
->size
, oo_objects(s
->oo
),
4903 (1 << oo_order(s
->oo
)));
4904 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4905 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4911 static const struct seq_operations slabinfo_op
= {
4918 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4920 return seq_open(file
, &slabinfo_op
);
4923 static const struct file_operations proc_slabinfo_operations
= {
4924 .open
= slabinfo_open
,
4926 .llseek
= seq_lseek
,
4927 .release
= seq_release
,
4930 static int __init
slab_proc_init(void)
4932 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4935 module_init(slab_proc_init
);
4936 #endif /* CONFIG_SLABINFO */