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 or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/kmemcheck.h>
23 #include <linux/cpu.h>
24 #include <linux/cpuset.h>
25 #include <linux/mempolicy.h>
26 #include <linux/ctype.h>
27 #include <linux/debugobjects.h>
28 #include <linux/kallsyms.h>
29 #include <linux/memory.h>
30 #include <linux/math64.h>
31 #include <linux/fault-inject.h>
32 #include <linux/stacktrace.h>
33 #include <linux/prefetch.h>
35 #include <trace/events/kmem.h>
39 * 1. slab_mutex (Global Mutex)
41 * 3. slab_lock(page) (Only on some arches and for debugging)
45 * The role of the slab_mutex is to protect the list of all the slabs
46 * and to synchronize major metadata changes to slab cache structures.
48 * The slab_lock is only used for debugging and on arches that do not
49 * have the ability to do a cmpxchg_double. It only protects the second
50 * double word in the page struct. Meaning
51 * A. page->freelist -> List of object free in a page
52 * B. page->counters -> Counters of objects
53 * C. page->frozen -> frozen state
55 * If a slab is frozen then it is exempt from list management. It is not
56 * on any list. The processor that froze the slab is the one who can
57 * perform list operations on the page. Other processors may put objects
58 * onto the freelist but the processor that froze the slab is the only
59 * one that can retrieve the objects from the page's freelist.
61 * The list_lock protects the partial and full list on each node and
62 * the partial slab counter. If taken then no new slabs may be added or
63 * removed from the lists nor make the number of partial slabs be modified.
64 * (Note that the total number of slabs is an atomic value that may be
65 * modified without taking the list lock).
67 * The list_lock is a centralized lock and thus we avoid taking it as
68 * much as possible. As long as SLUB does not have to handle partial
69 * slabs, operations can continue without any centralized lock. F.e.
70 * allocating a long series of objects that fill up slabs does not require
72 * Interrupts are disabled during allocation and deallocation in order to
73 * make the slab allocator safe to use in the context of an irq. In addition
74 * interrupts are disabled to ensure that the processor does not change
75 * while handling per_cpu slabs, due to kernel preemption.
77 * SLUB assigns one slab for allocation to each processor.
78 * Allocations only occur from these slabs called cpu slabs.
80 * Slabs with free elements are kept on a partial list and during regular
81 * operations no list for full slabs is used. If an object in a full slab is
82 * freed then the slab will show up again on the partial lists.
83 * We track full slabs for debugging purposes though because otherwise we
84 * cannot scan all objects.
86 * Slabs are freed when they become empty. Teardown and setup is
87 * minimal so we rely on the page allocators per cpu caches for
88 * fast frees and allocs.
90 * Overloading of page flags that are otherwise used for LRU management.
92 * PageActive The slab is frozen and exempt from list processing.
93 * This means that the slab is dedicated to a purpose
94 * such as satisfying allocations for a specific
95 * processor. Objects may be freed in the slab while
96 * it is frozen but slab_free will then skip the usual
97 * list operations. It is up to the processor holding
98 * the slab to integrate the slab into the slab lists
99 * when the slab is no longer needed.
101 * One use of this flag is to mark slabs that are
102 * used for allocations. Then such a slab becomes a cpu
103 * slab. The cpu slab may be equipped with an additional
104 * freelist that allows lockless access to
105 * free objects in addition to the regular freelist
106 * that requires the slab lock.
108 * PageError Slab requires special handling due to debug
109 * options set. This moves slab handling out of
110 * the fast path and disables lockless freelists.
113 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
114 SLAB_TRACE | SLAB_DEBUG_FREE)
116 static inline int kmem_cache_debug(struct kmem_cache
*s
)
118 #ifdef CONFIG_SLUB_DEBUG
119 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
126 * Issues still to be resolved:
128 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
130 * - Variable sizing of the per node arrays
133 /* Enable to test recovery from slab corruption on boot */
134 #undef SLUB_RESILIENCY_TEST
136 /* Enable to log cmpxchg failures */
137 #undef SLUB_DEBUG_CMPXCHG
140 * Mininum number of partial slabs. These will be left on the partial
141 * lists even if they are empty. kmem_cache_shrink may reclaim them.
143 #define MIN_PARTIAL 5
146 * Maximum number of desirable partial slabs.
147 * The existence of more partial slabs makes kmem_cache_shrink
148 * sort the partial list by the number of objects in the.
150 #define MAX_PARTIAL 10
152 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
153 SLAB_POISON | SLAB_STORE_USER)
156 * Debugging flags that require metadata to be stored in the slab. These get
157 * disabled when slub_debug=O is used and a cache's min order increases with
160 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
163 * Set of flags that will prevent slab merging
165 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
166 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
169 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
170 SLAB_CACHE_DMA | SLAB_NOTRACK)
173 #define OO_MASK ((1 << OO_SHIFT) - 1)
174 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
176 /* Internal SLUB flags */
177 #define __OBJECT_POISON 0x80000000UL /* Poison object */
178 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
180 static int kmem_size
= sizeof(struct kmem_cache
);
183 static struct notifier_block slab_notifier
;
187 * Tracking user of a slab.
189 #define TRACK_ADDRS_COUNT 16
191 unsigned long addr
; /* Called from address */
192 #ifdef CONFIG_STACKTRACE
193 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
195 int cpu
; /* Was running on cpu */
196 int pid
; /* Pid context */
197 unsigned long when
; /* When did the operation occur */
200 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
203 static int sysfs_slab_add(struct kmem_cache
*);
204 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
205 static void sysfs_slab_remove(struct kmem_cache
*);
208 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
209 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
211 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
219 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
221 #ifdef CONFIG_SLUB_STATS
222 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
226 /********************************************************************
227 * Core slab cache functions
228 *******************************************************************/
230 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
232 return s
->node
[node
];
235 /* Verify that a pointer has an address that is valid within a slab page */
236 static inline int check_valid_pointer(struct kmem_cache
*s
,
237 struct page
*page
, const void *object
)
244 base
= page_address(page
);
245 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
246 (object
- base
) % s
->size
) {
253 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
255 return *(void **)(object
+ s
->offset
);
258 static void prefetch_freepointer(const struct kmem_cache
*s
, void *object
)
260 prefetch(object
+ s
->offset
);
263 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
267 #ifdef CONFIG_DEBUG_PAGEALLOC
268 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
270 p
= get_freepointer(s
, object
);
275 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
277 *(void **)(object
+ s
->offset
) = fp
;
280 /* Loop over all objects in a slab */
281 #define for_each_object(__p, __s, __addr, __objects) \
282 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
285 /* Determine object index from a given position */
286 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
288 return (p
- addr
) / s
->size
;
291 static inline size_t slab_ksize(const struct kmem_cache
*s
)
293 #ifdef CONFIG_SLUB_DEBUG
295 * Debugging requires use of the padding between object
296 * and whatever may come after it.
298 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
299 return s
->object_size
;
303 * If we have the need to store the freelist pointer
304 * back there or track user information then we can
305 * only use the space before that information.
307 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
310 * Else we can use all the padding etc for the allocation
315 static inline int order_objects(int order
, unsigned long size
, int reserved
)
317 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
320 static inline struct kmem_cache_order_objects
oo_make(int order
,
321 unsigned long size
, int reserved
)
323 struct kmem_cache_order_objects x
= {
324 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
330 static inline int oo_order(struct kmem_cache_order_objects x
)
332 return x
.x
>> OO_SHIFT
;
335 static inline int oo_objects(struct kmem_cache_order_objects x
)
337 return x
.x
& OO_MASK
;
341 * Per slab locking using the pagelock
343 static __always_inline
void slab_lock(struct page
*page
)
345 bit_spin_lock(PG_locked
, &page
->flags
);
348 static __always_inline
void slab_unlock(struct page
*page
)
350 __bit_spin_unlock(PG_locked
, &page
->flags
);
353 /* Interrupts must be disabled (for the fallback code to work right) */
354 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
355 void *freelist_old
, unsigned long counters_old
,
356 void *freelist_new
, unsigned long counters_new
,
359 VM_BUG_ON(!irqs_disabled());
360 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
361 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
362 if (s
->flags
& __CMPXCHG_DOUBLE
) {
363 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
364 freelist_old
, counters_old
,
365 freelist_new
, counters_new
))
371 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
372 page
->freelist
= freelist_new
;
373 page
->counters
= counters_new
;
381 stat(s
, CMPXCHG_DOUBLE_FAIL
);
383 #ifdef SLUB_DEBUG_CMPXCHG
384 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
390 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
391 void *freelist_old
, unsigned long counters_old
,
392 void *freelist_new
, unsigned long counters_new
,
395 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
396 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
397 if (s
->flags
& __CMPXCHG_DOUBLE
) {
398 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
399 freelist_old
, counters_old
,
400 freelist_new
, counters_new
))
407 local_irq_save(flags
);
409 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
410 page
->freelist
= freelist_new
;
411 page
->counters
= counters_new
;
413 local_irq_restore(flags
);
417 local_irq_restore(flags
);
421 stat(s
, CMPXCHG_DOUBLE_FAIL
);
423 #ifdef SLUB_DEBUG_CMPXCHG
424 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
430 #ifdef CONFIG_SLUB_DEBUG
432 * Determine a map of object in use on a page.
434 * Node listlock must be held to guarantee that the page does
435 * not vanish from under us.
437 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
440 void *addr
= page_address(page
);
442 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
443 set_bit(slab_index(p
, s
, addr
), map
);
449 #ifdef CONFIG_SLUB_DEBUG_ON
450 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
452 static int slub_debug
;
455 static char *slub_debug_slabs
;
456 static int disable_higher_order_debug
;
461 static void print_section(char *text
, u8
*addr
, unsigned int length
)
463 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
467 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
468 enum track_item alloc
)
473 p
= object
+ s
->offset
+ sizeof(void *);
475 p
= object
+ s
->inuse
;
480 static void set_track(struct kmem_cache
*s
, void *object
,
481 enum track_item alloc
, unsigned long addr
)
483 struct track
*p
= get_track(s
, object
, alloc
);
486 #ifdef CONFIG_STACKTRACE
487 struct stack_trace trace
;
490 trace
.nr_entries
= 0;
491 trace
.max_entries
= TRACK_ADDRS_COUNT
;
492 trace
.entries
= p
->addrs
;
494 save_stack_trace(&trace
);
496 /* See rant in lockdep.c */
497 if (trace
.nr_entries
!= 0 &&
498 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
501 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
505 p
->cpu
= smp_processor_id();
506 p
->pid
= current
->pid
;
509 memset(p
, 0, sizeof(struct track
));
512 static void init_tracking(struct kmem_cache
*s
, void *object
)
514 if (!(s
->flags
& SLAB_STORE_USER
))
517 set_track(s
, object
, TRACK_FREE
, 0UL);
518 set_track(s
, object
, TRACK_ALLOC
, 0UL);
521 static void print_track(const char *s
, struct track
*t
)
526 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
527 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
528 #ifdef CONFIG_STACKTRACE
531 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
533 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
540 static void print_tracking(struct kmem_cache
*s
, void *object
)
542 if (!(s
->flags
& SLAB_STORE_USER
))
545 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
546 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
549 static void print_page_info(struct page
*page
)
551 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
552 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
556 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
562 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
564 printk(KERN_ERR
"========================================"
565 "=====================================\n");
566 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
567 printk(KERN_ERR
"----------------------------------------"
568 "-------------------------------------\n\n");
571 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
577 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
579 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
582 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
584 unsigned int off
; /* Offset of last byte */
585 u8
*addr
= page_address(page
);
587 print_tracking(s
, p
);
589 print_page_info(page
);
591 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
592 p
, p
- addr
, get_freepointer(s
, p
));
595 print_section("Bytes b4 ", p
- 16, 16);
597 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
599 if (s
->flags
& SLAB_RED_ZONE
)
600 print_section("Redzone ", p
+ s
->object_size
,
601 s
->inuse
- s
->object_size
);
604 off
= s
->offset
+ sizeof(void *);
608 if (s
->flags
& SLAB_STORE_USER
)
609 off
+= 2 * sizeof(struct track
);
612 /* Beginning of the filler is the free pointer */
613 print_section("Padding ", p
+ off
, s
->size
- off
);
618 static void object_err(struct kmem_cache
*s
, struct page
*page
,
619 u8
*object
, char *reason
)
621 slab_bug(s
, "%s", reason
);
622 print_trailer(s
, page
, object
);
625 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
631 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
633 slab_bug(s
, "%s", buf
);
634 print_page_info(page
);
638 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
642 if (s
->flags
& __OBJECT_POISON
) {
643 memset(p
, POISON_FREE
, s
->object_size
- 1);
644 p
[s
->object_size
- 1] = POISON_END
;
647 if (s
->flags
& SLAB_RED_ZONE
)
648 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
651 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
652 void *from
, void *to
)
654 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
655 memset(from
, data
, to
- from
);
658 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
659 u8
*object
, char *what
,
660 u8
*start
, unsigned int value
, unsigned int bytes
)
665 fault
= memchr_inv(start
, value
, bytes
);
670 while (end
> fault
&& end
[-1] == value
)
673 slab_bug(s
, "%s overwritten", what
);
674 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
675 fault
, end
- 1, fault
[0], value
);
676 print_trailer(s
, page
, object
);
678 restore_bytes(s
, what
, value
, fault
, end
);
686 * Bytes of the object to be managed.
687 * If the freepointer may overlay the object then the free
688 * pointer is the first word of the object.
690 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
693 * object + s->object_size
694 * Padding to reach word boundary. This is also used for Redzoning.
695 * Padding is extended by another word if Redzoning is enabled and
696 * object_size == inuse.
698 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
699 * 0xcc (RED_ACTIVE) for objects in use.
702 * Meta data starts here.
704 * A. Free pointer (if we cannot overwrite object on free)
705 * B. Tracking data for SLAB_STORE_USER
706 * C. Padding to reach required alignment boundary or at mininum
707 * one word if debugging is on to be able to detect writes
708 * before the word boundary.
710 * Padding is done using 0x5a (POISON_INUSE)
713 * Nothing is used beyond s->size.
715 * If slabcaches are merged then the object_size and inuse boundaries are mostly
716 * ignored. And therefore no slab options that rely on these boundaries
717 * may be used with merged slabcaches.
720 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
722 unsigned long off
= s
->inuse
; /* The end of info */
725 /* Freepointer is placed after the object. */
726 off
+= sizeof(void *);
728 if (s
->flags
& SLAB_STORE_USER
)
729 /* We also have user information there */
730 off
+= 2 * sizeof(struct track
);
735 return check_bytes_and_report(s
, page
, p
, "Object padding",
736 p
+ off
, POISON_INUSE
, s
->size
- off
);
739 /* Check the pad bytes at the end of a slab page */
740 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
748 if (!(s
->flags
& SLAB_POISON
))
751 start
= page_address(page
);
752 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
753 end
= start
+ length
;
754 remainder
= length
% s
->size
;
758 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
761 while (end
> fault
&& end
[-1] == POISON_INUSE
)
764 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
765 print_section("Padding ", end
- remainder
, remainder
);
767 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
771 static int check_object(struct kmem_cache
*s
, struct page
*page
,
772 void *object
, u8 val
)
775 u8
*endobject
= object
+ s
->object_size
;
777 if (s
->flags
& SLAB_RED_ZONE
) {
778 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
779 endobject
, val
, s
->inuse
- s
->object_size
))
782 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
783 check_bytes_and_report(s
, page
, p
, "Alignment padding",
784 endobject
, POISON_INUSE
, s
->inuse
- s
->object_size
);
788 if (s
->flags
& SLAB_POISON
) {
789 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
790 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
791 POISON_FREE
, s
->object_size
- 1) ||
792 !check_bytes_and_report(s
, page
, p
, "Poison",
793 p
+ s
->object_size
- 1, POISON_END
, 1)))
796 * check_pad_bytes cleans up on its own.
798 check_pad_bytes(s
, page
, p
);
801 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
803 * Object and freepointer overlap. Cannot check
804 * freepointer while object is allocated.
808 /* Check free pointer validity */
809 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
810 object_err(s
, page
, p
, "Freepointer corrupt");
812 * No choice but to zap it and thus lose the remainder
813 * of the free objects in this slab. May cause
814 * another error because the object count is now wrong.
816 set_freepointer(s
, p
, NULL
);
822 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
826 VM_BUG_ON(!irqs_disabled());
828 if (!PageSlab(page
)) {
829 slab_err(s
, page
, "Not a valid slab page");
833 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
834 if (page
->objects
> maxobj
) {
835 slab_err(s
, page
, "objects %u > max %u",
836 s
->name
, page
->objects
, maxobj
);
839 if (page
->inuse
> page
->objects
) {
840 slab_err(s
, page
, "inuse %u > max %u",
841 s
->name
, page
->inuse
, page
->objects
);
844 /* Slab_pad_check fixes things up after itself */
845 slab_pad_check(s
, page
);
850 * Determine if a certain object on a page is on the freelist. Must hold the
851 * slab lock to guarantee that the chains are in a consistent state.
853 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
858 unsigned long max_objects
;
861 while (fp
&& nr
<= page
->objects
) {
864 if (!check_valid_pointer(s
, page
, fp
)) {
866 object_err(s
, page
, object
,
867 "Freechain corrupt");
868 set_freepointer(s
, object
, NULL
);
871 slab_err(s
, page
, "Freepointer corrupt");
872 page
->freelist
= NULL
;
873 page
->inuse
= page
->objects
;
874 slab_fix(s
, "Freelist cleared");
880 fp
= get_freepointer(s
, object
);
884 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
885 if (max_objects
> MAX_OBJS_PER_PAGE
)
886 max_objects
= MAX_OBJS_PER_PAGE
;
888 if (page
->objects
!= max_objects
) {
889 slab_err(s
, page
, "Wrong number of objects. Found %d but "
890 "should be %d", page
->objects
, max_objects
);
891 page
->objects
= max_objects
;
892 slab_fix(s
, "Number of objects adjusted.");
894 if (page
->inuse
!= page
->objects
- nr
) {
895 slab_err(s
, page
, "Wrong object count. Counter is %d but "
896 "counted were %d", page
->inuse
, page
->objects
- nr
);
897 page
->inuse
= page
->objects
- nr
;
898 slab_fix(s
, "Object count adjusted.");
900 return search
== NULL
;
903 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
906 if (s
->flags
& SLAB_TRACE
) {
907 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
909 alloc
? "alloc" : "free",
914 print_section("Object ", (void *)object
, s
->object_size
);
921 * Hooks for other subsystems that check memory allocations. In a typical
922 * production configuration these hooks all should produce no code at all.
924 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
926 flags
&= gfp_allowed_mask
;
927 lockdep_trace_alloc(flags
);
928 might_sleep_if(flags
& __GFP_WAIT
);
930 return should_failslab(s
->object_size
, flags
, s
->flags
);
933 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
935 flags
&= gfp_allowed_mask
;
936 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
937 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
940 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
942 kmemleak_free_recursive(x
, s
->flags
);
945 * Trouble is that we may no longer disable interupts in the fast path
946 * So in order to make the debug calls that expect irqs to be
947 * disabled we need to disable interrupts temporarily.
949 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
953 local_irq_save(flags
);
954 kmemcheck_slab_free(s
, x
, s
->object_size
);
955 debug_check_no_locks_freed(x
, s
->object_size
);
956 local_irq_restore(flags
);
959 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
960 debug_check_no_obj_freed(x
, s
->object_size
);
964 * Tracking of fully allocated slabs for debugging purposes.
966 * list_lock must be held.
968 static void add_full(struct kmem_cache
*s
,
969 struct kmem_cache_node
*n
, struct page
*page
)
971 if (!(s
->flags
& SLAB_STORE_USER
))
974 list_add(&page
->lru
, &n
->full
);
978 * list_lock must be held.
980 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
982 if (!(s
->flags
& SLAB_STORE_USER
))
985 list_del(&page
->lru
);
988 /* Tracking of the number of slabs for debugging purposes */
989 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
991 struct kmem_cache_node
*n
= get_node(s
, node
);
993 return atomic_long_read(&n
->nr_slabs
);
996 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
998 return atomic_long_read(&n
->nr_slabs
);
1001 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1003 struct kmem_cache_node
*n
= get_node(s
, node
);
1006 * May be called early in order to allocate a slab for the
1007 * kmem_cache_node structure. Solve the chicken-egg
1008 * dilemma by deferring the increment of the count during
1009 * bootstrap (see early_kmem_cache_node_alloc).
1012 atomic_long_inc(&n
->nr_slabs
);
1013 atomic_long_add(objects
, &n
->total_objects
);
1016 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1018 struct kmem_cache_node
*n
= get_node(s
, node
);
1020 atomic_long_dec(&n
->nr_slabs
);
1021 atomic_long_sub(objects
, &n
->total_objects
);
1024 /* Object debug checks for alloc/free paths */
1025 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1028 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1031 init_object(s
, object
, SLUB_RED_INACTIVE
);
1032 init_tracking(s
, object
);
1035 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
1036 void *object
, unsigned long addr
)
1038 if (!check_slab(s
, page
))
1041 if (!check_valid_pointer(s
, page
, object
)) {
1042 object_err(s
, page
, object
, "Freelist Pointer check fails");
1046 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1049 /* Success perform special debug activities for allocs */
1050 if (s
->flags
& SLAB_STORE_USER
)
1051 set_track(s
, object
, TRACK_ALLOC
, addr
);
1052 trace(s
, page
, object
, 1);
1053 init_object(s
, object
, SLUB_RED_ACTIVE
);
1057 if (PageSlab(page
)) {
1059 * If this is a slab page then lets do the best we can
1060 * to avoid issues in the future. Marking all objects
1061 * as used avoids touching the remaining objects.
1063 slab_fix(s
, "Marking all objects used");
1064 page
->inuse
= page
->objects
;
1065 page
->freelist
= NULL
;
1070 static noinline
int free_debug_processing(struct kmem_cache
*s
,
1071 struct page
*page
, void *object
, unsigned long addr
)
1073 unsigned long flags
;
1076 local_irq_save(flags
);
1079 if (!check_slab(s
, page
))
1082 if (!check_valid_pointer(s
, page
, object
)) {
1083 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1087 if (on_freelist(s
, page
, object
)) {
1088 object_err(s
, page
, object
, "Object already free");
1092 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1095 if (unlikely(s
!= page
->slab
)) {
1096 if (!PageSlab(page
)) {
1097 slab_err(s
, page
, "Attempt to free object(0x%p) "
1098 "outside of slab", object
);
1099 } else if (!page
->slab
) {
1101 "SLUB <none>: no slab for object 0x%p.\n",
1105 object_err(s
, page
, object
,
1106 "page slab pointer corrupt.");
1110 if (s
->flags
& SLAB_STORE_USER
)
1111 set_track(s
, object
, TRACK_FREE
, addr
);
1112 trace(s
, page
, object
, 0);
1113 init_object(s
, object
, SLUB_RED_INACTIVE
);
1117 local_irq_restore(flags
);
1121 slab_fix(s
, "Object at 0x%p not freed", object
);
1125 static int __init
setup_slub_debug(char *str
)
1127 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1128 if (*str
++ != '=' || !*str
)
1130 * No options specified. Switch on full debugging.
1136 * No options but restriction on slabs. This means full
1137 * debugging for slabs matching a pattern.
1141 if (tolower(*str
) == 'o') {
1143 * Avoid enabling debugging on caches if its minimum order
1144 * would increase as a result.
1146 disable_higher_order_debug
= 1;
1153 * Switch off all debugging measures.
1158 * Determine which debug features should be switched on
1160 for (; *str
&& *str
!= ','; str
++) {
1161 switch (tolower(*str
)) {
1163 slub_debug
|= SLAB_DEBUG_FREE
;
1166 slub_debug
|= SLAB_RED_ZONE
;
1169 slub_debug
|= SLAB_POISON
;
1172 slub_debug
|= SLAB_STORE_USER
;
1175 slub_debug
|= SLAB_TRACE
;
1178 slub_debug
|= SLAB_FAILSLAB
;
1181 printk(KERN_ERR
"slub_debug option '%c' "
1182 "unknown. skipped\n", *str
);
1188 slub_debug_slabs
= str
+ 1;
1193 __setup("slub_debug", setup_slub_debug
);
1195 static unsigned long kmem_cache_flags(unsigned long object_size
,
1196 unsigned long flags
, const char *name
,
1197 void (*ctor
)(void *))
1200 * Enable debugging if selected on the kernel commandline.
1202 if (slub_debug
&& (!slub_debug_slabs
||
1203 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1204 flags
|= slub_debug
;
1209 static inline void setup_object_debug(struct kmem_cache
*s
,
1210 struct page
*page
, void *object
) {}
1212 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1213 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1215 static inline int free_debug_processing(struct kmem_cache
*s
,
1216 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1218 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1220 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1221 void *object
, u8 val
) { return 1; }
1222 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1223 struct page
*page
) {}
1224 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1225 static inline unsigned long kmem_cache_flags(unsigned long object_size
,
1226 unsigned long flags
, const char *name
,
1227 void (*ctor
)(void *))
1231 #define slub_debug 0
1233 #define disable_higher_order_debug 0
1235 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1237 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1239 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1241 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1244 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1247 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1250 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1252 #endif /* CONFIG_SLUB_DEBUG */
1255 * Slab allocation and freeing
1257 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1258 struct kmem_cache_order_objects oo
)
1260 int order
= oo_order(oo
);
1262 flags
|= __GFP_NOTRACK
;
1264 if (node
== NUMA_NO_NODE
)
1265 return alloc_pages(flags
, order
);
1267 return alloc_pages_exact_node(node
, flags
, order
);
1270 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1273 struct kmem_cache_order_objects oo
= s
->oo
;
1276 flags
&= gfp_allowed_mask
;
1278 if (flags
& __GFP_WAIT
)
1281 flags
|= s
->allocflags
;
1284 * Let the initial higher-order allocation fail under memory pressure
1285 * so we fall-back to the minimum order allocation.
1287 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1289 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1290 if (unlikely(!page
)) {
1293 * Allocation may have failed due to fragmentation.
1294 * Try a lower order alloc if possible
1296 page
= alloc_slab_page(flags
, node
, oo
);
1299 stat(s
, ORDER_FALLBACK
);
1302 if (kmemcheck_enabled
&& page
1303 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1304 int pages
= 1 << oo_order(oo
);
1306 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1309 * Objects from caches that have a constructor don't get
1310 * cleared when they're allocated, so we need to do it here.
1313 kmemcheck_mark_uninitialized_pages(page
, pages
);
1315 kmemcheck_mark_unallocated_pages(page
, pages
);
1318 if (flags
& __GFP_WAIT
)
1319 local_irq_disable();
1323 page
->objects
= oo_objects(oo
);
1324 mod_zone_page_state(page_zone(page
),
1325 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1326 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1332 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1335 setup_object_debug(s
, page
, object
);
1336 if (unlikely(s
->ctor
))
1340 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1347 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1349 page
= allocate_slab(s
,
1350 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1354 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1356 __SetPageSlab(page
);
1358 start
= page_address(page
);
1360 if (unlikely(s
->flags
& SLAB_POISON
))
1361 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1364 for_each_object(p
, s
, start
, page
->objects
) {
1365 setup_object(s
, page
, last
);
1366 set_freepointer(s
, last
, p
);
1369 setup_object(s
, page
, last
);
1370 set_freepointer(s
, last
, NULL
);
1372 page
->freelist
= start
;
1373 page
->inuse
= page
->objects
;
1379 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1381 int order
= compound_order(page
);
1382 int pages
= 1 << order
;
1384 if (kmem_cache_debug(s
)) {
1387 slab_pad_check(s
, page
);
1388 for_each_object(p
, s
, page_address(page
),
1390 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1393 kmemcheck_free_shadow(page
, compound_order(page
));
1395 mod_zone_page_state(page_zone(page
),
1396 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1397 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1400 __ClearPageSlab(page
);
1401 reset_page_mapcount(page
);
1402 if (current
->reclaim_state
)
1403 current
->reclaim_state
->reclaimed_slab
+= pages
;
1404 __free_pages(page
, order
);
1407 #define need_reserve_slab_rcu \
1408 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1410 static void rcu_free_slab(struct rcu_head
*h
)
1414 if (need_reserve_slab_rcu
)
1415 page
= virt_to_head_page(h
);
1417 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1419 __free_slab(page
->slab
, page
);
1422 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1424 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1425 struct rcu_head
*head
;
1427 if (need_reserve_slab_rcu
) {
1428 int order
= compound_order(page
);
1429 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1431 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1432 head
= page_address(page
) + offset
;
1435 * RCU free overloads the RCU head over the LRU
1437 head
= (void *)&page
->lru
;
1440 call_rcu(head
, rcu_free_slab
);
1442 __free_slab(s
, page
);
1445 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1447 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1452 * Management of partially allocated slabs.
1454 * list_lock must be held.
1456 static inline void add_partial(struct kmem_cache_node
*n
,
1457 struct page
*page
, int tail
)
1460 if (tail
== DEACTIVATE_TO_TAIL
)
1461 list_add_tail(&page
->lru
, &n
->partial
);
1463 list_add(&page
->lru
, &n
->partial
);
1467 * list_lock must be held.
1469 static inline void remove_partial(struct kmem_cache_node
*n
,
1472 list_del(&page
->lru
);
1477 * Remove slab from the partial list, freeze it and
1478 * return the pointer to the freelist.
1480 * Returns a list of objects or NULL if it fails.
1482 * Must hold list_lock since we modify the partial list.
1484 static inline void *acquire_slab(struct kmem_cache
*s
,
1485 struct kmem_cache_node
*n
, struct page
*page
,
1489 unsigned long counters
;
1493 * Zap the freelist and set the frozen bit.
1494 * The old freelist is the list of objects for the
1495 * per cpu allocation list.
1497 freelist
= page
->freelist
;
1498 counters
= page
->counters
;
1499 new.counters
= counters
;
1501 new.inuse
= page
->objects
;
1502 new.freelist
= NULL
;
1504 new.freelist
= freelist
;
1507 VM_BUG_ON(new.frozen
);
1510 if (!__cmpxchg_double_slab(s
, page
,
1512 new.freelist
, new.counters
,
1516 remove_partial(n
, page
);
1521 static int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1524 * Try to allocate a partial slab from a specific node.
1526 static void *get_partial_node(struct kmem_cache
*s
,
1527 struct kmem_cache_node
*n
, struct kmem_cache_cpu
*c
)
1529 struct page
*page
, *page2
;
1530 void *object
= NULL
;
1533 * Racy check. If we mistakenly see no partial slabs then we
1534 * just allocate an empty slab. If we mistakenly try to get a
1535 * partial slab and there is none available then get_partials()
1538 if (!n
|| !n
->nr_partial
)
1541 spin_lock(&n
->list_lock
);
1542 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1543 void *t
= acquire_slab(s
, n
, page
, object
== NULL
);
1551 stat(s
, ALLOC_FROM_PARTIAL
);
1553 available
= page
->objects
- page
->inuse
;
1555 available
= put_cpu_partial(s
, page
, 0);
1556 stat(s
, CPU_PARTIAL_NODE
);
1558 if (kmem_cache_debug(s
) || available
> s
->cpu_partial
/ 2)
1562 spin_unlock(&n
->list_lock
);
1567 * Get a page from somewhere. Search in increasing NUMA distances.
1569 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1570 struct kmem_cache_cpu
*c
)
1573 struct zonelist
*zonelist
;
1576 enum zone_type high_zoneidx
= gfp_zone(flags
);
1578 unsigned int cpuset_mems_cookie
;
1581 * The defrag ratio allows a configuration of the tradeoffs between
1582 * inter node defragmentation and node local allocations. A lower
1583 * defrag_ratio increases the tendency to do local allocations
1584 * instead of attempting to obtain partial slabs from other nodes.
1586 * If the defrag_ratio is set to 0 then kmalloc() always
1587 * returns node local objects. If the ratio is higher then kmalloc()
1588 * may return off node objects because partial slabs are obtained
1589 * from other nodes and filled up.
1591 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1592 * defrag_ratio = 1000) then every (well almost) allocation will
1593 * first attempt to defrag slab caches on other nodes. This means
1594 * scanning over all nodes to look for partial slabs which may be
1595 * expensive if we do it every time we are trying to find a slab
1596 * with available objects.
1598 if (!s
->remote_node_defrag_ratio
||
1599 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1603 cpuset_mems_cookie
= get_mems_allowed();
1604 zonelist
= node_zonelist(slab_node(), flags
);
1605 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1606 struct kmem_cache_node
*n
;
1608 n
= get_node(s
, zone_to_nid(zone
));
1610 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1611 n
->nr_partial
> s
->min_partial
) {
1612 object
= get_partial_node(s
, n
, c
);
1615 * Return the object even if
1616 * put_mems_allowed indicated that
1617 * the cpuset mems_allowed was
1618 * updated in parallel. It's a
1619 * harmless race between the alloc
1620 * and the cpuset update.
1622 put_mems_allowed(cpuset_mems_cookie
);
1627 } while (!put_mems_allowed(cpuset_mems_cookie
));
1633 * Get a partial page, lock it and return it.
1635 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1636 struct kmem_cache_cpu
*c
)
1639 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1641 object
= get_partial_node(s
, get_node(s
, searchnode
), c
);
1642 if (object
|| node
!= NUMA_NO_NODE
)
1645 return get_any_partial(s
, flags
, c
);
1648 #ifdef CONFIG_PREEMPT
1650 * Calculate the next globally unique transaction for disambiguiation
1651 * during cmpxchg. The transactions start with the cpu number and are then
1652 * incremented by CONFIG_NR_CPUS.
1654 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1657 * No preemption supported therefore also no need to check for
1663 static inline unsigned long next_tid(unsigned long tid
)
1665 return tid
+ TID_STEP
;
1668 static inline unsigned int tid_to_cpu(unsigned long tid
)
1670 return tid
% TID_STEP
;
1673 static inline unsigned long tid_to_event(unsigned long tid
)
1675 return tid
/ TID_STEP
;
1678 static inline unsigned int init_tid(int cpu
)
1683 static inline void note_cmpxchg_failure(const char *n
,
1684 const struct kmem_cache
*s
, unsigned long tid
)
1686 #ifdef SLUB_DEBUG_CMPXCHG
1687 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1689 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1691 #ifdef CONFIG_PREEMPT
1692 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1693 printk("due to cpu change %d -> %d\n",
1694 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1697 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1698 printk("due to cpu running other code. Event %ld->%ld\n",
1699 tid_to_event(tid
), tid_to_event(actual_tid
));
1701 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1702 actual_tid
, tid
, next_tid(tid
));
1704 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1707 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1711 for_each_possible_cpu(cpu
)
1712 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1716 * Remove the cpu slab
1718 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
, void *freelist
)
1720 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1721 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1723 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1725 int tail
= DEACTIVATE_TO_HEAD
;
1729 if (page
->freelist
) {
1730 stat(s
, DEACTIVATE_REMOTE_FREES
);
1731 tail
= DEACTIVATE_TO_TAIL
;
1735 * Stage one: Free all available per cpu objects back
1736 * to the page freelist while it is still frozen. Leave the
1739 * There is no need to take the list->lock because the page
1742 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1744 unsigned long counters
;
1747 prior
= page
->freelist
;
1748 counters
= page
->counters
;
1749 set_freepointer(s
, freelist
, prior
);
1750 new.counters
= counters
;
1752 VM_BUG_ON(!new.frozen
);
1754 } while (!__cmpxchg_double_slab(s
, page
,
1756 freelist
, new.counters
,
1757 "drain percpu freelist"));
1759 freelist
= nextfree
;
1763 * Stage two: Ensure that the page is unfrozen while the
1764 * list presence reflects the actual number of objects
1767 * We setup the list membership and then perform a cmpxchg
1768 * with the count. If there is a mismatch then the page
1769 * is not unfrozen but the page is on the wrong list.
1771 * Then we restart the process which may have to remove
1772 * the page from the list that we just put it on again
1773 * because the number of objects in the slab may have
1778 old
.freelist
= page
->freelist
;
1779 old
.counters
= page
->counters
;
1780 VM_BUG_ON(!old
.frozen
);
1782 /* Determine target state of the slab */
1783 new.counters
= old
.counters
;
1786 set_freepointer(s
, freelist
, old
.freelist
);
1787 new.freelist
= freelist
;
1789 new.freelist
= old
.freelist
;
1793 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1795 else if (new.freelist
) {
1800 * Taking the spinlock removes the possiblity
1801 * that acquire_slab() will see a slab page that
1804 spin_lock(&n
->list_lock
);
1808 if (kmem_cache_debug(s
) && !lock
) {
1811 * This also ensures that the scanning of full
1812 * slabs from diagnostic functions will not see
1815 spin_lock(&n
->list_lock
);
1823 remove_partial(n
, page
);
1825 else if (l
== M_FULL
)
1827 remove_full(s
, page
);
1829 if (m
== M_PARTIAL
) {
1831 add_partial(n
, page
, tail
);
1834 } else if (m
== M_FULL
) {
1836 stat(s
, DEACTIVATE_FULL
);
1837 add_full(s
, n
, page
);
1843 if (!__cmpxchg_double_slab(s
, page
,
1844 old
.freelist
, old
.counters
,
1845 new.freelist
, new.counters
,
1850 spin_unlock(&n
->list_lock
);
1853 stat(s
, DEACTIVATE_EMPTY
);
1854 discard_slab(s
, page
);
1860 * Unfreeze all the cpu partial slabs.
1862 * This function must be called with interrupt disabled.
1864 static void unfreeze_partials(struct kmem_cache
*s
)
1866 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1867 struct kmem_cache_cpu
*c
= this_cpu_ptr(s
->cpu_slab
);
1868 struct page
*page
, *discard_page
= NULL
;
1870 while ((page
= c
->partial
)) {
1874 c
->partial
= page
->next
;
1876 n2
= get_node(s
, page_to_nid(page
));
1879 spin_unlock(&n
->list_lock
);
1882 spin_lock(&n
->list_lock
);
1887 old
.freelist
= page
->freelist
;
1888 old
.counters
= page
->counters
;
1889 VM_BUG_ON(!old
.frozen
);
1891 new.counters
= old
.counters
;
1892 new.freelist
= old
.freelist
;
1896 } while (!__cmpxchg_double_slab(s
, page
,
1897 old
.freelist
, old
.counters
,
1898 new.freelist
, new.counters
,
1899 "unfreezing slab"));
1901 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
)) {
1902 page
->next
= discard_page
;
1903 discard_page
= page
;
1905 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1906 stat(s
, FREE_ADD_PARTIAL
);
1911 spin_unlock(&n
->list_lock
);
1913 while (discard_page
) {
1914 page
= discard_page
;
1915 discard_page
= discard_page
->next
;
1917 stat(s
, DEACTIVATE_EMPTY
);
1918 discard_slab(s
, page
);
1924 * Put a page that was just frozen (in __slab_free) into a partial page
1925 * slot if available. This is done without interrupts disabled and without
1926 * preemption disabled. The cmpxchg is racy and may put the partial page
1927 * onto a random cpus partial slot.
1929 * If we did not find a slot then simply move all the partials to the
1930 * per node partial list.
1932 int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1934 struct page
*oldpage
;
1941 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1944 pobjects
= oldpage
->pobjects
;
1945 pages
= oldpage
->pages
;
1946 if (drain
&& pobjects
> s
->cpu_partial
) {
1947 unsigned long flags
;
1949 * partial array is full. Move the existing
1950 * set to the per node partial list.
1952 local_irq_save(flags
);
1953 unfreeze_partials(s
);
1954 local_irq_restore(flags
);
1957 stat(s
, CPU_PARTIAL_DRAIN
);
1962 pobjects
+= page
->objects
- page
->inuse
;
1964 page
->pages
= pages
;
1965 page
->pobjects
= pobjects
;
1966 page
->next
= oldpage
;
1968 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
) != oldpage
);
1972 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1974 stat(s
, CPUSLAB_FLUSH
);
1975 deactivate_slab(s
, c
->page
, c
->freelist
);
1977 c
->tid
= next_tid(c
->tid
);
1985 * Called from IPI handler with interrupts disabled.
1987 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1989 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1995 unfreeze_partials(s
);
1999 static void flush_cpu_slab(void *d
)
2001 struct kmem_cache
*s
= d
;
2003 __flush_cpu_slab(s
, smp_processor_id());
2006 static bool has_cpu_slab(int cpu
, void *info
)
2008 struct kmem_cache
*s
= info
;
2009 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2011 return c
->page
|| c
->partial
;
2014 static void flush_all(struct kmem_cache
*s
)
2016 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2020 * Check if the objects in a per cpu structure fit numa
2021 * locality expectations.
2023 static inline int node_match(struct page
*page
, int node
)
2026 if (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
)
2032 static int count_free(struct page
*page
)
2034 return page
->objects
- page
->inuse
;
2037 static unsigned long count_partial(struct kmem_cache_node
*n
,
2038 int (*get_count
)(struct page
*))
2040 unsigned long flags
;
2041 unsigned long x
= 0;
2044 spin_lock_irqsave(&n
->list_lock
, flags
);
2045 list_for_each_entry(page
, &n
->partial
, lru
)
2046 x
+= get_count(page
);
2047 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2051 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2053 #ifdef CONFIG_SLUB_DEBUG
2054 return atomic_long_read(&n
->total_objects
);
2060 static noinline
void
2061 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2066 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2068 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2069 "default order: %d, min order: %d\n", s
->name
, s
->object_size
,
2070 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2072 if (oo_order(s
->min
) > get_order(s
->object_size
))
2073 printk(KERN_WARNING
" %s debugging increased min order, use "
2074 "slub_debug=O to disable.\n", s
->name
);
2076 for_each_online_node(node
) {
2077 struct kmem_cache_node
*n
= get_node(s
, node
);
2078 unsigned long nr_slabs
;
2079 unsigned long nr_objs
;
2080 unsigned long nr_free
;
2085 nr_free
= count_partial(n
, count_free
);
2086 nr_slabs
= node_nr_slabs(n
);
2087 nr_objs
= node_nr_objs(n
);
2090 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2091 node
, nr_slabs
, nr_objs
, nr_free
);
2095 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2096 int node
, struct kmem_cache_cpu
**pc
)
2099 struct kmem_cache_cpu
*c
= *pc
;
2102 freelist
= get_partial(s
, flags
, node
, c
);
2107 page
= new_slab(s
, flags
, node
);
2109 c
= __this_cpu_ptr(s
->cpu_slab
);
2114 * No other reference to the page yet so we can
2115 * muck around with it freely without cmpxchg
2117 freelist
= page
->freelist
;
2118 page
->freelist
= NULL
;
2120 stat(s
, ALLOC_SLAB
);
2130 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2131 * or deactivate the page.
2133 * The page is still frozen if the return value is not NULL.
2135 * If this function returns NULL then the page has been unfrozen.
2137 * This function must be called with interrupt disabled.
2139 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2142 unsigned long counters
;
2146 freelist
= page
->freelist
;
2147 counters
= page
->counters
;
2149 new.counters
= counters
;
2150 VM_BUG_ON(!new.frozen
);
2152 new.inuse
= page
->objects
;
2153 new.frozen
= freelist
!= NULL
;
2155 } while (!__cmpxchg_double_slab(s
, page
,
2164 * Slow path. The lockless freelist is empty or we need to perform
2167 * Processing is still very fast if new objects have been freed to the
2168 * regular freelist. In that case we simply take over the regular freelist
2169 * as the lockless freelist and zap the regular freelist.
2171 * If that is not working then we fall back to the partial lists. We take the
2172 * first element of the freelist as the object to allocate now and move the
2173 * rest of the freelist to the lockless freelist.
2175 * And if we were unable to get a new slab from the partial slab lists then
2176 * we need to allocate a new slab. This is the slowest path since it involves
2177 * a call to the page allocator and the setup of a new slab.
2179 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2180 unsigned long addr
, struct kmem_cache_cpu
*c
)
2184 unsigned long flags
;
2186 local_irq_save(flags
);
2187 #ifdef CONFIG_PREEMPT
2189 * We may have been preempted and rescheduled on a different
2190 * cpu before disabling interrupts. Need to reload cpu area
2193 c
= this_cpu_ptr(s
->cpu_slab
);
2201 if (unlikely(!node_match(page
, node
))) {
2202 stat(s
, ALLOC_NODE_MISMATCH
);
2203 deactivate_slab(s
, page
, c
->freelist
);
2209 /* must check again c->freelist in case of cpu migration or IRQ */
2210 freelist
= c
->freelist
;
2214 stat(s
, ALLOC_SLOWPATH
);
2216 freelist
= get_freelist(s
, page
);
2220 stat(s
, DEACTIVATE_BYPASS
);
2224 stat(s
, ALLOC_REFILL
);
2228 * freelist is pointing to the list of objects to be used.
2229 * page is pointing to the page from which the objects are obtained.
2230 * That page must be frozen for per cpu allocations to work.
2232 VM_BUG_ON(!c
->page
->frozen
);
2233 c
->freelist
= get_freepointer(s
, freelist
);
2234 c
->tid
= next_tid(c
->tid
);
2235 local_irq_restore(flags
);
2241 page
= c
->page
= c
->partial
;
2242 c
->partial
= page
->next
;
2243 stat(s
, CPU_PARTIAL_ALLOC
);
2248 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2250 if (unlikely(!freelist
)) {
2251 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2252 slab_out_of_memory(s
, gfpflags
, node
);
2254 local_irq_restore(flags
);
2259 if (likely(!kmem_cache_debug(s
)))
2262 /* Only entered in the debug case */
2263 if (!alloc_debug_processing(s
, page
, freelist
, addr
))
2264 goto new_slab
; /* Slab failed checks. Next slab needed */
2266 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2269 local_irq_restore(flags
);
2274 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2275 * have the fastpath folded into their functions. So no function call
2276 * overhead for requests that can be satisfied on the fastpath.
2278 * The fastpath works by first checking if the lockless freelist can be used.
2279 * If not then __slab_alloc is called for slow processing.
2281 * Otherwise we can simply pick the next object from the lockless free list.
2283 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2284 gfp_t gfpflags
, int node
, unsigned long addr
)
2287 struct kmem_cache_cpu
*c
;
2291 if (slab_pre_alloc_hook(s
, gfpflags
))
2297 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2298 * enabled. We may switch back and forth between cpus while
2299 * reading from one cpu area. That does not matter as long
2300 * as we end up on the original cpu again when doing the cmpxchg.
2302 c
= __this_cpu_ptr(s
->cpu_slab
);
2305 * The transaction ids are globally unique per cpu and per operation on
2306 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2307 * occurs on the right processor and that there was no operation on the
2308 * linked list in between.
2313 object
= c
->freelist
;
2315 if (unlikely(!object
|| !node_match(page
, node
)))
2317 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2320 void *next_object
= get_freepointer_safe(s
, object
);
2323 * The cmpxchg will only match if there was no additional
2324 * operation and if we are on the right processor.
2326 * The cmpxchg does the following atomically (without lock semantics!)
2327 * 1. Relocate first pointer to the current per cpu area.
2328 * 2. Verify that tid and freelist have not been changed
2329 * 3. If they were not changed replace tid and freelist
2331 * Since this is without lock semantics the protection is only against
2332 * code executing on this cpu *not* from access by other cpus.
2334 if (unlikely(!this_cpu_cmpxchg_double(
2335 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2337 next_object
, next_tid(tid
)))) {
2339 note_cmpxchg_failure("slab_alloc", s
, tid
);
2342 prefetch_freepointer(s
, next_object
);
2343 stat(s
, ALLOC_FASTPATH
);
2346 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2347 memset(object
, 0, s
->object_size
);
2349 slab_post_alloc_hook(s
, gfpflags
, object
);
2354 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2356 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2358 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
, s
->size
, gfpflags
);
2362 EXPORT_SYMBOL(kmem_cache_alloc
);
2364 #ifdef CONFIG_TRACING
2365 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2367 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2368 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2371 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2373 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2375 void *ret
= kmalloc_order(size
, flags
, order
);
2376 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2379 EXPORT_SYMBOL(kmalloc_order_trace
);
2383 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2385 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2387 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2388 s
->object_size
, s
->size
, gfpflags
, node
);
2392 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2394 #ifdef CONFIG_TRACING
2395 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2397 int node
, size_t size
)
2399 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2401 trace_kmalloc_node(_RET_IP_
, ret
,
2402 size
, s
->size
, gfpflags
, node
);
2405 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2410 * Slow patch handling. This may still be called frequently since objects
2411 * have a longer lifetime than the cpu slabs in most processing loads.
2413 * So we still attempt to reduce cache line usage. Just take the slab
2414 * lock and free the item. If there is no additional partial page
2415 * handling required then we can return immediately.
2417 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2418 void *x
, unsigned long addr
)
2421 void **object
= (void *)x
;
2425 unsigned long counters
;
2426 struct kmem_cache_node
*n
= NULL
;
2427 unsigned long uninitialized_var(flags
);
2429 stat(s
, FREE_SLOWPATH
);
2431 if (kmem_cache_debug(s
) && !free_debug_processing(s
, page
, x
, addr
))
2435 prior
= page
->freelist
;
2436 counters
= page
->counters
;
2437 set_freepointer(s
, object
, prior
);
2438 new.counters
= counters
;
2439 was_frozen
= new.frozen
;
2441 if ((!new.inuse
|| !prior
) && !was_frozen
&& !n
) {
2443 if (!kmem_cache_debug(s
) && !prior
)
2446 * Slab was on no list before and will be partially empty
2447 * We can defer the list move and instead freeze it.
2451 else { /* Needs to be taken off a list */
2453 n
= get_node(s
, page_to_nid(page
));
2455 * Speculatively acquire the list_lock.
2456 * If the cmpxchg does not succeed then we may
2457 * drop the list_lock without any processing.
2459 * Otherwise the list_lock will synchronize with
2460 * other processors updating the list of slabs.
2462 spin_lock_irqsave(&n
->list_lock
, flags
);
2468 } while (!cmpxchg_double_slab(s
, page
,
2470 object
, new.counters
,
2476 * If we just froze the page then put it onto the
2477 * per cpu partial list.
2479 if (new.frozen
&& !was_frozen
) {
2480 put_cpu_partial(s
, page
, 1);
2481 stat(s
, CPU_PARTIAL_FREE
);
2484 * The list lock was not taken therefore no list
2485 * activity can be necessary.
2488 stat(s
, FREE_FROZEN
);
2493 * was_frozen may have been set after we acquired the list_lock in
2494 * an earlier loop. So we need to check it here again.
2497 stat(s
, FREE_FROZEN
);
2499 if (unlikely(!inuse
&& n
->nr_partial
> s
->min_partial
))
2503 * Objects left in the slab. If it was not on the partial list before
2506 if (unlikely(!prior
)) {
2507 remove_full(s
, page
);
2508 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2509 stat(s
, FREE_ADD_PARTIAL
);
2512 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2518 * Slab on the partial list.
2520 remove_partial(n
, page
);
2521 stat(s
, FREE_REMOVE_PARTIAL
);
2523 /* Slab must be on the full list */
2524 remove_full(s
, page
);
2526 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2528 discard_slab(s
, page
);
2532 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2533 * can perform fastpath freeing without additional function calls.
2535 * The fastpath is only possible if we are freeing to the current cpu slab
2536 * of this processor. This typically the case if we have just allocated
2539 * If fastpath is not possible then fall back to __slab_free where we deal
2540 * with all sorts of special processing.
2542 static __always_inline
void slab_free(struct kmem_cache
*s
,
2543 struct page
*page
, void *x
, unsigned long addr
)
2545 void **object
= (void *)x
;
2546 struct kmem_cache_cpu
*c
;
2549 slab_free_hook(s
, x
);
2553 * Determine the currently cpus per cpu slab.
2554 * The cpu may change afterward. However that does not matter since
2555 * data is retrieved via this pointer. If we are on the same cpu
2556 * during the cmpxchg then the free will succedd.
2558 c
= __this_cpu_ptr(s
->cpu_slab
);
2563 if (likely(page
== c
->page
)) {
2564 set_freepointer(s
, object
, c
->freelist
);
2566 if (unlikely(!this_cpu_cmpxchg_double(
2567 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2569 object
, next_tid(tid
)))) {
2571 note_cmpxchg_failure("slab_free", s
, tid
);
2574 stat(s
, FREE_FASTPATH
);
2576 __slab_free(s
, page
, x
, addr
);
2580 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2584 page
= virt_to_head_page(x
);
2586 slab_free(s
, page
, x
, _RET_IP_
);
2588 trace_kmem_cache_free(_RET_IP_
, x
);
2590 EXPORT_SYMBOL(kmem_cache_free
);
2593 * Object placement in a slab is made very easy because we always start at
2594 * offset 0. If we tune the size of the object to the alignment then we can
2595 * get the required alignment by putting one properly sized object after
2598 * Notice that the allocation order determines the sizes of the per cpu
2599 * caches. Each processor has always one slab available for allocations.
2600 * Increasing the allocation order reduces the number of times that slabs
2601 * must be moved on and off the partial lists and is therefore a factor in
2606 * Mininum / Maximum order of slab pages. This influences locking overhead
2607 * and slab fragmentation. A higher order reduces the number of partial slabs
2608 * and increases the number of allocations possible without having to
2609 * take the list_lock.
2611 static int slub_min_order
;
2612 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2613 static int slub_min_objects
;
2616 * Merge control. If this is set then no merging of slab caches will occur.
2617 * (Could be removed. This was introduced to pacify the merge skeptics.)
2619 static int slub_nomerge
;
2622 * Calculate the order of allocation given an slab object size.
2624 * The order of allocation has significant impact on performance and other
2625 * system components. Generally order 0 allocations should be preferred since
2626 * order 0 does not cause fragmentation in the page allocator. Larger objects
2627 * be problematic to put into order 0 slabs because there may be too much
2628 * unused space left. We go to a higher order if more than 1/16th of the slab
2631 * In order to reach satisfactory performance we must ensure that a minimum
2632 * number of objects is in one slab. Otherwise we may generate too much
2633 * activity on the partial lists which requires taking the list_lock. This is
2634 * less a concern for large slabs though which are rarely used.
2636 * slub_max_order specifies the order where we begin to stop considering the
2637 * number of objects in a slab as critical. If we reach slub_max_order then
2638 * we try to keep the page order as low as possible. So we accept more waste
2639 * of space in favor of a small page order.
2641 * Higher order allocations also allow the placement of more objects in a
2642 * slab and thereby reduce object handling overhead. If the user has
2643 * requested a higher mininum order then we start with that one instead of
2644 * the smallest order which will fit the object.
2646 static inline int slab_order(int size
, int min_objects
,
2647 int max_order
, int fract_leftover
, int reserved
)
2651 int min_order
= slub_min_order
;
2653 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2654 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2656 for (order
= max(min_order
,
2657 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2658 order
<= max_order
; order
++) {
2660 unsigned long slab_size
= PAGE_SIZE
<< order
;
2662 if (slab_size
< min_objects
* size
+ reserved
)
2665 rem
= (slab_size
- reserved
) % size
;
2667 if (rem
<= slab_size
/ fract_leftover
)
2675 static inline int calculate_order(int size
, int reserved
)
2683 * Attempt to find best configuration for a slab. This
2684 * works by first attempting to generate a layout with
2685 * the best configuration and backing off gradually.
2687 * First we reduce the acceptable waste in a slab. Then
2688 * we reduce the minimum objects required in a slab.
2690 min_objects
= slub_min_objects
;
2692 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2693 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2694 min_objects
= min(min_objects
, max_objects
);
2696 while (min_objects
> 1) {
2698 while (fraction
>= 4) {
2699 order
= slab_order(size
, min_objects
,
2700 slub_max_order
, fraction
, reserved
);
2701 if (order
<= slub_max_order
)
2709 * We were unable to place multiple objects in a slab. Now
2710 * lets see if we can place a single object there.
2712 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2713 if (order
<= slub_max_order
)
2717 * Doh this slab cannot be placed using slub_max_order.
2719 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2720 if (order
< MAX_ORDER
)
2726 * Figure out what the alignment of the objects will be.
2728 static unsigned long calculate_alignment(unsigned long flags
,
2729 unsigned long align
, unsigned long size
)
2732 * If the user wants hardware cache aligned objects then follow that
2733 * suggestion if the object is sufficiently large.
2735 * The hardware cache alignment cannot override the specified
2736 * alignment though. If that is greater then use it.
2738 if (flags
& SLAB_HWCACHE_ALIGN
) {
2739 unsigned long ralign
= cache_line_size();
2740 while (size
<= ralign
/ 2)
2742 align
= max(align
, ralign
);
2745 if (align
< ARCH_SLAB_MINALIGN
)
2746 align
= ARCH_SLAB_MINALIGN
;
2748 return ALIGN(align
, sizeof(void *));
2752 init_kmem_cache_node(struct kmem_cache_node
*n
)
2755 spin_lock_init(&n
->list_lock
);
2756 INIT_LIST_HEAD(&n
->partial
);
2757 #ifdef CONFIG_SLUB_DEBUG
2758 atomic_long_set(&n
->nr_slabs
, 0);
2759 atomic_long_set(&n
->total_objects
, 0);
2760 INIT_LIST_HEAD(&n
->full
);
2764 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2766 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2767 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2770 * Must align to double word boundary for the double cmpxchg
2771 * instructions to work; see __pcpu_double_call_return_bool().
2773 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2774 2 * sizeof(void *));
2779 init_kmem_cache_cpus(s
);
2784 static struct kmem_cache
*kmem_cache_node
;
2787 * No kmalloc_node yet so do it by hand. We know that this is the first
2788 * slab on the node for this slabcache. There are no concurrent accesses
2791 * Note that this function only works on the kmalloc_node_cache
2792 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2793 * memory on a fresh node that has no slab structures yet.
2795 static void early_kmem_cache_node_alloc(int node
)
2798 struct kmem_cache_node
*n
;
2800 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2802 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2805 if (page_to_nid(page
) != node
) {
2806 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2808 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2809 "in order to be able to continue\n");
2814 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2817 kmem_cache_node
->node
[node
] = n
;
2818 #ifdef CONFIG_SLUB_DEBUG
2819 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2820 init_tracking(kmem_cache_node
, n
);
2822 init_kmem_cache_node(n
);
2823 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2825 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2828 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2832 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2833 struct kmem_cache_node
*n
= s
->node
[node
];
2836 kmem_cache_free(kmem_cache_node
, n
);
2838 s
->node
[node
] = NULL
;
2842 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2846 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2847 struct kmem_cache_node
*n
;
2849 if (slab_state
== DOWN
) {
2850 early_kmem_cache_node_alloc(node
);
2853 n
= kmem_cache_alloc_node(kmem_cache_node
,
2857 free_kmem_cache_nodes(s
);
2862 init_kmem_cache_node(n
);
2867 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2869 if (min
< MIN_PARTIAL
)
2871 else if (min
> MAX_PARTIAL
)
2873 s
->min_partial
= min
;
2877 * calculate_sizes() determines the order and the distribution of data within
2880 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2882 unsigned long flags
= s
->flags
;
2883 unsigned long size
= s
->object_size
;
2884 unsigned long align
= s
->align
;
2888 * Round up object size to the next word boundary. We can only
2889 * place the free pointer at word boundaries and this determines
2890 * the possible location of the free pointer.
2892 size
= ALIGN(size
, sizeof(void *));
2894 #ifdef CONFIG_SLUB_DEBUG
2896 * Determine if we can poison the object itself. If the user of
2897 * the slab may touch the object after free or before allocation
2898 * then we should never poison the object itself.
2900 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2902 s
->flags
|= __OBJECT_POISON
;
2904 s
->flags
&= ~__OBJECT_POISON
;
2908 * If we are Redzoning then check if there is some space between the
2909 * end of the object and the free pointer. If not then add an
2910 * additional word to have some bytes to store Redzone information.
2912 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
2913 size
+= sizeof(void *);
2917 * With that we have determined the number of bytes in actual use
2918 * by the object. This is the potential offset to the free pointer.
2922 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2925 * Relocate free pointer after the object if it is not
2926 * permitted to overwrite the first word of the object on
2929 * This is the case if we do RCU, have a constructor or
2930 * destructor or are poisoning the objects.
2933 size
+= sizeof(void *);
2936 #ifdef CONFIG_SLUB_DEBUG
2937 if (flags
& SLAB_STORE_USER
)
2939 * Need to store information about allocs and frees after
2942 size
+= 2 * sizeof(struct track
);
2944 if (flags
& SLAB_RED_ZONE
)
2946 * Add some empty padding so that we can catch
2947 * overwrites from earlier objects rather than let
2948 * tracking information or the free pointer be
2949 * corrupted if a user writes before the start
2952 size
+= sizeof(void *);
2956 * Determine the alignment based on various parameters that the
2957 * user specified and the dynamic determination of cache line size
2960 align
= calculate_alignment(flags
, align
, s
->object_size
);
2964 * SLUB stores one object immediately after another beginning from
2965 * offset 0. In order to align the objects we have to simply size
2966 * each object to conform to the alignment.
2968 size
= ALIGN(size
, align
);
2970 if (forced_order
>= 0)
2971 order
= forced_order
;
2973 order
= calculate_order(size
, s
->reserved
);
2980 s
->allocflags
|= __GFP_COMP
;
2982 if (s
->flags
& SLAB_CACHE_DMA
)
2983 s
->allocflags
|= SLUB_DMA
;
2985 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2986 s
->allocflags
|= __GFP_RECLAIMABLE
;
2989 * Determine the number of objects per slab
2991 s
->oo
= oo_make(order
, size
, s
->reserved
);
2992 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
2993 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2996 return !!oo_objects(s
->oo
);
3000 static int kmem_cache_open(struct kmem_cache
*s
,
3001 const char *name
, size_t size
,
3002 size_t align
, unsigned long flags
,
3003 void (*ctor
)(void *))
3005 memset(s
, 0, kmem_size
);
3008 s
->object_size
= size
;
3010 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
3013 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3014 s
->reserved
= sizeof(struct rcu_head
);
3016 if (!calculate_sizes(s
, -1))
3018 if (disable_higher_order_debug
) {
3020 * Disable debugging flags that store metadata if the min slab
3023 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3024 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3026 if (!calculate_sizes(s
, -1))
3031 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3032 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3033 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3034 /* Enable fast mode */
3035 s
->flags
|= __CMPXCHG_DOUBLE
;
3039 * The larger the object size is, the more pages we want on the partial
3040 * list to avoid pounding the page allocator excessively.
3042 set_min_partial(s
, ilog2(s
->size
) / 2);
3045 * cpu_partial determined the maximum number of objects kept in the
3046 * per cpu partial lists of a processor.
3048 * Per cpu partial lists mainly contain slabs that just have one
3049 * object freed. If they are used for allocation then they can be
3050 * filled up again with minimal effort. The slab will never hit the
3051 * per node partial lists and therefore no locking will be required.
3053 * This setting also determines
3055 * A) The number of objects from per cpu partial slabs dumped to the
3056 * per node list when we reach the limit.
3057 * B) The number of objects in cpu partial slabs to extract from the
3058 * per node list when we run out of per cpu objects. We only fetch 50%
3059 * to keep some capacity around for frees.
3061 if (kmem_cache_debug(s
))
3063 else if (s
->size
>= PAGE_SIZE
)
3065 else if (s
->size
>= 1024)
3067 else if (s
->size
>= 256)
3068 s
->cpu_partial
= 13;
3070 s
->cpu_partial
= 30;
3074 s
->remote_node_defrag_ratio
= 1000;
3076 if (!init_kmem_cache_nodes(s
))
3079 if (alloc_kmem_cache_cpus(s
))
3082 free_kmem_cache_nodes(s
);
3084 if (flags
& SLAB_PANIC
)
3085 panic("Cannot create slab %s size=%lu realsize=%u "
3086 "order=%u offset=%u flags=%lx\n",
3087 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
3093 * Determine the size of a slab object
3095 unsigned int kmem_cache_size(struct kmem_cache
*s
)
3097 return s
->object_size
;
3099 EXPORT_SYMBOL(kmem_cache_size
);
3101 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3104 #ifdef CONFIG_SLUB_DEBUG
3105 void *addr
= page_address(page
);
3107 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3108 sizeof(long), GFP_ATOMIC
);
3111 slab_err(s
, page
, "%s", text
);
3114 get_map(s
, page
, map
);
3115 for_each_object(p
, s
, addr
, page
->objects
) {
3117 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3118 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3120 print_tracking(s
, p
);
3129 * Attempt to free all partial slabs on a node.
3130 * This is called from kmem_cache_close(). We must be the last thread
3131 * using the cache and therefore we do not need to lock anymore.
3133 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3135 struct page
*page
, *h
;
3137 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3139 remove_partial(n
, page
);
3140 discard_slab(s
, page
);
3142 list_slab_objects(s
, page
,
3143 "Objects remaining on kmem_cache_close()");
3149 * Release all resources used by a slab cache.
3151 static inline int kmem_cache_close(struct kmem_cache
*s
)
3156 free_percpu(s
->cpu_slab
);
3157 /* Attempt to free all objects */
3158 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3159 struct kmem_cache_node
*n
= get_node(s
, node
);
3162 if (n
->nr_partial
|| slabs_node(s
, node
))
3165 free_kmem_cache_nodes(s
);
3170 * Close a cache and release the kmem_cache structure
3171 * (must be used for caches created using kmem_cache_create)
3173 void kmem_cache_destroy(struct kmem_cache
*s
)
3175 mutex_lock(&slab_mutex
);
3179 mutex_unlock(&slab_mutex
);
3180 if (kmem_cache_close(s
)) {
3181 printk(KERN_ERR
"SLUB %s: %s called for cache that "
3182 "still has objects.\n", s
->name
, __func__
);
3185 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
3187 sysfs_slab_remove(s
);
3189 mutex_unlock(&slab_mutex
);
3191 EXPORT_SYMBOL(kmem_cache_destroy
);
3193 /********************************************************************
3195 *******************************************************************/
3197 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
3198 EXPORT_SYMBOL(kmalloc_caches
);
3200 static struct kmem_cache
*kmem_cache
;
3202 #ifdef CONFIG_ZONE_DMA
3203 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
3206 static int __init
setup_slub_min_order(char *str
)
3208 get_option(&str
, &slub_min_order
);
3213 __setup("slub_min_order=", setup_slub_min_order
);
3215 static int __init
setup_slub_max_order(char *str
)
3217 get_option(&str
, &slub_max_order
);
3218 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3223 __setup("slub_max_order=", setup_slub_max_order
);
3225 static int __init
setup_slub_min_objects(char *str
)
3227 get_option(&str
, &slub_min_objects
);
3232 __setup("slub_min_objects=", setup_slub_min_objects
);
3234 static int __init
setup_slub_nomerge(char *str
)
3240 __setup("slub_nomerge", setup_slub_nomerge
);
3242 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
3243 int size
, unsigned int flags
)
3245 struct kmem_cache
*s
;
3247 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3250 * This function is called with IRQs disabled during early-boot on
3251 * single CPU so there's no need to take slab_mutex here.
3253 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
3257 list_add(&s
->list
, &slab_caches
);
3261 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
3266 * Conversion table for small slabs sizes / 8 to the index in the
3267 * kmalloc array. This is necessary for slabs < 192 since we have non power
3268 * of two cache sizes there. The size of larger slabs can be determined using
3271 static s8 size_index
[24] = {
3298 static inline int size_index_elem(size_t bytes
)
3300 return (bytes
- 1) / 8;
3303 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
3309 return ZERO_SIZE_PTR
;
3311 index
= size_index
[size_index_elem(size
)];
3313 index
= fls(size
- 1);
3315 #ifdef CONFIG_ZONE_DMA
3316 if (unlikely((flags
& SLUB_DMA
)))
3317 return kmalloc_dma_caches
[index
];
3320 return kmalloc_caches
[index
];
3323 void *__kmalloc(size_t size
, gfp_t flags
)
3325 struct kmem_cache
*s
;
3328 if (unlikely(size
> SLUB_MAX_SIZE
))
3329 return kmalloc_large(size
, flags
);
3331 s
= get_slab(size
, flags
);
3333 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3336 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
3338 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3342 EXPORT_SYMBOL(__kmalloc
);
3345 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3350 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3351 page
= alloc_pages_node(node
, flags
, get_order(size
));
3353 ptr
= page_address(page
);
3355 kmemleak_alloc(ptr
, size
, 1, flags
);
3359 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3361 struct kmem_cache
*s
;
3364 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3365 ret
= kmalloc_large_node(size
, flags
, node
);
3367 trace_kmalloc_node(_RET_IP_
, ret
,
3368 size
, PAGE_SIZE
<< get_order(size
),
3374 s
= get_slab(size
, flags
);
3376 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3379 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
3381 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3385 EXPORT_SYMBOL(__kmalloc_node
);
3388 size_t ksize(const void *object
)
3392 if (unlikely(object
== ZERO_SIZE_PTR
))
3395 page
= virt_to_head_page(object
);
3397 if (unlikely(!PageSlab(page
))) {
3398 WARN_ON(!PageCompound(page
));
3399 return PAGE_SIZE
<< compound_order(page
);
3402 return slab_ksize(page
->slab
);
3404 EXPORT_SYMBOL(ksize
);
3406 #ifdef CONFIG_SLUB_DEBUG
3407 bool verify_mem_not_deleted(const void *x
)
3410 void *object
= (void *)x
;
3411 unsigned long flags
;
3414 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3417 local_irq_save(flags
);
3419 page
= virt_to_head_page(x
);
3420 if (unlikely(!PageSlab(page
))) {
3421 /* maybe it was from stack? */
3427 if (on_freelist(page
->slab
, page
, object
)) {
3428 object_err(page
->slab
, page
, object
, "Object is on free-list");
3436 local_irq_restore(flags
);
3439 EXPORT_SYMBOL(verify_mem_not_deleted
);
3442 void kfree(const void *x
)
3445 void *object
= (void *)x
;
3447 trace_kfree(_RET_IP_
, x
);
3449 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3452 page
= virt_to_head_page(x
);
3453 if (unlikely(!PageSlab(page
))) {
3454 BUG_ON(!PageCompound(page
));
3459 slab_free(page
->slab
, page
, object
, _RET_IP_
);
3461 EXPORT_SYMBOL(kfree
);
3464 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3465 * the remaining slabs by the number of items in use. The slabs with the
3466 * most items in use come first. New allocations will then fill those up
3467 * and thus they can be removed from the partial lists.
3469 * The slabs with the least items are placed last. This results in them
3470 * being allocated from last increasing the chance that the last objects
3471 * are freed in them.
3473 int kmem_cache_shrink(struct kmem_cache
*s
)
3477 struct kmem_cache_node
*n
;
3480 int objects
= oo_objects(s
->max
);
3481 struct list_head
*slabs_by_inuse
=
3482 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3483 unsigned long flags
;
3485 if (!slabs_by_inuse
)
3489 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3490 n
= get_node(s
, node
);
3495 for (i
= 0; i
< objects
; i
++)
3496 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3498 spin_lock_irqsave(&n
->list_lock
, flags
);
3501 * Build lists indexed by the items in use in each slab.
3503 * Note that concurrent frees may occur while we hold the
3504 * list_lock. page->inuse here is the upper limit.
3506 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3507 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3513 * Rebuild the partial list with the slabs filled up most
3514 * first and the least used slabs at the end.
3516 for (i
= objects
- 1; i
> 0; i
--)
3517 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3519 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3521 /* Release empty slabs */
3522 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3523 discard_slab(s
, page
);
3526 kfree(slabs_by_inuse
);
3529 EXPORT_SYMBOL(kmem_cache_shrink
);
3531 #if defined(CONFIG_MEMORY_HOTPLUG)
3532 static int slab_mem_going_offline_callback(void *arg
)
3534 struct kmem_cache
*s
;
3536 mutex_lock(&slab_mutex
);
3537 list_for_each_entry(s
, &slab_caches
, list
)
3538 kmem_cache_shrink(s
);
3539 mutex_unlock(&slab_mutex
);
3544 static void slab_mem_offline_callback(void *arg
)
3546 struct kmem_cache_node
*n
;
3547 struct kmem_cache
*s
;
3548 struct memory_notify
*marg
= arg
;
3551 offline_node
= marg
->status_change_nid
;
3554 * If the node still has available memory. we need kmem_cache_node
3557 if (offline_node
< 0)
3560 mutex_lock(&slab_mutex
);
3561 list_for_each_entry(s
, &slab_caches
, list
) {
3562 n
= get_node(s
, offline_node
);
3565 * if n->nr_slabs > 0, slabs still exist on the node
3566 * that is going down. We were unable to free them,
3567 * and offline_pages() function shouldn't call this
3568 * callback. So, we must fail.
3570 BUG_ON(slabs_node(s
, offline_node
));
3572 s
->node
[offline_node
] = NULL
;
3573 kmem_cache_free(kmem_cache_node
, n
);
3576 mutex_unlock(&slab_mutex
);
3579 static int slab_mem_going_online_callback(void *arg
)
3581 struct kmem_cache_node
*n
;
3582 struct kmem_cache
*s
;
3583 struct memory_notify
*marg
= arg
;
3584 int nid
= marg
->status_change_nid
;
3588 * If the node's memory is already available, then kmem_cache_node is
3589 * already created. Nothing to do.
3595 * We are bringing a node online. No memory is available yet. We must
3596 * allocate a kmem_cache_node structure in order to bring the node
3599 mutex_lock(&slab_mutex
);
3600 list_for_each_entry(s
, &slab_caches
, list
) {
3602 * XXX: kmem_cache_alloc_node will fallback to other nodes
3603 * since memory is not yet available from the node that
3606 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3611 init_kmem_cache_node(n
);
3615 mutex_unlock(&slab_mutex
);
3619 static int slab_memory_callback(struct notifier_block
*self
,
3620 unsigned long action
, void *arg
)
3625 case MEM_GOING_ONLINE
:
3626 ret
= slab_mem_going_online_callback(arg
);
3628 case MEM_GOING_OFFLINE
:
3629 ret
= slab_mem_going_offline_callback(arg
);
3632 case MEM_CANCEL_ONLINE
:
3633 slab_mem_offline_callback(arg
);
3636 case MEM_CANCEL_OFFLINE
:
3640 ret
= notifier_from_errno(ret
);
3646 #endif /* CONFIG_MEMORY_HOTPLUG */
3648 /********************************************************************
3649 * Basic setup of slabs
3650 *******************************************************************/
3653 * Used for early kmem_cache structures that were allocated using
3654 * the page allocator
3657 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3661 list_add(&s
->list
, &slab_caches
);
3664 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3665 struct kmem_cache_node
*n
= get_node(s
, node
);
3669 list_for_each_entry(p
, &n
->partial
, lru
)
3672 #ifdef CONFIG_SLUB_DEBUG
3673 list_for_each_entry(p
, &n
->full
, lru
)
3680 void __init
kmem_cache_init(void)
3684 struct kmem_cache
*temp_kmem_cache
;
3686 struct kmem_cache
*temp_kmem_cache_node
;
3687 unsigned long kmalloc_size
;
3689 if (debug_guardpage_minorder())
3692 kmem_size
= offsetof(struct kmem_cache
, node
) +
3693 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3695 /* Allocate two kmem_caches from the page allocator */
3696 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3697 order
= get_order(2 * kmalloc_size
);
3698 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3701 * Must first have the slab cache available for the allocations of the
3702 * struct kmem_cache_node's. There is special bootstrap code in
3703 * kmem_cache_open for slab_state == DOWN.
3705 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3707 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3708 sizeof(struct kmem_cache_node
),
3709 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3711 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3713 /* Able to allocate the per node structures */
3714 slab_state
= PARTIAL
;
3716 temp_kmem_cache
= kmem_cache
;
3717 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3718 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3719 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3720 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3723 * Allocate kmem_cache_node properly from the kmem_cache slab.
3724 * kmem_cache_node is separately allocated so no need to
3725 * update any list pointers.
3727 temp_kmem_cache_node
= kmem_cache_node
;
3729 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3730 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3732 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3735 kmem_cache_bootstrap_fixup(kmem_cache
);
3737 /* Free temporary boot structure */
3738 free_pages((unsigned long)temp_kmem_cache
, order
);
3740 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3743 * Patch up the size_index table if we have strange large alignment
3744 * requirements for the kmalloc array. This is only the case for
3745 * MIPS it seems. The standard arches will not generate any code here.
3747 * Largest permitted alignment is 256 bytes due to the way we
3748 * handle the index determination for the smaller caches.
3750 * Make sure that nothing crazy happens if someone starts tinkering
3751 * around with ARCH_KMALLOC_MINALIGN
3753 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3754 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3756 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3757 int elem
= size_index_elem(i
);
3758 if (elem
>= ARRAY_SIZE(size_index
))
3760 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3763 if (KMALLOC_MIN_SIZE
== 64) {
3765 * The 96 byte size cache is not used if the alignment
3768 for (i
= 64 + 8; i
<= 96; i
+= 8)
3769 size_index
[size_index_elem(i
)] = 7;
3770 } else if (KMALLOC_MIN_SIZE
== 128) {
3772 * The 192 byte sized cache is not used if the alignment
3773 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3776 for (i
= 128 + 8; i
<= 192; i
+= 8)
3777 size_index
[size_index_elem(i
)] = 8;
3780 /* Caches that are not of the two-to-the-power-of size */
3781 if (KMALLOC_MIN_SIZE
<= 32) {
3782 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3786 if (KMALLOC_MIN_SIZE
<= 64) {
3787 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3791 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3792 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3798 /* Provide the correct kmalloc names now that the caches are up */
3799 if (KMALLOC_MIN_SIZE
<= 32) {
3800 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3801 BUG_ON(!kmalloc_caches
[1]->name
);
3804 if (KMALLOC_MIN_SIZE
<= 64) {
3805 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3806 BUG_ON(!kmalloc_caches
[2]->name
);
3809 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3810 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3813 kmalloc_caches
[i
]->name
= s
;
3817 register_cpu_notifier(&slab_notifier
);
3820 #ifdef CONFIG_ZONE_DMA
3821 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3822 struct kmem_cache
*s
= kmalloc_caches
[i
];
3825 char *name
= kasprintf(GFP_NOWAIT
,
3826 "dma-kmalloc-%d", s
->object_size
);
3829 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3830 s
->object_size
, SLAB_CACHE_DMA
);
3835 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3836 " CPUs=%d, Nodes=%d\n",
3837 caches
, cache_line_size(),
3838 slub_min_order
, slub_max_order
, slub_min_objects
,
3839 nr_cpu_ids
, nr_node_ids
);
3842 void __init
kmem_cache_init_late(void)
3847 * Find a mergeable slab cache
3849 static int slab_unmergeable(struct kmem_cache
*s
)
3851 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3858 * We may have set a slab to be unmergeable during bootstrap.
3860 if (s
->refcount
< 0)
3866 static struct kmem_cache
*find_mergeable(size_t size
,
3867 size_t align
, unsigned long flags
, const char *name
,
3868 void (*ctor
)(void *))
3870 struct kmem_cache
*s
;
3872 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3878 size
= ALIGN(size
, sizeof(void *));
3879 align
= calculate_alignment(flags
, align
, size
);
3880 size
= ALIGN(size
, align
);
3881 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3883 list_for_each_entry(s
, &slab_caches
, list
) {
3884 if (slab_unmergeable(s
))
3890 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3893 * Check if alignment is compatible.
3894 * Courtesy of Adrian Drzewiecki
3896 if ((s
->size
& ~(align
- 1)) != s
->size
)
3899 if (s
->size
- size
>= sizeof(void *))
3907 struct kmem_cache
*__kmem_cache_create(const char *name
, size_t size
,
3908 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3910 struct kmem_cache
*s
;
3913 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3917 * Adjust the object sizes so that we clear
3918 * the complete object on kzalloc.
3920 s
->object_size
= max(s
->object_size
, (int)size
);
3921 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3923 if (sysfs_slab_alias(s
, name
)) {
3930 n
= kstrdup(name
, GFP_KERNEL
);
3934 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3936 if (kmem_cache_open(s
, n
,
3937 size
, align
, flags
, ctor
)) {
3940 list_add(&s
->list
, &slab_caches
);
3941 mutex_unlock(&slab_mutex
);
3942 r
= sysfs_slab_add(s
);
3943 mutex_lock(&slab_mutex
);
3949 kmem_cache_close(s
);
3959 * Use the cpu notifier to insure that the cpu slabs are flushed when
3962 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3963 unsigned long action
, void *hcpu
)
3965 long cpu
= (long)hcpu
;
3966 struct kmem_cache
*s
;
3967 unsigned long flags
;
3970 case CPU_UP_CANCELED
:
3971 case CPU_UP_CANCELED_FROZEN
:
3973 case CPU_DEAD_FROZEN
:
3974 mutex_lock(&slab_mutex
);
3975 list_for_each_entry(s
, &slab_caches
, list
) {
3976 local_irq_save(flags
);
3977 __flush_cpu_slab(s
, cpu
);
3978 local_irq_restore(flags
);
3980 mutex_unlock(&slab_mutex
);
3988 static struct notifier_block __cpuinitdata slab_notifier
= {
3989 .notifier_call
= slab_cpuup_callback
3994 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3996 struct kmem_cache
*s
;
3999 if (unlikely(size
> SLUB_MAX_SIZE
))
4000 return kmalloc_large(size
, gfpflags
);
4002 s
= get_slab(size
, gfpflags
);
4004 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4007 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
4009 /* Honor the call site pointer we received. */
4010 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
4016 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
4017 int node
, unsigned long caller
)
4019 struct kmem_cache
*s
;
4022 if (unlikely(size
> SLUB_MAX_SIZE
)) {
4023 ret
= kmalloc_large_node(size
, gfpflags
, node
);
4025 trace_kmalloc_node(caller
, ret
,
4026 size
, PAGE_SIZE
<< get_order(size
),
4032 s
= get_slab(size
, gfpflags
);
4034 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4037 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
4039 /* Honor the call site pointer we received. */
4040 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
4047 static int count_inuse(struct page
*page
)
4052 static int count_total(struct page
*page
)
4054 return page
->objects
;
4058 #ifdef CONFIG_SLUB_DEBUG
4059 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
4063 void *addr
= page_address(page
);
4065 if (!check_slab(s
, page
) ||
4066 !on_freelist(s
, page
, NULL
))
4069 /* Now we know that a valid freelist exists */
4070 bitmap_zero(map
, page
->objects
);
4072 get_map(s
, page
, map
);
4073 for_each_object(p
, s
, addr
, page
->objects
) {
4074 if (test_bit(slab_index(p
, s
, addr
), map
))
4075 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
4079 for_each_object(p
, s
, addr
, page
->objects
)
4080 if (!test_bit(slab_index(p
, s
, addr
), map
))
4081 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
4086 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
4090 validate_slab(s
, page
, map
);
4094 static int validate_slab_node(struct kmem_cache
*s
,
4095 struct kmem_cache_node
*n
, unsigned long *map
)
4097 unsigned long count
= 0;
4099 unsigned long flags
;
4101 spin_lock_irqsave(&n
->list_lock
, flags
);
4103 list_for_each_entry(page
, &n
->partial
, lru
) {
4104 validate_slab_slab(s
, page
, map
);
4107 if (count
!= n
->nr_partial
)
4108 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
4109 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
4111 if (!(s
->flags
& SLAB_STORE_USER
))
4114 list_for_each_entry(page
, &n
->full
, lru
) {
4115 validate_slab_slab(s
, page
, map
);
4118 if (count
!= atomic_long_read(&n
->nr_slabs
))
4119 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
4120 "counter=%ld\n", s
->name
, count
,
4121 atomic_long_read(&n
->nr_slabs
));
4124 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4128 static long validate_slab_cache(struct kmem_cache
*s
)
4131 unsigned long count
= 0;
4132 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4133 sizeof(unsigned long), GFP_KERNEL
);
4139 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4140 struct kmem_cache_node
*n
= get_node(s
, node
);
4142 count
+= validate_slab_node(s
, n
, map
);
4148 * Generate lists of code addresses where slabcache objects are allocated
4153 unsigned long count
;
4160 DECLARE_BITMAP(cpus
, NR_CPUS
);
4166 unsigned long count
;
4167 struct location
*loc
;
4170 static void free_loc_track(struct loc_track
*t
)
4173 free_pages((unsigned long)t
->loc
,
4174 get_order(sizeof(struct location
) * t
->max
));
4177 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4182 order
= get_order(sizeof(struct location
) * max
);
4184 l
= (void *)__get_free_pages(flags
, order
);
4189 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4197 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4198 const struct track
*track
)
4200 long start
, end
, pos
;
4202 unsigned long caddr
;
4203 unsigned long age
= jiffies
- track
->when
;
4209 pos
= start
+ (end
- start
+ 1) / 2;
4212 * There is nothing at "end". If we end up there
4213 * we need to add something to before end.
4218 caddr
= t
->loc
[pos
].addr
;
4219 if (track
->addr
== caddr
) {
4225 if (age
< l
->min_time
)
4227 if (age
> l
->max_time
)
4230 if (track
->pid
< l
->min_pid
)
4231 l
->min_pid
= track
->pid
;
4232 if (track
->pid
> l
->max_pid
)
4233 l
->max_pid
= track
->pid
;
4235 cpumask_set_cpu(track
->cpu
,
4236 to_cpumask(l
->cpus
));
4238 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4242 if (track
->addr
< caddr
)
4249 * Not found. Insert new tracking element.
4251 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4257 (t
->count
- pos
) * sizeof(struct location
));
4260 l
->addr
= track
->addr
;
4264 l
->min_pid
= track
->pid
;
4265 l
->max_pid
= track
->pid
;
4266 cpumask_clear(to_cpumask(l
->cpus
));
4267 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4268 nodes_clear(l
->nodes
);
4269 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4273 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4274 struct page
*page
, enum track_item alloc
,
4277 void *addr
= page_address(page
);
4280 bitmap_zero(map
, page
->objects
);
4281 get_map(s
, page
, map
);
4283 for_each_object(p
, s
, addr
, page
->objects
)
4284 if (!test_bit(slab_index(p
, s
, addr
), map
))
4285 add_location(t
, s
, get_track(s
, p
, alloc
));
4288 static int list_locations(struct kmem_cache
*s
, char *buf
,
4289 enum track_item alloc
)
4293 struct loc_track t
= { 0, 0, NULL
};
4295 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4296 sizeof(unsigned long), GFP_KERNEL
);
4298 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4301 return sprintf(buf
, "Out of memory\n");
4303 /* Push back cpu slabs */
4306 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4307 struct kmem_cache_node
*n
= get_node(s
, node
);
4308 unsigned long flags
;
4311 if (!atomic_long_read(&n
->nr_slabs
))
4314 spin_lock_irqsave(&n
->list_lock
, flags
);
4315 list_for_each_entry(page
, &n
->partial
, lru
)
4316 process_slab(&t
, s
, page
, alloc
, map
);
4317 list_for_each_entry(page
, &n
->full
, lru
)
4318 process_slab(&t
, s
, page
, alloc
, map
);
4319 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4322 for (i
= 0; i
< t
.count
; i
++) {
4323 struct location
*l
= &t
.loc
[i
];
4325 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4327 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4330 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4332 len
+= sprintf(buf
+ len
, "<not-available>");
4334 if (l
->sum_time
!= l
->min_time
) {
4335 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4337 (long)div_u64(l
->sum_time
, l
->count
),
4340 len
+= sprintf(buf
+ len
, " age=%ld",
4343 if (l
->min_pid
!= l
->max_pid
)
4344 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4345 l
->min_pid
, l
->max_pid
);
4347 len
+= sprintf(buf
+ len
, " pid=%ld",
4350 if (num_online_cpus() > 1 &&
4351 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4352 len
< PAGE_SIZE
- 60) {
4353 len
+= sprintf(buf
+ len
, " cpus=");
4354 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4355 to_cpumask(l
->cpus
));
4358 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4359 len
< PAGE_SIZE
- 60) {
4360 len
+= sprintf(buf
+ len
, " nodes=");
4361 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4365 len
+= sprintf(buf
+ len
, "\n");
4371 len
+= sprintf(buf
, "No data\n");
4376 #ifdef SLUB_RESILIENCY_TEST
4377 static void resiliency_test(void)
4381 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
4383 printk(KERN_ERR
"SLUB resiliency testing\n");
4384 printk(KERN_ERR
"-----------------------\n");
4385 printk(KERN_ERR
"A. Corruption after allocation\n");
4387 p
= kzalloc(16, GFP_KERNEL
);
4389 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4390 " 0x12->0x%p\n\n", p
+ 16);
4392 validate_slab_cache(kmalloc_caches
[4]);
4394 /* Hmmm... The next two are dangerous */
4395 p
= kzalloc(32, GFP_KERNEL
);
4396 p
[32 + sizeof(void *)] = 0x34;
4397 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4398 " 0x34 -> -0x%p\n", p
);
4400 "If allocated object is overwritten then not detectable\n\n");
4402 validate_slab_cache(kmalloc_caches
[5]);
4403 p
= kzalloc(64, GFP_KERNEL
);
4404 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4406 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4409 "If allocated object is overwritten then not detectable\n\n");
4410 validate_slab_cache(kmalloc_caches
[6]);
4412 printk(KERN_ERR
"\nB. Corruption after free\n");
4413 p
= kzalloc(128, GFP_KERNEL
);
4416 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4417 validate_slab_cache(kmalloc_caches
[7]);
4419 p
= kzalloc(256, GFP_KERNEL
);
4422 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4424 validate_slab_cache(kmalloc_caches
[8]);
4426 p
= kzalloc(512, GFP_KERNEL
);
4429 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4430 validate_slab_cache(kmalloc_caches
[9]);
4434 static void resiliency_test(void) {};
4439 enum slab_stat_type
{
4440 SL_ALL
, /* All slabs */
4441 SL_PARTIAL
, /* Only partially allocated slabs */
4442 SL_CPU
, /* Only slabs used for cpu caches */
4443 SL_OBJECTS
, /* Determine allocated objects not slabs */
4444 SL_TOTAL
/* Determine object capacity not slabs */
4447 #define SO_ALL (1 << SL_ALL)
4448 #define SO_PARTIAL (1 << SL_PARTIAL)
4449 #define SO_CPU (1 << SL_CPU)
4450 #define SO_OBJECTS (1 << SL_OBJECTS)
4451 #define SO_TOTAL (1 << SL_TOTAL)
4453 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4454 char *buf
, unsigned long flags
)
4456 unsigned long total
= 0;
4459 unsigned long *nodes
;
4460 unsigned long *per_cpu
;
4462 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4465 per_cpu
= nodes
+ nr_node_ids
;
4467 if (flags
& SO_CPU
) {
4470 for_each_possible_cpu(cpu
) {
4471 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4475 page
= ACCESS_ONCE(c
->page
);
4479 node
= page_to_nid(page
);
4480 if (flags
& SO_TOTAL
)
4482 else if (flags
& SO_OBJECTS
)
4490 page
= ACCESS_ONCE(c
->partial
);
4501 lock_memory_hotplug();
4502 #ifdef CONFIG_SLUB_DEBUG
4503 if (flags
& SO_ALL
) {
4504 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4505 struct kmem_cache_node
*n
= get_node(s
, node
);
4507 if (flags
& SO_TOTAL
)
4508 x
= atomic_long_read(&n
->total_objects
);
4509 else if (flags
& SO_OBJECTS
)
4510 x
= atomic_long_read(&n
->total_objects
) -
4511 count_partial(n
, count_free
);
4514 x
= atomic_long_read(&n
->nr_slabs
);
4521 if (flags
& SO_PARTIAL
) {
4522 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4523 struct kmem_cache_node
*n
= get_node(s
, node
);
4525 if (flags
& SO_TOTAL
)
4526 x
= count_partial(n
, count_total
);
4527 else if (flags
& SO_OBJECTS
)
4528 x
= count_partial(n
, count_inuse
);
4535 x
= sprintf(buf
, "%lu", total
);
4537 for_each_node_state(node
, N_NORMAL_MEMORY
)
4539 x
+= sprintf(buf
+ x
, " N%d=%lu",
4542 unlock_memory_hotplug();
4544 return x
+ sprintf(buf
+ x
, "\n");
4547 #ifdef CONFIG_SLUB_DEBUG
4548 static int any_slab_objects(struct kmem_cache
*s
)
4552 for_each_online_node(node
) {
4553 struct kmem_cache_node
*n
= get_node(s
, node
);
4558 if (atomic_long_read(&n
->total_objects
))
4565 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4566 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4568 struct slab_attribute
{
4569 struct attribute attr
;
4570 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4571 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4574 #define SLAB_ATTR_RO(_name) \
4575 static struct slab_attribute _name##_attr = \
4576 __ATTR(_name, 0400, _name##_show, NULL)
4578 #define SLAB_ATTR(_name) \
4579 static struct slab_attribute _name##_attr = \
4580 __ATTR(_name, 0600, _name##_show, _name##_store)
4582 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4584 return sprintf(buf
, "%d\n", s
->size
);
4586 SLAB_ATTR_RO(slab_size
);
4588 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4590 return sprintf(buf
, "%d\n", s
->align
);
4592 SLAB_ATTR_RO(align
);
4594 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4596 return sprintf(buf
, "%d\n", s
->object_size
);
4598 SLAB_ATTR_RO(object_size
);
4600 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4602 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4604 SLAB_ATTR_RO(objs_per_slab
);
4606 static ssize_t
order_store(struct kmem_cache
*s
,
4607 const char *buf
, size_t length
)
4609 unsigned long order
;
4612 err
= strict_strtoul(buf
, 10, &order
);
4616 if (order
> slub_max_order
|| order
< slub_min_order
)
4619 calculate_sizes(s
, order
);
4623 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4625 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4629 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4631 return sprintf(buf
, "%lu\n", s
->min_partial
);
4634 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4640 err
= strict_strtoul(buf
, 10, &min
);
4644 set_min_partial(s
, min
);
4647 SLAB_ATTR(min_partial
);
4649 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4651 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4654 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4657 unsigned long objects
;
4660 err
= strict_strtoul(buf
, 10, &objects
);
4663 if (objects
&& kmem_cache_debug(s
))
4666 s
->cpu_partial
= objects
;
4670 SLAB_ATTR(cpu_partial
);
4672 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4676 return sprintf(buf
, "%pS\n", s
->ctor
);
4680 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4682 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4684 SLAB_ATTR_RO(aliases
);
4686 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4688 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4690 SLAB_ATTR_RO(partial
);
4692 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4694 return show_slab_objects(s
, buf
, SO_CPU
);
4696 SLAB_ATTR_RO(cpu_slabs
);
4698 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4700 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4702 SLAB_ATTR_RO(objects
);
4704 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4706 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4708 SLAB_ATTR_RO(objects_partial
);
4710 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4717 for_each_online_cpu(cpu
) {
4718 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4721 pages
+= page
->pages
;
4722 objects
+= page
->pobjects
;
4726 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4729 for_each_online_cpu(cpu
) {
4730 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4732 if (page
&& len
< PAGE_SIZE
- 20)
4733 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4734 page
->pobjects
, page
->pages
);
4737 return len
+ sprintf(buf
+ len
, "\n");
4739 SLAB_ATTR_RO(slabs_cpu_partial
);
4741 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4743 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4746 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4747 const char *buf
, size_t length
)
4749 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4751 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4754 SLAB_ATTR(reclaim_account
);
4756 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4758 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4760 SLAB_ATTR_RO(hwcache_align
);
4762 #ifdef CONFIG_ZONE_DMA
4763 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4765 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4767 SLAB_ATTR_RO(cache_dma
);
4770 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4772 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4774 SLAB_ATTR_RO(destroy_by_rcu
);
4776 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4778 return sprintf(buf
, "%d\n", s
->reserved
);
4780 SLAB_ATTR_RO(reserved
);
4782 #ifdef CONFIG_SLUB_DEBUG
4783 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4785 return show_slab_objects(s
, buf
, SO_ALL
);
4787 SLAB_ATTR_RO(slabs
);
4789 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4791 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4793 SLAB_ATTR_RO(total_objects
);
4795 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4797 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4800 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4801 const char *buf
, size_t length
)
4803 s
->flags
&= ~SLAB_DEBUG_FREE
;
4804 if (buf
[0] == '1') {
4805 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4806 s
->flags
|= SLAB_DEBUG_FREE
;
4810 SLAB_ATTR(sanity_checks
);
4812 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4814 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4817 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4820 s
->flags
&= ~SLAB_TRACE
;
4821 if (buf
[0] == '1') {
4822 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4823 s
->flags
|= SLAB_TRACE
;
4829 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4831 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4834 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4835 const char *buf
, size_t length
)
4837 if (any_slab_objects(s
))
4840 s
->flags
&= ~SLAB_RED_ZONE
;
4841 if (buf
[0] == '1') {
4842 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4843 s
->flags
|= SLAB_RED_ZONE
;
4845 calculate_sizes(s
, -1);
4848 SLAB_ATTR(red_zone
);
4850 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4852 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4855 static ssize_t
poison_store(struct kmem_cache
*s
,
4856 const char *buf
, size_t length
)
4858 if (any_slab_objects(s
))
4861 s
->flags
&= ~SLAB_POISON
;
4862 if (buf
[0] == '1') {
4863 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4864 s
->flags
|= SLAB_POISON
;
4866 calculate_sizes(s
, -1);
4871 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4873 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4876 static ssize_t
store_user_store(struct kmem_cache
*s
,
4877 const char *buf
, size_t length
)
4879 if (any_slab_objects(s
))
4882 s
->flags
&= ~SLAB_STORE_USER
;
4883 if (buf
[0] == '1') {
4884 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4885 s
->flags
|= SLAB_STORE_USER
;
4887 calculate_sizes(s
, -1);
4890 SLAB_ATTR(store_user
);
4892 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4897 static ssize_t
validate_store(struct kmem_cache
*s
,
4898 const char *buf
, size_t length
)
4902 if (buf
[0] == '1') {
4903 ret
= validate_slab_cache(s
);
4909 SLAB_ATTR(validate
);
4911 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4913 if (!(s
->flags
& SLAB_STORE_USER
))
4915 return list_locations(s
, buf
, TRACK_ALLOC
);
4917 SLAB_ATTR_RO(alloc_calls
);
4919 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4921 if (!(s
->flags
& SLAB_STORE_USER
))
4923 return list_locations(s
, buf
, TRACK_FREE
);
4925 SLAB_ATTR_RO(free_calls
);
4926 #endif /* CONFIG_SLUB_DEBUG */
4928 #ifdef CONFIG_FAILSLAB
4929 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4931 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4934 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4937 s
->flags
&= ~SLAB_FAILSLAB
;
4939 s
->flags
|= SLAB_FAILSLAB
;
4942 SLAB_ATTR(failslab
);
4945 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4950 static ssize_t
shrink_store(struct kmem_cache
*s
,
4951 const char *buf
, size_t length
)
4953 if (buf
[0] == '1') {
4954 int rc
= kmem_cache_shrink(s
);
4965 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4967 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4970 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4971 const char *buf
, size_t length
)
4973 unsigned long ratio
;
4976 err
= strict_strtoul(buf
, 10, &ratio
);
4981 s
->remote_node_defrag_ratio
= ratio
* 10;
4985 SLAB_ATTR(remote_node_defrag_ratio
);
4988 #ifdef CONFIG_SLUB_STATS
4989 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4991 unsigned long sum
= 0;
4994 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4999 for_each_online_cpu(cpu
) {
5000 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
5006 len
= sprintf(buf
, "%lu", sum
);
5009 for_each_online_cpu(cpu
) {
5010 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
5011 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
5015 return len
+ sprintf(buf
+ len
, "\n");
5018 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
5022 for_each_online_cpu(cpu
)
5023 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
5026 #define STAT_ATTR(si, text) \
5027 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5029 return show_stat(s, buf, si); \
5031 static ssize_t text##_store(struct kmem_cache *s, \
5032 const char *buf, size_t length) \
5034 if (buf[0] != '0') \
5036 clear_stat(s, si); \
5041 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5042 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
5043 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
5044 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
5045 STAT_ATTR(FREE_FROZEN
, free_frozen
);
5046 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
5047 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
5048 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
5049 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
5050 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
5051 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
5052 STAT_ATTR(FREE_SLAB
, free_slab
);
5053 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
5054 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
5055 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
5056 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
5057 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
5058 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
5059 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
5060 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
5061 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
5062 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
5063 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
5064 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
5065 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
5066 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
5069 static struct attribute
*slab_attrs
[] = {
5070 &slab_size_attr
.attr
,
5071 &object_size_attr
.attr
,
5072 &objs_per_slab_attr
.attr
,
5074 &min_partial_attr
.attr
,
5075 &cpu_partial_attr
.attr
,
5077 &objects_partial_attr
.attr
,
5079 &cpu_slabs_attr
.attr
,
5083 &hwcache_align_attr
.attr
,
5084 &reclaim_account_attr
.attr
,
5085 &destroy_by_rcu_attr
.attr
,
5087 &reserved_attr
.attr
,
5088 &slabs_cpu_partial_attr
.attr
,
5089 #ifdef CONFIG_SLUB_DEBUG
5090 &total_objects_attr
.attr
,
5092 &sanity_checks_attr
.attr
,
5094 &red_zone_attr
.attr
,
5096 &store_user_attr
.attr
,
5097 &validate_attr
.attr
,
5098 &alloc_calls_attr
.attr
,
5099 &free_calls_attr
.attr
,
5101 #ifdef CONFIG_ZONE_DMA
5102 &cache_dma_attr
.attr
,
5105 &remote_node_defrag_ratio_attr
.attr
,
5107 #ifdef CONFIG_SLUB_STATS
5108 &alloc_fastpath_attr
.attr
,
5109 &alloc_slowpath_attr
.attr
,
5110 &free_fastpath_attr
.attr
,
5111 &free_slowpath_attr
.attr
,
5112 &free_frozen_attr
.attr
,
5113 &free_add_partial_attr
.attr
,
5114 &free_remove_partial_attr
.attr
,
5115 &alloc_from_partial_attr
.attr
,
5116 &alloc_slab_attr
.attr
,
5117 &alloc_refill_attr
.attr
,
5118 &alloc_node_mismatch_attr
.attr
,
5119 &free_slab_attr
.attr
,
5120 &cpuslab_flush_attr
.attr
,
5121 &deactivate_full_attr
.attr
,
5122 &deactivate_empty_attr
.attr
,
5123 &deactivate_to_head_attr
.attr
,
5124 &deactivate_to_tail_attr
.attr
,
5125 &deactivate_remote_frees_attr
.attr
,
5126 &deactivate_bypass_attr
.attr
,
5127 &order_fallback_attr
.attr
,
5128 &cmpxchg_double_fail_attr
.attr
,
5129 &cmpxchg_double_cpu_fail_attr
.attr
,
5130 &cpu_partial_alloc_attr
.attr
,
5131 &cpu_partial_free_attr
.attr
,
5132 &cpu_partial_node_attr
.attr
,
5133 &cpu_partial_drain_attr
.attr
,
5135 #ifdef CONFIG_FAILSLAB
5136 &failslab_attr
.attr
,
5142 static struct attribute_group slab_attr_group
= {
5143 .attrs
= slab_attrs
,
5146 static ssize_t
slab_attr_show(struct kobject
*kobj
,
5147 struct attribute
*attr
,
5150 struct slab_attribute
*attribute
;
5151 struct kmem_cache
*s
;
5154 attribute
= to_slab_attr(attr
);
5157 if (!attribute
->show
)
5160 err
= attribute
->show(s
, buf
);
5165 static ssize_t
slab_attr_store(struct kobject
*kobj
,
5166 struct attribute
*attr
,
5167 const char *buf
, size_t len
)
5169 struct slab_attribute
*attribute
;
5170 struct kmem_cache
*s
;
5173 attribute
= to_slab_attr(attr
);
5176 if (!attribute
->store
)
5179 err
= attribute
->store(s
, buf
, len
);
5184 static void kmem_cache_release(struct kobject
*kobj
)
5186 struct kmem_cache
*s
= to_slab(kobj
);
5192 static const struct sysfs_ops slab_sysfs_ops
= {
5193 .show
= slab_attr_show
,
5194 .store
= slab_attr_store
,
5197 static struct kobj_type slab_ktype
= {
5198 .sysfs_ops
= &slab_sysfs_ops
,
5199 .release
= kmem_cache_release
5202 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5204 struct kobj_type
*ktype
= get_ktype(kobj
);
5206 if (ktype
== &slab_ktype
)
5211 static const struct kset_uevent_ops slab_uevent_ops
= {
5212 .filter
= uevent_filter
,
5215 static struct kset
*slab_kset
;
5217 #define ID_STR_LENGTH 64
5219 /* Create a unique string id for a slab cache:
5221 * Format :[flags-]size
5223 static char *create_unique_id(struct kmem_cache
*s
)
5225 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5232 * First flags affecting slabcache operations. We will only
5233 * get here for aliasable slabs so we do not need to support
5234 * too many flags. The flags here must cover all flags that
5235 * are matched during merging to guarantee that the id is
5238 if (s
->flags
& SLAB_CACHE_DMA
)
5240 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5242 if (s
->flags
& SLAB_DEBUG_FREE
)
5244 if (!(s
->flags
& SLAB_NOTRACK
))
5248 p
+= sprintf(p
, "%07d", s
->size
);
5249 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5253 static int sysfs_slab_add(struct kmem_cache
*s
)
5259 if (slab_state
< FULL
)
5260 /* Defer until later */
5263 unmergeable
= slab_unmergeable(s
);
5266 * Slabcache can never be merged so we can use the name proper.
5267 * This is typically the case for debug situations. In that
5268 * case we can catch duplicate names easily.
5270 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5274 * Create a unique name for the slab as a target
5277 name
= create_unique_id(s
);
5280 s
->kobj
.kset
= slab_kset
;
5281 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5283 kobject_put(&s
->kobj
);
5287 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5289 kobject_del(&s
->kobj
);
5290 kobject_put(&s
->kobj
);
5293 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5295 /* Setup first alias */
5296 sysfs_slab_alias(s
, s
->name
);
5302 static void sysfs_slab_remove(struct kmem_cache
*s
)
5304 if (slab_state
< FULL
)
5306 * Sysfs has not been setup yet so no need to remove the
5311 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5312 kobject_del(&s
->kobj
);
5313 kobject_put(&s
->kobj
);
5317 * Need to buffer aliases during bootup until sysfs becomes
5318 * available lest we lose that information.
5320 struct saved_alias
{
5321 struct kmem_cache
*s
;
5323 struct saved_alias
*next
;
5326 static struct saved_alias
*alias_list
;
5328 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5330 struct saved_alias
*al
;
5332 if (slab_state
== FULL
) {
5334 * If we have a leftover link then remove it.
5336 sysfs_remove_link(&slab_kset
->kobj
, name
);
5337 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5340 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5346 al
->next
= alias_list
;
5351 static int __init
slab_sysfs_init(void)
5353 struct kmem_cache
*s
;
5356 mutex_lock(&slab_mutex
);
5358 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5360 mutex_unlock(&slab_mutex
);
5361 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5367 list_for_each_entry(s
, &slab_caches
, list
) {
5368 err
= sysfs_slab_add(s
);
5370 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5371 " to sysfs\n", s
->name
);
5374 while (alias_list
) {
5375 struct saved_alias
*al
= alias_list
;
5377 alias_list
= alias_list
->next
;
5378 err
= sysfs_slab_alias(al
->s
, al
->name
);
5380 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5381 " %s to sysfs\n", al
->name
);
5385 mutex_unlock(&slab_mutex
);
5390 __initcall(slab_sysfs_init
);
5391 #endif /* CONFIG_SYSFS */
5394 * The /proc/slabinfo ABI
5396 #ifdef CONFIG_SLABINFO
5397 static void print_slabinfo_header(struct seq_file
*m
)
5399 seq_puts(m
, "slabinfo - version: 2.1\n");
5400 seq_puts(m
, "# name <active_objs> <num_objs> <object_size> "
5401 "<objperslab> <pagesperslab>");
5402 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
5403 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5407 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
5411 mutex_lock(&slab_mutex
);
5413 print_slabinfo_header(m
);
5415 return seq_list_start(&slab_caches
, *pos
);
5418 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
5420 return seq_list_next(p
, &slab_caches
, pos
);
5423 static void s_stop(struct seq_file
*m
, void *p
)
5425 mutex_unlock(&slab_mutex
);
5428 static int s_show(struct seq_file
*m
, void *p
)
5430 unsigned long nr_partials
= 0;
5431 unsigned long nr_slabs
= 0;
5432 unsigned long nr_inuse
= 0;
5433 unsigned long nr_objs
= 0;
5434 unsigned long nr_free
= 0;
5435 struct kmem_cache
*s
;
5438 s
= list_entry(p
, struct kmem_cache
, list
);
5440 for_each_online_node(node
) {
5441 struct kmem_cache_node
*n
= get_node(s
, node
);
5446 nr_partials
+= n
->nr_partial
;
5447 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
5448 nr_objs
+= atomic_long_read(&n
->total_objects
);
5449 nr_free
+= count_partial(n
, count_free
);
5452 nr_inuse
= nr_objs
- nr_free
;
5454 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
5455 nr_objs
, s
->size
, oo_objects(s
->oo
),
5456 (1 << oo_order(s
->oo
)));
5457 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
5458 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
5464 static const struct seq_operations slabinfo_op
= {
5471 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
5473 return seq_open(file
, &slabinfo_op
);
5476 static const struct file_operations proc_slabinfo_operations
= {
5477 .open
= slabinfo_open
,
5479 .llseek
= seq_lseek
,
5480 .release
= seq_release
,
5483 static int __init
slab_proc_init(void)
5485 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
5488 module_init(slab_proc_init
);
5489 #endif /* CONFIG_SLABINFO */