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>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/kmemcheck.h>
22 #include <linux/cpu.h>
23 #include <linux/cpuset.h>
24 #include <linux/mempolicy.h>
25 #include <linux/ctype.h>
26 #include <linux/debugobjects.h>
27 #include <linux/kallsyms.h>
28 #include <linux/memory.h>
29 #include <linux/math64.h>
30 #include <linux/fault-inject.h>
31 #include <linux/stacktrace.h>
33 #include <trace/events/kmem.h>
37 * 1. slub_lock (Global Semaphore)
39 * 3. slab_lock(page) (Only on some arches and for debugging)
43 * The role of the slub_lock is to protect the list of all the slabs
44 * and to synchronize major metadata changes to slab cache structures.
46 * The slab_lock is only used for debugging and on arches that do not
47 * have the ability to do a cmpxchg_double. It only protects the second
48 * double word in the page struct. Meaning
49 * A. page->freelist -> List of object free in a page
50 * B. page->counters -> Counters of objects
51 * C. page->frozen -> frozen state
53 * If a slab is frozen then it is exempt from list management. It is not
54 * on any list. The processor that froze the slab is the one who can
55 * perform list operations on the page. Other processors may put objects
56 * onto the freelist but the processor that froze the slab is the only
57 * one that can retrieve the objects from the page's freelist.
59 * The list_lock protects the partial and full list on each node and
60 * the partial slab counter. If taken then no new slabs may be added or
61 * removed from the lists nor make the number of partial slabs be modified.
62 * (Note that the total number of slabs is an atomic value that may be
63 * modified without taking the list lock).
65 * The list_lock is a centralized lock and thus we avoid taking it as
66 * much as possible. As long as SLUB does not have to handle partial
67 * slabs, operations can continue without any centralized lock. F.e.
68 * allocating a long series of objects that fill up slabs does not require
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache
*s
)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
134 /* Enable to log cmpxchg failures */
135 #undef SLUB_DEBUG_CMPXCHG
138 * Mininum number of partial slabs. These will be left on the partial
139 * lists even if they are empty. kmem_cache_shrink may reclaim them.
141 #define MIN_PARTIAL 5
144 * Maximum number of desirable partial slabs.
145 * The existence of more partial slabs makes kmem_cache_shrink
146 * sort the partial list by the number of objects in the.
148 #define MAX_PARTIAL 10
150 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
151 SLAB_POISON | SLAB_STORE_USER)
154 * Debugging flags that require metadata to be stored in the slab. These get
155 * disabled when slub_debug=O is used and a cache's min order increases with
158 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
161 * Set of flags that will prevent slab merging
163 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
164 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
167 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
168 SLAB_CACHE_DMA | SLAB_NOTRACK)
171 #define OO_MASK ((1 << OO_SHIFT) - 1)
172 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
174 /* Internal SLUB flags */
175 #define __OBJECT_POISON 0x80000000UL /* Poison object */
176 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
178 static int kmem_size
= sizeof(struct kmem_cache
);
181 static struct notifier_block slab_notifier
;
185 DOWN
, /* No slab functionality available */
186 PARTIAL
, /* Kmem_cache_node works */
187 UP
, /* Everything works but does not show up in sysfs */
191 /* A list of all slab caches on the system */
192 static DECLARE_RWSEM(slub_lock
);
193 static LIST_HEAD(slab_caches
);
196 * Tracking user of a slab.
198 #define TRACK_ADDRS_COUNT 16
200 unsigned long addr
; /* Called from address */
201 #ifdef CONFIG_STACKTRACE
202 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
204 int cpu
; /* Was running on cpu */
205 int pid
; /* Pid context */
206 unsigned long when
; /* When did the operation occur */
209 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
212 static int sysfs_slab_add(struct kmem_cache
*);
213 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
214 static void sysfs_slab_remove(struct kmem_cache
*);
217 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
218 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
220 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
228 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
230 #ifdef CONFIG_SLUB_STATS
231 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
235 /********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
239 int slab_is_available(void)
241 return slab_state
>= UP
;
244 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
246 return s
->node
[node
];
249 /* Verify that a pointer has an address that is valid within a slab page */
250 static inline int check_valid_pointer(struct kmem_cache
*s
,
251 struct page
*page
, const void *object
)
258 base
= page_address(page
);
259 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
260 (object
- base
) % s
->size
) {
267 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
269 return *(void **)(object
+ s
->offset
);
272 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
276 #ifdef CONFIG_DEBUG_PAGEALLOC
277 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
279 p
= get_freepointer(s
, object
);
284 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
286 *(void **)(object
+ s
->offset
) = fp
;
289 /* Loop over all objects in a slab */
290 #define for_each_object(__p, __s, __addr, __objects) \
291 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
294 /* Determine object index from a given position */
295 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
297 return (p
- addr
) / s
->size
;
300 static inline size_t slab_ksize(const struct kmem_cache
*s
)
302 #ifdef CONFIG_SLUB_DEBUG
304 * Debugging requires use of the padding between object
305 * and whatever may come after it.
307 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
312 * If we have the need to store the freelist pointer
313 * back there or track user information then we can
314 * only use the space before that information.
316 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
319 * Else we can use all the padding etc for the allocation
324 static inline int order_objects(int order
, unsigned long size
, int reserved
)
326 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
329 static inline struct kmem_cache_order_objects
oo_make(int order
,
330 unsigned long size
, int reserved
)
332 struct kmem_cache_order_objects x
= {
333 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
339 static inline int oo_order(struct kmem_cache_order_objects x
)
341 return x
.x
>> OO_SHIFT
;
344 static inline int oo_objects(struct kmem_cache_order_objects x
)
346 return x
.x
& OO_MASK
;
350 * Per slab locking using the pagelock
352 static __always_inline
void slab_lock(struct page
*page
)
354 bit_spin_lock(PG_locked
, &page
->flags
);
357 static __always_inline
void slab_unlock(struct page
*page
)
359 __bit_spin_unlock(PG_locked
, &page
->flags
);
362 /* Interrupts must be disabled (for the fallback code to work right) */
363 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
364 void *freelist_old
, unsigned long counters_old
,
365 void *freelist_new
, unsigned long counters_new
,
368 VM_BUG_ON(!irqs_disabled());
369 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
370 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
371 if (s
->flags
& __CMPXCHG_DOUBLE
) {
372 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
373 freelist_old
, counters_old
,
374 freelist_new
, counters_new
))
380 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
381 page
->freelist
= freelist_new
;
382 page
->counters
= counters_new
;
390 stat(s
, CMPXCHG_DOUBLE_FAIL
);
392 #ifdef SLUB_DEBUG_CMPXCHG
393 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
399 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
400 void *freelist_old
, unsigned long counters_old
,
401 void *freelist_new
, unsigned long counters_new
,
404 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
405 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
406 if (s
->flags
& __CMPXCHG_DOUBLE
) {
407 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
408 freelist_old
, counters_old
,
409 freelist_new
, counters_new
))
416 local_irq_save(flags
);
418 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
419 page
->freelist
= freelist_new
;
420 page
->counters
= counters_new
;
422 local_irq_restore(flags
);
426 local_irq_restore(flags
);
430 stat(s
, CMPXCHG_DOUBLE_FAIL
);
432 #ifdef SLUB_DEBUG_CMPXCHG
433 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
439 #ifdef CONFIG_SLUB_DEBUG
441 * Determine a map of object in use on a page.
443 * Node listlock must be held to guarantee that the page does
444 * not vanish from under us.
446 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
449 void *addr
= page_address(page
);
451 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
452 set_bit(slab_index(p
, s
, addr
), map
);
458 #ifdef CONFIG_SLUB_DEBUG_ON
459 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
461 static int slub_debug
;
464 static char *slub_debug_slabs
;
465 static int disable_higher_order_debug
;
470 static void print_section(char *text
, u8
*addr
, unsigned int length
)
472 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
476 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
477 enum track_item alloc
)
482 p
= object
+ s
->offset
+ sizeof(void *);
484 p
= object
+ s
->inuse
;
489 static void set_track(struct kmem_cache
*s
, void *object
,
490 enum track_item alloc
, unsigned long addr
)
492 struct track
*p
= get_track(s
, object
, alloc
);
495 #ifdef CONFIG_STACKTRACE
496 struct stack_trace trace
;
499 trace
.nr_entries
= 0;
500 trace
.max_entries
= TRACK_ADDRS_COUNT
;
501 trace
.entries
= p
->addrs
;
503 save_stack_trace(&trace
);
505 /* See rant in lockdep.c */
506 if (trace
.nr_entries
!= 0 &&
507 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
510 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
514 p
->cpu
= smp_processor_id();
515 p
->pid
= current
->pid
;
518 memset(p
, 0, sizeof(struct track
));
521 static void init_tracking(struct kmem_cache
*s
, void *object
)
523 if (!(s
->flags
& SLAB_STORE_USER
))
526 set_track(s
, object
, TRACK_FREE
, 0UL);
527 set_track(s
, object
, TRACK_ALLOC
, 0UL);
530 static void print_track(const char *s
, struct track
*t
)
535 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
536 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
537 #ifdef CONFIG_STACKTRACE
540 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
542 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
549 static void print_tracking(struct kmem_cache
*s
, void *object
)
551 if (!(s
->flags
& SLAB_STORE_USER
))
554 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
555 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
558 static void print_page_info(struct page
*page
)
560 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
561 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
565 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
571 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
573 printk(KERN_ERR
"========================================"
574 "=====================================\n");
575 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
576 printk(KERN_ERR
"----------------------------------------"
577 "-------------------------------------\n\n");
580 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
586 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
588 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
591 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
593 unsigned int off
; /* Offset of last byte */
594 u8
*addr
= page_address(page
);
596 print_tracking(s
, p
);
598 print_page_info(page
);
600 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
601 p
, p
- addr
, get_freepointer(s
, p
));
604 print_section("Bytes b4 ", p
- 16, 16);
606 print_section("Object ", p
, min_t(unsigned long, s
->objsize
,
608 if (s
->flags
& SLAB_RED_ZONE
)
609 print_section("Redzone ", p
+ s
->objsize
,
610 s
->inuse
- s
->objsize
);
613 off
= s
->offset
+ sizeof(void *);
617 if (s
->flags
& SLAB_STORE_USER
)
618 off
+= 2 * sizeof(struct track
);
621 /* Beginning of the filler is the free pointer */
622 print_section("Padding ", p
+ off
, s
->size
- off
);
627 static void object_err(struct kmem_cache
*s
, struct page
*page
,
628 u8
*object
, char *reason
)
630 slab_bug(s
, "%s", reason
);
631 print_trailer(s
, page
, object
);
634 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
640 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
642 slab_bug(s
, "%s", buf
);
643 print_page_info(page
);
647 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
651 if (s
->flags
& __OBJECT_POISON
) {
652 memset(p
, POISON_FREE
, s
->objsize
- 1);
653 p
[s
->objsize
- 1] = POISON_END
;
656 if (s
->flags
& SLAB_RED_ZONE
)
657 memset(p
+ s
->objsize
, val
, s
->inuse
- s
->objsize
);
660 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
661 void *from
, void *to
)
663 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
664 memset(from
, data
, to
- from
);
667 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
668 u8
*object
, char *what
,
669 u8
*start
, unsigned int value
, unsigned int bytes
)
674 fault
= memchr_inv(start
, value
, bytes
);
679 while (end
> fault
&& end
[-1] == value
)
682 slab_bug(s
, "%s overwritten", what
);
683 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
684 fault
, end
- 1, fault
[0], value
);
685 print_trailer(s
, page
, object
);
687 restore_bytes(s
, what
, value
, fault
, end
);
695 * Bytes of the object to be managed.
696 * If the freepointer may overlay the object then the free
697 * pointer is the first word of the object.
699 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
702 * object + s->objsize
703 * Padding to reach word boundary. This is also used for Redzoning.
704 * Padding is extended by another word if Redzoning is enabled and
707 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
708 * 0xcc (RED_ACTIVE) for objects in use.
711 * Meta data starts here.
713 * A. Free pointer (if we cannot overwrite object on free)
714 * B. Tracking data for SLAB_STORE_USER
715 * C. Padding to reach required alignment boundary or at mininum
716 * one word if debugging is on to be able to detect writes
717 * before the word boundary.
719 * Padding is done using 0x5a (POISON_INUSE)
722 * Nothing is used beyond s->size.
724 * If slabcaches are merged then the objsize and inuse boundaries are mostly
725 * ignored. And therefore no slab options that rely on these boundaries
726 * may be used with merged slabcaches.
729 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
731 unsigned long off
= s
->inuse
; /* The end of info */
734 /* Freepointer is placed after the object. */
735 off
+= sizeof(void *);
737 if (s
->flags
& SLAB_STORE_USER
)
738 /* We also have user information there */
739 off
+= 2 * sizeof(struct track
);
744 return check_bytes_and_report(s
, page
, p
, "Object padding",
745 p
+ off
, POISON_INUSE
, s
->size
- off
);
748 /* Check the pad bytes at the end of a slab page */
749 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
757 if (!(s
->flags
& SLAB_POISON
))
760 start
= page_address(page
);
761 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
762 end
= start
+ length
;
763 remainder
= length
% s
->size
;
767 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
770 while (end
> fault
&& end
[-1] == POISON_INUSE
)
773 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
774 print_section("Padding ", end
- remainder
, remainder
);
776 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
780 static int check_object(struct kmem_cache
*s
, struct page
*page
,
781 void *object
, u8 val
)
784 u8
*endobject
= object
+ s
->objsize
;
786 if (s
->flags
& SLAB_RED_ZONE
) {
787 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
788 endobject
, val
, s
->inuse
- s
->objsize
))
791 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
792 check_bytes_and_report(s
, page
, p
, "Alignment padding",
793 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
797 if (s
->flags
& SLAB_POISON
) {
798 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
799 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
800 POISON_FREE
, s
->objsize
- 1) ||
801 !check_bytes_and_report(s
, page
, p
, "Poison",
802 p
+ s
->objsize
- 1, POISON_END
, 1)))
805 * check_pad_bytes cleans up on its own.
807 check_pad_bytes(s
, page
, p
);
810 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
812 * Object and freepointer overlap. Cannot check
813 * freepointer while object is allocated.
817 /* Check free pointer validity */
818 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
819 object_err(s
, page
, p
, "Freepointer corrupt");
821 * No choice but to zap it and thus lose the remainder
822 * of the free objects in this slab. May cause
823 * another error because the object count is now wrong.
825 set_freepointer(s
, p
, NULL
);
831 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
835 VM_BUG_ON(!irqs_disabled());
837 if (!PageSlab(page
)) {
838 slab_err(s
, page
, "Not a valid slab page");
842 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
843 if (page
->objects
> maxobj
) {
844 slab_err(s
, page
, "objects %u > max %u",
845 s
->name
, page
->objects
, maxobj
);
848 if (page
->inuse
> page
->objects
) {
849 slab_err(s
, page
, "inuse %u > max %u",
850 s
->name
, page
->inuse
, page
->objects
);
853 /* Slab_pad_check fixes things up after itself */
854 slab_pad_check(s
, page
);
859 * Determine if a certain object on a page is on the freelist. Must hold the
860 * slab lock to guarantee that the chains are in a consistent state.
862 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
867 unsigned long max_objects
;
870 while (fp
&& nr
<= page
->objects
) {
873 if (!check_valid_pointer(s
, page
, fp
)) {
875 object_err(s
, page
, object
,
876 "Freechain corrupt");
877 set_freepointer(s
, object
, NULL
);
880 slab_err(s
, page
, "Freepointer corrupt");
881 page
->freelist
= NULL
;
882 page
->inuse
= page
->objects
;
883 slab_fix(s
, "Freelist cleared");
889 fp
= get_freepointer(s
, object
);
893 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
894 if (max_objects
> MAX_OBJS_PER_PAGE
)
895 max_objects
= MAX_OBJS_PER_PAGE
;
897 if (page
->objects
!= max_objects
) {
898 slab_err(s
, page
, "Wrong number of objects. Found %d but "
899 "should be %d", page
->objects
, max_objects
);
900 page
->objects
= max_objects
;
901 slab_fix(s
, "Number of objects adjusted.");
903 if (page
->inuse
!= page
->objects
- nr
) {
904 slab_err(s
, page
, "Wrong object count. Counter is %d but "
905 "counted were %d", page
->inuse
, page
->objects
- nr
);
906 page
->inuse
= page
->objects
- nr
;
907 slab_fix(s
, "Object count adjusted.");
909 return search
== NULL
;
912 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
915 if (s
->flags
& SLAB_TRACE
) {
916 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
918 alloc
? "alloc" : "free",
923 print_section("Object ", (void *)object
, s
->objsize
);
930 * Hooks for other subsystems that check memory allocations. In a typical
931 * production configuration these hooks all should produce no code at all.
933 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
935 flags
&= gfp_allowed_mask
;
936 lockdep_trace_alloc(flags
);
937 might_sleep_if(flags
& __GFP_WAIT
);
939 return should_failslab(s
->objsize
, flags
, s
->flags
);
942 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
944 flags
&= gfp_allowed_mask
;
945 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
946 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
949 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
951 kmemleak_free_recursive(x
, s
->flags
);
954 * Trouble is that we may no longer disable interupts in the fast path
955 * So in order to make the debug calls that expect irqs to be
956 * disabled we need to disable interrupts temporarily.
958 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
962 local_irq_save(flags
);
963 kmemcheck_slab_free(s
, x
, s
->objsize
);
964 debug_check_no_locks_freed(x
, s
->objsize
);
965 local_irq_restore(flags
);
968 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
969 debug_check_no_obj_freed(x
, s
->objsize
);
973 * Tracking of fully allocated slabs for debugging purposes.
975 * list_lock must be held.
977 static void add_full(struct kmem_cache
*s
,
978 struct kmem_cache_node
*n
, struct page
*page
)
980 if (!(s
->flags
& SLAB_STORE_USER
))
983 list_add(&page
->lru
, &n
->full
);
987 * list_lock must be held.
989 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
991 if (!(s
->flags
& SLAB_STORE_USER
))
994 list_del(&page
->lru
);
997 /* Tracking of the number of slabs for debugging purposes */
998 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1000 struct kmem_cache_node
*n
= get_node(s
, node
);
1002 return atomic_long_read(&n
->nr_slabs
);
1005 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1007 return atomic_long_read(&n
->nr_slabs
);
1010 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1012 struct kmem_cache_node
*n
= get_node(s
, node
);
1015 * May be called early in order to allocate a slab for the
1016 * kmem_cache_node structure. Solve the chicken-egg
1017 * dilemma by deferring the increment of the count during
1018 * bootstrap (see early_kmem_cache_node_alloc).
1021 atomic_long_inc(&n
->nr_slabs
);
1022 atomic_long_add(objects
, &n
->total_objects
);
1025 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1027 struct kmem_cache_node
*n
= get_node(s
, node
);
1029 atomic_long_dec(&n
->nr_slabs
);
1030 atomic_long_sub(objects
, &n
->total_objects
);
1033 /* Object debug checks for alloc/free paths */
1034 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1037 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1040 init_object(s
, object
, SLUB_RED_INACTIVE
);
1041 init_tracking(s
, object
);
1044 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
1045 void *object
, unsigned long addr
)
1047 if (!check_slab(s
, page
))
1050 if (!check_valid_pointer(s
, page
, object
)) {
1051 object_err(s
, page
, object
, "Freelist Pointer check fails");
1055 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1058 /* Success perform special debug activities for allocs */
1059 if (s
->flags
& SLAB_STORE_USER
)
1060 set_track(s
, object
, TRACK_ALLOC
, addr
);
1061 trace(s
, page
, object
, 1);
1062 init_object(s
, object
, SLUB_RED_ACTIVE
);
1066 if (PageSlab(page
)) {
1068 * If this is a slab page then lets do the best we can
1069 * to avoid issues in the future. Marking all objects
1070 * as used avoids touching the remaining objects.
1072 slab_fix(s
, "Marking all objects used");
1073 page
->inuse
= page
->objects
;
1074 page
->freelist
= NULL
;
1079 static noinline
int free_debug_processing(struct kmem_cache
*s
,
1080 struct page
*page
, void *object
, unsigned long addr
)
1082 unsigned long flags
;
1085 local_irq_save(flags
);
1088 if (!check_slab(s
, page
))
1091 if (!check_valid_pointer(s
, page
, object
)) {
1092 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1096 if (on_freelist(s
, page
, object
)) {
1097 object_err(s
, page
, object
, "Object already free");
1101 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1104 if (unlikely(s
!= page
->slab
)) {
1105 if (!PageSlab(page
)) {
1106 slab_err(s
, page
, "Attempt to free object(0x%p) "
1107 "outside of slab", object
);
1108 } else if (!page
->slab
) {
1110 "SLUB <none>: no slab for object 0x%p.\n",
1114 object_err(s
, page
, object
,
1115 "page slab pointer corrupt.");
1119 if (s
->flags
& SLAB_STORE_USER
)
1120 set_track(s
, object
, TRACK_FREE
, addr
);
1121 trace(s
, page
, object
, 0);
1122 init_object(s
, object
, SLUB_RED_INACTIVE
);
1126 local_irq_restore(flags
);
1130 slab_fix(s
, "Object at 0x%p not freed", object
);
1134 static int __init
setup_slub_debug(char *str
)
1136 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1137 if (*str
++ != '=' || !*str
)
1139 * No options specified. Switch on full debugging.
1145 * No options but restriction on slabs. This means full
1146 * debugging for slabs matching a pattern.
1150 if (tolower(*str
) == 'o') {
1152 * Avoid enabling debugging on caches if its minimum order
1153 * would increase as a result.
1155 disable_higher_order_debug
= 1;
1162 * Switch off all debugging measures.
1167 * Determine which debug features should be switched on
1169 for (; *str
&& *str
!= ','; str
++) {
1170 switch (tolower(*str
)) {
1172 slub_debug
|= SLAB_DEBUG_FREE
;
1175 slub_debug
|= SLAB_RED_ZONE
;
1178 slub_debug
|= SLAB_POISON
;
1181 slub_debug
|= SLAB_STORE_USER
;
1184 slub_debug
|= SLAB_TRACE
;
1187 slub_debug
|= SLAB_FAILSLAB
;
1190 printk(KERN_ERR
"slub_debug option '%c' "
1191 "unknown. skipped\n", *str
);
1197 slub_debug_slabs
= str
+ 1;
1202 __setup("slub_debug", setup_slub_debug
);
1204 static unsigned long kmem_cache_flags(unsigned long objsize
,
1205 unsigned long flags
, const char *name
,
1206 void (*ctor
)(void *))
1209 * Enable debugging if selected on the kernel commandline.
1211 if (slub_debug
&& (!slub_debug_slabs
||
1212 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1213 flags
|= slub_debug
;
1218 static inline void setup_object_debug(struct kmem_cache
*s
,
1219 struct page
*page
, void *object
) {}
1221 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1222 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1224 static inline int free_debug_processing(struct kmem_cache
*s
,
1225 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1227 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1229 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1230 void *object
, u8 val
) { return 1; }
1231 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1232 struct page
*page
) {}
1233 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1234 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1235 unsigned long flags
, const char *name
,
1236 void (*ctor
)(void *))
1240 #define slub_debug 0
1242 #define disable_higher_order_debug 0
1244 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1246 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1248 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1250 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1253 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1256 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1259 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1261 #endif /* CONFIG_SLUB_DEBUG */
1264 * Slab allocation and freeing
1266 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1267 struct kmem_cache_order_objects oo
)
1269 int order
= oo_order(oo
);
1271 flags
|= __GFP_NOTRACK
;
1273 if (node
== NUMA_NO_NODE
)
1274 return alloc_pages(flags
, order
);
1276 return alloc_pages_exact_node(node
, flags
, order
);
1279 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1282 struct kmem_cache_order_objects oo
= s
->oo
;
1285 flags
&= gfp_allowed_mask
;
1287 if (flags
& __GFP_WAIT
)
1290 flags
|= s
->allocflags
;
1293 * Let the initial higher-order allocation fail under memory pressure
1294 * so we fall-back to the minimum order allocation.
1296 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1298 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1299 if (unlikely(!page
)) {
1302 * Allocation may have failed due to fragmentation.
1303 * Try a lower order alloc if possible
1305 page
= alloc_slab_page(flags
, node
, oo
);
1308 stat(s
, ORDER_FALLBACK
);
1311 if (flags
& __GFP_WAIT
)
1312 local_irq_disable();
1317 if (kmemcheck_enabled
1318 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1319 int pages
= 1 << oo_order(oo
);
1321 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1324 * Objects from caches that have a constructor don't get
1325 * cleared when they're allocated, so we need to do it here.
1328 kmemcheck_mark_uninitialized_pages(page
, pages
);
1330 kmemcheck_mark_unallocated_pages(page
, pages
);
1333 page
->objects
= oo_objects(oo
);
1334 mod_zone_page_state(page_zone(page
),
1335 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1336 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1342 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1345 setup_object_debug(s
, page
, object
);
1346 if (unlikely(s
->ctor
))
1350 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1357 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1359 page
= allocate_slab(s
,
1360 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1364 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1366 page
->flags
|= 1 << PG_slab
;
1368 start
= page_address(page
);
1370 if (unlikely(s
->flags
& SLAB_POISON
))
1371 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1374 for_each_object(p
, s
, start
, page
->objects
) {
1375 setup_object(s
, page
, last
);
1376 set_freepointer(s
, last
, p
);
1379 setup_object(s
, page
, last
);
1380 set_freepointer(s
, last
, NULL
);
1382 page
->freelist
= start
;
1383 page
->inuse
= page
->objects
;
1389 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1391 int order
= compound_order(page
);
1392 int pages
= 1 << order
;
1394 if (kmem_cache_debug(s
)) {
1397 slab_pad_check(s
, page
);
1398 for_each_object(p
, s
, page_address(page
),
1400 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1403 kmemcheck_free_shadow(page
, compound_order(page
));
1405 mod_zone_page_state(page_zone(page
),
1406 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1407 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1410 __ClearPageSlab(page
);
1411 reset_page_mapcount(page
);
1412 if (current
->reclaim_state
)
1413 current
->reclaim_state
->reclaimed_slab
+= pages
;
1414 __free_pages(page
, order
);
1417 #define need_reserve_slab_rcu \
1418 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1420 static void rcu_free_slab(struct rcu_head
*h
)
1424 if (need_reserve_slab_rcu
)
1425 page
= virt_to_head_page(h
);
1427 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1429 __free_slab(page
->slab
, page
);
1432 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1434 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1435 struct rcu_head
*head
;
1437 if (need_reserve_slab_rcu
) {
1438 int order
= compound_order(page
);
1439 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1441 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1442 head
= page_address(page
) + offset
;
1445 * RCU free overloads the RCU head over the LRU
1447 head
= (void *)&page
->lru
;
1450 call_rcu(head
, rcu_free_slab
);
1452 __free_slab(s
, page
);
1455 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1457 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1462 * Management of partially allocated slabs.
1464 * list_lock must be held.
1466 static inline void add_partial(struct kmem_cache_node
*n
,
1467 struct page
*page
, int tail
)
1470 if (tail
== DEACTIVATE_TO_TAIL
)
1471 list_add_tail(&page
->lru
, &n
->partial
);
1473 list_add(&page
->lru
, &n
->partial
);
1477 * list_lock must be held.
1479 static inline void remove_partial(struct kmem_cache_node
*n
,
1482 list_del(&page
->lru
);
1487 * Lock slab, remove from the partial list and put the object into the
1490 * Returns a list of objects or NULL if it fails.
1492 * Must hold list_lock.
1494 static inline void *acquire_slab(struct kmem_cache
*s
,
1495 struct kmem_cache_node
*n
, struct page
*page
,
1499 unsigned long counters
;
1503 * Zap the freelist and set the frozen bit.
1504 * The old freelist is the list of objects for the
1505 * per cpu allocation list.
1508 freelist
= page
->freelist
;
1509 counters
= page
->counters
;
1510 new.counters
= counters
;
1512 new.inuse
= page
->objects
;
1514 VM_BUG_ON(new.frozen
);
1517 } while (!__cmpxchg_double_slab(s
, page
,
1520 "lock and freeze"));
1522 remove_partial(n
, page
);
1526 static int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1529 * Try to allocate a partial slab from a specific node.
1531 static void *get_partial_node(struct kmem_cache
*s
,
1532 struct kmem_cache_node
*n
, struct kmem_cache_cpu
*c
)
1534 struct page
*page
, *page2
;
1535 void *object
= NULL
;
1538 * Racy check. If we mistakenly see no partial slabs then we
1539 * just allocate an empty slab. If we mistakenly try to get a
1540 * partial slab and there is none available then get_partials()
1543 if (!n
|| !n
->nr_partial
)
1546 spin_lock(&n
->list_lock
);
1547 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1548 void *t
= acquire_slab(s
, n
, page
, object
== NULL
);
1556 c
->node
= page_to_nid(page
);
1557 stat(s
, ALLOC_FROM_PARTIAL
);
1559 available
= page
->objects
- page
->inuse
;
1562 available
= put_cpu_partial(s
, page
, 0);
1564 if (kmem_cache_debug(s
) || available
> s
->cpu_partial
/ 2)
1568 spin_unlock(&n
->list_lock
);
1573 * Get a page from somewhere. Search in increasing NUMA distances.
1575 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1576 struct kmem_cache_cpu
*c
)
1579 struct zonelist
*zonelist
;
1582 enum zone_type high_zoneidx
= gfp_zone(flags
);
1584 unsigned int cpuset_mems_cookie
;
1587 * The defrag ratio allows a configuration of the tradeoffs between
1588 * inter node defragmentation and node local allocations. A lower
1589 * defrag_ratio increases the tendency to do local allocations
1590 * instead of attempting to obtain partial slabs from other nodes.
1592 * If the defrag_ratio is set to 0 then kmalloc() always
1593 * returns node local objects. If the ratio is higher then kmalloc()
1594 * may return off node objects because partial slabs are obtained
1595 * from other nodes and filled up.
1597 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1598 * defrag_ratio = 1000) then every (well almost) allocation will
1599 * first attempt to defrag slab caches on other nodes. This means
1600 * scanning over all nodes to look for partial slabs which may be
1601 * expensive if we do it every time we are trying to find a slab
1602 * with available objects.
1604 if (!s
->remote_node_defrag_ratio
||
1605 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1609 cpuset_mems_cookie
= get_mems_allowed();
1610 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1611 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1612 struct kmem_cache_node
*n
;
1614 n
= get_node(s
, zone_to_nid(zone
));
1616 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1617 n
->nr_partial
> s
->min_partial
) {
1618 object
= get_partial_node(s
, n
, c
);
1621 * Return the object even if
1622 * put_mems_allowed indicated that
1623 * the cpuset mems_allowed was
1624 * updated in parallel. It's a
1625 * harmless race between the alloc
1626 * and the cpuset update.
1628 put_mems_allowed(cpuset_mems_cookie
);
1633 } while (!put_mems_allowed(cpuset_mems_cookie
));
1639 * Get a partial page, lock it and return it.
1641 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1642 struct kmem_cache_cpu
*c
)
1645 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1647 object
= get_partial_node(s
, get_node(s
, searchnode
), c
);
1648 if (object
|| node
!= NUMA_NO_NODE
)
1651 return get_any_partial(s
, flags
, c
);
1654 #ifdef CONFIG_PREEMPT
1656 * Calculate the next globally unique transaction for disambiguiation
1657 * during cmpxchg. The transactions start with the cpu number and are then
1658 * incremented by CONFIG_NR_CPUS.
1660 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1663 * No preemption supported therefore also no need to check for
1669 static inline unsigned long next_tid(unsigned long tid
)
1671 return tid
+ TID_STEP
;
1674 static inline unsigned int tid_to_cpu(unsigned long tid
)
1676 return tid
% TID_STEP
;
1679 static inline unsigned long tid_to_event(unsigned long tid
)
1681 return tid
/ TID_STEP
;
1684 static inline unsigned int init_tid(int cpu
)
1689 static inline void note_cmpxchg_failure(const char *n
,
1690 const struct kmem_cache
*s
, unsigned long tid
)
1692 #ifdef SLUB_DEBUG_CMPXCHG
1693 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1695 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1697 #ifdef CONFIG_PREEMPT
1698 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1699 printk("due to cpu change %d -> %d\n",
1700 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1703 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1704 printk("due to cpu running other code. Event %ld->%ld\n",
1705 tid_to_event(tid
), tid_to_event(actual_tid
));
1707 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1708 actual_tid
, tid
, next_tid(tid
));
1710 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1713 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1717 for_each_possible_cpu(cpu
)
1718 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1722 * Remove the cpu slab
1724 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1726 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1727 struct page
*page
= c
->page
;
1728 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1730 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1733 int tail
= DEACTIVATE_TO_HEAD
;
1737 if (page
->freelist
) {
1738 stat(s
, DEACTIVATE_REMOTE_FREES
);
1739 tail
= DEACTIVATE_TO_TAIL
;
1742 c
->tid
= next_tid(c
->tid
);
1744 freelist
= c
->freelist
;
1748 * Stage one: Free all available per cpu objects back
1749 * to the page freelist while it is still frozen. Leave the
1752 * There is no need to take the list->lock because the page
1755 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1757 unsigned long counters
;
1760 prior
= page
->freelist
;
1761 counters
= page
->counters
;
1762 set_freepointer(s
, freelist
, prior
);
1763 new.counters
= counters
;
1765 VM_BUG_ON(!new.frozen
);
1767 } while (!__cmpxchg_double_slab(s
, page
,
1769 freelist
, new.counters
,
1770 "drain percpu freelist"));
1772 freelist
= nextfree
;
1776 * Stage two: Ensure that the page is unfrozen while the
1777 * list presence reflects the actual number of objects
1780 * We setup the list membership and then perform a cmpxchg
1781 * with the count. If there is a mismatch then the page
1782 * is not unfrozen but the page is on the wrong list.
1784 * Then we restart the process which may have to remove
1785 * the page from the list that we just put it on again
1786 * because the number of objects in the slab may have
1791 old
.freelist
= page
->freelist
;
1792 old
.counters
= page
->counters
;
1793 VM_BUG_ON(!old
.frozen
);
1795 /* Determine target state of the slab */
1796 new.counters
= old
.counters
;
1799 set_freepointer(s
, freelist
, old
.freelist
);
1800 new.freelist
= freelist
;
1802 new.freelist
= old
.freelist
;
1806 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1808 else if (new.freelist
) {
1813 * Taking the spinlock removes the possiblity
1814 * that acquire_slab() will see a slab page that
1817 spin_lock(&n
->list_lock
);
1821 if (kmem_cache_debug(s
) && !lock
) {
1824 * This also ensures that the scanning of full
1825 * slabs from diagnostic functions will not see
1828 spin_lock(&n
->list_lock
);
1836 remove_partial(n
, page
);
1838 else if (l
== M_FULL
)
1840 remove_full(s
, page
);
1842 if (m
== M_PARTIAL
) {
1844 add_partial(n
, page
, tail
);
1847 } else if (m
== M_FULL
) {
1849 stat(s
, DEACTIVATE_FULL
);
1850 add_full(s
, n
, page
);
1856 if (!__cmpxchg_double_slab(s
, page
,
1857 old
.freelist
, old
.counters
,
1858 new.freelist
, new.counters
,
1863 spin_unlock(&n
->list_lock
);
1866 stat(s
, DEACTIVATE_EMPTY
);
1867 discard_slab(s
, page
);
1872 /* Unfreeze all the cpu partial slabs */
1873 static void unfreeze_partials(struct kmem_cache
*s
)
1875 struct kmem_cache_node
*n
= NULL
;
1876 struct kmem_cache_cpu
*c
= this_cpu_ptr(s
->cpu_slab
);
1877 struct page
*page
, *discard_page
= NULL
;
1879 while ((page
= c
->partial
)) {
1880 enum slab_modes
{ M_PARTIAL
, M_FREE
};
1881 enum slab_modes l
, m
;
1885 c
->partial
= page
->next
;
1890 old
.freelist
= page
->freelist
;
1891 old
.counters
= page
->counters
;
1892 VM_BUG_ON(!old
.frozen
);
1894 new.counters
= old
.counters
;
1895 new.freelist
= old
.freelist
;
1899 if (!new.inuse
&& (!n
|| n
->nr_partial
> s
->min_partial
))
1902 struct kmem_cache_node
*n2
= get_node(s
,
1908 spin_unlock(&n
->list_lock
);
1911 spin_lock(&n
->list_lock
);
1916 if (l
== M_PARTIAL
) {
1917 remove_partial(n
, page
);
1918 stat(s
, FREE_REMOVE_PARTIAL
);
1920 add_partial(n
, page
,
1921 DEACTIVATE_TO_TAIL
);
1922 stat(s
, FREE_ADD_PARTIAL
);
1928 } while (!cmpxchg_double_slab(s
, page
,
1929 old
.freelist
, old
.counters
,
1930 new.freelist
, new.counters
,
1931 "unfreezing slab"));
1934 page
->next
= discard_page
;
1935 discard_page
= page
;
1940 spin_unlock(&n
->list_lock
);
1942 while (discard_page
) {
1943 page
= discard_page
;
1944 discard_page
= discard_page
->next
;
1946 stat(s
, DEACTIVATE_EMPTY
);
1947 discard_slab(s
, page
);
1953 * Put a page that was just frozen (in __slab_free) into a partial page
1954 * slot if available. This is done without interrupts disabled and without
1955 * preemption disabled. The cmpxchg is racy and may put the partial page
1956 * onto a random cpus partial slot.
1958 * If we did not find a slot then simply move all the partials to the
1959 * per node partial list.
1961 int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1963 struct page
*oldpage
;
1970 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1973 pobjects
= oldpage
->pobjects
;
1974 pages
= oldpage
->pages
;
1975 if (drain
&& pobjects
> s
->cpu_partial
) {
1976 unsigned long flags
;
1978 * partial array is full. Move the existing
1979 * set to the per node partial list.
1981 local_irq_save(flags
);
1982 unfreeze_partials(s
);
1983 local_irq_restore(flags
);
1990 pobjects
+= page
->objects
- page
->inuse
;
1992 page
->pages
= pages
;
1993 page
->pobjects
= pobjects
;
1994 page
->next
= oldpage
;
1996 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
) != oldpage
);
1997 stat(s
, CPU_PARTIAL_FREE
);
2001 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2003 stat(s
, CPUSLAB_FLUSH
);
2004 deactivate_slab(s
, c
);
2010 * Called from IPI handler with interrupts disabled.
2012 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2014 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2020 unfreeze_partials(s
);
2024 static void flush_cpu_slab(void *d
)
2026 struct kmem_cache
*s
= d
;
2028 __flush_cpu_slab(s
, smp_processor_id());
2031 static void flush_all(struct kmem_cache
*s
)
2033 on_each_cpu(flush_cpu_slab
, s
, 1);
2037 * Check if the objects in a per cpu structure fit numa
2038 * locality expectations.
2040 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
2043 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
2049 static int count_free(struct page
*page
)
2051 return page
->objects
- page
->inuse
;
2054 static unsigned long count_partial(struct kmem_cache_node
*n
,
2055 int (*get_count
)(struct page
*))
2057 unsigned long flags
;
2058 unsigned long x
= 0;
2061 spin_lock_irqsave(&n
->list_lock
, flags
);
2062 list_for_each_entry(page
, &n
->partial
, lru
)
2063 x
+= get_count(page
);
2064 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2068 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2070 #ifdef CONFIG_SLUB_DEBUG
2071 return atomic_long_read(&n
->total_objects
);
2077 static noinline
void
2078 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2083 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2085 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2086 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
2087 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2089 if (oo_order(s
->min
) > get_order(s
->objsize
))
2090 printk(KERN_WARNING
" %s debugging increased min order, use "
2091 "slub_debug=O to disable.\n", s
->name
);
2093 for_each_online_node(node
) {
2094 struct kmem_cache_node
*n
= get_node(s
, node
);
2095 unsigned long nr_slabs
;
2096 unsigned long nr_objs
;
2097 unsigned long nr_free
;
2102 nr_free
= count_partial(n
, count_free
);
2103 nr_slabs
= node_nr_slabs(n
);
2104 nr_objs
= node_nr_objs(n
);
2107 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2108 node
, nr_slabs
, nr_objs
, nr_free
);
2112 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2113 int node
, struct kmem_cache_cpu
**pc
)
2116 struct kmem_cache_cpu
*c
;
2117 struct page
*page
= new_slab(s
, flags
, node
);
2120 c
= __this_cpu_ptr(s
->cpu_slab
);
2125 * No other reference to the page yet so we can
2126 * muck around with it freely without cmpxchg
2128 object
= page
->freelist
;
2129 page
->freelist
= NULL
;
2131 stat(s
, ALLOC_SLAB
);
2132 c
->node
= page_to_nid(page
);
2142 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2143 * or deactivate the page.
2145 * The page is still frozen if the return value is not NULL.
2147 * If this function returns NULL then the page has been unfrozen.
2149 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2152 unsigned long counters
;
2156 freelist
= page
->freelist
;
2157 counters
= page
->counters
;
2158 new.counters
= counters
;
2159 VM_BUG_ON(!new.frozen
);
2161 new.inuse
= page
->objects
;
2162 new.frozen
= freelist
!= NULL
;
2164 } while (!cmpxchg_double_slab(s
, page
,
2173 * Slow path. The lockless freelist is empty or we need to perform
2176 * Processing is still very fast if new objects have been freed to the
2177 * regular freelist. In that case we simply take over the regular freelist
2178 * as the lockless freelist and zap the regular freelist.
2180 * If that is not working then we fall back to the partial lists. We take the
2181 * first element of the freelist as the object to allocate now and move the
2182 * rest of the freelist to the lockless freelist.
2184 * And if we were unable to get a new slab from the partial slab lists then
2185 * we need to allocate a new slab. This is the slowest path since it involves
2186 * a call to the page allocator and the setup of a new slab.
2188 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2189 unsigned long addr
, struct kmem_cache_cpu
*c
)
2192 unsigned long flags
;
2194 local_irq_save(flags
);
2195 #ifdef CONFIG_PREEMPT
2197 * We may have been preempted and rescheduled on a different
2198 * cpu before disabling interrupts. Need to reload cpu area
2201 c
= this_cpu_ptr(s
->cpu_slab
);
2207 if (unlikely(!node_match(c
, node
))) {
2208 stat(s
, ALLOC_NODE_MISMATCH
);
2209 deactivate_slab(s
, c
);
2213 /* must check again c->freelist in case of cpu migration or IRQ */
2214 object
= c
->freelist
;
2218 stat(s
, ALLOC_SLOWPATH
);
2220 object
= get_freelist(s
, c
->page
);
2224 stat(s
, DEACTIVATE_BYPASS
);
2228 stat(s
, ALLOC_REFILL
);
2231 c
->freelist
= get_freepointer(s
, object
);
2232 c
->tid
= next_tid(c
->tid
);
2233 local_irq_restore(flags
);
2239 c
->page
= c
->partial
;
2240 c
->partial
= c
->page
->next
;
2241 c
->node
= page_to_nid(c
->page
);
2242 stat(s
, CPU_PARTIAL_ALLOC
);
2247 /* Then do expensive stuff like retrieving pages from the partial lists */
2248 object
= get_partial(s
, gfpflags
, node
, c
);
2250 if (unlikely(!object
)) {
2252 object
= new_slab_objects(s
, gfpflags
, node
, &c
);
2254 if (unlikely(!object
)) {
2255 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2256 slab_out_of_memory(s
, gfpflags
, node
);
2258 local_irq_restore(flags
);
2263 if (likely(!kmem_cache_debug(s
)))
2266 /* Only entered in the debug case */
2267 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
2268 goto new_slab
; /* Slab failed checks. Next slab needed */
2270 c
->freelist
= get_freepointer(s
, object
);
2271 deactivate_slab(s
, c
);
2272 c
->node
= NUMA_NO_NODE
;
2273 local_irq_restore(flags
);
2278 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2279 * have the fastpath folded into their functions. So no function call
2280 * overhead for requests that can be satisfied on the fastpath.
2282 * The fastpath works by first checking if the lockless freelist can be used.
2283 * If not then __slab_alloc is called for slow processing.
2285 * Otherwise we can simply pick the next object from the lockless free list.
2287 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2288 gfp_t gfpflags
, int node
, unsigned long addr
)
2291 struct kmem_cache_cpu
*c
;
2294 if (slab_pre_alloc_hook(s
, gfpflags
))
2300 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2301 * enabled. We may switch back and forth between cpus while
2302 * reading from one cpu area. That does not matter as long
2303 * as we end up on the original cpu again when doing the cmpxchg.
2305 c
= __this_cpu_ptr(s
->cpu_slab
);
2308 * The transaction ids are globally unique per cpu and per operation on
2309 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2310 * occurs on the right processor and that there was no operation on the
2311 * linked list in between.
2316 object
= c
->freelist
;
2317 if (unlikely(!object
|| !node_match(c
, node
)))
2319 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
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 get_freepointer_safe(s
, object
), next_tid(tid
)))) {
2339 note_cmpxchg_failure("slab_alloc", s
, tid
);
2342 stat(s
, ALLOC_FASTPATH
);
2345 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2346 memset(object
, 0, s
->objsize
);
2348 slab_post_alloc_hook(s
, gfpflags
, object
);
2353 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2355 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2357 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
2361 EXPORT_SYMBOL(kmem_cache_alloc
);
2363 #ifdef CONFIG_TRACING
2364 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2366 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2367 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2370 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2372 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2374 void *ret
= kmalloc_order(size
, flags
, order
);
2375 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2378 EXPORT_SYMBOL(kmalloc_order_trace
);
2382 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2384 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2386 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2387 s
->objsize
, s
->size
, gfpflags
, node
);
2391 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2393 #ifdef CONFIG_TRACING
2394 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2396 int node
, size_t size
)
2398 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2400 trace_kmalloc_node(_RET_IP_
, ret
,
2401 size
, s
->size
, gfpflags
, node
);
2404 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2409 * Slow patch handling. This may still be called frequently since objects
2410 * have a longer lifetime than the cpu slabs in most processing loads.
2412 * So we still attempt to reduce cache line usage. Just take the slab
2413 * lock and free the item. If there is no additional partial page
2414 * handling required then we can return immediately.
2416 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2417 void *x
, unsigned long addr
)
2420 void **object
= (void *)x
;
2424 unsigned long counters
;
2425 struct kmem_cache_node
*n
= NULL
;
2426 unsigned long uninitialized_var(flags
);
2428 stat(s
, FREE_SLOWPATH
);
2430 if (kmem_cache_debug(s
) && !free_debug_processing(s
, page
, x
, addr
))
2434 prior
= page
->freelist
;
2435 counters
= page
->counters
;
2436 set_freepointer(s
, object
, prior
);
2437 new.counters
= counters
;
2438 was_frozen
= new.frozen
;
2440 if ((!new.inuse
|| !prior
) && !was_frozen
&& !n
) {
2442 if (!kmem_cache_debug(s
) && !prior
)
2445 * Slab was on no list before and will be partially empty
2446 * We can defer the list move and instead freeze it.
2450 else { /* Needs to be taken off a list */
2452 n
= get_node(s
, page_to_nid(page
));
2454 * Speculatively acquire the list_lock.
2455 * If the cmpxchg does not succeed then we may
2456 * drop the list_lock without any processing.
2458 * Otherwise the list_lock will synchronize with
2459 * other processors updating the list of slabs.
2461 spin_lock_irqsave(&n
->list_lock
, flags
);
2467 } while (!cmpxchg_double_slab(s
, page
,
2469 object
, new.counters
,
2475 * If we just froze the page then put it onto the
2476 * per cpu partial list.
2478 if (new.frozen
&& !was_frozen
)
2479 put_cpu_partial(s
, page
, 1);
2482 * The list lock was not taken therefore no list
2483 * activity can be necessary.
2486 stat(s
, FREE_FROZEN
);
2491 * was_frozen may have been set after we acquired the list_lock in
2492 * an earlier loop. So we need to check it here again.
2495 stat(s
, FREE_FROZEN
);
2497 if (unlikely(!inuse
&& n
->nr_partial
> s
->min_partial
))
2501 * Objects left in the slab. If it was not on the partial list before
2504 if (unlikely(!prior
)) {
2505 remove_full(s
, page
);
2506 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2507 stat(s
, FREE_ADD_PARTIAL
);
2510 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2516 * Slab on the partial list.
2518 remove_partial(n
, page
);
2519 stat(s
, FREE_REMOVE_PARTIAL
);
2521 /* Slab must be on the full list */
2522 remove_full(s
, page
);
2524 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2526 discard_slab(s
, page
);
2530 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2531 * can perform fastpath freeing without additional function calls.
2533 * The fastpath is only possible if we are freeing to the current cpu slab
2534 * of this processor. This typically the case if we have just allocated
2537 * If fastpath is not possible then fall back to __slab_free where we deal
2538 * with all sorts of special processing.
2540 static __always_inline
void slab_free(struct kmem_cache
*s
,
2541 struct page
*page
, void *x
, unsigned long addr
)
2543 void **object
= (void *)x
;
2544 struct kmem_cache_cpu
*c
;
2547 slab_free_hook(s
, x
);
2551 * Determine the currently cpus per cpu slab.
2552 * The cpu may change afterward. However that does not matter since
2553 * data is retrieved via this pointer. If we are on the same cpu
2554 * during the cmpxchg then the free will succedd.
2556 c
= __this_cpu_ptr(s
->cpu_slab
);
2561 if (likely(page
== c
->page
)) {
2562 set_freepointer(s
, object
, c
->freelist
);
2564 if (unlikely(!this_cpu_cmpxchg_double(
2565 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2567 object
, next_tid(tid
)))) {
2569 note_cmpxchg_failure("slab_free", s
, tid
);
2572 stat(s
, FREE_FASTPATH
);
2574 __slab_free(s
, page
, x
, addr
);
2578 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2582 page
= virt_to_head_page(x
);
2584 slab_free(s
, page
, x
, _RET_IP_
);
2586 trace_kmem_cache_free(_RET_IP_
, x
);
2588 EXPORT_SYMBOL(kmem_cache_free
);
2591 * Object placement in a slab is made very easy because we always start at
2592 * offset 0. If we tune the size of the object to the alignment then we can
2593 * get the required alignment by putting one properly sized object after
2596 * Notice that the allocation order determines the sizes of the per cpu
2597 * caches. Each processor has always one slab available for allocations.
2598 * Increasing the allocation order reduces the number of times that slabs
2599 * must be moved on and off the partial lists and is therefore a factor in
2604 * Mininum / Maximum order of slab pages. This influences locking overhead
2605 * and slab fragmentation. A higher order reduces the number of partial slabs
2606 * and increases the number of allocations possible without having to
2607 * take the list_lock.
2609 static int slub_min_order
;
2610 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2611 static int slub_min_objects
;
2614 * Merge control. If this is set then no merging of slab caches will occur.
2615 * (Could be removed. This was introduced to pacify the merge skeptics.)
2617 static int slub_nomerge
;
2620 * Calculate the order of allocation given an slab object size.
2622 * The order of allocation has significant impact on performance and other
2623 * system components. Generally order 0 allocations should be preferred since
2624 * order 0 does not cause fragmentation in the page allocator. Larger objects
2625 * be problematic to put into order 0 slabs because there may be too much
2626 * unused space left. We go to a higher order if more than 1/16th of the slab
2629 * In order to reach satisfactory performance we must ensure that a minimum
2630 * number of objects is in one slab. Otherwise we may generate too much
2631 * activity on the partial lists which requires taking the list_lock. This is
2632 * less a concern for large slabs though which are rarely used.
2634 * slub_max_order specifies the order where we begin to stop considering the
2635 * number of objects in a slab as critical. If we reach slub_max_order then
2636 * we try to keep the page order as low as possible. So we accept more waste
2637 * of space in favor of a small page order.
2639 * Higher order allocations also allow the placement of more objects in a
2640 * slab and thereby reduce object handling overhead. If the user has
2641 * requested a higher mininum order then we start with that one instead of
2642 * the smallest order which will fit the object.
2644 static inline int slab_order(int size
, int min_objects
,
2645 int max_order
, int fract_leftover
, int reserved
)
2649 int min_order
= slub_min_order
;
2651 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2652 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2654 for (order
= max(min_order
,
2655 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2656 order
<= max_order
; order
++) {
2658 unsigned long slab_size
= PAGE_SIZE
<< order
;
2660 if (slab_size
< min_objects
* size
+ reserved
)
2663 rem
= (slab_size
- reserved
) % size
;
2665 if (rem
<= slab_size
/ fract_leftover
)
2673 static inline int calculate_order(int size
, int reserved
)
2681 * Attempt to find best configuration for a slab. This
2682 * works by first attempting to generate a layout with
2683 * the best configuration and backing off gradually.
2685 * First we reduce the acceptable waste in a slab. Then
2686 * we reduce the minimum objects required in a slab.
2688 min_objects
= slub_min_objects
;
2690 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2691 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2692 min_objects
= min(min_objects
, max_objects
);
2694 while (min_objects
> 1) {
2696 while (fraction
>= 4) {
2697 order
= slab_order(size
, min_objects
,
2698 slub_max_order
, fraction
, reserved
);
2699 if (order
<= slub_max_order
)
2707 * We were unable to place multiple objects in a slab. Now
2708 * lets see if we can place a single object there.
2710 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2711 if (order
<= slub_max_order
)
2715 * Doh this slab cannot be placed using slub_max_order.
2717 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2718 if (order
< MAX_ORDER
)
2724 * Figure out what the alignment of the objects will be.
2726 static unsigned long calculate_alignment(unsigned long flags
,
2727 unsigned long align
, unsigned long size
)
2730 * If the user wants hardware cache aligned objects then follow that
2731 * suggestion if the object is sufficiently large.
2733 * The hardware cache alignment cannot override the specified
2734 * alignment though. If that is greater then use it.
2736 if (flags
& SLAB_HWCACHE_ALIGN
) {
2737 unsigned long ralign
= cache_line_size();
2738 while (size
<= ralign
/ 2)
2740 align
= max(align
, ralign
);
2743 if (align
< ARCH_SLAB_MINALIGN
)
2744 align
= ARCH_SLAB_MINALIGN
;
2746 return ALIGN(align
, sizeof(void *));
2750 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2753 spin_lock_init(&n
->list_lock
);
2754 INIT_LIST_HEAD(&n
->partial
);
2755 #ifdef CONFIG_SLUB_DEBUG
2756 atomic_long_set(&n
->nr_slabs
, 0);
2757 atomic_long_set(&n
->total_objects
, 0);
2758 INIT_LIST_HEAD(&n
->full
);
2762 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2764 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2765 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2768 * Must align to double word boundary for the double cmpxchg
2769 * instructions to work; see __pcpu_double_call_return_bool().
2771 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2772 2 * sizeof(void *));
2777 init_kmem_cache_cpus(s
);
2782 static struct kmem_cache
*kmem_cache_node
;
2785 * No kmalloc_node yet so do it by hand. We know that this is the first
2786 * slab on the node for this slabcache. There are no concurrent accesses
2789 * Note that this function only works on the kmalloc_node_cache
2790 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2791 * memory on a fresh node that has no slab structures yet.
2793 static void early_kmem_cache_node_alloc(int node
)
2796 struct kmem_cache_node
*n
;
2798 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2800 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2803 if (page_to_nid(page
) != node
) {
2804 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2806 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2807 "in order to be able to continue\n");
2812 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2815 kmem_cache_node
->node
[node
] = n
;
2816 #ifdef CONFIG_SLUB_DEBUG
2817 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2818 init_tracking(kmem_cache_node
, n
);
2820 init_kmem_cache_node(n
, kmem_cache_node
);
2821 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2823 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2826 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2830 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2831 struct kmem_cache_node
*n
= s
->node
[node
];
2834 kmem_cache_free(kmem_cache_node
, n
);
2836 s
->node
[node
] = NULL
;
2840 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2844 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2845 struct kmem_cache_node
*n
;
2847 if (slab_state
== DOWN
) {
2848 early_kmem_cache_node_alloc(node
);
2851 n
= kmem_cache_alloc_node(kmem_cache_node
,
2855 free_kmem_cache_nodes(s
);
2860 init_kmem_cache_node(n
, s
);
2865 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2867 if (min
< MIN_PARTIAL
)
2869 else if (min
> MAX_PARTIAL
)
2871 s
->min_partial
= min
;
2875 * calculate_sizes() determines the order and the distribution of data within
2878 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2880 unsigned long flags
= s
->flags
;
2881 unsigned long size
= s
->objsize
;
2882 unsigned long align
= s
->align
;
2886 * Round up object size to the next word boundary. We can only
2887 * place the free pointer at word boundaries and this determines
2888 * the possible location of the free pointer.
2890 size
= ALIGN(size
, sizeof(void *));
2892 #ifdef CONFIG_SLUB_DEBUG
2894 * Determine if we can poison the object itself. If the user of
2895 * the slab may touch the object after free or before allocation
2896 * then we should never poison the object itself.
2898 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2900 s
->flags
|= __OBJECT_POISON
;
2902 s
->flags
&= ~__OBJECT_POISON
;
2906 * If we are Redzoning then check if there is some space between the
2907 * end of the object and the free pointer. If not then add an
2908 * additional word to have some bytes to store Redzone information.
2910 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2911 size
+= sizeof(void *);
2915 * With that we have determined the number of bytes in actual use
2916 * by the object. This is the potential offset to the free pointer.
2920 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2923 * Relocate free pointer after the object if it is not
2924 * permitted to overwrite the first word of the object on
2927 * This is the case if we do RCU, have a constructor or
2928 * destructor or are poisoning the objects.
2931 size
+= sizeof(void *);
2934 #ifdef CONFIG_SLUB_DEBUG
2935 if (flags
& SLAB_STORE_USER
)
2937 * Need to store information about allocs and frees after
2940 size
+= 2 * sizeof(struct track
);
2942 if (flags
& SLAB_RED_ZONE
)
2944 * Add some empty padding so that we can catch
2945 * overwrites from earlier objects rather than let
2946 * tracking information or the free pointer be
2947 * corrupted if a user writes before the start
2950 size
+= sizeof(void *);
2954 * Determine the alignment based on various parameters that the
2955 * user specified and the dynamic determination of cache line size
2958 align
= calculate_alignment(flags
, align
, s
->objsize
);
2962 * SLUB stores one object immediately after another beginning from
2963 * offset 0. In order to align the objects we have to simply size
2964 * each object to conform to the alignment.
2966 size
= ALIGN(size
, align
);
2968 if (forced_order
>= 0)
2969 order
= forced_order
;
2971 order
= calculate_order(size
, s
->reserved
);
2978 s
->allocflags
|= __GFP_COMP
;
2980 if (s
->flags
& SLAB_CACHE_DMA
)
2981 s
->allocflags
|= SLUB_DMA
;
2983 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2984 s
->allocflags
|= __GFP_RECLAIMABLE
;
2987 * Determine the number of objects per slab
2989 s
->oo
= oo_make(order
, size
, s
->reserved
);
2990 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
2991 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2994 return !!oo_objects(s
->oo
);
2998 static int kmem_cache_open(struct kmem_cache
*s
,
2999 const char *name
, size_t size
,
3000 size_t align
, unsigned long flags
,
3001 void (*ctor
)(void *))
3003 memset(s
, 0, kmem_size
);
3008 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
3011 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3012 s
->reserved
= sizeof(struct rcu_head
);
3014 if (!calculate_sizes(s
, -1))
3016 if (disable_higher_order_debug
) {
3018 * Disable debugging flags that store metadata if the min slab
3021 if (get_order(s
->size
) > get_order(s
->objsize
)) {
3022 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3024 if (!calculate_sizes(s
, -1))
3029 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3030 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3031 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3032 /* Enable fast mode */
3033 s
->flags
|= __CMPXCHG_DOUBLE
;
3037 * The larger the object size is, the more pages we want on the partial
3038 * list to avoid pounding the page allocator excessively.
3040 set_min_partial(s
, ilog2(s
->size
) / 2);
3043 * cpu_partial determined the maximum number of objects kept in the
3044 * per cpu partial lists of a processor.
3046 * Per cpu partial lists mainly contain slabs that just have one
3047 * object freed. If they are used for allocation then they can be
3048 * filled up again with minimal effort. The slab will never hit the
3049 * per node partial lists and therefore no locking will be required.
3051 * This setting also determines
3053 * A) The number of objects from per cpu partial slabs dumped to the
3054 * per node list when we reach the limit.
3055 * B) The number of objects in cpu partial slabs to extract from the
3056 * per node list when we run out of per cpu objects. We only fetch 50%
3057 * to keep some capacity around for frees.
3059 if (kmem_cache_debug(s
))
3061 else if (s
->size
>= PAGE_SIZE
)
3063 else if (s
->size
>= 1024)
3065 else if (s
->size
>= 256)
3066 s
->cpu_partial
= 13;
3068 s
->cpu_partial
= 30;
3072 s
->remote_node_defrag_ratio
= 1000;
3074 if (!init_kmem_cache_nodes(s
))
3077 if (alloc_kmem_cache_cpus(s
))
3080 free_kmem_cache_nodes(s
);
3082 if (flags
& SLAB_PANIC
)
3083 panic("Cannot create slab %s size=%lu realsize=%u "
3084 "order=%u offset=%u flags=%lx\n",
3085 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
3091 * Determine the size of a slab object
3093 unsigned int kmem_cache_size(struct kmem_cache
*s
)
3097 EXPORT_SYMBOL(kmem_cache_size
);
3099 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3102 #ifdef CONFIG_SLUB_DEBUG
3103 void *addr
= page_address(page
);
3105 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3106 sizeof(long), GFP_ATOMIC
);
3109 slab_err(s
, page
, "%s", text
);
3112 get_map(s
, page
, map
);
3113 for_each_object(p
, s
, addr
, page
->objects
) {
3115 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3116 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3118 print_tracking(s
, p
);
3127 * Attempt to free all partial slabs on a node.
3128 * This is called from kmem_cache_close(). We must be the last thread
3129 * using the cache and therefore we do not need to lock anymore.
3131 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3133 struct page
*page
, *h
;
3135 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3137 remove_partial(n
, page
);
3138 discard_slab(s
, page
);
3140 list_slab_objects(s
, page
,
3141 "Objects remaining on kmem_cache_close()");
3147 * Release all resources used by a slab cache.
3149 static inline int kmem_cache_close(struct kmem_cache
*s
)
3154 free_percpu(s
->cpu_slab
);
3155 /* Attempt to free all objects */
3156 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3157 struct kmem_cache_node
*n
= get_node(s
, node
);
3160 if (n
->nr_partial
|| slabs_node(s
, node
))
3163 free_kmem_cache_nodes(s
);
3168 * Close a cache and release the kmem_cache structure
3169 * (must be used for caches created using kmem_cache_create)
3171 void kmem_cache_destroy(struct kmem_cache
*s
)
3173 down_write(&slub_lock
);
3177 up_write(&slub_lock
);
3178 if (kmem_cache_close(s
)) {
3179 printk(KERN_ERR
"SLUB %s: %s called for cache that "
3180 "still has objects.\n", s
->name
, __func__
);
3183 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
3185 sysfs_slab_remove(s
);
3187 up_write(&slub_lock
);
3189 EXPORT_SYMBOL(kmem_cache_destroy
);
3191 /********************************************************************
3193 *******************************************************************/
3195 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
3196 EXPORT_SYMBOL(kmalloc_caches
);
3198 static struct kmem_cache
*kmem_cache
;
3200 #ifdef CONFIG_ZONE_DMA
3201 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
3204 static int __init
setup_slub_min_order(char *str
)
3206 get_option(&str
, &slub_min_order
);
3211 __setup("slub_min_order=", setup_slub_min_order
);
3213 static int __init
setup_slub_max_order(char *str
)
3215 get_option(&str
, &slub_max_order
);
3216 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3221 __setup("slub_max_order=", setup_slub_max_order
);
3223 static int __init
setup_slub_min_objects(char *str
)
3225 get_option(&str
, &slub_min_objects
);
3230 __setup("slub_min_objects=", setup_slub_min_objects
);
3232 static int __init
setup_slub_nomerge(char *str
)
3238 __setup("slub_nomerge", setup_slub_nomerge
);
3240 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
3241 int size
, unsigned int flags
)
3243 struct kmem_cache
*s
;
3245 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3248 * This function is called with IRQs disabled during early-boot on
3249 * single CPU so there's no need to take slub_lock here.
3251 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
3255 list_add(&s
->list
, &slab_caches
);
3259 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
3264 * Conversion table for small slabs sizes / 8 to the index in the
3265 * kmalloc array. This is necessary for slabs < 192 since we have non power
3266 * of two cache sizes there. The size of larger slabs can be determined using
3269 static s8 size_index
[24] = {
3296 static inline int size_index_elem(size_t bytes
)
3298 return (bytes
- 1) / 8;
3301 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
3307 return ZERO_SIZE_PTR
;
3309 index
= size_index
[size_index_elem(size
)];
3311 index
= fls(size
- 1);
3313 #ifdef CONFIG_ZONE_DMA
3314 if (unlikely((flags
& SLUB_DMA
)))
3315 return kmalloc_dma_caches
[index
];
3318 return kmalloc_caches
[index
];
3321 void *__kmalloc(size_t size
, gfp_t flags
)
3323 struct kmem_cache
*s
;
3326 if (unlikely(size
> SLUB_MAX_SIZE
))
3327 return kmalloc_large(size
, flags
);
3329 s
= get_slab(size
, flags
);
3331 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3334 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
3336 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3340 EXPORT_SYMBOL(__kmalloc
);
3343 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3348 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3349 page
= alloc_pages_node(node
, flags
, get_order(size
));
3351 ptr
= page_address(page
);
3353 kmemleak_alloc(ptr
, size
, 1, flags
);
3357 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3359 struct kmem_cache
*s
;
3362 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3363 ret
= kmalloc_large_node(size
, flags
, node
);
3365 trace_kmalloc_node(_RET_IP_
, ret
,
3366 size
, PAGE_SIZE
<< get_order(size
),
3372 s
= get_slab(size
, flags
);
3374 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3377 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
3379 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3383 EXPORT_SYMBOL(__kmalloc_node
);
3386 size_t ksize(const void *object
)
3390 if (unlikely(object
== ZERO_SIZE_PTR
))
3393 page
= virt_to_head_page(object
);
3395 if (unlikely(!PageSlab(page
))) {
3396 WARN_ON(!PageCompound(page
));
3397 return PAGE_SIZE
<< compound_order(page
);
3400 return slab_ksize(page
->slab
);
3402 EXPORT_SYMBOL(ksize
);
3404 #ifdef CONFIG_SLUB_DEBUG
3405 bool verify_mem_not_deleted(const void *x
)
3408 void *object
= (void *)x
;
3409 unsigned long flags
;
3412 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3415 local_irq_save(flags
);
3417 page
= virt_to_head_page(x
);
3418 if (unlikely(!PageSlab(page
))) {
3419 /* maybe it was from stack? */
3425 if (on_freelist(page
->slab
, page
, object
)) {
3426 object_err(page
->slab
, page
, object
, "Object is on free-list");
3434 local_irq_restore(flags
);
3437 EXPORT_SYMBOL(verify_mem_not_deleted
);
3440 void kfree(const void *x
)
3443 void *object
= (void *)x
;
3445 trace_kfree(_RET_IP_
, x
);
3447 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3450 page
= virt_to_head_page(x
);
3451 if (unlikely(!PageSlab(page
))) {
3452 BUG_ON(!PageCompound(page
));
3457 slab_free(page
->slab
, page
, object
, _RET_IP_
);
3459 EXPORT_SYMBOL(kfree
);
3462 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3463 * the remaining slabs by the number of items in use. The slabs with the
3464 * most items in use come first. New allocations will then fill those up
3465 * and thus they can be removed from the partial lists.
3467 * The slabs with the least items are placed last. This results in them
3468 * being allocated from last increasing the chance that the last objects
3469 * are freed in them.
3471 int kmem_cache_shrink(struct kmem_cache
*s
)
3475 struct kmem_cache_node
*n
;
3478 int objects
= oo_objects(s
->max
);
3479 struct list_head
*slabs_by_inuse
=
3480 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3481 unsigned long flags
;
3483 if (!slabs_by_inuse
)
3487 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3488 n
= get_node(s
, node
);
3493 for (i
= 0; i
< objects
; i
++)
3494 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3496 spin_lock_irqsave(&n
->list_lock
, flags
);
3499 * Build lists indexed by the items in use in each slab.
3501 * Note that concurrent frees may occur while we hold the
3502 * list_lock. page->inuse here is the upper limit.
3504 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3505 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3511 * Rebuild the partial list with the slabs filled up most
3512 * first and the least used slabs at the end.
3514 for (i
= objects
- 1; i
> 0; i
--)
3515 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3517 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3519 /* Release empty slabs */
3520 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3521 discard_slab(s
, page
);
3524 kfree(slabs_by_inuse
);
3527 EXPORT_SYMBOL(kmem_cache_shrink
);
3529 #if defined(CONFIG_MEMORY_HOTPLUG)
3530 static int slab_mem_going_offline_callback(void *arg
)
3532 struct kmem_cache
*s
;
3534 down_read(&slub_lock
);
3535 list_for_each_entry(s
, &slab_caches
, list
)
3536 kmem_cache_shrink(s
);
3537 up_read(&slub_lock
);
3542 static void slab_mem_offline_callback(void *arg
)
3544 struct kmem_cache_node
*n
;
3545 struct kmem_cache
*s
;
3546 struct memory_notify
*marg
= arg
;
3549 offline_node
= marg
->status_change_nid
;
3552 * If the node still has available memory. we need kmem_cache_node
3555 if (offline_node
< 0)
3558 down_read(&slub_lock
);
3559 list_for_each_entry(s
, &slab_caches
, list
) {
3560 n
= get_node(s
, offline_node
);
3563 * if n->nr_slabs > 0, slabs still exist on the node
3564 * that is going down. We were unable to free them,
3565 * and offline_pages() function shouldn't call this
3566 * callback. So, we must fail.
3568 BUG_ON(slabs_node(s
, offline_node
));
3570 s
->node
[offline_node
] = NULL
;
3571 kmem_cache_free(kmem_cache_node
, n
);
3574 up_read(&slub_lock
);
3577 static int slab_mem_going_online_callback(void *arg
)
3579 struct kmem_cache_node
*n
;
3580 struct kmem_cache
*s
;
3581 struct memory_notify
*marg
= arg
;
3582 int nid
= marg
->status_change_nid
;
3586 * If the node's memory is already available, then kmem_cache_node is
3587 * already created. Nothing to do.
3593 * We are bringing a node online. No memory is available yet. We must
3594 * allocate a kmem_cache_node structure in order to bring the node
3597 down_read(&slub_lock
);
3598 list_for_each_entry(s
, &slab_caches
, list
) {
3600 * XXX: kmem_cache_alloc_node will fallback to other nodes
3601 * since memory is not yet available from the node that
3604 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3609 init_kmem_cache_node(n
, s
);
3613 up_read(&slub_lock
);
3617 static int slab_memory_callback(struct notifier_block
*self
,
3618 unsigned long action
, void *arg
)
3623 case MEM_GOING_ONLINE
:
3624 ret
= slab_mem_going_online_callback(arg
);
3626 case MEM_GOING_OFFLINE
:
3627 ret
= slab_mem_going_offline_callback(arg
);
3630 case MEM_CANCEL_ONLINE
:
3631 slab_mem_offline_callback(arg
);
3634 case MEM_CANCEL_OFFLINE
:
3638 ret
= notifier_from_errno(ret
);
3644 #endif /* CONFIG_MEMORY_HOTPLUG */
3646 /********************************************************************
3647 * Basic setup of slabs
3648 *******************************************************************/
3651 * Used for early kmem_cache structures that were allocated using
3652 * the page allocator
3655 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3659 list_add(&s
->list
, &slab_caches
);
3662 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3663 struct kmem_cache_node
*n
= get_node(s
, node
);
3667 list_for_each_entry(p
, &n
->partial
, lru
)
3670 #ifdef CONFIG_SLUB_DEBUG
3671 list_for_each_entry(p
, &n
->full
, lru
)
3678 void __init
kmem_cache_init(void)
3682 struct kmem_cache
*temp_kmem_cache
;
3684 struct kmem_cache
*temp_kmem_cache_node
;
3685 unsigned long kmalloc_size
;
3687 if (debug_guardpage_minorder())
3690 kmem_size
= offsetof(struct kmem_cache
, node
) +
3691 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3693 /* Allocate two kmem_caches from the page allocator */
3694 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3695 order
= get_order(2 * kmalloc_size
);
3696 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3699 * Must first have the slab cache available for the allocations of the
3700 * struct kmem_cache_node's. There is special bootstrap code in
3701 * kmem_cache_open for slab_state == DOWN.
3703 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3705 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3706 sizeof(struct kmem_cache_node
),
3707 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3709 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3711 /* Able to allocate the per node structures */
3712 slab_state
= PARTIAL
;
3714 temp_kmem_cache
= kmem_cache
;
3715 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3716 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3717 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3718 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3721 * Allocate kmem_cache_node properly from the kmem_cache slab.
3722 * kmem_cache_node is separately allocated so no need to
3723 * update any list pointers.
3725 temp_kmem_cache_node
= kmem_cache_node
;
3727 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3728 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3730 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3733 kmem_cache_bootstrap_fixup(kmem_cache
);
3735 /* Free temporary boot structure */
3736 free_pages((unsigned long)temp_kmem_cache
, order
);
3738 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3741 * Patch up the size_index table if we have strange large alignment
3742 * requirements for the kmalloc array. This is only the case for
3743 * MIPS it seems. The standard arches will not generate any code here.
3745 * Largest permitted alignment is 256 bytes due to the way we
3746 * handle the index determination for the smaller caches.
3748 * Make sure that nothing crazy happens if someone starts tinkering
3749 * around with ARCH_KMALLOC_MINALIGN
3751 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3752 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3754 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3755 int elem
= size_index_elem(i
);
3756 if (elem
>= ARRAY_SIZE(size_index
))
3758 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3761 if (KMALLOC_MIN_SIZE
== 64) {
3763 * The 96 byte size cache is not used if the alignment
3766 for (i
= 64 + 8; i
<= 96; i
+= 8)
3767 size_index
[size_index_elem(i
)] = 7;
3768 } else if (KMALLOC_MIN_SIZE
== 128) {
3770 * The 192 byte sized cache is not used if the alignment
3771 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3774 for (i
= 128 + 8; i
<= 192; i
+= 8)
3775 size_index
[size_index_elem(i
)] = 8;
3778 /* Caches that are not of the two-to-the-power-of size */
3779 if (KMALLOC_MIN_SIZE
<= 32) {
3780 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3784 if (KMALLOC_MIN_SIZE
<= 64) {
3785 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3789 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3790 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3796 /* Provide the correct kmalloc names now that the caches are up */
3797 if (KMALLOC_MIN_SIZE
<= 32) {
3798 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3799 BUG_ON(!kmalloc_caches
[1]->name
);
3802 if (KMALLOC_MIN_SIZE
<= 64) {
3803 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3804 BUG_ON(!kmalloc_caches
[2]->name
);
3807 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3808 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3811 kmalloc_caches
[i
]->name
= s
;
3815 register_cpu_notifier(&slab_notifier
);
3818 #ifdef CONFIG_ZONE_DMA
3819 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3820 struct kmem_cache
*s
= kmalloc_caches
[i
];
3823 char *name
= kasprintf(GFP_NOWAIT
,
3824 "dma-kmalloc-%d", s
->objsize
);
3827 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3828 s
->objsize
, SLAB_CACHE_DMA
);
3833 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3834 " CPUs=%d, Nodes=%d\n",
3835 caches
, cache_line_size(),
3836 slub_min_order
, slub_max_order
, slub_min_objects
,
3837 nr_cpu_ids
, nr_node_ids
);
3840 void __init
kmem_cache_init_late(void)
3845 * Find a mergeable slab cache
3847 static int slab_unmergeable(struct kmem_cache
*s
)
3849 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3856 * We may have set a slab to be unmergeable during bootstrap.
3858 if (s
->refcount
< 0)
3864 static struct kmem_cache
*find_mergeable(size_t size
,
3865 size_t align
, unsigned long flags
, const char *name
,
3866 void (*ctor
)(void *))
3868 struct kmem_cache
*s
;
3870 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3876 size
= ALIGN(size
, sizeof(void *));
3877 align
= calculate_alignment(flags
, align
, size
);
3878 size
= ALIGN(size
, align
);
3879 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3881 list_for_each_entry(s
, &slab_caches
, list
) {
3882 if (slab_unmergeable(s
))
3888 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3891 * Check if alignment is compatible.
3892 * Courtesy of Adrian Drzewiecki
3894 if ((s
->size
& ~(align
- 1)) != s
->size
)
3897 if (s
->size
- size
>= sizeof(void *))
3905 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3906 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3908 struct kmem_cache
*s
;
3914 down_write(&slub_lock
);
3915 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3919 * Adjust the object sizes so that we clear
3920 * the complete object on kzalloc.
3922 s
->objsize
= max(s
->objsize
, (int)size
);
3923 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3925 if (sysfs_slab_alias(s
, name
)) {
3929 up_write(&slub_lock
);
3933 n
= kstrdup(name
, GFP_KERNEL
);
3937 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3939 if (kmem_cache_open(s
, n
,
3940 size
, align
, flags
, ctor
)) {
3941 list_add(&s
->list
, &slab_caches
);
3942 if (sysfs_slab_add(s
)) {
3948 up_write(&slub_lock
);
3955 up_write(&slub_lock
);
3957 if (flags
& SLAB_PANIC
)
3958 panic("Cannot create slabcache %s\n", name
);
3963 EXPORT_SYMBOL(kmem_cache_create
);
3967 * Use the cpu notifier to insure that the cpu slabs are flushed when
3970 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3971 unsigned long action
, void *hcpu
)
3973 long cpu
= (long)hcpu
;
3974 struct kmem_cache
*s
;
3975 unsigned long flags
;
3978 case CPU_UP_CANCELED
:
3979 case CPU_UP_CANCELED_FROZEN
:
3981 case CPU_DEAD_FROZEN
:
3982 down_read(&slub_lock
);
3983 list_for_each_entry(s
, &slab_caches
, list
) {
3984 local_irq_save(flags
);
3985 __flush_cpu_slab(s
, cpu
);
3986 local_irq_restore(flags
);
3988 up_read(&slub_lock
);
3996 static struct notifier_block __cpuinitdata slab_notifier
= {
3997 .notifier_call
= slab_cpuup_callback
4002 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
4004 struct kmem_cache
*s
;
4007 if (unlikely(size
> SLUB_MAX_SIZE
))
4008 return kmalloc_large(size
, gfpflags
);
4010 s
= get_slab(size
, gfpflags
);
4012 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4015 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
4017 /* Honor the call site pointer we received. */
4018 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
4024 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
4025 int node
, unsigned long caller
)
4027 struct kmem_cache
*s
;
4030 if (unlikely(size
> SLUB_MAX_SIZE
)) {
4031 ret
= kmalloc_large_node(size
, gfpflags
, node
);
4033 trace_kmalloc_node(caller
, ret
,
4034 size
, PAGE_SIZE
<< get_order(size
),
4040 s
= get_slab(size
, gfpflags
);
4042 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4045 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
4047 /* Honor the call site pointer we received. */
4048 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
4055 static int count_inuse(struct page
*page
)
4060 static int count_total(struct page
*page
)
4062 return page
->objects
;
4066 #ifdef CONFIG_SLUB_DEBUG
4067 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
4071 void *addr
= page_address(page
);
4073 if (!check_slab(s
, page
) ||
4074 !on_freelist(s
, page
, NULL
))
4077 /* Now we know that a valid freelist exists */
4078 bitmap_zero(map
, page
->objects
);
4080 get_map(s
, page
, map
);
4081 for_each_object(p
, s
, addr
, page
->objects
) {
4082 if (test_bit(slab_index(p
, s
, addr
), map
))
4083 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
4087 for_each_object(p
, s
, addr
, page
->objects
)
4088 if (!test_bit(slab_index(p
, s
, addr
), map
))
4089 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
4094 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
4098 validate_slab(s
, page
, map
);
4102 static int validate_slab_node(struct kmem_cache
*s
,
4103 struct kmem_cache_node
*n
, unsigned long *map
)
4105 unsigned long count
= 0;
4107 unsigned long flags
;
4109 spin_lock_irqsave(&n
->list_lock
, flags
);
4111 list_for_each_entry(page
, &n
->partial
, lru
) {
4112 validate_slab_slab(s
, page
, map
);
4115 if (count
!= n
->nr_partial
)
4116 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
4117 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
4119 if (!(s
->flags
& SLAB_STORE_USER
))
4122 list_for_each_entry(page
, &n
->full
, lru
) {
4123 validate_slab_slab(s
, page
, map
);
4126 if (count
!= atomic_long_read(&n
->nr_slabs
))
4127 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
4128 "counter=%ld\n", s
->name
, count
,
4129 atomic_long_read(&n
->nr_slabs
));
4132 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4136 static long validate_slab_cache(struct kmem_cache
*s
)
4139 unsigned long count
= 0;
4140 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4141 sizeof(unsigned long), GFP_KERNEL
);
4147 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4148 struct kmem_cache_node
*n
= get_node(s
, node
);
4150 count
+= validate_slab_node(s
, n
, map
);
4156 * Generate lists of code addresses where slabcache objects are allocated
4161 unsigned long count
;
4168 DECLARE_BITMAP(cpus
, NR_CPUS
);
4174 unsigned long count
;
4175 struct location
*loc
;
4178 static void free_loc_track(struct loc_track
*t
)
4181 free_pages((unsigned long)t
->loc
,
4182 get_order(sizeof(struct location
) * t
->max
));
4185 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4190 order
= get_order(sizeof(struct location
) * max
);
4192 l
= (void *)__get_free_pages(flags
, order
);
4197 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4205 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4206 const struct track
*track
)
4208 long start
, end
, pos
;
4210 unsigned long caddr
;
4211 unsigned long age
= jiffies
- track
->when
;
4217 pos
= start
+ (end
- start
+ 1) / 2;
4220 * There is nothing at "end". If we end up there
4221 * we need to add something to before end.
4226 caddr
= t
->loc
[pos
].addr
;
4227 if (track
->addr
== caddr
) {
4233 if (age
< l
->min_time
)
4235 if (age
> l
->max_time
)
4238 if (track
->pid
< l
->min_pid
)
4239 l
->min_pid
= track
->pid
;
4240 if (track
->pid
> l
->max_pid
)
4241 l
->max_pid
= track
->pid
;
4243 cpumask_set_cpu(track
->cpu
,
4244 to_cpumask(l
->cpus
));
4246 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4250 if (track
->addr
< caddr
)
4257 * Not found. Insert new tracking element.
4259 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4265 (t
->count
- pos
) * sizeof(struct location
));
4268 l
->addr
= track
->addr
;
4272 l
->min_pid
= track
->pid
;
4273 l
->max_pid
= track
->pid
;
4274 cpumask_clear(to_cpumask(l
->cpus
));
4275 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4276 nodes_clear(l
->nodes
);
4277 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4281 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4282 struct page
*page
, enum track_item alloc
,
4285 void *addr
= page_address(page
);
4288 bitmap_zero(map
, page
->objects
);
4289 get_map(s
, page
, map
);
4291 for_each_object(p
, s
, addr
, page
->objects
)
4292 if (!test_bit(slab_index(p
, s
, addr
), map
))
4293 add_location(t
, s
, get_track(s
, p
, alloc
));
4296 static int list_locations(struct kmem_cache
*s
, char *buf
,
4297 enum track_item alloc
)
4301 struct loc_track t
= { 0, 0, NULL
};
4303 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4304 sizeof(unsigned long), GFP_KERNEL
);
4306 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4309 return sprintf(buf
, "Out of memory\n");
4311 /* Push back cpu slabs */
4314 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4315 struct kmem_cache_node
*n
= get_node(s
, node
);
4316 unsigned long flags
;
4319 if (!atomic_long_read(&n
->nr_slabs
))
4322 spin_lock_irqsave(&n
->list_lock
, flags
);
4323 list_for_each_entry(page
, &n
->partial
, lru
)
4324 process_slab(&t
, s
, page
, alloc
, map
);
4325 list_for_each_entry(page
, &n
->full
, lru
)
4326 process_slab(&t
, s
, page
, alloc
, map
);
4327 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4330 for (i
= 0; i
< t
.count
; i
++) {
4331 struct location
*l
= &t
.loc
[i
];
4333 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4335 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4338 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4340 len
+= sprintf(buf
+ len
, "<not-available>");
4342 if (l
->sum_time
!= l
->min_time
) {
4343 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4345 (long)div_u64(l
->sum_time
, l
->count
),
4348 len
+= sprintf(buf
+ len
, " age=%ld",
4351 if (l
->min_pid
!= l
->max_pid
)
4352 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4353 l
->min_pid
, l
->max_pid
);
4355 len
+= sprintf(buf
+ len
, " pid=%ld",
4358 if (num_online_cpus() > 1 &&
4359 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4360 len
< PAGE_SIZE
- 60) {
4361 len
+= sprintf(buf
+ len
, " cpus=");
4362 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4363 to_cpumask(l
->cpus
));
4366 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4367 len
< PAGE_SIZE
- 60) {
4368 len
+= sprintf(buf
+ len
, " nodes=");
4369 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4373 len
+= sprintf(buf
+ len
, "\n");
4379 len
+= sprintf(buf
, "No data\n");
4384 #ifdef SLUB_RESILIENCY_TEST
4385 static void resiliency_test(void)
4389 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
4391 printk(KERN_ERR
"SLUB resiliency testing\n");
4392 printk(KERN_ERR
"-----------------------\n");
4393 printk(KERN_ERR
"A. Corruption after allocation\n");
4395 p
= kzalloc(16, GFP_KERNEL
);
4397 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4398 " 0x12->0x%p\n\n", p
+ 16);
4400 validate_slab_cache(kmalloc_caches
[4]);
4402 /* Hmmm... The next two are dangerous */
4403 p
= kzalloc(32, GFP_KERNEL
);
4404 p
[32 + sizeof(void *)] = 0x34;
4405 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4406 " 0x34 -> -0x%p\n", p
);
4408 "If allocated object is overwritten then not detectable\n\n");
4410 validate_slab_cache(kmalloc_caches
[5]);
4411 p
= kzalloc(64, GFP_KERNEL
);
4412 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4414 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4417 "If allocated object is overwritten then not detectable\n\n");
4418 validate_slab_cache(kmalloc_caches
[6]);
4420 printk(KERN_ERR
"\nB. Corruption after free\n");
4421 p
= kzalloc(128, GFP_KERNEL
);
4424 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4425 validate_slab_cache(kmalloc_caches
[7]);
4427 p
= kzalloc(256, GFP_KERNEL
);
4430 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4432 validate_slab_cache(kmalloc_caches
[8]);
4434 p
= kzalloc(512, GFP_KERNEL
);
4437 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4438 validate_slab_cache(kmalloc_caches
[9]);
4442 static void resiliency_test(void) {};
4447 enum slab_stat_type
{
4448 SL_ALL
, /* All slabs */
4449 SL_PARTIAL
, /* Only partially allocated slabs */
4450 SL_CPU
, /* Only slabs used for cpu caches */
4451 SL_OBJECTS
, /* Determine allocated objects not slabs */
4452 SL_TOTAL
/* Determine object capacity not slabs */
4455 #define SO_ALL (1 << SL_ALL)
4456 #define SO_PARTIAL (1 << SL_PARTIAL)
4457 #define SO_CPU (1 << SL_CPU)
4458 #define SO_OBJECTS (1 << SL_OBJECTS)
4459 #define SO_TOTAL (1 << SL_TOTAL)
4461 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4462 char *buf
, unsigned long flags
)
4464 unsigned long total
= 0;
4467 unsigned long *nodes
;
4468 unsigned long *per_cpu
;
4470 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4473 per_cpu
= nodes
+ nr_node_ids
;
4475 if (flags
& SO_CPU
) {
4478 for_each_possible_cpu(cpu
) {
4479 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4480 int node
= ACCESS_ONCE(c
->node
);
4485 page
= ACCESS_ONCE(c
->page
);
4487 if (flags
& SO_TOTAL
)
4489 else if (flags
& SO_OBJECTS
)
4508 lock_memory_hotplug();
4509 #ifdef CONFIG_SLUB_DEBUG
4510 if (flags
& SO_ALL
) {
4511 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4512 struct kmem_cache_node
*n
= get_node(s
, node
);
4514 if (flags
& SO_TOTAL
)
4515 x
= atomic_long_read(&n
->total_objects
);
4516 else if (flags
& SO_OBJECTS
)
4517 x
= atomic_long_read(&n
->total_objects
) -
4518 count_partial(n
, count_free
);
4521 x
= atomic_long_read(&n
->nr_slabs
);
4528 if (flags
& SO_PARTIAL
) {
4529 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4530 struct kmem_cache_node
*n
= get_node(s
, node
);
4532 if (flags
& SO_TOTAL
)
4533 x
= count_partial(n
, count_total
);
4534 else if (flags
& SO_OBJECTS
)
4535 x
= count_partial(n
, count_inuse
);
4542 x
= sprintf(buf
, "%lu", total
);
4544 for_each_node_state(node
, N_NORMAL_MEMORY
)
4546 x
+= sprintf(buf
+ x
, " N%d=%lu",
4549 unlock_memory_hotplug();
4551 return x
+ sprintf(buf
+ x
, "\n");
4554 #ifdef CONFIG_SLUB_DEBUG
4555 static int any_slab_objects(struct kmem_cache
*s
)
4559 for_each_online_node(node
) {
4560 struct kmem_cache_node
*n
= get_node(s
, node
);
4565 if (atomic_long_read(&n
->total_objects
))
4572 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4573 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4575 struct slab_attribute
{
4576 struct attribute attr
;
4577 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4578 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4581 #define SLAB_ATTR_RO(_name) \
4582 static struct slab_attribute _name##_attr = \
4583 __ATTR(_name, 0400, _name##_show, NULL)
4585 #define SLAB_ATTR(_name) \
4586 static struct slab_attribute _name##_attr = \
4587 __ATTR(_name, 0600, _name##_show, _name##_store)
4589 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4591 return sprintf(buf
, "%d\n", s
->size
);
4593 SLAB_ATTR_RO(slab_size
);
4595 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4597 return sprintf(buf
, "%d\n", s
->align
);
4599 SLAB_ATTR_RO(align
);
4601 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4603 return sprintf(buf
, "%d\n", s
->objsize
);
4605 SLAB_ATTR_RO(object_size
);
4607 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4609 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4611 SLAB_ATTR_RO(objs_per_slab
);
4613 static ssize_t
order_store(struct kmem_cache
*s
,
4614 const char *buf
, size_t length
)
4616 unsigned long order
;
4619 err
= strict_strtoul(buf
, 10, &order
);
4623 if (order
> slub_max_order
|| order
< slub_min_order
)
4626 calculate_sizes(s
, order
);
4630 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4632 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4636 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4638 return sprintf(buf
, "%lu\n", s
->min_partial
);
4641 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4647 err
= strict_strtoul(buf
, 10, &min
);
4651 set_min_partial(s
, min
);
4654 SLAB_ATTR(min_partial
);
4656 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4658 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4661 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4664 unsigned long objects
;
4667 err
= strict_strtoul(buf
, 10, &objects
);
4670 if (objects
&& kmem_cache_debug(s
))
4673 s
->cpu_partial
= objects
;
4677 SLAB_ATTR(cpu_partial
);
4679 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4683 return sprintf(buf
, "%pS\n", s
->ctor
);
4687 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4689 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4691 SLAB_ATTR_RO(aliases
);
4693 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4695 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4697 SLAB_ATTR_RO(partial
);
4699 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4701 return show_slab_objects(s
, buf
, SO_CPU
);
4703 SLAB_ATTR_RO(cpu_slabs
);
4705 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4707 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4709 SLAB_ATTR_RO(objects
);
4711 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4713 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4715 SLAB_ATTR_RO(objects_partial
);
4717 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4724 for_each_online_cpu(cpu
) {
4725 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4728 pages
+= page
->pages
;
4729 objects
+= page
->pobjects
;
4733 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4736 for_each_online_cpu(cpu
) {
4737 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4739 if (page
&& len
< PAGE_SIZE
- 20)
4740 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4741 page
->pobjects
, page
->pages
);
4744 return len
+ sprintf(buf
+ len
, "\n");
4746 SLAB_ATTR_RO(slabs_cpu_partial
);
4748 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4750 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4753 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4754 const char *buf
, size_t length
)
4756 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4758 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4761 SLAB_ATTR(reclaim_account
);
4763 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4765 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4767 SLAB_ATTR_RO(hwcache_align
);
4769 #ifdef CONFIG_ZONE_DMA
4770 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4772 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4774 SLAB_ATTR_RO(cache_dma
);
4777 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4779 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4781 SLAB_ATTR_RO(destroy_by_rcu
);
4783 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4785 return sprintf(buf
, "%d\n", s
->reserved
);
4787 SLAB_ATTR_RO(reserved
);
4789 #ifdef CONFIG_SLUB_DEBUG
4790 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4792 return show_slab_objects(s
, buf
, SO_ALL
);
4794 SLAB_ATTR_RO(slabs
);
4796 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4798 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4800 SLAB_ATTR_RO(total_objects
);
4802 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4804 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4807 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4808 const char *buf
, size_t length
)
4810 s
->flags
&= ~SLAB_DEBUG_FREE
;
4811 if (buf
[0] == '1') {
4812 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4813 s
->flags
|= SLAB_DEBUG_FREE
;
4817 SLAB_ATTR(sanity_checks
);
4819 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4821 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4824 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4827 s
->flags
&= ~SLAB_TRACE
;
4828 if (buf
[0] == '1') {
4829 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4830 s
->flags
|= SLAB_TRACE
;
4836 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4838 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4841 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4842 const char *buf
, size_t length
)
4844 if (any_slab_objects(s
))
4847 s
->flags
&= ~SLAB_RED_ZONE
;
4848 if (buf
[0] == '1') {
4849 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4850 s
->flags
|= SLAB_RED_ZONE
;
4852 calculate_sizes(s
, -1);
4855 SLAB_ATTR(red_zone
);
4857 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4859 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4862 static ssize_t
poison_store(struct kmem_cache
*s
,
4863 const char *buf
, size_t length
)
4865 if (any_slab_objects(s
))
4868 s
->flags
&= ~SLAB_POISON
;
4869 if (buf
[0] == '1') {
4870 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4871 s
->flags
|= SLAB_POISON
;
4873 calculate_sizes(s
, -1);
4878 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4880 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4883 static ssize_t
store_user_store(struct kmem_cache
*s
,
4884 const char *buf
, size_t length
)
4886 if (any_slab_objects(s
))
4889 s
->flags
&= ~SLAB_STORE_USER
;
4890 if (buf
[0] == '1') {
4891 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4892 s
->flags
|= SLAB_STORE_USER
;
4894 calculate_sizes(s
, -1);
4897 SLAB_ATTR(store_user
);
4899 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4904 static ssize_t
validate_store(struct kmem_cache
*s
,
4905 const char *buf
, size_t length
)
4909 if (buf
[0] == '1') {
4910 ret
= validate_slab_cache(s
);
4916 SLAB_ATTR(validate
);
4918 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4920 if (!(s
->flags
& SLAB_STORE_USER
))
4922 return list_locations(s
, buf
, TRACK_ALLOC
);
4924 SLAB_ATTR_RO(alloc_calls
);
4926 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4928 if (!(s
->flags
& SLAB_STORE_USER
))
4930 return list_locations(s
, buf
, TRACK_FREE
);
4932 SLAB_ATTR_RO(free_calls
);
4933 #endif /* CONFIG_SLUB_DEBUG */
4935 #ifdef CONFIG_FAILSLAB
4936 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4938 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4941 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4944 s
->flags
&= ~SLAB_FAILSLAB
;
4946 s
->flags
|= SLAB_FAILSLAB
;
4949 SLAB_ATTR(failslab
);
4952 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4957 static ssize_t
shrink_store(struct kmem_cache
*s
,
4958 const char *buf
, size_t length
)
4960 if (buf
[0] == '1') {
4961 int rc
= kmem_cache_shrink(s
);
4972 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4974 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4977 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4978 const char *buf
, size_t length
)
4980 unsigned long ratio
;
4983 err
= strict_strtoul(buf
, 10, &ratio
);
4988 s
->remote_node_defrag_ratio
= ratio
* 10;
4992 SLAB_ATTR(remote_node_defrag_ratio
);
4995 #ifdef CONFIG_SLUB_STATS
4996 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4998 unsigned long sum
= 0;
5001 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
5006 for_each_online_cpu(cpu
) {
5007 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
5013 len
= sprintf(buf
, "%lu", sum
);
5016 for_each_online_cpu(cpu
) {
5017 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
5018 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
5022 return len
+ sprintf(buf
+ len
, "\n");
5025 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
5029 for_each_online_cpu(cpu
)
5030 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
5033 #define STAT_ATTR(si, text) \
5034 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5036 return show_stat(s, buf, si); \
5038 static ssize_t text##_store(struct kmem_cache *s, \
5039 const char *buf, size_t length) \
5041 if (buf[0] != '0') \
5043 clear_stat(s, si); \
5048 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5049 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
5050 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
5051 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
5052 STAT_ATTR(FREE_FROZEN
, free_frozen
);
5053 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
5054 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
5055 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
5056 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
5057 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
5058 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
5059 STAT_ATTR(FREE_SLAB
, free_slab
);
5060 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
5061 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
5062 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
5063 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
5064 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
5065 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
5066 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
5067 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
5068 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
5069 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
5070 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
5071 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
5074 static struct attribute
*slab_attrs
[] = {
5075 &slab_size_attr
.attr
,
5076 &object_size_attr
.attr
,
5077 &objs_per_slab_attr
.attr
,
5079 &min_partial_attr
.attr
,
5080 &cpu_partial_attr
.attr
,
5082 &objects_partial_attr
.attr
,
5084 &cpu_slabs_attr
.attr
,
5088 &hwcache_align_attr
.attr
,
5089 &reclaim_account_attr
.attr
,
5090 &destroy_by_rcu_attr
.attr
,
5092 &reserved_attr
.attr
,
5093 &slabs_cpu_partial_attr
.attr
,
5094 #ifdef CONFIG_SLUB_DEBUG
5095 &total_objects_attr
.attr
,
5097 &sanity_checks_attr
.attr
,
5099 &red_zone_attr
.attr
,
5101 &store_user_attr
.attr
,
5102 &validate_attr
.attr
,
5103 &alloc_calls_attr
.attr
,
5104 &free_calls_attr
.attr
,
5106 #ifdef CONFIG_ZONE_DMA
5107 &cache_dma_attr
.attr
,
5110 &remote_node_defrag_ratio_attr
.attr
,
5112 #ifdef CONFIG_SLUB_STATS
5113 &alloc_fastpath_attr
.attr
,
5114 &alloc_slowpath_attr
.attr
,
5115 &free_fastpath_attr
.attr
,
5116 &free_slowpath_attr
.attr
,
5117 &free_frozen_attr
.attr
,
5118 &free_add_partial_attr
.attr
,
5119 &free_remove_partial_attr
.attr
,
5120 &alloc_from_partial_attr
.attr
,
5121 &alloc_slab_attr
.attr
,
5122 &alloc_refill_attr
.attr
,
5123 &alloc_node_mismatch_attr
.attr
,
5124 &free_slab_attr
.attr
,
5125 &cpuslab_flush_attr
.attr
,
5126 &deactivate_full_attr
.attr
,
5127 &deactivate_empty_attr
.attr
,
5128 &deactivate_to_head_attr
.attr
,
5129 &deactivate_to_tail_attr
.attr
,
5130 &deactivate_remote_frees_attr
.attr
,
5131 &deactivate_bypass_attr
.attr
,
5132 &order_fallback_attr
.attr
,
5133 &cmpxchg_double_fail_attr
.attr
,
5134 &cmpxchg_double_cpu_fail_attr
.attr
,
5135 &cpu_partial_alloc_attr
.attr
,
5136 &cpu_partial_free_attr
.attr
,
5138 #ifdef CONFIG_FAILSLAB
5139 &failslab_attr
.attr
,
5145 static struct attribute_group slab_attr_group
= {
5146 .attrs
= slab_attrs
,
5149 static ssize_t
slab_attr_show(struct kobject
*kobj
,
5150 struct attribute
*attr
,
5153 struct slab_attribute
*attribute
;
5154 struct kmem_cache
*s
;
5157 attribute
= to_slab_attr(attr
);
5160 if (!attribute
->show
)
5163 err
= attribute
->show(s
, buf
);
5168 static ssize_t
slab_attr_store(struct kobject
*kobj
,
5169 struct attribute
*attr
,
5170 const char *buf
, size_t len
)
5172 struct slab_attribute
*attribute
;
5173 struct kmem_cache
*s
;
5176 attribute
= to_slab_attr(attr
);
5179 if (!attribute
->store
)
5182 err
= attribute
->store(s
, buf
, len
);
5187 static void kmem_cache_release(struct kobject
*kobj
)
5189 struct kmem_cache
*s
= to_slab(kobj
);
5195 static const struct sysfs_ops slab_sysfs_ops
= {
5196 .show
= slab_attr_show
,
5197 .store
= slab_attr_store
,
5200 static struct kobj_type slab_ktype
= {
5201 .sysfs_ops
= &slab_sysfs_ops
,
5202 .release
= kmem_cache_release
5205 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5207 struct kobj_type
*ktype
= get_ktype(kobj
);
5209 if (ktype
== &slab_ktype
)
5214 static const struct kset_uevent_ops slab_uevent_ops
= {
5215 .filter
= uevent_filter
,
5218 static struct kset
*slab_kset
;
5220 #define ID_STR_LENGTH 64
5222 /* Create a unique string id for a slab cache:
5224 * Format :[flags-]size
5226 static char *create_unique_id(struct kmem_cache
*s
)
5228 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5235 * First flags affecting slabcache operations. We will only
5236 * get here for aliasable slabs so we do not need to support
5237 * too many flags. The flags here must cover all flags that
5238 * are matched during merging to guarantee that the id is
5241 if (s
->flags
& SLAB_CACHE_DMA
)
5243 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5245 if (s
->flags
& SLAB_DEBUG_FREE
)
5247 if (!(s
->flags
& SLAB_NOTRACK
))
5251 p
+= sprintf(p
, "%07d", s
->size
);
5252 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5256 static int sysfs_slab_add(struct kmem_cache
*s
)
5262 if (slab_state
< SYSFS
)
5263 /* Defer until later */
5266 unmergeable
= slab_unmergeable(s
);
5269 * Slabcache can never be merged so we can use the name proper.
5270 * This is typically the case for debug situations. In that
5271 * case we can catch duplicate names easily.
5273 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5277 * Create a unique name for the slab as a target
5280 name
= create_unique_id(s
);
5283 s
->kobj
.kset
= slab_kset
;
5284 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5286 kobject_put(&s
->kobj
);
5290 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5292 kobject_del(&s
->kobj
);
5293 kobject_put(&s
->kobj
);
5296 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5298 /* Setup first alias */
5299 sysfs_slab_alias(s
, s
->name
);
5305 static void sysfs_slab_remove(struct kmem_cache
*s
)
5307 if (slab_state
< SYSFS
)
5309 * Sysfs has not been setup yet so no need to remove the
5314 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5315 kobject_del(&s
->kobj
);
5316 kobject_put(&s
->kobj
);
5320 * Need to buffer aliases during bootup until sysfs becomes
5321 * available lest we lose that information.
5323 struct saved_alias
{
5324 struct kmem_cache
*s
;
5326 struct saved_alias
*next
;
5329 static struct saved_alias
*alias_list
;
5331 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5333 struct saved_alias
*al
;
5335 if (slab_state
== SYSFS
) {
5337 * If we have a leftover link then remove it.
5339 sysfs_remove_link(&slab_kset
->kobj
, name
);
5340 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5343 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5349 al
->next
= alias_list
;
5354 static int __init
slab_sysfs_init(void)
5356 struct kmem_cache
*s
;
5359 down_write(&slub_lock
);
5361 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5363 up_write(&slub_lock
);
5364 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5370 list_for_each_entry(s
, &slab_caches
, list
) {
5371 err
= sysfs_slab_add(s
);
5373 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5374 " to sysfs\n", s
->name
);
5377 while (alias_list
) {
5378 struct saved_alias
*al
= alias_list
;
5380 alias_list
= alias_list
->next
;
5381 err
= sysfs_slab_alias(al
->s
, al
->name
);
5383 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5384 " %s to sysfs\n", s
->name
);
5388 up_write(&slub_lock
);
5393 __initcall(slab_sysfs_init
);
5394 #endif /* CONFIG_SYSFS */
5397 * The /proc/slabinfo ABI
5399 #ifdef CONFIG_SLABINFO
5400 static void print_slabinfo_header(struct seq_file
*m
)
5402 seq_puts(m
, "slabinfo - version: 2.1\n");
5403 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
5404 "<objperslab> <pagesperslab>");
5405 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
5406 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5410 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
5414 down_read(&slub_lock
);
5416 print_slabinfo_header(m
);
5418 return seq_list_start(&slab_caches
, *pos
);
5421 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
5423 return seq_list_next(p
, &slab_caches
, pos
);
5426 static void s_stop(struct seq_file
*m
, void *p
)
5428 up_read(&slub_lock
);
5431 static int s_show(struct seq_file
*m
, void *p
)
5433 unsigned long nr_partials
= 0;
5434 unsigned long nr_slabs
= 0;
5435 unsigned long nr_inuse
= 0;
5436 unsigned long nr_objs
= 0;
5437 unsigned long nr_free
= 0;
5438 struct kmem_cache
*s
;
5441 s
= list_entry(p
, struct kmem_cache
, list
);
5443 for_each_online_node(node
) {
5444 struct kmem_cache_node
*n
= get_node(s
, node
);
5449 nr_partials
+= n
->nr_partial
;
5450 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
5451 nr_objs
+= atomic_long_read(&n
->total_objects
);
5452 nr_free
+= count_partial(n
, count_free
);
5455 nr_inuse
= nr_objs
- nr_free
;
5457 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
5458 nr_objs
, s
->size
, oo_objects(s
->oo
),
5459 (1 << oo_order(s
->oo
)));
5460 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
5461 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
5467 static const struct seq_operations slabinfo_op
= {
5474 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
5476 return seq_open(file
, &slabinfo_op
);
5479 static const struct file_operations proc_slabinfo_operations
= {
5480 .open
= slabinfo_open
,
5482 .llseek
= seq_lseek
,
5483 .release
= seq_release
,
5486 static int __init
slab_proc_init(void)
5488 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
5491 module_init(slab_proc_init
);
5492 #endif /* CONFIG_SLABINFO */