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>
32 #include <linux/prefetch.h>
34 #include <trace/events/kmem.h>
38 * 1. slub_lock (Global Semaphore)
40 * 3. slab_lock(page) (Only on some arches and for debugging)
44 * The role of the slub_lock is to protect the list of all the slabs
45 * and to synchronize major metadata changes to slab cache structures.
47 * The slab_lock is only used for debugging and on arches that do not
48 * have the ability to do a cmpxchg_double. It only protects the second
49 * double word in the page struct. Meaning
50 * A. page->freelist -> List of object free in a page
51 * B. page->counters -> Counters of objects
52 * C. page->frozen -> frozen state
54 * If a slab is frozen then it is exempt from list management. It is not
55 * on any list. The processor that froze the slab is the one who can
56 * perform list operations on the page. Other processors may put objects
57 * onto the freelist but the processor that froze the slab is the only
58 * one that can retrieve the objects from the page's freelist.
60 * The list_lock protects the partial and full list on each node and
61 * the partial slab counter. If taken then no new slabs may be added or
62 * removed from the lists nor make the number of partial slabs be modified.
63 * (Note that the total number of slabs is an atomic value that may be
64 * modified without taking the list lock).
66 * The list_lock is a centralized lock and thus we avoid taking it as
67 * much as possible. As long as SLUB does not have to handle partial
68 * slabs, operations can continue without any centralized lock. F.e.
69 * allocating a long series of objects that fill up slabs does not require
71 * Interrupts are disabled during allocation and deallocation in order to
72 * make the slab allocator safe to use in the context of an irq. In addition
73 * interrupts are disabled to ensure that the processor does not change
74 * while handling per_cpu slabs, due to kernel preemption.
76 * SLUB assigns one slab for allocation to each processor.
77 * Allocations only occur from these slabs called cpu slabs.
79 * Slabs with free elements are kept on a partial list and during regular
80 * operations no list for full slabs is used. If an object in a full slab is
81 * freed then the slab will show up again on the partial lists.
82 * We track full slabs for debugging purposes though because otherwise we
83 * cannot scan all objects.
85 * Slabs are freed when they become empty. Teardown and setup is
86 * minimal so we rely on the page allocators per cpu caches for
87 * fast frees and allocs.
89 * Overloading of page flags that are otherwise used for LRU management.
91 * PageActive The slab is frozen and exempt from list processing.
92 * This means that the slab is dedicated to a purpose
93 * such as satisfying allocations for a specific
94 * processor. Objects may be freed in the slab while
95 * it is frozen but slab_free will then skip the usual
96 * list operations. It is up to the processor holding
97 * the slab to integrate the slab into the slab lists
98 * when the slab is no longer needed.
100 * One use of this flag is to mark slabs that are
101 * used for allocations. Then such a slab becomes a cpu
102 * slab. The cpu slab may be equipped with an additional
103 * freelist that allows lockless access to
104 * free objects in addition to the regular freelist
105 * that requires the slab lock.
107 * PageError Slab requires special handling due to debug
108 * options set. This moves slab handling out of
109 * the fast path and disables lockless freelists.
112 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
113 SLAB_TRACE | SLAB_DEBUG_FREE)
115 static inline int kmem_cache_debug(struct kmem_cache
*s
)
117 #ifdef CONFIG_SLUB_DEBUG
118 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
125 * Issues still to be resolved:
127 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
129 * - Variable sizing of the per node arrays
132 /* Enable to test recovery from slab corruption on boot */
133 #undef SLUB_RESILIENCY_TEST
135 /* Enable to log cmpxchg failures */
136 #undef SLUB_DEBUG_CMPXCHG
139 * Mininum number of partial slabs. These will be left on the partial
140 * lists even if they are empty. kmem_cache_shrink may reclaim them.
142 #define MIN_PARTIAL 5
145 * Maximum number of desirable partial slabs.
146 * The existence of more partial slabs makes kmem_cache_shrink
147 * sort the partial list by the number of objects in the.
149 #define MAX_PARTIAL 10
151 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
152 SLAB_POISON | SLAB_STORE_USER)
155 * Debugging flags that require metadata to be stored in the slab. These get
156 * disabled when slub_debug=O is used and a cache's min order increases with
159 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
162 * Set of flags that will prevent slab merging
164 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
165 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
168 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
169 SLAB_CACHE_DMA | SLAB_NOTRACK)
172 #define OO_MASK ((1 << OO_SHIFT) - 1)
173 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
175 /* Internal SLUB flags */
176 #define __OBJECT_POISON 0x80000000UL /* Poison object */
177 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
179 static int kmem_size
= sizeof(struct kmem_cache
);
182 static struct notifier_block slab_notifier
;
186 DOWN
, /* No slab functionality available */
187 PARTIAL
, /* Kmem_cache_node works */
188 UP
, /* Everything works but does not show up in sysfs */
192 /* A list of all slab caches on the system */
193 static DECLARE_RWSEM(slub_lock
);
194 static LIST_HEAD(slab_caches
);
197 * Tracking user of a slab.
199 #define TRACK_ADDRS_COUNT 16
201 unsigned long addr
; /* Called from address */
202 #ifdef CONFIG_STACKTRACE
203 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
205 int cpu
; /* Was running on cpu */
206 int pid
; /* Pid context */
207 unsigned long when
; /* When did the operation occur */
210 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
213 static int sysfs_slab_add(struct kmem_cache
*);
214 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
215 static void sysfs_slab_remove(struct kmem_cache
*);
218 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
219 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
221 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
229 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
231 #ifdef CONFIG_SLUB_STATS
232 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
236 /********************************************************************
237 * Core slab cache functions
238 *******************************************************************/
240 int slab_is_available(void)
242 return slab_state
>= UP
;
245 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
247 return s
->node
[node
];
250 /* Verify that a pointer has an address that is valid within a slab page */
251 static inline int check_valid_pointer(struct kmem_cache
*s
,
252 struct page
*page
, const void *object
)
259 base
= page_address(page
);
260 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
261 (object
- base
) % s
->size
) {
268 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
270 return *(void **)(object
+ s
->offset
);
273 static void prefetch_freepointer(const struct kmem_cache
*s
, void *object
)
275 prefetch(object
+ s
->offset
);
278 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
282 #ifdef CONFIG_DEBUG_PAGEALLOC
283 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
285 p
= get_freepointer(s
, object
);
290 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
292 *(void **)(object
+ s
->offset
) = fp
;
295 /* Loop over all objects in a slab */
296 #define for_each_object(__p, __s, __addr, __objects) \
297 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
300 /* Determine object index from a given position */
301 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
303 return (p
- addr
) / s
->size
;
306 static inline size_t slab_ksize(const struct kmem_cache
*s
)
308 #ifdef CONFIG_SLUB_DEBUG
310 * Debugging requires use of the padding between object
311 * and whatever may come after it.
313 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
314 return s
->object_size
;
318 * If we have the need to store the freelist pointer
319 * back there or track user information then we can
320 * only use the space before that information.
322 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
325 * Else we can use all the padding etc for the allocation
330 static inline int order_objects(int order
, unsigned long size
, int reserved
)
332 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
335 static inline struct kmem_cache_order_objects
oo_make(int order
,
336 unsigned long size
, int reserved
)
338 struct kmem_cache_order_objects x
= {
339 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
345 static inline int oo_order(struct kmem_cache_order_objects x
)
347 return x
.x
>> OO_SHIFT
;
350 static inline int oo_objects(struct kmem_cache_order_objects x
)
352 return x
.x
& OO_MASK
;
356 * Per slab locking using the pagelock
358 static __always_inline
void slab_lock(struct page
*page
)
360 bit_spin_lock(PG_locked
, &page
->flags
);
363 static __always_inline
void slab_unlock(struct page
*page
)
365 __bit_spin_unlock(PG_locked
, &page
->flags
);
368 /* Interrupts must be disabled (for the fallback code to work right) */
369 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
370 void *freelist_old
, unsigned long counters_old
,
371 void *freelist_new
, unsigned long counters_new
,
374 VM_BUG_ON(!irqs_disabled());
375 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
376 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
377 if (s
->flags
& __CMPXCHG_DOUBLE
) {
378 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
379 freelist_old
, counters_old
,
380 freelist_new
, counters_new
))
386 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
387 page
->freelist
= freelist_new
;
388 page
->counters
= counters_new
;
396 stat(s
, CMPXCHG_DOUBLE_FAIL
);
398 #ifdef SLUB_DEBUG_CMPXCHG
399 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
405 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
406 void *freelist_old
, unsigned long counters_old
,
407 void *freelist_new
, unsigned long counters_new
,
410 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
411 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
412 if (s
->flags
& __CMPXCHG_DOUBLE
) {
413 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
414 freelist_old
, counters_old
,
415 freelist_new
, counters_new
))
422 local_irq_save(flags
);
424 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
425 page
->freelist
= freelist_new
;
426 page
->counters
= counters_new
;
428 local_irq_restore(flags
);
432 local_irq_restore(flags
);
436 stat(s
, CMPXCHG_DOUBLE_FAIL
);
438 #ifdef SLUB_DEBUG_CMPXCHG
439 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
445 #ifdef CONFIG_SLUB_DEBUG
447 * Determine a map of object in use on a page.
449 * Node listlock must be held to guarantee that the page does
450 * not vanish from under us.
452 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
455 void *addr
= page_address(page
);
457 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
458 set_bit(slab_index(p
, s
, addr
), map
);
464 #ifdef CONFIG_SLUB_DEBUG_ON
465 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
467 static int slub_debug
;
470 static char *slub_debug_slabs
;
471 static int disable_higher_order_debug
;
476 static void print_section(char *text
, u8
*addr
, unsigned int length
)
478 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
482 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
483 enum track_item alloc
)
488 p
= object
+ s
->offset
+ sizeof(void *);
490 p
= object
+ s
->inuse
;
495 static void set_track(struct kmem_cache
*s
, void *object
,
496 enum track_item alloc
, unsigned long addr
)
498 struct track
*p
= get_track(s
, object
, alloc
);
501 #ifdef CONFIG_STACKTRACE
502 struct stack_trace trace
;
505 trace
.nr_entries
= 0;
506 trace
.max_entries
= TRACK_ADDRS_COUNT
;
507 trace
.entries
= p
->addrs
;
509 save_stack_trace(&trace
);
511 /* See rant in lockdep.c */
512 if (trace
.nr_entries
!= 0 &&
513 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
516 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
520 p
->cpu
= smp_processor_id();
521 p
->pid
= current
->pid
;
524 memset(p
, 0, sizeof(struct track
));
527 static void init_tracking(struct kmem_cache
*s
, void *object
)
529 if (!(s
->flags
& SLAB_STORE_USER
))
532 set_track(s
, object
, TRACK_FREE
, 0UL);
533 set_track(s
, object
, TRACK_ALLOC
, 0UL);
536 static void print_track(const char *s
, struct track
*t
)
541 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
542 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
543 #ifdef CONFIG_STACKTRACE
546 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
548 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
555 static void print_tracking(struct kmem_cache
*s
, void *object
)
557 if (!(s
->flags
& SLAB_STORE_USER
))
560 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
561 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
564 static void print_page_info(struct page
*page
)
566 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
567 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
571 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
577 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
579 printk(KERN_ERR
"========================================"
580 "=====================================\n");
581 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
582 printk(KERN_ERR
"----------------------------------------"
583 "-------------------------------------\n\n");
586 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
592 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
594 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
597 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
599 unsigned int off
; /* Offset of last byte */
600 u8
*addr
= page_address(page
);
602 print_tracking(s
, p
);
604 print_page_info(page
);
606 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
607 p
, p
- addr
, get_freepointer(s
, p
));
610 print_section("Bytes b4 ", p
- 16, 16);
612 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
614 if (s
->flags
& SLAB_RED_ZONE
)
615 print_section("Redzone ", p
+ s
->object_size
,
616 s
->inuse
- s
->object_size
);
619 off
= s
->offset
+ sizeof(void *);
623 if (s
->flags
& SLAB_STORE_USER
)
624 off
+= 2 * sizeof(struct track
);
627 /* Beginning of the filler is the free pointer */
628 print_section("Padding ", p
+ off
, s
->size
- off
);
633 static void object_err(struct kmem_cache
*s
, struct page
*page
,
634 u8
*object
, char *reason
)
636 slab_bug(s
, "%s", reason
);
637 print_trailer(s
, page
, object
);
640 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
646 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
648 slab_bug(s
, "%s", buf
);
649 print_page_info(page
);
653 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
657 if (s
->flags
& __OBJECT_POISON
) {
658 memset(p
, POISON_FREE
, s
->object_size
- 1);
659 p
[s
->object_size
- 1] = POISON_END
;
662 if (s
->flags
& SLAB_RED_ZONE
)
663 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
666 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
667 void *from
, void *to
)
669 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
670 memset(from
, data
, to
- from
);
673 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
674 u8
*object
, char *what
,
675 u8
*start
, unsigned int value
, unsigned int bytes
)
680 fault
= memchr_inv(start
, value
, bytes
);
685 while (end
> fault
&& end
[-1] == value
)
688 slab_bug(s
, "%s overwritten", what
);
689 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
690 fault
, end
- 1, fault
[0], value
);
691 print_trailer(s
, page
, object
);
693 restore_bytes(s
, what
, value
, fault
, end
);
701 * Bytes of the object to be managed.
702 * If the freepointer may overlay the object then the free
703 * pointer is the first word of the object.
705 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
708 * object + s->object_size
709 * Padding to reach word boundary. This is also used for Redzoning.
710 * Padding is extended by another word if Redzoning is enabled and
711 * object_size == inuse.
713 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
714 * 0xcc (RED_ACTIVE) for objects in use.
717 * Meta data starts here.
719 * A. Free pointer (if we cannot overwrite object on free)
720 * B. Tracking data for SLAB_STORE_USER
721 * C. Padding to reach required alignment boundary or at mininum
722 * one word if debugging is on to be able to detect writes
723 * before the word boundary.
725 * Padding is done using 0x5a (POISON_INUSE)
728 * Nothing is used beyond s->size.
730 * If slabcaches are merged then the object_size and inuse boundaries are mostly
731 * ignored. And therefore no slab options that rely on these boundaries
732 * may be used with merged slabcaches.
735 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
737 unsigned long off
= s
->inuse
; /* The end of info */
740 /* Freepointer is placed after the object. */
741 off
+= sizeof(void *);
743 if (s
->flags
& SLAB_STORE_USER
)
744 /* We also have user information there */
745 off
+= 2 * sizeof(struct track
);
750 return check_bytes_and_report(s
, page
, p
, "Object padding",
751 p
+ off
, POISON_INUSE
, s
->size
- off
);
754 /* Check the pad bytes at the end of a slab page */
755 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
763 if (!(s
->flags
& SLAB_POISON
))
766 start
= page_address(page
);
767 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
768 end
= start
+ length
;
769 remainder
= length
% s
->size
;
773 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
776 while (end
> fault
&& end
[-1] == POISON_INUSE
)
779 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
780 print_section("Padding ", end
- remainder
, remainder
);
782 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
786 static int check_object(struct kmem_cache
*s
, struct page
*page
,
787 void *object
, u8 val
)
790 u8
*endobject
= object
+ s
->object_size
;
792 if (s
->flags
& SLAB_RED_ZONE
) {
793 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
794 endobject
, val
, s
->inuse
- s
->object_size
))
797 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
798 check_bytes_and_report(s
, page
, p
, "Alignment padding",
799 endobject
, POISON_INUSE
, s
->inuse
- s
->object_size
);
803 if (s
->flags
& SLAB_POISON
) {
804 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
805 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
806 POISON_FREE
, s
->object_size
- 1) ||
807 !check_bytes_and_report(s
, page
, p
, "Poison",
808 p
+ s
->object_size
- 1, POISON_END
, 1)))
811 * check_pad_bytes cleans up on its own.
813 check_pad_bytes(s
, page
, p
);
816 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
818 * Object and freepointer overlap. Cannot check
819 * freepointer while object is allocated.
823 /* Check free pointer validity */
824 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
825 object_err(s
, page
, p
, "Freepointer corrupt");
827 * No choice but to zap it and thus lose the remainder
828 * of the free objects in this slab. May cause
829 * another error because the object count is now wrong.
831 set_freepointer(s
, p
, NULL
);
837 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
841 VM_BUG_ON(!irqs_disabled());
843 if (!PageSlab(page
)) {
844 slab_err(s
, page
, "Not a valid slab page");
848 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
849 if (page
->objects
> maxobj
) {
850 slab_err(s
, page
, "objects %u > max %u",
851 s
->name
, page
->objects
, maxobj
);
854 if (page
->inuse
> page
->objects
) {
855 slab_err(s
, page
, "inuse %u > max %u",
856 s
->name
, page
->inuse
, page
->objects
);
859 /* Slab_pad_check fixes things up after itself */
860 slab_pad_check(s
, page
);
865 * Determine if a certain object on a page is on the freelist. Must hold the
866 * slab lock to guarantee that the chains are in a consistent state.
868 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
873 unsigned long max_objects
;
876 while (fp
&& nr
<= page
->objects
) {
879 if (!check_valid_pointer(s
, page
, fp
)) {
881 object_err(s
, page
, object
,
882 "Freechain corrupt");
883 set_freepointer(s
, object
, NULL
);
886 slab_err(s
, page
, "Freepointer corrupt");
887 page
->freelist
= NULL
;
888 page
->inuse
= page
->objects
;
889 slab_fix(s
, "Freelist cleared");
895 fp
= get_freepointer(s
, object
);
899 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
900 if (max_objects
> MAX_OBJS_PER_PAGE
)
901 max_objects
= MAX_OBJS_PER_PAGE
;
903 if (page
->objects
!= max_objects
) {
904 slab_err(s
, page
, "Wrong number of objects. Found %d but "
905 "should be %d", page
->objects
, max_objects
);
906 page
->objects
= max_objects
;
907 slab_fix(s
, "Number of objects adjusted.");
909 if (page
->inuse
!= page
->objects
- nr
) {
910 slab_err(s
, page
, "Wrong object count. Counter is %d but "
911 "counted were %d", page
->inuse
, page
->objects
- nr
);
912 page
->inuse
= page
->objects
- nr
;
913 slab_fix(s
, "Object count adjusted.");
915 return search
== NULL
;
918 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
921 if (s
->flags
& SLAB_TRACE
) {
922 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
924 alloc
? "alloc" : "free",
929 print_section("Object ", (void *)object
, s
->object_size
);
936 * Hooks for other subsystems that check memory allocations. In a typical
937 * production configuration these hooks all should produce no code at all.
939 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
941 flags
&= gfp_allowed_mask
;
942 lockdep_trace_alloc(flags
);
943 might_sleep_if(flags
& __GFP_WAIT
);
945 return should_failslab(s
->object_size
, flags
, s
->flags
);
948 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
950 flags
&= gfp_allowed_mask
;
951 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
952 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
955 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
957 kmemleak_free_recursive(x
, s
->flags
);
960 * Trouble is that we may no longer disable interupts in the fast path
961 * So in order to make the debug calls that expect irqs to be
962 * disabled we need to disable interrupts temporarily.
964 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
968 local_irq_save(flags
);
969 kmemcheck_slab_free(s
, x
, s
->object_size
);
970 debug_check_no_locks_freed(x
, s
->object_size
);
971 local_irq_restore(flags
);
974 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
975 debug_check_no_obj_freed(x
, s
->object_size
);
979 * Tracking of fully allocated slabs for debugging purposes.
981 * list_lock must be held.
983 static void add_full(struct kmem_cache
*s
,
984 struct kmem_cache_node
*n
, struct page
*page
)
986 if (!(s
->flags
& SLAB_STORE_USER
))
989 list_add(&page
->lru
, &n
->full
);
993 * list_lock must be held.
995 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
997 if (!(s
->flags
& SLAB_STORE_USER
))
1000 list_del(&page
->lru
);
1003 /* Tracking of the number of slabs for debugging purposes */
1004 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1006 struct kmem_cache_node
*n
= get_node(s
, node
);
1008 return atomic_long_read(&n
->nr_slabs
);
1011 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1013 return atomic_long_read(&n
->nr_slabs
);
1016 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1018 struct kmem_cache_node
*n
= get_node(s
, node
);
1021 * May be called early in order to allocate a slab for the
1022 * kmem_cache_node structure. Solve the chicken-egg
1023 * dilemma by deferring the increment of the count during
1024 * bootstrap (see early_kmem_cache_node_alloc).
1027 atomic_long_inc(&n
->nr_slabs
);
1028 atomic_long_add(objects
, &n
->total_objects
);
1031 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1033 struct kmem_cache_node
*n
= get_node(s
, node
);
1035 atomic_long_dec(&n
->nr_slabs
);
1036 atomic_long_sub(objects
, &n
->total_objects
);
1039 /* Object debug checks for alloc/free paths */
1040 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1043 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1046 init_object(s
, object
, SLUB_RED_INACTIVE
);
1047 init_tracking(s
, object
);
1050 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
1051 void *object
, unsigned long addr
)
1053 if (!check_slab(s
, page
))
1056 if (!check_valid_pointer(s
, page
, object
)) {
1057 object_err(s
, page
, object
, "Freelist Pointer check fails");
1061 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1064 /* Success perform special debug activities for allocs */
1065 if (s
->flags
& SLAB_STORE_USER
)
1066 set_track(s
, object
, TRACK_ALLOC
, addr
);
1067 trace(s
, page
, object
, 1);
1068 init_object(s
, object
, SLUB_RED_ACTIVE
);
1072 if (PageSlab(page
)) {
1074 * If this is a slab page then lets do the best we can
1075 * to avoid issues in the future. Marking all objects
1076 * as used avoids touching the remaining objects.
1078 slab_fix(s
, "Marking all objects used");
1079 page
->inuse
= page
->objects
;
1080 page
->freelist
= NULL
;
1085 static noinline
int free_debug_processing(struct kmem_cache
*s
,
1086 struct page
*page
, void *object
, unsigned long addr
)
1088 unsigned long flags
;
1091 local_irq_save(flags
);
1094 if (!check_slab(s
, page
))
1097 if (!check_valid_pointer(s
, page
, object
)) {
1098 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1102 if (on_freelist(s
, page
, object
)) {
1103 object_err(s
, page
, object
, "Object already free");
1107 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1110 if (unlikely(s
!= page
->slab
)) {
1111 if (!PageSlab(page
)) {
1112 slab_err(s
, page
, "Attempt to free object(0x%p) "
1113 "outside of slab", object
);
1114 } else if (!page
->slab
) {
1116 "SLUB <none>: no slab for object 0x%p.\n",
1120 object_err(s
, page
, object
,
1121 "page slab pointer corrupt.");
1125 if (s
->flags
& SLAB_STORE_USER
)
1126 set_track(s
, object
, TRACK_FREE
, addr
);
1127 trace(s
, page
, object
, 0);
1128 init_object(s
, object
, SLUB_RED_INACTIVE
);
1132 local_irq_restore(flags
);
1136 slab_fix(s
, "Object at 0x%p not freed", object
);
1140 static int __init
setup_slub_debug(char *str
)
1142 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1143 if (*str
++ != '=' || !*str
)
1145 * No options specified. Switch on full debugging.
1151 * No options but restriction on slabs. This means full
1152 * debugging for slabs matching a pattern.
1156 if (tolower(*str
) == 'o') {
1158 * Avoid enabling debugging on caches if its minimum order
1159 * would increase as a result.
1161 disable_higher_order_debug
= 1;
1168 * Switch off all debugging measures.
1173 * Determine which debug features should be switched on
1175 for (; *str
&& *str
!= ','; str
++) {
1176 switch (tolower(*str
)) {
1178 slub_debug
|= SLAB_DEBUG_FREE
;
1181 slub_debug
|= SLAB_RED_ZONE
;
1184 slub_debug
|= SLAB_POISON
;
1187 slub_debug
|= SLAB_STORE_USER
;
1190 slub_debug
|= SLAB_TRACE
;
1193 slub_debug
|= SLAB_FAILSLAB
;
1196 printk(KERN_ERR
"slub_debug option '%c' "
1197 "unknown. skipped\n", *str
);
1203 slub_debug_slabs
= str
+ 1;
1208 __setup("slub_debug", setup_slub_debug
);
1210 static unsigned long kmem_cache_flags(unsigned long object_size
,
1211 unsigned long flags
, const char *name
,
1212 void (*ctor
)(void *))
1215 * Enable debugging if selected on the kernel commandline.
1217 if (slub_debug
&& (!slub_debug_slabs
||
1218 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1219 flags
|= slub_debug
;
1224 static inline void setup_object_debug(struct kmem_cache
*s
,
1225 struct page
*page
, void *object
) {}
1227 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1228 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1230 static inline int free_debug_processing(struct kmem_cache
*s
,
1231 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1233 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1235 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1236 void *object
, u8 val
) { return 1; }
1237 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1238 struct page
*page
) {}
1239 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1240 static inline unsigned long kmem_cache_flags(unsigned long object_size
,
1241 unsigned long flags
, const char *name
,
1242 void (*ctor
)(void *))
1246 #define slub_debug 0
1248 #define disable_higher_order_debug 0
1250 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1252 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1254 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1256 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1259 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1262 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1265 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1267 #endif /* CONFIG_SLUB_DEBUG */
1270 * Slab allocation and freeing
1272 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1273 struct kmem_cache_order_objects oo
)
1275 int order
= oo_order(oo
);
1277 flags
|= __GFP_NOTRACK
;
1279 if (node
== NUMA_NO_NODE
)
1280 return alloc_pages(flags
, order
);
1282 return alloc_pages_exact_node(node
, flags
, order
);
1285 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1288 struct kmem_cache_order_objects oo
= s
->oo
;
1291 flags
&= gfp_allowed_mask
;
1293 if (flags
& __GFP_WAIT
)
1296 flags
|= s
->allocflags
;
1299 * Let the initial higher-order allocation fail under memory pressure
1300 * so we fall-back to the minimum order allocation.
1302 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1304 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1305 if (unlikely(!page
)) {
1308 * Allocation may have failed due to fragmentation.
1309 * Try a lower order alloc if possible
1311 page
= alloc_slab_page(flags
, node
, oo
);
1314 stat(s
, ORDER_FALLBACK
);
1317 if (flags
& __GFP_WAIT
)
1318 local_irq_disable();
1323 if (kmemcheck_enabled
1324 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1325 int pages
= 1 << oo_order(oo
);
1327 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1330 * Objects from caches that have a constructor don't get
1331 * cleared when they're allocated, so we need to do it here.
1334 kmemcheck_mark_uninitialized_pages(page
, pages
);
1336 kmemcheck_mark_unallocated_pages(page
, pages
);
1339 page
->objects
= oo_objects(oo
);
1340 mod_zone_page_state(page_zone(page
),
1341 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1342 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1348 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1351 setup_object_debug(s
, page
, object
);
1352 if (unlikely(s
->ctor
))
1356 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1363 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1365 page
= allocate_slab(s
,
1366 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1370 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1372 __SetPageSlab(page
);
1374 start
= page_address(page
);
1376 if (unlikely(s
->flags
& SLAB_POISON
))
1377 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1380 for_each_object(p
, s
, start
, page
->objects
) {
1381 setup_object(s
, page
, last
);
1382 set_freepointer(s
, last
, p
);
1385 setup_object(s
, page
, last
);
1386 set_freepointer(s
, last
, NULL
);
1388 page
->freelist
= start
;
1389 page
->inuse
= page
->objects
;
1395 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1397 int order
= compound_order(page
);
1398 int pages
= 1 << order
;
1400 if (kmem_cache_debug(s
)) {
1403 slab_pad_check(s
, page
);
1404 for_each_object(p
, s
, page_address(page
),
1406 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1409 kmemcheck_free_shadow(page
, compound_order(page
));
1411 mod_zone_page_state(page_zone(page
),
1412 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1413 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1416 __ClearPageSlab(page
);
1417 reset_page_mapcount(page
);
1418 if (current
->reclaim_state
)
1419 current
->reclaim_state
->reclaimed_slab
+= pages
;
1420 __free_pages(page
, order
);
1423 #define need_reserve_slab_rcu \
1424 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1426 static void rcu_free_slab(struct rcu_head
*h
)
1430 if (need_reserve_slab_rcu
)
1431 page
= virt_to_head_page(h
);
1433 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1435 __free_slab(page
->slab
, page
);
1438 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1440 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1441 struct rcu_head
*head
;
1443 if (need_reserve_slab_rcu
) {
1444 int order
= compound_order(page
);
1445 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1447 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1448 head
= page_address(page
) + offset
;
1451 * RCU free overloads the RCU head over the LRU
1453 head
= (void *)&page
->lru
;
1456 call_rcu(head
, rcu_free_slab
);
1458 __free_slab(s
, page
);
1461 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1463 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1468 * Management of partially allocated slabs.
1470 * list_lock must be held.
1472 static inline void add_partial(struct kmem_cache_node
*n
,
1473 struct page
*page
, int tail
)
1476 if (tail
== DEACTIVATE_TO_TAIL
)
1477 list_add_tail(&page
->lru
, &n
->partial
);
1479 list_add(&page
->lru
, &n
->partial
);
1483 * list_lock must be held.
1485 static inline void remove_partial(struct kmem_cache_node
*n
,
1488 list_del(&page
->lru
);
1493 * Remove slab from the partial list, freeze it and
1494 * return the pointer to the freelist.
1496 * Returns a list of objects or NULL if it fails.
1498 * Must hold list_lock since we modify the partial list.
1500 static inline void *acquire_slab(struct kmem_cache
*s
,
1501 struct kmem_cache_node
*n
, struct page
*page
,
1505 unsigned long counters
;
1509 * Zap the freelist and set the frozen bit.
1510 * The old freelist is the list of objects for the
1511 * per cpu allocation list.
1513 freelist
= page
->freelist
;
1514 counters
= page
->counters
;
1515 new.counters
= counters
;
1517 new.inuse
= page
->objects
;
1518 new.freelist
= NULL
;
1520 new.freelist
= freelist
;
1523 VM_BUG_ON(new.frozen
);
1526 if (!__cmpxchg_double_slab(s
, page
,
1528 new.freelist
, new.counters
,
1532 remove_partial(n
, page
);
1537 static int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1540 * Try to allocate a partial slab from a specific node.
1542 static void *get_partial_node(struct kmem_cache
*s
,
1543 struct kmem_cache_node
*n
, struct kmem_cache_cpu
*c
)
1545 struct page
*page
, *page2
;
1546 void *object
= NULL
;
1549 * Racy check. If we mistakenly see no partial slabs then we
1550 * just allocate an empty slab. If we mistakenly try to get a
1551 * partial slab and there is none available then get_partials()
1554 if (!n
|| !n
->nr_partial
)
1557 spin_lock(&n
->list_lock
);
1558 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1559 void *t
= acquire_slab(s
, n
, page
, object
== NULL
);
1567 stat(s
, ALLOC_FROM_PARTIAL
);
1569 available
= page
->objects
- page
->inuse
;
1571 available
= put_cpu_partial(s
, page
, 0);
1572 stat(s
, CPU_PARTIAL_NODE
);
1574 if (kmem_cache_debug(s
) || available
> s
->cpu_partial
/ 2)
1578 spin_unlock(&n
->list_lock
);
1583 * Get a page from somewhere. Search in increasing NUMA distances.
1585 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1586 struct kmem_cache_cpu
*c
)
1589 struct zonelist
*zonelist
;
1592 enum zone_type high_zoneidx
= gfp_zone(flags
);
1594 unsigned int cpuset_mems_cookie
;
1597 * The defrag ratio allows a configuration of the tradeoffs between
1598 * inter node defragmentation and node local allocations. A lower
1599 * defrag_ratio increases the tendency to do local allocations
1600 * instead of attempting to obtain partial slabs from other nodes.
1602 * If the defrag_ratio is set to 0 then kmalloc() always
1603 * returns node local objects. If the ratio is higher then kmalloc()
1604 * may return off node objects because partial slabs are obtained
1605 * from other nodes and filled up.
1607 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1608 * defrag_ratio = 1000) then every (well almost) allocation will
1609 * first attempt to defrag slab caches on other nodes. This means
1610 * scanning over all nodes to look for partial slabs which may be
1611 * expensive if we do it every time we are trying to find a slab
1612 * with available objects.
1614 if (!s
->remote_node_defrag_ratio
||
1615 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1619 cpuset_mems_cookie
= get_mems_allowed();
1620 zonelist
= node_zonelist(slab_node(), flags
);
1621 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1622 struct kmem_cache_node
*n
;
1624 n
= get_node(s
, zone_to_nid(zone
));
1626 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1627 n
->nr_partial
> s
->min_partial
) {
1628 object
= get_partial_node(s
, n
, c
);
1631 * Return the object even if
1632 * put_mems_allowed indicated that
1633 * the cpuset mems_allowed was
1634 * updated in parallel. It's a
1635 * harmless race between the alloc
1636 * and the cpuset update.
1638 put_mems_allowed(cpuset_mems_cookie
);
1643 } while (!put_mems_allowed(cpuset_mems_cookie
));
1649 * Get a partial page, lock it and return it.
1651 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1652 struct kmem_cache_cpu
*c
)
1655 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1657 object
= get_partial_node(s
, get_node(s
, searchnode
), c
);
1658 if (object
|| node
!= NUMA_NO_NODE
)
1661 return get_any_partial(s
, flags
, c
);
1664 #ifdef CONFIG_PREEMPT
1666 * Calculate the next globally unique transaction for disambiguiation
1667 * during cmpxchg. The transactions start with the cpu number and are then
1668 * incremented by CONFIG_NR_CPUS.
1670 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1673 * No preemption supported therefore also no need to check for
1679 static inline unsigned long next_tid(unsigned long tid
)
1681 return tid
+ TID_STEP
;
1684 static inline unsigned int tid_to_cpu(unsigned long tid
)
1686 return tid
% TID_STEP
;
1689 static inline unsigned long tid_to_event(unsigned long tid
)
1691 return tid
/ TID_STEP
;
1694 static inline unsigned int init_tid(int cpu
)
1699 static inline void note_cmpxchg_failure(const char *n
,
1700 const struct kmem_cache
*s
, unsigned long tid
)
1702 #ifdef SLUB_DEBUG_CMPXCHG
1703 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1705 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1707 #ifdef CONFIG_PREEMPT
1708 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1709 printk("due to cpu change %d -> %d\n",
1710 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1713 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1714 printk("due to cpu running other code. Event %ld->%ld\n",
1715 tid_to_event(tid
), tid_to_event(actual_tid
));
1717 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1718 actual_tid
, tid
, next_tid(tid
));
1720 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1723 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1727 for_each_possible_cpu(cpu
)
1728 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1732 * Remove the cpu slab
1734 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
, void *freelist
)
1736 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1737 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1739 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1741 int tail
= DEACTIVATE_TO_HEAD
;
1745 if (page
->freelist
) {
1746 stat(s
, DEACTIVATE_REMOTE_FREES
);
1747 tail
= DEACTIVATE_TO_TAIL
;
1751 * Stage one: Free all available per cpu objects back
1752 * to the page freelist while it is still frozen. Leave the
1755 * There is no need to take the list->lock because the page
1758 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1760 unsigned long counters
;
1763 prior
= page
->freelist
;
1764 counters
= page
->counters
;
1765 set_freepointer(s
, freelist
, prior
);
1766 new.counters
= counters
;
1768 VM_BUG_ON(!new.frozen
);
1770 } while (!__cmpxchg_double_slab(s
, page
,
1772 freelist
, new.counters
,
1773 "drain percpu freelist"));
1775 freelist
= nextfree
;
1779 * Stage two: Ensure that the page is unfrozen while the
1780 * list presence reflects the actual number of objects
1783 * We setup the list membership and then perform a cmpxchg
1784 * with the count. If there is a mismatch then the page
1785 * is not unfrozen but the page is on the wrong list.
1787 * Then we restart the process which may have to remove
1788 * the page from the list that we just put it on again
1789 * because the number of objects in the slab may have
1794 old
.freelist
= page
->freelist
;
1795 old
.counters
= page
->counters
;
1796 VM_BUG_ON(!old
.frozen
);
1798 /* Determine target state of the slab */
1799 new.counters
= old
.counters
;
1802 set_freepointer(s
, freelist
, old
.freelist
);
1803 new.freelist
= freelist
;
1805 new.freelist
= old
.freelist
;
1809 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1811 else if (new.freelist
) {
1816 * Taking the spinlock removes the possiblity
1817 * that acquire_slab() will see a slab page that
1820 spin_lock(&n
->list_lock
);
1824 if (kmem_cache_debug(s
) && !lock
) {
1827 * This also ensures that the scanning of full
1828 * slabs from diagnostic functions will not see
1831 spin_lock(&n
->list_lock
);
1839 remove_partial(n
, page
);
1841 else if (l
== M_FULL
)
1843 remove_full(s
, page
);
1845 if (m
== M_PARTIAL
) {
1847 add_partial(n
, page
, tail
);
1850 } else if (m
== M_FULL
) {
1852 stat(s
, DEACTIVATE_FULL
);
1853 add_full(s
, n
, page
);
1859 if (!__cmpxchg_double_slab(s
, page
,
1860 old
.freelist
, old
.counters
,
1861 new.freelist
, new.counters
,
1866 spin_unlock(&n
->list_lock
);
1869 stat(s
, DEACTIVATE_EMPTY
);
1870 discard_slab(s
, page
);
1876 * Unfreeze all the cpu partial slabs.
1878 * This function must be called with interrupt disabled.
1880 static void unfreeze_partials(struct kmem_cache
*s
)
1882 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1883 struct kmem_cache_cpu
*c
= this_cpu_ptr(s
->cpu_slab
);
1884 struct page
*page
, *discard_page
= NULL
;
1886 while ((page
= c
->partial
)) {
1890 c
->partial
= page
->next
;
1892 n2
= get_node(s
, page_to_nid(page
));
1895 spin_unlock(&n
->list_lock
);
1898 spin_lock(&n
->list_lock
);
1903 old
.freelist
= page
->freelist
;
1904 old
.counters
= page
->counters
;
1905 VM_BUG_ON(!old
.frozen
);
1907 new.counters
= old
.counters
;
1908 new.freelist
= old
.freelist
;
1912 } while (!__cmpxchg_double_slab(s
, page
,
1913 old
.freelist
, old
.counters
,
1914 new.freelist
, new.counters
,
1915 "unfreezing slab"));
1917 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
)) {
1918 page
->next
= discard_page
;
1919 discard_page
= page
;
1921 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1922 stat(s
, FREE_ADD_PARTIAL
);
1927 spin_unlock(&n
->list_lock
);
1929 while (discard_page
) {
1930 page
= discard_page
;
1931 discard_page
= discard_page
->next
;
1933 stat(s
, DEACTIVATE_EMPTY
);
1934 discard_slab(s
, page
);
1940 * Put a page that was just frozen (in __slab_free) into a partial page
1941 * slot if available. This is done without interrupts disabled and without
1942 * preemption disabled. The cmpxchg is racy and may put the partial page
1943 * onto a random cpus partial slot.
1945 * If we did not find a slot then simply move all the partials to the
1946 * per node partial list.
1948 int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1950 struct page
*oldpage
;
1957 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1960 pobjects
= oldpage
->pobjects
;
1961 pages
= oldpage
->pages
;
1962 if (drain
&& pobjects
> s
->cpu_partial
) {
1963 unsigned long flags
;
1965 * partial array is full. Move the existing
1966 * set to the per node partial list.
1968 local_irq_save(flags
);
1969 unfreeze_partials(s
);
1970 local_irq_restore(flags
);
1973 stat(s
, CPU_PARTIAL_DRAIN
);
1978 pobjects
+= page
->objects
- page
->inuse
;
1980 page
->pages
= pages
;
1981 page
->pobjects
= pobjects
;
1982 page
->next
= oldpage
;
1984 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
) != oldpage
);
1988 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1990 stat(s
, CPUSLAB_FLUSH
);
1991 deactivate_slab(s
, c
->page
, c
->freelist
);
1993 c
->tid
= next_tid(c
->tid
);
2001 * Called from IPI handler with interrupts disabled.
2003 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2005 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2011 unfreeze_partials(s
);
2015 static void flush_cpu_slab(void *d
)
2017 struct kmem_cache
*s
= d
;
2019 __flush_cpu_slab(s
, smp_processor_id());
2022 static bool has_cpu_slab(int cpu
, void *info
)
2024 struct kmem_cache
*s
= info
;
2025 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2027 return c
->page
|| c
->partial
;
2030 static void flush_all(struct kmem_cache
*s
)
2032 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2036 * Check if the objects in a per cpu structure fit numa
2037 * locality expectations.
2039 static inline int node_match(struct page
*page
, int node
)
2042 if (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
)
2048 static int count_free(struct page
*page
)
2050 return page
->objects
- page
->inuse
;
2053 static unsigned long count_partial(struct kmem_cache_node
*n
,
2054 int (*get_count
)(struct page
*))
2056 unsigned long flags
;
2057 unsigned long x
= 0;
2060 spin_lock_irqsave(&n
->list_lock
, flags
);
2061 list_for_each_entry(page
, &n
->partial
, lru
)
2062 x
+= get_count(page
);
2063 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2067 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2069 #ifdef CONFIG_SLUB_DEBUG
2070 return atomic_long_read(&n
->total_objects
);
2076 static noinline
void
2077 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2082 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2084 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2085 "default order: %d, min order: %d\n", s
->name
, s
->object_size
,
2086 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2088 if (oo_order(s
->min
) > get_order(s
->object_size
))
2089 printk(KERN_WARNING
" %s debugging increased min order, use "
2090 "slub_debug=O to disable.\n", s
->name
);
2092 for_each_online_node(node
) {
2093 struct kmem_cache_node
*n
= get_node(s
, node
);
2094 unsigned long nr_slabs
;
2095 unsigned long nr_objs
;
2096 unsigned long nr_free
;
2101 nr_free
= count_partial(n
, count_free
);
2102 nr_slabs
= node_nr_slabs(n
);
2103 nr_objs
= node_nr_objs(n
);
2106 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2107 node
, nr_slabs
, nr_objs
, nr_free
);
2111 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2112 int node
, struct kmem_cache_cpu
**pc
)
2115 struct kmem_cache_cpu
*c
= *pc
;
2118 freelist
= get_partial(s
, flags
, node
, c
);
2123 page
= new_slab(s
, flags
, node
);
2125 c
= __this_cpu_ptr(s
->cpu_slab
);
2130 * No other reference to the page yet so we can
2131 * muck around with it freely without cmpxchg
2133 freelist
= page
->freelist
;
2134 page
->freelist
= NULL
;
2136 stat(s
, ALLOC_SLAB
);
2146 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2147 * or deactivate the page.
2149 * The page is still frozen if the return value is not NULL.
2151 * If this function returns NULL then the page has been unfrozen.
2153 * This function must be called with interrupt disabled.
2155 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2158 unsigned long counters
;
2162 freelist
= page
->freelist
;
2163 counters
= page
->counters
;
2165 new.counters
= counters
;
2166 VM_BUG_ON(!new.frozen
);
2168 new.inuse
= page
->objects
;
2169 new.frozen
= freelist
!= NULL
;
2171 } while (!__cmpxchg_double_slab(s
, page
,
2180 * Slow path. The lockless freelist is empty or we need to perform
2183 * Processing is still very fast if new objects have been freed to the
2184 * regular freelist. In that case we simply take over the regular freelist
2185 * as the lockless freelist and zap the regular freelist.
2187 * If that is not working then we fall back to the partial lists. We take the
2188 * first element of the freelist as the object to allocate now and move the
2189 * rest of the freelist to the lockless freelist.
2191 * And if we were unable to get a new slab from the partial slab lists then
2192 * we need to allocate a new slab. This is the slowest path since it involves
2193 * a call to the page allocator and the setup of a new slab.
2195 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2196 unsigned long addr
, struct kmem_cache_cpu
*c
)
2200 unsigned long flags
;
2202 local_irq_save(flags
);
2203 #ifdef CONFIG_PREEMPT
2205 * We may have been preempted and rescheduled on a different
2206 * cpu before disabling interrupts. Need to reload cpu area
2209 c
= this_cpu_ptr(s
->cpu_slab
);
2217 if (unlikely(!node_match(page
, node
))) {
2218 stat(s
, ALLOC_NODE_MISMATCH
);
2219 deactivate_slab(s
, page
, c
->freelist
);
2225 /* must check again c->freelist in case of cpu migration or IRQ */
2226 freelist
= c
->freelist
;
2230 stat(s
, ALLOC_SLOWPATH
);
2232 freelist
= get_freelist(s
, page
);
2236 stat(s
, DEACTIVATE_BYPASS
);
2240 stat(s
, ALLOC_REFILL
);
2244 * freelist is pointing to the list of objects to be used.
2245 * page is pointing to the page from which the objects are obtained.
2246 * That page must be frozen for per cpu allocations to work.
2248 VM_BUG_ON(!c
->page
->frozen
);
2249 c
->freelist
= get_freepointer(s
, freelist
);
2250 c
->tid
= next_tid(c
->tid
);
2251 local_irq_restore(flags
);
2257 page
= c
->page
= c
->partial
;
2258 c
->partial
= page
->next
;
2259 stat(s
, CPU_PARTIAL_ALLOC
);
2264 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2266 if (unlikely(!freelist
)) {
2267 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2268 slab_out_of_memory(s
, gfpflags
, node
);
2270 local_irq_restore(flags
);
2275 if (likely(!kmem_cache_debug(s
)))
2278 /* Only entered in the debug case */
2279 if (!alloc_debug_processing(s
, page
, freelist
, addr
))
2280 goto new_slab
; /* Slab failed checks. Next slab needed */
2282 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2285 local_irq_restore(flags
);
2290 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2291 * have the fastpath folded into their functions. So no function call
2292 * overhead for requests that can be satisfied on the fastpath.
2294 * The fastpath works by first checking if the lockless freelist can be used.
2295 * If not then __slab_alloc is called for slow processing.
2297 * Otherwise we can simply pick the next object from the lockless free list.
2299 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2300 gfp_t gfpflags
, int node
, unsigned long addr
)
2303 struct kmem_cache_cpu
*c
;
2307 if (slab_pre_alloc_hook(s
, gfpflags
))
2313 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2314 * enabled. We may switch back and forth between cpus while
2315 * reading from one cpu area. That does not matter as long
2316 * as we end up on the original cpu again when doing the cmpxchg.
2318 c
= __this_cpu_ptr(s
->cpu_slab
);
2321 * The transaction ids are globally unique per cpu and per operation on
2322 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2323 * occurs on the right processor and that there was no operation on the
2324 * linked list in between.
2329 object
= c
->freelist
;
2331 if (unlikely(!object
|| !node_match(page
, node
)))
2333 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2336 void *next_object
= get_freepointer_safe(s
, object
);
2339 * The cmpxchg will only match if there was no additional
2340 * operation and if we are on the right processor.
2342 * The cmpxchg does the following atomically (without lock semantics!)
2343 * 1. Relocate first pointer to the current per cpu area.
2344 * 2. Verify that tid and freelist have not been changed
2345 * 3. If they were not changed replace tid and freelist
2347 * Since this is without lock semantics the protection is only against
2348 * code executing on this cpu *not* from access by other cpus.
2350 if (unlikely(!this_cpu_cmpxchg_double(
2351 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2353 next_object
, next_tid(tid
)))) {
2355 note_cmpxchg_failure("slab_alloc", s
, tid
);
2358 prefetch_freepointer(s
, next_object
);
2359 stat(s
, ALLOC_FASTPATH
);
2362 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2363 memset(object
, 0, s
->object_size
);
2365 slab_post_alloc_hook(s
, gfpflags
, object
);
2370 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2372 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2374 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
, s
->size
, gfpflags
);
2378 EXPORT_SYMBOL(kmem_cache_alloc
);
2380 #ifdef CONFIG_TRACING
2381 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2383 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2384 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2387 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2389 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2391 void *ret
= kmalloc_order(size
, flags
, order
);
2392 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2395 EXPORT_SYMBOL(kmalloc_order_trace
);
2399 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2401 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2403 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2404 s
->object_size
, s
->size
, gfpflags
, node
);
2408 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2410 #ifdef CONFIG_TRACING
2411 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2413 int node
, size_t size
)
2415 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2417 trace_kmalloc_node(_RET_IP_
, ret
,
2418 size
, s
->size
, gfpflags
, node
);
2421 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2426 * Slow patch handling. This may still be called frequently since objects
2427 * have a longer lifetime than the cpu slabs in most processing loads.
2429 * So we still attempt to reduce cache line usage. Just take the slab
2430 * lock and free the item. If there is no additional partial page
2431 * handling required then we can return immediately.
2433 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2434 void *x
, unsigned long addr
)
2437 void **object
= (void *)x
;
2441 unsigned long counters
;
2442 struct kmem_cache_node
*n
= NULL
;
2443 unsigned long uninitialized_var(flags
);
2445 stat(s
, FREE_SLOWPATH
);
2447 if (kmem_cache_debug(s
) && !free_debug_processing(s
, page
, x
, addr
))
2451 prior
= page
->freelist
;
2452 counters
= page
->counters
;
2453 set_freepointer(s
, object
, prior
);
2454 new.counters
= counters
;
2455 was_frozen
= new.frozen
;
2457 if ((!new.inuse
|| !prior
) && !was_frozen
&& !n
) {
2459 if (!kmem_cache_debug(s
) && !prior
)
2462 * Slab was on no list before and will be partially empty
2463 * We can defer the list move and instead freeze it.
2467 else { /* Needs to be taken off a list */
2469 n
= get_node(s
, page_to_nid(page
));
2471 * Speculatively acquire the list_lock.
2472 * If the cmpxchg does not succeed then we may
2473 * drop the list_lock without any processing.
2475 * Otherwise the list_lock will synchronize with
2476 * other processors updating the list of slabs.
2478 spin_lock_irqsave(&n
->list_lock
, flags
);
2484 } while (!cmpxchg_double_slab(s
, page
,
2486 object
, new.counters
,
2492 * If we just froze the page then put it onto the
2493 * per cpu partial list.
2495 if (new.frozen
&& !was_frozen
) {
2496 put_cpu_partial(s
, page
, 1);
2497 stat(s
, CPU_PARTIAL_FREE
);
2500 * The list lock was not taken therefore no list
2501 * activity can be necessary.
2504 stat(s
, FREE_FROZEN
);
2509 * was_frozen may have been set after we acquired the list_lock in
2510 * an earlier loop. So we need to check it here again.
2513 stat(s
, FREE_FROZEN
);
2515 if (unlikely(!inuse
&& n
->nr_partial
> s
->min_partial
))
2519 * Objects left in the slab. If it was not on the partial list before
2522 if (unlikely(!prior
)) {
2523 remove_full(s
, page
);
2524 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2525 stat(s
, FREE_ADD_PARTIAL
);
2528 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2534 * Slab on the partial list.
2536 remove_partial(n
, page
);
2537 stat(s
, FREE_REMOVE_PARTIAL
);
2539 /* Slab must be on the full list */
2540 remove_full(s
, page
);
2542 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2544 discard_slab(s
, page
);
2548 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2549 * can perform fastpath freeing without additional function calls.
2551 * The fastpath is only possible if we are freeing to the current cpu slab
2552 * of this processor. This typically the case if we have just allocated
2555 * If fastpath is not possible then fall back to __slab_free where we deal
2556 * with all sorts of special processing.
2558 static __always_inline
void slab_free(struct kmem_cache
*s
,
2559 struct page
*page
, void *x
, unsigned long addr
)
2561 void **object
= (void *)x
;
2562 struct kmem_cache_cpu
*c
;
2565 slab_free_hook(s
, x
);
2569 * Determine the currently cpus per cpu slab.
2570 * The cpu may change afterward. However that does not matter since
2571 * data is retrieved via this pointer. If we are on the same cpu
2572 * during the cmpxchg then the free will succedd.
2574 c
= __this_cpu_ptr(s
->cpu_slab
);
2579 if (likely(page
== c
->page
)) {
2580 set_freepointer(s
, object
, c
->freelist
);
2582 if (unlikely(!this_cpu_cmpxchg_double(
2583 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2585 object
, next_tid(tid
)))) {
2587 note_cmpxchg_failure("slab_free", s
, tid
);
2590 stat(s
, FREE_FASTPATH
);
2592 __slab_free(s
, page
, x
, addr
);
2596 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2600 page
= virt_to_head_page(x
);
2602 slab_free(s
, page
, x
, _RET_IP_
);
2604 trace_kmem_cache_free(_RET_IP_
, x
);
2606 EXPORT_SYMBOL(kmem_cache_free
);
2609 * Object placement in a slab is made very easy because we always start at
2610 * offset 0. If we tune the size of the object to the alignment then we can
2611 * get the required alignment by putting one properly sized object after
2614 * Notice that the allocation order determines the sizes of the per cpu
2615 * caches. Each processor has always one slab available for allocations.
2616 * Increasing the allocation order reduces the number of times that slabs
2617 * must be moved on and off the partial lists and is therefore a factor in
2622 * Mininum / Maximum order of slab pages. This influences locking overhead
2623 * and slab fragmentation. A higher order reduces the number of partial slabs
2624 * and increases the number of allocations possible without having to
2625 * take the list_lock.
2627 static int slub_min_order
;
2628 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2629 static int slub_min_objects
;
2632 * Merge control. If this is set then no merging of slab caches will occur.
2633 * (Could be removed. This was introduced to pacify the merge skeptics.)
2635 static int slub_nomerge
;
2638 * Calculate the order of allocation given an slab object size.
2640 * The order of allocation has significant impact on performance and other
2641 * system components. Generally order 0 allocations should be preferred since
2642 * order 0 does not cause fragmentation in the page allocator. Larger objects
2643 * be problematic to put into order 0 slabs because there may be too much
2644 * unused space left. We go to a higher order if more than 1/16th of the slab
2647 * In order to reach satisfactory performance we must ensure that a minimum
2648 * number of objects is in one slab. Otherwise we may generate too much
2649 * activity on the partial lists which requires taking the list_lock. This is
2650 * less a concern for large slabs though which are rarely used.
2652 * slub_max_order specifies the order where we begin to stop considering the
2653 * number of objects in a slab as critical. If we reach slub_max_order then
2654 * we try to keep the page order as low as possible. So we accept more waste
2655 * of space in favor of a small page order.
2657 * Higher order allocations also allow the placement of more objects in a
2658 * slab and thereby reduce object handling overhead. If the user has
2659 * requested a higher mininum order then we start with that one instead of
2660 * the smallest order which will fit the object.
2662 static inline int slab_order(int size
, int min_objects
,
2663 int max_order
, int fract_leftover
, int reserved
)
2667 int min_order
= slub_min_order
;
2669 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2670 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2672 for (order
= max(min_order
,
2673 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2674 order
<= max_order
; order
++) {
2676 unsigned long slab_size
= PAGE_SIZE
<< order
;
2678 if (slab_size
< min_objects
* size
+ reserved
)
2681 rem
= (slab_size
- reserved
) % size
;
2683 if (rem
<= slab_size
/ fract_leftover
)
2691 static inline int calculate_order(int size
, int reserved
)
2699 * Attempt to find best configuration for a slab. This
2700 * works by first attempting to generate a layout with
2701 * the best configuration and backing off gradually.
2703 * First we reduce the acceptable waste in a slab. Then
2704 * we reduce the minimum objects required in a slab.
2706 min_objects
= slub_min_objects
;
2708 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2709 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2710 min_objects
= min(min_objects
, max_objects
);
2712 while (min_objects
> 1) {
2714 while (fraction
>= 4) {
2715 order
= slab_order(size
, min_objects
,
2716 slub_max_order
, fraction
, reserved
);
2717 if (order
<= slub_max_order
)
2725 * We were unable to place multiple objects in a slab. Now
2726 * lets see if we can place a single object there.
2728 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2729 if (order
<= slub_max_order
)
2733 * Doh this slab cannot be placed using slub_max_order.
2735 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2736 if (order
< MAX_ORDER
)
2742 * Figure out what the alignment of the objects will be.
2744 static unsigned long calculate_alignment(unsigned long flags
,
2745 unsigned long align
, unsigned long size
)
2748 * If the user wants hardware cache aligned objects then follow that
2749 * suggestion if the object is sufficiently large.
2751 * The hardware cache alignment cannot override the specified
2752 * alignment though. If that is greater then use it.
2754 if (flags
& SLAB_HWCACHE_ALIGN
) {
2755 unsigned long ralign
= cache_line_size();
2756 while (size
<= ralign
/ 2)
2758 align
= max(align
, ralign
);
2761 if (align
< ARCH_SLAB_MINALIGN
)
2762 align
= ARCH_SLAB_MINALIGN
;
2764 return ALIGN(align
, sizeof(void *));
2768 init_kmem_cache_node(struct kmem_cache_node
*n
)
2771 spin_lock_init(&n
->list_lock
);
2772 INIT_LIST_HEAD(&n
->partial
);
2773 #ifdef CONFIG_SLUB_DEBUG
2774 atomic_long_set(&n
->nr_slabs
, 0);
2775 atomic_long_set(&n
->total_objects
, 0);
2776 INIT_LIST_HEAD(&n
->full
);
2780 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2782 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2783 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2786 * Must align to double word boundary for the double cmpxchg
2787 * instructions to work; see __pcpu_double_call_return_bool().
2789 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2790 2 * sizeof(void *));
2795 init_kmem_cache_cpus(s
);
2800 static struct kmem_cache
*kmem_cache_node
;
2803 * No kmalloc_node yet so do it by hand. We know that this is the first
2804 * slab on the node for this slabcache. There are no concurrent accesses
2807 * Note that this function only works on the kmalloc_node_cache
2808 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2809 * memory on a fresh node that has no slab structures yet.
2811 static void early_kmem_cache_node_alloc(int node
)
2814 struct kmem_cache_node
*n
;
2816 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2818 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2821 if (page_to_nid(page
) != node
) {
2822 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2824 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2825 "in order to be able to continue\n");
2830 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2833 kmem_cache_node
->node
[node
] = n
;
2834 #ifdef CONFIG_SLUB_DEBUG
2835 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2836 init_tracking(kmem_cache_node
, n
);
2838 init_kmem_cache_node(n
);
2839 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2841 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2844 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2848 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2849 struct kmem_cache_node
*n
= s
->node
[node
];
2852 kmem_cache_free(kmem_cache_node
, n
);
2854 s
->node
[node
] = NULL
;
2858 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2862 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2863 struct kmem_cache_node
*n
;
2865 if (slab_state
== DOWN
) {
2866 early_kmem_cache_node_alloc(node
);
2869 n
= kmem_cache_alloc_node(kmem_cache_node
,
2873 free_kmem_cache_nodes(s
);
2878 init_kmem_cache_node(n
);
2883 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2885 if (min
< MIN_PARTIAL
)
2887 else if (min
> MAX_PARTIAL
)
2889 s
->min_partial
= min
;
2893 * calculate_sizes() determines the order and the distribution of data within
2896 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2898 unsigned long flags
= s
->flags
;
2899 unsigned long size
= s
->object_size
;
2900 unsigned long align
= s
->align
;
2904 * Round up object size to the next word boundary. We can only
2905 * place the free pointer at word boundaries and this determines
2906 * the possible location of the free pointer.
2908 size
= ALIGN(size
, sizeof(void *));
2910 #ifdef CONFIG_SLUB_DEBUG
2912 * Determine if we can poison the object itself. If the user of
2913 * the slab may touch the object after free or before allocation
2914 * then we should never poison the object itself.
2916 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2918 s
->flags
|= __OBJECT_POISON
;
2920 s
->flags
&= ~__OBJECT_POISON
;
2924 * If we are Redzoning then check if there is some space between the
2925 * end of the object and the free pointer. If not then add an
2926 * additional word to have some bytes to store Redzone information.
2928 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
2929 size
+= sizeof(void *);
2933 * With that we have determined the number of bytes in actual use
2934 * by the object. This is the potential offset to the free pointer.
2938 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2941 * Relocate free pointer after the object if it is not
2942 * permitted to overwrite the first word of the object on
2945 * This is the case if we do RCU, have a constructor or
2946 * destructor or are poisoning the objects.
2949 size
+= sizeof(void *);
2952 #ifdef CONFIG_SLUB_DEBUG
2953 if (flags
& SLAB_STORE_USER
)
2955 * Need to store information about allocs and frees after
2958 size
+= 2 * sizeof(struct track
);
2960 if (flags
& SLAB_RED_ZONE
)
2962 * Add some empty padding so that we can catch
2963 * overwrites from earlier objects rather than let
2964 * tracking information or the free pointer be
2965 * corrupted if a user writes before the start
2968 size
+= sizeof(void *);
2972 * Determine the alignment based on various parameters that the
2973 * user specified and the dynamic determination of cache line size
2976 align
= calculate_alignment(flags
, align
, s
->object_size
);
2980 * SLUB stores one object immediately after another beginning from
2981 * offset 0. In order to align the objects we have to simply size
2982 * each object to conform to the alignment.
2984 size
= ALIGN(size
, align
);
2986 if (forced_order
>= 0)
2987 order
= forced_order
;
2989 order
= calculate_order(size
, s
->reserved
);
2996 s
->allocflags
|= __GFP_COMP
;
2998 if (s
->flags
& SLAB_CACHE_DMA
)
2999 s
->allocflags
|= SLUB_DMA
;
3001 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3002 s
->allocflags
|= __GFP_RECLAIMABLE
;
3005 * Determine the number of objects per slab
3007 s
->oo
= oo_make(order
, size
, s
->reserved
);
3008 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
3009 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3012 return !!oo_objects(s
->oo
);
3016 static int kmem_cache_open(struct kmem_cache
*s
,
3017 const char *name
, size_t size
,
3018 size_t align
, unsigned long flags
,
3019 void (*ctor
)(void *))
3021 memset(s
, 0, kmem_size
);
3024 s
->object_size
= size
;
3026 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
3029 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3030 s
->reserved
= sizeof(struct rcu_head
);
3032 if (!calculate_sizes(s
, -1))
3034 if (disable_higher_order_debug
) {
3036 * Disable debugging flags that store metadata if the min slab
3039 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3040 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3042 if (!calculate_sizes(s
, -1))
3047 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3048 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3049 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3050 /* Enable fast mode */
3051 s
->flags
|= __CMPXCHG_DOUBLE
;
3055 * The larger the object size is, the more pages we want on the partial
3056 * list to avoid pounding the page allocator excessively.
3058 set_min_partial(s
, ilog2(s
->size
) / 2);
3061 * cpu_partial determined the maximum number of objects kept in the
3062 * per cpu partial lists of a processor.
3064 * Per cpu partial lists mainly contain slabs that just have one
3065 * object freed. If they are used for allocation then they can be
3066 * filled up again with minimal effort. The slab will never hit the
3067 * per node partial lists and therefore no locking will be required.
3069 * This setting also determines
3071 * A) The number of objects from per cpu partial slabs dumped to the
3072 * per node list when we reach the limit.
3073 * B) The number of objects in cpu partial slabs to extract from the
3074 * per node list when we run out of per cpu objects. We only fetch 50%
3075 * to keep some capacity around for frees.
3077 if (kmem_cache_debug(s
))
3079 else if (s
->size
>= PAGE_SIZE
)
3081 else if (s
->size
>= 1024)
3083 else if (s
->size
>= 256)
3084 s
->cpu_partial
= 13;
3086 s
->cpu_partial
= 30;
3090 s
->remote_node_defrag_ratio
= 1000;
3092 if (!init_kmem_cache_nodes(s
))
3095 if (alloc_kmem_cache_cpus(s
))
3098 free_kmem_cache_nodes(s
);
3100 if (flags
& SLAB_PANIC
)
3101 panic("Cannot create slab %s size=%lu realsize=%u "
3102 "order=%u offset=%u flags=%lx\n",
3103 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
3109 * Determine the size of a slab object
3111 unsigned int kmem_cache_size(struct kmem_cache
*s
)
3113 return s
->object_size
;
3115 EXPORT_SYMBOL(kmem_cache_size
);
3117 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3120 #ifdef CONFIG_SLUB_DEBUG
3121 void *addr
= page_address(page
);
3123 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3124 sizeof(long), GFP_ATOMIC
);
3127 slab_err(s
, page
, "%s", text
);
3130 get_map(s
, page
, map
);
3131 for_each_object(p
, s
, addr
, page
->objects
) {
3133 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3134 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3136 print_tracking(s
, p
);
3145 * Attempt to free all partial slabs on a node.
3146 * This is called from kmem_cache_close(). We must be the last thread
3147 * using the cache and therefore we do not need to lock anymore.
3149 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3151 struct page
*page
, *h
;
3153 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3155 remove_partial(n
, page
);
3156 discard_slab(s
, page
);
3158 list_slab_objects(s
, page
,
3159 "Objects remaining on kmem_cache_close()");
3165 * Release all resources used by a slab cache.
3167 static inline int kmem_cache_close(struct kmem_cache
*s
)
3172 free_percpu(s
->cpu_slab
);
3173 /* Attempt to free all objects */
3174 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3175 struct kmem_cache_node
*n
= get_node(s
, node
);
3178 if (n
->nr_partial
|| slabs_node(s
, node
))
3181 free_kmem_cache_nodes(s
);
3186 * Close a cache and release the kmem_cache structure
3187 * (must be used for caches created using kmem_cache_create)
3189 void kmem_cache_destroy(struct kmem_cache
*s
)
3191 down_write(&slub_lock
);
3195 up_write(&slub_lock
);
3196 if (kmem_cache_close(s
)) {
3197 printk(KERN_ERR
"SLUB %s: %s called for cache that "
3198 "still has objects.\n", s
->name
, __func__
);
3201 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
3203 sysfs_slab_remove(s
);
3205 up_write(&slub_lock
);
3207 EXPORT_SYMBOL(kmem_cache_destroy
);
3209 /********************************************************************
3211 *******************************************************************/
3213 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
3214 EXPORT_SYMBOL(kmalloc_caches
);
3216 static struct kmem_cache
*kmem_cache
;
3218 #ifdef CONFIG_ZONE_DMA
3219 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
3222 static int __init
setup_slub_min_order(char *str
)
3224 get_option(&str
, &slub_min_order
);
3229 __setup("slub_min_order=", setup_slub_min_order
);
3231 static int __init
setup_slub_max_order(char *str
)
3233 get_option(&str
, &slub_max_order
);
3234 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3239 __setup("slub_max_order=", setup_slub_max_order
);
3241 static int __init
setup_slub_min_objects(char *str
)
3243 get_option(&str
, &slub_min_objects
);
3248 __setup("slub_min_objects=", setup_slub_min_objects
);
3250 static int __init
setup_slub_nomerge(char *str
)
3256 __setup("slub_nomerge", setup_slub_nomerge
);
3258 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
3259 int size
, unsigned int flags
)
3261 struct kmem_cache
*s
;
3263 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3266 * This function is called with IRQs disabled during early-boot on
3267 * single CPU so there's no need to take slub_lock here.
3269 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
3273 list_add(&s
->list
, &slab_caches
);
3277 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
3282 * Conversion table for small slabs sizes / 8 to the index in the
3283 * kmalloc array. This is necessary for slabs < 192 since we have non power
3284 * of two cache sizes there. The size of larger slabs can be determined using
3287 static s8 size_index
[24] = {
3314 static inline int size_index_elem(size_t bytes
)
3316 return (bytes
- 1) / 8;
3319 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
3325 return ZERO_SIZE_PTR
;
3327 index
= size_index
[size_index_elem(size
)];
3329 index
= fls(size
- 1);
3331 #ifdef CONFIG_ZONE_DMA
3332 if (unlikely((flags
& SLUB_DMA
)))
3333 return kmalloc_dma_caches
[index
];
3336 return kmalloc_caches
[index
];
3339 void *__kmalloc(size_t size
, gfp_t flags
)
3341 struct kmem_cache
*s
;
3344 if (unlikely(size
> SLUB_MAX_SIZE
))
3345 return kmalloc_large(size
, flags
);
3347 s
= get_slab(size
, flags
);
3349 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3352 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
3354 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3358 EXPORT_SYMBOL(__kmalloc
);
3361 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3366 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3367 page
= alloc_pages_node(node
, flags
, get_order(size
));
3369 ptr
= page_address(page
);
3371 kmemleak_alloc(ptr
, size
, 1, flags
);
3375 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3377 struct kmem_cache
*s
;
3380 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3381 ret
= kmalloc_large_node(size
, flags
, node
);
3383 trace_kmalloc_node(_RET_IP_
, ret
,
3384 size
, PAGE_SIZE
<< get_order(size
),
3390 s
= get_slab(size
, flags
);
3392 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3395 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
3397 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3401 EXPORT_SYMBOL(__kmalloc_node
);
3404 size_t ksize(const void *object
)
3408 if (unlikely(object
== ZERO_SIZE_PTR
))
3411 page
= virt_to_head_page(object
);
3413 if (unlikely(!PageSlab(page
))) {
3414 WARN_ON(!PageCompound(page
));
3415 return PAGE_SIZE
<< compound_order(page
);
3418 return slab_ksize(page
->slab
);
3420 EXPORT_SYMBOL(ksize
);
3422 #ifdef CONFIG_SLUB_DEBUG
3423 bool verify_mem_not_deleted(const void *x
)
3426 void *object
= (void *)x
;
3427 unsigned long flags
;
3430 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3433 local_irq_save(flags
);
3435 page
= virt_to_head_page(x
);
3436 if (unlikely(!PageSlab(page
))) {
3437 /* maybe it was from stack? */
3443 if (on_freelist(page
->slab
, page
, object
)) {
3444 object_err(page
->slab
, page
, object
, "Object is on free-list");
3452 local_irq_restore(flags
);
3455 EXPORT_SYMBOL(verify_mem_not_deleted
);
3458 void kfree(const void *x
)
3461 void *object
= (void *)x
;
3463 trace_kfree(_RET_IP_
, x
);
3465 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3468 page
= virt_to_head_page(x
);
3469 if (unlikely(!PageSlab(page
))) {
3470 BUG_ON(!PageCompound(page
));
3475 slab_free(page
->slab
, page
, object
, _RET_IP_
);
3477 EXPORT_SYMBOL(kfree
);
3480 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3481 * the remaining slabs by the number of items in use. The slabs with the
3482 * most items in use come first. New allocations will then fill those up
3483 * and thus they can be removed from the partial lists.
3485 * The slabs with the least items are placed last. This results in them
3486 * being allocated from last increasing the chance that the last objects
3487 * are freed in them.
3489 int kmem_cache_shrink(struct kmem_cache
*s
)
3493 struct kmem_cache_node
*n
;
3496 int objects
= oo_objects(s
->max
);
3497 struct list_head
*slabs_by_inuse
=
3498 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3499 unsigned long flags
;
3501 if (!slabs_by_inuse
)
3505 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3506 n
= get_node(s
, node
);
3511 for (i
= 0; i
< objects
; i
++)
3512 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3514 spin_lock_irqsave(&n
->list_lock
, flags
);
3517 * Build lists indexed by the items in use in each slab.
3519 * Note that concurrent frees may occur while we hold the
3520 * list_lock. page->inuse here is the upper limit.
3522 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3523 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3529 * Rebuild the partial list with the slabs filled up most
3530 * first and the least used slabs at the end.
3532 for (i
= objects
- 1; i
> 0; i
--)
3533 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3535 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3537 /* Release empty slabs */
3538 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3539 discard_slab(s
, page
);
3542 kfree(slabs_by_inuse
);
3545 EXPORT_SYMBOL(kmem_cache_shrink
);
3547 #if defined(CONFIG_MEMORY_HOTPLUG)
3548 static int slab_mem_going_offline_callback(void *arg
)
3550 struct kmem_cache
*s
;
3552 down_read(&slub_lock
);
3553 list_for_each_entry(s
, &slab_caches
, list
)
3554 kmem_cache_shrink(s
);
3555 up_read(&slub_lock
);
3560 static void slab_mem_offline_callback(void *arg
)
3562 struct kmem_cache_node
*n
;
3563 struct kmem_cache
*s
;
3564 struct memory_notify
*marg
= arg
;
3567 offline_node
= marg
->status_change_nid
;
3570 * If the node still has available memory. we need kmem_cache_node
3573 if (offline_node
< 0)
3576 down_read(&slub_lock
);
3577 list_for_each_entry(s
, &slab_caches
, list
) {
3578 n
= get_node(s
, offline_node
);
3581 * if n->nr_slabs > 0, slabs still exist on the node
3582 * that is going down. We were unable to free them,
3583 * and offline_pages() function shouldn't call this
3584 * callback. So, we must fail.
3586 BUG_ON(slabs_node(s
, offline_node
));
3588 s
->node
[offline_node
] = NULL
;
3589 kmem_cache_free(kmem_cache_node
, n
);
3592 up_read(&slub_lock
);
3595 static int slab_mem_going_online_callback(void *arg
)
3597 struct kmem_cache_node
*n
;
3598 struct kmem_cache
*s
;
3599 struct memory_notify
*marg
= arg
;
3600 int nid
= marg
->status_change_nid
;
3604 * If the node's memory is already available, then kmem_cache_node is
3605 * already created. Nothing to do.
3611 * We are bringing a node online. No memory is available yet. We must
3612 * allocate a kmem_cache_node structure in order to bring the node
3615 down_read(&slub_lock
);
3616 list_for_each_entry(s
, &slab_caches
, list
) {
3618 * XXX: kmem_cache_alloc_node will fallback to other nodes
3619 * since memory is not yet available from the node that
3622 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3627 init_kmem_cache_node(n
);
3631 up_read(&slub_lock
);
3635 static int slab_memory_callback(struct notifier_block
*self
,
3636 unsigned long action
, void *arg
)
3641 case MEM_GOING_ONLINE
:
3642 ret
= slab_mem_going_online_callback(arg
);
3644 case MEM_GOING_OFFLINE
:
3645 ret
= slab_mem_going_offline_callback(arg
);
3648 case MEM_CANCEL_ONLINE
:
3649 slab_mem_offline_callback(arg
);
3652 case MEM_CANCEL_OFFLINE
:
3656 ret
= notifier_from_errno(ret
);
3662 #endif /* CONFIG_MEMORY_HOTPLUG */
3664 /********************************************************************
3665 * Basic setup of slabs
3666 *******************************************************************/
3669 * Used for early kmem_cache structures that were allocated using
3670 * the page allocator
3673 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3677 list_add(&s
->list
, &slab_caches
);
3680 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3681 struct kmem_cache_node
*n
= get_node(s
, node
);
3685 list_for_each_entry(p
, &n
->partial
, lru
)
3688 #ifdef CONFIG_SLUB_DEBUG
3689 list_for_each_entry(p
, &n
->full
, lru
)
3696 void __init
kmem_cache_init(void)
3700 struct kmem_cache
*temp_kmem_cache
;
3702 struct kmem_cache
*temp_kmem_cache_node
;
3703 unsigned long kmalloc_size
;
3705 if (debug_guardpage_minorder())
3708 kmem_size
= offsetof(struct kmem_cache
, node
) +
3709 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3711 /* Allocate two kmem_caches from the page allocator */
3712 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3713 order
= get_order(2 * kmalloc_size
);
3714 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3717 * Must first have the slab cache available for the allocations of the
3718 * struct kmem_cache_node's. There is special bootstrap code in
3719 * kmem_cache_open for slab_state == DOWN.
3721 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3723 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3724 sizeof(struct kmem_cache_node
),
3725 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3727 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3729 /* Able to allocate the per node structures */
3730 slab_state
= PARTIAL
;
3732 temp_kmem_cache
= kmem_cache
;
3733 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3734 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3735 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3736 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3739 * Allocate kmem_cache_node properly from the kmem_cache slab.
3740 * kmem_cache_node is separately allocated so no need to
3741 * update any list pointers.
3743 temp_kmem_cache_node
= kmem_cache_node
;
3745 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3746 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3748 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3751 kmem_cache_bootstrap_fixup(kmem_cache
);
3753 /* Free temporary boot structure */
3754 free_pages((unsigned long)temp_kmem_cache
, order
);
3756 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3759 * Patch up the size_index table if we have strange large alignment
3760 * requirements for the kmalloc array. This is only the case for
3761 * MIPS it seems. The standard arches will not generate any code here.
3763 * Largest permitted alignment is 256 bytes due to the way we
3764 * handle the index determination for the smaller caches.
3766 * Make sure that nothing crazy happens if someone starts tinkering
3767 * around with ARCH_KMALLOC_MINALIGN
3769 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3770 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3772 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3773 int elem
= size_index_elem(i
);
3774 if (elem
>= ARRAY_SIZE(size_index
))
3776 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3779 if (KMALLOC_MIN_SIZE
== 64) {
3781 * The 96 byte size cache is not used if the alignment
3784 for (i
= 64 + 8; i
<= 96; i
+= 8)
3785 size_index
[size_index_elem(i
)] = 7;
3786 } else if (KMALLOC_MIN_SIZE
== 128) {
3788 * The 192 byte sized cache is not used if the alignment
3789 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3792 for (i
= 128 + 8; i
<= 192; i
+= 8)
3793 size_index
[size_index_elem(i
)] = 8;
3796 /* Caches that are not of the two-to-the-power-of size */
3797 if (KMALLOC_MIN_SIZE
<= 32) {
3798 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3802 if (KMALLOC_MIN_SIZE
<= 64) {
3803 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3807 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3808 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3814 /* Provide the correct kmalloc names now that the caches are up */
3815 if (KMALLOC_MIN_SIZE
<= 32) {
3816 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3817 BUG_ON(!kmalloc_caches
[1]->name
);
3820 if (KMALLOC_MIN_SIZE
<= 64) {
3821 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3822 BUG_ON(!kmalloc_caches
[2]->name
);
3825 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3826 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3829 kmalloc_caches
[i
]->name
= s
;
3833 register_cpu_notifier(&slab_notifier
);
3836 #ifdef CONFIG_ZONE_DMA
3837 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3838 struct kmem_cache
*s
= kmalloc_caches
[i
];
3841 char *name
= kasprintf(GFP_NOWAIT
,
3842 "dma-kmalloc-%d", s
->object_size
);
3845 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3846 s
->object_size
, SLAB_CACHE_DMA
);
3851 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3852 " CPUs=%d, Nodes=%d\n",
3853 caches
, cache_line_size(),
3854 slub_min_order
, slub_max_order
, slub_min_objects
,
3855 nr_cpu_ids
, nr_node_ids
);
3858 void __init
kmem_cache_init_late(void)
3863 * Find a mergeable slab cache
3865 static int slab_unmergeable(struct kmem_cache
*s
)
3867 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3874 * We may have set a slab to be unmergeable during bootstrap.
3876 if (s
->refcount
< 0)
3882 static struct kmem_cache
*find_mergeable(size_t size
,
3883 size_t align
, unsigned long flags
, const char *name
,
3884 void (*ctor
)(void *))
3886 struct kmem_cache
*s
;
3888 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3894 size
= ALIGN(size
, sizeof(void *));
3895 align
= calculate_alignment(flags
, align
, size
);
3896 size
= ALIGN(size
, align
);
3897 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3899 list_for_each_entry(s
, &slab_caches
, list
) {
3900 if (slab_unmergeable(s
))
3906 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3909 * Check if alignment is compatible.
3910 * Courtesy of Adrian Drzewiecki
3912 if ((s
->size
& ~(align
- 1)) != s
->size
)
3915 if (s
->size
- size
>= sizeof(void *))
3923 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3924 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3926 struct kmem_cache
*s
;
3932 down_write(&slub_lock
);
3933 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3937 * Adjust the object sizes so that we clear
3938 * the complete object on kzalloc.
3940 s
->object_size
= max(s
->object_size
, (int)size
);
3941 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3943 if (sysfs_slab_alias(s
, name
)) {
3947 up_write(&slub_lock
);
3951 n
= kstrdup(name
, GFP_KERNEL
);
3955 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3957 if (kmem_cache_open(s
, n
,
3958 size
, align
, flags
, ctor
)) {
3959 list_add(&s
->list
, &slab_caches
);
3960 up_write(&slub_lock
);
3961 if (sysfs_slab_add(s
)) {
3962 down_write(&slub_lock
);
3974 up_write(&slub_lock
);
3976 if (flags
& SLAB_PANIC
)
3977 panic("Cannot create slabcache %s\n", name
);
3982 EXPORT_SYMBOL(kmem_cache_create
);
3986 * Use the cpu notifier to insure that the cpu slabs are flushed when
3989 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3990 unsigned long action
, void *hcpu
)
3992 long cpu
= (long)hcpu
;
3993 struct kmem_cache
*s
;
3994 unsigned long flags
;
3997 case CPU_UP_CANCELED
:
3998 case CPU_UP_CANCELED_FROZEN
:
4000 case CPU_DEAD_FROZEN
:
4001 down_read(&slub_lock
);
4002 list_for_each_entry(s
, &slab_caches
, list
) {
4003 local_irq_save(flags
);
4004 __flush_cpu_slab(s
, cpu
);
4005 local_irq_restore(flags
);
4007 up_read(&slub_lock
);
4015 static struct notifier_block __cpuinitdata slab_notifier
= {
4016 .notifier_call
= slab_cpuup_callback
4021 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
4023 struct kmem_cache
*s
;
4026 if (unlikely(size
> SLUB_MAX_SIZE
))
4027 return kmalloc_large(size
, gfpflags
);
4029 s
= get_slab(size
, gfpflags
);
4031 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4034 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
4036 /* Honor the call site pointer we received. */
4037 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
4043 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
4044 int node
, unsigned long caller
)
4046 struct kmem_cache
*s
;
4049 if (unlikely(size
> SLUB_MAX_SIZE
)) {
4050 ret
= kmalloc_large_node(size
, gfpflags
, node
);
4052 trace_kmalloc_node(caller
, ret
,
4053 size
, PAGE_SIZE
<< get_order(size
),
4059 s
= get_slab(size
, gfpflags
);
4061 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4064 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
4066 /* Honor the call site pointer we received. */
4067 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
4074 static int count_inuse(struct page
*page
)
4079 static int count_total(struct page
*page
)
4081 return page
->objects
;
4085 #ifdef CONFIG_SLUB_DEBUG
4086 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
4090 void *addr
= page_address(page
);
4092 if (!check_slab(s
, page
) ||
4093 !on_freelist(s
, page
, NULL
))
4096 /* Now we know that a valid freelist exists */
4097 bitmap_zero(map
, page
->objects
);
4099 get_map(s
, page
, map
);
4100 for_each_object(p
, s
, addr
, page
->objects
) {
4101 if (test_bit(slab_index(p
, s
, addr
), map
))
4102 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
4106 for_each_object(p
, s
, addr
, page
->objects
)
4107 if (!test_bit(slab_index(p
, s
, addr
), map
))
4108 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
4113 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
4117 validate_slab(s
, page
, map
);
4121 static int validate_slab_node(struct kmem_cache
*s
,
4122 struct kmem_cache_node
*n
, unsigned long *map
)
4124 unsigned long count
= 0;
4126 unsigned long flags
;
4128 spin_lock_irqsave(&n
->list_lock
, flags
);
4130 list_for_each_entry(page
, &n
->partial
, lru
) {
4131 validate_slab_slab(s
, page
, map
);
4134 if (count
!= n
->nr_partial
)
4135 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
4136 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
4138 if (!(s
->flags
& SLAB_STORE_USER
))
4141 list_for_each_entry(page
, &n
->full
, lru
) {
4142 validate_slab_slab(s
, page
, map
);
4145 if (count
!= atomic_long_read(&n
->nr_slabs
))
4146 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
4147 "counter=%ld\n", s
->name
, count
,
4148 atomic_long_read(&n
->nr_slabs
));
4151 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4155 static long validate_slab_cache(struct kmem_cache
*s
)
4158 unsigned long count
= 0;
4159 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4160 sizeof(unsigned long), GFP_KERNEL
);
4166 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4167 struct kmem_cache_node
*n
= get_node(s
, node
);
4169 count
+= validate_slab_node(s
, n
, map
);
4175 * Generate lists of code addresses where slabcache objects are allocated
4180 unsigned long count
;
4187 DECLARE_BITMAP(cpus
, NR_CPUS
);
4193 unsigned long count
;
4194 struct location
*loc
;
4197 static void free_loc_track(struct loc_track
*t
)
4200 free_pages((unsigned long)t
->loc
,
4201 get_order(sizeof(struct location
) * t
->max
));
4204 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4209 order
= get_order(sizeof(struct location
) * max
);
4211 l
= (void *)__get_free_pages(flags
, order
);
4216 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4224 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4225 const struct track
*track
)
4227 long start
, end
, pos
;
4229 unsigned long caddr
;
4230 unsigned long age
= jiffies
- track
->when
;
4236 pos
= start
+ (end
- start
+ 1) / 2;
4239 * There is nothing at "end". If we end up there
4240 * we need to add something to before end.
4245 caddr
= t
->loc
[pos
].addr
;
4246 if (track
->addr
== caddr
) {
4252 if (age
< l
->min_time
)
4254 if (age
> l
->max_time
)
4257 if (track
->pid
< l
->min_pid
)
4258 l
->min_pid
= track
->pid
;
4259 if (track
->pid
> l
->max_pid
)
4260 l
->max_pid
= track
->pid
;
4262 cpumask_set_cpu(track
->cpu
,
4263 to_cpumask(l
->cpus
));
4265 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4269 if (track
->addr
< caddr
)
4276 * Not found. Insert new tracking element.
4278 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4284 (t
->count
- pos
) * sizeof(struct location
));
4287 l
->addr
= track
->addr
;
4291 l
->min_pid
= track
->pid
;
4292 l
->max_pid
= track
->pid
;
4293 cpumask_clear(to_cpumask(l
->cpus
));
4294 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4295 nodes_clear(l
->nodes
);
4296 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4300 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4301 struct page
*page
, enum track_item alloc
,
4304 void *addr
= page_address(page
);
4307 bitmap_zero(map
, page
->objects
);
4308 get_map(s
, page
, map
);
4310 for_each_object(p
, s
, addr
, page
->objects
)
4311 if (!test_bit(slab_index(p
, s
, addr
), map
))
4312 add_location(t
, s
, get_track(s
, p
, alloc
));
4315 static int list_locations(struct kmem_cache
*s
, char *buf
,
4316 enum track_item alloc
)
4320 struct loc_track t
= { 0, 0, NULL
};
4322 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4323 sizeof(unsigned long), GFP_KERNEL
);
4325 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4328 return sprintf(buf
, "Out of memory\n");
4330 /* Push back cpu slabs */
4333 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4334 struct kmem_cache_node
*n
= get_node(s
, node
);
4335 unsigned long flags
;
4338 if (!atomic_long_read(&n
->nr_slabs
))
4341 spin_lock_irqsave(&n
->list_lock
, flags
);
4342 list_for_each_entry(page
, &n
->partial
, lru
)
4343 process_slab(&t
, s
, page
, alloc
, map
);
4344 list_for_each_entry(page
, &n
->full
, lru
)
4345 process_slab(&t
, s
, page
, alloc
, map
);
4346 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4349 for (i
= 0; i
< t
.count
; i
++) {
4350 struct location
*l
= &t
.loc
[i
];
4352 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4354 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4357 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4359 len
+= sprintf(buf
+ len
, "<not-available>");
4361 if (l
->sum_time
!= l
->min_time
) {
4362 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4364 (long)div_u64(l
->sum_time
, l
->count
),
4367 len
+= sprintf(buf
+ len
, " age=%ld",
4370 if (l
->min_pid
!= l
->max_pid
)
4371 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4372 l
->min_pid
, l
->max_pid
);
4374 len
+= sprintf(buf
+ len
, " pid=%ld",
4377 if (num_online_cpus() > 1 &&
4378 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4379 len
< PAGE_SIZE
- 60) {
4380 len
+= sprintf(buf
+ len
, " cpus=");
4381 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4382 to_cpumask(l
->cpus
));
4385 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4386 len
< PAGE_SIZE
- 60) {
4387 len
+= sprintf(buf
+ len
, " nodes=");
4388 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4392 len
+= sprintf(buf
+ len
, "\n");
4398 len
+= sprintf(buf
, "No data\n");
4403 #ifdef SLUB_RESILIENCY_TEST
4404 static void resiliency_test(void)
4408 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
4410 printk(KERN_ERR
"SLUB resiliency testing\n");
4411 printk(KERN_ERR
"-----------------------\n");
4412 printk(KERN_ERR
"A. Corruption after allocation\n");
4414 p
= kzalloc(16, GFP_KERNEL
);
4416 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4417 " 0x12->0x%p\n\n", p
+ 16);
4419 validate_slab_cache(kmalloc_caches
[4]);
4421 /* Hmmm... The next two are dangerous */
4422 p
= kzalloc(32, GFP_KERNEL
);
4423 p
[32 + sizeof(void *)] = 0x34;
4424 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4425 " 0x34 -> -0x%p\n", p
);
4427 "If allocated object is overwritten then not detectable\n\n");
4429 validate_slab_cache(kmalloc_caches
[5]);
4430 p
= kzalloc(64, GFP_KERNEL
);
4431 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4433 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4436 "If allocated object is overwritten then not detectable\n\n");
4437 validate_slab_cache(kmalloc_caches
[6]);
4439 printk(KERN_ERR
"\nB. Corruption after free\n");
4440 p
= kzalloc(128, GFP_KERNEL
);
4443 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4444 validate_slab_cache(kmalloc_caches
[7]);
4446 p
= kzalloc(256, GFP_KERNEL
);
4449 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4451 validate_slab_cache(kmalloc_caches
[8]);
4453 p
= kzalloc(512, GFP_KERNEL
);
4456 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4457 validate_slab_cache(kmalloc_caches
[9]);
4461 static void resiliency_test(void) {};
4466 enum slab_stat_type
{
4467 SL_ALL
, /* All slabs */
4468 SL_PARTIAL
, /* Only partially allocated slabs */
4469 SL_CPU
, /* Only slabs used for cpu caches */
4470 SL_OBJECTS
, /* Determine allocated objects not slabs */
4471 SL_TOTAL
/* Determine object capacity not slabs */
4474 #define SO_ALL (1 << SL_ALL)
4475 #define SO_PARTIAL (1 << SL_PARTIAL)
4476 #define SO_CPU (1 << SL_CPU)
4477 #define SO_OBJECTS (1 << SL_OBJECTS)
4478 #define SO_TOTAL (1 << SL_TOTAL)
4480 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4481 char *buf
, unsigned long flags
)
4483 unsigned long total
= 0;
4486 unsigned long *nodes
;
4487 unsigned long *per_cpu
;
4489 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4492 per_cpu
= nodes
+ nr_node_ids
;
4494 if (flags
& SO_CPU
) {
4497 for_each_possible_cpu(cpu
) {
4498 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4502 page
= ACCESS_ONCE(c
->page
);
4506 node
= page_to_nid(page
);
4507 if (flags
& SO_TOTAL
)
4509 else if (flags
& SO_OBJECTS
)
4517 page
= ACCESS_ONCE(c
->partial
);
4528 lock_memory_hotplug();
4529 #ifdef CONFIG_SLUB_DEBUG
4530 if (flags
& SO_ALL
) {
4531 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4532 struct kmem_cache_node
*n
= get_node(s
, node
);
4534 if (flags
& SO_TOTAL
)
4535 x
= atomic_long_read(&n
->total_objects
);
4536 else if (flags
& SO_OBJECTS
)
4537 x
= atomic_long_read(&n
->total_objects
) -
4538 count_partial(n
, count_free
);
4541 x
= atomic_long_read(&n
->nr_slabs
);
4548 if (flags
& SO_PARTIAL
) {
4549 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4550 struct kmem_cache_node
*n
= get_node(s
, node
);
4552 if (flags
& SO_TOTAL
)
4553 x
= count_partial(n
, count_total
);
4554 else if (flags
& SO_OBJECTS
)
4555 x
= count_partial(n
, count_inuse
);
4562 x
= sprintf(buf
, "%lu", total
);
4564 for_each_node_state(node
, N_NORMAL_MEMORY
)
4566 x
+= sprintf(buf
+ x
, " N%d=%lu",
4569 unlock_memory_hotplug();
4571 return x
+ sprintf(buf
+ x
, "\n");
4574 #ifdef CONFIG_SLUB_DEBUG
4575 static int any_slab_objects(struct kmem_cache
*s
)
4579 for_each_online_node(node
) {
4580 struct kmem_cache_node
*n
= get_node(s
, node
);
4585 if (atomic_long_read(&n
->total_objects
))
4592 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4593 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4595 struct slab_attribute
{
4596 struct attribute attr
;
4597 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4598 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4601 #define SLAB_ATTR_RO(_name) \
4602 static struct slab_attribute _name##_attr = \
4603 __ATTR(_name, 0400, _name##_show, NULL)
4605 #define SLAB_ATTR(_name) \
4606 static struct slab_attribute _name##_attr = \
4607 __ATTR(_name, 0600, _name##_show, _name##_store)
4609 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4611 return sprintf(buf
, "%d\n", s
->size
);
4613 SLAB_ATTR_RO(slab_size
);
4615 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4617 return sprintf(buf
, "%d\n", s
->align
);
4619 SLAB_ATTR_RO(align
);
4621 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4623 return sprintf(buf
, "%d\n", s
->object_size
);
4625 SLAB_ATTR_RO(object_size
);
4627 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4629 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4631 SLAB_ATTR_RO(objs_per_slab
);
4633 static ssize_t
order_store(struct kmem_cache
*s
,
4634 const char *buf
, size_t length
)
4636 unsigned long order
;
4639 err
= strict_strtoul(buf
, 10, &order
);
4643 if (order
> slub_max_order
|| order
< slub_min_order
)
4646 calculate_sizes(s
, order
);
4650 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4652 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4656 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4658 return sprintf(buf
, "%lu\n", s
->min_partial
);
4661 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4667 err
= strict_strtoul(buf
, 10, &min
);
4671 set_min_partial(s
, min
);
4674 SLAB_ATTR(min_partial
);
4676 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4678 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4681 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4684 unsigned long objects
;
4687 err
= strict_strtoul(buf
, 10, &objects
);
4690 if (objects
&& kmem_cache_debug(s
))
4693 s
->cpu_partial
= objects
;
4697 SLAB_ATTR(cpu_partial
);
4699 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4703 return sprintf(buf
, "%pS\n", s
->ctor
);
4707 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4709 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4711 SLAB_ATTR_RO(aliases
);
4713 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4715 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4717 SLAB_ATTR_RO(partial
);
4719 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4721 return show_slab_objects(s
, buf
, SO_CPU
);
4723 SLAB_ATTR_RO(cpu_slabs
);
4725 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4727 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4729 SLAB_ATTR_RO(objects
);
4731 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4733 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4735 SLAB_ATTR_RO(objects_partial
);
4737 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4744 for_each_online_cpu(cpu
) {
4745 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4748 pages
+= page
->pages
;
4749 objects
+= page
->pobjects
;
4753 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4756 for_each_online_cpu(cpu
) {
4757 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4759 if (page
&& len
< PAGE_SIZE
- 20)
4760 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4761 page
->pobjects
, page
->pages
);
4764 return len
+ sprintf(buf
+ len
, "\n");
4766 SLAB_ATTR_RO(slabs_cpu_partial
);
4768 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4770 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4773 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4774 const char *buf
, size_t length
)
4776 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4778 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4781 SLAB_ATTR(reclaim_account
);
4783 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4785 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4787 SLAB_ATTR_RO(hwcache_align
);
4789 #ifdef CONFIG_ZONE_DMA
4790 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4792 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4794 SLAB_ATTR_RO(cache_dma
);
4797 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4799 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4801 SLAB_ATTR_RO(destroy_by_rcu
);
4803 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4805 return sprintf(buf
, "%d\n", s
->reserved
);
4807 SLAB_ATTR_RO(reserved
);
4809 #ifdef CONFIG_SLUB_DEBUG
4810 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4812 return show_slab_objects(s
, buf
, SO_ALL
);
4814 SLAB_ATTR_RO(slabs
);
4816 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4818 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4820 SLAB_ATTR_RO(total_objects
);
4822 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4824 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4827 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4828 const char *buf
, size_t length
)
4830 s
->flags
&= ~SLAB_DEBUG_FREE
;
4831 if (buf
[0] == '1') {
4832 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4833 s
->flags
|= SLAB_DEBUG_FREE
;
4837 SLAB_ATTR(sanity_checks
);
4839 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4841 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4844 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4847 s
->flags
&= ~SLAB_TRACE
;
4848 if (buf
[0] == '1') {
4849 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4850 s
->flags
|= SLAB_TRACE
;
4856 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4858 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4861 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4862 const char *buf
, size_t length
)
4864 if (any_slab_objects(s
))
4867 s
->flags
&= ~SLAB_RED_ZONE
;
4868 if (buf
[0] == '1') {
4869 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4870 s
->flags
|= SLAB_RED_ZONE
;
4872 calculate_sizes(s
, -1);
4875 SLAB_ATTR(red_zone
);
4877 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4879 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4882 static ssize_t
poison_store(struct kmem_cache
*s
,
4883 const char *buf
, size_t length
)
4885 if (any_slab_objects(s
))
4888 s
->flags
&= ~SLAB_POISON
;
4889 if (buf
[0] == '1') {
4890 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4891 s
->flags
|= SLAB_POISON
;
4893 calculate_sizes(s
, -1);
4898 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4900 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4903 static ssize_t
store_user_store(struct kmem_cache
*s
,
4904 const char *buf
, size_t length
)
4906 if (any_slab_objects(s
))
4909 s
->flags
&= ~SLAB_STORE_USER
;
4910 if (buf
[0] == '1') {
4911 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4912 s
->flags
|= SLAB_STORE_USER
;
4914 calculate_sizes(s
, -1);
4917 SLAB_ATTR(store_user
);
4919 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4924 static ssize_t
validate_store(struct kmem_cache
*s
,
4925 const char *buf
, size_t length
)
4929 if (buf
[0] == '1') {
4930 ret
= validate_slab_cache(s
);
4936 SLAB_ATTR(validate
);
4938 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4940 if (!(s
->flags
& SLAB_STORE_USER
))
4942 return list_locations(s
, buf
, TRACK_ALLOC
);
4944 SLAB_ATTR_RO(alloc_calls
);
4946 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4948 if (!(s
->flags
& SLAB_STORE_USER
))
4950 return list_locations(s
, buf
, TRACK_FREE
);
4952 SLAB_ATTR_RO(free_calls
);
4953 #endif /* CONFIG_SLUB_DEBUG */
4955 #ifdef CONFIG_FAILSLAB
4956 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4958 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4961 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4964 s
->flags
&= ~SLAB_FAILSLAB
;
4966 s
->flags
|= SLAB_FAILSLAB
;
4969 SLAB_ATTR(failslab
);
4972 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4977 static ssize_t
shrink_store(struct kmem_cache
*s
,
4978 const char *buf
, size_t length
)
4980 if (buf
[0] == '1') {
4981 int rc
= kmem_cache_shrink(s
);
4992 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4994 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4997 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4998 const char *buf
, size_t length
)
5000 unsigned long ratio
;
5003 err
= strict_strtoul(buf
, 10, &ratio
);
5008 s
->remote_node_defrag_ratio
= ratio
* 10;
5012 SLAB_ATTR(remote_node_defrag_ratio
);
5015 #ifdef CONFIG_SLUB_STATS
5016 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
5018 unsigned long sum
= 0;
5021 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
5026 for_each_online_cpu(cpu
) {
5027 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
5033 len
= sprintf(buf
, "%lu", sum
);
5036 for_each_online_cpu(cpu
) {
5037 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
5038 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
5042 return len
+ sprintf(buf
+ len
, "\n");
5045 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
5049 for_each_online_cpu(cpu
)
5050 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
5053 #define STAT_ATTR(si, text) \
5054 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5056 return show_stat(s, buf, si); \
5058 static ssize_t text##_store(struct kmem_cache *s, \
5059 const char *buf, size_t length) \
5061 if (buf[0] != '0') \
5063 clear_stat(s, si); \
5068 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5069 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
5070 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
5071 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
5072 STAT_ATTR(FREE_FROZEN
, free_frozen
);
5073 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
5074 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
5075 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
5076 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
5077 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
5078 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
5079 STAT_ATTR(FREE_SLAB
, free_slab
);
5080 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
5081 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
5082 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
5083 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
5084 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
5085 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
5086 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
5087 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
5088 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
5089 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
5090 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
5091 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
5092 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
5093 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
5096 static struct attribute
*slab_attrs
[] = {
5097 &slab_size_attr
.attr
,
5098 &object_size_attr
.attr
,
5099 &objs_per_slab_attr
.attr
,
5101 &min_partial_attr
.attr
,
5102 &cpu_partial_attr
.attr
,
5104 &objects_partial_attr
.attr
,
5106 &cpu_slabs_attr
.attr
,
5110 &hwcache_align_attr
.attr
,
5111 &reclaim_account_attr
.attr
,
5112 &destroy_by_rcu_attr
.attr
,
5114 &reserved_attr
.attr
,
5115 &slabs_cpu_partial_attr
.attr
,
5116 #ifdef CONFIG_SLUB_DEBUG
5117 &total_objects_attr
.attr
,
5119 &sanity_checks_attr
.attr
,
5121 &red_zone_attr
.attr
,
5123 &store_user_attr
.attr
,
5124 &validate_attr
.attr
,
5125 &alloc_calls_attr
.attr
,
5126 &free_calls_attr
.attr
,
5128 #ifdef CONFIG_ZONE_DMA
5129 &cache_dma_attr
.attr
,
5132 &remote_node_defrag_ratio_attr
.attr
,
5134 #ifdef CONFIG_SLUB_STATS
5135 &alloc_fastpath_attr
.attr
,
5136 &alloc_slowpath_attr
.attr
,
5137 &free_fastpath_attr
.attr
,
5138 &free_slowpath_attr
.attr
,
5139 &free_frozen_attr
.attr
,
5140 &free_add_partial_attr
.attr
,
5141 &free_remove_partial_attr
.attr
,
5142 &alloc_from_partial_attr
.attr
,
5143 &alloc_slab_attr
.attr
,
5144 &alloc_refill_attr
.attr
,
5145 &alloc_node_mismatch_attr
.attr
,
5146 &free_slab_attr
.attr
,
5147 &cpuslab_flush_attr
.attr
,
5148 &deactivate_full_attr
.attr
,
5149 &deactivate_empty_attr
.attr
,
5150 &deactivate_to_head_attr
.attr
,
5151 &deactivate_to_tail_attr
.attr
,
5152 &deactivate_remote_frees_attr
.attr
,
5153 &deactivate_bypass_attr
.attr
,
5154 &order_fallback_attr
.attr
,
5155 &cmpxchg_double_fail_attr
.attr
,
5156 &cmpxchg_double_cpu_fail_attr
.attr
,
5157 &cpu_partial_alloc_attr
.attr
,
5158 &cpu_partial_free_attr
.attr
,
5159 &cpu_partial_node_attr
.attr
,
5160 &cpu_partial_drain_attr
.attr
,
5162 #ifdef CONFIG_FAILSLAB
5163 &failslab_attr
.attr
,
5169 static struct attribute_group slab_attr_group
= {
5170 .attrs
= slab_attrs
,
5173 static ssize_t
slab_attr_show(struct kobject
*kobj
,
5174 struct attribute
*attr
,
5177 struct slab_attribute
*attribute
;
5178 struct kmem_cache
*s
;
5181 attribute
= to_slab_attr(attr
);
5184 if (!attribute
->show
)
5187 err
= attribute
->show(s
, buf
);
5192 static ssize_t
slab_attr_store(struct kobject
*kobj
,
5193 struct attribute
*attr
,
5194 const char *buf
, size_t len
)
5196 struct slab_attribute
*attribute
;
5197 struct kmem_cache
*s
;
5200 attribute
= to_slab_attr(attr
);
5203 if (!attribute
->store
)
5206 err
= attribute
->store(s
, buf
, len
);
5211 static void kmem_cache_release(struct kobject
*kobj
)
5213 struct kmem_cache
*s
= to_slab(kobj
);
5219 static const struct sysfs_ops slab_sysfs_ops
= {
5220 .show
= slab_attr_show
,
5221 .store
= slab_attr_store
,
5224 static struct kobj_type slab_ktype
= {
5225 .sysfs_ops
= &slab_sysfs_ops
,
5226 .release
= kmem_cache_release
5229 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5231 struct kobj_type
*ktype
= get_ktype(kobj
);
5233 if (ktype
== &slab_ktype
)
5238 static const struct kset_uevent_ops slab_uevent_ops
= {
5239 .filter
= uevent_filter
,
5242 static struct kset
*slab_kset
;
5244 #define ID_STR_LENGTH 64
5246 /* Create a unique string id for a slab cache:
5248 * Format :[flags-]size
5250 static char *create_unique_id(struct kmem_cache
*s
)
5252 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5259 * First flags affecting slabcache operations. We will only
5260 * get here for aliasable slabs so we do not need to support
5261 * too many flags. The flags here must cover all flags that
5262 * are matched during merging to guarantee that the id is
5265 if (s
->flags
& SLAB_CACHE_DMA
)
5267 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5269 if (s
->flags
& SLAB_DEBUG_FREE
)
5271 if (!(s
->flags
& SLAB_NOTRACK
))
5275 p
+= sprintf(p
, "%07d", s
->size
);
5276 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5280 static int sysfs_slab_add(struct kmem_cache
*s
)
5286 if (slab_state
< SYSFS
)
5287 /* Defer until later */
5290 unmergeable
= slab_unmergeable(s
);
5293 * Slabcache can never be merged so we can use the name proper.
5294 * This is typically the case for debug situations. In that
5295 * case we can catch duplicate names easily.
5297 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5301 * Create a unique name for the slab as a target
5304 name
= create_unique_id(s
);
5307 s
->kobj
.kset
= slab_kset
;
5308 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5310 kobject_put(&s
->kobj
);
5314 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5316 kobject_del(&s
->kobj
);
5317 kobject_put(&s
->kobj
);
5320 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5322 /* Setup first alias */
5323 sysfs_slab_alias(s
, s
->name
);
5329 static void sysfs_slab_remove(struct kmem_cache
*s
)
5331 if (slab_state
< SYSFS
)
5333 * Sysfs has not been setup yet so no need to remove the
5338 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5339 kobject_del(&s
->kobj
);
5340 kobject_put(&s
->kobj
);
5344 * Need to buffer aliases during bootup until sysfs becomes
5345 * available lest we lose that information.
5347 struct saved_alias
{
5348 struct kmem_cache
*s
;
5350 struct saved_alias
*next
;
5353 static struct saved_alias
*alias_list
;
5355 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5357 struct saved_alias
*al
;
5359 if (slab_state
== SYSFS
) {
5361 * If we have a leftover link then remove it.
5363 sysfs_remove_link(&slab_kset
->kobj
, name
);
5364 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5367 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5373 al
->next
= alias_list
;
5378 static int __init
slab_sysfs_init(void)
5380 struct kmem_cache
*s
;
5383 down_write(&slub_lock
);
5385 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5387 up_write(&slub_lock
);
5388 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5394 list_for_each_entry(s
, &slab_caches
, list
) {
5395 err
= sysfs_slab_add(s
);
5397 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5398 " to sysfs\n", s
->name
);
5401 while (alias_list
) {
5402 struct saved_alias
*al
= alias_list
;
5404 alias_list
= alias_list
->next
;
5405 err
= sysfs_slab_alias(al
->s
, al
->name
);
5407 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5408 " %s to sysfs\n", s
->name
);
5412 up_write(&slub_lock
);
5417 __initcall(slab_sysfs_init
);
5418 #endif /* CONFIG_SYSFS */
5421 * The /proc/slabinfo ABI
5423 #ifdef CONFIG_SLABINFO
5424 static void print_slabinfo_header(struct seq_file
*m
)
5426 seq_puts(m
, "slabinfo - version: 2.1\n");
5427 seq_puts(m
, "# name <active_objs> <num_objs> <object_size> "
5428 "<objperslab> <pagesperslab>");
5429 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
5430 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5434 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
5438 down_read(&slub_lock
);
5440 print_slabinfo_header(m
);
5442 return seq_list_start(&slab_caches
, *pos
);
5445 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
5447 return seq_list_next(p
, &slab_caches
, pos
);
5450 static void s_stop(struct seq_file
*m
, void *p
)
5452 up_read(&slub_lock
);
5455 static int s_show(struct seq_file
*m
, void *p
)
5457 unsigned long nr_partials
= 0;
5458 unsigned long nr_slabs
= 0;
5459 unsigned long nr_inuse
= 0;
5460 unsigned long nr_objs
= 0;
5461 unsigned long nr_free
= 0;
5462 struct kmem_cache
*s
;
5465 s
= list_entry(p
, struct kmem_cache
, list
);
5467 for_each_online_node(node
) {
5468 struct kmem_cache_node
*n
= get_node(s
, node
);
5473 nr_partials
+= n
->nr_partial
;
5474 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
5475 nr_objs
+= atomic_long_read(&n
->total_objects
);
5476 nr_free
+= count_partial(n
, count_free
);
5479 nr_inuse
= nr_objs
- nr_free
;
5481 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
5482 nr_objs
, s
->size
, oo_objects(s
->oo
),
5483 (1 << oo_order(s
->oo
)));
5484 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
5485 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
5491 static const struct seq_operations slabinfo_op
= {
5498 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
5500 return seq_open(file
, &slabinfo_op
);
5503 static const struct file_operations proc_slabinfo_operations
= {
5504 .open
= slabinfo_open
,
5506 .llseek
= seq_lseek
,
5507 .release
= seq_release
,
5510 static int __init
slab_proc_init(void)
5512 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
5515 module_init(slab_proc_init
);
5516 #endif /* CONFIG_SLABINFO */