2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/kmemcheck.h>
23 #include <linux/cpu.h>
24 #include <linux/cpuset.h>
25 #include <linux/mempolicy.h>
26 #include <linux/ctype.h>
27 #include <linux/debugobjects.h>
28 #include <linux/kallsyms.h>
29 #include <linux/memory.h>
30 #include <linux/math64.h>
31 #include <linux/fault-inject.h>
32 #include <linux/stacktrace.h>
33 #include <linux/prefetch.h>
35 #include <trace/events/kmem.h>
39 * 1. slub_lock (Global Semaphore)
41 * 3. slab_lock(page) (Only on some arches and for debugging)
45 * The role of the slub_lock is to protect the list of all the slabs
46 * and to synchronize major metadata changes to slab cache structures.
48 * The slab_lock is only used for debugging and on arches that do not
49 * have the ability to do a cmpxchg_double. It only protects the second
50 * double word in the page struct. Meaning
51 * A. page->freelist -> List of object free in a page
52 * B. page->counters -> Counters of objects
53 * C. page->frozen -> frozen state
55 * If a slab is frozen then it is exempt from list management. It is not
56 * on any list. The processor that froze the slab is the one who can
57 * perform list operations on the page. Other processors may put objects
58 * onto the freelist but the processor that froze the slab is the only
59 * one that can retrieve the objects from the page's freelist.
61 * The list_lock protects the partial and full list on each node and
62 * the partial slab counter. If taken then no new slabs may be added or
63 * removed from the lists nor make the number of partial slabs be modified.
64 * (Note that the total number of slabs is an atomic value that may be
65 * modified without taking the list lock).
67 * The list_lock is a centralized lock and thus we avoid taking it as
68 * much as possible. As long as SLUB does not have to handle partial
69 * slabs, operations can continue without any centralized lock. F.e.
70 * allocating a long series of objects that fill up slabs does not require
72 * Interrupts are disabled during allocation and deallocation in order to
73 * make the slab allocator safe to use in the context of an irq. In addition
74 * interrupts are disabled to ensure that the processor does not change
75 * while handling per_cpu slabs, due to kernel preemption.
77 * SLUB assigns one slab for allocation to each processor.
78 * Allocations only occur from these slabs called cpu slabs.
80 * Slabs with free elements are kept on a partial list and during regular
81 * operations no list for full slabs is used. If an object in a full slab is
82 * freed then the slab will show up again on the partial lists.
83 * We track full slabs for debugging purposes though because otherwise we
84 * cannot scan all objects.
86 * Slabs are freed when they become empty. Teardown and setup is
87 * minimal so we rely on the page allocators per cpu caches for
88 * fast frees and allocs.
90 * Overloading of page flags that are otherwise used for LRU management.
92 * PageActive The slab is frozen and exempt from list processing.
93 * This means that the slab is dedicated to a purpose
94 * such as satisfying allocations for a specific
95 * processor. Objects may be freed in the slab while
96 * it is frozen but slab_free will then skip the usual
97 * list operations. It is up to the processor holding
98 * the slab to integrate the slab into the slab lists
99 * when the slab is no longer needed.
101 * One use of this flag is to mark slabs that are
102 * used for allocations. Then such a slab becomes a cpu
103 * slab. The cpu slab may be equipped with an additional
104 * freelist that allows lockless access to
105 * free objects in addition to the regular freelist
106 * that requires the slab lock.
108 * PageError Slab requires special handling due to debug
109 * options set. This moves slab handling out of
110 * the fast path and disables lockless freelists.
113 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
114 SLAB_TRACE | SLAB_DEBUG_FREE)
116 static inline int kmem_cache_debug(struct kmem_cache
*s
)
118 #ifdef CONFIG_SLUB_DEBUG
119 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
126 * Issues still to be resolved:
128 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
130 * - Variable sizing of the per node arrays
133 /* Enable to test recovery from slab corruption on boot */
134 #undef SLUB_RESILIENCY_TEST
136 /* Enable to log cmpxchg failures */
137 #undef SLUB_DEBUG_CMPXCHG
140 * Mininum number of partial slabs. These will be left on the partial
141 * lists even if they are empty. kmem_cache_shrink may reclaim them.
143 #define MIN_PARTIAL 5
146 * Maximum number of desirable partial slabs.
147 * The existence of more partial slabs makes kmem_cache_shrink
148 * sort the partial list by the number of objects in the.
150 #define MAX_PARTIAL 10
152 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
153 SLAB_POISON | SLAB_STORE_USER)
156 * Debugging flags that require metadata to be stored in the slab. These get
157 * disabled when slub_debug=O is used and a cache's min order increases with
160 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
163 * Set of flags that will prevent slab merging
165 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
166 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
169 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
170 SLAB_CACHE_DMA | SLAB_NOTRACK)
173 #define OO_MASK ((1 << OO_SHIFT) - 1)
174 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
176 /* Internal SLUB flags */
177 #define __OBJECT_POISON 0x80000000UL /* Poison object */
178 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
180 static int kmem_size
= sizeof(struct kmem_cache
);
183 static struct notifier_block slab_notifier
;
186 /* A list of all slab caches on the system */
187 static DECLARE_RWSEM(slub_lock
);
188 static LIST_HEAD(slab_caches
);
191 * Tracking user of a slab.
193 #define TRACK_ADDRS_COUNT 16
195 unsigned long addr
; /* Called from address */
196 #ifdef CONFIG_STACKTRACE
197 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
199 int cpu
; /* Was running on cpu */
200 int pid
; /* Pid context */
201 unsigned long when
; /* When did the operation occur */
204 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
207 static int sysfs_slab_add(struct kmem_cache
*);
208 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
209 static void sysfs_slab_remove(struct kmem_cache
*);
212 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
213 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
215 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
223 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
225 #ifdef CONFIG_SLUB_STATS
226 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
230 /********************************************************************
231 * Core slab cache functions
232 *******************************************************************/
234 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
236 return s
->node
[node
];
239 /* Verify that a pointer has an address that is valid within a slab page */
240 static inline int check_valid_pointer(struct kmem_cache
*s
,
241 struct page
*page
, const void *object
)
248 base
= page_address(page
);
249 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
250 (object
- base
) % s
->size
) {
257 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
259 return *(void **)(object
+ s
->offset
);
262 static void prefetch_freepointer(const struct kmem_cache
*s
, void *object
)
264 prefetch(object
+ s
->offset
);
267 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
271 #ifdef CONFIG_DEBUG_PAGEALLOC
272 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
274 p
= get_freepointer(s
, object
);
279 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
281 *(void **)(object
+ s
->offset
) = fp
;
284 /* Loop over all objects in a slab */
285 #define for_each_object(__p, __s, __addr, __objects) \
286 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
289 /* Determine object index from a given position */
290 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
292 return (p
- addr
) / s
->size
;
295 static inline size_t slab_ksize(const struct kmem_cache
*s
)
297 #ifdef CONFIG_SLUB_DEBUG
299 * Debugging requires use of the padding between object
300 * and whatever may come after it.
302 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
303 return s
->object_size
;
307 * If we have the need to store the freelist pointer
308 * back there or track user information then we can
309 * only use the space before that information.
311 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
314 * Else we can use all the padding etc for the allocation
319 static inline int order_objects(int order
, unsigned long size
, int reserved
)
321 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
324 static inline struct kmem_cache_order_objects
oo_make(int order
,
325 unsigned long size
, int reserved
)
327 struct kmem_cache_order_objects x
= {
328 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
334 static inline int oo_order(struct kmem_cache_order_objects x
)
336 return x
.x
>> OO_SHIFT
;
339 static inline int oo_objects(struct kmem_cache_order_objects x
)
341 return x
.x
& OO_MASK
;
345 * Per slab locking using the pagelock
347 static __always_inline
void slab_lock(struct page
*page
)
349 bit_spin_lock(PG_locked
, &page
->flags
);
352 static __always_inline
void slab_unlock(struct page
*page
)
354 __bit_spin_unlock(PG_locked
, &page
->flags
);
357 /* Interrupts must be disabled (for the fallback code to work right) */
358 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
359 void *freelist_old
, unsigned long counters_old
,
360 void *freelist_new
, unsigned long counters_new
,
363 VM_BUG_ON(!irqs_disabled());
364 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
365 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
366 if (s
->flags
& __CMPXCHG_DOUBLE
) {
367 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
368 freelist_old
, counters_old
,
369 freelist_new
, counters_new
))
375 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
376 page
->freelist
= freelist_new
;
377 page
->counters
= counters_new
;
385 stat(s
, CMPXCHG_DOUBLE_FAIL
);
387 #ifdef SLUB_DEBUG_CMPXCHG
388 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
394 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
395 void *freelist_old
, unsigned long counters_old
,
396 void *freelist_new
, unsigned long counters_new
,
399 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
400 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
401 if (s
->flags
& __CMPXCHG_DOUBLE
) {
402 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
403 freelist_old
, counters_old
,
404 freelist_new
, counters_new
))
411 local_irq_save(flags
);
413 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
414 page
->freelist
= freelist_new
;
415 page
->counters
= counters_new
;
417 local_irq_restore(flags
);
421 local_irq_restore(flags
);
425 stat(s
, CMPXCHG_DOUBLE_FAIL
);
427 #ifdef SLUB_DEBUG_CMPXCHG
428 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
434 #ifdef CONFIG_SLUB_DEBUG
436 * Determine a map of object in use on a page.
438 * Node listlock must be held to guarantee that the page does
439 * not vanish from under us.
441 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
444 void *addr
= page_address(page
);
446 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
447 set_bit(slab_index(p
, s
, addr
), map
);
453 #ifdef CONFIG_SLUB_DEBUG_ON
454 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
456 static int slub_debug
;
459 static char *slub_debug_slabs
;
460 static int disable_higher_order_debug
;
465 static void print_section(char *text
, u8
*addr
, unsigned int length
)
467 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
471 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
472 enum track_item alloc
)
477 p
= object
+ s
->offset
+ sizeof(void *);
479 p
= object
+ s
->inuse
;
484 static void set_track(struct kmem_cache
*s
, void *object
,
485 enum track_item alloc
, unsigned long addr
)
487 struct track
*p
= get_track(s
, object
, alloc
);
490 #ifdef CONFIG_STACKTRACE
491 struct stack_trace trace
;
494 trace
.nr_entries
= 0;
495 trace
.max_entries
= TRACK_ADDRS_COUNT
;
496 trace
.entries
= p
->addrs
;
498 save_stack_trace(&trace
);
500 /* See rant in lockdep.c */
501 if (trace
.nr_entries
!= 0 &&
502 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
505 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
509 p
->cpu
= smp_processor_id();
510 p
->pid
= current
->pid
;
513 memset(p
, 0, sizeof(struct track
));
516 static void init_tracking(struct kmem_cache
*s
, void *object
)
518 if (!(s
->flags
& SLAB_STORE_USER
))
521 set_track(s
, object
, TRACK_FREE
, 0UL);
522 set_track(s
, object
, TRACK_ALLOC
, 0UL);
525 static void print_track(const char *s
, struct track
*t
)
530 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
531 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
532 #ifdef CONFIG_STACKTRACE
535 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
537 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
544 static void print_tracking(struct kmem_cache
*s
, void *object
)
546 if (!(s
->flags
& SLAB_STORE_USER
))
549 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
550 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
553 static void print_page_info(struct page
*page
)
555 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
556 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
560 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
566 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
568 printk(KERN_ERR
"========================================"
569 "=====================================\n");
570 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
571 printk(KERN_ERR
"----------------------------------------"
572 "-------------------------------------\n\n");
575 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
581 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
583 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
586 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
588 unsigned int off
; /* Offset of last byte */
589 u8
*addr
= page_address(page
);
591 print_tracking(s
, p
);
593 print_page_info(page
);
595 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
596 p
, p
- addr
, get_freepointer(s
, p
));
599 print_section("Bytes b4 ", p
- 16, 16);
601 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
603 if (s
->flags
& SLAB_RED_ZONE
)
604 print_section("Redzone ", p
+ s
->object_size
,
605 s
->inuse
- s
->object_size
);
608 off
= s
->offset
+ sizeof(void *);
612 if (s
->flags
& SLAB_STORE_USER
)
613 off
+= 2 * sizeof(struct track
);
616 /* Beginning of the filler is the free pointer */
617 print_section("Padding ", p
+ off
, s
->size
- off
);
622 static void object_err(struct kmem_cache
*s
, struct page
*page
,
623 u8
*object
, char *reason
)
625 slab_bug(s
, "%s", reason
);
626 print_trailer(s
, page
, object
);
629 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
635 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
637 slab_bug(s
, "%s", buf
);
638 print_page_info(page
);
642 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
646 if (s
->flags
& __OBJECT_POISON
) {
647 memset(p
, POISON_FREE
, s
->object_size
- 1);
648 p
[s
->object_size
- 1] = POISON_END
;
651 if (s
->flags
& SLAB_RED_ZONE
)
652 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
655 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
656 void *from
, void *to
)
658 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
659 memset(from
, data
, to
- from
);
662 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
663 u8
*object
, char *what
,
664 u8
*start
, unsigned int value
, unsigned int bytes
)
669 fault
= memchr_inv(start
, value
, bytes
);
674 while (end
> fault
&& end
[-1] == value
)
677 slab_bug(s
, "%s overwritten", what
);
678 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
679 fault
, end
- 1, fault
[0], value
);
680 print_trailer(s
, page
, object
);
682 restore_bytes(s
, what
, value
, fault
, end
);
690 * Bytes of the object to be managed.
691 * If the freepointer may overlay the object then the free
692 * pointer is the first word of the object.
694 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
697 * object + s->object_size
698 * Padding to reach word boundary. This is also used for Redzoning.
699 * Padding is extended by another word if Redzoning is enabled and
700 * object_size == inuse.
702 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
703 * 0xcc (RED_ACTIVE) for objects in use.
706 * Meta data starts here.
708 * A. Free pointer (if we cannot overwrite object on free)
709 * B. Tracking data for SLAB_STORE_USER
710 * C. Padding to reach required alignment boundary or at mininum
711 * one word if debugging is on to be able to detect writes
712 * before the word boundary.
714 * Padding is done using 0x5a (POISON_INUSE)
717 * Nothing is used beyond s->size.
719 * If slabcaches are merged then the object_size and inuse boundaries are mostly
720 * ignored. And therefore no slab options that rely on these boundaries
721 * may be used with merged slabcaches.
724 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
726 unsigned long off
= s
->inuse
; /* The end of info */
729 /* Freepointer is placed after the object. */
730 off
+= sizeof(void *);
732 if (s
->flags
& SLAB_STORE_USER
)
733 /* We also have user information there */
734 off
+= 2 * sizeof(struct track
);
739 return check_bytes_and_report(s
, page
, p
, "Object padding",
740 p
+ off
, POISON_INUSE
, s
->size
- off
);
743 /* Check the pad bytes at the end of a slab page */
744 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
752 if (!(s
->flags
& SLAB_POISON
))
755 start
= page_address(page
);
756 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
757 end
= start
+ length
;
758 remainder
= length
% s
->size
;
762 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
765 while (end
> fault
&& end
[-1] == POISON_INUSE
)
768 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
769 print_section("Padding ", end
- remainder
, remainder
);
771 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
775 static int check_object(struct kmem_cache
*s
, struct page
*page
,
776 void *object
, u8 val
)
779 u8
*endobject
= object
+ s
->object_size
;
781 if (s
->flags
& SLAB_RED_ZONE
) {
782 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
783 endobject
, val
, s
->inuse
- s
->object_size
))
786 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
787 check_bytes_and_report(s
, page
, p
, "Alignment padding",
788 endobject
, POISON_INUSE
, s
->inuse
- s
->object_size
);
792 if (s
->flags
& SLAB_POISON
) {
793 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
794 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
795 POISON_FREE
, s
->object_size
- 1) ||
796 !check_bytes_and_report(s
, page
, p
, "Poison",
797 p
+ s
->object_size
- 1, POISON_END
, 1)))
800 * check_pad_bytes cleans up on its own.
802 check_pad_bytes(s
, page
, p
);
805 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
807 * Object and freepointer overlap. Cannot check
808 * freepointer while object is allocated.
812 /* Check free pointer validity */
813 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
814 object_err(s
, page
, p
, "Freepointer corrupt");
816 * No choice but to zap it and thus lose the remainder
817 * of the free objects in this slab. May cause
818 * another error because the object count is now wrong.
820 set_freepointer(s
, p
, NULL
);
826 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
830 VM_BUG_ON(!irqs_disabled());
832 if (!PageSlab(page
)) {
833 slab_err(s
, page
, "Not a valid slab page");
837 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
838 if (page
->objects
> maxobj
) {
839 slab_err(s
, page
, "objects %u > max %u",
840 s
->name
, page
->objects
, maxobj
);
843 if (page
->inuse
> page
->objects
) {
844 slab_err(s
, page
, "inuse %u > max %u",
845 s
->name
, page
->inuse
, page
->objects
);
848 /* Slab_pad_check fixes things up after itself */
849 slab_pad_check(s
, page
);
854 * Determine if a certain object on a page is on the freelist. Must hold the
855 * slab lock to guarantee that the chains are in a consistent state.
857 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
862 unsigned long max_objects
;
865 while (fp
&& nr
<= page
->objects
) {
868 if (!check_valid_pointer(s
, page
, fp
)) {
870 object_err(s
, page
, object
,
871 "Freechain corrupt");
872 set_freepointer(s
, object
, NULL
);
875 slab_err(s
, page
, "Freepointer corrupt");
876 page
->freelist
= NULL
;
877 page
->inuse
= page
->objects
;
878 slab_fix(s
, "Freelist cleared");
884 fp
= get_freepointer(s
, object
);
888 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
889 if (max_objects
> MAX_OBJS_PER_PAGE
)
890 max_objects
= MAX_OBJS_PER_PAGE
;
892 if (page
->objects
!= max_objects
) {
893 slab_err(s
, page
, "Wrong number of objects. Found %d but "
894 "should be %d", page
->objects
, max_objects
);
895 page
->objects
= max_objects
;
896 slab_fix(s
, "Number of objects adjusted.");
898 if (page
->inuse
!= page
->objects
- nr
) {
899 slab_err(s
, page
, "Wrong object count. Counter is %d but "
900 "counted were %d", page
->inuse
, page
->objects
- nr
);
901 page
->inuse
= page
->objects
- nr
;
902 slab_fix(s
, "Object count adjusted.");
904 return search
== NULL
;
907 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
910 if (s
->flags
& SLAB_TRACE
) {
911 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
913 alloc
? "alloc" : "free",
918 print_section("Object ", (void *)object
, s
->object_size
);
925 * Hooks for other subsystems that check memory allocations. In a typical
926 * production configuration these hooks all should produce no code at all.
928 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
930 flags
&= gfp_allowed_mask
;
931 lockdep_trace_alloc(flags
);
932 might_sleep_if(flags
& __GFP_WAIT
);
934 return should_failslab(s
->object_size
, flags
, s
->flags
);
937 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
939 flags
&= gfp_allowed_mask
;
940 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
941 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
944 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
946 kmemleak_free_recursive(x
, s
->flags
);
949 * Trouble is that we may no longer disable interupts in the fast path
950 * So in order to make the debug calls that expect irqs to be
951 * disabled we need to disable interrupts temporarily.
953 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
957 local_irq_save(flags
);
958 kmemcheck_slab_free(s
, x
, s
->object_size
);
959 debug_check_no_locks_freed(x
, s
->object_size
);
960 local_irq_restore(flags
);
963 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
964 debug_check_no_obj_freed(x
, s
->object_size
);
968 * Tracking of fully allocated slabs for debugging purposes.
970 * list_lock must be held.
972 static void add_full(struct kmem_cache
*s
,
973 struct kmem_cache_node
*n
, struct page
*page
)
975 if (!(s
->flags
& SLAB_STORE_USER
))
978 list_add(&page
->lru
, &n
->full
);
982 * list_lock must be held.
984 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
986 if (!(s
->flags
& SLAB_STORE_USER
))
989 list_del(&page
->lru
);
992 /* Tracking of the number of slabs for debugging purposes */
993 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
995 struct kmem_cache_node
*n
= get_node(s
, node
);
997 return atomic_long_read(&n
->nr_slabs
);
1000 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1002 return atomic_long_read(&n
->nr_slabs
);
1005 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1007 struct kmem_cache_node
*n
= get_node(s
, node
);
1010 * May be called early in order to allocate a slab for the
1011 * kmem_cache_node structure. Solve the chicken-egg
1012 * dilemma by deferring the increment of the count during
1013 * bootstrap (see early_kmem_cache_node_alloc).
1016 atomic_long_inc(&n
->nr_slabs
);
1017 atomic_long_add(objects
, &n
->total_objects
);
1020 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1022 struct kmem_cache_node
*n
= get_node(s
, node
);
1024 atomic_long_dec(&n
->nr_slabs
);
1025 atomic_long_sub(objects
, &n
->total_objects
);
1028 /* Object debug checks for alloc/free paths */
1029 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1032 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1035 init_object(s
, object
, SLUB_RED_INACTIVE
);
1036 init_tracking(s
, object
);
1039 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
1040 void *object
, unsigned long addr
)
1042 if (!check_slab(s
, page
))
1045 if (!check_valid_pointer(s
, page
, object
)) {
1046 object_err(s
, page
, object
, "Freelist Pointer check fails");
1050 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1053 /* Success perform special debug activities for allocs */
1054 if (s
->flags
& SLAB_STORE_USER
)
1055 set_track(s
, object
, TRACK_ALLOC
, addr
);
1056 trace(s
, page
, object
, 1);
1057 init_object(s
, object
, SLUB_RED_ACTIVE
);
1061 if (PageSlab(page
)) {
1063 * If this is a slab page then lets do the best we can
1064 * to avoid issues in the future. Marking all objects
1065 * as used avoids touching the remaining objects.
1067 slab_fix(s
, "Marking all objects used");
1068 page
->inuse
= page
->objects
;
1069 page
->freelist
= NULL
;
1074 static noinline
int free_debug_processing(struct kmem_cache
*s
,
1075 struct page
*page
, void *object
, unsigned long addr
)
1077 unsigned long flags
;
1080 local_irq_save(flags
);
1083 if (!check_slab(s
, page
))
1086 if (!check_valid_pointer(s
, page
, object
)) {
1087 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1091 if (on_freelist(s
, page
, object
)) {
1092 object_err(s
, page
, object
, "Object already free");
1096 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1099 if (unlikely(s
!= page
->slab
)) {
1100 if (!PageSlab(page
)) {
1101 slab_err(s
, page
, "Attempt to free object(0x%p) "
1102 "outside of slab", object
);
1103 } else if (!page
->slab
) {
1105 "SLUB <none>: no slab for object 0x%p.\n",
1109 object_err(s
, page
, object
,
1110 "page slab pointer corrupt.");
1114 if (s
->flags
& SLAB_STORE_USER
)
1115 set_track(s
, object
, TRACK_FREE
, addr
);
1116 trace(s
, page
, object
, 0);
1117 init_object(s
, object
, SLUB_RED_INACTIVE
);
1121 local_irq_restore(flags
);
1125 slab_fix(s
, "Object at 0x%p not freed", object
);
1129 static int __init
setup_slub_debug(char *str
)
1131 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1132 if (*str
++ != '=' || !*str
)
1134 * No options specified. Switch on full debugging.
1140 * No options but restriction on slabs. This means full
1141 * debugging for slabs matching a pattern.
1145 if (tolower(*str
) == 'o') {
1147 * Avoid enabling debugging on caches if its minimum order
1148 * would increase as a result.
1150 disable_higher_order_debug
= 1;
1157 * Switch off all debugging measures.
1162 * Determine which debug features should be switched on
1164 for (; *str
&& *str
!= ','; str
++) {
1165 switch (tolower(*str
)) {
1167 slub_debug
|= SLAB_DEBUG_FREE
;
1170 slub_debug
|= SLAB_RED_ZONE
;
1173 slub_debug
|= SLAB_POISON
;
1176 slub_debug
|= SLAB_STORE_USER
;
1179 slub_debug
|= SLAB_TRACE
;
1182 slub_debug
|= SLAB_FAILSLAB
;
1185 printk(KERN_ERR
"slub_debug option '%c' "
1186 "unknown. skipped\n", *str
);
1192 slub_debug_slabs
= str
+ 1;
1197 __setup("slub_debug", setup_slub_debug
);
1199 static unsigned long kmem_cache_flags(unsigned long object_size
,
1200 unsigned long flags
, const char *name
,
1201 void (*ctor
)(void *))
1204 * Enable debugging if selected on the kernel commandline.
1206 if (slub_debug
&& (!slub_debug_slabs
||
1207 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1208 flags
|= slub_debug
;
1213 static inline void setup_object_debug(struct kmem_cache
*s
,
1214 struct page
*page
, void *object
) {}
1216 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1217 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1219 static inline int free_debug_processing(struct kmem_cache
*s
,
1220 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1222 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1224 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1225 void *object
, u8 val
) { return 1; }
1226 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1227 struct page
*page
) {}
1228 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1229 static inline unsigned long kmem_cache_flags(unsigned long object_size
,
1230 unsigned long flags
, const char *name
,
1231 void (*ctor
)(void *))
1235 #define slub_debug 0
1237 #define disable_higher_order_debug 0
1239 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1241 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1243 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1245 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1248 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1251 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1254 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1256 #endif /* CONFIG_SLUB_DEBUG */
1259 * Slab allocation and freeing
1261 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1262 struct kmem_cache_order_objects oo
)
1264 int order
= oo_order(oo
);
1266 flags
|= __GFP_NOTRACK
;
1268 if (node
== NUMA_NO_NODE
)
1269 return alloc_pages(flags
, order
);
1271 return alloc_pages_exact_node(node
, flags
, order
);
1274 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1277 struct kmem_cache_order_objects oo
= s
->oo
;
1280 flags
&= gfp_allowed_mask
;
1282 if (flags
& __GFP_WAIT
)
1285 flags
|= s
->allocflags
;
1288 * Let the initial higher-order allocation fail under memory pressure
1289 * so we fall-back to the minimum order allocation.
1291 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1293 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1294 if (unlikely(!page
)) {
1297 * Allocation may have failed due to fragmentation.
1298 * Try a lower order alloc if possible
1300 page
= alloc_slab_page(flags
, node
, oo
);
1303 stat(s
, ORDER_FALLBACK
);
1306 if (flags
& __GFP_WAIT
)
1307 local_irq_disable();
1312 if (kmemcheck_enabled
1313 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1314 int pages
= 1 << oo_order(oo
);
1316 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1319 * Objects from caches that have a constructor don't get
1320 * cleared when they're allocated, so we need to do it here.
1323 kmemcheck_mark_uninitialized_pages(page
, pages
);
1325 kmemcheck_mark_unallocated_pages(page
, pages
);
1328 page
->objects
= oo_objects(oo
);
1329 mod_zone_page_state(page_zone(page
),
1330 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1331 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1337 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1340 setup_object_debug(s
, page
, object
);
1341 if (unlikely(s
->ctor
))
1345 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1352 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1354 page
= allocate_slab(s
,
1355 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1359 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1361 __SetPageSlab(page
);
1363 start
= page_address(page
);
1365 if (unlikely(s
->flags
& SLAB_POISON
))
1366 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1369 for_each_object(p
, s
, start
, page
->objects
) {
1370 setup_object(s
, page
, last
);
1371 set_freepointer(s
, last
, p
);
1374 setup_object(s
, page
, last
);
1375 set_freepointer(s
, last
, NULL
);
1377 page
->freelist
= start
;
1378 page
->inuse
= page
->objects
;
1384 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1386 int order
= compound_order(page
);
1387 int pages
= 1 << order
;
1389 if (kmem_cache_debug(s
)) {
1392 slab_pad_check(s
, page
);
1393 for_each_object(p
, s
, page_address(page
),
1395 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1398 kmemcheck_free_shadow(page
, compound_order(page
));
1400 mod_zone_page_state(page_zone(page
),
1401 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1402 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1405 __ClearPageSlab(page
);
1406 reset_page_mapcount(page
);
1407 if (current
->reclaim_state
)
1408 current
->reclaim_state
->reclaimed_slab
+= pages
;
1409 __free_pages(page
, order
);
1412 #define need_reserve_slab_rcu \
1413 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1415 static void rcu_free_slab(struct rcu_head
*h
)
1419 if (need_reserve_slab_rcu
)
1420 page
= virt_to_head_page(h
);
1422 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1424 __free_slab(page
->slab
, page
);
1427 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1429 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1430 struct rcu_head
*head
;
1432 if (need_reserve_slab_rcu
) {
1433 int order
= compound_order(page
);
1434 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1436 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1437 head
= page_address(page
) + offset
;
1440 * RCU free overloads the RCU head over the LRU
1442 head
= (void *)&page
->lru
;
1445 call_rcu(head
, rcu_free_slab
);
1447 __free_slab(s
, page
);
1450 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1452 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1457 * Management of partially allocated slabs.
1459 * list_lock must be held.
1461 static inline void add_partial(struct kmem_cache_node
*n
,
1462 struct page
*page
, int tail
)
1465 if (tail
== DEACTIVATE_TO_TAIL
)
1466 list_add_tail(&page
->lru
, &n
->partial
);
1468 list_add(&page
->lru
, &n
->partial
);
1472 * list_lock must be held.
1474 static inline void remove_partial(struct kmem_cache_node
*n
,
1477 list_del(&page
->lru
);
1482 * Remove slab from the partial list, freeze it and
1483 * return the pointer to the freelist.
1485 * Returns a list of objects or NULL if it fails.
1487 * Must hold list_lock since we modify the partial list.
1489 static inline void *acquire_slab(struct kmem_cache
*s
,
1490 struct kmem_cache_node
*n
, struct page
*page
,
1494 unsigned long counters
;
1498 * Zap the freelist and set the frozen bit.
1499 * The old freelist is the list of objects for the
1500 * per cpu allocation list.
1502 freelist
= page
->freelist
;
1503 counters
= page
->counters
;
1504 new.counters
= counters
;
1506 new.inuse
= page
->objects
;
1507 new.freelist
= NULL
;
1509 new.freelist
= freelist
;
1512 VM_BUG_ON(new.frozen
);
1515 if (!__cmpxchg_double_slab(s
, page
,
1517 new.freelist
, new.counters
,
1521 remove_partial(n
, page
);
1526 static int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1529 * Try to allocate a partial slab from a specific node.
1531 static void *get_partial_node(struct kmem_cache
*s
,
1532 struct kmem_cache_node
*n
, struct kmem_cache_cpu
*c
)
1534 struct page
*page
, *page2
;
1535 void *object
= NULL
;
1538 * Racy check. If we mistakenly see no partial slabs then we
1539 * just allocate an empty slab. If we mistakenly try to get a
1540 * partial slab and there is none available then get_partials()
1543 if (!n
|| !n
->nr_partial
)
1546 spin_lock(&n
->list_lock
);
1547 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1548 void *t
= acquire_slab(s
, n
, page
, object
== NULL
);
1556 stat(s
, ALLOC_FROM_PARTIAL
);
1558 available
= page
->objects
- page
->inuse
;
1560 available
= put_cpu_partial(s
, page
, 0);
1561 stat(s
, CPU_PARTIAL_NODE
);
1563 if (kmem_cache_debug(s
) || available
> s
->cpu_partial
/ 2)
1567 spin_unlock(&n
->list_lock
);
1572 * Get a page from somewhere. Search in increasing NUMA distances.
1574 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1575 struct kmem_cache_cpu
*c
)
1578 struct zonelist
*zonelist
;
1581 enum zone_type high_zoneidx
= gfp_zone(flags
);
1583 unsigned int cpuset_mems_cookie
;
1586 * The defrag ratio allows a configuration of the tradeoffs between
1587 * inter node defragmentation and node local allocations. A lower
1588 * defrag_ratio increases the tendency to do local allocations
1589 * instead of attempting to obtain partial slabs from other nodes.
1591 * If the defrag_ratio is set to 0 then kmalloc() always
1592 * returns node local objects. If the ratio is higher then kmalloc()
1593 * may return off node objects because partial slabs are obtained
1594 * from other nodes and filled up.
1596 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1597 * defrag_ratio = 1000) then every (well almost) allocation will
1598 * first attempt to defrag slab caches on other nodes. This means
1599 * scanning over all nodes to look for partial slabs which may be
1600 * expensive if we do it every time we are trying to find a slab
1601 * with available objects.
1603 if (!s
->remote_node_defrag_ratio
||
1604 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1608 cpuset_mems_cookie
= get_mems_allowed();
1609 zonelist
= node_zonelist(slab_node(), flags
);
1610 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1611 struct kmem_cache_node
*n
;
1613 n
= get_node(s
, zone_to_nid(zone
));
1615 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1616 n
->nr_partial
> s
->min_partial
) {
1617 object
= get_partial_node(s
, n
, c
);
1620 * Return the object even if
1621 * put_mems_allowed indicated that
1622 * the cpuset mems_allowed was
1623 * updated in parallel. It's a
1624 * harmless race between the alloc
1625 * and the cpuset update.
1627 put_mems_allowed(cpuset_mems_cookie
);
1632 } while (!put_mems_allowed(cpuset_mems_cookie
));
1638 * Get a partial page, lock it and return it.
1640 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1641 struct kmem_cache_cpu
*c
)
1644 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1646 object
= get_partial_node(s
, get_node(s
, searchnode
), c
);
1647 if (object
|| node
!= NUMA_NO_NODE
)
1650 return get_any_partial(s
, flags
, c
);
1653 #ifdef CONFIG_PREEMPT
1655 * Calculate the next globally unique transaction for disambiguiation
1656 * during cmpxchg. The transactions start with the cpu number and are then
1657 * incremented by CONFIG_NR_CPUS.
1659 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1662 * No preemption supported therefore also no need to check for
1668 static inline unsigned long next_tid(unsigned long tid
)
1670 return tid
+ TID_STEP
;
1673 static inline unsigned int tid_to_cpu(unsigned long tid
)
1675 return tid
% TID_STEP
;
1678 static inline unsigned long tid_to_event(unsigned long tid
)
1680 return tid
/ TID_STEP
;
1683 static inline unsigned int init_tid(int cpu
)
1688 static inline void note_cmpxchg_failure(const char *n
,
1689 const struct kmem_cache
*s
, unsigned long tid
)
1691 #ifdef SLUB_DEBUG_CMPXCHG
1692 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1694 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1696 #ifdef CONFIG_PREEMPT
1697 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1698 printk("due to cpu change %d -> %d\n",
1699 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1702 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1703 printk("due to cpu running other code. Event %ld->%ld\n",
1704 tid_to_event(tid
), tid_to_event(actual_tid
));
1706 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1707 actual_tid
, tid
, next_tid(tid
));
1709 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1712 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1716 for_each_possible_cpu(cpu
)
1717 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1721 * Remove the cpu slab
1723 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
, void *freelist
)
1725 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1726 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1728 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1730 int tail
= DEACTIVATE_TO_HEAD
;
1734 if (page
->freelist
) {
1735 stat(s
, DEACTIVATE_REMOTE_FREES
);
1736 tail
= DEACTIVATE_TO_TAIL
;
1740 * Stage one: Free all available per cpu objects back
1741 * to the page freelist while it is still frozen. Leave the
1744 * There is no need to take the list->lock because the page
1747 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1749 unsigned long counters
;
1752 prior
= page
->freelist
;
1753 counters
= page
->counters
;
1754 set_freepointer(s
, freelist
, prior
);
1755 new.counters
= counters
;
1757 VM_BUG_ON(!new.frozen
);
1759 } while (!__cmpxchg_double_slab(s
, page
,
1761 freelist
, new.counters
,
1762 "drain percpu freelist"));
1764 freelist
= nextfree
;
1768 * Stage two: Ensure that the page is unfrozen while the
1769 * list presence reflects the actual number of objects
1772 * We setup the list membership and then perform a cmpxchg
1773 * with the count. If there is a mismatch then the page
1774 * is not unfrozen but the page is on the wrong list.
1776 * Then we restart the process which may have to remove
1777 * the page from the list that we just put it on again
1778 * because the number of objects in the slab may have
1783 old
.freelist
= page
->freelist
;
1784 old
.counters
= page
->counters
;
1785 VM_BUG_ON(!old
.frozen
);
1787 /* Determine target state of the slab */
1788 new.counters
= old
.counters
;
1791 set_freepointer(s
, freelist
, old
.freelist
);
1792 new.freelist
= freelist
;
1794 new.freelist
= old
.freelist
;
1798 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1800 else if (new.freelist
) {
1805 * Taking the spinlock removes the possiblity
1806 * that acquire_slab() will see a slab page that
1809 spin_lock(&n
->list_lock
);
1813 if (kmem_cache_debug(s
) && !lock
) {
1816 * This also ensures that the scanning of full
1817 * slabs from diagnostic functions will not see
1820 spin_lock(&n
->list_lock
);
1828 remove_partial(n
, page
);
1830 else if (l
== M_FULL
)
1832 remove_full(s
, page
);
1834 if (m
== M_PARTIAL
) {
1836 add_partial(n
, page
, tail
);
1839 } else if (m
== M_FULL
) {
1841 stat(s
, DEACTIVATE_FULL
);
1842 add_full(s
, n
, page
);
1848 if (!__cmpxchg_double_slab(s
, page
,
1849 old
.freelist
, old
.counters
,
1850 new.freelist
, new.counters
,
1855 spin_unlock(&n
->list_lock
);
1858 stat(s
, DEACTIVATE_EMPTY
);
1859 discard_slab(s
, page
);
1865 * Unfreeze all the cpu partial slabs.
1867 * This function must be called with interrupt disabled.
1869 static void unfreeze_partials(struct kmem_cache
*s
)
1871 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1872 struct kmem_cache_cpu
*c
= this_cpu_ptr(s
->cpu_slab
);
1873 struct page
*page
, *discard_page
= NULL
;
1875 while ((page
= c
->partial
)) {
1879 c
->partial
= page
->next
;
1881 n2
= get_node(s
, page_to_nid(page
));
1884 spin_unlock(&n
->list_lock
);
1887 spin_lock(&n
->list_lock
);
1892 old
.freelist
= page
->freelist
;
1893 old
.counters
= page
->counters
;
1894 VM_BUG_ON(!old
.frozen
);
1896 new.counters
= old
.counters
;
1897 new.freelist
= old
.freelist
;
1901 } while (!__cmpxchg_double_slab(s
, page
,
1902 old
.freelist
, old
.counters
,
1903 new.freelist
, new.counters
,
1904 "unfreezing slab"));
1906 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
)) {
1907 page
->next
= discard_page
;
1908 discard_page
= page
;
1910 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1911 stat(s
, FREE_ADD_PARTIAL
);
1916 spin_unlock(&n
->list_lock
);
1918 while (discard_page
) {
1919 page
= discard_page
;
1920 discard_page
= discard_page
->next
;
1922 stat(s
, DEACTIVATE_EMPTY
);
1923 discard_slab(s
, page
);
1929 * Put a page that was just frozen (in __slab_free) into a partial page
1930 * slot if available. This is done without interrupts disabled and without
1931 * preemption disabled. The cmpxchg is racy and may put the partial page
1932 * onto a random cpus partial slot.
1934 * If we did not find a slot then simply move all the partials to the
1935 * per node partial list.
1937 int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1939 struct page
*oldpage
;
1946 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1949 pobjects
= oldpage
->pobjects
;
1950 pages
= oldpage
->pages
;
1951 if (drain
&& pobjects
> s
->cpu_partial
) {
1952 unsigned long flags
;
1954 * partial array is full. Move the existing
1955 * set to the per node partial list.
1957 local_irq_save(flags
);
1958 unfreeze_partials(s
);
1959 local_irq_restore(flags
);
1962 stat(s
, CPU_PARTIAL_DRAIN
);
1967 pobjects
+= page
->objects
- page
->inuse
;
1969 page
->pages
= pages
;
1970 page
->pobjects
= pobjects
;
1971 page
->next
= oldpage
;
1973 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
) != oldpage
);
1977 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1979 stat(s
, CPUSLAB_FLUSH
);
1980 deactivate_slab(s
, c
->page
, c
->freelist
);
1982 c
->tid
= next_tid(c
->tid
);
1990 * Called from IPI handler with interrupts disabled.
1992 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1994 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2000 unfreeze_partials(s
);
2004 static void flush_cpu_slab(void *d
)
2006 struct kmem_cache
*s
= d
;
2008 __flush_cpu_slab(s
, smp_processor_id());
2011 static bool has_cpu_slab(int cpu
, void *info
)
2013 struct kmem_cache
*s
= info
;
2014 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2016 return c
->page
|| c
->partial
;
2019 static void flush_all(struct kmem_cache
*s
)
2021 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2025 * Check if the objects in a per cpu structure fit numa
2026 * locality expectations.
2028 static inline int node_match(struct page
*page
, int node
)
2031 if (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
)
2037 static int count_free(struct page
*page
)
2039 return page
->objects
- page
->inuse
;
2042 static unsigned long count_partial(struct kmem_cache_node
*n
,
2043 int (*get_count
)(struct page
*))
2045 unsigned long flags
;
2046 unsigned long x
= 0;
2049 spin_lock_irqsave(&n
->list_lock
, flags
);
2050 list_for_each_entry(page
, &n
->partial
, lru
)
2051 x
+= get_count(page
);
2052 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2056 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2058 #ifdef CONFIG_SLUB_DEBUG
2059 return atomic_long_read(&n
->total_objects
);
2065 static noinline
void
2066 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2071 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2073 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2074 "default order: %d, min order: %d\n", s
->name
, s
->object_size
,
2075 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2077 if (oo_order(s
->min
) > get_order(s
->object_size
))
2078 printk(KERN_WARNING
" %s debugging increased min order, use "
2079 "slub_debug=O to disable.\n", s
->name
);
2081 for_each_online_node(node
) {
2082 struct kmem_cache_node
*n
= get_node(s
, node
);
2083 unsigned long nr_slabs
;
2084 unsigned long nr_objs
;
2085 unsigned long nr_free
;
2090 nr_free
= count_partial(n
, count_free
);
2091 nr_slabs
= node_nr_slabs(n
);
2092 nr_objs
= node_nr_objs(n
);
2095 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2096 node
, nr_slabs
, nr_objs
, nr_free
);
2100 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2101 int node
, struct kmem_cache_cpu
**pc
)
2104 struct kmem_cache_cpu
*c
= *pc
;
2107 freelist
= get_partial(s
, flags
, node
, c
);
2112 page
= new_slab(s
, flags
, node
);
2114 c
= __this_cpu_ptr(s
->cpu_slab
);
2119 * No other reference to the page yet so we can
2120 * muck around with it freely without cmpxchg
2122 freelist
= page
->freelist
;
2123 page
->freelist
= NULL
;
2125 stat(s
, ALLOC_SLAB
);
2135 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2136 * or deactivate the page.
2138 * The page is still frozen if the return value is not NULL.
2140 * If this function returns NULL then the page has been unfrozen.
2142 * This function must be called with interrupt disabled.
2144 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2147 unsigned long counters
;
2151 freelist
= page
->freelist
;
2152 counters
= page
->counters
;
2154 new.counters
= counters
;
2155 VM_BUG_ON(!new.frozen
);
2157 new.inuse
= page
->objects
;
2158 new.frozen
= freelist
!= NULL
;
2160 } while (!__cmpxchg_double_slab(s
, page
,
2169 * Slow path. The lockless freelist is empty or we need to perform
2172 * Processing is still very fast if new objects have been freed to the
2173 * regular freelist. In that case we simply take over the regular freelist
2174 * as the lockless freelist and zap the regular freelist.
2176 * If that is not working then we fall back to the partial lists. We take the
2177 * first element of the freelist as the object to allocate now and move the
2178 * rest of the freelist to the lockless freelist.
2180 * And if we were unable to get a new slab from the partial slab lists then
2181 * we need to allocate a new slab. This is the slowest path since it involves
2182 * a call to the page allocator and the setup of a new slab.
2184 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2185 unsigned long addr
, struct kmem_cache_cpu
*c
)
2189 unsigned long flags
;
2191 local_irq_save(flags
);
2192 #ifdef CONFIG_PREEMPT
2194 * We may have been preempted and rescheduled on a different
2195 * cpu before disabling interrupts. Need to reload cpu area
2198 c
= this_cpu_ptr(s
->cpu_slab
);
2206 if (unlikely(!node_match(page
, node
))) {
2207 stat(s
, ALLOC_NODE_MISMATCH
);
2208 deactivate_slab(s
, page
, c
->freelist
);
2214 /* must check again c->freelist in case of cpu migration or IRQ */
2215 freelist
= c
->freelist
;
2219 stat(s
, ALLOC_SLOWPATH
);
2221 freelist
= get_freelist(s
, page
);
2225 stat(s
, DEACTIVATE_BYPASS
);
2229 stat(s
, ALLOC_REFILL
);
2233 * freelist is pointing to the list of objects to be used.
2234 * page is pointing to the page from which the objects are obtained.
2235 * That page must be frozen for per cpu allocations to work.
2237 VM_BUG_ON(!c
->page
->frozen
);
2238 c
->freelist
= get_freepointer(s
, freelist
);
2239 c
->tid
= next_tid(c
->tid
);
2240 local_irq_restore(flags
);
2246 page
= c
->page
= c
->partial
;
2247 c
->partial
= page
->next
;
2248 stat(s
, CPU_PARTIAL_ALLOC
);
2253 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2255 if (unlikely(!freelist
)) {
2256 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2257 slab_out_of_memory(s
, gfpflags
, node
);
2259 local_irq_restore(flags
);
2264 if (likely(!kmem_cache_debug(s
)))
2267 /* Only entered in the debug case */
2268 if (!alloc_debug_processing(s
, page
, freelist
, addr
))
2269 goto new_slab
; /* Slab failed checks. Next slab needed */
2271 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2274 local_irq_restore(flags
);
2279 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2280 * have the fastpath folded into their functions. So no function call
2281 * overhead for requests that can be satisfied on the fastpath.
2283 * The fastpath works by first checking if the lockless freelist can be used.
2284 * If not then __slab_alloc is called for slow processing.
2286 * Otherwise we can simply pick the next object from the lockless free list.
2288 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2289 gfp_t gfpflags
, int node
, unsigned long addr
)
2292 struct kmem_cache_cpu
*c
;
2296 if (slab_pre_alloc_hook(s
, gfpflags
))
2302 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2303 * enabled. We may switch back and forth between cpus while
2304 * reading from one cpu area. That does not matter as long
2305 * as we end up on the original cpu again when doing the cmpxchg.
2307 c
= __this_cpu_ptr(s
->cpu_slab
);
2310 * The transaction ids are globally unique per cpu and per operation on
2311 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2312 * occurs on the right processor and that there was no operation on the
2313 * linked list in between.
2318 object
= c
->freelist
;
2320 if (unlikely(!object
|| !node_match(page
, node
)))
2322 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2325 void *next_object
= get_freepointer_safe(s
, object
);
2328 * The cmpxchg will only match if there was no additional
2329 * operation and if we are on the right processor.
2331 * The cmpxchg does the following atomically (without lock semantics!)
2332 * 1. Relocate first pointer to the current per cpu area.
2333 * 2. Verify that tid and freelist have not been changed
2334 * 3. If they were not changed replace tid and freelist
2336 * Since this is without lock semantics the protection is only against
2337 * code executing on this cpu *not* from access by other cpus.
2339 if (unlikely(!this_cpu_cmpxchg_double(
2340 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2342 next_object
, next_tid(tid
)))) {
2344 note_cmpxchg_failure("slab_alloc", s
, tid
);
2347 prefetch_freepointer(s
, next_object
);
2348 stat(s
, ALLOC_FASTPATH
);
2351 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2352 memset(object
, 0, s
->object_size
);
2354 slab_post_alloc_hook(s
, gfpflags
, object
);
2359 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2361 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2363 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
, s
->size
, gfpflags
);
2367 EXPORT_SYMBOL(kmem_cache_alloc
);
2369 #ifdef CONFIG_TRACING
2370 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2372 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2373 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2376 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2378 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2380 void *ret
= kmalloc_order(size
, flags
, order
);
2381 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2384 EXPORT_SYMBOL(kmalloc_order_trace
);
2388 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2390 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2392 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2393 s
->object_size
, s
->size
, gfpflags
, node
);
2397 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2399 #ifdef CONFIG_TRACING
2400 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2402 int node
, size_t size
)
2404 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2406 trace_kmalloc_node(_RET_IP_
, ret
,
2407 size
, s
->size
, gfpflags
, node
);
2410 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2415 * Slow patch handling. This may still be called frequently since objects
2416 * have a longer lifetime than the cpu slabs in most processing loads.
2418 * So we still attempt to reduce cache line usage. Just take the slab
2419 * lock and free the item. If there is no additional partial page
2420 * handling required then we can return immediately.
2422 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2423 void *x
, unsigned long addr
)
2426 void **object
= (void *)x
;
2430 unsigned long counters
;
2431 struct kmem_cache_node
*n
= NULL
;
2432 unsigned long uninitialized_var(flags
);
2434 stat(s
, FREE_SLOWPATH
);
2436 if (kmem_cache_debug(s
) && !free_debug_processing(s
, page
, x
, addr
))
2440 prior
= page
->freelist
;
2441 counters
= page
->counters
;
2442 set_freepointer(s
, object
, prior
);
2443 new.counters
= counters
;
2444 was_frozen
= new.frozen
;
2446 if ((!new.inuse
|| !prior
) && !was_frozen
&& !n
) {
2448 if (!kmem_cache_debug(s
) && !prior
)
2451 * Slab was on no list before and will be partially empty
2452 * We can defer the list move and instead freeze it.
2456 else { /* Needs to be taken off a list */
2458 n
= get_node(s
, page_to_nid(page
));
2460 * Speculatively acquire the list_lock.
2461 * If the cmpxchg does not succeed then we may
2462 * drop the list_lock without any processing.
2464 * Otherwise the list_lock will synchronize with
2465 * other processors updating the list of slabs.
2467 spin_lock_irqsave(&n
->list_lock
, flags
);
2473 } while (!cmpxchg_double_slab(s
, page
,
2475 object
, new.counters
,
2481 * If we just froze the page then put it onto the
2482 * per cpu partial list.
2484 if (new.frozen
&& !was_frozen
) {
2485 put_cpu_partial(s
, page
, 1);
2486 stat(s
, CPU_PARTIAL_FREE
);
2489 * The list lock was not taken therefore no list
2490 * activity can be necessary.
2493 stat(s
, FREE_FROZEN
);
2498 * was_frozen may have been set after we acquired the list_lock in
2499 * an earlier loop. So we need to check it here again.
2502 stat(s
, FREE_FROZEN
);
2504 if (unlikely(!inuse
&& n
->nr_partial
> s
->min_partial
))
2508 * Objects left in the slab. If it was not on the partial list before
2511 if (unlikely(!prior
)) {
2512 remove_full(s
, page
);
2513 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2514 stat(s
, FREE_ADD_PARTIAL
);
2517 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2523 * Slab on the partial list.
2525 remove_partial(n
, page
);
2526 stat(s
, FREE_REMOVE_PARTIAL
);
2528 /* Slab must be on the full list */
2529 remove_full(s
, page
);
2531 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2533 discard_slab(s
, page
);
2537 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2538 * can perform fastpath freeing without additional function calls.
2540 * The fastpath is only possible if we are freeing to the current cpu slab
2541 * of this processor. This typically the case if we have just allocated
2544 * If fastpath is not possible then fall back to __slab_free where we deal
2545 * with all sorts of special processing.
2547 static __always_inline
void slab_free(struct kmem_cache
*s
,
2548 struct page
*page
, void *x
, unsigned long addr
)
2550 void **object
= (void *)x
;
2551 struct kmem_cache_cpu
*c
;
2554 slab_free_hook(s
, x
);
2558 * Determine the currently cpus per cpu slab.
2559 * The cpu may change afterward. However that does not matter since
2560 * data is retrieved via this pointer. If we are on the same cpu
2561 * during the cmpxchg then the free will succedd.
2563 c
= __this_cpu_ptr(s
->cpu_slab
);
2568 if (likely(page
== c
->page
)) {
2569 set_freepointer(s
, object
, c
->freelist
);
2571 if (unlikely(!this_cpu_cmpxchg_double(
2572 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2574 object
, next_tid(tid
)))) {
2576 note_cmpxchg_failure("slab_free", s
, tid
);
2579 stat(s
, FREE_FASTPATH
);
2581 __slab_free(s
, page
, x
, addr
);
2585 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2589 page
= virt_to_head_page(x
);
2591 slab_free(s
, page
, x
, _RET_IP_
);
2593 trace_kmem_cache_free(_RET_IP_
, x
);
2595 EXPORT_SYMBOL(kmem_cache_free
);
2598 * Object placement in a slab is made very easy because we always start at
2599 * offset 0. If we tune the size of the object to the alignment then we can
2600 * get the required alignment by putting one properly sized object after
2603 * Notice that the allocation order determines the sizes of the per cpu
2604 * caches. Each processor has always one slab available for allocations.
2605 * Increasing the allocation order reduces the number of times that slabs
2606 * must be moved on and off the partial lists and is therefore a factor in
2611 * Mininum / Maximum order of slab pages. This influences locking overhead
2612 * and slab fragmentation. A higher order reduces the number of partial slabs
2613 * and increases the number of allocations possible without having to
2614 * take the list_lock.
2616 static int slub_min_order
;
2617 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2618 static int slub_min_objects
;
2621 * Merge control. If this is set then no merging of slab caches will occur.
2622 * (Could be removed. This was introduced to pacify the merge skeptics.)
2624 static int slub_nomerge
;
2627 * Calculate the order of allocation given an slab object size.
2629 * The order of allocation has significant impact on performance and other
2630 * system components. Generally order 0 allocations should be preferred since
2631 * order 0 does not cause fragmentation in the page allocator. Larger objects
2632 * be problematic to put into order 0 slabs because there may be too much
2633 * unused space left. We go to a higher order if more than 1/16th of the slab
2636 * In order to reach satisfactory performance we must ensure that a minimum
2637 * number of objects is in one slab. Otherwise we may generate too much
2638 * activity on the partial lists which requires taking the list_lock. This is
2639 * less a concern for large slabs though which are rarely used.
2641 * slub_max_order specifies the order where we begin to stop considering the
2642 * number of objects in a slab as critical. If we reach slub_max_order then
2643 * we try to keep the page order as low as possible. So we accept more waste
2644 * of space in favor of a small page order.
2646 * Higher order allocations also allow the placement of more objects in a
2647 * slab and thereby reduce object handling overhead. If the user has
2648 * requested a higher mininum order then we start with that one instead of
2649 * the smallest order which will fit the object.
2651 static inline int slab_order(int size
, int min_objects
,
2652 int max_order
, int fract_leftover
, int reserved
)
2656 int min_order
= slub_min_order
;
2658 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2659 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2661 for (order
= max(min_order
,
2662 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2663 order
<= max_order
; order
++) {
2665 unsigned long slab_size
= PAGE_SIZE
<< order
;
2667 if (slab_size
< min_objects
* size
+ reserved
)
2670 rem
= (slab_size
- reserved
) % size
;
2672 if (rem
<= slab_size
/ fract_leftover
)
2680 static inline int calculate_order(int size
, int reserved
)
2688 * Attempt to find best configuration for a slab. This
2689 * works by first attempting to generate a layout with
2690 * the best configuration and backing off gradually.
2692 * First we reduce the acceptable waste in a slab. Then
2693 * we reduce the minimum objects required in a slab.
2695 min_objects
= slub_min_objects
;
2697 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2698 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2699 min_objects
= min(min_objects
, max_objects
);
2701 while (min_objects
> 1) {
2703 while (fraction
>= 4) {
2704 order
= slab_order(size
, min_objects
,
2705 slub_max_order
, fraction
, reserved
);
2706 if (order
<= slub_max_order
)
2714 * We were unable to place multiple objects in a slab. Now
2715 * lets see if we can place a single object there.
2717 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2718 if (order
<= slub_max_order
)
2722 * Doh this slab cannot be placed using slub_max_order.
2724 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2725 if (order
< MAX_ORDER
)
2731 * Figure out what the alignment of the objects will be.
2733 static unsigned long calculate_alignment(unsigned long flags
,
2734 unsigned long align
, unsigned long size
)
2737 * If the user wants hardware cache aligned objects then follow that
2738 * suggestion if the object is sufficiently large.
2740 * The hardware cache alignment cannot override the specified
2741 * alignment though. If that is greater then use it.
2743 if (flags
& SLAB_HWCACHE_ALIGN
) {
2744 unsigned long ralign
= cache_line_size();
2745 while (size
<= ralign
/ 2)
2747 align
= max(align
, ralign
);
2750 if (align
< ARCH_SLAB_MINALIGN
)
2751 align
= ARCH_SLAB_MINALIGN
;
2753 return ALIGN(align
, sizeof(void *));
2757 init_kmem_cache_node(struct kmem_cache_node
*n
)
2760 spin_lock_init(&n
->list_lock
);
2761 INIT_LIST_HEAD(&n
->partial
);
2762 #ifdef CONFIG_SLUB_DEBUG
2763 atomic_long_set(&n
->nr_slabs
, 0);
2764 atomic_long_set(&n
->total_objects
, 0);
2765 INIT_LIST_HEAD(&n
->full
);
2769 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2771 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2772 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2775 * Must align to double word boundary for the double cmpxchg
2776 * instructions to work; see __pcpu_double_call_return_bool().
2778 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2779 2 * sizeof(void *));
2784 init_kmem_cache_cpus(s
);
2789 static struct kmem_cache
*kmem_cache_node
;
2792 * No kmalloc_node yet so do it by hand. We know that this is the first
2793 * slab on the node for this slabcache. There are no concurrent accesses
2796 * Note that this function only works on the kmalloc_node_cache
2797 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2798 * memory on a fresh node that has no slab structures yet.
2800 static void early_kmem_cache_node_alloc(int node
)
2803 struct kmem_cache_node
*n
;
2805 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2807 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2810 if (page_to_nid(page
) != node
) {
2811 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2813 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2814 "in order to be able to continue\n");
2819 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2822 kmem_cache_node
->node
[node
] = n
;
2823 #ifdef CONFIG_SLUB_DEBUG
2824 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2825 init_tracking(kmem_cache_node
, n
);
2827 init_kmem_cache_node(n
);
2828 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2830 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2833 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2837 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2838 struct kmem_cache_node
*n
= s
->node
[node
];
2841 kmem_cache_free(kmem_cache_node
, n
);
2843 s
->node
[node
] = NULL
;
2847 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2851 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2852 struct kmem_cache_node
*n
;
2854 if (slab_state
== DOWN
) {
2855 early_kmem_cache_node_alloc(node
);
2858 n
= kmem_cache_alloc_node(kmem_cache_node
,
2862 free_kmem_cache_nodes(s
);
2867 init_kmem_cache_node(n
);
2872 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2874 if (min
< MIN_PARTIAL
)
2876 else if (min
> MAX_PARTIAL
)
2878 s
->min_partial
= min
;
2882 * calculate_sizes() determines the order and the distribution of data within
2885 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2887 unsigned long flags
= s
->flags
;
2888 unsigned long size
= s
->object_size
;
2889 unsigned long align
= s
->align
;
2893 * Round up object size to the next word boundary. We can only
2894 * place the free pointer at word boundaries and this determines
2895 * the possible location of the free pointer.
2897 size
= ALIGN(size
, sizeof(void *));
2899 #ifdef CONFIG_SLUB_DEBUG
2901 * Determine if we can poison the object itself. If the user of
2902 * the slab may touch the object after free or before allocation
2903 * then we should never poison the object itself.
2905 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2907 s
->flags
|= __OBJECT_POISON
;
2909 s
->flags
&= ~__OBJECT_POISON
;
2913 * If we are Redzoning then check if there is some space between the
2914 * end of the object and the free pointer. If not then add an
2915 * additional word to have some bytes to store Redzone information.
2917 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
2918 size
+= sizeof(void *);
2922 * With that we have determined the number of bytes in actual use
2923 * by the object. This is the potential offset to the free pointer.
2927 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2930 * Relocate free pointer after the object if it is not
2931 * permitted to overwrite the first word of the object on
2934 * This is the case if we do RCU, have a constructor or
2935 * destructor or are poisoning the objects.
2938 size
+= sizeof(void *);
2941 #ifdef CONFIG_SLUB_DEBUG
2942 if (flags
& SLAB_STORE_USER
)
2944 * Need to store information about allocs and frees after
2947 size
+= 2 * sizeof(struct track
);
2949 if (flags
& SLAB_RED_ZONE
)
2951 * Add some empty padding so that we can catch
2952 * overwrites from earlier objects rather than let
2953 * tracking information or the free pointer be
2954 * corrupted if a user writes before the start
2957 size
+= sizeof(void *);
2961 * Determine the alignment based on various parameters that the
2962 * user specified and the dynamic determination of cache line size
2965 align
= calculate_alignment(flags
, align
, s
->object_size
);
2969 * SLUB stores one object immediately after another beginning from
2970 * offset 0. In order to align the objects we have to simply size
2971 * each object to conform to the alignment.
2973 size
= ALIGN(size
, align
);
2975 if (forced_order
>= 0)
2976 order
= forced_order
;
2978 order
= calculate_order(size
, s
->reserved
);
2985 s
->allocflags
|= __GFP_COMP
;
2987 if (s
->flags
& SLAB_CACHE_DMA
)
2988 s
->allocflags
|= SLUB_DMA
;
2990 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2991 s
->allocflags
|= __GFP_RECLAIMABLE
;
2994 * Determine the number of objects per slab
2996 s
->oo
= oo_make(order
, size
, s
->reserved
);
2997 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
2998 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3001 return !!oo_objects(s
->oo
);
3005 static int kmem_cache_open(struct kmem_cache
*s
,
3006 const char *name
, size_t size
,
3007 size_t align
, unsigned long flags
,
3008 void (*ctor
)(void *))
3010 memset(s
, 0, kmem_size
);
3013 s
->object_size
= size
;
3015 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
3018 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3019 s
->reserved
= sizeof(struct rcu_head
);
3021 if (!calculate_sizes(s
, -1))
3023 if (disable_higher_order_debug
) {
3025 * Disable debugging flags that store metadata if the min slab
3028 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3029 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3031 if (!calculate_sizes(s
, -1))
3036 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3037 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3038 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3039 /* Enable fast mode */
3040 s
->flags
|= __CMPXCHG_DOUBLE
;
3044 * The larger the object size is, the more pages we want on the partial
3045 * list to avoid pounding the page allocator excessively.
3047 set_min_partial(s
, ilog2(s
->size
) / 2);
3050 * cpu_partial determined the maximum number of objects kept in the
3051 * per cpu partial lists of a processor.
3053 * Per cpu partial lists mainly contain slabs that just have one
3054 * object freed. If they are used for allocation then they can be
3055 * filled up again with minimal effort. The slab will never hit the
3056 * per node partial lists and therefore no locking will be required.
3058 * This setting also determines
3060 * A) The number of objects from per cpu partial slabs dumped to the
3061 * per node list when we reach the limit.
3062 * B) The number of objects in cpu partial slabs to extract from the
3063 * per node list when we run out of per cpu objects. We only fetch 50%
3064 * to keep some capacity around for frees.
3066 if (kmem_cache_debug(s
))
3068 else if (s
->size
>= PAGE_SIZE
)
3070 else if (s
->size
>= 1024)
3072 else if (s
->size
>= 256)
3073 s
->cpu_partial
= 13;
3075 s
->cpu_partial
= 30;
3079 s
->remote_node_defrag_ratio
= 1000;
3081 if (!init_kmem_cache_nodes(s
))
3084 if (alloc_kmem_cache_cpus(s
))
3087 free_kmem_cache_nodes(s
);
3089 if (flags
& SLAB_PANIC
)
3090 panic("Cannot create slab %s size=%lu realsize=%u "
3091 "order=%u offset=%u flags=%lx\n",
3092 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
3098 * Determine the size of a slab object
3100 unsigned int kmem_cache_size(struct kmem_cache
*s
)
3102 return s
->object_size
;
3104 EXPORT_SYMBOL(kmem_cache_size
);
3106 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3109 #ifdef CONFIG_SLUB_DEBUG
3110 void *addr
= page_address(page
);
3112 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3113 sizeof(long), GFP_ATOMIC
);
3116 slab_err(s
, page
, "%s", text
);
3119 get_map(s
, page
, map
);
3120 for_each_object(p
, s
, addr
, page
->objects
) {
3122 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3123 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3125 print_tracking(s
, p
);
3134 * Attempt to free all partial slabs on a node.
3135 * This is called from kmem_cache_close(). We must be the last thread
3136 * using the cache and therefore we do not need to lock anymore.
3138 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3140 struct page
*page
, *h
;
3142 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3144 remove_partial(n
, page
);
3145 discard_slab(s
, page
);
3147 list_slab_objects(s
, page
,
3148 "Objects remaining on kmem_cache_close()");
3154 * Release all resources used by a slab cache.
3156 static inline int kmem_cache_close(struct kmem_cache
*s
)
3161 free_percpu(s
->cpu_slab
);
3162 /* Attempt to free all objects */
3163 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3164 struct kmem_cache_node
*n
= get_node(s
, node
);
3167 if (n
->nr_partial
|| slabs_node(s
, node
))
3170 free_kmem_cache_nodes(s
);
3175 * Close a cache and release the kmem_cache structure
3176 * (must be used for caches created using kmem_cache_create)
3178 void kmem_cache_destroy(struct kmem_cache
*s
)
3180 down_write(&slub_lock
);
3184 up_write(&slub_lock
);
3185 if (kmem_cache_close(s
)) {
3186 printk(KERN_ERR
"SLUB %s: %s called for cache that "
3187 "still has objects.\n", s
->name
, __func__
);
3190 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
3192 sysfs_slab_remove(s
);
3194 up_write(&slub_lock
);
3196 EXPORT_SYMBOL(kmem_cache_destroy
);
3198 /********************************************************************
3200 *******************************************************************/
3202 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
3203 EXPORT_SYMBOL(kmalloc_caches
);
3205 static struct kmem_cache
*kmem_cache
;
3207 #ifdef CONFIG_ZONE_DMA
3208 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
3211 static int __init
setup_slub_min_order(char *str
)
3213 get_option(&str
, &slub_min_order
);
3218 __setup("slub_min_order=", setup_slub_min_order
);
3220 static int __init
setup_slub_max_order(char *str
)
3222 get_option(&str
, &slub_max_order
);
3223 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3228 __setup("slub_max_order=", setup_slub_max_order
);
3230 static int __init
setup_slub_min_objects(char *str
)
3232 get_option(&str
, &slub_min_objects
);
3237 __setup("slub_min_objects=", setup_slub_min_objects
);
3239 static int __init
setup_slub_nomerge(char *str
)
3245 __setup("slub_nomerge", setup_slub_nomerge
);
3247 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
3248 int size
, unsigned int flags
)
3250 struct kmem_cache
*s
;
3252 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3255 * This function is called with IRQs disabled during early-boot on
3256 * single CPU so there's no need to take slub_lock here.
3258 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
3262 list_add(&s
->list
, &slab_caches
);
3266 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
3271 * Conversion table for small slabs sizes / 8 to the index in the
3272 * kmalloc array. This is necessary for slabs < 192 since we have non power
3273 * of two cache sizes there. The size of larger slabs can be determined using
3276 static s8 size_index
[24] = {
3303 static inline int size_index_elem(size_t bytes
)
3305 return (bytes
- 1) / 8;
3308 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
3314 return ZERO_SIZE_PTR
;
3316 index
= size_index
[size_index_elem(size
)];
3318 index
= fls(size
- 1);
3320 #ifdef CONFIG_ZONE_DMA
3321 if (unlikely((flags
& SLUB_DMA
)))
3322 return kmalloc_dma_caches
[index
];
3325 return kmalloc_caches
[index
];
3328 void *__kmalloc(size_t size
, gfp_t flags
)
3330 struct kmem_cache
*s
;
3333 if (unlikely(size
> SLUB_MAX_SIZE
))
3334 return kmalloc_large(size
, flags
);
3336 s
= get_slab(size
, flags
);
3338 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3341 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
3343 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3347 EXPORT_SYMBOL(__kmalloc
);
3350 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3355 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3356 page
= alloc_pages_node(node
, flags
, get_order(size
));
3358 ptr
= page_address(page
);
3360 kmemleak_alloc(ptr
, size
, 1, flags
);
3364 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3366 struct kmem_cache
*s
;
3369 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3370 ret
= kmalloc_large_node(size
, flags
, node
);
3372 trace_kmalloc_node(_RET_IP_
, ret
,
3373 size
, PAGE_SIZE
<< get_order(size
),
3379 s
= get_slab(size
, flags
);
3381 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3384 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
3386 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3390 EXPORT_SYMBOL(__kmalloc_node
);
3393 size_t ksize(const void *object
)
3397 if (unlikely(object
== ZERO_SIZE_PTR
))
3400 page
= virt_to_head_page(object
);
3402 if (unlikely(!PageSlab(page
))) {
3403 WARN_ON(!PageCompound(page
));
3404 return PAGE_SIZE
<< compound_order(page
);
3407 return slab_ksize(page
->slab
);
3409 EXPORT_SYMBOL(ksize
);
3411 #ifdef CONFIG_SLUB_DEBUG
3412 bool verify_mem_not_deleted(const void *x
)
3415 void *object
= (void *)x
;
3416 unsigned long flags
;
3419 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3422 local_irq_save(flags
);
3424 page
= virt_to_head_page(x
);
3425 if (unlikely(!PageSlab(page
))) {
3426 /* maybe it was from stack? */
3432 if (on_freelist(page
->slab
, page
, object
)) {
3433 object_err(page
->slab
, page
, object
, "Object is on free-list");
3441 local_irq_restore(flags
);
3444 EXPORT_SYMBOL(verify_mem_not_deleted
);
3447 void kfree(const void *x
)
3450 void *object
= (void *)x
;
3452 trace_kfree(_RET_IP_
, x
);
3454 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3457 page
= virt_to_head_page(x
);
3458 if (unlikely(!PageSlab(page
))) {
3459 BUG_ON(!PageCompound(page
));
3464 slab_free(page
->slab
, page
, object
, _RET_IP_
);
3466 EXPORT_SYMBOL(kfree
);
3469 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3470 * the remaining slabs by the number of items in use. The slabs with the
3471 * most items in use come first. New allocations will then fill those up
3472 * and thus they can be removed from the partial lists.
3474 * The slabs with the least items are placed last. This results in them
3475 * being allocated from last increasing the chance that the last objects
3476 * are freed in them.
3478 int kmem_cache_shrink(struct kmem_cache
*s
)
3482 struct kmem_cache_node
*n
;
3485 int objects
= oo_objects(s
->max
);
3486 struct list_head
*slabs_by_inuse
=
3487 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3488 unsigned long flags
;
3490 if (!slabs_by_inuse
)
3494 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3495 n
= get_node(s
, node
);
3500 for (i
= 0; i
< objects
; i
++)
3501 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3503 spin_lock_irqsave(&n
->list_lock
, flags
);
3506 * Build lists indexed by the items in use in each slab.
3508 * Note that concurrent frees may occur while we hold the
3509 * list_lock. page->inuse here is the upper limit.
3511 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3512 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3518 * Rebuild the partial list with the slabs filled up most
3519 * first and the least used slabs at the end.
3521 for (i
= objects
- 1; i
> 0; i
--)
3522 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3524 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3526 /* Release empty slabs */
3527 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3528 discard_slab(s
, page
);
3531 kfree(slabs_by_inuse
);
3534 EXPORT_SYMBOL(kmem_cache_shrink
);
3536 #if defined(CONFIG_MEMORY_HOTPLUG)
3537 static int slab_mem_going_offline_callback(void *arg
)
3539 struct kmem_cache
*s
;
3541 down_read(&slub_lock
);
3542 list_for_each_entry(s
, &slab_caches
, list
)
3543 kmem_cache_shrink(s
);
3544 up_read(&slub_lock
);
3549 static void slab_mem_offline_callback(void *arg
)
3551 struct kmem_cache_node
*n
;
3552 struct kmem_cache
*s
;
3553 struct memory_notify
*marg
= arg
;
3556 offline_node
= marg
->status_change_nid
;
3559 * If the node still has available memory. we need kmem_cache_node
3562 if (offline_node
< 0)
3565 down_read(&slub_lock
);
3566 list_for_each_entry(s
, &slab_caches
, list
) {
3567 n
= get_node(s
, offline_node
);
3570 * if n->nr_slabs > 0, slabs still exist on the node
3571 * that is going down. We were unable to free them,
3572 * and offline_pages() function shouldn't call this
3573 * callback. So, we must fail.
3575 BUG_ON(slabs_node(s
, offline_node
));
3577 s
->node
[offline_node
] = NULL
;
3578 kmem_cache_free(kmem_cache_node
, n
);
3581 up_read(&slub_lock
);
3584 static int slab_mem_going_online_callback(void *arg
)
3586 struct kmem_cache_node
*n
;
3587 struct kmem_cache
*s
;
3588 struct memory_notify
*marg
= arg
;
3589 int nid
= marg
->status_change_nid
;
3593 * If the node's memory is already available, then kmem_cache_node is
3594 * already created. Nothing to do.
3600 * We are bringing a node online. No memory is available yet. We must
3601 * allocate a kmem_cache_node structure in order to bring the node
3604 down_read(&slub_lock
);
3605 list_for_each_entry(s
, &slab_caches
, list
) {
3607 * XXX: kmem_cache_alloc_node will fallback to other nodes
3608 * since memory is not yet available from the node that
3611 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3616 init_kmem_cache_node(n
);
3620 up_read(&slub_lock
);
3624 static int slab_memory_callback(struct notifier_block
*self
,
3625 unsigned long action
, void *arg
)
3630 case MEM_GOING_ONLINE
:
3631 ret
= slab_mem_going_online_callback(arg
);
3633 case MEM_GOING_OFFLINE
:
3634 ret
= slab_mem_going_offline_callback(arg
);
3637 case MEM_CANCEL_ONLINE
:
3638 slab_mem_offline_callback(arg
);
3641 case MEM_CANCEL_OFFLINE
:
3645 ret
= notifier_from_errno(ret
);
3651 #endif /* CONFIG_MEMORY_HOTPLUG */
3653 /********************************************************************
3654 * Basic setup of slabs
3655 *******************************************************************/
3658 * Used for early kmem_cache structures that were allocated using
3659 * the page allocator
3662 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3666 list_add(&s
->list
, &slab_caches
);
3669 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3670 struct kmem_cache_node
*n
= get_node(s
, node
);
3674 list_for_each_entry(p
, &n
->partial
, lru
)
3677 #ifdef CONFIG_SLUB_DEBUG
3678 list_for_each_entry(p
, &n
->full
, lru
)
3685 void __init
kmem_cache_init(void)
3689 struct kmem_cache
*temp_kmem_cache
;
3691 struct kmem_cache
*temp_kmem_cache_node
;
3692 unsigned long kmalloc_size
;
3694 if (debug_guardpage_minorder())
3697 kmem_size
= offsetof(struct kmem_cache
, node
) +
3698 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3700 /* Allocate two kmem_caches from the page allocator */
3701 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3702 order
= get_order(2 * kmalloc_size
);
3703 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3706 * Must first have the slab cache available for the allocations of the
3707 * struct kmem_cache_node's. There is special bootstrap code in
3708 * kmem_cache_open for slab_state == DOWN.
3710 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3712 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3713 sizeof(struct kmem_cache_node
),
3714 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3716 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3718 /* Able to allocate the per node structures */
3719 slab_state
= PARTIAL
;
3721 temp_kmem_cache
= kmem_cache
;
3722 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3723 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3724 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3725 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3728 * Allocate kmem_cache_node properly from the kmem_cache slab.
3729 * kmem_cache_node is separately allocated so no need to
3730 * update any list pointers.
3732 temp_kmem_cache_node
= kmem_cache_node
;
3734 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3735 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3737 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3740 kmem_cache_bootstrap_fixup(kmem_cache
);
3742 /* Free temporary boot structure */
3743 free_pages((unsigned long)temp_kmem_cache
, order
);
3745 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3748 * Patch up the size_index table if we have strange large alignment
3749 * requirements for the kmalloc array. This is only the case for
3750 * MIPS it seems. The standard arches will not generate any code here.
3752 * Largest permitted alignment is 256 bytes due to the way we
3753 * handle the index determination for the smaller caches.
3755 * Make sure that nothing crazy happens if someone starts tinkering
3756 * around with ARCH_KMALLOC_MINALIGN
3758 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3759 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3761 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3762 int elem
= size_index_elem(i
);
3763 if (elem
>= ARRAY_SIZE(size_index
))
3765 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3768 if (KMALLOC_MIN_SIZE
== 64) {
3770 * The 96 byte size cache is not used if the alignment
3773 for (i
= 64 + 8; i
<= 96; i
+= 8)
3774 size_index
[size_index_elem(i
)] = 7;
3775 } else if (KMALLOC_MIN_SIZE
== 128) {
3777 * The 192 byte sized cache is not used if the alignment
3778 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3781 for (i
= 128 + 8; i
<= 192; i
+= 8)
3782 size_index
[size_index_elem(i
)] = 8;
3785 /* Caches that are not of the two-to-the-power-of size */
3786 if (KMALLOC_MIN_SIZE
<= 32) {
3787 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3791 if (KMALLOC_MIN_SIZE
<= 64) {
3792 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3796 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3797 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3803 /* Provide the correct kmalloc names now that the caches are up */
3804 if (KMALLOC_MIN_SIZE
<= 32) {
3805 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3806 BUG_ON(!kmalloc_caches
[1]->name
);
3809 if (KMALLOC_MIN_SIZE
<= 64) {
3810 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3811 BUG_ON(!kmalloc_caches
[2]->name
);
3814 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3815 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3818 kmalloc_caches
[i
]->name
= s
;
3822 register_cpu_notifier(&slab_notifier
);
3825 #ifdef CONFIG_ZONE_DMA
3826 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3827 struct kmem_cache
*s
= kmalloc_caches
[i
];
3830 char *name
= kasprintf(GFP_NOWAIT
,
3831 "dma-kmalloc-%d", s
->object_size
);
3834 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3835 s
->object_size
, SLAB_CACHE_DMA
);
3840 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3841 " CPUs=%d, Nodes=%d\n",
3842 caches
, cache_line_size(),
3843 slub_min_order
, slub_max_order
, slub_min_objects
,
3844 nr_cpu_ids
, nr_node_ids
);
3847 void __init
kmem_cache_init_late(void)
3852 * Find a mergeable slab cache
3854 static int slab_unmergeable(struct kmem_cache
*s
)
3856 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3863 * We may have set a slab to be unmergeable during bootstrap.
3865 if (s
->refcount
< 0)
3871 static struct kmem_cache
*find_mergeable(size_t size
,
3872 size_t align
, unsigned long flags
, const char *name
,
3873 void (*ctor
)(void *))
3875 struct kmem_cache
*s
;
3877 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3883 size
= ALIGN(size
, sizeof(void *));
3884 align
= calculate_alignment(flags
, align
, size
);
3885 size
= ALIGN(size
, align
);
3886 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3888 list_for_each_entry(s
, &slab_caches
, list
) {
3889 if (slab_unmergeable(s
))
3895 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3898 * Check if alignment is compatible.
3899 * Courtesy of Adrian Drzewiecki
3901 if ((s
->size
& ~(align
- 1)) != s
->size
)
3904 if (s
->size
- size
>= sizeof(void *))
3912 struct kmem_cache
*__kmem_cache_create(const char *name
, size_t size
,
3913 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3915 struct kmem_cache
*s
;
3918 down_write(&slub_lock
);
3919 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3923 * Adjust the object sizes so that we clear
3924 * the complete object on kzalloc.
3926 s
->object_size
= max(s
->object_size
, (int)size
);
3927 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3929 if (sysfs_slab_alias(s
, name
)) {
3933 up_write(&slub_lock
);
3937 n
= kstrdup(name
, GFP_KERNEL
);
3941 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3943 if (kmem_cache_open(s
, n
,
3944 size
, align
, flags
, ctor
)) {
3945 list_add(&s
->list
, &slab_caches
);
3946 up_write(&slub_lock
);
3947 if (sysfs_slab_add(s
)) {
3948 down_write(&slub_lock
);
3960 up_write(&slub_lock
);
3966 * Use the cpu notifier to insure that the cpu slabs are flushed when
3969 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3970 unsigned long action
, void *hcpu
)
3972 long cpu
= (long)hcpu
;
3973 struct kmem_cache
*s
;
3974 unsigned long flags
;
3977 case CPU_UP_CANCELED
:
3978 case CPU_UP_CANCELED_FROZEN
:
3980 case CPU_DEAD_FROZEN
:
3981 down_read(&slub_lock
);
3982 list_for_each_entry(s
, &slab_caches
, list
) {
3983 local_irq_save(flags
);
3984 __flush_cpu_slab(s
, cpu
);
3985 local_irq_restore(flags
);
3987 up_read(&slub_lock
);
3995 static struct notifier_block __cpuinitdata slab_notifier
= {
3996 .notifier_call
= slab_cpuup_callback
4001 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
4003 struct kmem_cache
*s
;
4006 if (unlikely(size
> SLUB_MAX_SIZE
))
4007 return kmalloc_large(size
, gfpflags
);
4009 s
= get_slab(size
, gfpflags
);
4011 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4014 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
4016 /* Honor the call site pointer we received. */
4017 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
4023 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
4024 int node
, unsigned long caller
)
4026 struct kmem_cache
*s
;
4029 if (unlikely(size
> SLUB_MAX_SIZE
)) {
4030 ret
= kmalloc_large_node(size
, gfpflags
, node
);
4032 trace_kmalloc_node(caller
, ret
,
4033 size
, PAGE_SIZE
<< get_order(size
),
4039 s
= get_slab(size
, gfpflags
);
4041 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4044 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
4046 /* Honor the call site pointer we received. */
4047 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
4054 static int count_inuse(struct page
*page
)
4059 static int count_total(struct page
*page
)
4061 return page
->objects
;
4065 #ifdef CONFIG_SLUB_DEBUG
4066 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
4070 void *addr
= page_address(page
);
4072 if (!check_slab(s
, page
) ||
4073 !on_freelist(s
, page
, NULL
))
4076 /* Now we know that a valid freelist exists */
4077 bitmap_zero(map
, page
->objects
);
4079 get_map(s
, page
, map
);
4080 for_each_object(p
, s
, addr
, page
->objects
) {
4081 if (test_bit(slab_index(p
, s
, addr
), map
))
4082 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
4086 for_each_object(p
, s
, addr
, page
->objects
)
4087 if (!test_bit(slab_index(p
, s
, addr
), map
))
4088 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
4093 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
4097 validate_slab(s
, page
, map
);
4101 static int validate_slab_node(struct kmem_cache
*s
,
4102 struct kmem_cache_node
*n
, unsigned long *map
)
4104 unsigned long count
= 0;
4106 unsigned long flags
;
4108 spin_lock_irqsave(&n
->list_lock
, flags
);
4110 list_for_each_entry(page
, &n
->partial
, lru
) {
4111 validate_slab_slab(s
, page
, map
);
4114 if (count
!= n
->nr_partial
)
4115 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
4116 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
4118 if (!(s
->flags
& SLAB_STORE_USER
))
4121 list_for_each_entry(page
, &n
->full
, lru
) {
4122 validate_slab_slab(s
, page
, map
);
4125 if (count
!= atomic_long_read(&n
->nr_slabs
))
4126 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
4127 "counter=%ld\n", s
->name
, count
,
4128 atomic_long_read(&n
->nr_slabs
));
4131 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4135 static long validate_slab_cache(struct kmem_cache
*s
)
4138 unsigned long count
= 0;
4139 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4140 sizeof(unsigned long), GFP_KERNEL
);
4146 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4147 struct kmem_cache_node
*n
= get_node(s
, node
);
4149 count
+= validate_slab_node(s
, n
, map
);
4155 * Generate lists of code addresses where slabcache objects are allocated
4160 unsigned long count
;
4167 DECLARE_BITMAP(cpus
, NR_CPUS
);
4173 unsigned long count
;
4174 struct location
*loc
;
4177 static void free_loc_track(struct loc_track
*t
)
4180 free_pages((unsigned long)t
->loc
,
4181 get_order(sizeof(struct location
) * t
->max
));
4184 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4189 order
= get_order(sizeof(struct location
) * max
);
4191 l
= (void *)__get_free_pages(flags
, order
);
4196 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4204 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4205 const struct track
*track
)
4207 long start
, end
, pos
;
4209 unsigned long caddr
;
4210 unsigned long age
= jiffies
- track
->when
;
4216 pos
= start
+ (end
- start
+ 1) / 2;
4219 * There is nothing at "end". If we end up there
4220 * we need to add something to before end.
4225 caddr
= t
->loc
[pos
].addr
;
4226 if (track
->addr
== caddr
) {
4232 if (age
< l
->min_time
)
4234 if (age
> l
->max_time
)
4237 if (track
->pid
< l
->min_pid
)
4238 l
->min_pid
= track
->pid
;
4239 if (track
->pid
> l
->max_pid
)
4240 l
->max_pid
= track
->pid
;
4242 cpumask_set_cpu(track
->cpu
,
4243 to_cpumask(l
->cpus
));
4245 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4249 if (track
->addr
< caddr
)
4256 * Not found. Insert new tracking element.
4258 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4264 (t
->count
- pos
) * sizeof(struct location
));
4267 l
->addr
= track
->addr
;
4271 l
->min_pid
= track
->pid
;
4272 l
->max_pid
= track
->pid
;
4273 cpumask_clear(to_cpumask(l
->cpus
));
4274 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4275 nodes_clear(l
->nodes
);
4276 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4280 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4281 struct page
*page
, enum track_item alloc
,
4284 void *addr
= page_address(page
);
4287 bitmap_zero(map
, page
->objects
);
4288 get_map(s
, page
, map
);
4290 for_each_object(p
, s
, addr
, page
->objects
)
4291 if (!test_bit(slab_index(p
, s
, addr
), map
))
4292 add_location(t
, s
, get_track(s
, p
, alloc
));
4295 static int list_locations(struct kmem_cache
*s
, char *buf
,
4296 enum track_item alloc
)
4300 struct loc_track t
= { 0, 0, NULL
};
4302 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4303 sizeof(unsigned long), GFP_KERNEL
);
4305 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4308 return sprintf(buf
, "Out of memory\n");
4310 /* Push back cpu slabs */
4313 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4314 struct kmem_cache_node
*n
= get_node(s
, node
);
4315 unsigned long flags
;
4318 if (!atomic_long_read(&n
->nr_slabs
))
4321 spin_lock_irqsave(&n
->list_lock
, flags
);
4322 list_for_each_entry(page
, &n
->partial
, lru
)
4323 process_slab(&t
, s
, page
, alloc
, map
);
4324 list_for_each_entry(page
, &n
->full
, lru
)
4325 process_slab(&t
, s
, page
, alloc
, map
);
4326 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4329 for (i
= 0; i
< t
.count
; i
++) {
4330 struct location
*l
= &t
.loc
[i
];
4332 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4334 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4337 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4339 len
+= sprintf(buf
+ len
, "<not-available>");
4341 if (l
->sum_time
!= l
->min_time
) {
4342 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4344 (long)div_u64(l
->sum_time
, l
->count
),
4347 len
+= sprintf(buf
+ len
, " age=%ld",
4350 if (l
->min_pid
!= l
->max_pid
)
4351 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4352 l
->min_pid
, l
->max_pid
);
4354 len
+= sprintf(buf
+ len
, " pid=%ld",
4357 if (num_online_cpus() > 1 &&
4358 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4359 len
< PAGE_SIZE
- 60) {
4360 len
+= sprintf(buf
+ len
, " cpus=");
4361 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4362 to_cpumask(l
->cpus
));
4365 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4366 len
< PAGE_SIZE
- 60) {
4367 len
+= sprintf(buf
+ len
, " nodes=");
4368 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4372 len
+= sprintf(buf
+ len
, "\n");
4378 len
+= sprintf(buf
, "No data\n");
4383 #ifdef SLUB_RESILIENCY_TEST
4384 static void resiliency_test(void)
4388 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
4390 printk(KERN_ERR
"SLUB resiliency testing\n");
4391 printk(KERN_ERR
"-----------------------\n");
4392 printk(KERN_ERR
"A. Corruption after allocation\n");
4394 p
= kzalloc(16, GFP_KERNEL
);
4396 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4397 " 0x12->0x%p\n\n", p
+ 16);
4399 validate_slab_cache(kmalloc_caches
[4]);
4401 /* Hmmm... The next two are dangerous */
4402 p
= kzalloc(32, GFP_KERNEL
);
4403 p
[32 + sizeof(void *)] = 0x34;
4404 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4405 " 0x34 -> -0x%p\n", p
);
4407 "If allocated object is overwritten then not detectable\n\n");
4409 validate_slab_cache(kmalloc_caches
[5]);
4410 p
= kzalloc(64, GFP_KERNEL
);
4411 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4413 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4416 "If allocated object is overwritten then not detectable\n\n");
4417 validate_slab_cache(kmalloc_caches
[6]);
4419 printk(KERN_ERR
"\nB. Corruption after free\n");
4420 p
= kzalloc(128, GFP_KERNEL
);
4423 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4424 validate_slab_cache(kmalloc_caches
[7]);
4426 p
= kzalloc(256, GFP_KERNEL
);
4429 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4431 validate_slab_cache(kmalloc_caches
[8]);
4433 p
= kzalloc(512, GFP_KERNEL
);
4436 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4437 validate_slab_cache(kmalloc_caches
[9]);
4441 static void resiliency_test(void) {};
4446 enum slab_stat_type
{
4447 SL_ALL
, /* All slabs */
4448 SL_PARTIAL
, /* Only partially allocated slabs */
4449 SL_CPU
, /* Only slabs used for cpu caches */
4450 SL_OBJECTS
, /* Determine allocated objects not slabs */
4451 SL_TOTAL
/* Determine object capacity not slabs */
4454 #define SO_ALL (1 << SL_ALL)
4455 #define SO_PARTIAL (1 << SL_PARTIAL)
4456 #define SO_CPU (1 << SL_CPU)
4457 #define SO_OBJECTS (1 << SL_OBJECTS)
4458 #define SO_TOTAL (1 << SL_TOTAL)
4460 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4461 char *buf
, unsigned long flags
)
4463 unsigned long total
= 0;
4466 unsigned long *nodes
;
4467 unsigned long *per_cpu
;
4469 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4472 per_cpu
= nodes
+ nr_node_ids
;
4474 if (flags
& SO_CPU
) {
4477 for_each_possible_cpu(cpu
) {
4478 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4482 page
= ACCESS_ONCE(c
->page
);
4486 node
= page_to_nid(page
);
4487 if (flags
& SO_TOTAL
)
4489 else if (flags
& SO_OBJECTS
)
4497 page
= ACCESS_ONCE(c
->partial
);
4508 lock_memory_hotplug();
4509 #ifdef CONFIG_SLUB_DEBUG
4510 if (flags
& SO_ALL
) {
4511 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4512 struct kmem_cache_node
*n
= get_node(s
, node
);
4514 if (flags
& SO_TOTAL
)
4515 x
= atomic_long_read(&n
->total_objects
);
4516 else if (flags
& SO_OBJECTS
)
4517 x
= atomic_long_read(&n
->total_objects
) -
4518 count_partial(n
, count_free
);
4521 x
= atomic_long_read(&n
->nr_slabs
);
4528 if (flags
& SO_PARTIAL
) {
4529 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4530 struct kmem_cache_node
*n
= get_node(s
, node
);
4532 if (flags
& SO_TOTAL
)
4533 x
= count_partial(n
, count_total
);
4534 else if (flags
& SO_OBJECTS
)
4535 x
= count_partial(n
, count_inuse
);
4542 x
= sprintf(buf
, "%lu", total
);
4544 for_each_node_state(node
, N_NORMAL_MEMORY
)
4546 x
+= sprintf(buf
+ x
, " N%d=%lu",
4549 unlock_memory_hotplug();
4551 return x
+ sprintf(buf
+ x
, "\n");
4554 #ifdef CONFIG_SLUB_DEBUG
4555 static int any_slab_objects(struct kmem_cache
*s
)
4559 for_each_online_node(node
) {
4560 struct kmem_cache_node
*n
= get_node(s
, node
);
4565 if (atomic_long_read(&n
->total_objects
))
4572 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4573 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4575 struct slab_attribute
{
4576 struct attribute attr
;
4577 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4578 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4581 #define SLAB_ATTR_RO(_name) \
4582 static struct slab_attribute _name##_attr = \
4583 __ATTR(_name, 0400, _name##_show, NULL)
4585 #define SLAB_ATTR(_name) \
4586 static struct slab_attribute _name##_attr = \
4587 __ATTR(_name, 0600, _name##_show, _name##_store)
4589 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4591 return sprintf(buf
, "%d\n", s
->size
);
4593 SLAB_ATTR_RO(slab_size
);
4595 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4597 return sprintf(buf
, "%d\n", s
->align
);
4599 SLAB_ATTR_RO(align
);
4601 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4603 return sprintf(buf
, "%d\n", s
->object_size
);
4605 SLAB_ATTR_RO(object_size
);
4607 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4609 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4611 SLAB_ATTR_RO(objs_per_slab
);
4613 static ssize_t
order_store(struct kmem_cache
*s
,
4614 const char *buf
, size_t length
)
4616 unsigned long order
;
4619 err
= strict_strtoul(buf
, 10, &order
);
4623 if (order
> slub_max_order
|| order
< slub_min_order
)
4626 calculate_sizes(s
, order
);
4630 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4632 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4636 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4638 return sprintf(buf
, "%lu\n", s
->min_partial
);
4641 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4647 err
= strict_strtoul(buf
, 10, &min
);
4651 set_min_partial(s
, min
);
4654 SLAB_ATTR(min_partial
);
4656 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4658 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4661 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4664 unsigned long objects
;
4667 err
= strict_strtoul(buf
, 10, &objects
);
4670 if (objects
&& kmem_cache_debug(s
))
4673 s
->cpu_partial
= objects
;
4677 SLAB_ATTR(cpu_partial
);
4679 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4683 return sprintf(buf
, "%pS\n", s
->ctor
);
4687 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4689 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4691 SLAB_ATTR_RO(aliases
);
4693 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4695 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4697 SLAB_ATTR_RO(partial
);
4699 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4701 return show_slab_objects(s
, buf
, SO_CPU
);
4703 SLAB_ATTR_RO(cpu_slabs
);
4705 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4707 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4709 SLAB_ATTR_RO(objects
);
4711 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4713 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4715 SLAB_ATTR_RO(objects_partial
);
4717 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4724 for_each_online_cpu(cpu
) {
4725 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4728 pages
+= page
->pages
;
4729 objects
+= page
->pobjects
;
4733 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4736 for_each_online_cpu(cpu
) {
4737 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4739 if (page
&& len
< PAGE_SIZE
- 20)
4740 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4741 page
->pobjects
, page
->pages
);
4744 return len
+ sprintf(buf
+ len
, "\n");
4746 SLAB_ATTR_RO(slabs_cpu_partial
);
4748 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4750 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4753 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4754 const char *buf
, size_t length
)
4756 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4758 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4761 SLAB_ATTR(reclaim_account
);
4763 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4765 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4767 SLAB_ATTR_RO(hwcache_align
);
4769 #ifdef CONFIG_ZONE_DMA
4770 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4772 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4774 SLAB_ATTR_RO(cache_dma
);
4777 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4779 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4781 SLAB_ATTR_RO(destroy_by_rcu
);
4783 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4785 return sprintf(buf
, "%d\n", s
->reserved
);
4787 SLAB_ATTR_RO(reserved
);
4789 #ifdef CONFIG_SLUB_DEBUG
4790 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4792 return show_slab_objects(s
, buf
, SO_ALL
);
4794 SLAB_ATTR_RO(slabs
);
4796 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4798 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4800 SLAB_ATTR_RO(total_objects
);
4802 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4804 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4807 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4808 const char *buf
, size_t length
)
4810 s
->flags
&= ~SLAB_DEBUG_FREE
;
4811 if (buf
[0] == '1') {
4812 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4813 s
->flags
|= SLAB_DEBUG_FREE
;
4817 SLAB_ATTR(sanity_checks
);
4819 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4821 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4824 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4827 s
->flags
&= ~SLAB_TRACE
;
4828 if (buf
[0] == '1') {
4829 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4830 s
->flags
|= SLAB_TRACE
;
4836 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4838 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4841 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4842 const char *buf
, size_t length
)
4844 if (any_slab_objects(s
))
4847 s
->flags
&= ~SLAB_RED_ZONE
;
4848 if (buf
[0] == '1') {
4849 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4850 s
->flags
|= SLAB_RED_ZONE
;
4852 calculate_sizes(s
, -1);
4855 SLAB_ATTR(red_zone
);
4857 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4859 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4862 static ssize_t
poison_store(struct kmem_cache
*s
,
4863 const char *buf
, size_t length
)
4865 if (any_slab_objects(s
))
4868 s
->flags
&= ~SLAB_POISON
;
4869 if (buf
[0] == '1') {
4870 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4871 s
->flags
|= SLAB_POISON
;
4873 calculate_sizes(s
, -1);
4878 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4880 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4883 static ssize_t
store_user_store(struct kmem_cache
*s
,
4884 const char *buf
, size_t length
)
4886 if (any_slab_objects(s
))
4889 s
->flags
&= ~SLAB_STORE_USER
;
4890 if (buf
[0] == '1') {
4891 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4892 s
->flags
|= SLAB_STORE_USER
;
4894 calculate_sizes(s
, -1);
4897 SLAB_ATTR(store_user
);
4899 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4904 static ssize_t
validate_store(struct kmem_cache
*s
,
4905 const char *buf
, size_t length
)
4909 if (buf
[0] == '1') {
4910 ret
= validate_slab_cache(s
);
4916 SLAB_ATTR(validate
);
4918 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4920 if (!(s
->flags
& SLAB_STORE_USER
))
4922 return list_locations(s
, buf
, TRACK_ALLOC
);
4924 SLAB_ATTR_RO(alloc_calls
);
4926 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4928 if (!(s
->flags
& SLAB_STORE_USER
))
4930 return list_locations(s
, buf
, TRACK_FREE
);
4932 SLAB_ATTR_RO(free_calls
);
4933 #endif /* CONFIG_SLUB_DEBUG */
4935 #ifdef CONFIG_FAILSLAB
4936 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4938 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4941 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4944 s
->flags
&= ~SLAB_FAILSLAB
;
4946 s
->flags
|= SLAB_FAILSLAB
;
4949 SLAB_ATTR(failslab
);
4952 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4957 static ssize_t
shrink_store(struct kmem_cache
*s
,
4958 const char *buf
, size_t length
)
4960 if (buf
[0] == '1') {
4961 int rc
= kmem_cache_shrink(s
);
4972 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4974 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4977 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4978 const char *buf
, size_t length
)
4980 unsigned long ratio
;
4983 err
= strict_strtoul(buf
, 10, &ratio
);
4988 s
->remote_node_defrag_ratio
= ratio
* 10;
4992 SLAB_ATTR(remote_node_defrag_ratio
);
4995 #ifdef CONFIG_SLUB_STATS
4996 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4998 unsigned long sum
= 0;
5001 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
5006 for_each_online_cpu(cpu
) {
5007 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
5013 len
= sprintf(buf
, "%lu", sum
);
5016 for_each_online_cpu(cpu
) {
5017 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
5018 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
5022 return len
+ sprintf(buf
+ len
, "\n");
5025 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
5029 for_each_online_cpu(cpu
)
5030 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
5033 #define STAT_ATTR(si, text) \
5034 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5036 return show_stat(s, buf, si); \
5038 static ssize_t text##_store(struct kmem_cache *s, \
5039 const char *buf, size_t length) \
5041 if (buf[0] != '0') \
5043 clear_stat(s, si); \
5048 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5049 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
5050 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
5051 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
5052 STAT_ATTR(FREE_FROZEN
, free_frozen
);
5053 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
5054 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
5055 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
5056 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
5057 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
5058 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
5059 STAT_ATTR(FREE_SLAB
, free_slab
);
5060 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
5061 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
5062 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
5063 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
5064 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
5065 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
5066 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
5067 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
5068 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
5069 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
5070 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
5071 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
5072 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
5073 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
5076 static struct attribute
*slab_attrs
[] = {
5077 &slab_size_attr
.attr
,
5078 &object_size_attr
.attr
,
5079 &objs_per_slab_attr
.attr
,
5081 &min_partial_attr
.attr
,
5082 &cpu_partial_attr
.attr
,
5084 &objects_partial_attr
.attr
,
5086 &cpu_slabs_attr
.attr
,
5090 &hwcache_align_attr
.attr
,
5091 &reclaim_account_attr
.attr
,
5092 &destroy_by_rcu_attr
.attr
,
5094 &reserved_attr
.attr
,
5095 &slabs_cpu_partial_attr
.attr
,
5096 #ifdef CONFIG_SLUB_DEBUG
5097 &total_objects_attr
.attr
,
5099 &sanity_checks_attr
.attr
,
5101 &red_zone_attr
.attr
,
5103 &store_user_attr
.attr
,
5104 &validate_attr
.attr
,
5105 &alloc_calls_attr
.attr
,
5106 &free_calls_attr
.attr
,
5108 #ifdef CONFIG_ZONE_DMA
5109 &cache_dma_attr
.attr
,
5112 &remote_node_defrag_ratio_attr
.attr
,
5114 #ifdef CONFIG_SLUB_STATS
5115 &alloc_fastpath_attr
.attr
,
5116 &alloc_slowpath_attr
.attr
,
5117 &free_fastpath_attr
.attr
,
5118 &free_slowpath_attr
.attr
,
5119 &free_frozen_attr
.attr
,
5120 &free_add_partial_attr
.attr
,
5121 &free_remove_partial_attr
.attr
,
5122 &alloc_from_partial_attr
.attr
,
5123 &alloc_slab_attr
.attr
,
5124 &alloc_refill_attr
.attr
,
5125 &alloc_node_mismatch_attr
.attr
,
5126 &free_slab_attr
.attr
,
5127 &cpuslab_flush_attr
.attr
,
5128 &deactivate_full_attr
.attr
,
5129 &deactivate_empty_attr
.attr
,
5130 &deactivate_to_head_attr
.attr
,
5131 &deactivate_to_tail_attr
.attr
,
5132 &deactivate_remote_frees_attr
.attr
,
5133 &deactivate_bypass_attr
.attr
,
5134 &order_fallback_attr
.attr
,
5135 &cmpxchg_double_fail_attr
.attr
,
5136 &cmpxchg_double_cpu_fail_attr
.attr
,
5137 &cpu_partial_alloc_attr
.attr
,
5138 &cpu_partial_free_attr
.attr
,
5139 &cpu_partial_node_attr
.attr
,
5140 &cpu_partial_drain_attr
.attr
,
5142 #ifdef CONFIG_FAILSLAB
5143 &failslab_attr
.attr
,
5149 static struct attribute_group slab_attr_group
= {
5150 .attrs
= slab_attrs
,
5153 static ssize_t
slab_attr_show(struct kobject
*kobj
,
5154 struct attribute
*attr
,
5157 struct slab_attribute
*attribute
;
5158 struct kmem_cache
*s
;
5161 attribute
= to_slab_attr(attr
);
5164 if (!attribute
->show
)
5167 err
= attribute
->show(s
, buf
);
5172 static ssize_t
slab_attr_store(struct kobject
*kobj
,
5173 struct attribute
*attr
,
5174 const char *buf
, size_t len
)
5176 struct slab_attribute
*attribute
;
5177 struct kmem_cache
*s
;
5180 attribute
= to_slab_attr(attr
);
5183 if (!attribute
->store
)
5186 err
= attribute
->store(s
, buf
, len
);
5191 static void kmem_cache_release(struct kobject
*kobj
)
5193 struct kmem_cache
*s
= to_slab(kobj
);
5199 static const struct sysfs_ops slab_sysfs_ops
= {
5200 .show
= slab_attr_show
,
5201 .store
= slab_attr_store
,
5204 static struct kobj_type slab_ktype
= {
5205 .sysfs_ops
= &slab_sysfs_ops
,
5206 .release
= kmem_cache_release
5209 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5211 struct kobj_type
*ktype
= get_ktype(kobj
);
5213 if (ktype
== &slab_ktype
)
5218 static const struct kset_uevent_ops slab_uevent_ops
= {
5219 .filter
= uevent_filter
,
5222 static struct kset
*slab_kset
;
5224 #define ID_STR_LENGTH 64
5226 /* Create a unique string id for a slab cache:
5228 * Format :[flags-]size
5230 static char *create_unique_id(struct kmem_cache
*s
)
5232 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5239 * First flags affecting slabcache operations. We will only
5240 * get here for aliasable slabs so we do not need to support
5241 * too many flags. The flags here must cover all flags that
5242 * are matched during merging to guarantee that the id is
5245 if (s
->flags
& SLAB_CACHE_DMA
)
5247 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5249 if (s
->flags
& SLAB_DEBUG_FREE
)
5251 if (!(s
->flags
& SLAB_NOTRACK
))
5255 p
+= sprintf(p
, "%07d", s
->size
);
5256 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5260 static int sysfs_slab_add(struct kmem_cache
*s
)
5266 if (slab_state
< FULL
)
5267 /* Defer until later */
5270 unmergeable
= slab_unmergeable(s
);
5273 * Slabcache can never be merged so we can use the name proper.
5274 * This is typically the case for debug situations. In that
5275 * case we can catch duplicate names easily.
5277 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5281 * Create a unique name for the slab as a target
5284 name
= create_unique_id(s
);
5287 s
->kobj
.kset
= slab_kset
;
5288 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5290 kobject_put(&s
->kobj
);
5294 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5296 kobject_del(&s
->kobj
);
5297 kobject_put(&s
->kobj
);
5300 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5302 /* Setup first alias */
5303 sysfs_slab_alias(s
, s
->name
);
5309 static void sysfs_slab_remove(struct kmem_cache
*s
)
5311 if (slab_state
< FULL
)
5313 * Sysfs has not been setup yet so no need to remove the
5318 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5319 kobject_del(&s
->kobj
);
5320 kobject_put(&s
->kobj
);
5324 * Need to buffer aliases during bootup until sysfs becomes
5325 * available lest we lose that information.
5327 struct saved_alias
{
5328 struct kmem_cache
*s
;
5330 struct saved_alias
*next
;
5333 static struct saved_alias
*alias_list
;
5335 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5337 struct saved_alias
*al
;
5339 if (slab_state
== FULL
) {
5341 * If we have a leftover link then remove it.
5343 sysfs_remove_link(&slab_kset
->kobj
, name
);
5344 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5347 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5353 al
->next
= alias_list
;
5358 static int __init
slab_sysfs_init(void)
5360 struct kmem_cache
*s
;
5363 down_write(&slub_lock
);
5365 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5367 up_write(&slub_lock
);
5368 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5374 list_for_each_entry(s
, &slab_caches
, list
) {
5375 err
= sysfs_slab_add(s
);
5377 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5378 " to sysfs\n", s
->name
);
5381 while (alias_list
) {
5382 struct saved_alias
*al
= alias_list
;
5384 alias_list
= alias_list
->next
;
5385 err
= sysfs_slab_alias(al
->s
, al
->name
);
5387 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5388 " %s to sysfs\n", al
->name
);
5392 up_write(&slub_lock
);
5397 __initcall(slab_sysfs_init
);
5398 #endif /* CONFIG_SYSFS */
5401 * The /proc/slabinfo ABI
5403 #ifdef CONFIG_SLABINFO
5404 static void print_slabinfo_header(struct seq_file
*m
)
5406 seq_puts(m
, "slabinfo - version: 2.1\n");
5407 seq_puts(m
, "# name <active_objs> <num_objs> <object_size> "
5408 "<objperslab> <pagesperslab>");
5409 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
5410 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5414 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
5418 down_read(&slub_lock
);
5420 print_slabinfo_header(m
);
5422 return seq_list_start(&slab_caches
, *pos
);
5425 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
5427 return seq_list_next(p
, &slab_caches
, pos
);
5430 static void s_stop(struct seq_file
*m
, void *p
)
5432 up_read(&slub_lock
);
5435 static int s_show(struct seq_file
*m
, void *p
)
5437 unsigned long nr_partials
= 0;
5438 unsigned long nr_slabs
= 0;
5439 unsigned long nr_inuse
= 0;
5440 unsigned long nr_objs
= 0;
5441 unsigned long nr_free
= 0;
5442 struct kmem_cache
*s
;
5445 s
= list_entry(p
, struct kmem_cache
, list
);
5447 for_each_online_node(node
) {
5448 struct kmem_cache_node
*n
= get_node(s
, node
);
5453 nr_partials
+= n
->nr_partial
;
5454 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
5455 nr_objs
+= atomic_long_read(&n
->total_objects
);
5456 nr_free
+= count_partial(n
, count_free
);
5459 nr_inuse
= nr_objs
- nr_free
;
5461 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
5462 nr_objs
, s
->size
, oo_objects(s
->oo
),
5463 (1 << oo_order(s
->oo
)));
5464 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
5465 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
5471 static const struct seq_operations slabinfo_op
= {
5478 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
5480 return seq_open(file
, &slabinfo_op
);
5483 static const struct file_operations proc_slabinfo_operations
= {
5484 .open
= slabinfo_open
,
5486 .llseek
= seq_lseek
,
5487 .release
= seq_release
,
5490 static int __init
slab_proc_init(void)
5492 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
5495 module_init(slab_proc_init
);
5496 #endif /* CONFIG_SLABINFO */