2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/kmemcheck.h>
22 #include <linux/cpu.h>
23 #include <linux/cpuset.h>
24 #include <linux/mempolicy.h>
25 #include <linux/ctype.h>
26 #include <linux/debugobjects.h>
27 #include <linux/kallsyms.h>
28 #include <linux/memory.h>
29 #include <linux/math64.h>
30 #include <linux/fault-inject.h>
31 #include <linux/stacktrace.h>
33 #include <trace/events/kmem.h>
37 * 1. slub_lock (Global Semaphore)
39 * 3. slab_lock(page) (Only on some arches and for debugging)
43 * The role of the slub_lock is to protect the list of all the slabs
44 * and to synchronize major metadata changes to slab cache structures.
46 * The slab_lock is only used for debugging and on arches that do not
47 * have the ability to do a cmpxchg_double. It only protects the second
48 * double word in the page struct. Meaning
49 * A. page->freelist -> List of object free in a page
50 * B. page->counters -> Counters of objects
51 * C. page->frozen -> frozen state
53 * If a slab is frozen then it is exempt from list management. It is not
54 * on any list. The processor that froze the slab is the one who can
55 * perform list operations on the page. Other processors may put objects
56 * onto the freelist but the processor that froze the slab is the only
57 * one that can retrieve the objects from the page's freelist.
59 * The list_lock protects the partial and full list on each node and
60 * the partial slab counter. If taken then no new slabs may be added or
61 * removed from the lists nor make the number of partial slabs be modified.
62 * (Note that the total number of slabs is an atomic value that may be
63 * modified without taking the list lock).
65 * The list_lock is a centralized lock and thus we avoid taking it as
66 * much as possible. As long as SLUB does not have to handle partial
67 * slabs, operations can continue without any centralized lock. F.e.
68 * allocating a long series of objects that fill up slabs does not require
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache
*s
)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
134 /* Enable to log cmpxchg failures */
135 #undef SLUB_DEBUG_CMPXCHG
138 * Mininum number of partial slabs. These will be left on the partial
139 * lists even if they are empty. kmem_cache_shrink may reclaim them.
141 #define MIN_PARTIAL 5
144 * Maximum number of desirable partial slabs.
145 * The existence of more partial slabs makes kmem_cache_shrink
146 * sort the partial list by the number of objects in the.
148 #define MAX_PARTIAL 10
150 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
151 SLAB_POISON | SLAB_STORE_USER)
154 * Debugging flags that require metadata to be stored in the slab. These get
155 * disabled when slub_debug=O is used and a cache's min order increases with
158 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
161 * Set of flags that will prevent slab merging
163 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
164 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
167 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
168 SLAB_CACHE_DMA | SLAB_NOTRACK)
171 #define OO_MASK ((1 << OO_SHIFT) - 1)
172 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
174 /* Internal SLUB flags */
175 #define __OBJECT_POISON 0x80000000UL /* Poison object */
176 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
178 static int kmem_size
= sizeof(struct kmem_cache
);
181 static struct notifier_block slab_notifier
;
185 DOWN
, /* No slab functionality available */
186 PARTIAL
, /* Kmem_cache_node works */
187 UP
, /* Everything works but does not show up in sysfs */
191 /* A list of all slab caches on the system */
192 static DECLARE_RWSEM(slub_lock
);
193 static LIST_HEAD(slab_caches
);
196 * Tracking user of a slab.
198 #define TRACK_ADDRS_COUNT 16
200 unsigned long addr
; /* Called from address */
201 #ifdef CONFIG_STACKTRACE
202 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
204 int cpu
; /* Was running on cpu */
205 int pid
; /* Pid context */
206 unsigned long when
; /* When did the operation occur */
209 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
212 static int sysfs_slab_add(struct kmem_cache
*);
213 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
214 static void sysfs_slab_remove(struct kmem_cache
*);
217 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
218 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
220 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
228 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
230 #ifdef CONFIG_SLUB_STATS
231 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
235 /********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
239 int slab_is_available(void)
241 return slab_state
>= UP
;
244 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
246 return s
->node
[node
];
249 /* Verify that a pointer has an address that is valid within a slab page */
250 static inline int check_valid_pointer(struct kmem_cache
*s
,
251 struct page
*page
, const void *object
)
258 base
= page_address(page
);
259 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
260 (object
- base
) % s
->size
) {
267 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
269 return *(void **)(object
+ s
->offset
);
272 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
276 #ifdef CONFIG_DEBUG_PAGEALLOC
277 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
279 p
= get_freepointer(s
, object
);
284 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
286 *(void **)(object
+ s
->offset
) = fp
;
289 /* Loop over all objects in a slab */
290 #define for_each_object(__p, __s, __addr, __objects) \
291 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
294 /* Determine object index from a given position */
295 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
297 return (p
- addr
) / s
->size
;
300 static inline size_t slab_ksize(const struct kmem_cache
*s
)
302 #ifdef CONFIG_SLUB_DEBUG
304 * Debugging requires use of the padding between object
305 * and whatever may come after it.
307 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
312 * If we have the need to store the freelist pointer
313 * back there or track user information then we can
314 * only use the space before that information.
316 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
319 * Else we can use all the padding etc for the allocation
324 static inline int order_objects(int order
, unsigned long size
, int reserved
)
326 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
329 static inline struct kmem_cache_order_objects
oo_make(int order
,
330 unsigned long size
, int reserved
)
332 struct kmem_cache_order_objects x
= {
333 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
339 static inline int oo_order(struct kmem_cache_order_objects x
)
341 return x
.x
>> OO_SHIFT
;
344 static inline int oo_objects(struct kmem_cache_order_objects x
)
346 return x
.x
& OO_MASK
;
350 * Per slab locking using the pagelock
352 static __always_inline
void slab_lock(struct page
*page
)
354 bit_spin_lock(PG_locked
, &page
->flags
);
357 static __always_inline
void slab_unlock(struct page
*page
)
359 __bit_spin_unlock(PG_locked
, &page
->flags
);
362 /* Interrupts must be disabled (for the fallback code to work right) */
363 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
364 void *freelist_old
, unsigned long counters_old
,
365 void *freelist_new
, unsigned long counters_new
,
368 VM_BUG_ON(!irqs_disabled());
369 #ifdef CONFIG_CMPXCHG_DOUBLE
370 if (s
->flags
& __CMPXCHG_DOUBLE
) {
371 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
372 freelist_old
, counters_old
,
373 freelist_new
, counters_new
))
379 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
380 page
->freelist
= freelist_new
;
381 page
->counters
= counters_new
;
389 stat(s
, CMPXCHG_DOUBLE_FAIL
);
391 #ifdef SLUB_DEBUG_CMPXCHG
392 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
398 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
399 void *freelist_old
, unsigned long counters_old
,
400 void *freelist_new
, unsigned long counters_new
,
403 #ifdef CONFIG_CMPXCHG_DOUBLE
404 if (s
->flags
& __CMPXCHG_DOUBLE
) {
405 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
406 freelist_old
, counters_old
,
407 freelist_new
, counters_new
))
414 local_irq_save(flags
);
416 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
417 page
->freelist
= freelist_new
;
418 page
->counters
= counters_new
;
420 local_irq_restore(flags
);
424 local_irq_restore(flags
);
428 stat(s
, CMPXCHG_DOUBLE_FAIL
);
430 #ifdef SLUB_DEBUG_CMPXCHG
431 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
437 #ifdef CONFIG_SLUB_DEBUG
439 * Determine a map of object in use on a page.
441 * Node listlock must be held to guarantee that the page does
442 * not vanish from under us.
444 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
447 void *addr
= page_address(page
);
449 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
450 set_bit(slab_index(p
, s
, addr
), map
);
456 #ifdef CONFIG_SLUB_DEBUG_ON
457 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
459 static int slub_debug
;
462 static char *slub_debug_slabs
;
463 static int disable_higher_order_debug
;
468 static void print_section(char *text
, u8
*addr
, unsigned int length
)
470 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
474 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
475 enum track_item alloc
)
480 p
= object
+ s
->offset
+ sizeof(void *);
482 p
= object
+ s
->inuse
;
487 static void set_track(struct kmem_cache
*s
, void *object
,
488 enum track_item alloc
, unsigned long addr
)
490 struct track
*p
= get_track(s
, object
, alloc
);
493 #ifdef CONFIG_STACKTRACE
494 struct stack_trace trace
;
497 trace
.nr_entries
= 0;
498 trace
.max_entries
= TRACK_ADDRS_COUNT
;
499 trace
.entries
= p
->addrs
;
501 save_stack_trace(&trace
);
503 /* See rant in lockdep.c */
504 if (trace
.nr_entries
!= 0 &&
505 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
508 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
512 p
->cpu
= smp_processor_id();
513 p
->pid
= current
->pid
;
516 memset(p
, 0, sizeof(struct track
));
519 static void init_tracking(struct kmem_cache
*s
, void *object
)
521 if (!(s
->flags
& SLAB_STORE_USER
))
524 set_track(s
, object
, TRACK_FREE
, 0UL);
525 set_track(s
, object
, TRACK_ALLOC
, 0UL);
528 static void print_track(const char *s
, struct track
*t
)
533 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
534 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
535 #ifdef CONFIG_STACKTRACE
538 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
540 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
547 static void print_tracking(struct kmem_cache
*s
, void *object
)
549 if (!(s
->flags
& SLAB_STORE_USER
))
552 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
553 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
556 static void print_page_info(struct page
*page
)
558 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
559 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
563 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
569 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
571 printk(KERN_ERR
"========================================"
572 "=====================================\n");
573 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
574 printk(KERN_ERR
"----------------------------------------"
575 "-------------------------------------\n\n");
578 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
584 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
586 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
589 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
591 unsigned int off
; /* Offset of last byte */
592 u8
*addr
= page_address(page
);
594 print_tracking(s
, p
);
596 print_page_info(page
);
598 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
599 p
, p
- addr
, get_freepointer(s
, p
));
602 print_section("Bytes b4 ", p
- 16, 16);
604 print_section("Object ", p
, min_t(unsigned long, s
->objsize
,
606 if (s
->flags
& SLAB_RED_ZONE
)
607 print_section("Redzone ", p
+ s
->objsize
,
608 s
->inuse
- s
->objsize
);
611 off
= s
->offset
+ sizeof(void *);
615 if (s
->flags
& SLAB_STORE_USER
)
616 off
+= 2 * sizeof(struct track
);
619 /* Beginning of the filler is the free pointer */
620 print_section("Padding ", p
+ off
, s
->size
- off
);
625 static void object_err(struct kmem_cache
*s
, struct page
*page
,
626 u8
*object
, char *reason
)
628 slab_bug(s
, "%s", reason
);
629 print_trailer(s
, page
, object
);
632 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
638 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
640 slab_bug(s
, "%s", buf
);
641 print_page_info(page
);
645 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
649 if (s
->flags
& __OBJECT_POISON
) {
650 memset(p
, POISON_FREE
, s
->objsize
- 1);
651 p
[s
->objsize
- 1] = POISON_END
;
654 if (s
->flags
& SLAB_RED_ZONE
)
655 memset(p
+ s
->objsize
, val
, s
->inuse
- s
->objsize
);
658 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
659 void *from
, void *to
)
661 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
662 memset(from
, data
, to
- from
);
665 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
666 u8
*object
, char *what
,
667 u8
*start
, unsigned int value
, unsigned int bytes
)
672 fault
= memchr_inv(start
, value
, bytes
);
677 while (end
> fault
&& end
[-1] == value
)
680 slab_bug(s
, "%s overwritten", what
);
681 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
682 fault
, end
- 1, fault
[0], value
);
683 print_trailer(s
, page
, object
);
685 restore_bytes(s
, what
, value
, fault
, end
);
693 * Bytes of the object to be managed.
694 * If the freepointer may overlay the object then the free
695 * pointer is the first word of the object.
697 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
700 * object + s->objsize
701 * Padding to reach word boundary. This is also used for Redzoning.
702 * Padding is extended by another word if Redzoning is enabled and
705 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
706 * 0xcc (RED_ACTIVE) for objects in use.
709 * Meta data starts here.
711 * A. Free pointer (if we cannot overwrite object on free)
712 * B. Tracking data for SLAB_STORE_USER
713 * C. Padding to reach required alignment boundary or at mininum
714 * one word if debugging is on to be able to detect writes
715 * before the word boundary.
717 * Padding is done using 0x5a (POISON_INUSE)
720 * Nothing is used beyond s->size.
722 * If slabcaches are merged then the objsize and inuse boundaries are mostly
723 * ignored. And therefore no slab options that rely on these boundaries
724 * may be used with merged slabcaches.
727 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
729 unsigned long off
= s
->inuse
; /* The end of info */
732 /* Freepointer is placed after the object. */
733 off
+= sizeof(void *);
735 if (s
->flags
& SLAB_STORE_USER
)
736 /* We also have user information there */
737 off
+= 2 * sizeof(struct track
);
742 return check_bytes_and_report(s
, page
, p
, "Object padding",
743 p
+ off
, POISON_INUSE
, s
->size
- off
);
746 /* Check the pad bytes at the end of a slab page */
747 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
755 if (!(s
->flags
& SLAB_POISON
))
758 start
= page_address(page
);
759 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
760 end
= start
+ length
;
761 remainder
= length
% s
->size
;
765 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
768 while (end
> fault
&& end
[-1] == POISON_INUSE
)
771 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
772 print_section("Padding ", end
- remainder
, remainder
);
774 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
778 static int check_object(struct kmem_cache
*s
, struct page
*page
,
779 void *object
, u8 val
)
782 u8
*endobject
= object
+ s
->objsize
;
784 if (s
->flags
& SLAB_RED_ZONE
) {
785 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
786 endobject
, val
, s
->inuse
- s
->objsize
))
789 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
790 check_bytes_and_report(s
, page
, p
, "Alignment padding",
791 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
795 if (s
->flags
& SLAB_POISON
) {
796 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
797 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
798 POISON_FREE
, s
->objsize
- 1) ||
799 !check_bytes_and_report(s
, page
, p
, "Poison",
800 p
+ s
->objsize
- 1, POISON_END
, 1)))
803 * check_pad_bytes cleans up on its own.
805 check_pad_bytes(s
, page
, p
);
808 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
810 * Object and freepointer overlap. Cannot check
811 * freepointer while object is allocated.
815 /* Check free pointer validity */
816 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
817 object_err(s
, page
, p
, "Freepointer corrupt");
819 * No choice but to zap it and thus lose the remainder
820 * of the free objects in this slab. May cause
821 * another error because the object count is now wrong.
823 set_freepointer(s
, p
, NULL
);
829 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
833 VM_BUG_ON(!irqs_disabled());
835 if (!PageSlab(page
)) {
836 slab_err(s
, page
, "Not a valid slab page");
840 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
841 if (page
->objects
> maxobj
) {
842 slab_err(s
, page
, "objects %u > max %u",
843 s
->name
, page
->objects
, maxobj
);
846 if (page
->inuse
> page
->objects
) {
847 slab_err(s
, page
, "inuse %u > max %u",
848 s
->name
, page
->inuse
, page
->objects
);
851 /* Slab_pad_check fixes things up after itself */
852 slab_pad_check(s
, page
);
857 * Determine if a certain object on a page is on the freelist. Must hold the
858 * slab lock to guarantee that the chains are in a consistent state.
860 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
865 unsigned long max_objects
;
868 while (fp
&& nr
<= page
->objects
) {
871 if (!check_valid_pointer(s
, page
, fp
)) {
873 object_err(s
, page
, object
,
874 "Freechain corrupt");
875 set_freepointer(s
, object
, NULL
);
878 slab_err(s
, page
, "Freepointer corrupt");
879 page
->freelist
= NULL
;
880 page
->inuse
= page
->objects
;
881 slab_fix(s
, "Freelist cleared");
887 fp
= get_freepointer(s
, object
);
891 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
892 if (max_objects
> MAX_OBJS_PER_PAGE
)
893 max_objects
= MAX_OBJS_PER_PAGE
;
895 if (page
->objects
!= max_objects
) {
896 slab_err(s
, page
, "Wrong number of objects. Found %d but "
897 "should be %d", page
->objects
, max_objects
);
898 page
->objects
= max_objects
;
899 slab_fix(s
, "Number of objects adjusted.");
901 if (page
->inuse
!= page
->objects
- nr
) {
902 slab_err(s
, page
, "Wrong object count. Counter is %d but "
903 "counted were %d", page
->inuse
, page
->objects
- nr
);
904 page
->inuse
= page
->objects
- nr
;
905 slab_fix(s
, "Object count adjusted.");
907 return search
== NULL
;
910 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
913 if (s
->flags
& SLAB_TRACE
) {
914 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
916 alloc
? "alloc" : "free",
921 print_section("Object ", (void *)object
, s
->objsize
);
928 * Hooks for other subsystems that check memory allocations. In a typical
929 * production configuration these hooks all should produce no code at all.
931 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
933 flags
&= gfp_allowed_mask
;
934 lockdep_trace_alloc(flags
);
935 might_sleep_if(flags
& __GFP_WAIT
);
937 return should_failslab(s
->objsize
, flags
, s
->flags
);
940 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
942 flags
&= gfp_allowed_mask
;
943 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
944 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
947 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
949 kmemleak_free_recursive(x
, s
->flags
);
952 * Trouble is that we may no longer disable interupts in the fast path
953 * So in order to make the debug calls that expect irqs to be
954 * disabled we need to disable interrupts temporarily.
956 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
960 local_irq_save(flags
);
961 kmemcheck_slab_free(s
, x
, s
->objsize
);
962 debug_check_no_locks_freed(x
, s
->objsize
);
963 local_irq_restore(flags
);
966 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
967 debug_check_no_obj_freed(x
, s
->objsize
);
971 * Tracking of fully allocated slabs for debugging purposes.
973 * list_lock must be held.
975 static void add_full(struct kmem_cache
*s
,
976 struct kmem_cache_node
*n
, struct page
*page
)
978 if (!(s
->flags
& SLAB_STORE_USER
))
981 list_add(&page
->lru
, &n
->full
);
985 * list_lock must be held.
987 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
989 if (!(s
->flags
& SLAB_STORE_USER
))
992 list_del(&page
->lru
);
995 /* Tracking of the number of slabs for debugging purposes */
996 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
998 struct kmem_cache_node
*n
= get_node(s
, node
);
1000 return atomic_long_read(&n
->nr_slabs
);
1003 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1005 return atomic_long_read(&n
->nr_slabs
);
1008 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1010 struct kmem_cache_node
*n
= get_node(s
, node
);
1013 * May be called early in order to allocate a slab for the
1014 * kmem_cache_node structure. Solve the chicken-egg
1015 * dilemma by deferring the increment of the count during
1016 * bootstrap (see early_kmem_cache_node_alloc).
1019 atomic_long_inc(&n
->nr_slabs
);
1020 atomic_long_add(objects
, &n
->total_objects
);
1023 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1025 struct kmem_cache_node
*n
= get_node(s
, node
);
1027 atomic_long_dec(&n
->nr_slabs
);
1028 atomic_long_sub(objects
, &n
->total_objects
);
1031 /* Object debug checks for alloc/free paths */
1032 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1035 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1038 init_object(s
, object
, SLUB_RED_INACTIVE
);
1039 init_tracking(s
, object
);
1042 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
1043 void *object
, unsigned long addr
)
1045 if (!check_slab(s
, page
))
1048 if (!check_valid_pointer(s
, page
, object
)) {
1049 object_err(s
, page
, object
, "Freelist Pointer check fails");
1053 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1056 /* Success perform special debug activities for allocs */
1057 if (s
->flags
& SLAB_STORE_USER
)
1058 set_track(s
, object
, TRACK_ALLOC
, addr
);
1059 trace(s
, page
, object
, 1);
1060 init_object(s
, object
, SLUB_RED_ACTIVE
);
1064 if (PageSlab(page
)) {
1066 * If this is a slab page then lets do the best we can
1067 * to avoid issues in the future. Marking all objects
1068 * as used avoids touching the remaining objects.
1070 slab_fix(s
, "Marking all objects used");
1071 page
->inuse
= page
->objects
;
1072 page
->freelist
= NULL
;
1077 static noinline
int free_debug_processing(struct kmem_cache
*s
,
1078 struct page
*page
, void *object
, unsigned long addr
)
1080 unsigned long flags
;
1083 local_irq_save(flags
);
1086 if (!check_slab(s
, page
))
1089 if (!check_valid_pointer(s
, page
, object
)) {
1090 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1094 if (on_freelist(s
, page
, object
)) {
1095 object_err(s
, page
, object
, "Object already free");
1099 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1102 if (unlikely(s
!= page
->slab
)) {
1103 if (!PageSlab(page
)) {
1104 slab_err(s
, page
, "Attempt to free object(0x%p) "
1105 "outside of slab", object
);
1106 } else if (!page
->slab
) {
1108 "SLUB <none>: no slab for object 0x%p.\n",
1112 object_err(s
, page
, object
,
1113 "page slab pointer corrupt.");
1117 if (s
->flags
& SLAB_STORE_USER
)
1118 set_track(s
, object
, TRACK_FREE
, addr
);
1119 trace(s
, page
, object
, 0);
1120 init_object(s
, object
, SLUB_RED_INACTIVE
);
1124 local_irq_restore(flags
);
1128 slab_fix(s
, "Object at 0x%p not freed", object
);
1132 static int __init
setup_slub_debug(char *str
)
1134 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1135 if (*str
++ != '=' || !*str
)
1137 * No options specified. Switch on full debugging.
1143 * No options but restriction on slabs. This means full
1144 * debugging for slabs matching a pattern.
1148 if (tolower(*str
) == 'o') {
1150 * Avoid enabling debugging on caches if its minimum order
1151 * would increase as a result.
1153 disable_higher_order_debug
= 1;
1160 * Switch off all debugging measures.
1165 * Determine which debug features should be switched on
1167 for (; *str
&& *str
!= ','; str
++) {
1168 switch (tolower(*str
)) {
1170 slub_debug
|= SLAB_DEBUG_FREE
;
1173 slub_debug
|= SLAB_RED_ZONE
;
1176 slub_debug
|= SLAB_POISON
;
1179 slub_debug
|= SLAB_STORE_USER
;
1182 slub_debug
|= SLAB_TRACE
;
1185 slub_debug
|= SLAB_FAILSLAB
;
1188 printk(KERN_ERR
"slub_debug option '%c' "
1189 "unknown. skipped\n", *str
);
1195 slub_debug_slabs
= str
+ 1;
1200 __setup("slub_debug", setup_slub_debug
);
1202 static unsigned long kmem_cache_flags(unsigned long objsize
,
1203 unsigned long flags
, const char *name
,
1204 void (*ctor
)(void *))
1207 * Enable debugging if selected on the kernel commandline.
1209 if (slub_debug
&& (!slub_debug_slabs
||
1210 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1211 flags
|= slub_debug
;
1216 static inline void setup_object_debug(struct kmem_cache
*s
,
1217 struct page
*page
, void *object
) {}
1219 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1220 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1222 static inline int free_debug_processing(struct kmem_cache
*s
,
1223 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1225 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1227 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1228 void *object
, u8 val
) { return 1; }
1229 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1230 struct page
*page
) {}
1231 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1232 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1233 unsigned long flags
, const char *name
,
1234 void (*ctor
)(void *))
1238 #define slub_debug 0
1240 #define disable_higher_order_debug 0
1242 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1244 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1246 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1248 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1251 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1254 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1257 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1259 #endif /* CONFIG_SLUB_DEBUG */
1262 * Slab allocation and freeing
1264 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1265 struct kmem_cache_order_objects oo
)
1267 int order
= oo_order(oo
);
1269 flags
|= __GFP_NOTRACK
;
1271 if (node
== NUMA_NO_NODE
)
1272 return alloc_pages(flags
, order
);
1274 return alloc_pages_exact_node(node
, flags
, order
);
1277 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1280 struct kmem_cache_order_objects oo
= s
->oo
;
1283 flags
&= gfp_allowed_mask
;
1285 if (flags
& __GFP_WAIT
)
1288 flags
|= s
->allocflags
;
1291 * Let the initial higher-order allocation fail under memory pressure
1292 * so we fall-back to the minimum order allocation.
1294 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1296 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1297 if (unlikely(!page
)) {
1300 * Allocation may have failed due to fragmentation.
1301 * Try a lower order alloc if possible
1303 page
= alloc_slab_page(flags
, node
, oo
);
1306 stat(s
, ORDER_FALLBACK
);
1309 if (flags
& __GFP_WAIT
)
1310 local_irq_disable();
1315 if (kmemcheck_enabled
1316 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1317 int pages
= 1 << oo_order(oo
);
1319 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1322 * Objects from caches that have a constructor don't get
1323 * cleared when they're allocated, so we need to do it here.
1326 kmemcheck_mark_uninitialized_pages(page
, pages
);
1328 kmemcheck_mark_unallocated_pages(page
, pages
);
1331 page
->objects
= oo_objects(oo
);
1332 mod_zone_page_state(page_zone(page
),
1333 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1334 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1340 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1343 setup_object_debug(s
, page
, object
);
1344 if (unlikely(s
->ctor
))
1348 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1355 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1357 page
= allocate_slab(s
,
1358 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1362 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1364 page
->flags
|= 1 << PG_slab
;
1366 start
= page_address(page
);
1368 if (unlikely(s
->flags
& SLAB_POISON
))
1369 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1372 for_each_object(p
, s
, start
, page
->objects
) {
1373 setup_object(s
, page
, last
);
1374 set_freepointer(s
, last
, p
);
1377 setup_object(s
, page
, last
);
1378 set_freepointer(s
, last
, NULL
);
1380 page
->freelist
= start
;
1381 page
->inuse
= page
->objects
;
1387 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1389 int order
= compound_order(page
);
1390 int pages
= 1 << order
;
1392 if (kmem_cache_debug(s
)) {
1395 slab_pad_check(s
, page
);
1396 for_each_object(p
, s
, page_address(page
),
1398 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1401 kmemcheck_free_shadow(page
, compound_order(page
));
1403 mod_zone_page_state(page_zone(page
),
1404 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1405 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1408 __ClearPageSlab(page
);
1409 reset_page_mapcount(page
);
1410 if (current
->reclaim_state
)
1411 current
->reclaim_state
->reclaimed_slab
+= pages
;
1412 __free_pages(page
, order
);
1415 #define need_reserve_slab_rcu \
1416 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1418 static void rcu_free_slab(struct rcu_head
*h
)
1422 if (need_reserve_slab_rcu
)
1423 page
= virt_to_head_page(h
);
1425 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1427 __free_slab(page
->slab
, page
);
1430 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1432 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1433 struct rcu_head
*head
;
1435 if (need_reserve_slab_rcu
) {
1436 int order
= compound_order(page
);
1437 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1439 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1440 head
= page_address(page
) + offset
;
1443 * RCU free overloads the RCU head over the LRU
1445 head
= (void *)&page
->lru
;
1448 call_rcu(head
, rcu_free_slab
);
1450 __free_slab(s
, page
);
1453 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1455 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1460 * Management of partially allocated slabs.
1462 * list_lock must be held.
1464 static inline void add_partial(struct kmem_cache_node
*n
,
1465 struct page
*page
, int tail
)
1468 if (tail
== DEACTIVATE_TO_TAIL
)
1469 list_add_tail(&page
->lru
, &n
->partial
);
1471 list_add(&page
->lru
, &n
->partial
);
1475 * list_lock must be held.
1477 static inline void remove_partial(struct kmem_cache_node
*n
,
1480 list_del(&page
->lru
);
1485 * Lock slab, remove from the partial list and put the object into the
1488 * Returns a list of objects or NULL if it fails.
1490 * Must hold list_lock.
1492 static inline void *acquire_slab(struct kmem_cache
*s
,
1493 struct kmem_cache_node
*n
, struct page
*page
,
1497 unsigned long counters
;
1501 * Zap the freelist and set the frozen bit.
1502 * The old freelist is the list of objects for the
1503 * per cpu allocation list.
1506 freelist
= page
->freelist
;
1507 counters
= page
->counters
;
1508 new.counters
= counters
;
1510 new.inuse
= page
->objects
;
1512 VM_BUG_ON(new.frozen
);
1515 } while (!__cmpxchg_double_slab(s
, page
,
1518 "lock and freeze"));
1520 remove_partial(n
, page
);
1524 static int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1527 * Try to allocate a partial slab from a specific node.
1529 static void *get_partial_node(struct kmem_cache
*s
,
1530 struct kmem_cache_node
*n
, struct kmem_cache_cpu
*c
)
1532 struct page
*page
, *page2
;
1533 void *object
= NULL
;
1536 * Racy check. If we mistakenly see no partial slabs then we
1537 * just allocate an empty slab. If we mistakenly try to get a
1538 * partial slab and there is none available then get_partials()
1541 if (!n
|| !n
->nr_partial
)
1544 spin_lock(&n
->list_lock
);
1545 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1546 void *t
= acquire_slab(s
, n
, page
, object
== NULL
);
1554 c
->node
= page_to_nid(page
);
1555 stat(s
, ALLOC_FROM_PARTIAL
);
1557 available
= page
->objects
- page
->inuse
;
1560 available
= put_cpu_partial(s
, page
, 0);
1562 if (kmem_cache_debug(s
) || available
> s
->cpu_partial
/ 2)
1566 spin_unlock(&n
->list_lock
);
1571 * Get a page from somewhere. Search in increasing NUMA distances.
1573 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1574 struct kmem_cache_cpu
*c
)
1577 struct zonelist
*zonelist
;
1580 enum zone_type high_zoneidx
= gfp_zone(flags
);
1584 * The defrag ratio allows a configuration of the tradeoffs between
1585 * inter node defragmentation and node local allocations. A lower
1586 * defrag_ratio increases the tendency to do local allocations
1587 * instead of attempting to obtain partial slabs from other nodes.
1589 * If the defrag_ratio is set to 0 then kmalloc() always
1590 * returns node local objects. If the ratio is higher then kmalloc()
1591 * may return off node objects because partial slabs are obtained
1592 * from other nodes and filled up.
1594 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1595 * defrag_ratio = 1000) then every (well almost) allocation will
1596 * first attempt to defrag slab caches on other nodes. This means
1597 * scanning over all nodes to look for partial slabs which may be
1598 * expensive if we do it every time we are trying to find a slab
1599 * with available objects.
1601 if (!s
->remote_node_defrag_ratio
||
1602 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1606 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1607 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1608 struct kmem_cache_node
*n
;
1610 n
= get_node(s
, zone_to_nid(zone
));
1612 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1613 n
->nr_partial
> s
->min_partial
) {
1614 object
= get_partial_node(s
, n
, c
);
1627 * Get a partial page, lock it and return it.
1629 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1630 struct kmem_cache_cpu
*c
)
1633 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1635 object
= get_partial_node(s
, get_node(s
, searchnode
), c
);
1636 if (object
|| node
!= NUMA_NO_NODE
)
1639 return get_any_partial(s
, flags
, c
);
1642 #ifdef CONFIG_PREEMPT
1644 * Calculate the next globally unique transaction for disambiguiation
1645 * during cmpxchg. The transactions start with the cpu number and are then
1646 * incremented by CONFIG_NR_CPUS.
1648 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1651 * No preemption supported therefore also no need to check for
1657 static inline unsigned long next_tid(unsigned long tid
)
1659 return tid
+ TID_STEP
;
1662 static inline unsigned int tid_to_cpu(unsigned long tid
)
1664 return tid
% TID_STEP
;
1667 static inline unsigned long tid_to_event(unsigned long tid
)
1669 return tid
/ TID_STEP
;
1672 static inline unsigned int init_tid(int cpu
)
1677 static inline void note_cmpxchg_failure(const char *n
,
1678 const struct kmem_cache
*s
, unsigned long tid
)
1680 #ifdef SLUB_DEBUG_CMPXCHG
1681 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1683 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1685 #ifdef CONFIG_PREEMPT
1686 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1687 printk("due to cpu change %d -> %d\n",
1688 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1691 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1692 printk("due to cpu running other code. Event %ld->%ld\n",
1693 tid_to_event(tid
), tid_to_event(actual_tid
));
1695 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1696 actual_tid
, tid
, next_tid(tid
));
1698 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1701 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1705 for_each_possible_cpu(cpu
)
1706 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1710 * Remove the cpu slab
1712 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1714 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1715 struct page
*page
= c
->page
;
1716 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1718 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1721 int tail
= DEACTIVATE_TO_HEAD
;
1725 if (page
->freelist
) {
1726 stat(s
, DEACTIVATE_REMOTE_FREES
);
1727 tail
= DEACTIVATE_TO_TAIL
;
1730 c
->tid
= next_tid(c
->tid
);
1732 freelist
= c
->freelist
;
1736 * Stage one: Free all available per cpu objects back
1737 * to the page freelist while it is still frozen. Leave the
1740 * There is no need to take the list->lock because the page
1743 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1745 unsigned long counters
;
1748 prior
= page
->freelist
;
1749 counters
= page
->counters
;
1750 set_freepointer(s
, freelist
, prior
);
1751 new.counters
= counters
;
1753 VM_BUG_ON(!new.frozen
);
1755 } while (!__cmpxchg_double_slab(s
, page
,
1757 freelist
, new.counters
,
1758 "drain percpu freelist"));
1760 freelist
= nextfree
;
1764 * Stage two: Ensure that the page is unfrozen while the
1765 * list presence reflects the actual number of objects
1768 * We setup the list membership and then perform a cmpxchg
1769 * with the count. If there is a mismatch then the page
1770 * is not unfrozen but the page is on the wrong list.
1772 * Then we restart the process which may have to remove
1773 * the page from the list that we just put it on again
1774 * because the number of objects in the slab may have
1779 old
.freelist
= page
->freelist
;
1780 old
.counters
= page
->counters
;
1781 VM_BUG_ON(!old
.frozen
);
1783 /* Determine target state of the slab */
1784 new.counters
= old
.counters
;
1787 set_freepointer(s
, freelist
, old
.freelist
);
1788 new.freelist
= freelist
;
1790 new.freelist
= old
.freelist
;
1794 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1796 else if (new.freelist
) {
1801 * Taking the spinlock removes the possiblity
1802 * that acquire_slab() will see a slab page that
1805 spin_lock(&n
->list_lock
);
1809 if (kmem_cache_debug(s
) && !lock
) {
1812 * This also ensures that the scanning of full
1813 * slabs from diagnostic functions will not see
1816 spin_lock(&n
->list_lock
);
1824 remove_partial(n
, page
);
1826 else if (l
== M_FULL
)
1828 remove_full(s
, page
);
1830 if (m
== M_PARTIAL
) {
1832 add_partial(n
, page
, tail
);
1835 } else if (m
== M_FULL
) {
1837 stat(s
, DEACTIVATE_FULL
);
1838 add_full(s
, n
, page
);
1844 if (!__cmpxchg_double_slab(s
, page
,
1845 old
.freelist
, old
.counters
,
1846 new.freelist
, new.counters
,
1851 spin_unlock(&n
->list_lock
);
1854 stat(s
, DEACTIVATE_EMPTY
);
1855 discard_slab(s
, page
);
1860 /* Unfreeze all the cpu partial slabs */
1861 static void unfreeze_partials(struct kmem_cache
*s
)
1863 struct kmem_cache_node
*n
= NULL
;
1864 struct kmem_cache_cpu
*c
= this_cpu_ptr(s
->cpu_slab
);
1865 struct page
*page
, *discard_page
= NULL
;
1867 while ((page
= c
->partial
)) {
1868 enum slab_modes
{ M_PARTIAL
, M_FREE
};
1869 enum slab_modes l
, m
;
1873 c
->partial
= page
->next
;
1878 old
.freelist
= page
->freelist
;
1879 old
.counters
= page
->counters
;
1880 VM_BUG_ON(!old
.frozen
);
1882 new.counters
= old
.counters
;
1883 new.freelist
= old
.freelist
;
1887 if (!new.inuse
&& (!n
|| n
->nr_partial
> s
->min_partial
))
1890 struct kmem_cache_node
*n2
= get_node(s
,
1896 spin_unlock(&n
->list_lock
);
1899 spin_lock(&n
->list_lock
);
1904 if (l
== M_PARTIAL
) {
1905 remove_partial(n
, page
);
1906 stat(s
, FREE_REMOVE_PARTIAL
);
1908 add_partial(n
, page
,
1909 DEACTIVATE_TO_TAIL
);
1910 stat(s
, FREE_ADD_PARTIAL
);
1916 } while (!cmpxchg_double_slab(s
, page
,
1917 old
.freelist
, old
.counters
,
1918 new.freelist
, new.counters
,
1919 "unfreezing slab"));
1922 page
->next
= discard_page
;
1923 discard_page
= page
;
1928 spin_unlock(&n
->list_lock
);
1930 while (discard_page
) {
1931 page
= discard_page
;
1932 discard_page
= discard_page
->next
;
1934 stat(s
, DEACTIVATE_EMPTY
);
1935 discard_slab(s
, page
);
1941 * Put a page that was just frozen (in __slab_free) into a partial page
1942 * slot if available. This is done without interrupts disabled and without
1943 * preemption disabled. The cmpxchg is racy and may put the partial page
1944 * onto a random cpus partial slot.
1946 * If we did not find a slot then simply move all the partials to the
1947 * per node partial list.
1949 int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1951 struct page
*oldpage
;
1958 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1961 pobjects
= oldpage
->pobjects
;
1962 pages
= oldpage
->pages
;
1963 if (drain
&& pobjects
> s
->cpu_partial
) {
1964 unsigned long flags
;
1966 * partial array is full. Move the existing
1967 * set to the per node partial list.
1969 local_irq_save(flags
);
1970 unfreeze_partials(s
);
1971 local_irq_restore(flags
);
1978 pobjects
+= page
->objects
- page
->inuse
;
1980 page
->pages
= pages
;
1981 page
->pobjects
= pobjects
;
1982 page
->next
= oldpage
;
1984 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
) != oldpage
);
1985 stat(s
, CPU_PARTIAL_FREE
);
1989 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1991 stat(s
, CPUSLAB_FLUSH
);
1992 deactivate_slab(s
, c
);
1998 * Called from IPI handler with interrupts disabled.
2000 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2002 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2008 unfreeze_partials(s
);
2012 static void flush_cpu_slab(void *d
)
2014 struct kmem_cache
*s
= d
;
2016 __flush_cpu_slab(s
, smp_processor_id());
2019 static void flush_all(struct kmem_cache
*s
)
2021 on_each_cpu(flush_cpu_slab
, s
, 1);
2025 * Check if the objects in a per cpu structure fit numa
2026 * locality expectations.
2028 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
2031 if (node
!= NUMA_NO_NODE
&& c
->node
!= 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
->objsize
,
2075 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2077 if (oo_order(s
->min
) > get_order(s
->objsize
))
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
;
2105 struct page
*page
= new_slab(s
, flags
, node
);
2108 c
= __this_cpu_ptr(s
->cpu_slab
);
2113 * No other reference to the page yet so we can
2114 * muck around with it freely without cmpxchg
2116 object
= page
->freelist
;
2117 page
->freelist
= NULL
;
2119 stat(s
, ALLOC_SLAB
);
2120 c
->node
= page_to_nid(page
);
2130 * Check the page->freelist of a page and either transfer the freelist to the per cpu freelist
2131 * or deactivate the page.
2133 * The page is still frozen if the return value is not NULL.
2135 * If this function returns NULL then the page has been unfrozen.
2137 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2140 unsigned long counters
;
2144 freelist
= page
->freelist
;
2145 counters
= page
->counters
;
2146 new.counters
= counters
;
2147 VM_BUG_ON(!new.frozen
);
2149 new.inuse
= page
->objects
;
2150 new.frozen
= freelist
!= NULL
;
2152 } while (!cmpxchg_double_slab(s
, page
,
2161 * Slow path. The lockless freelist is empty or we need to perform
2164 * Processing is still very fast if new objects have been freed to the
2165 * regular freelist. In that case we simply take over the regular freelist
2166 * as the lockless freelist and zap the regular freelist.
2168 * If that is not working then we fall back to the partial lists. We take the
2169 * first element of the freelist as the object to allocate now and move the
2170 * rest of the freelist to the lockless freelist.
2172 * And if we were unable to get a new slab from the partial slab lists then
2173 * we need to allocate a new slab. This is the slowest path since it involves
2174 * a call to the page allocator and the setup of a new slab.
2176 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2177 unsigned long addr
, struct kmem_cache_cpu
*c
)
2180 unsigned long flags
;
2182 local_irq_save(flags
);
2183 #ifdef CONFIG_PREEMPT
2185 * We may have been preempted and rescheduled on a different
2186 * cpu before disabling interrupts. Need to reload cpu area
2189 c
= this_cpu_ptr(s
->cpu_slab
);
2195 if (unlikely(!node_match(c
, node
))) {
2196 stat(s
, ALLOC_NODE_MISMATCH
);
2197 deactivate_slab(s
, c
);
2201 /* must check again c->freelist in case of cpu migration or IRQ */
2202 object
= c
->freelist
;
2206 stat(s
, ALLOC_SLOWPATH
);
2208 object
= get_freelist(s
, c
->page
);
2212 stat(s
, DEACTIVATE_BYPASS
);
2216 stat(s
, ALLOC_REFILL
);
2219 c
->freelist
= get_freepointer(s
, object
);
2220 c
->tid
= next_tid(c
->tid
);
2221 local_irq_restore(flags
);
2227 c
->page
= c
->partial
;
2228 c
->partial
= c
->page
->next
;
2229 c
->node
= page_to_nid(c
->page
);
2230 stat(s
, CPU_PARTIAL_ALLOC
);
2235 /* Then do expensive stuff like retrieving pages from the partial lists */
2236 object
= get_partial(s
, gfpflags
, node
, c
);
2238 if (unlikely(!object
)) {
2240 object
= new_slab_objects(s
, gfpflags
, node
, &c
);
2242 if (unlikely(!object
)) {
2243 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2244 slab_out_of_memory(s
, gfpflags
, node
);
2246 local_irq_restore(flags
);
2251 if (likely(!kmem_cache_debug(s
)))
2254 /* Only entered in the debug case */
2255 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
2256 goto new_slab
; /* Slab failed checks. Next slab needed */
2258 c
->freelist
= get_freepointer(s
, object
);
2259 deactivate_slab(s
, c
);
2260 c
->node
= NUMA_NO_NODE
;
2261 local_irq_restore(flags
);
2266 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2267 * have the fastpath folded into their functions. So no function call
2268 * overhead for requests that can be satisfied on the fastpath.
2270 * The fastpath works by first checking if the lockless freelist can be used.
2271 * If not then __slab_alloc is called for slow processing.
2273 * Otherwise we can simply pick the next object from the lockless free list.
2275 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2276 gfp_t gfpflags
, int node
, unsigned long addr
)
2279 struct kmem_cache_cpu
*c
;
2282 if (slab_pre_alloc_hook(s
, gfpflags
))
2288 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2289 * enabled. We may switch back and forth between cpus while
2290 * reading from one cpu area. That does not matter as long
2291 * as we end up on the original cpu again when doing the cmpxchg.
2293 c
= __this_cpu_ptr(s
->cpu_slab
);
2296 * The transaction ids are globally unique per cpu and per operation on
2297 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2298 * occurs on the right processor and that there was no operation on the
2299 * linked list in between.
2304 object
= c
->freelist
;
2305 if (unlikely(!object
|| !node_match(c
, node
)))
2307 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2311 * The cmpxchg will only match if there was no additional
2312 * operation and if we are on the right processor.
2314 * The cmpxchg does the following atomically (without lock semantics!)
2315 * 1. Relocate first pointer to the current per cpu area.
2316 * 2. Verify that tid and freelist have not been changed
2317 * 3. If they were not changed replace tid and freelist
2319 * Since this is without lock semantics the protection is only against
2320 * code executing on this cpu *not* from access by other cpus.
2322 if (unlikely(!this_cpu_cmpxchg_double(
2323 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2325 get_freepointer_safe(s
, object
), next_tid(tid
)))) {
2327 note_cmpxchg_failure("slab_alloc", s
, tid
);
2330 stat(s
, ALLOC_FASTPATH
);
2333 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2334 memset(object
, 0, s
->objsize
);
2336 slab_post_alloc_hook(s
, gfpflags
, object
);
2341 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2343 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2345 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
2349 EXPORT_SYMBOL(kmem_cache_alloc
);
2351 #ifdef CONFIG_TRACING
2352 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2354 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2355 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2358 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2360 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2362 void *ret
= kmalloc_order(size
, flags
, order
);
2363 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2366 EXPORT_SYMBOL(kmalloc_order_trace
);
2370 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2372 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2374 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2375 s
->objsize
, s
->size
, gfpflags
, node
);
2379 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2381 #ifdef CONFIG_TRACING
2382 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2384 int node
, size_t size
)
2386 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2388 trace_kmalloc_node(_RET_IP_
, ret
,
2389 size
, s
->size
, gfpflags
, node
);
2392 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2397 * Slow patch handling. This may still be called frequently since objects
2398 * have a longer lifetime than the cpu slabs in most processing loads.
2400 * So we still attempt to reduce cache line usage. Just take the slab
2401 * lock and free the item. If there is no additional partial page
2402 * handling required then we can return immediately.
2404 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2405 void *x
, unsigned long addr
)
2408 void **object
= (void *)x
;
2412 unsigned long counters
;
2413 struct kmem_cache_node
*n
= NULL
;
2414 unsigned long uninitialized_var(flags
);
2416 stat(s
, FREE_SLOWPATH
);
2418 if (kmem_cache_debug(s
) && !free_debug_processing(s
, page
, x
, addr
))
2422 prior
= page
->freelist
;
2423 counters
= page
->counters
;
2424 set_freepointer(s
, object
, prior
);
2425 new.counters
= counters
;
2426 was_frozen
= new.frozen
;
2428 if ((!new.inuse
|| !prior
) && !was_frozen
&& !n
) {
2430 if (!kmem_cache_debug(s
) && !prior
)
2433 * Slab was on no list before and will be partially empty
2434 * We can defer the list move and instead freeze it.
2438 else { /* Needs to be taken off a list */
2440 n
= get_node(s
, page_to_nid(page
));
2442 * Speculatively acquire the list_lock.
2443 * If the cmpxchg does not succeed then we may
2444 * drop the list_lock without any processing.
2446 * Otherwise the list_lock will synchronize with
2447 * other processors updating the list of slabs.
2449 spin_lock_irqsave(&n
->list_lock
, flags
);
2455 } while (!cmpxchg_double_slab(s
, page
,
2457 object
, new.counters
,
2463 * If we just froze the page then put it onto the
2464 * per cpu partial list.
2466 if (new.frozen
&& !was_frozen
)
2467 put_cpu_partial(s
, page
, 1);
2470 * The list lock was not taken therefore no list
2471 * activity can be necessary.
2474 stat(s
, FREE_FROZEN
);
2479 * was_frozen may have been set after we acquired the list_lock in
2480 * an earlier loop. So we need to check it here again.
2483 stat(s
, FREE_FROZEN
);
2485 if (unlikely(!inuse
&& n
->nr_partial
> s
->min_partial
))
2489 * Objects left in the slab. If it was not on the partial list before
2492 if (unlikely(!prior
)) {
2493 remove_full(s
, page
);
2494 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2495 stat(s
, FREE_ADD_PARTIAL
);
2498 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2504 * Slab on the partial list.
2506 remove_partial(n
, page
);
2507 stat(s
, FREE_REMOVE_PARTIAL
);
2509 /* Slab must be on the full list */
2510 remove_full(s
, page
);
2512 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2514 discard_slab(s
, page
);
2518 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2519 * can perform fastpath freeing without additional function calls.
2521 * The fastpath is only possible if we are freeing to the current cpu slab
2522 * of this processor. This typically the case if we have just allocated
2525 * If fastpath is not possible then fall back to __slab_free where we deal
2526 * with all sorts of special processing.
2528 static __always_inline
void slab_free(struct kmem_cache
*s
,
2529 struct page
*page
, void *x
, unsigned long addr
)
2531 void **object
= (void *)x
;
2532 struct kmem_cache_cpu
*c
;
2535 slab_free_hook(s
, x
);
2539 * Determine the currently cpus per cpu slab.
2540 * The cpu may change afterward. However that does not matter since
2541 * data is retrieved via this pointer. If we are on the same cpu
2542 * during the cmpxchg then the free will succedd.
2544 c
= __this_cpu_ptr(s
->cpu_slab
);
2549 if (likely(page
== c
->page
)) {
2550 set_freepointer(s
, object
, c
->freelist
);
2552 if (unlikely(!this_cpu_cmpxchg_double(
2553 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2555 object
, next_tid(tid
)))) {
2557 note_cmpxchg_failure("slab_free", s
, tid
);
2560 stat(s
, FREE_FASTPATH
);
2562 __slab_free(s
, page
, x
, addr
);
2566 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2570 page
= virt_to_head_page(x
);
2572 slab_free(s
, page
, x
, _RET_IP_
);
2574 trace_kmem_cache_free(_RET_IP_
, x
);
2576 EXPORT_SYMBOL(kmem_cache_free
);
2579 * Object placement in a slab is made very easy because we always start at
2580 * offset 0. If we tune the size of the object to the alignment then we can
2581 * get the required alignment by putting one properly sized object after
2584 * Notice that the allocation order determines the sizes of the per cpu
2585 * caches. Each processor has always one slab available for allocations.
2586 * Increasing the allocation order reduces the number of times that slabs
2587 * must be moved on and off the partial lists and is therefore a factor in
2592 * Mininum / Maximum order of slab pages. This influences locking overhead
2593 * and slab fragmentation. A higher order reduces the number of partial slabs
2594 * and increases the number of allocations possible without having to
2595 * take the list_lock.
2597 static int slub_min_order
;
2598 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2599 static int slub_min_objects
;
2602 * Merge control. If this is set then no merging of slab caches will occur.
2603 * (Could be removed. This was introduced to pacify the merge skeptics.)
2605 static int slub_nomerge
;
2608 * Calculate the order of allocation given an slab object size.
2610 * The order of allocation has significant impact on performance and other
2611 * system components. Generally order 0 allocations should be preferred since
2612 * order 0 does not cause fragmentation in the page allocator. Larger objects
2613 * be problematic to put into order 0 slabs because there may be too much
2614 * unused space left. We go to a higher order if more than 1/16th of the slab
2617 * In order to reach satisfactory performance we must ensure that a minimum
2618 * number of objects is in one slab. Otherwise we may generate too much
2619 * activity on the partial lists which requires taking the list_lock. This is
2620 * less a concern for large slabs though which are rarely used.
2622 * slub_max_order specifies the order where we begin to stop considering the
2623 * number of objects in a slab as critical. If we reach slub_max_order then
2624 * we try to keep the page order as low as possible. So we accept more waste
2625 * of space in favor of a small page order.
2627 * Higher order allocations also allow the placement of more objects in a
2628 * slab and thereby reduce object handling overhead. If the user has
2629 * requested a higher mininum order then we start with that one instead of
2630 * the smallest order which will fit the object.
2632 static inline int slab_order(int size
, int min_objects
,
2633 int max_order
, int fract_leftover
, int reserved
)
2637 int min_order
= slub_min_order
;
2639 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2640 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2642 for (order
= max(min_order
,
2643 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2644 order
<= max_order
; order
++) {
2646 unsigned long slab_size
= PAGE_SIZE
<< order
;
2648 if (slab_size
< min_objects
* size
+ reserved
)
2651 rem
= (slab_size
- reserved
) % size
;
2653 if (rem
<= slab_size
/ fract_leftover
)
2661 static inline int calculate_order(int size
, int reserved
)
2669 * Attempt to find best configuration for a slab. This
2670 * works by first attempting to generate a layout with
2671 * the best configuration and backing off gradually.
2673 * First we reduce the acceptable waste in a slab. Then
2674 * we reduce the minimum objects required in a slab.
2676 min_objects
= slub_min_objects
;
2678 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2679 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2680 min_objects
= min(min_objects
, max_objects
);
2682 while (min_objects
> 1) {
2684 while (fraction
>= 4) {
2685 order
= slab_order(size
, min_objects
,
2686 slub_max_order
, fraction
, reserved
);
2687 if (order
<= slub_max_order
)
2695 * We were unable to place multiple objects in a slab. Now
2696 * lets see if we can place a single object there.
2698 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2699 if (order
<= slub_max_order
)
2703 * Doh this slab cannot be placed using slub_max_order.
2705 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2706 if (order
< MAX_ORDER
)
2712 * Figure out what the alignment of the objects will be.
2714 static unsigned long calculate_alignment(unsigned long flags
,
2715 unsigned long align
, unsigned long size
)
2718 * If the user wants hardware cache aligned objects then follow that
2719 * suggestion if the object is sufficiently large.
2721 * The hardware cache alignment cannot override the specified
2722 * alignment though. If that is greater then use it.
2724 if (flags
& SLAB_HWCACHE_ALIGN
) {
2725 unsigned long ralign
= cache_line_size();
2726 while (size
<= ralign
/ 2)
2728 align
= max(align
, ralign
);
2731 if (align
< ARCH_SLAB_MINALIGN
)
2732 align
= ARCH_SLAB_MINALIGN
;
2734 return ALIGN(align
, sizeof(void *));
2738 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2741 spin_lock_init(&n
->list_lock
);
2742 INIT_LIST_HEAD(&n
->partial
);
2743 #ifdef CONFIG_SLUB_DEBUG
2744 atomic_long_set(&n
->nr_slabs
, 0);
2745 atomic_long_set(&n
->total_objects
, 0);
2746 INIT_LIST_HEAD(&n
->full
);
2750 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2752 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2753 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2756 * Must align to double word boundary for the double cmpxchg
2757 * instructions to work; see __pcpu_double_call_return_bool().
2759 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2760 2 * sizeof(void *));
2765 init_kmem_cache_cpus(s
);
2770 static struct kmem_cache
*kmem_cache_node
;
2773 * No kmalloc_node yet so do it by hand. We know that this is the first
2774 * slab on the node for this slabcache. There are no concurrent accesses
2777 * Note that this function only works on the kmalloc_node_cache
2778 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2779 * memory on a fresh node that has no slab structures yet.
2781 static void early_kmem_cache_node_alloc(int node
)
2784 struct kmem_cache_node
*n
;
2786 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2788 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2791 if (page_to_nid(page
) != node
) {
2792 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2794 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2795 "in order to be able to continue\n");
2800 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2803 kmem_cache_node
->node
[node
] = n
;
2804 #ifdef CONFIG_SLUB_DEBUG
2805 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2806 init_tracking(kmem_cache_node
, n
);
2808 init_kmem_cache_node(n
, kmem_cache_node
);
2809 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2811 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2814 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2818 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2819 struct kmem_cache_node
*n
= s
->node
[node
];
2822 kmem_cache_free(kmem_cache_node
, n
);
2824 s
->node
[node
] = NULL
;
2828 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2832 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2833 struct kmem_cache_node
*n
;
2835 if (slab_state
== DOWN
) {
2836 early_kmem_cache_node_alloc(node
);
2839 n
= kmem_cache_alloc_node(kmem_cache_node
,
2843 free_kmem_cache_nodes(s
);
2848 init_kmem_cache_node(n
, s
);
2853 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2855 if (min
< MIN_PARTIAL
)
2857 else if (min
> MAX_PARTIAL
)
2859 s
->min_partial
= min
;
2863 * calculate_sizes() determines the order and the distribution of data within
2866 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2868 unsigned long flags
= s
->flags
;
2869 unsigned long size
= s
->objsize
;
2870 unsigned long align
= s
->align
;
2874 * Round up object size to the next word boundary. We can only
2875 * place the free pointer at word boundaries and this determines
2876 * the possible location of the free pointer.
2878 size
= ALIGN(size
, sizeof(void *));
2880 #ifdef CONFIG_SLUB_DEBUG
2882 * Determine if we can poison the object itself. If the user of
2883 * the slab may touch the object after free or before allocation
2884 * then we should never poison the object itself.
2886 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2888 s
->flags
|= __OBJECT_POISON
;
2890 s
->flags
&= ~__OBJECT_POISON
;
2894 * If we are Redzoning then check if there is some space between the
2895 * end of the object and the free pointer. If not then add an
2896 * additional word to have some bytes to store Redzone information.
2898 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2899 size
+= sizeof(void *);
2903 * With that we have determined the number of bytes in actual use
2904 * by the object. This is the potential offset to the free pointer.
2908 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2911 * Relocate free pointer after the object if it is not
2912 * permitted to overwrite the first word of the object on
2915 * This is the case if we do RCU, have a constructor or
2916 * destructor or are poisoning the objects.
2919 size
+= sizeof(void *);
2922 #ifdef CONFIG_SLUB_DEBUG
2923 if (flags
& SLAB_STORE_USER
)
2925 * Need to store information about allocs and frees after
2928 size
+= 2 * sizeof(struct track
);
2930 if (flags
& SLAB_RED_ZONE
)
2932 * Add some empty padding so that we can catch
2933 * overwrites from earlier objects rather than let
2934 * tracking information or the free pointer be
2935 * corrupted if a user writes before the start
2938 size
+= sizeof(void *);
2942 * Determine the alignment based on various parameters that the
2943 * user specified and the dynamic determination of cache line size
2946 align
= calculate_alignment(flags
, align
, s
->objsize
);
2950 * SLUB stores one object immediately after another beginning from
2951 * offset 0. In order to align the objects we have to simply size
2952 * each object to conform to the alignment.
2954 size
= ALIGN(size
, align
);
2956 if (forced_order
>= 0)
2957 order
= forced_order
;
2959 order
= calculate_order(size
, s
->reserved
);
2966 s
->allocflags
|= __GFP_COMP
;
2968 if (s
->flags
& SLAB_CACHE_DMA
)
2969 s
->allocflags
|= SLUB_DMA
;
2971 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2972 s
->allocflags
|= __GFP_RECLAIMABLE
;
2975 * Determine the number of objects per slab
2977 s
->oo
= oo_make(order
, size
, s
->reserved
);
2978 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
2979 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2982 return !!oo_objects(s
->oo
);
2986 static int kmem_cache_open(struct kmem_cache
*s
,
2987 const char *name
, size_t size
,
2988 size_t align
, unsigned long flags
,
2989 void (*ctor
)(void *))
2991 memset(s
, 0, kmem_size
);
2996 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2999 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3000 s
->reserved
= sizeof(struct rcu_head
);
3002 if (!calculate_sizes(s
, -1))
3004 if (disable_higher_order_debug
) {
3006 * Disable debugging flags that store metadata if the min slab
3009 if (get_order(s
->size
) > get_order(s
->objsize
)) {
3010 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3012 if (!calculate_sizes(s
, -1))
3017 #ifdef CONFIG_CMPXCHG_DOUBLE
3018 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3019 /* Enable fast mode */
3020 s
->flags
|= __CMPXCHG_DOUBLE
;
3024 * The larger the object size is, the more pages we want on the partial
3025 * list to avoid pounding the page allocator excessively.
3027 set_min_partial(s
, ilog2(s
->size
) / 2);
3030 * cpu_partial determined the maximum number of objects kept in the
3031 * per cpu partial lists of a processor.
3033 * Per cpu partial lists mainly contain slabs that just have one
3034 * object freed. If they are used for allocation then they can be
3035 * filled up again with minimal effort. The slab will never hit the
3036 * per node partial lists and therefore no locking will be required.
3038 * This setting also determines
3040 * A) The number of objects from per cpu partial slabs dumped to the
3041 * per node list when we reach the limit.
3042 * B) The number of objects in cpu partial slabs to extract from the
3043 * per node list when we run out of per cpu objects. We only fetch 50%
3044 * to keep some capacity around for frees.
3046 if (kmem_cache_debug(s
))
3048 else if (s
->size
>= PAGE_SIZE
)
3050 else if (s
->size
>= 1024)
3052 else if (s
->size
>= 256)
3053 s
->cpu_partial
= 13;
3055 s
->cpu_partial
= 30;
3059 s
->remote_node_defrag_ratio
= 1000;
3061 if (!init_kmem_cache_nodes(s
))
3064 if (alloc_kmem_cache_cpus(s
))
3067 free_kmem_cache_nodes(s
);
3069 if (flags
& SLAB_PANIC
)
3070 panic("Cannot create slab %s size=%lu realsize=%u "
3071 "order=%u offset=%u flags=%lx\n",
3072 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
3078 * Determine the size of a slab object
3080 unsigned int kmem_cache_size(struct kmem_cache
*s
)
3084 EXPORT_SYMBOL(kmem_cache_size
);
3086 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3089 #ifdef CONFIG_SLUB_DEBUG
3090 void *addr
= page_address(page
);
3092 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3093 sizeof(long), GFP_ATOMIC
);
3096 slab_err(s
, page
, "%s", text
);
3099 get_map(s
, page
, map
);
3100 for_each_object(p
, s
, addr
, page
->objects
) {
3102 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3103 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3105 print_tracking(s
, p
);
3114 * Attempt to free all partial slabs on a node.
3115 * This is called from kmem_cache_close(). We must be the last thread
3116 * using the cache and therefore we do not need to lock anymore.
3118 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3120 struct page
*page
, *h
;
3122 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3124 remove_partial(n
, page
);
3125 discard_slab(s
, page
);
3127 list_slab_objects(s
, page
,
3128 "Objects remaining on kmem_cache_close()");
3134 * Release all resources used by a slab cache.
3136 static inline int kmem_cache_close(struct kmem_cache
*s
)
3141 free_percpu(s
->cpu_slab
);
3142 /* Attempt to free all objects */
3143 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3144 struct kmem_cache_node
*n
= get_node(s
, node
);
3147 if (n
->nr_partial
|| slabs_node(s
, node
))
3150 free_kmem_cache_nodes(s
);
3155 * Close a cache and release the kmem_cache structure
3156 * (must be used for caches created using kmem_cache_create)
3158 void kmem_cache_destroy(struct kmem_cache
*s
)
3160 down_write(&slub_lock
);
3164 up_write(&slub_lock
);
3165 if (kmem_cache_close(s
)) {
3166 printk(KERN_ERR
"SLUB %s: %s called for cache that "
3167 "still has objects.\n", s
->name
, __func__
);
3170 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
3172 sysfs_slab_remove(s
);
3174 up_write(&slub_lock
);
3176 EXPORT_SYMBOL(kmem_cache_destroy
);
3178 /********************************************************************
3180 *******************************************************************/
3182 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
3183 EXPORT_SYMBOL(kmalloc_caches
);
3185 static struct kmem_cache
*kmem_cache
;
3187 #ifdef CONFIG_ZONE_DMA
3188 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
3191 static int __init
setup_slub_min_order(char *str
)
3193 get_option(&str
, &slub_min_order
);
3198 __setup("slub_min_order=", setup_slub_min_order
);
3200 static int __init
setup_slub_max_order(char *str
)
3202 get_option(&str
, &slub_max_order
);
3203 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3208 __setup("slub_max_order=", setup_slub_max_order
);
3210 static int __init
setup_slub_min_objects(char *str
)
3212 get_option(&str
, &slub_min_objects
);
3217 __setup("slub_min_objects=", setup_slub_min_objects
);
3219 static int __init
setup_slub_nomerge(char *str
)
3225 __setup("slub_nomerge", setup_slub_nomerge
);
3227 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
3228 int size
, unsigned int flags
)
3230 struct kmem_cache
*s
;
3232 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3235 * This function is called with IRQs disabled during early-boot on
3236 * single CPU so there's no need to take slub_lock here.
3238 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
3242 list_add(&s
->list
, &slab_caches
);
3246 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
3251 * Conversion table for small slabs sizes / 8 to the index in the
3252 * kmalloc array. This is necessary for slabs < 192 since we have non power
3253 * of two cache sizes there. The size of larger slabs can be determined using
3256 static s8 size_index
[24] = {
3283 static inline int size_index_elem(size_t bytes
)
3285 return (bytes
- 1) / 8;
3288 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
3294 return ZERO_SIZE_PTR
;
3296 index
= size_index
[size_index_elem(size
)];
3298 index
= fls(size
- 1);
3300 #ifdef CONFIG_ZONE_DMA
3301 if (unlikely((flags
& SLUB_DMA
)))
3302 return kmalloc_dma_caches
[index
];
3305 return kmalloc_caches
[index
];
3308 void *__kmalloc(size_t size
, gfp_t flags
)
3310 struct kmem_cache
*s
;
3313 if (unlikely(size
> SLUB_MAX_SIZE
))
3314 return kmalloc_large(size
, flags
);
3316 s
= get_slab(size
, flags
);
3318 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3321 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
3323 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3327 EXPORT_SYMBOL(__kmalloc
);
3330 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3335 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3336 page
= alloc_pages_node(node
, flags
, get_order(size
));
3338 ptr
= page_address(page
);
3340 kmemleak_alloc(ptr
, size
, 1, flags
);
3344 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3346 struct kmem_cache
*s
;
3349 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3350 ret
= kmalloc_large_node(size
, flags
, node
);
3352 trace_kmalloc_node(_RET_IP_
, ret
,
3353 size
, PAGE_SIZE
<< get_order(size
),
3359 s
= get_slab(size
, flags
);
3361 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3364 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
3366 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3370 EXPORT_SYMBOL(__kmalloc_node
);
3373 size_t ksize(const void *object
)
3377 if (unlikely(object
== ZERO_SIZE_PTR
))
3380 page
= virt_to_head_page(object
);
3382 if (unlikely(!PageSlab(page
))) {
3383 WARN_ON(!PageCompound(page
));
3384 return PAGE_SIZE
<< compound_order(page
);
3387 return slab_ksize(page
->slab
);
3389 EXPORT_SYMBOL(ksize
);
3391 #ifdef CONFIG_SLUB_DEBUG
3392 bool verify_mem_not_deleted(const void *x
)
3395 void *object
= (void *)x
;
3396 unsigned long flags
;
3399 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3402 local_irq_save(flags
);
3404 page
= virt_to_head_page(x
);
3405 if (unlikely(!PageSlab(page
))) {
3406 /* maybe it was from stack? */
3412 if (on_freelist(page
->slab
, page
, object
)) {
3413 object_err(page
->slab
, page
, object
, "Object is on free-list");
3421 local_irq_restore(flags
);
3424 EXPORT_SYMBOL(verify_mem_not_deleted
);
3427 void kfree(const void *x
)
3430 void *object
= (void *)x
;
3432 trace_kfree(_RET_IP_
, x
);
3434 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3437 page
= virt_to_head_page(x
);
3438 if (unlikely(!PageSlab(page
))) {
3439 BUG_ON(!PageCompound(page
));
3444 slab_free(page
->slab
, page
, object
, _RET_IP_
);
3446 EXPORT_SYMBOL(kfree
);
3449 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3450 * the remaining slabs by the number of items in use. The slabs with the
3451 * most items in use come first. New allocations will then fill those up
3452 * and thus they can be removed from the partial lists.
3454 * The slabs with the least items are placed last. This results in them
3455 * being allocated from last increasing the chance that the last objects
3456 * are freed in them.
3458 int kmem_cache_shrink(struct kmem_cache
*s
)
3462 struct kmem_cache_node
*n
;
3465 int objects
= oo_objects(s
->max
);
3466 struct list_head
*slabs_by_inuse
=
3467 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3468 unsigned long flags
;
3470 if (!slabs_by_inuse
)
3474 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3475 n
= get_node(s
, node
);
3480 for (i
= 0; i
< objects
; i
++)
3481 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3483 spin_lock_irqsave(&n
->list_lock
, flags
);
3486 * Build lists indexed by the items in use in each slab.
3488 * Note that concurrent frees may occur while we hold the
3489 * list_lock. page->inuse here is the upper limit.
3491 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3492 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3498 * Rebuild the partial list with the slabs filled up most
3499 * first and the least used slabs at the end.
3501 for (i
= objects
- 1; i
> 0; i
--)
3502 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3504 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3506 /* Release empty slabs */
3507 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3508 discard_slab(s
, page
);
3511 kfree(slabs_by_inuse
);
3514 EXPORT_SYMBOL(kmem_cache_shrink
);
3516 #if defined(CONFIG_MEMORY_HOTPLUG)
3517 static int slab_mem_going_offline_callback(void *arg
)
3519 struct kmem_cache
*s
;
3521 down_read(&slub_lock
);
3522 list_for_each_entry(s
, &slab_caches
, list
)
3523 kmem_cache_shrink(s
);
3524 up_read(&slub_lock
);
3529 static void slab_mem_offline_callback(void *arg
)
3531 struct kmem_cache_node
*n
;
3532 struct kmem_cache
*s
;
3533 struct memory_notify
*marg
= arg
;
3536 offline_node
= marg
->status_change_nid
;
3539 * If the node still has available memory. we need kmem_cache_node
3542 if (offline_node
< 0)
3545 down_read(&slub_lock
);
3546 list_for_each_entry(s
, &slab_caches
, list
) {
3547 n
= get_node(s
, offline_node
);
3550 * if n->nr_slabs > 0, slabs still exist on the node
3551 * that is going down. We were unable to free them,
3552 * and offline_pages() function shouldn't call this
3553 * callback. So, we must fail.
3555 BUG_ON(slabs_node(s
, offline_node
));
3557 s
->node
[offline_node
] = NULL
;
3558 kmem_cache_free(kmem_cache_node
, n
);
3561 up_read(&slub_lock
);
3564 static int slab_mem_going_online_callback(void *arg
)
3566 struct kmem_cache_node
*n
;
3567 struct kmem_cache
*s
;
3568 struct memory_notify
*marg
= arg
;
3569 int nid
= marg
->status_change_nid
;
3573 * If the node's memory is already available, then kmem_cache_node is
3574 * already created. Nothing to do.
3580 * We are bringing a node online. No memory is available yet. We must
3581 * allocate a kmem_cache_node structure in order to bring the node
3584 down_read(&slub_lock
);
3585 list_for_each_entry(s
, &slab_caches
, list
) {
3587 * XXX: kmem_cache_alloc_node will fallback to other nodes
3588 * since memory is not yet available from the node that
3591 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3596 init_kmem_cache_node(n
, s
);
3600 up_read(&slub_lock
);
3604 static int slab_memory_callback(struct notifier_block
*self
,
3605 unsigned long action
, void *arg
)
3610 case MEM_GOING_ONLINE
:
3611 ret
= slab_mem_going_online_callback(arg
);
3613 case MEM_GOING_OFFLINE
:
3614 ret
= slab_mem_going_offline_callback(arg
);
3617 case MEM_CANCEL_ONLINE
:
3618 slab_mem_offline_callback(arg
);
3621 case MEM_CANCEL_OFFLINE
:
3625 ret
= notifier_from_errno(ret
);
3631 #endif /* CONFIG_MEMORY_HOTPLUG */
3633 /********************************************************************
3634 * Basic setup of slabs
3635 *******************************************************************/
3638 * Used for early kmem_cache structures that were allocated using
3639 * the page allocator
3642 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3646 list_add(&s
->list
, &slab_caches
);
3649 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3650 struct kmem_cache_node
*n
= get_node(s
, node
);
3654 list_for_each_entry(p
, &n
->partial
, lru
)
3657 #ifdef CONFIG_SLUB_DEBUG
3658 list_for_each_entry(p
, &n
->full
, lru
)
3665 void __init
kmem_cache_init(void)
3669 struct kmem_cache
*temp_kmem_cache
;
3671 struct kmem_cache
*temp_kmem_cache_node
;
3672 unsigned long kmalloc_size
;
3674 if (debug_guardpage_minorder())
3677 kmem_size
= offsetof(struct kmem_cache
, node
) +
3678 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3680 /* Allocate two kmem_caches from the page allocator */
3681 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3682 order
= get_order(2 * kmalloc_size
);
3683 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3686 * Must first have the slab cache available for the allocations of the
3687 * struct kmem_cache_node's. There is special bootstrap code in
3688 * kmem_cache_open for slab_state == DOWN.
3690 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3692 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3693 sizeof(struct kmem_cache_node
),
3694 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3696 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3698 /* Able to allocate the per node structures */
3699 slab_state
= PARTIAL
;
3701 temp_kmem_cache
= kmem_cache
;
3702 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3703 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3704 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3705 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3708 * Allocate kmem_cache_node properly from the kmem_cache slab.
3709 * kmem_cache_node is separately allocated so no need to
3710 * update any list pointers.
3712 temp_kmem_cache_node
= kmem_cache_node
;
3714 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3715 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3717 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3720 kmem_cache_bootstrap_fixup(kmem_cache
);
3722 /* Free temporary boot structure */
3723 free_pages((unsigned long)temp_kmem_cache
, order
);
3725 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3728 * Patch up the size_index table if we have strange large alignment
3729 * requirements for the kmalloc array. This is only the case for
3730 * MIPS it seems. The standard arches will not generate any code here.
3732 * Largest permitted alignment is 256 bytes due to the way we
3733 * handle the index determination for the smaller caches.
3735 * Make sure that nothing crazy happens if someone starts tinkering
3736 * around with ARCH_KMALLOC_MINALIGN
3738 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3739 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3741 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3742 int elem
= size_index_elem(i
);
3743 if (elem
>= ARRAY_SIZE(size_index
))
3745 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3748 if (KMALLOC_MIN_SIZE
== 64) {
3750 * The 96 byte size cache is not used if the alignment
3753 for (i
= 64 + 8; i
<= 96; i
+= 8)
3754 size_index
[size_index_elem(i
)] = 7;
3755 } else if (KMALLOC_MIN_SIZE
== 128) {
3757 * The 192 byte sized cache is not used if the alignment
3758 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3761 for (i
= 128 + 8; i
<= 192; i
+= 8)
3762 size_index
[size_index_elem(i
)] = 8;
3765 /* Caches that are not of the two-to-the-power-of size */
3766 if (KMALLOC_MIN_SIZE
<= 32) {
3767 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3771 if (KMALLOC_MIN_SIZE
<= 64) {
3772 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3776 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3777 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3783 /* Provide the correct kmalloc names now that the caches are up */
3784 if (KMALLOC_MIN_SIZE
<= 32) {
3785 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3786 BUG_ON(!kmalloc_caches
[1]->name
);
3789 if (KMALLOC_MIN_SIZE
<= 64) {
3790 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3791 BUG_ON(!kmalloc_caches
[2]->name
);
3794 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3795 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3798 kmalloc_caches
[i
]->name
= s
;
3802 register_cpu_notifier(&slab_notifier
);
3805 #ifdef CONFIG_ZONE_DMA
3806 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3807 struct kmem_cache
*s
= kmalloc_caches
[i
];
3810 char *name
= kasprintf(GFP_NOWAIT
,
3811 "dma-kmalloc-%d", s
->objsize
);
3814 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3815 s
->objsize
, SLAB_CACHE_DMA
);
3820 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3821 " CPUs=%d, Nodes=%d\n",
3822 caches
, cache_line_size(),
3823 slub_min_order
, slub_max_order
, slub_min_objects
,
3824 nr_cpu_ids
, nr_node_ids
);
3827 void __init
kmem_cache_init_late(void)
3832 * Find a mergeable slab cache
3834 static int slab_unmergeable(struct kmem_cache
*s
)
3836 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3843 * We may have set a slab to be unmergeable during bootstrap.
3845 if (s
->refcount
< 0)
3851 static struct kmem_cache
*find_mergeable(size_t size
,
3852 size_t align
, unsigned long flags
, const char *name
,
3853 void (*ctor
)(void *))
3855 struct kmem_cache
*s
;
3857 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3863 size
= ALIGN(size
, sizeof(void *));
3864 align
= calculate_alignment(flags
, align
, size
);
3865 size
= ALIGN(size
, align
);
3866 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3868 list_for_each_entry(s
, &slab_caches
, list
) {
3869 if (slab_unmergeable(s
))
3875 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3878 * Check if alignment is compatible.
3879 * Courtesy of Adrian Drzewiecki
3881 if ((s
->size
& ~(align
- 1)) != s
->size
)
3884 if (s
->size
- size
>= sizeof(void *))
3892 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3893 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3895 struct kmem_cache
*s
;
3901 down_write(&slub_lock
);
3902 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3906 * Adjust the object sizes so that we clear
3907 * the complete object on kzalloc.
3909 s
->objsize
= max(s
->objsize
, (int)size
);
3910 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3912 if (sysfs_slab_alias(s
, name
)) {
3916 up_write(&slub_lock
);
3920 n
= kstrdup(name
, GFP_KERNEL
);
3924 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3926 if (kmem_cache_open(s
, n
,
3927 size
, align
, flags
, ctor
)) {
3928 list_add(&s
->list
, &slab_caches
);
3929 if (sysfs_slab_add(s
)) {
3935 up_write(&slub_lock
);
3942 up_write(&slub_lock
);
3944 if (flags
& SLAB_PANIC
)
3945 panic("Cannot create slabcache %s\n", name
);
3950 EXPORT_SYMBOL(kmem_cache_create
);
3954 * Use the cpu notifier to insure that the cpu slabs are flushed when
3957 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3958 unsigned long action
, void *hcpu
)
3960 long cpu
= (long)hcpu
;
3961 struct kmem_cache
*s
;
3962 unsigned long flags
;
3965 case CPU_UP_CANCELED
:
3966 case CPU_UP_CANCELED_FROZEN
:
3968 case CPU_DEAD_FROZEN
:
3969 down_read(&slub_lock
);
3970 list_for_each_entry(s
, &slab_caches
, list
) {
3971 local_irq_save(flags
);
3972 __flush_cpu_slab(s
, cpu
);
3973 local_irq_restore(flags
);
3975 up_read(&slub_lock
);
3983 static struct notifier_block __cpuinitdata slab_notifier
= {
3984 .notifier_call
= slab_cpuup_callback
3989 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3991 struct kmem_cache
*s
;
3994 if (unlikely(size
> SLUB_MAX_SIZE
))
3995 return kmalloc_large(size
, gfpflags
);
3997 s
= get_slab(size
, gfpflags
);
3999 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4002 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
4004 /* Honor the call site pointer we received. */
4005 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
4011 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
4012 int node
, unsigned long caller
)
4014 struct kmem_cache
*s
;
4017 if (unlikely(size
> SLUB_MAX_SIZE
)) {
4018 ret
= kmalloc_large_node(size
, gfpflags
, node
);
4020 trace_kmalloc_node(caller
, ret
,
4021 size
, PAGE_SIZE
<< get_order(size
),
4027 s
= get_slab(size
, gfpflags
);
4029 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4032 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
4034 /* Honor the call site pointer we received. */
4035 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
4042 static int count_inuse(struct page
*page
)
4047 static int count_total(struct page
*page
)
4049 return page
->objects
;
4053 #ifdef CONFIG_SLUB_DEBUG
4054 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
4058 void *addr
= page_address(page
);
4060 if (!check_slab(s
, page
) ||
4061 !on_freelist(s
, page
, NULL
))
4064 /* Now we know that a valid freelist exists */
4065 bitmap_zero(map
, page
->objects
);
4067 get_map(s
, page
, map
);
4068 for_each_object(p
, s
, addr
, page
->objects
) {
4069 if (test_bit(slab_index(p
, s
, addr
), map
))
4070 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
4074 for_each_object(p
, s
, addr
, page
->objects
)
4075 if (!test_bit(slab_index(p
, s
, addr
), map
))
4076 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
4081 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
4085 validate_slab(s
, page
, map
);
4089 static int validate_slab_node(struct kmem_cache
*s
,
4090 struct kmem_cache_node
*n
, unsigned long *map
)
4092 unsigned long count
= 0;
4094 unsigned long flags
;
4096 spin_lock_irqsave(&n
->list_lock
, flags
);
4098 list_for_each_entry(page
, &n
->partial
, lru
) {
4099 validate_slab_slab(s
, page
, map
);
4102 if (count
!= n
->nr_partial
)
4103 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
4104 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
4106 if (!(s
->flags
& SLAB_STORE_USER
))
4109 list_for_each_entry(page
, &n
->full
, lru
) {
4110 validate_slab_slab(s
, page
, map
);
4113 if (count
!= atomic_long_read(&n
->nr_slabs
))
4114 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
4115 "counter=%ld\n", s
->name
, count
,
4116 atomic_long_read(&n
->nr_slabs
));
4119 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4123 static long validate_slab_cache(struct kmem_cache
*s
)
4126 unsigned long count
= 0;
4127 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4128 sizeof(unsigned long), GFP_KERNEL
);
4134 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4135 struct kmem_cache_node
*n
= get_node(s
, node
);
4137 count
+= validate_slab_node(s
, n
, map
);
4143 * Generate lists of code addresses where slabcache objects are allocated
4148 unsigned long count
;
4155 DECLARE_BITMAP(cpus
, NR_CPUS
);
4161 unsigned long count
;
4162 struct location
*loc
;
4165 static void free_loc_track(struct loc_track
*t
)
4168 free_pages((unsigned long)t
->loc
,
4169 get_order(sizeof(struct location
) * t
->max
));
4172 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4177 order
= get_order(sizeof(struct location
) * max
);
4179 l
= (void *)__get_free_pages(flags
, order
);
4184 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4192 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4193 const struct track
*track
)
4195 long start
, end
, pos
;
4197 unsigned long caddr
;
4198 unsigned long age
= jiffies
- track
->when
;
4204 pos
= start
+ (end
- start
+ 1) / 2;
4207 * There is nothing at "end". If we end up there
4208 * we need to add something to before end.
4213 caddr
= t
->loc
[pos
].addr
;
4214 if (track
->addr
== caddr
) {
4220 if (age
< l
->min_time
)
4222 if (age
> l
->max_time
)
4225 if (track
->pid
< l
->min_pid
)
4226 l
->min_pid
= track
->pid
;
4227 if (track
->pid
> l
->max_pid
)
4228 l
->max_pid
= track
->pid
;
4230 cpumask_set_cpu(track
->cpu
,
4231 to_cpumask(l
->cpus
));
4233 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4237 if (track
->addr
< caddr
)
4244 * Not found. Insert new tracking element.
4246 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4252 (t
->count
- pos
) * sizeof(struct location
));
4255 l
->addr
= track
->addr
;
4259 l
->min_pid
= track
->pid
;
4260 l
->max_pid
= track
->pid
;
4261 cpumask_clear(to_cpumask(l
->cpus
));
4262 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4263 nodes_clear(l
->nodes
);
4264 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4268 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4269 struct page
*page
, enum track_item alloc
,
4272 void *addr
= page_address(page
);
4275 bitmap_zero(map
, page
->objects
);
4276 get_map(s
, page
, map
);
4278 for_each_object(p
, s
, addr
, page
->objects
)
4279 if (!test_bit(slab_index(p
, s
, addr
), map
))
4280 add_location(t
, s
, get_track(s
, p
, alloc
));
4283 static int list_locations(struct kmem_cache
*s
, char *buf
,
4284 enum track_item alloc
)
4288 struct loc_track t
= { 0, 0, NULL
};
4290 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4291 sizeof(unsigned long), GFP_KERNEL
);
4293 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4296 return sprintf(buf
, "Out of memory\n");
4298 /* Push back cpu slabs */
4301 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4302 struct kmem_cache_node
*n
= get_node(s
, node
);
4303 unsigned long flags
;
4306 if (!atomic_long_read(&n
->nr_slabs
))
4309 spin_lock_irqsave(&n
->list_lock
, flags
);
4310 list_for_each_entry(page
, &n
->partial
, lru
)
4311 process_slab(&t
, s
, page
, alloc
, map
);
4312 list_for_each_entry(page
, &n
->full
, lru
)
4313 process_slab(&t
, s
, page
, alloc
, map
);
4314 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4317 for (i
= 0; i
< t
.count
; i
++) {
4318 struct location
*l
= &t
.loc
[i
];
4320 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4322 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4325 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4327 len
+= sprintf(buf
+ len
, "<not-available>");
4329 if (l
->sum_time
!= l
->min_time
) {
4330 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4332 (long)div_u64(l
->sum_time
, l
->count
),
4335 len
+= sprintf(buf
+ len
, " age=%ld",
4338 if (l
->min_pid
!= l
->max_pid
)
4339 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4340 l
->min_pid
, l
->max_pid
);
4342 len
+= sprintf(buf
+ len
, " pid=%ld",
4345 if (num_online_cpus() > 1 &&
4346 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4347 len
< PAGE_SIZE
- 60) {
4348 len
+= sprintf(buf
+ len
, " cpus=");
4349 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4350 to_cpumask(l
->cpus
));
4353 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4354 len
< PAGE_SIZE
- 60) {
4355 len
+= sprintf(buf
+ len
, " nodes=");
4356 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4360 len
+= sprintf(buf
+ len
, "\n");
4366 len
+= sprintf(buf
, "No data\n");
4371 #ifdef SLUB_RESILIENCY_TEST
4372 static void resiliency_test(void)
4376 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
4378 printk(KERN_ERR
"SLUB resiliency testing\n");
4379 printk(KERN_ERR
"-----------------------\n");
4380 printk(KERN_ERR
"A. Corruption after allocation\n");
4382 p
= kzalloc(16, GFP_KERNEL
);
4384 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4385 " 0x12->0x%p\n\n", p
+ 16);
4387 validate_slab_cache(kmalloc_caches
[4]);
4389 /* Hmmm... The next two are dangerous */
4390 p
= kzalloc(32, GFP_KERNEL
);
4391 p
[32 + sizeof(void *)] = 0x34;
4392 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4393 " 0x34 -> -0x%p\n", p
);
4395 "If allocated object is overwritten then not detectable\n\n");
4397 validate_slab_cache(kmalloc_caches
[5]);
4398 p
= kzalloc(64, GFP_KERNEL
);
4399 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4401 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4404 "If allocated object is overwritten then not detectable\n\n");
4405 validate_slab_cache(kmalloc_caches
[6]);
4407 printk(KERN_ERR
"\nB. Corruption after free\n");
4408 p
= kzalloc(128, GFP_KERNEL
);
4411 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4412 validate_slab_cache(kmalloc_caches
[7]);
4414 p
= kzalloc(256, GFP_KERNEL
);
4417 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4419 validate_slab_cache(kmalloc_caches
[8]);
4421 p
= kzalloc(512, GFP_KERNEL
);
4424 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4425 validate_slab_cache(kmalloc_caches
[9]);
4429 static void resiliency_test(void) {};
4434 enum slab_stat_type
{
4435 SL_ALL
, /* All slabs */
4436 SL_PARTIAL
, /* Only partially allocated slabs */
4437 SL_CPU
, /* Only slabs used for cpu caches */
4438 SL_OBJECTS
, /* Determine allocated objects not slabs */
4439 SL_TOTAL
/* Determine object capacity not slabs */
4442 #define SO_ALL (1 << SL_ALL)
4443 #define SO_PARTIAL (1 << SL_PARTIAL)
4444 #define SO_CPU (1 << SL_CPU)
4445 #define SO_OBJECTS (1 << SL_OBJECTS)
4446 #define SO_TOTAL (1 << SL_TOTAL)
4448 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4449 char *buf
, unsigned long flags
)
4451 unsigned long total
= 0;
4454 unsigned long *nodes
;
4455 unsigned long *per_cpu
;
4457 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4460 per_cpu
= nodes
+ nr_node_ids
;
4462 if (flags
& SO_CPU
) {
4465 for_each_possible_cpu(cpu
) {
4466 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4467 int node
= ACCESS_ONCE(c
->node
);
4472 page
= ACCESS_ONCE(c
->page
);
4474 if (flags
& SO_TOTAL
)
4476 else if (flags
& SO_OBJECTS
)
4495 lock_memory_hotplug();
4496 #ifdef CONFIG_SLUB_DEBUG
4497 if (flags
& SO_ALL
) {
4498 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4499 struct kmem_cache_node
*n
= get_node(s
, node
);
4501 if (flags
& SO_TOTAL
)
4502 x
= atomic_long_read(&n
->total_objects
);
4503 else if (flags
& SO_OBJECTS
)
4504 x
= atomic_long_read(&n
->total_objects
) -
4505 count_partial(n
, count_free
);
4508 x
= atomic_long_read(&n
->nr_slabs
);
4515 if (flags
& SO_PARTIAL
) {
4516 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4517 struct kmem_cache_node
*n
= get_node(s
, node
);
4519 if (flags
& SO_TOTAL
)
4520 x
= count_partial(n
, count_total
);
4521 else if (flags
& SO_OBJECTS
)
4522 x
= count_partial(n
, count_inuse
);
4529 x
= sprintf(buf
, "%lu", total
);
4531 for_each_node_state(node
, N_NORMAL_MEMORY
)
4533 x
+= sprintf(buf
+ x
, " N%d=%lu",
4536 unlock_memory_hotplug();
4538 return x
+ sprintf(buf
+ x
, "\n");
4541 #ifdef CONFIG_SLUB_DEBUG
4542 static int any_slab_objects(struct kmem_cache
*s
)
4546 for_each_online_node(node
) {
4547 struct kmem_cache_node
*n
= get_node(s
, node
);
4552 if (atomic_long_read(&n
->total_objects
))
4559 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4560 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4562 struct slab_attribute
{
4563 struct attribute attr
;
4564 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4565 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4568 #define SLAB_ATTR_RO(_name) \
4569 static struct slab_attribute _name##_attr = \
4570 __ATTR(_name, 0400, _name##_show, NULL)
4572 #define SLAB_ATTR(_name) \
4573 static struct slab_attribute _name##_attr = \
4574 __ATTR(_name, 0600, _name##_show, _name##_store)
4576 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4578 return sprintf(buf
, "%d\n", s
->size
);
4580 SLAB_ATTR_RO(slab_size
);
4582 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4584 return sprintf(buf
, "%d\n", s
->align
);
4586 SLAB_ATTR_RO(align
);
4588 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4590 return sprintf(buf
, "%d\n", s
->objsize
);
4592 SLAB_ATTR_RO(object_size
);
4594 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4596 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4598 SLAB_ATTR_RO(objs_per_slab
);
4600 static ssize_t
order_store(struct kmem_cache
*s
,
4601 const char *buf
, size_t length
)
4603 unsigned long order
;
4606 err
= strict_strtoul(buf
, 10, &order
);
4610 if (order
> slub_max_order
|| order
< slub_min_order
)
4613 calculate_sizes(s
, order
);
4617 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4619 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4623 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4625 return sprintf(buf
, "%lu\n", s
->min_partial
);
4628 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4634 err
= strict_strtoul(buf
, 10, &min
);
4638 set_min_partial(s
, min
);
4641 SLAB_ATTR(min_partial
);
4643 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4645 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4648 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4651 unsigned long objects
;
4654 err
= strict_strtoul(buf
, 10, &objects
);
4657 if (objects
&& kmem_cache_debug(s
))
4660 s
->cpu_partial
= objects
;
4664 SLAB_ATTR(cpu_partial
);
4666 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4670 return sprintf(buf
, "%pS\n", s
->ctor
);
4674 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4676 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4678 SLAB_ATTR_RO(aliases
);
4680 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4682 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4684 SLAB_ATTR_RO(partial
);
4686 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4688 return show_slab_objects(s
, buf
, SO_CPU
);
4690 SLAB_ATTR_RO(cpu_slabs
);
4692 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4694 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4696 SLAB_ATTR_RO(objects
);
4698 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4700 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4702 SLAB_ATTR_RO(objects_partial
);
4704 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4711 for_each_online_cpu(cpu
) {
4712 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4715 pages
+= page
->pages
;
4716 objects
+= page
->pobjects
;
4720 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4723 for_each_online_cpu(cpu
) {
4724 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4726 if (page
&& len
< PAGE_SIZE
- 20)
4727 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4728 page
->pobjects
, page
->pages
);
4731 return len
+ sprintf(buf
+ len
, "\n");
4733 SLAB_ATTR_RO(slabs_cpu_partial
);
4735 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4737 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4740 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4741 const char *buf
, size_t length
)
4743 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4745 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4748 SLAB_ATTR(reclaim_account
);
4750 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4752 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4754 SLAB_ATTR_RO(hwcache_align
);
4756 #ifdef CONFIG_ZONE_DMA
4757 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4759 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4761 SLAB_ATTR_RO(cache_dma
);
4764 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4766 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4768 SLAB_ATTR_RO(destroy_by_rcu
);
4770 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4772 return sprintf(buf
, "%d\n", s
->reserved
);
4774 SLAB_ATTR_RO(reserved
);
4776 #ifdef CONFIG_SLUB_DEBUG
4777 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4779 return show_slab_objects(s
, buf
, SO_ALL
);
4781 SLAB_ATTR_RO(slabs
);
4783 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4785 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4787 SLAB_ATTR_RO(total_objects
);
4789 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4791 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4794 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4795 const char *buf
, size_t length
)
4797 s
->flags
&= ~SLAB_DEBUG_FREE
;
4798 if (buf
[0] == '1') {
4799 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4800 s
->flags
|= SLAB_DEBUG_FREE
;
4804 SLAB_ATTR(sanity_checks
);
4806 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4808 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4811 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4814 s
->flags
&= ~SLAB_TRACE
;
4815 if (buf
[0] == '1') {
4816 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4817 s
->flags
|= SLAB_TRACE
;
4823 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4825 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4828 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4829 const char *buf
, size_t length
)
4831 if (any_slab_objects(s
))
4834 s
->flags
&= ~SLAB_RED_ZONE
;
4835 if (buf
[0] == '1') {
4836 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4837 s
->flags
|= SLAB_RED_ZONE
;
4839 calculate_sizes(s
, -1);
4842 SLAB_ATTR(red_zone
);
4844 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4846 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4849 static ssize_t
poison_store(struct kmem_cache
*s
,
4850 const char *buf
, size_t length
)
4852 if (any_slab_objects(s
))
4855 s
->flags
&= ~SLAB_POISON
;
4856 if (buf
[0] == '1') {
4857 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4858 s
->flags
|= SLAB_POISON
;
4860 calculate_sizes(s
, -1);
4865 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4867 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4870 static ssize_t
store_user_store(struct kmem_cache
*s
,
4871 const char *buf
, size_t length
)
4873 if (any_slab_objects(s
))
4876 s
->flags
&= ~SLAB_STORE_USER
;
4877 if (buf
[0] == '1') {
4878 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4879 s
->flags
|= SLAB_STORE_USER
;
4881 calculate_sizes(s
, -1);
4884 SLAB_ATTR(store_user
);
4886 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4891 static ssize_t
validate_store(struct kmem_cache
*s
,
4892 const char *buf
, size_t length
)
4896 if (buf
[0] == '1') {
4897 ret
= validate_slab_cache(s
);
4903 SLAB_ATTR(validate
);
4905 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4907 if (!(s
->flags
& SLAB_STORE_USER
))
4909 return list_locations(s
, buf
, TRACK_ALLOC
);
4911 SLAB_ATTR_RO(alloc_calls
);
4913 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4915 if (!(s
->flags
& SLAB_STORE_USER
))
4917 return list_locations(s
, buf
, TRACK_FREE
);
4919 SLAB_ATTR_RO(free_calls
);
4920 #endif /* CONFIG_SLUB_DEBUG */
4922 #ifdef CONFIG_FAILSLAB
4923 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4925 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4928 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4931 s
->flags
&= ~SLAB_FAILSLAB
;
4933 s
->flags
|= SLAB_FAILSLAB
;
4936 SLAB_ATTR(failslab
);
4939 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4944 static ssize_t
shrink_store(struct kmem_cache
*s
,
4945 const char *buf
, size_t length
)
4947 if (buf
[0] == '1') {
4948 int rc
= kmem_cache_shrink(s
);
4959 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4961 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4964 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4965 const char *buf
, size_t length
)
4967 unsigned long ratio
;
4970 err
= strict_strtoul(buf
, 10, &ratio
);
4975 s
->remote_node_defrag_ratio
= ratio
* 10;
4979 SLAB_ATTR(remote_node_defrag_ratio
);
4982 #ifdef CONFIG_SLUB_STATS
4983 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4985 unsigned long sum
= 0;
4988 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4993 for_each_online_cpu(cpu
) {
4994 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
5000 len
= sprintf(buf
, "%lu", sum
);
5003 for_each_online_cpu(cpu
) {
5004 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
5005 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
5009 return len
+ sprintf(buf
+ len
, "\n");
5012 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
5016 for_each_online_cpu(cpu
)
5017 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
5020 #define STAT_ATTR(si, text) \
5021 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5023 return show_stat(s, buf, si); \
5025 static ssize_t text##_store(struct kmem_cache *s, \
5026 const char *buf, size_t length) \
5028 if (buf[0] != '0') \
5030 clear_stat(s, si); \
5035 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5036 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
5037 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
5038 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
5039 STAT_ATTR(FREE_FROZEN
, free_frozen
);
5040 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
5041 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
5042 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
5043 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
5044 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
5045 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
5046 STAT_ATTR(FREE_SLAB
, free_slab
);
5047 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
5048 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
5049 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
5050 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
5051 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
5052 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
5053 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
5054 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
5055 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
5056 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
5057 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
5058 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
5061 static struct attribute
*slab_attrs
[] = {
5062 &slab_size_attr
.attr
,
5063 &object_size_attr
.attr
,
5064 &objs_per_slab_attr
.attr
,
5066 &min_partial_attr
.attr
,
5067 &cpu_partial_attr
.attr
,
5069 &objects_partial_attr
.attr
,
5071 &cpu_slabs_attr
.attr
,
5075 &hwcache_align_attr
.attr
,
5076 &reclaim_account_attr
.attr
,
5077 &destroy_by_rcu_attr
.attr
,
5079 &reserved_attr
.attr
,
5080 &slabs_cpu_partial_attr
.attr
,
5081 #ifdef CONFIG_SLUB_DEBUG
5082 &total_objects_attr
.attr
,
5084 &sanity_checks_attr
.attr
,
5086 &red_zone_attr
.attr
,
5088 &store_user_attr
.attr
,
5089 &validate_attr
.attr
,
5090 &alloc_calls_attr
.attr
,
5091 &free_calls_attr
.attr
,
5093 #ifdef CONFIG_ZONE_DMA
5094 &cache_dma_attr
.attr
,
5097 &remote_node_defrag_ratio_attr
.attr
,
5099 #ifdef CONFIG_SLUB_STATS
5100 &alloc_fastpath_attr
.attr
,
5101 &alloc_slowpath_attr
.attr
,
5102 &free_fastpath_attr
.attr
,
5103 &free_slowpath_attr
.attr
,
5104 &free_frozen_attr
.attr
,
5105 &free_add_partial_attr
.attr
,
5106 &free_remove_partial_attr
.attr
,
5107 &alloc_from_partial_attr
.attr
,
5108 &alloc_slab_attr
.attr
,
5109 &alloc_refill_attr
.attr
,
5110 &alloc_node_mismatch_attr
.attr
,
5111 &free_slab_attr
.attr
,
5112 &cpuslab_flush_attr
.attr
,
5113 &deactivate_full_attr
.attr
,
5114 &deactivate_empty_attr
.attr
,
5115 &deactivate_to_head_attr
.attr
,
5116 &deactivate_to_tail_attr
.attr
,
5117 &deactivate_remote_frees_attr
.attr
,
5118 &deactivate_bypass_attr
.attr
,
5119 &order_fallback_attr
.attr
,
5120 &cmpxchg_double_fail_attr
.attr
,
5121 &cmpxchg_double_cpu_fail_attr
.attr
,
5122 &cpu_partial_alloc_attr
.attr
,
5123 &cpu_partial_free_attr
.attr
,
5125 #ifdef CONFIG_FAILSLAB
5126 &failslab_attr
.attr
,
5132 static struct attribute_group slab_attr_group
= {
5133 .attrs
= slab_attrs
,
5136 static ssize_t
slab_attr_show(struct kobject
*kobj
,
5137 struct attribute
*attr
,
5140 struct slab_attribute
*attribute
;
5141 struct kmem_cache
*s
;
5144 attribute
= to_slab_attr(attr
);
5147 if (!attribute
->show
)
5150 err
= attribute
->show(s
, buf
);
5155 static ssize_t
slab_attr_store(struct kobject
*kobj
,
5156 struct attribute
*attr
,
5157 const char *buf
, size_t len
)
5159 struct slab_attribute
*attribute
;
5160 struct kmem_cache
*s
;
5163 attribute
= to_slab_attr(attr
);
5166 if (!attribute
->store
)
5169 err
= attribute
->store(s
, buf
, len
);
5174 static void kmem_cache_release(struct kobject
*kobj
)
5176 struct kmem_cache
*s
= to_slab(kobj
);
5182 static const struct sysfs_ops slab_sysfs_ops
= {
5183 .show
= slab_attr_show
,
5184 .store
= slab_attr_store
,
5187 static struct kobj_type slab_ktype
= {
5188 .sysfs_ops
= &slab_sysfs_ops
,
5189 .release
= kmem_cache_release
5192 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5194 struct kobj_type
*ktype
= get_ktype(kobj
);
5196 if (ktype
== &slab_ktype
)
5201 static const struct kset_uevent_ops slab_uevent_ops
= {
5202 .filter
= uevent_filter
,
5205 static struct kset
*slab_kset
;
5207 #define ID_STR_LENGTH 64
5209 /* Create a unique string id for a slab cache:
5211 * Format :[flags-]size
5213 static char *create_unique_id(struct kmem_cache
*s
)
5215 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5222 * First flags affecting slabcache operations. We will only
5223 * get here for aliasable slabs so we do not need to support
5224 * too many flags. The flags here must cover all flags that
5225 * are matched during merging to guarantee that the id is
5228 if (s
->flags
& SLAB_CACHE_DMA
)
5230 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5232 if (s
->flags
& SLAB_DEBUG_FREE
)
5234 if (!(s
->flags
& SLAB_NOTRACK
))
5238 p
+= sprintf(p
, "%07d", s
->size
);
5239 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5243 static int sysfs_slab_add(struct kmem_cache
*s
)
5249 if (slab_state
< SYSFS
)
5250 /* Defer until later */
5253 unmergeable
= slab_unmergeable(s
);
5256 * Slabcache can never be merged so we can use the name proper.
5257 * This is typically the case for debug situations. In that
5258 * case we can catch duplicate names easily.
5260 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5264 * Create a unique name for the slab as a target
5267 name
= create_unique_id(s
);
5270 s
->kobj
.kset
= slab_kset
;
5271 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5273 kobject_put(&s
->kobj
);
5277 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5279 kobject_del(&s
->kobj
);
5280 kobject_put(&s
->kobj
);
5283 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5285 /* Setup first alias */
5286 sysfs_slab_alias(s
, s
->name
);
5292 static void sysfs_slab_remove(struct kmem_cache
*s
)
5294 if (slab_state
< SYSFS
)
5296 * Sysfs has not been setup yet so no need to remove the
5301 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5302 kobject_del(&s
->kobj
);
5303 kobject_put(&s
->kobj
);
5307 * Need to buffer aliases during bootup until sysfs becomes
5308 * available lest we lose that information.
5310 struct saved_alias
{
5311 struct kmem_cache
*s
;
5313 struct saved_alias
*next
;
5316 static struct saved_alias
*alias_list
;
5318 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5320 struct saved_alias
*al
;
5322 if (slab_state
== SYSFS
) {
5324 * If we have a leftover link then remove it.
5326 sysfs_remove_link(&slab_kset
->kobj
, name
);
5327 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5330 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5336 al
->next
= alias_list
;
5341 static int __init
slab_sysfs_init(void)
5343 struct kmem_cache
*s
;
5346 down_write(&slub_lock
);
5348 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5350 up_write(&slub_lock
);
5351 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5357 list_for_each_entry(s
, &slab_caches
, list
) {
5358 err
= sysfs_slab_add(s
);
5360 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5361 " to sysfs\n", s
->name
);
5364 while (alias_list
) {
5365 struct saved_alias
*al
= alias_list
;
5367 alias_list
= alias_list
->next
;
5368 err
= sysfs_slab_alias(al
->s
, al
->name
);
5370 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5371 " %s to sysfs\n", s
->name
);
5375 up_write(&slub_lock
);
5380 __initcall(slab_sysfs_init
);
5381 #endif /* CONFIG_SYSFS */
5384 * The /proc/slabinfo ABI
5386 #ifdef CONFIG_SLABINFO
5387 static void print_slabinfo_header(struct seq_file
*m
)
5389 seq_puts(m
, "slabinfo - version: 2.1\n");
5390 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
5391 "<objperslab> <pagesperslab>");
5392 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
5393 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5397 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
5401 down_read(&slub_lock
);
5403 print_slabinfo_header(m
);
5405 return seq_list_start(&slab_caches
, *pos
);
5408 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
5410 return seq_list_next(p
, &slab_caches
, pos
);
5413 static void s_stop(struct seq_file
*m
, void *p
)
5415 up_read(&slub_lock
);
5418 static int s_show(struct seq_file
*m
, void *p
)
5420 unsigned long nr_partials
= 0;
5421 unsigned long nr_slabs
= 0;
5422 unsigned long nr_inuse
= 0;
5423 unsigned long nr_objs
= 0;
5424 unsigned long nr_free
= 0;
5425 struct kmem_cache
*s
;
5428 s
= list_entry(p
, struct kmem_cache
, list
);
5430 for_each_online_node(node
) {
5431 struct kmem_cache_node
*n
= get_node(s
, node
);
5436 nr_partials
+= n
->nr_partial
;
5437 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
5438 nr_objs
+= atomic_long_read(&n
->total_objects
);
5439 nr_free
+= count_partial(n
, count_free
);
5442 nr_inuse
= nr_objs
- nr_free
;
5444 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
5445 nr_objs
, s
->size
, oo_objects(s
->oo
),
5446 (1 << oo_order(s
->oo
)));
5447 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
5448 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
5454 static const struct seq_operations slabinfo_op
= {
5461 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
5463 return seq_open(file
, &slabinfo_op
);
5466 static const struct file_operations proc_slabinfo_operations
= {
5467 .open
= slabinfo_open
,
5469 .llseek
= seq_lseek
,
5470 .release
= seq_release
,
5473 static int __init
slab_proc_init(void)
5475 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
5478 module_init(slab_proc_init
);
5479 #endif /* CONFIG_SLABINFO */