2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
20 #include <linux/proc_fs.h>
21 #include <linux/notifier.h>
22 #include <linux/seq_file.h>
23 #include <linux/kmemcheck.h>
24 #include <linux/cpu.h>
25 #include <linux/cpuset.h>
26 #include <linux/mempolicy.h>
27 #include <linux/ctype.h>
28 #include <linux/debugobjects.h>
29 #include <linux/kallsyms.h>
30 #include <linux/memory.h>
31 #include <linux/math64.h>
32 #include <linux/fault-inject.h>
33 #include <linux/stacktrace.h>
34 #include <linux/prefetch.h>
35 #include <linux/memcontrol.h>
37 #include <trace/events/kmem.h>
43 * 1. slab_mutex (Global Mutex)
45 * 3. slab_lock(page) (Only on some arches and for debugging)
49 * The role of the slab_mutex is to protect the list of all the slabs
50 * and to synchronize major metadata changes to slab cache structures.
52 * The slab_lock is only used for debugging and on arches that do not
53 * have the ability to do a cmpxchg_double. It only protects the second
54 * double word in the page struct. Meaning
55 * A. page->freelist -> List of object free in a page
56 * B. page->counters -> Counters of objects
57 * C. page->frozen -> frozen state
59 * If a slab is frozen then it is exempt from list management. It is not
60 * on any list. The processor that froze the slab is the one who can
61 * perform list operations on the page. Other processors may put objects
62 * onto the freelist but the processor that froze the slab is the only
63 * one that can retrieve the objects from the page's freelist.
65 * The list_lock protects the partial and full list on each node and
66 * the partial slab counter. If taken then no new slabs may be added or
67 * removed from the lists nor make the number of partial slabs be modified.
68 * (Note that the total number of slabs is an atomic value that may be
69 * modified without taking the list lock).
71 * The list_lock is a centralized lock and thus we avoid taking it as
72 * much as possible. As long as SLUB does not have to handle partial
73 * slabs, operations can continue without any centralized lock. F.e.
74 * allocating a long series of objects that fill up slabs does not require
76 * Interrupts are disabled during allocation and deallocation in order to
77 * make the slab allocator safe to use in the context of an irq. In addition
78 * interrupts are disabled to ensure that the processor does not change
79 * while handling per_cpu slabs, due to kernel preemption.
81 * SLUB assigns one slab for allocation to each processor.
82 * Allocations only occur from these slabs called cpu slabs.
84 * Slabs with free elements are kept on a partial list and during regular
85 * operations no list for full slabs is used. If an object in a full slab is
86 * freed then the slab will show up again on the partial lists.
87 * We track full slabs for debugging purposes though because otherwise we
88 * cannot scan all objects.
90 * Slabs are freed when they become empty. Teardown and setup is
91 * minimal so we rely on the page allocators per cpu caches for
92 * fast frees and allocs.
94 * Overloading of page flags that are otherwise used for LRU management.
96 * PageActive The slab is frozen and exempt from list processing.
97 * This means that the slab is dedicated to a purpose
98 * such as satisfying allocations for a specific
99 * processor. Objects may be freed in the slab while
100 * it is frozen but slab_free will then skip the usual
101 * list operations. It is up to the processor holding
102 * the slab to integrate the slab into the slab lists
103 * when the slab is no longer needed.
105 * One use of this flag is to mark slabs that are
106 * used for allocations. Then such a slab becomes a cpu
107 * slab. The cpu slab may be equipped with an additional
108 * freelist that allows lockless access to
109 * free objects in addition to the regular freelist
110 * that requires the slab lock.
112 * PageError Slab requires special handling due to debug
113 * options set. This moves slab handling out of
114 * the fast path and disables lockless freelists.
117 static inline int kmem_cache_debug(struct kmem_cache
*s
)
119 #ifdef CONFIG_SLUB_DEBUG
120 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
126 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache
*s
)
128 #ifdef CONFIG_SLUB_CPU_PARTIAL
129 return !kmem_cache_debug(s
);
136 * Issues still to be resolved:
138 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
140 * - Variable sizing of the per node arrays
143 /* Enable to test recovery from slab corruption on boot */
144 #undef SLUB_RESILIENCY_TEST
146 /* Enable to log cmpxchg failures */
147 #undef SLUB_DEBUG_CMPXCHG
150 * Mininum number of partial slabs. These will be left on the partial
151 * lists even if they are empty. kmem_cache_shrink may reclaim them.
153 #define MIN_PARTIAL 5
156 * Maximum number of desirable partial slabs.
157 * The existence of more partial slabs makes kmem_cache_shrink
158 * sort the partial list by the number of objects in use.
160 #define MAX_PARTIAL 10
162 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
163 SLAB_POISON | SLAB_STORE_USER)
166 * Debugging flags that require metadata to be stored in the slab. These get
167 * disabled when slub_debug=O is used and a cache's min order increases with
170 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
173 * Set of flags that will prevent slab merging
175 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
176 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
179 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
180 SLAB_CACHE_DMA | SLAB_NOTRACK)
183 #define OO_MASK ((1 << OO_SHIFT) - 1)
184 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
186 /* Internal SLUB flags */
187 #define __OBJECT_POISON 0x80000000UL /* Poison object */
188 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
191 static struct notifier_block slab_notifier
;
195 * Tracking user of a slab.
197 #define TRACK_ADDRS_COUNT 16
199 unsigned long addr
; /* Called from address */
200 #ifdef CONFIG_STACKTRACE
201 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
203 int cpu
; /* Was running on cpu */
204 int pid
; /* Pid context */
205 unsigned long when
; /* When did the operation occur */
208 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
211 static int sysfs_slab_add(struct kmem_cache
*);
212 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
213 static void sysfs_slab_remove(struct kmem_cache
*);
214 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
);
216 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
217 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
219 static inline void sysfs_slab_remove(struct kmem_cache
*s
) { }
221 static inline void memcg_propagate_slab_attrs(struct kmem_cache
*s
) { }
224 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
226 #ifdef CONFIG_SLUB_STATS
227 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
231 /********************************************************************
232 * Core slab cache functions
233 *******************************************************************/
235 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
237 return s
->node
[node
];
240 /* Verify that a pointer has an address that is valid within a slab page */
241 static inline int check_valid_pointer(struct kmem_cache
*s
,
242 struct page
*page
, const void *object
)
249 base
= page_address(page
);
250 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
251 (object
- base
) % s
->size
) {
258 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
260 return *(void **)(object
+ s
->offset
);
263 static void prefetch_freepointer(const struct kmem_cache
*s
, void *object
)
265 prefetch(object
+ s
->offset
);
268 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
272 #ifdef CONFIG_DEBUG_PAGEALLOC
273 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
275 p
= get_freepointer(s
, object
);
280 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
282 *(void **)(object
+ s
->offset
) = fp
;
285 /* Loop over all objects in a slab */
286 #define for_each_object(__p, __s, __addr, __objects) \
287 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
290 /* Determine object index from a given position */
291 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
293 return (p
- addr
) / s
->size
;
296 static inline size_t slab_ksize(const struct kmem_cache
*s
)
298 #ifdef CONFIG_SLUB_DEBUG
300 * Debugging requires use of the padding between object
301 * and whatever may come after it.
303 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
304 return s
->object_size
;
308 * If we have the need to store the freelist pointer
309 * back there or track user information then we can
310 * only use the space before that information.
312 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
315 * Else we can use all the padding etc for the allocation
320 static inline int order_objects(int order
, unsigned long size
, int reserved
)
322 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
325 static inline struct kmem_cache_order_objects
oo_make(int order
,
326 unsigned long size
, int reserved
)
328 struct kmem_cache_order_objects x
= {
329 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
335 static inline int oo_order(struct kmem_cache_order_objects x
)
337 return x
.x
>> OO_SHIFT
;
340 static inline int oo_objects(struct kmem_cache_order_objects x
)
342 return x
.x
& OO_MASK
;
346 * Per slab locking using the pagelock
348 static __always_inline
void slab_lock(struct page
*page
)
350 bit_spin_lock(PG_locked
, &page
->flags
);
353 static __always_inline
void slab_unlock(struct page
*page
)
355 __bit_spin_unlock(PG_locked
, &page
->flags
);
358 static inline void set_page_slub_counters(struct page
*page
, unsigned long counters_new
)
361 tmp
.counters
= counters_new
;
363 * page->counters can cover frozen/inuse/objects as well
364 * as page->_count. If we assign to ->counters directly
365 * we run the risk of losing updates to page->_count, so
366 * be careful and only assign to the fields we need.
368 page
->frozen
= tmp
.frozen
;
369 page
->inuse
= tmp
.inuse
;
370 page
->objects
= tmp
.objects
;
373 /* Interrupts must be disabled (for the fallback code to work right) */
374 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
375 void *freelist_old
, unsigned long counters_old
,
376 void *freelist_new
, unsigned long counters_new
,
379 VM_BUG_ON(!irqs_disabled());
380 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
381 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
382 if (s
->flags
& __CMPXCHG_DOUBLE
) {
383 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
384 freelist_old
, counters_old
,
385 freelist_new
, counters_new
))
391 if (page
->freelist
== freelist_old
&&
392 page
->counters
== counters_old
) {
393 page
->freelist
= freelist_new
;
394 set_page_slub_counters(page
, counters_new
);
402 stat(s
, CMPXCHG_DOUBLE_FAIL
);
404 #ifdef SLUB_DEBUG_CMPXCHG
405 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
411 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
412 void *freelist_old
, unsigned long counters_old
,
413 void *freelist_new
, unsigned long counters_new
,
416 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
417 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
418 if (s
->flags
& __CMPXCHG_DOUBLE
) {
419 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
420 freelist_old
, counters_old
,
421 freelist_new
, counters_new
))
428 local_irq_save(flags
);
430 if (page
->freelist
== freelist_old
&&
431 page
->counters
== counters_old
) {
432 page
->freelist
= freelist_new
;
433 set_page_slub_counters(page
, counters_new
);
435 local_irq_restore(flags
);
439 local_irq_restore(flags
);
443 stat(s
, CMPXCHG_DOUBLE_FAIL
);
445 #ifdef SLUB_DEBUG_CMPXCHG
446 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
452 #ifdef CONFIG_SLUB_DEBUG
454 * Determine a map of object in use on a page.
456 * Node listlock must be held to guarantee that the page does
457 * not vanish from under us.
459 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
462 void *addr
= page_address(page
);
464 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
465 set_bit(slab_index(p
, s
, addr
), map
);
471 #ifdef CONFIG_SLUB_DEBUG_ON
472 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
474 static int slub_debug
;
477 static char *slub_debug_slabs
;
478 static int disable_higher_order_debug
;
483 static void print_section(char *text
, u8
*addr
, unsigned int length
)
485 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
489 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
490 enum track_item alloc
)
495 p
= object
+ s
->offset
+ sizeof(void *);
497 p
= object
+ s
->inuse
;
502 static void set_track(struct kmem_cache
*s
, void *object
,
503 enum track_item alloc
, unsigned long addr
)
505 struct track
*p
= get_track(s
, object
, alloc
);
508 #ifdef CONFIG_STACKTRACE
509 struct stack_trace trace
;
512 trace
.nr_entries
= 0;
513 trace
.max_entries
= TRACK_ADDRS_COUNT
;
514 trace
.entries
= p
->addrs
;
516 save_stack_trace(&trace
);
518 /* See rant in lockdep.c */
519 if (trace
.nr_entries
!= 0 &&
520 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
523 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
527 p
->cpu
= smp_processor_id();
528 p
->pid
= current
->pid
;
531 memset(p
, 0, sizeof(struct track
));
534 static void init_tracking(struct kmem_cache
*s
, void *object
)
536 if (!(s
->flags
& SLAB_STORE_USER
))
539 set_track(s
, object
, TRACK_FREE
, 0UL);
540 set_track(s
, object
, TRACK_ALLOC
, 0UL);
543 static void print_track(const char *s
, struct track
*t
)
548 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
549 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
550 #ifdef CONFIG_STACKTRACE
553 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
555 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
562 static void print_tracking(struct kmem_cache
*s
, void *object
)
564 if (!(s
->flags
& SLAB_STORE_USER
))
567 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
568 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
571 static void print_page_info(struct page
*page
)
574 "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
575 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
579 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
585 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
587 printk(KERN_ERR
"========================================"
588 "=====================================\n");
589 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
590 printk(KERN_ERR
"----------------------------------------"
591 "-------------------------------------\n\n");
593 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
596 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
602 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
604 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
607 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
609 unsigned int off
; /* Offset of last byte */
610 u8
*addr
= page_address(page
);
612 print_tracking(s
, p
);
614 print_page_info(page
);
616 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
617 p
, p
- addr
, get_freepointer(s
, p
));
620 print_section("Bytes b4 ", p
- 16, 16);
622 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
624 if (s
->flags
& SLAB_RED_ZONE
)
625 print_section("Redzone ", p
+ s
->object_size
,
626 s
->inuse
- s
->object_size
);
629 off
= s
->offset
+ sizeof(void *);
633 if (s
->flags
& SLAB_STORE_USER
)
634 off
+= 2 * sizeof(struct track
);
637 /* Beginning of the filler is the free pointer */
638 print_section("Padding ", p
+ off
, s
->size
- off
);
643 static void object_err(struct kmem_cache
*s
, struct page
*page
,
644 u8
*object
, char *reason
)
646 slab_bug(s
, "%s", reason
);
647 print_trailer(s
, page
, object
);
650 static void slab_err(struct kmem_cache
*s
, struct page
*page
,
651 const char *fmt
, ...)
657 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
659 slab_bug(s
, "%s", buf
);
660 print_page_info(page
);
664 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
668 if (s
->flags
& __OBJECT_POISON
) {
669 memset(p
, POISON_FREE
, s
->object_size
- 1);
670 p
[s
->object_size
- 1] = POISON_END
;
673 if (s
->flags
& SLAB_RED_ZONE
)
674 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
677 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
678 void *from
, void *to
)
680 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
681 memset(from
, data
, to
- from
);
684 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
685 u8
*object
, char *what
,
686 u8
*start
, unsigned int value
, unsigned int bytes
)
691 fault
= memchr_inv(start
, value
, bytes
);
696 while (end
> fault
&& end
[-1] == value
)
699 slab_bug(s
, "%s overwritten", what
);
700 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
701 fault
, end
- 1, fault
[0], value
);
702 print_trailer(s
, page
, object
);
704 restore_bytes(s
, what
, value
, fault
, end
);
712 * Bytes of the object to be managed.
713 * If the freepointer may overlay the object then the free
714 * pointer is the first word of the object.
716 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
719 * object + s->object_size
720 * Padding to reach word boundary. This is also used for Redzoning.
721 * Padding is extended by another word if Redzoning is enabled and
722 * object_size == inuse.
724 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
725 * 0xcc (RED_ACTIVE) for objects in use.
728 * Meta data starts here.
730 * A. Free pointer (if we cannot overwrite object on free)
731 * B. Tracking data for SLAB_STORE_USER
732 * C. Padding to reach required alignment boundary or at mininum
733 * one word if debugging is on to be able to detect writes
734 * before the word boundary.
736 * Padding is done using 0x5a (POISON_INUSE)
739 * Nothing is used beyond s->size.
741 * If slabcaches are merged then the object_size and inuse boundaries are mostly
742 * ignored. And therefore no slab options that rely on these boundaries
743 * may be used with merged slabcaches.
746 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
748 unsigned long off
= s
->inuse
; /* The end of info */
751 /* Freepointer is placed after the object. */
752 off
+= sizeof(void *);
754 if (s
->flags
& SLAB_STORE_USER
)
755 /* We also have user information there */
756 off
+= 2 * sizeof(struct track
);
761 return check_bytes_and_report(s
, page
, p
, "Object padding",
762 p
+ off
, POISON_INUSE
, s
->size
- off
);
765 /* Check the pad bytes at the end of a slab page */
766 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
774 if (!(s
->flags
& SLAB_POISON
))
777 start
= page_address(page
);
778 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
779 end
= start
+ length
;
780 remainder
= length
% s
->size
;
784 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
787 while (end
> fault
&& end
[-1] == POISON_INUSE
)
790 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
791 print_section("Padding ", end
- remainder
, remainder
);
793 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
797 static int check_object(struct kmem_cache
*s
, struct page
*page
,
798 void *object
, u8 val
)
801 u8
*endobject
= object
+ s
->object_size
;
803 if (s
->flags
& SLAB_RED_ZONE
) {
804 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
805 endobject
, val
, s
->inuse
- s
->object_size
))
808 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
809 check_bytes_and_report(s
, page
, p
, "Alignment padding",
810 endobject
, POISON_INUSE
,
811 s
->inuse
- s
->object_size
);
815 if (s
->flags
& SLAB_POISON
) {
816 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
817 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
818 POISON_FREE
, s
->object_size
- 1) ||
819 !check_bytes_and_report(s
, page
, p
, "Poison",
820 p
+ s
->object_size
- 1, POISON_END
, 1)))
823 * check_pad_bytes cleans up on its own.
825 check_pad_bytes(s
, page
, p
);
828 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
830 * Object and freepointer overlap. Cannot check
831 * freepointer while object is allocated.
835 /* Check free pointer validity */
836 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
837 object_err(s
, page
, p
, "Freepointer corrupt");
839 * No choice but to zap it and thus lose the remainder
840 * of the free objects in this slab. May cause
841 * another error because the object count is now wrong.
843 set_freepointer(s
, p
, NULL
);
849 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
853 VM_BUG_ON(!irqs_disabled());
855 if (!PageSlab(page
)) {
856 slab_err(s
, page
, "Not a valid slab page");
860 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
861 if (page
->objects
> maxobj
) {
862 slab_err(s
, page
, "objects %u > max %u",
863 s
->name
, page
->objects
, maxobj
);
866 if (page
->inuse
> page
->objects
) {
867 slab_err(s
, page
, "inuse %u > max %u",
868 s
->name
, page
->inuse
, page
->objects
);
871 /* Slab_pad_check fixes things up after itself */
872 slab_pad_check(s
, page
);
877 * Determine if a certain object on a page is on the freelist. Must hold the
878 * slab lock to guarantee that the chains are in a consistent state.
880 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
885 unsigned long max_objects
;
888 while (fp
&& nr
<= page
->objects
) {
891 if (!check_valid_pointer(s
, page
, fp
)) {
893 object_err(s
, page
, object
,
894 "Freechain corrupt");
895 set_freepointer(s
, object
, NULL
);
897 slab_err(s
, page
, "Freepointer corrupt");
898 page
->freelist
= NULL
;
899 page
->inuse
= page
->objects
;
900 slab_fix(s
, "Freelist cleared");
906 fp
= get_freepointer(s
, object
);
910 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
911 if (max_objects
> MAX_OBJS_PER_PAGE
)
912 max_objects
= MAX_OBJS_PER_PAGE
;
914 if (page
->objects
!= max_objects
) {
915 slab_err(s
, page
, "Wrong number of objects. Found %d but "
916 "should be %d", page
->objects
, max_objects
);
917 page
->objects
= max_objects
;
918 slab_fix(s
, "Number of objects adjusted.");
920 if (page
->inuse
!= page
->objects
- nr
) {
921 slab_err(s
, page
, "Wrong object count. Counter is %d but "
922 "counted were %d", page
->inuse
, page
->objects
- nr
);
923 page
->inuse
= page
->objects
- nr
;
924 slab_fix(s
, "Object count adjusted.");
926 return search
== NULL
;
929 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
932 if (s
->flags
& SLAB_TRACE
) {
933 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
935 alloc
? "alloc" : "free",
940 print_section("Object ", (void *)object
,
948 * Hooks for other subsystems that check memory allocations. In a typical
949 * production configuration these hooks all should produce no code at all.
951 static inline void kmalloc_large_node_hook(void *ptr
, size_t size
, gfp_t flags
)
953 kmemleak_alloc(ptr
, size
, 1, flags
);
956 static inline void kfree_hook(const void *x
)
961 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
963 flags
&= gfp_allowed_mask
;
964 lockdep_trace_alloc(flags
);
965 might_sleep_if(flags
& __GFP_WAIT
);
967 return should_failslab(s
->object_size
, flags
, s
->flags
);
970 static inline void slab_post_alloc_hook(struct kmem_cache
*s
,
971 gfp_t flags
, void *object
)
973 flags
&= gfp_allowed_mask
;
974 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
975 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
978 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
980 kmemleak_free_recursive(x
, s
->flags
);
983 * Trouble is that we may no longer disable interrupts in the fast path
984 * So in order to make the debug calls that expect irqs to be
985 * disabled we need to disable interrupts temporarily.
987 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
991 local_irq_save(flags
);
992 kmemcheck_slab_free(s
, x
, s
->object_size
);
993 debug_check_no_locks_freed(x
, s
->object_size
);
994 local_irq_restore(flags
);
997 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
998 debug_check_no_obj_freed(x
, s
->object_size
);
1002 * Tracking of fully allocated slabs for debugging purposes.
1004 * list_lock must be held.
1006 static void add_full(struct kmem_cache
*s
,
1007 struct kmem_cache_node
*n
, struct page
*page
)
1009 if (!(s
->flags
& SLAB_STORE_USER
))
1012 list_add(&page
->lru
, &n
->full
);
1016 * list_lock must be held.
1018 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
1020 if (!(s
->flags
& SLAB_STORE_USER
))
1023 list_del(&page
->lru
);
1026 /* Tracking of the number of slabs for debugging purposes */
1027 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1029 struct kmem_cache_node
*n
= get_node(s
, node
);
1031 return atomic_long_read(&n
->nr_slabs
);
1034 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1036 return atomic_long_read(&n
->nr_slabs
);
1039 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1041 struct kmem_cache_node
*n
= get_node(s
, node
);
1044 * May be called early in order to allocate a slab for the
1045 * kmem_cache_node structure. Solve the chicken-egg
1046 * dilemma by deferring the increment of the count during
1047 * bootstrap (see early_kmem_cache_node_alloc).
1050 atomic_long_inc(&n
->nr_slabs
);
1051 atomic_long_add(objects
, &n
->total_objects
);
1054 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1056 struct kmem_cache_node
*n
= get_node(s
, node
);
1058 atomic_long_dec(&n
->nr_slabs
);
1059 atomic_long_sub(objects
, &n
->total_objects
);
1062 /* Object debug checks for alloc/free paths */
1063 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1066 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1069 init_object(s
, object
, SLUB_RED_INACTIVE
);
1070 init_tracking(s
, object
);
1073 static noinline
int alloc_debug_processing(struct kmem_cache
*s
,
1075 void *object
, unsigned long addr
)
1077 if (!check_slab(s
, page
))
1080 if (!check_valid_pointer(s
, page
, object
)) {
1081 object_err(s
, page
, object
, "Freelist Pointer check fails");
1085 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1088 /* Success perform special debug activities for allocs */
1089 if (s
->flags
& SLAB_STORE_USER
)
1090 set_track(s
, object
, TRACK_ALLOC
, addr
);
1091 trace(s
, page
, object
, 1);
1092 init_object(s
, object
, SLUB_RED_ACTIVE
);
1096 if (PageSlab(page
)) {
1098 * If this is a slab page then lets do the best we can
1099 * to avoid issues in the future. Marking all objects
1100 * as used avoids touching the remaining objects.
1102 slab_fix(s
, "Marking all objects used");
1103 page
->inuse
= page
->objects
;
1104 page
->freelist
= NULL
;
1109 static noinline
struct kmem_cache_node
*free_debug_processing(
1110 struct kmem_cache
*s
, struct page
*page
, void *object
,
1111 unsigned long addr
, unsigned long *flags
)
1113 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1115 spin_lock_irqsave(&n
->list_lock
, *flags
);
1118 if (!check_slab(s
, page
))
1121 if (!check_valid_pointer(s
, page
, object
)) {
1122 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1126 if (on_freelist(s
, page
, object
)) {
1127 object_err(s
, page
, object
, "Object already free");
1131 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1134 if (unlikely(s
!= page
->slab_cache
)) {
1135 if (!PageSlab(page
)) {
1136 slab_err(s
, page
, "Attempt to free object(0x%p) "
1137 "outside of slab", object
);
1138 } else if (!page
->slab_cache
) {
1140 "SLUB <none>: no slab for object 0x%p.\n",
1144 object_err(s
, page
, object
,
1145 "page slab pointer corrupt.");
1149 if (s
->flags
& SLAB_STORE_USER
)
1150 set_track(s
, object
, TRACK_FREE
, addr
);
1151 trace(s
, page
, object
, 0);
1152 init_object(s
, object
, SLUB_RED_INACTIVE
);
1156 * Keep node_lock to preserve integrity
1157 * until the object is actually freed
1163 spin_unlock_irqrestore(&n
->list_lock
, *flags
);
1164 slab_fix(s
, "Object at 0x%p not freed", object
);
1168 static int __init
setup_slub_debug(char *str
)
1170 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1171 if (*str
++ != '=' || !*str
)
1173 * No options specified. Switch on full debugging.
1179 * No options but restriction on slabs. This means full
1180 * debugging for slabs matching a pattern.
1184 if (tolower(*str
) == 'o') {
1186 * Avoid enabling debugging on caches if its minimum order
1187 * would increase as a result.
1189 disable_higher_order_debug
= 1;
1196 * Switch off all debugging measures.
1201 * Determine which debug features should be switched on
1203 for (; *str
&& *str
!= ','; str
++) {
1204 switch (tolower(*str
)) {
1206 slub_debug
|= SLAB_DEBUG_FREE
;
1209 slub_debug
|= SLAB_RED_ZONE
;
1212 slub_debug
|= SLAB_POISON
;
1215 slub_debug
|= SLAB_STORE_USER
;
1218 slub_debug
|= SLAB_TRACE
;
1221 slub_debug
|= SLAB_FAILSLAB
;
1224 printk(KERN_ERR
"slub_debug option '%c' "
1225 "unknown. skipped\n", *str
);
1231 slub_debug_slabs
= str
+ 1;
1236 __setup("slub_debug", setup_slub_debug
);
1238 static unsigned long kmem_cache_flags(unsigned long object_size
,
1239 unsigned long flags
, const char *name
,
1240 void (*ctor
)(void *))
1243 * Enable debugging if selected on the kernel commandline.
1245 if (slub_debug
&& (!slub_debug_slabs
|| (name
&&
1246 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)))))
1247 flags
|= slub_debug
;
1252 static inline void setup_object_debug(struct kmem_cache
*s
,
1253 struct page
*page
, void *object
) {}
1255 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1256 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1258 static inline struct kmem_cache_node
*free_debug_processing(
1259 struct kmem_cache
*s
, struct page
*page
, void *object
,
1260 unsigned long addr
, unsigned long *flags
) { return NULL
; }
1262 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1264 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1265 void *object
, u8 val
) { return 1; }
1266 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1267 struct page
*page
) {}
1268 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1269 static inline unsigned long kmem_cache_flags(unsigned long object_size
,
1270 unsigned long flags
, const char *name
,
1271 void (*ctor
)(void *))
1275 #define slub_debug 0
1277 #define disable_higher_order_debug 0
1279 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1281 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1283 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1285 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1288 static inline void kmalloc_large_node_hook(void *ptr
, size_t size
, gfp_t flags
)
1290 kmemleak_alloc(ptr
, size
, 1, flags
);
1293 static inline void kfree_hook(const void *x
)
1298 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1301 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1304 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
,
1305 flags
& gfp_allowed_mask
);
1308 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
1310 kmemleak_free_recursive(x
, s
->flags
);
1313 #endif /* CONFIG_SLUB_DEBUG */
1316 * Slab allocation and freeing
1318 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1319 struct kmem_cache_order_objects oo
)
1321 int order
= oo_order(oo
);
1323 flags
|= __GFP_NOTRACK
;
1325 if (node
== NUMA_NO_NODE
)
1326 return alloc_pages(flags
, order
);
1328 return alloc_pages_exact_node(node
, flags
, order
);
1331 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1334 struct kmem_cache_order_objects oo
= s
->oo
;
1337 flags
&= gfp_allowed_mask
;
1339 if (flags
& __GFP_WAIT
)
1342 flags
|= s
->allocflags
;
1345 * Let the initial higher-order allocation fail under memory pressure
1346 * so we fall-back to the minimum order allocation.
1348 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1350 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1351 if (unlikely(!page
)) {
1354 * Allocation may have failed due to fragmentation.
1355 * Try a lower order alloc if possible
1357 page
= alloc_slab_page(flags
, node
, oo
);
1360 stat(s
, ORDER_FALLBACK
);
1363 if (kmemcheck_enabled
&& page
1364 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1365 int pages
= 1 << oo_order(oo
);
1367 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1370 * Objects from caches that have a constructor don't get
1371 * cleared when they're allocated, so we need to do it here.
1374 kmemcheck_mark_uninitialized_pages(page
, pages
);
1376 kmemcheck_mark_unallocated_pages(page
, pages
);
1379 if (flags
& __GFP_WAIT
)
1380 local_irq_disable();
1384 page
->objects
= oo_objects(oo
);
1385 mod_zone_page_state(page_zone(page
),
1386 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1387 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1393 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1396 setup_object_debug(s
, page
, object
);
1397 if (unlikely(s
->ctor
))
1401 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1409 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1411 page
= allocate_slab(s
,
1412 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1416 order
= compound_order(page
);
1417 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1418 memcg_bind_pages(s
, order
);
1419 page
->slab_cache
= s
;
1420 __SetPageSlab(page
);
1421 if (page
->pfmemalloc
)
1422 SetPageSlabPfmemalloc(page
);
1424 start
= page_address(page
);
1426 if (unlikely(s
->flags
& SLAB_POISON
))
1427 memset(start
, POISON_INUSE
, PAGE_SIZE
<< order
);
1430 for_each_object(p
, s
, start
, page
->objects
) {
1431 setup_object(s
, page
, last
);
1432 set_freepointer(s
, last
, p
);
1435 setup_object(s
, page
, last
);
1436 set_freepointer(s
, last
, NULL
);
1438 page
->freelist
= start
;
1439 page
->inuse
= page
->objects
;
1445 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1447 int order
= compound_order(page
);
1448 int pages
= 1 << order
;
1450 if (kmem_cache_debug(s
)) {
1453 slab_pad_check(s
, page
);
1454 for_each_object(p
, s
, page_address(page
),
1456 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1459 kmemcheck_free_shadow(page
, compound_order(page
));
1461 mod_zone_page_state(page_zone(page
),
1462 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1463 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1466 __ClearPageSlabPfmemalloc(page
);
1467 __ClearPageSlab(page
);
1469 memcg_release_pages(s
, order
);
1470 page_mapcount_reset(page
);
1471 if (current
->reclaim_state
)
1472 current
->reclaim_state
->reclaimed_slab
+= pages
;
1473 __free_memcg_kmem_pages(page
, order
);
1476 #define need_reserve_slab_rcu \
1477 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1479 static void rcu_free_slab(struct rcu_head
*h
)
1483 if (need_reserve_slab_rcu
)
1484 page
= virt_to_head_page(h
);
1486 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1488 __free_slab(page
->slab_cache
, page
);
1491 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1493 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1494 struct rcu_head
*head
;
1496 if (need_reserve_slab_rcu
) {
1497 int order
= compound_order(page
);
1498 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1500 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1501 head
= page_address(page
) + offset
;
1504 * RCU free overloads the RCU head over the LRU
1506 head
= (void *)&page
->lru
;
1509 call_rcu(head
, rcu_free_slab
);
1511 __free_slab(s
, page
);
1514 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1516 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1521 * Management of partially allocated slabs.
1523 * list_lock must be held.
1525 static inline void add_partial(struct kmem_cache_node
*n
,
1526 struct page
*page
, int tail
)
1529 if (tail
== DEACTIVATE_TO_TAIL
)
1530 list_add_tail(&page
->lru
, &n
->partial
);
1532 list_add(&page
->lru
, &n
->partial
);
1536 * list_lock must be held.
1538 static inline void remove_partial(struct kmem_cache_node
*n
,
1541 list_del(&page
->lru
);
1546 * Remove slab from the partial list, freeze it and
1547 * return the pointer to the freelist.
1549 * Returns a list of objects or NULL if it fails.
1551 * Must hold list_lock since we modify the partial list.
1553 static inline void *acquire_slab(struct kmem_cache
*s
,
1554 struct kmem_cache_node
*n
, struct page
*page
,
1555 int mode
, int *objects
)
1558 unsigned long counters
;
1562 * Zap the freelist and set the frozen bit.
1563 * The old freelist is the list of objects for the
1564 * per cpu allocation list.
1566 freelist
= page
->freelist
;
1567 counters
= page
->counters
;
1568 new.counters
= counters
;
1569 *objects
= new.objects
- new.inuse
;
1571 new.inuse
= page
->objects
;
1572 new.freelist
= NULL
;
1574 new.freelist
= freelist
;
1577 VM_BUG_ON(new.frozen
);
1580 if (!__cmpxchg_double_slab(s
, page
,
1582 new.freelist
, new.counters
,
1586 remove_partial(n
, page
);
1591 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1592 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
);
1595 * Try to allocate a partial slab from a specific node.
1597 static void *get_partial_node(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1598 struct kmem_cache_cpu
*c
, gfp_t flags
)
1600 struct page
*page
, *page2
;
1601 void *object
= NULL
;
1606 * Racy check. If we mistakenly see no partial slabs then we
1607 * just allocate an empty slab. If we mistakenly try to get a
1608 * partial slab and there is none available then get_partials()
1611 if (!n
|| !n
->nr_partial
)
1614 spin_lock(&n
->list_lock
);
1615 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1618 if (!pfmemalloc_match(page
, flags
))
1621 t
= acquire_slab(s
, n
, page
, object
== NULL
, &objects
);
1625 available
+= objects
;
1628 stat(s
, ALLOC_FROM_PARTIAL
);
1631 put_cpu_partial(s
, page
, 0);
1632 stat(s
, CPU_PARTIAL_NODE
);
1634 if (!kmem_cache_has_cpu_partial(s
)
1635 || available
> s
->cpu_partial
/ 2)
1639 spin_unlock(&n
->list_lock
);
1644 * Get a page from somewhere. Search in increasing NUMA distances.
1646 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1647 struct kmem_cache_cpu
*c
)
1650 struct zonelist
*zonelist
;
1653 enum zone_type high_zoneidx
= gfp_zone(flags
);
1655 unsigned int cpuset_mems_cookie
;
1658 * The defrag ratio allows a configuration of the tradeoffs between
1659 * inter node defragmentation and node local allocations. A lower
1660 * defrag_ratio increases the tendency to do local allocations
1661 * instead of attempting to obtain partial slabs from other nodes.
1663 * If the defrag_ratio is set to 0 then kmalloc() always
1664 * returns node local objects. If the ratio is higher then kmalloc()
1665 * may return off node objects because partial slabs are obtained
1666 * from other nodes and filled up.
1668 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1669 * defrag_ratio = 1000) then every (well almost) allocation will
1670 * first attempt to defrag slab caches on other nodes. This means
1671 * scanning over all nodes to look for partial slabs which may be
1672 * expensive if we do it every time we are trying to find a slab
1673 * with available objects.
1675 if (!s
->remote_node_defrag_ratio
||
1676 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1680 cpuset_mems_cookie
= get_mems_allowed();
1681 zonelist
= node_zonelist(slab_node(), flags
);
1682 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1683 struct kmem_cache_node
*n
;
1685 n
= get_node(s
, zone_to_nid(zone
));
1687 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1688 n
->nr_partial
> s
->min_partial
) {
1689 object
= get_partial_node(s
, n
, c
, flags
);
1692 * Return the object even if
1693 * put_mems_allowed indicated that
1694 * the cpuset mems_allowed was
1695 * updated in parallel. It's a
1696 * harmless race between the alloc
1697 * and the cpuset update.
1699 put_mems_allowed(cpuset_mems_cookie
);
1704 } while (!put_mems_allowed(cpuset_mems_cookie
));
1710 * Get a partial page, lock it and return it.
1712 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1713 struct kmem_cache_cpu
*c
)
1716 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1718 object
= get_partial_node(s
, get_node(s
, searchnode
), c
, flags
);
1719 if (object
|| node
!= NUMA_NO_NODE
)
1722 return get_any_partial(s
, flags
, c
);
1725 #ifdef CONFIG_PREEMPT
1727 * Calculate the next globally unique transaction for disambiguiation
1728 * during cmpxchg. The transactions start with the cpu number and are then
1729 * incremented by CONFIG_NR_CPUS.
1731 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1734 * No preemption supported therefore also no need to check for
1740 static inline unsigned long next_tid(unsigned long tid
)
1742 return tid
+ TID_STEP
;
1745 static inline unsigned int tid_to_cpu(unsigned long tid
)
1747 return tid
% TID_STEP
;
1750 static inline unsigned long tid_to_event(unsigned long tid
)
1752 return tid
/ TID_STEP
;
1755 static inline unsigned int init_tid(int cpu
)
1760 static inline void note_cmpxchg_failure(const char *n
,
1761 const struct kmem_cache
*s
, unsigned long tid
)
1763 #ifdef SLUB_DEBUG_CMPXCHG
1764 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1766 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1768 #ifdef CONFIG_PREEMPT
1769 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1770 printk("due to cpu change %d -> %d\n",
1771 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1774 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1775 printk("due to cpu running other code. Event %ld->%ld\n",
1776 tid_to_event(tid
), tid_to_event(actual_tid
));
1778 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1779 actual_tid
, tid
, next_tid(tid
));
1781 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1784 static void init_kmem_cache_cpus(struct kmem_cache
*s
)
1788 for_each_possible_cpu(cpu
)
1789 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1793 * Remove the cpu slab
1795 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
,
1798 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1799 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1801 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1803 int tail
= DEACTIVATE_TO_HEAD
;
1807 if (page
->freelist
) {
1808 stat(s
, DEACTIVATE_REMOTE_FREES
);
1809 tail
= DEACTIVATE_TO_TAIL
;
1813 * Stage one: Free all available per cpu objects back
1814 * to the page freelist while it is still frozen. Leave the
1817 * There is no need to take the list->lock because the page
1820 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1822 unsigned long counters
;
1825 prior
= page
->freelist
;
1826 counters
= page
->counters
;
1827 set_freepointer(s
, freelist
, prior
);
1828 new.counters
= counters
;
1830 VM_BUG_ON(!new.frozen
);
1832 } while (!__cmpxchg_double_slab(s
, page
,
1834 freelist
, new.counters
,
1835 "drain percpu freelist"));
1837 freelist
= nextfree
;
1841 * Stage two: Ensure that the page is unfrozen while the
1842 * list presence reflects the actual number of objects
1845 * We setup the list membership and then perform a cmpxchg
1846 * with the count. If there is a mismatch then the page
1847 * is not unfrozen but the page is on the wrong list.
1849 * Then we restart the process which may have to remove
1850 * the page from the list that we just put it on again
1851 * because the number of objects in the slab may have
1856 old
.freelist
= page
->freelist
;
1857 old
.counters
= page
->counters
;
1858 VM_BUG_ON(!old
.frozen
);
1860 /* Determine target state of the slab */
1861 new.counters
= old
.counters
;
1864 set_freepointer(s
, freelist
, old
.freelist
);
1865 new.freelist
= freelist
;
1867 new.freelist
= old
.freelist
;
1871 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1873 else if (new.freelist
) {
1878 * Taking the spinlock removes the possiblity
1879 * that acquire_slab() will see a slab page that
1882 spin_lock(&n
->list_lock
);
1886 if (kmem_cache_debug(s
) && !lock
) {
1889 * This also ensures that the scanning of full
1890 * slabs from diagnostic functions will not see
1893 spin_lock(&n
->list_lock
);
1901 remove_partial(n
, page
);
1903 else if (l
== M_FULL
)
1905 remove_full(s
, page
);
1907 if (m
== M_PARTIAL
) {
1909 add_partial(n
, page
, tail
);
1912 } else if (m
== M_FULL
) {
1914 stat(s
, DEACTIVATE_FULL
);
1915 add_full(s
, n
, page
);
1921 if (!__cmpxchg_double_slab(s
, page
,
1922 old
.freelist
, old
.counters
,
1923 new.freelist
, new.counters
,
1928 spin_unlock(&n
->list_lock
);
1931 stat(s
, DEACTIVATE_EMPTY
);
1932 discard_slab(s
, page
);
1938 * Unfreeze all the cpu partial slabs.
1940 * This function must be called with interrupts disabled
1941 * for the cpu using c (or some other guarantee must be there
1942 * to guarantee no concurrent accesses).
1944 static void unfreeze_partials(struct kmem_cache
*s
,
1945 struct kmem_cache_cpu
*c
)
1947 #ifdef CONFIG_SLUB_CPU_PARTIAL
1948 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1949 struct page
*page
, *discard_page
= NULL
;
1951 while ((page
= c
->partial
)) {
1955 c
->partial
= page
->next
;
1957 n2
= get_node(s
, page_to_nid(page
));
1960 spin_unlock(&n
->list_lock
);
1963 spin_lock(&n
->list_lock
);
1968 old
.freelist
= page
->freelist
;
1969 old
.counters
= page
->counters
;
1970 VM_BUG_ON(!old
.frozen
);
1972 new.counters
= old
.counters
;
1973 new.freelist
= old
.freelist
;
1977 } while (!__cmpxchg_double_slab(s
, page
,
1978 old
.freelist
, old
.counters
,
1979 new.freelist
, new.counters
,
1980 "unfreezing slab"));
1982 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
)) {
1983 page
->next
= discard_page
;
1984 discard_page
= page
;
1986 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1987 stat(s
, FREE_ADD_PARTIAL
);
1992 spin_unlock(&n
->list_lock
);
1994 while (discard_page
) {
1995 page
= discard_page
;
1996 discard_page
= discard_page
->next
;
1998 stat(s
, DEACTIVATE_EMPTY
);
1999 discard_slab(s
, page
);
2006 * Put a page that was just frozen (in __slab_free) into a partial page
2007 * slot if available. This is done without interrupts disabled and without
2008 * preemption disabled. The cmpxchg is racy and may put the partial page
2009 * onto a random cpus partial slot.
2011 * If we did not find a slot then simply move all the partials to the
2012 * per node partial list.
2014 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
2016 #ifdef CONFIG_SLUB_CPU_PARTIAL
2017 struct page
*oldpage
;
2024 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
2027 pobjects
= oldpage
->pobjects
;
2028 pages
= oldpage
->pages
;
2029 if (drain
&& pobjects
> s
->cpu_partial
) {
2030 unsigned long flags
;
2032 * partial array is full. Move the existing
2033 * set to the per node partial list.
2035 local_irq_save(flags
);
2036 unfreeze_partials(s
, this_cpu_ptr(s
->cpu_slab
));
2037 local_irq_restore(flags
);
2041 stat(s
, CPU_PARTIAL_DRAIN
);
2046 pobjects
+= page
->objects
- page
->inuse
;
2048 page
->pages
= pages
;
2049 page
->pobjects
= pobjects
;
2050 page
->next
= oldpage
;
2052 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
)
2057 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2059 stat(s
, CPUSLAB_FLUSH
);
2060 deactivate_slab(s
, c
->page
, c
->freelist
);
2062 c
->tid
= next_tid(c
->tid
);
2070 * Called from IPI handler with interrupts disabled.
2072 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2074 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2080 unfreeze_partials(s
, c
);
2084 static void flush_cpu_slab(void *d
)
2086 struct kmem_cache
*s
= d
;
2088 __flush_cpu_slab(s
, smp_processor_id());
2091 static bool has_cpu_slab(int cpu
, void *info
)
2093 struct kmem_cache
*s
= info
;
2094 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2096 return c
->page
|| c
->partial
;
2099 static void flush_all(struct kmem_cache
*s
)
2101 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2105 * Check if the objects in a per cpu structure fit numa
2106 * locality expectations.
2108 static inline int node_match(struct page
*page
, int node
)
2111 if (!page
|| (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
))
2117 static int count_free(struct page
*page
)
2119 return page
->objects
- page
->inuse
;
2122 static unsigned long count_partial(struct kmem_cache_node
*n
,
2123 int (*get_count
)(struct page
*))
2125 unsigned long flags
;
2126 unsigned long x
= 0;
2129 spin_lock_irqsave(&n
->list_lock
, flags
);
2130 list_for_each_entry(page
, &n
->partial
, lru
)
2131 x
+= get_count(page
);
2132 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2136 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2138 #ifdef CONFIG_SLUB_DEBUG
2139 return atomic_long_read(&n
->total_objects
);
2145 static noinline
void
2146 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2151 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2153 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2154 "default order: %d, min order: %d\n", s
->name
, s
->object_size
,
2155 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2157 if (oo_order(s
->min
) > get_order(s
->object_size
))
2158 printk(KERN_WARNING
" %s debugging increased min order, use "
2159 "slub_debug=O to disable.\n", s
->name
);
2161 for_each_online_node(node
) {
2162 struct kmem_cache_node
*n
= get_node(s
, node
);
2163 unsigned long nr_slabs
;
2164 unsigned long nr_objs
;
2165 unsigned long nr_free
;
2170 nr_free
= count_partial(n
, count_free
);
2171 nr_slabs
= node_nr_slabs(n
);
2172 nr_objs
= node_nr_objs(n
);
2175 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2176 node
, nr_slabs
, nr_objs
, nr_free
);
2180 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2181 int node
, struct kmem_cache_cpu
**pc
)
2184 struct kmem_cache_cpu
*c
= *pc
;
2187 freelist
= get_partial(s
, flags
, node
, c
);
2192 page
= new_slab(s
, flags
, node
);
2194 c
= __this_cpu_ptr(s
->cpu_slab
);
2199 * No other reference to the page yet so we can
2200 * muck around with it freely without cmpxchg
2202 freelist
= page
->freelist
;
2203 page
->freelist
= NULL
;
2205 stat(s
, ALLOC_SLAB
);
2214 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
)
2216 if (unlikely(PageSlabPfmemalloc(page
)))
2217 return gfp_pfmemalloc_allowed(gfpflags
);
2223 * Check the page->freelist of a page and either transfer the freelist to the
2224 * per cpu freelist or deactivate the page.
2226 * The page is still frozen if the return value is not NULL.
2228 * If this function returns NULL then the page has been unfrozen.
2230 * This function must be called with interrupt disabled.
2232 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2235 unsigned long counters
;
2239 freelist
= page
->freelist
;
2240 counters
= page
->counters
;
2242 new.counters
= counters
;
2243 VM_BUG_ON(!new.frozen
);
2245 new.inuse
= page
->objects
;
2246 new.frozen
= freelist
!= NULL
;
2248 } while (!__cmpxchg_double_slab(s
, page
,
2257 * Slow path. The lockless freelist is empty or we need to perform
2260 * Processing is still very fast if new objects have been freed to the
2261 * regular freelist. In that case we simply take over the regular freelist
2262 * as the lockless freelist and zap the regular freelist.
2264 * If that is not working then we fall back to the partial lists. We take the
2265 * first element of the freelist as the object to allocate now and move the
2266 * rest of the freelist to the lockless freelist.
2268 * And if we were unable to get a new slab from the partial slab lists then
2269 * we need to allocate a new slab. This is the slowest path since it involves
2270 * a call to the page allocator and the setup of a new slab.
2272 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2273 unsigned long addr
, struct kmem_cache_cpu
*c
)
2277 unsigned long flags
;
2279 local_irq_save(flags
);
2280 #ifdef CONFIG_PREEMPT
2282 * We may have been preempted and rescheduled on a different
2283 * cpu before disabling interrupts. Need to reload cpu area
2286 c
= this_cpu_ptr(s
->cpu_slab
);
2294 if (unlikely(!node_match(page
, node
))) {
2295 stat(s
, ALLOC_NODE_MISMATCH
);
2296 deactivate_slab(s
, page
, c
->freelist
);
2303 * By rights, we should be searching for a slab page that was
2304 * PFMEMALLOC but right now, we are losing the pfmemalloc
2305 * information when the page leaves the per-cpu allocator
2307 if (unlikely(!pfmemalloc_match(page
, gfpflags
))) {
2308 deactivate_slab(s
, page
, c
->freelist
);
2314 /* must check again c->freelist in case of cpu migration or IRQ */
2315 freelist
= c
->freelist
;
2319 stat(s
, ALLOC_SLOWPATH
);
2321 freelist
= get_freelist(s
, page
);
2325 stat(s
, DEACTIVATE_BYPASS
);
2329 stat(s
, ALLOC_REFILL
);
2333 * freelist is pointing to the list of objects to be used.
2334 * page is pointing to the page from which the objects are obtained.
2335 * That page must be frozen for per cpu allocations to work.
2337 VM_BUG_ON(!c
->page
->frozen
);
2338 c
->freelist
= get_freepointer(s
, freelist
);
2339 c
->tid
= next_tid(c
->tid
);
2340 local_irq_restore(flags
);
2346 page
= c
->page
= c
->partial
;
2347 c
->partial
= page
->next
;
2348 stat(s
, CPU_PARTIAL_ALLOC
);
2353 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2355 if (unlikely(!freelist
)) {
2356 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2357 slab_out_of_memory(s
, gfpflags
, node
);
2359 local_irq_restore(flags
);
2364 if (likely(!kmem_cache_debug(s
) && pfmemalloc_match(page
, gfpflags
)))
2367 /* Only entered in the debug case */
2368 if (kmem_cache_debug(s
) &&
2369 !alloc_debug_processing(s
, page
, freelist
, addr
))
2370 goto new_slab
; /* Slab failed checks. Next slab needed */
2372 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2375 local_irq_restore(flags
);
2380 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2381 * have the fastpath folded into their functions. So no function call
2382 * overhead for requests that can be satisfied on the fastpath.
2384 * The fastpath works by first checking if the lockless freelist can be used.
2385 * If not then __slab_alloc is called for slow processing.
2387 * Otherwise we can simply pick the next object from the lockless free list.
2389 static __always_inline
void *slab_alloc_node(struct kmem_cache
*s
,
2390 gfp_t gfpflags
, int node
, unsigned long addr
)
2393 struct kmem_cache_cpu
*c
;
2397 if (slab_pre_alloc_hook(s
, gfpflags
))
2400 s
= memcg_kmem_get_cache(s
, gfpflags
);
2403 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2404 * enabled. We may switch back and forth between cpus while
2405 * reading from one cpu area. That does not matter as long
2406 * as we end up on the original cpu again when doing the cmpxchg.
2408 * Preemption is disabled for the retrieval of the tid because that
2409 * must occur from the current processor. We cannot allow rescheduling
2410 * on a different processor between the determination of the pointer
2411 * and the retrieval of the tid.
2414 c
= __this_cpu_ptr(s
->cpu_slab
);
2417 * The transaction ids are globally unique per cpu and per operation on
2418 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2419 * occurs on the right processor and that there was no operation on the
2420 * linked list in between.
2425 object
= c
->freelist
;
2427 if (unlikely(!object
|| !node_match(page
, node
)))
2428 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2431 void *next_object
= get_freepointer_safe(s
, object
);
2434 * The cmpxchg will only match if there was no additional
2435 * operation and if we are on the right processor.
2437 * The cmpxchg does the following atomically (without lock
2439 * 1. Relocate first pointer to the current per cpu area.
2440 * 2. Verify that tid and freelist have not been changed
2441 * 3. If they were not changed replace tid and freelist
2443 * Since this is without lock semantics the protection is only
2444 * against code executing on this cpu *not* from access by
2447 if (unlikely(!this_cpu_cmpxchg_double(
2448 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2450 next_object
, next_tid(tid
)))) {
2452 note_cmpxchg_failure("slab_alloc", s
, tid
);
2455 prefetch_freepointer(s
, next_object
);
2456 stat(s
, ALLOC_FASTPATH
);
2459 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2460 memset(object
, 0, s
->object_size
);
2462 slab_post_alloc_hook(s
, gfpflags
, object
);
2467 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2468 gfp_t gfpflags
, unsigned long addr
)
2470 return slab_alloc_node(s
, gfpflags
, NUMA_NO_NODE
, addr
);
2473 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2475 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2477 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
,
2482 EXPORT_SYMBOL(kmem_cache_alloc
);
2484 #ifdef CONFIG_TRACING
2485 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2487 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2488 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2491 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2495 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2497 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2499 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2500 s
->object_size
, s
->size
, gfpflags
, node
);
2504 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2506 #ifdef CONFIG_TRACING
2507 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2509 int node
, size_t size
)
2511 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2513 trace_kmalloc_node(_RET_IP_
, ret
,
2514 size
, s
->size
, gfpflags
, node
);
2517 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2522 * Slow patch handling. This may still be called frequently since objects
2523 * have a longer lifetime than the cpu slabs in most processing loads.
2525 * So we still attempt to reduce cache line usage. Just take the slab
2526 * lock and free the item. If there is no additional partial page
2527 * handling required then we can return immediately.
2529 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2530 void *x
, unsigned long addr
)
2533 void **object
= (void *)x
;
2536 unsigned long counters
;
2537 struct kmem_cache_node
*n
= NULL
;
2538 unsigned long uninitialized_var(flags
);
2540 stat(s
, FREE_SLOWPATH
);
2542 if (kmem_cache_debug(s
) &&
2543 !(n
= free_debug_processing(s
, page
, x
, addr
, &flags
)))
2548 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2551 prior
= page
->freelist
;
2552 counters
= page
->counters
;
2553 set_freepointer(s
, object
, prior
);
2554 new.counters
= counters
;
2555 was_frozen
= new.frozen
;
2557 if ((!new.inuse
|| !prior
) && !was_frozen
) {
2559 if (kmem_cache_has_cpu_partial(s
) && !prior
)
2562 * Slab was on no list before and will be
2564 * We can defer the list move and instead
2569 else { /* Needs to be taken off a list */
2571 n
= get_node(s
, page_to_nid(page
));
2573 * Speculatively acquire the list_lock.
2574 * If the cmpxchg does not succeed then we may
2575 * drop the list_lock without any processing.
2577 * Otherwise the list_lock will synchronize with
2578 * other processors updating the list of slabs.
2580 spin_lock_irqsave(&n
->list_lock
, flags
);
2585 } while (!cmpxchg_double_slab(s
, page
,
2587 object
, new.counters
,
2593 * If we just froze the page then put it onto the
2594 * per cpu partial list.
2596 if (new.frozen
&& !was_frozen
) {
2597 put_cpu_partial(s
, page
, 1);
2598 stat(s
, CPU_PARTIAL_FREE
);
2601 * The list lock was not taken therefore no list
2602 * activity can be necessary.
2605 stat(s
, FREE_FROZEN
);
2609 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
))
2613 * Objects left in the slab. If it was not on the partial list before
2616 if (!kmem_cache_has_cpu_partial(s
) && unlikely(!prior
)) {
2617 if (kmem_cache_debug(s
))
2618 remove_full(s
, page
);
2619 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2620 stat(s
, FREE_ADD_PARTIAL
);
2622 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2628 * Slab on the partial list.
2630 remove_partial(n
, page
);
2631 stat(s
, FREE_REMOVE_PARTIAL
);
2633 /* Slab must be on the full list */
2634 remove_full(s
, page
);
2636 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2638 discard_slab(s
, page
);
2642 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2643 * can perform fastpath freeing without additional function calls.
2645 * The fastpath is only possible if we are freeing to the current cpu slab
2646 * of this processor. This typically the case if we have just allocated
2649 * If fastpath is not possible then fall back to __slab_free where we deal
2650 * with all sorts of special processing.
2652 static __always_inline
void slab_free(struct kmem_cache
*s
,
2653 struct page
*page
, void *x
, unsigned long addr
)
2655 void **object
= (void *)x
;
2656 struct kmem_cache_cpu
*c
;
2659 slab_free_hook(s
, x
);
2663 * Determine the currently cpus per cpu slab.
2664 * The cpu may change afterward. However that does not matter since
2665 * data is retrieved via this pointer. If we are on the same cpu
2666 * during the cmpxchg then the free will succedd.
2669 c
= __this_cpu_ptr(s
->cpu_slab
);
2674 if (likely(page
== c
->page
)) {
2675 set_freepointer(s
, object
, c
->freelist
);
2677 if (unlikely(!this_cpu_cmpxchg_double(
2678 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2680 object
, next_tid(tid
)))) {
2682 note_cmpxchg_failure("slab_free", s
, tid
);
2685 stat(s
, FREE_FASTPATH
);
2687 __slab_free(s
, page
, x
, addr
);
2691 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2693 s
= cache_from_obj(s
, x
);
2696 slab_free(s
, virt_to_head_page(x
), x
, _RET_IP_
);
2697 trace_kmem_cache_free(_RET_IP_
, x
);
2699 EXPORT_SYMBOL(kmem_cache_free
);
2702 * Object placement in a slab is made very easy because we always start at
2703 * offset 0. If we tune the size of the object to the alignment then we can
2704 * get the required alignment by putting one properly sized object after
2707 * Notice that the allocation order determines the sizes of the per cpu
2708 * caches. Each processor has always one slab available for allocations.
2709 * Increasing the allocation order reduces the number of times that slabs
2710 * must be moved on and off the partial lists and is therefore a factor in
2715 * Mininum / Maximum order of slab pages. This influences locking overhead
2716 * and slab fragmentation. A higher order reduces the number of partial slabs
2717 * and increases the number of allocations possible without having to
2718 * take the list_lock.
2720 static int slub_min_order
;
2721 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2722 static int slub_min_objects
;
2725 * Merge control. If this is set then no merging of slab caches will occur.
2726 * (Could be removed. This was introduced to pacify the merge skeptics.)
2728 static int slub_nomerge
;
2731 * Calculate the order of allocation given an slab object size.
2733 * The order of allocation has significant impact on performance and other
2734 * system components. Generally order 0 allocations should be preferred since
2735 * order 0 does not cause fragmentation in the page allocator. Larger objects
2736 * be problematic to put into order 0 slabs because there may be too much
2737 * unused space left. We go to a higher order if more than 1/16th of the slab
2740 * In order to reach satisfactory performance we must ensure that a minimum
2741 * number of objects is in one slab. Otherwise we may generate too much
2742 * activity on the partial lists which requires taking the list_lock. This is
2743 * less a concern for large slabs though which are rarely used.
2745 * slub_max_order specifies the order where we begin to stop considering the
2746 * number of objects in a slab as critical. If we reach slub_max_order then
2747 * we try to keep the page order as low as possible. So we accept more waste
2748 * of space in favor of a small page order.
2750 * Higher order allocations also allow the placement of more objects in a
2751 * slab and thereby reduce object handling overhead. If the user has
2752 * requested a higher mininum order then we start with that one instead of
2753 * the smallest order which will fit the object.
2755 static inline int slab_order(int size
, int min_objects
,
2756 int max_order
, int fract_leftover
, int reserved
)
2760 int min_order
= slub_min_order
;
2762 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2763 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2765 for (order
= max(min_order
,
2766 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2767 order
<= max_order
; order
++) {
2769 unsigned long slab_size
= PAGE_SIZE
<< order
;
2771 if (slab_size
< min_objects
* size
+ reserved
)
2774 rem
= (slab_size
- reserved
) % size
;
2776 if (rem
<= slab_size
/ fract_leftover
)
2784 static inline int calculate_order(int size
, int reserved
)
2792 * Attempt to find best configuration for a slab. This
2793 * works by first attempting to generate a layout with
2794 * the best configuration and backing off gradually.
2796 * First we reduce the acceptable waste in a slab. Then
2797 * we reduce the minimum objects required in a slab.
2799 min_objects
= slub_min_objects
;
2801 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2802 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2803 min_objects
= min(min_objects
, max_objects
);
2805 while (min_objects
> 1) {
2807 while (fraction
>= 4) {
2808 order
= slab_order(size
, min_objects
,
2809 slub_max_order
, fraction
, reserved
);
2810 if (order
<= slub_max_order
)
2818 * We were unable to place multiple objects in a slab. Now
2819 * lets see if we can place a single object there.
2821 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2822 if (order
<= slub_max_order
)
2826 * Doh this slab cannot be placed using slub_max_order.
2828 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2829 if (order
< MAX_ORDER
)
2835 init_kmem_cache_node(struct kmem_cache_node
*n
)
2838 spin_lock_init(&n
->list_lock
);
2839 INIT_LIST_HEAD(&n
->partial
);
2840 #ifdef CONFIG_SLUB_DEBUG
2841 atomic_long_set(&n
->nr_slabs
, 0);
2842 atomic_long_set(&n
->total_objects
, 0);
2843 INIT_LIST_HEAD(&n
->full
);
2847 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2849 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2850 KMALLOC_SHIFT_HIGH
* sizeof(struct kmem_cache_cpu
));
2853 * Must align to double word boundary for the double cmpxchg
2854 * instructions to work; see __pcpu_double_call_return_bool().
2856 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2857 2 * sizeof(void *));
2862 init_kmem_cache_cpus(s
);
2867 static struct kmem_cache
*kmem_cache_node
;
2870 * No kmalloc_node yet so do it by hand. We know that this is the first
2871 * slab on the node for this slabcache. There are no concurrent accesses
2874 * Note that this function only works on the kmem_cache_node
2875 * when allocating for the kmem_cache_node. This is used for bootstrapping
2876 * memory on a fresh node that has no slab structures yet.
2878 static void early_kmem_cache_node_alloc(int node
)
2881 struct kmem_cache_node
*n
;
2883 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2885 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2888 if (page_to_nid(page
) != node
) {
2889 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2891 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2892 "in order to be able to continue\n");
2897 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2900 kmem_cache_node
->node
[node
] = n
;
2901 #ifdef CONFIG_SLUB_DEBUG
2902 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2903 init_tracking(kmem_cache_node
, n
);
2905 init_kmem_cache_node(n
);
2906 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2908 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2911 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2915 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2916 struct kmem_cache_node
*n
= s
->node
[node
];
2919 kmem_cache_free(kmem_cache_node
, n
);
2921 s
->node
[node
] = NULL
;
2925 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2929 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2930 struct kmem_cache_node
*n
;
2932 if (slab_state
== DOWN
) {
2933 early_kmem_cache_node_alloc(node
);
2936 n
= kmem_cache_alloc_node(kmem_cache_node
,
2940 free_kmem_cache_nodes(s
);
2945 init_kmem_cache_node(n
);
2950 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2952 if (min
< MIN_PARTIAL
)
2954 else if (min
> MAX_PARTIAL
)
2956 s
->min_partial
= min
;
2960 * calculate_sizes() determines the order and the distribution of data within
2963 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2965 unsigned long flags
= s
->flags
;
2966 unsigned long size
= s
->object_size
;
2970 * Round up object size to the next word boundary. We can only
2971 * place the free pointer at word boundaries and this determines
2972 * the possible location of the free pointer.
2974 size
= ALIGN(size
, sizeof(void *));
2976 #ifdef CONFIG_SLUB_DEBUG
2978 * Determine if we can poison the object itself. If the user of
2979 * the slab may touch the object after free or before allocation
2980 * then we should never poison the object itself.
2982 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2984 s
->flags
|= __OBJECT_POISON
;
2986 s
->flags
&= ~__OBJECT_POISON
;
2990 * If we are Redzoning then check if there is some space between the
2991 * end of the object and the free pointer. If not then add an
2992 * additional word to have some bytes to store Redzone information.
2994 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
2995 size
+= sizeof(void *);
2999 * With that we have determined the number of bytes in actual use
3000 * by the object. This is the potential offset to the free pointer.
3004 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
3007 * Relocate free pointer after the object if it is not
3008 * permitted to overwrite the first word of the object on
3011 * This is the case if we do RCU, have a constructor or
3012 * destructor or are poisoning the objects.
3015 size
+= sizeof(void *);
3018 #ifdef CONFIG_SLUB_DEBUG
3019 if (flags
& SLAB_STORE_USER
)
3021 * Need to store information about allocs and frees after
3024 size
+= 2 * sizeof(struct track
);
3026 if (flags
& SLAB_RED_ZONE
)
3028 * Add some empty padding so that we can catch
3029 * overwrites from earlier objects rather than let
3030 * tracking information or the free pointer be
3031 * corrupted if a user writes before the start
3034 size
+= sizeof(void *);
3038 * SLUB stores one object immediately after another beginning from
3039 * offset 0. In order to align the objects we have to simply size
3040 * each object to conform to the alignment.
3042 size
= ALIGN(size
, s
->align
);
3044 if (forced_order
>= 0)
3045 order
= forced_order
;
3047 order
= calculate_order(size
, s
->reserved
);
3054 s
->allocflags
|= __GFP_COMP
;
3056 if (s
->flags
& SLAB_CACHE_DMA
)
3057 s
->allocflags
|= GFP_DMA
;
3059 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3060 s
->allocflags
|= __GFP_RECLAIMABLE
;
3063 * Determine the number of objects per slab
3065 s
->oo
= oo_make(order
, size
, s
->reserved
);
3066 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
3067 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3070 return !!oo_objects(s
->oo
);
3073 static int kmem_cache_open(struct kmem_cache
*s
, unsigned long flags
)
3075 s
->flags
= kmem_cache_flags(s
->size
, flags
, s
->name
, s
->ctor
);
3078 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3079 s
->reserved
= sizeof(struct rcu_head
);
3081 if (!calculate_sizes(s
, -1))
3083 if (disable_higher_order_debug
) {
3085 * Disable debugging flags that store metadata if the min slab
3088 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3089 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3091 if (!calculate_sizes(s
, -1))
3096 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3097 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3098 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3099 /* Enable fast mode */
3100 s
->flags
|= __CMPXCHG_DOUBLE
;
3104 * The larger the object size is, the more pages we want on the partial
3105 * list to avoid pounding the page allocator excessively.
3107 set_min_partial(s
, ilog2(s
->size
) / 2);
3110 * cpu_partial determined the maximum number of objects kept in the
3111 * per cpu partial lists of a processor.
3113 * Per cpu partial lists mainly contain slabs that just have one
3114 * object freed. If they are used for allocation then they can be
3115 * filled up again with minimal effort. The slab will never hit the
3116 * per node partial lists and therefore no locking will be required.
3118 * This setting also determines
3120 * A) The number of objects from per cpu partial slabs dumped to the
3121 * per node list when we reach the limit.
3122 * B) The number of objects in cpu partial slabs to extract from the
3123 * per node list when we run out of per cpu objects. We only fetch
3124 * 50% to keep some capacity around for frees.
3126 if (!kmem_cache_has_cpu_partial(s
))
3128 else if (s
->size
>= PAGE_SIZE
)
3130 else if (s
->size
>= 1024)
3132 else if (s
->size
>= 256)
3133 s
->cpu_partial
= 13;
3135 s
->cpu_partial
= 30;
3138 s
->remote_node_defrag_ratio
= 1000;
3140 if (!init_kmem_cache_nodes(s
))
3143 if (alloc_kmem_cache_cpus(s
))
3146 free_kmem_cache_nodes(s
);
3148 if (flags
& SLAB_PANIC
)
3149 panic("Cannot create slab %s size=%lu realsize=%u "
3150 "order=%u offset=%u flags=%lx\n",
3151 s
->name
, (unsigned long)s
->size
, s
->size
,
3152 oo_order(s
->oo
), s
->offset
, flags
);
3156 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3159 #ifdef CONFIG_SLUB_DEBUG
3160 void *addr
= page_address(page
);
3162 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3163 sizeof(long), GFP_ATOMIC
);
3166 slab_err(s
, page
, text
, s
->name
);
3169 get_map(s
, page
, map
);
3170 for_each_object(p
, s
, addr
, page
->objects
) {
3172 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3173 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3175 print_tracking(s
, p
);
3184 * Attempt to free all partial slabs on a node.
3185 * This is called from kmem_cache_close(). We must be the last thread
3186 * using the cache and therefore we do not need to lock anymore.
3188 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3190 struct page
*page
, *h
;
3192 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3194 remove_partial(n
, page
);
3195 discard_slab(s
, page
);
3197 list_slab_objects(s
, page
,
3198 "Objects remaining in %s on kmem_cache_close()");
3204 * Release all resources used by a slab cache.
3206 static inline int kmem_cache_close(struct kmem_cache
*s
)
3211 /* Attempt to free all objects */
3212 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3213 struct kmem_cache_node
*n
= get_node(s
, node
);
3216 if (n
->nr_partial
|| slabs_node(s
, node
))
3219 free_percpu(s
->cpu_slab
);
3220 free_kmem_cache_nodes(s
);
3224 int __kmem_cache_shutdown(struct kmem_cache
*s
)
3226 int rc
= kmem_cache_close(s
);
3230 * We do the same lock strategy around sysfs_slab_add, see
3231 * __kmem_cache_create. Because this is pretty much the last
3232 * operation we do and the lock will be released shortly after
3233 * that in slab_common.c, we could just move sysfs_slab_remove
3234 * to a later point in common code. We should do that when we
3235 * have a common sysfs framework for all allocators.
3237 mutex_unlock(&slab_mutex
);
3238 sysfs_slab_remove(s
);
3239 mutex_lock(&slab_mutex
);
3245 /********************************************************************
3247 *******************************************************************/
3249 static int __init
setup_slub_min_order(char *str
)
3251 get_option(&str
, &slub_min_order
);
3256 __setup("slub_min_order=", setup_slub_min_order
);
3258 static int __init
setup_slub_max_order(char *str
)
3260 get_option(&str
, &slub_max_order
);
3261 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3266 __setup("slub_max_order=", setup_slub_max_order
);
3268 static int __init
setup_slub_min_objects(char *str
)
3270 get_option(&str
, &slub_min_objects
);
3275 __setup("slub_min_objects=", setup_slub_min_objects
);
3277 static int __init
setup_slub_nomerge(char *str
)
3283 __setup("slub_nomerge", setup_slub_nomerge
);
3285 void *__kmalloc(size_t size
, gfp_t flags
)
3287 struct kmem_cache
*s
;
3290 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3291 return kmalloc_large(size
, flags
);
3293 s
= kmalloc_slab(size
, flags
);
3295 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3298 ret
= slab_alloc(s
, flags
, _RET_IP_
);
3300 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3304 EXPORT_SYMBOL(__kmalloc
);
3307 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3312 flags
|= __GFP_COMP
| __GFP_NOTRACK
| __GFP_KMEMCG
;
3313 page
= alloc_pages_node(node
, flags
, get_order(size
));
3315 ptr
= page_address(page
);
3317 kmalloc_large_node_hook(ptr
, size
, flags
);
3321 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3323 struct kmem_cache
*s
;
3326 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3327 ret
= kmalloc_large_node(size
, flags
, node
);
3329 trace_kmalloc_node(_RET_IP_
, ret
,
3330 size
, PAGE_SIZE
<< get_order(size
),
3336 s
= kmalloc_slab(size
, flags
);
3338 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3341 ret
= slab_alloc_node(s
, flags
, node
, _RET_IP_
);
3343 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3347 EXPORT_SYMBOL(__kmalloc_node
);
3350 size_t ksize(const void *object
)
3354 if (unlikely(object
== ZERO_SIZE_PTR
))
3357 page
= virt_to_head_page(object
);
3359 if (unlikely(!PageSlab(page
))) {
3360 WARN_ON(!PageCompound(page
));
3361 return PAGE_SIZE
<< compound_order(page
);
3364 return slab_ksize(page
->slab_cache
);
3366 EXPORT_SYMBOL(ksize
);
3368 void kfree(const void *x
)
3371 void *object
= (void *)x
;
3373 trace_kfree(_RET_IP_
, x
);
3375 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3378 page
= virt_to_head_page(x
);
3379 if (unlikely(!PageSlab(page
))) {
3380 BUG_ON(!PageCompound(page
));
3382 __free_memcg_kmem_pages(page
, compound_order(page
));
3385 slab_free(page
->slab_cache
, page
, object
, _RET_IP_
);
3387 EXPORT_SYMBOL(kfree
);
3390 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3391 * the remaining slabs by the number of items in use. The slabs with the
3392 * most items in use come first. New allocations will then fill those up
3393 * and thus they can be removed from the partial lists.
3395 * The slabs with the least items are placed last. This results in them
3396 * being allocated from last increasing the chance that the last objects
3397 * are freed in them.
3399 int kmem_cache_shrink(struct kmem_cache
*s
)
3403 struct kmem_cache_node
*n
;
3406 int objects
= oo_objects(s
->max
);
3407 struct list_head
*slabs_by_inuse
=
3408 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3409 unsigned long flags
;
3411 if (!slabs_by_inuse
)
3415 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3416 n
= get_node(s
, node
);
3421 for (i
= 0; i
< objects
; i
++)
3422 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3424 spin_lock_irqsave(&n
->list_lock
, flags
);
3427 * Build lists indexed by the items in use in each slab.
3429 * Note that concurrent frees may occur while we hold the
3430 * list_lock. page->inuse here is the upper limit.
3432 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3433 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3439 * Rebuild the partial list with the slabs filled up most
3440 * first and the least used slabs at the end.
3442 for (i
= objects
- 1; i
> 0; i
--)
3443 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3445 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3447 /* Release empty slabs */
3448 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3449 discard_slab(s
, page
);
3452 kfree(slabs_by_inuse
);
3455 EXPORT_SYMBOL(kmem_cache_shrink
);
3457 static int slab_mem_going_offline_callback(void *arg
)
3459 struct kmem_cache
*s
;
3461 mutex_lock(&slab_mutex
);
3462 list_for_each_entry(s
, &slab_caches
, list
)
3463 kmem_cache_shrink(s
);
3464 mutex_unlock(&slab_mutex
);
3469 static void slab_mem_offline_callback(void *arg
)
3471 struct kmem_cache_node
*n
;
3472 struct kmem_cache
*s
;
3473 struct memory_notify
*marg
= arg
;
3476 offline_node
= marg
->status_change_nid_normal
;
3479 * If the node still has available memory. we need kmem_cache_node
3482 if (offline_node
< 0)
3485 mutex_lock(&slab_mutex
);
3486 list_for_each_entry(s
, &slab_caches
, list
) {
3487 n
= get_node(s
, offline_node
);
3490 * if n->nr_slabs > 0, slabs still exist on the node
3491 * that is going down. We were unable to free them,
3492 * and offline_pages() function shouldn't call this
3493 * callback. So, we must fail.
3495 BUG_ON(slabs_node(s
, offline_node
));
3497 s
->node
[offline_node
] = NULL
;
3498 kmem_cache_free(kmem_cache_node
, n
);
3501 mutex_unlock(&slab_mutex
);
3504 static int slab_mem_going_online_callback(void *arg
)
3506 struct kmem_cache_node
*n
;
3507 struct kmem_cache
*s
;
3508 struct memory_notify
*marg
= arg
;
3509 int nid
= marg
->status_change_nid_normal
;
3513 * If the node's memory is already available, then kmem_cache_node is
3514 * already created. Nothing to do.
3520 * We are bringing a node online. No memory is available yet. We must
3521 * allocate a kmem_cache_node structure in order to bring the node
3524 mutex_lock(&slab_mutex
);
3525 list_for_each_entry(s
, &slab_caches
, list
) {
3527 * XXX: kmem_cache_alloc_node will fallback to other nodes
3528 * since memory is not yet available from the node that
3531 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3536 init_kmem_cache_node(n
);
3540 mutex_unlock(&slab_mutex
);
3544 static int slab_memory_callback(struct notifier_block
*self
,
3545 unsigned long action
, void *arg
)
3550 case MEM_GOING_ONLINE
:
3551 ret
= slab_mem_going_online_callback(arg
);
3553 case MEM_GOING_OFFLINE
:
3554 ret
= slab_mem_going_offline_callback(arg
);
3557 case MEM_CANCEL_ONLINE
:
3558 slab_mem_offline_callback(arg
);
3561 case MEM_CANCEL_OFFLINE
:
3565 ret
= notifier_from_errno(ret
);
3571 static struct notifier_block slab_memory_callback_nb
= {
3572 .notifier_call
= slab_memory_callback
,
3573 .priority
= SLAB_CALLBACK_PRI
,
3576 /********************************************************************
3577 * Basic setup of slabs
3578 *******************************************************************/
3581 * Used for early kmem_cache structures that were allocated using
3582 * the page allocator. Allocate them properly then fix up the pointers
3583 * that may be pointing to the wrong kmem_cache structure.
3586 static struct kmem_cache
* __init
bootstrap(struct kmem_cache
*static_cache
)
3589 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
3591 memcpy(s
, static_cache
, kmem_cache
->object_size
);
3594 * This runs very early, and only the boot processor is supposed to be
3595 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3598 __flush_cpu_slab(s
, smp_processor_id());
3599 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3600 struct kmem_cache_node
*n
= get_node(s
, node
);
3604 list_for_each_entry(p
, &n
->partial
, lru
)
3607 #ifdef CONFIG_SLUB_DEBUG
3608 list_for_each_entry(p
, &n
->full
, lru
)
3613 list_add(&s
->list
, &slab_caches
);
3617 void __init
kmem_cache_init(void)
3619 static __initdata
struct kmem_cache boot_kmem_cache
,
3620 boot_kmem_cache_node
;
3622 if (debug_guardpage_minorder())
3625 kmem_cache_node
= &boot_kmem_cache_node
;
3626 kmem_cache
= &boot_kmem_cache
;
3628 create_boot_cache(kmem_cache_node
, "kmem_cache_node",
3629 sizeof(struct kmem_cache_node
), SLAB_HWCACHE_ALIGN
);
3631 register_hotmemory_notifier(&slab_memory_callback_nb
);
3633 /* Able to allocate the per node structures */
3634 slab_state
= PARTIAL
;
3636 create_boot_cache(kmem_cache
, "kmem_cache",
3637 offsetof(struct kmem_cache
, node
) +
3638 nr_node_ids
* sizeof(struct kmem_cache_node
*),
3639 SLAB_HWCACHE_ALIGN
);
3641 kmem_cache
= bootstrap(&boot_kmem_cache
);
3644 * Allocate kmem_cache_node properly from the kmem_cache slab.
3645 * kmem_cache_node is separately allocated so no need to
3646 * update any list pointers.
3648 kmem_cache_node
= bootstrap(&boot_kmem_cache_node
);
3650 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3651 create_kmalloc_caches(0);
3654 register_cpu_notifier(&slab_notifier
);
3658 "SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d,"
3659 " CPUs=%d, Nodes=%d\n",
3661 slub_min_order
, slub_max_order
, slub_min_objects
,
3662 nr_cpu_ids
, nr_node_ids
);
3665 void __init
kmem_cache_init_late(void)
3670 * Find a mergeable slab cache
3672 static int slab_unmergeable(struct kmem_cache
*s
)
3674 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3681 * We may have set a slab to be unmergeable during bootstrap.
3683 if (s
->refcount
< 0)
3689 static struct kmem_cache
*find_mergeable(struct mem_cgroup
*memcg
, size_t size
,
3690 size_t align
, unsigned long flags
, const char *name
,
3691 void (*ctor
)(void *))
3693 struct kmem_cache
*s
;
3695 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3701 size
= ALIGN(size
, sizeof(void *));
3702 align
= calculate_alignment(flags
, align
, size
);
3703 size
= ALIGN(size
, align
);
3704 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3706 list_for_each_entry(s
, &slab_caches
, list
) {
3707 if (slab_unmergeable(s
))
3713 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3716 * Check if alignment is compatible.
3717 * Courtesy of Adrian Drzewiecki
3719 if ((s
->size
& ~(align
- 1)) != s
->size
)
3722 if (s
->size
- size
>= sizeof(void *))
3725 if (!cache_match_memcg(s
, memcg
))
3734 __kmem_cache_alias(struct mem_cgroup
*memcg
, const char *name
, size_t size
,
3735 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3737 struct kmem_cache
*s
;
3739 s
= find_mergeable(memcg
, size
, align
, flags
, name
, ctor
);
3743 * Adjust the object sizes so that we clear
3744 * the complete object on kzalloc.
3746 s
->object_size
= max(s
->object_size
, (int)size
);
3747 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3749 if (sysfs_slab_alias(s
, name
)) {
3758 int __kmem_cache_create(struct kmem_cache
*s
, unsigned long flags
)
3762 err
= kmem_cache_open(s
, flags
);
3766 /* Mutex is not taken during early boot */
3767 if (slab_state
<= UP
)
3770 memcg_propagate_slab_attrs(s
);
3771 mutex_unlock(&slab_mutex
);
3772 err
= sysfs_slab_add(s
);
3773 mutex_lock(&slab_mutex
);
3776 kmem_cache_close(s
);
3783 * Use the cpu notifier to insure that the cpu slabs are flushed when
3786 static int slab_cpuup_callback(struct notifier_block
*nfb
,
3787 unsigned long action
, void *hcpu
)
3789 long cpu
= (long)hcpu
;
3790 struct kmem_cache
*s
;
3791 unsigned long flags
;
3794 case CPU_UP_CANCELED
:
3795 case CPU_UP_CANCELED_FROZEN
:
3797 case CPU_DEAD_FROZEN
:
3798 mutex_lock(&slab_mutex
);
3799 list_for_each_entry(s
, &slab_caches
, list
) {
3800 local_irq_save(flags
);
3801 __flush_cpu_slab(s
, cpu
);
3802 local_irq_restore(flags
);
3804 mutex_unlock(&slab_mutex
);
3812 static struct notifier_block slab_notifier
= {
3813 .notifier_call
= slab_cpuup_callback
3818 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3820 struct kmem_cache
*s
;
3823 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3824 return kmalloc_large(size
, gfpflags
);
3826 s
= kmalloc_slab(size
, gfpflags
);
3828 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3831 ret
= slab_alloc(s
, gfpflags
, caller
);
3833 /* Honor the call site pointer we received. */
3834 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3840 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3841 int node
, unsigned long caller
)
3843 struct kmem_cache
*s
;
3846 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3847 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3849 trace_kmalloc_node(caller
, ret
,
3850 size
, PAGE_SIZE
<< get_order(size
),
3856 s
= kmalloc_slab(size
, gfpflags
);
3858 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3861 ret
= slab_alloc_node(s
, gfpflags
, node
, caller
);
3863 /* Honor the call site pointer we received. */
3864 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3871 static int count_inuse(struct page
*page
)
3876 static int count_total(struct page
*page
)
3878 return page
->objects
;
3882 #ifdef CONFIG_SLUB_DEBUG
3883 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3887 void *addr
= page_address(page
);
3889 if (!check_slab(s
, page
) ||
3890 !on_freelist(s
, page
, NULL
))
3893 /* Now we know that a valid freelist exists */
3894 bitmap_zero(map
, page
->objects
);
3896 get_map(s
, page
, map
);
3897 for_each_object(p
, s
, addr
, page
->objects
) {
3898 if (test_bit(slab_index(p
, s
, addr
), map
))
3899 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
3903 for_each_object(p
, s
, addr
, page
->objects
)
3904 if (!test_bit(slab_index(p
, s
, addr
), map
))
3905 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
3910 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3914 validate_slab(s
, page
, map
);
3918 static int validate_slab_node(struct kmem_cache
*s
,
3919 struct kmem_cache_node
*n
, unsigned long *map
)
3921 unsigned long count
= 0;
3923 unsigned long flags
;
3925 spin_lock_irqsave(&n
->list_lock
, flags
);
3927 list_for_each_entry(page
, &n
->partial
, lru
) {
3928 validate_slab_slab(s
, page
, map
);
3931 if (count
!= n
->nr_partial
)
3932 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3933 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3935 if (!(s
->flags
& SLAB_STORE_USER
))
3938 list_for_each_entry(page
, &n
->full
, lru
) {
3939 validate_slab_slab(s
, page
, map
);
3942 if (count
!= atomic_long_read(&n
->nr_slabs
))
3943 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3944 "counter=%ld\n", s
->name
, count
,
3945 atomic_long_read(&n
->nr_slabs
));
3948 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3952 static long validate_slab_cache(struct kmem_cache
*s
)
3955 unsigned long count
= 0;
3956 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3957 sizeof(unsigned long), GFP_KERNEL
);
3963 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3964 struct kmem_cache_node
*n
= get_node(s
, node
);
3966 count
+= validate_slab_node(s
, n
, map
);
3972 * Generate lists of code addresses where slabcache objects are allocated
3977 unsigned long count
;
3984 DECLARE_BITMAP(cpus
, NR_CPUS
);
3990 unsigned long count
;
3991 struct location
*loc
;
3994 static void free_loc_track(struct loc_track
*t
)
3997 free_pages((unsigned long)t
->loc
,
3998 get_order(sizeof(struct location
) * t
->max
));
4001 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4006 order
= get_order(sizeof(struct location
) * max
);
4008 l
= (void *)__get_free_pages(flags
, order
);
4013 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4021 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4022 const struct track
*track
)
4024 long start
, end
, pos
;
4026 unsigned long caddr
;
4027 unsigned long age
= jiffies
- track
->when
;
4033 pos
= start
+ (end
- start
+ 1) / 2;
4036 * There is nothing at "end". If we end up there
4037 * we need to add something to before end.
4042 caddr
= t
->loc
[pos
].addr
;
4043 if (track
->addr
== caddr
) {
4049 if (age
< l
->min_time
)
4051 if (age
> l
->max_time
)
4054 if (track
->pid
< l
->min_pid
)
4055 l
->min_pid
= track
->pid
;
4056 if (track
->pid
> l
->max_pid
)
4057 l
->max_pid
= track
->pid
;
4059 cpumask_set_cpu(track
->cpu
,
4060 to_cpumask(l
->cpus
));
4062 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4066 if (track
->addr
< caddr
)
4073 * Not found. Insert new tracking element.
4075 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4081 (t
->count
- pos
) * sizeof(struct location
));
4084 l
->addr
= track
->addr
;
4088 l
->min_pid
= track
->pid
;
4089 l
->max_pid
= track
->pid
;
4090 cpumask_clear(to_cpumask(l
->cpus
));
4091 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4092 nodes_clear(l
->nodes
);
4093 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4097 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4098 struct page
*page
, enum track_item alloc
,
4101 void *addr
= page_address(page
);
4104 bitmap_zero(map
, page
->objects
);
4105 get_map(s
, page
, map
);
4107 for_each_object(p
, s
, addr
, page
->objects
)
4108 if (!test_bit(slab_index(p
, s
, addr
), map
))
4109 add_location(t
, s
, get_track(s
, p
, alloc
));
4112 static int list_locations(struct kmem_cache
*s
, char *buf
,
4113 enum track_item alloc
)
4117 struct loc_track t
= { 0, 0, NULL
};
4119 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4120 sizeof(unsigned long), GFP_KERNEL
);
4122 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4125 return sprintf(buf
, "Out of memory\n");
4127 /* Push back cpu slabs */
4130 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4131 struct kmem_cache_node
*n
= get_node(s
, node
);
4132 unsigned long flags
;
4135 if (!atomic_long_read(&n
->nr_slabs
))
4138 spin_lock_irqsave(&n
->list_lock
, flags
);
4139 list_for_each_entry(page
, &n
->partial
, lru
)
4140 process_slab(&t
, s
, page
, alloc
, map
);
4141 list_for_each_entry(page
, &n
->full
, lru
)
4142 process_slab(&t
, s
, page
, alloc
, map
);
4143 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4146 for (i
= 0; i
< t
.count
; i
++) {
4147 struct location
*l
= &t
.loc
[i
];
4149 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4151 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4154 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4156 len
+= sprintf(buf
+ len
, "<not-available>");
4158 if (l
->sum_time
!= l
->min_time
) {
4159 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4161 (long)div_u64(l
->sum_time
, l
->count
),
4164 len
+= sprintf(buf
+ len
, " age=%ld",
4167 if (l
->min_pid
!= l
->max_pid
)
4168 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4169 l
->min_pid
, l
->max_pid
);
4171 len
+= sprintf(buf
+ len
, " pid=%ld",
4174 if (num_online_cpus() > 1 &&
4175 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4176 len
< PAGE_SIZE
- 60) {
4177 len
+= sprintf(buf
+ len
, " cpus=");
4178 len
+= cpulist_scnprintf(buf
+ len
,
4179 PAGE_SIZE
- len
- 50,
4180 to_cpumask(l
->cpus
));
4183 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4184 len
< PAGE_SIZE
- 60) {
4185 len
+= sprintf(buf
+ len
, " nodes=");
4186 len
+= nodelist_scnprintf(buf
+ len
,
4187 PAGE_SIZE
- len
- 50,
4191 len
+= sprintf(buf
+ len
, "\n");
4197 len
+= sprintf(buf
, "No data\n");
4202 #ifdef SLUB_RESILIENCY_TEST
4203 static void resiliency_test(void)
4207 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || KMALLOC_SHIFT_HIGH
< 10);
4209 printk(KERN_ERR
"SLUB resiliency testing\n");
4210 printk(KERN_ERR
"-----------------------\n");
4211 printk(KERN_ERR
"A. Corruption after allocation\n");
4213 p
= kzalloc(16, GFP_KERNEL
);
4215 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4216 " 0x12->0x%p\n\n", p
+ 16);
4218 validate_slab_cache(kmalloc_caches
[4]);
4220 /* Hmmm... The next two are dangerous */
4221 p
= kzalloc(32, GFP_KERNEL
);
4222 p
[32 + sizeof(void *)] = 0x34;
4223 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4224 " 0x34 -> -0x%p\n", p
);
4226 "If allocated object is overwritten then not detectable\n\n");
4228 validate_slab_cache(kmalloc_caches
[5]);
4229 p
= kzalloc(64, GFP_KERNEL
);
4230 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4232 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4235 "If allocated object is overwritten then not detectable\n\n");
4236 validate_slab_cache(kmalloc_caches
[6]);
4238 printk(KERN_ERR
"\nB. Corruption after free\n");
4239 p
= kzalloc(128, GFP_KERNEL
);
4242 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4243 validate_slab_cache(kmalloc_caches
[7]);
4245 p
= kzalloc(256, GFP_KERNEL
);
4248 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4250 validate_slab_cache(kmalloc_caches
[8]);
4252 p
= kzalloc(512, GFP_KERNEL
);
4255 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4256 validate_slab_cache(kmalloc_caches
[9]);
4260 static void resiliency_test(void) {};
4265 enum slab_stat_type
{
4266 SL_ALL
, /* All slabs */
4267 SL_PARTIAL
, /* Only partially allocated slabs */
4268 SL_CPU
, /* Only slabs used for cpu caches */
4269 SL_OBJECTS
, /* Determine allocated objects not slabs */
4270 SL_TOTAL
/* Determine object capacity not slabs */
4273 #define SO_ALL (1 << SL_ALL)
4274 #define SO_PARTIAL (1 << SL_PARTIAL)
4275 #define SO_CPU (1 << SL_CPU)
4276 #define SO_OBJECTS (1 << SL_OBJECTS)
4277 #define SO_TOTAL (1 << SL_TOTAL)
4279 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4280 char *buf
, unsigned long flags
)
4282 unsigned long total
= 0;
4285 unsigned long *nodes
;
4287 nodes
= kzalloc(sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4291 if (flags
& SO_CPU
) {
4294 for_each_possible_cpu(cpu
) {
4295 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
,
4300 page
= ACCESS_ONCE(c
->page
);
4304 node
= page_to_nid(page
);
4305 if (flags
& SO_TOTAL
)
4307 else if (flags
& SO_OBJECTS
)
4315 page
= ACCESS_ONCE(c
->partial
);
4324 lock_memory_hotplug();
4325 #ifdef CONFIG_SLUB_DEBUG
4326 if (flags
& SO_ALL
) {
4327 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4328 struct kmem_cache_node
*n
= get_node(s
, node
);
4330 if (flags
& SO_TOTAL
)
4331 x
= atomic_long_read(&n
->total_objects
);
4332 else if (flags
& SO_OBJECTS
)
4333 x
= atomic_long_read(&n
->total_objects
) -
4334 count_partial(n
, count_free
);
4336 x
= atomic_long_read(&n
->nr_slabs
);
4343 if (flags
& SO_PARTIAL
) {
4344 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4345 struct kmem_cache_node
*n
= get_node(s
, node
);
4347 if (flags
& SO_TOTAL
)
4348 x
= count_partial(n
, count_total
);
4349 else if (flags
& SO_OBJECTS
)
4350 x
= count_partial(n
, count_inuse
);
4357 x
= sprintf(buf
, "%lu", total
);
4359 for_each_node_state(node
, N_NORMAL_MEMORY
)
4361 x
+= sprintf(buf
+ x
, " N%d=%lu",
4364 unlock_memory_hotplug();
4366 return x
+ sprintf(buf
+ x
, "\n");
4369 #ifdef CONFIG_SLUB_DEBUG
4370 static int any_slab_objects(struct kmem_cache
*s
)
4374 for_each_online_node(node
) {
4375 struct kmem_cache_node
*n
= get_node(s
, node
);
4380 if (atomic_long_read(&n
->total_objects
))
4387 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4388 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4390 struct slab_attribute
{
4391 struct attribute attr
;
4392 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4393 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4396 #define SLAB_ATTR_RO(_name) \
4397 static struct slab_attribute _name##_attr = \
4398 __ATTR(_name, 0400, _name##_show, NULL)
4400 #define SLAB_ATTR(_name) \
4401 static struct slab_attribute _name##_attr = \
4402 __ATTR(_name, 0600, _name##_show, _name##_store)
4404 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4406 return sprintf(buf
, "%d\n", s
->size
);
4408 SLAB_ATTR_RO(slab_size
);
4410 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4412 return sprintf(buf
, "%d\n", s
->align
);
4414 SLAB_ATTR_RO(align
);
4416 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4418 return sprintf(buf
, "%d\n", s
->object_size
);
4420 SLAB_ATTR_RO(object_size
);
4422 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4424 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4426 SLAB_ATTR_RO(objs_per_slab
);
4428 static ssize_t
order_store(struct kmem_cache
*s
,
4429 const char *buf
, size_t length
)
4431 unsigned long order
;
4434 err
= kstrtoul(buf
, 10, &order
);
4438 if (order
> slub_max_order
|| order
< slub_min_order
)
4441 calculate_sizes(s
, order
);
4445 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4447 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4451 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4453 return sprintf(buf
, "%lu\n", s
->min_partial
);
4456 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4462 err
= kstrtoul(buf
, 10, &min
);
4466 set_min_partial(s
, min
);
4469 SLAB_ATTR(min_partial
);
4471 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4473 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4476 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4479 unsigned long objects
;
4482 err
= kstrtoul(buf
, 10, &objects
);
4485 if (objects
&& !kmem_cache_has_cpu_partial(s
))
4488 s
->cpu_partial
= objects
;
4492 SLAB_ATTR(cpu_partial
);
4494 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4498 return sprintf(buf
, "%pS\n", s
->ctor
);
4502 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4504 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4506 SLAB_ATTR_RO(aliases
);
4508 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4510 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4512 SLAB_ATTR_RO(partial
);
4514 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4516 return show_slab_objects(s
, buf
, SO_CPU
);
4518 SLAB_ATTR_RO(cpu_slabs
);
4520 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4522 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4524 SLAB_ATTR_RO(objects
);
4526 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4528 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4530 SLAB_ATTR_RO(objects_partial
);
4532 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4539 for_each_online_cpu(cpu
) {
4540 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4543 pages
+= page
->pages
;
4544 objects
+= page
->pobjects
;
4548 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4551 for_each_online_cpu(cpu
) {
4552 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4554 if (page
&& len
< PAGE_SIZE
- 20)
4555 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4556 page
->pobjects
, page
->pages
);
4559 return len
+ sprintf(buf
+ len
, "\n");
4561 SLAB_ATTR_RO(slabs_cpu_partial
);
4563 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4565 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4568 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4569 const char *buf
, size_t length
)
4571 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4573 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4576 SLAB_ATTR(reclaim_account
);
4578 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4580 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4582 SLAB_ATTR_RO(hwcache_align
);
4584 #ifdef CONFIG_ZONE_DMA
4585 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4587 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4589 SLAB_ATTR_RO(cache_dma
);
4592 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4594 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4596 SLAB_ATTR_RO(destroy_by_rcu
);
4598 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4600 return sprintf(buf
, "%d\n", s
->reserved
);
4602 SLAB_ATTR_RO(reserved
);
4604 #ifdef CONFIG_SLUB_DEBUG
4605 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4607 return show_slab_objects(s
, buf
, SO_ALL
);
4609 SLAB_ATTR_RO(slabs
);
4611 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4613 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4615 SLAB_ATTR_RO(total_objects
);
4617 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4619 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4622 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4623 const char *buf
, size_t length
)
4625 s
->flags
&= ~SLAB_DEBUG_FREE
;
4626 if (buf
[0] == '1') {
4627 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4628 s
->flags
|= SLAB_DEBUG_FREE
;
4632 SLAB_ATTR(sanity_checks
);
4634 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4636 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4639 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4642 s
->flags
&= ~SLAB_TRACE
;
4643 if (buf
[0] == '1') {
4644 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4645 s
->flags
|= SLAB_TRACE
;
4651 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4653 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4656 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4657 const char *buf
, size_t length
)
4659 if (any_slab_objects(s
))
4662 s
->flags
&= ~SLAB_RED_ZONE
;
4663 if (buf
[0] == '1') {
4664 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4665 s
->flags
|= SLAB_RED_ZONE
;
4667 calculate_sizes(s
, -1);
4670 SLAB_ATTR(red_zone
);
4672 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4674 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4677 static ssize_t
poison_store(struct kmem_cache
*s
,
4678 const char *buf
, size_t length
)
4680 if (any_slab_objects(s
))
4683 s
->flags
&= ~SLAB_POISON
;
4684 if (buf
[0] == '1') {
4685 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4686 s
->flags
|= SLAB_POISON
;
4688 calculate_sizes(s
, -1);
4693 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4695 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4698 static ssize_t
store_user_store(struct kmem_cache
*s
,
4699 const char *buf
, size_t length
)
4701 if (any_slab_objects(s
))
4704 s
->flags
&= ~SLAB_STORE_USER
;
4705 if (buf
[0] == '1') {
4706 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4707 s
->flags
|= SLAB_STORE_USER
;
4709 calculate_sizes(s
, -1);
4712 SLAB_ATTR(store_user
);
4714 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4719 static ssize_t
validate_store(struct kmem_cache
*s
,
4720 const char *buf
, size_t length
)
4724 if (buf
[0] == '1') {
4725 ret
= validate_slab_cache(s
);
4731 SLAB_ATTR(validate
);
4733 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4735 if (!(s
->flags
& SLAB_STORE_USER
))
4737 return list_locations(s
, buf
, TRACK_ALLOC
);
4739 SLAB_ATTR_RO(alloc_calls
);
4741 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4743 if (!(s
->flags
& SLAB_STORE_USER
))
4745 return list_locations(s
, buf
, TRACK_FREE
);
4747 SLAB_ATTR_RO(free_calls
);
4748 #endif /* CONFIG_SLUB_DEBUG */
4750 #ifdef CONFIG_FAILSLAB
4751 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4753 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4756 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4759 s
->flags
&= ~SLAB_FAILSLAB
;
4761 s
->flags
|= SLAB_FAILSLAB
;
4764 SLAB_ATTR(failslab
);
4767 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4772 static ssize_t
shrink_store(struct kmem_cache
*s
,
4773 const char *buf
, size_t length
)
4775 if (buf
[0] == '1') {
4776 int rc
= kmem_cache_shrink(s
);
4787 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4789 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4792 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4793 const char *buf
, size_t length
)
4795 unsigned long ratio
;
4798 err
= kstrtoul(buf
, 10, &ratio
);
4803 s
->remote_node_defrag_ratio
= ratio
* 10;
4807 SLAB_ATTR(remote_node_defrag_ratio
);
4810 #ifdef CONFIG_SLUB_STATS
4811 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4813 unsigned long sum
= 0;
4816 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4821 for_each_online_cpu(cpu
) {
4822 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4828 len
= sprintf(buf
, "%lu", sum
);
4831 for_each_online_cpu(cpu
) {
4832 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4833 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4837 return len
+ sprintf(buf
+ len
, "\n");
4840 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4844 for_each_online_cpu(cpu
)
4845 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4848 #define STAT_ATTR(si, text) \
4849 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4851 return show_stat(s, buf, si); \
4853 static ssize_t text##_store(struct kmem_cache *s, \
4854 const char *buf, size_t length) \
4856 if (buf[0] != '0') \
4858 clear_stat(s, si); \
4863 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4864 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4865 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4866 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4867 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4868 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4869 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4870 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4871 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4872 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4873 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
4874 STAT_ATTR(FREE_SLAB
, free_slab
);
4875 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4876 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4877 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4878 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4879 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4880 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4881 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
4882 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4883 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
4884 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
4885 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
4886 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
4887 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
4888 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
4891 static struct attribute
*slab_attrs
[] = {
4892 &slab_size_attr
.attr
,
4893 &object_size_attr
.attr
,
4894 &objs_per_slab_attr
.attr
,
4896 &min_partial_attr
.attr
,
4897 &cpu_partial_attr
.attr
,
4899 &objects_partial_attr
.attr
,
4901 &cpu_slabs_attr
.attr
,
4905 &hwcache_align_attr
.attr
,
4906 &reclaim_account_attr
.attr
,
4907 &destroy_by_rcu_attr
.attr
,
4909 &reserved_attr
.attr
,
4910 &slabs_cpu_partial_attr
.attr
,
4911 #ifdef CONFIG_SLUB_DEBUG
4912 &total_objects_attr
.attr
,
4914 &sanity_checks_attr
.attr
,
4916 &red_zone_attr
.attr
,
4918 &store_user_attr
.attr
,
4919 &validate_attr
.attr
,
4920 &alloc_calls_attr
.attr
,
4921 &free_calls_attr
.attr
,
4923 #ifdef CONFIG_ZONE_DMA
4924 &cache_dma_attr
.attr
,
4927 &remote_node_defrag_ratio_attr
.attr
,
4929 #ifdef CONFIG_SLUB_STATS
4930 &alloc_fastpath_attr
.attr
,
4931 &alloc_slowpath_attr
.attr
,
4932 &free_fastpath_attr
.attr
,
4933 &free_slowpath_attr
.attr
,
4934 &free_frozen_attr
.attr
,
4935 &free_add_partial_attr
.attr
,
4936 &free_remove_partial_attr
.attr
,
4937 &alloc_from_partial_attr
.attr
,
4938 &alloc_slab_attr
.attr
,
4939 &alloc_refill_attr
.attr
,
4940 &alloc_node_mismatch_attr
.attr
,
4941 &free_slab_attr
.attr
,
4942 &cpuslab_flush_attr
.attr
,
4943 &deactivate_full_attr
.attr
,
4944 &deactivate_empty_attr
.attr
,
4945 &deactivate_to_head_attr
.attr
,
4946 &deactivate_to_tail_attr
.attr
,
4947 &deactivate_remote_frees_attr
.attr
,
4948 &deactivate_bypass_attr
.attr
,
4949 &order_fallback_attr
.attr
,
4950 &cmpxchg_double_fail_attr
.attr
,
4951 &cmpxchg_double_cpu_fail_attr
.attr
,
4952 &cpu_partial_alloc_attr
.attr
,
4953 &cpu_partial_free_attr
.attr
,
4954 &cpu_partial_node_attr
.attr
,
4955 &cpu_partial_drain_attr
.attr
,
4957 #ifdef CONFIG_FAILSLAB
4958 &failslab_attr
.attr
,
4964 static struct attribute_group slab_attr_group
= {
4965 .attrs
= slab_attrs
,
4968 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4969 struct attribute
*attr
,
4972 struct slab_attribute
*attribute
;
4973 struct kmem_cache
*s
;
4976 attribute
= to_slab_attr(attr
);
4979 if (!attribute
->show
)
4982 err
= attribute
->show(s
, buf
);
4987 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4988 struct attribute
*attr
,
4989 const char *buf
, size_t len
)
4991 struct slab_attribute
*attribute
;
4992 struct kmem_cache
*s
;
4995 attribute
= to_slab_attr(attr
);
4998 if (!attribute
->store
)
5001 err
= attribute
->store(s
, buf
, len
);
5002 #ifdef CONFIG_MEMCG_KMEM
5003 if (slab_state
>= FULL
&& err
>= 0 && is_root_cache(s
)) {
5006 mutex_lock(&slab_mutex
);
5007 if (s
->max_attr_size
< len
)
5008 s
->max_attr_size
= len
;
5011 * This is a best effort propagation, so this function's return
5012 * value will be determined by the parent cache only. This is
5013 * basically because not all attributes will have a well
5014 * defined semantics for rollbacks - most of the actions will
5015 * have permanent effects.
5017 * Returning the error value of any of the children that fail
5018 * is not 100 % defined, in the sense that users seeing the
5019 * error code won't be able to know anything about the state of
5022 * Only returning the error code for the parent cache at least
5023 * has well defined semantics. The cache being written to
5024 * directly either failed or succeeded, in which case we loop
5025 * through the descendants with best-effort propagation.
5027 for_each_memcg_cache_index(i
) {
5028 struct kmem_cache
*c
= cache_from_memcg_idx(s
, i
);
5030 attribute
->store(c
, buf
, len
);
5032 mutex_unlock(&slab_mutex
);
5038 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
)
5040 #ifdef CONFIG_MEMCG_KMEM
5042 char *buffer
= NULL
;
5044 if (!is_root_cache(s
))
5048 * This mean this cache had no attribute written. Therefore, no point
5049 * in copying default values around
5051 if (!s
->max_attr_size
)
5054 for (i
= 0; i
< ARRAY_SIZE(slab_attrs
); i
++) {
5057 struct slab_attribute
*attr
= to_slab_attr(slab_attrs
[i
]);
5059 if (!attr
|| !attr
->store
|| !attr
->show
)
5063 * It is really bad that we have to allocate here, so we will
5064 * do it only as a fallback. If we actually allocate, though,
5065 * we can just use the allocated buffer until the end.
5067 * Most of the slub attributes will tend to be very small in
5068 * size, but sysfs allows buffers up to a page, so they can
5069 * theoretically happen.
5073 else if (s
->max_attr_size
< ARRAY_SIZE(mbuf
))
5076 buffer
= (char *) get_zeroed_page(GFP_KERNEL
);
5077 if (WARN_ON(!buffer
))
5082 attr
->show(s
->memcg_params
->root_cache
, buf
);
5083 attr
->store(s
, buf
, strlen(buf
));
5087 free_page((unsigned long)buffer
);
5091 static const struct sysfs_ops slab_sysfs_ops
= {
5092 .show
= slab_attr_show
,
5093 .store
= slab_attr_store
,
5096 static struct kobj_type slab_ktype
= {
5097 .sysfs_ops
= &slab_sysfs_ops
,
5100 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5102 struct kobj_type
*ktype
= get_ktype(kobj
);
5104 if (ktype
== &slab_ktype
)
5109 static const struct kset_uevent_ops slab_uevent_ops
= {
5110 .filter
= uevent_filter
,
5113 static struct kset
*slab_kset
;
5115 #define ID_STR_LENGTH 64
5117 /* Create a unique string id for a slab cache:
5119 * Format :[flags-]size
5121 static char *create_unique_id(struct kmem_cache
*s
)
5123 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5130 * First flags affecting slabcache operations. We will only
5131 * get here for aliasable slabs so we do not need to support
5132 * too many flags. The flags here must cover all flags that
5133 * are matched during merging to guarantee that the id is
5136 if (s
->flags
& SLAB_CACHE_DMA
)
5138 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5140 if (s
->flags
& SLAB_DEBUG_FREE
)
5142 if (!(s
->flags
& SLAB_NOTRACK
))
5146 p
+= sprintf(p
, "%07d", s
->size
);
5148 #ifdef CONFIG_MEMCG_KMEM
5149 if (!is_root_cache(s
))
5150 p
+= sprintf(p
, "-%08d",
5151 memcg_cache_id(s
->memcg_params
->memcg
));
5154 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5158 static int sysfs_slab_add(struct kmem_cache
*s
)
5162 int unmergeable
= slab_unmergeable(s
);
5166 * Slabcache can never be merged so we can use the name proper.
5167 * This is typically the case for debug situations. In that
5168 * case we can catch duplicate names easily.
5170 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5174 * Create a unique name for the slab as a target
5177 name
= create_unique_id(s
);
5180 s
->kobj
.kset
= slab_kset
;
5181 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5183 kobject_put(&s
->kobj
);
5187 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5189 kobject_del(&s
->kobj
);
5190 kobject_put(&s
->kobj
);
5193 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5195 /* Setup first alias */
5196 sysfs_slab_alias(s
, s
->name
);
5202 static void sysfs_slab_remove(struct kmem_cache
*s
)
5204 if (slab_state
< FULL
)
5206 * Sysfs has not been setup yet so no need to remove the
5211 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5212 kobject_del(&s
->kobj
);
5213 kobject_put(&s
->kobj
);
5217 * Need to buffer aliases during bootup until sysfs becomes
5218 * available lest we lose that information.
5220 struct saved_alias
{
5221 struct kmem_cache
*s
;
5223 struct saved_alias
*next
;
5226 static struct saved_alias
*alias_list
;
5228 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5230 struct saved_alias
*al
;
5232 if (slab_state
== FULL
) {
5234 * If we have a leftover link then remove it.
5236 sysfs_remove_link(&slab_kset
->kobj
, name
);
5237 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5240 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5246 al
->next
= alias_list
;
5251 static int __init
slab_sysfs_init(void)
5253 struct kmem_cache
*s
;
5256 mutex_lock(&slab_mutex
);
5258 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5260 mutex_unlock(&slab_mutex
);
5261 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5267 list_for_each_entry(s
, &slab_caches
, list
) {
5268 err
= sysfs_slab_add(s
);
5270 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5271 " to sysfs\n", s
->name
);
5274 while (alias_list
) {
5275 struct saved_alias
*al
= alias_list
;
5277 alias_list
= alias_list
->next
;
5278 err
= sysfs_slab_alias(al
->s
, al
->name
);
5280 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5281 " %s to sysfs\n", al
->name
);
5285 mutex_unlock(&slab_mutex
);
5290 __initcall(slab_sysfs_init
);
5291 #endif /* CONFIG_SYSFS */
5294 * The /proc/slabinfo ABI
5296 #ifdef CONFIG_SLABINFO
5297 void get_slabinfo(struct kmem_cache
*s
, struct slabinfo
*sinfo
)
5299 unsigned long nr_slabs
= 0;
5300 unsigned long nr_objs
= 0;
5301 unsigned long nr_free
= 0;
5304 for_each_online_node(node
) {
5305 struct kmem_cache_node
*n
= get_node(s
, node
);
5310 nr_slabs
+= node_nr_slabs(n
);
5311 nr_objs
+= node_nr_objs(n
);
5312 nr_free
+= count_partial(n
, count_free
);
5315 sinfo
->active_objs
= nr_objs
- nr_free
;
5316 sinfo
->num_objs
= nr_objs
;
5317 sinfo
->active_slabs
= nr_slabs
;
5318 sinfo
->num_slabs
= nr_slabs
;
5319 sinfo
->objects_per_slab
= oo_objects(s
->oo
);
5320 sinfo
->cache_order
= oo_order(s
->oo
);
5323 void slabinfo_show_stats(struct seq_file
*m
, struct kmem_cache
*s
)
5327 ssize_t
slabinfo_write(struct file
*file
, const char __user
*buffer
,
5328 size_t count
, loff_t
*ppos
)
5332 #endif /* CONFIG_SLABINFO */