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 the.
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 /* Interrupts must be disabled (for the fallback code to work right) */
359 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
360 void *freelist_old
, unsigned long counters_old
,
361 void *freelist_new
, unsigned long counters_new
,
364 VM_BUG_ON(!irqs_disabled());
365 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
366 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
367 if (s
->flags
& __CMPXCHG_DOUBLE
) {
368 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
369 freelist_old
, counters_old
,
370 freelist_new
, counters_new
))
376 if (page
->freelist
== freelist_old
&&
377 page
->counters
== counters_old
) {
378 page
->freelist
= freelist_new
;
379 page
->counters
= counters_new
;
387 stat(s
, CMPXCHG_DOUBLE_FAIL
);
389 #ifdef SLUB_DEBUG_CMPXCHG
390 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
396 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
397 void *freelist_old
, unsigned long counters_old
,
398 void *freelist_new
, unsigned long counters_new
,
401 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
402 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
403 if (s
->flags
& __CMPXCHG_DOUBLE
) {
404 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
405 freelist_old
, counters_old
,
406 freelist_new
, counters_new
))
413 local_irq_save(flags
);
415 if (page
->freelist
== freelist_old
&&
416 page
->counters
== counters_old
) {
417 page
->freelist
= freelist_new
;
418 page
->counters
= counters_new
;
420 local_irq_restore(flags
);
424 local_irq_restore(flags
);
428 stat(s
, CMPXCHG_DOUBLE_FAIL
);
430 #ifdef SLUB_DEBUG_CMPXCHG
431 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
437 #ifdef CONFIG_SLUB_DEBUG
439 * Determine a map of object in use on a page.
441 * Node listlock must be held to guarantee that the page does
442 * not vanish from under us.
444 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
447 void *addr
= page_address(page
);
449 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
450 set_bit(slab_index(p
, s
, addr
), map
);
456 #ifdef CONFIG_SLUB_DEBUG_ON
457 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
459 static int slub_debug
;
462 static char *slub_debug_slabs
;
463 static int disable_higher_order_debug
;
468 static void print_section(char *text
, u8
*addr
, unsigned int length
)
470 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
474 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
475 enum track_item alloc
)
480 p
= object
+ s
->offset
+ sizeof(void *);
482 p
= object
+ s
->inuse
;
487 static void set_track(struct kmem_cache
*s
, void *object
,
488 enum track_item alloc
, unsigned long addr
)
490 struct track
*p
= get_track(s
, object
, alloc
);
493 #ifdef CONFIG_STACKTRACE
494 struct stack_trace trace
;
497 trace
.nr_entries
= 0;
498 trace
.max_entries
= TRACK_ADDRS_COUNT
;
499 trace
.entries
= p
->addrs
;
501 save_stack_trace(&trace
);
503 /* See rant in lockdep.c */
504 if (trace
.nr_entries
!= 0 &&
505 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
508 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
512 p
->cpu
= smp_processor_id();
513 p
->pid
= current
->pid
;
516 memset(p
, 0, sizeof(struct track
));
519 static void init_tracking(struct kmem_cache
*s
, void *object
)
521 if (!(s
->flags
& SLAB_STORE_USER
))
524 set_track(s
, object
, TRACK_FREE
, 0UL);
525 set_track(s
, object
, TRACK_ALLOC
, 0UL);
528 static void print_track(const char *s
, struct track
*t
)
533 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
534 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
535 #ifdef CONFIG_STACKTRACE
538 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
540 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
547 static void print_tracking(struct kmem_cache
*s
, void *object
)
549 if (!(s
->flags
& SLAB_STORE_USER
))
552 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
553 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
556 static void print_page_info(struct page
*page
)
559 "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
560 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
564 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
570 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
572 printk(KERN_ERR
"========================================"
573 "=====================================\n");
574 printk(KERN_ERR
"BUG %s (%s): %s\n", s
->name
, print_tainted(), buf
);
575 printk(KERN_ERR
"----------------------------------------"
576 "-------------------------------------\n\n");
578 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
581 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
587 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
589 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
592 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
594 unsigned int off
; /* Offset of last byte */
595 u8
*addr
= page_address(page
);
597 print_tracking(s
, p
);
599 print_page_info(page
);
601 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
602 p
, p
- addr
, get_freepointer(s
, p
));
605 print_section("Bytes b4 ", p
- 16, 16);
607 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
609 if (s
->flags
& SLAB_RED_ZONE
)
610 print_section("Redzone ", p
+ s
->object_size
,
611 s
->inuse
- s
->object_size
);
614 off
= s
->offset
+ sizeof(void *);
618 if (s
->flags
& SLAB_STORE_USER
)
619 off
+= 2 * sizeof(struct track
);
622 /* Beginning of the filler is the free pointer */
623 print_section("Padding ", p
+ off
, s
->size
- off
);
628 static void object_err(struct kmem_cache
*s
, struct page
*page
,
629 u8
*object
, char *reason
)
631 slab_bug(s
, "%s", reason
);
632 print_trailer(s
, page
, object
);
635 static void slab_err(struct kmem_cache
*s
, struct page
*page
,
636 const char *fmt
, ...)
642 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
644 slab_bug(s
, "%s", buf
);
645 print_page_info(page
);
649 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
653 if (s
->flags
& __OBJECT_POISON
) {
654 memset(p
, POISON_FREE
, s
->object_size
- 1);
655 p
[s
->object_size
- 1] = POISON_END
;
658 if (s
->flags
& SLAB_RED_ZONE
)
659 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
662 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
663 void *from
, void *to
)
665 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
666 memset(from
, data
, to
- from
);
669 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
670 u8
*object
, char *what
,
671 u8
*start
, unsigned int value
, unsigned int bytes
)
676 fault
= memchr_inv(start
, value
, bytes
);
681 while (end
> fault
&& end
[-1] == value
)
684 slab_bug(s
, "%s overwritten", what
);
685 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
686 fault
, end
- 1, fault
[0], value
);
687 print_trailer(s
, page
, object
);
689 restore_bytes(s
, what
, value
, fault
, end
);
697 * Bytes of the object to be managed.
698 * If the freepointer may overlay the object then the free
699 * pointer is the first word of the object.
701 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
704 * object + s->object_size
705 * Padding to reach word boundary. This is also used for Redzoning.
706 * Padding is extended by another word if Redzoning is enabled and
707 * object_size == inuse.
709 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
710 * 0xcc (RED_ACTIVE) for objects in use.
713 * Meta data starts here.
715 * A. Free pointer (if we cannot overwrite object on free)
716 * B. Tracking data for SLAB_STORE_USER
717 * C. Padding to reach required alignment boundary or at mininum
718 * one word if debugging is on to be able to detect writes
719 * before the word boundary.
721 * Padding is done using 0x5a (POISON_INUSE)
724 * Nothing is used beyond s->size.
726 * If slabcaches are merged then the object_size and inuse boundaries are mostly
727 * ignored. And therefore no slab options that rely on these boundaries
728 * may be used with merged slabcaches.
731 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
733 unsigned long off
= s
->inuse
; /* The end of info */
736 /* Freepointer is placed after the object. */
737 off
+= sizeof(void *);
739 if (s
->flags
& SLAB_STORE_USER
)
740 /* We also have user information there */
741 off
+= 2 * sizeof(struct track
);
746 return check_bytes_and_report(s
, page
, p
, "Object padding",
747 p
+ off
, POISON_INUSE
, s
->size
- off
);
750 /* Check the pad bytes at the end of a slab page */
751 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
759 if (!(s
->flags
& SLAB_POISON
))
762 start
= page_address(page
);
763 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
764 end
= start
+ length
;
765 remainder
= length
% s
->size
;
769 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
772 while (end
> fault
&& end
[-1] == POISON_INUSE
)
775 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
776 print_section("Padding ", end
- remainder
, remainder
);
778 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
782 static int check_object(struct kmem_cache
*s
, struct page
*page
,
783 void *object
, u8 val
)
786 u8
*endobject
= object
+ s
->object_size
;
788 if (s
->flags
& SLAB_RED_ZONE
) {
789 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
790 endobject
, val
, s
->inuse
- s
->object_size
))
793 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
794 check_bytes_and_report(s
, page
, p
, "Alignment padding",
795 endobject
, POISON_INUSE
,
796 s
->inuse
- s
->object_size
);
800 if (s
->flags
& SLAB_POISON
) {
801 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
802 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
803 POISON_FREE
, s
->object_size
- 1) ||
804 !check_bytes_and_report(s
, page
, p
, "Poison",
805 p
+ s
->object_size
- 1, POISON_END
, 1)))
808 * check_pad_bytes cleans up on its own.
810 check_pad_bytes(s
, page
, p
);
813 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
815 * Object and freepointer overlap. Cannot check
816 * freepointer while object is allocated.
820 /* Check free pointer validity */
821 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
822 object_err(s
, page
, p
, "Freepointer corrupt");
824 * No choice but to zap it and thus lose the remainder
825 * of the free objects in this slab. May cause
826 * another error because the object count is now wrong.
828 set_freepointer(s
, p
, NULL
);
834 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
838 VM_BUG_ON(!irqs_disabled());
840 if (!PageSlab(page
)) {
841 slab_err(s
, page
, "Not a valid slab page");
845 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
846 if (page
->objects
> maxobj
) {
847 slab_err(s
, page
, "objects %u > max %u",
848 s
->name
, page
->objects
, maxobj
);
851 if (page
->inuse
> page
->objects
) {
852 slab_err(s
, page
, "inuse %u > max %u",
853 s
->name
, page
->inuse
, page
->objects
);
856 /* Slab_pad_check fixes things up after itself */
857 slab_pad_check(s
, page
);
862 * Determine if a certain object on a page is on the freelist. Must hold the
863 * slab lock to guarantee that the chains are in a consistent state.
865 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
870 unsigned long max_objects
;
873 while (fp
&& nr
<= page
->objects
) {
876 if (!check_valid_pointer(s
, page
, fp
)) {
878 object_err(s
, page
, object
,
879 "Freechain corrupt");
880 set_freepointer(s
, object
, NULL
);
882 slab_err(s
, page
, "Freepointer corrupt");
883 page
->freelist
= NULL
;
884 page
->inuse
= page
->objects
;
885 slab_fix(s
, "Freelist cleared");
891 fp
= get_freepointer(s
, object
);
895 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
896 if (max_objects
> MAX_OBJS_PER_PAGE
)
897 max_objects
= MAX_OBJS_PER_PAGE
;
899 if (page
->objects
!= max_objects
) {
900 slab_err(s
, page
, "Wrong number of objects. Found %d but "
901 "should be %d", page
->objects
, max_objects
);
902 page
->objects
= max_objects
;
903 slab_fix(s
, "Number of objects adjusted.");
905 if (page
->inuse
!= page
->objects
- nr
) {
906 slab_err(s
, page
, "Wrong object count. Counter is %d but "
907 "counted were %d", page
->inuse
, page
->objects
- nr
);
908 page
->inuse
= page
->objects
- nr
;
909 slab_fix(s
, "Object count adjusted.");
911 return search
== NULL
;
914 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
917 if (s
->flags
& SLAB_TRACE
) {
918 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
920 alloc
? "alloc" : "free",
925 print_section("Object ", (void *)object
,
933 * Hooks for other subsystems that check memory allocations. In a typical
934 * production configuration these hooks all should produce no code at all.
936 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
938 flags
&= gfp_allowed_mask
;
939 lockdep_trace_alloc(flags
);
940 might_sleep_if(flags
& __GFP_WAIT
);
942 return should_failslab(s
->object_size
, flags
, s
->flags
);
945 static inline void slab_post_alloc_hook(struct kmem_cache
*s
,
946 gfp_t flags
, void *object
)
948 flags
&= gfp_allowed_mask
;
949 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
950 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
953 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
955 kmemleak_free_recursive(x
, s
->flags
);
958 * Trouble is that we may no longer disable interupts in the fast path
959 * So in order to make the debug calls that expect irqs to be
960 * disabled we need to disable interrupts temporarily.
962 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
966 local_irq_save(flags
);
967 kmemcheck_slab_free(s
, x
, s
->object_size
);
968 debug_check_no_locks_freed(x
, s
->object_size
);
969 local_irq_restore(flags
);
972 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
973 debug_check_no_obj_freed(x
, s
->object_size
);
977 * Tracking of fully allocated slabs for debugging purposes.
979 * list_lock must be held.
981 static void add_full(struct kmem_cache
*s
,
982 struct kmem_cache_node
*n
, struct page
*page
)
984 if (!(s
->flags
& SLAB_STORE_USER
))
987 list_add(&page
->lru
, &n
->full
);
991 * list_lock must be held.
993 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
995 if (!(s
->flags
& SLAB_STORE_USER
))
998 list_del(&page
->lru
);
1001 /* Tracking of the number of slabs for debugging purposes */
1002 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1004 struct kmem_cache_node
*n
= get_node(s
, node
);
1006 return atomic_long_read(&n
->nr_slabs
);
1009 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1011 return atomic_long_read(&n
->nr_slabs
);
1014 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1016 struct kmem_cache_node
*n
= get_node(s
, node
);
1019 * May be called early in order to allocate a slab for the
1020 * kmem_cache_node structure. Solve the chicken-egg
1021 * dilemma by deferring the increment of the count during
1022 * bootstrap (see early_kmem_cache_node_alloc).
1025 atomic_long_inc(&n
->nr_slabs
);
1026 atomic_long_add(objects
, &n
->total_objects
);
1029 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1031 struct kmem_cache_node
*n
= get_node(s
, node
);
1033 atomic_long_dec(&n
->nr_slabs
);
1034 atomic_long_sub(objects
, &n
->total_objects
);
1037 /* Object debug checks for alloc/free paths */
1038 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1041 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1044 init_object(s
, object
, SLUB_RED_INACTIVE
);
1045 init_tracking(s
, object
);
1048 static noinline
int alloc_debug_processing(struct kmem_cache
*s
,
1050 void *object
, unsigned long addr
)
1052 if (!check_slab(s
, page
))
1055 if (!check_valid_pointer(s
, page
, object
)) {
1056 object_err(s
, page
, object
, "Freelist Pointer check fails");
1060 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1063 /* Success perform special debug activities for allocs */
1064 if (s
->flags
& SLAB_STORE_USER
)
1065 set_track(s
, object
, TRACK_ALLOC
, addr
);
1066 trace(s
, page
, object
, 1);
1067 init_object(s
, object
, SLUB_RED_ACTIVE
);
1071 if (PageSlab(page
)) {
1073 * If this is a slab page then lets do the best we can
1074 * to avoid issues in the future. Marking all objects
1075 * as used avoids touching the remaining objects.
1077 slab_fix(s
, "Marking all objects used");
1078 page
->inuse
= page
->objects
;
1079 page
->freelist
= NULL
;
1084 static noinline
struct kmem_cache_node
*free_debug_processing(
1085 struct kmem_cache
*s
, struct page
*page
, void *object
,
1086 unsigned long addr
, unsigned long *flags
)
1088 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1090 spin_lock_irqsave(&n
->list_lock
, *flags
);
1093 if (!check_slab(s
, page
))
1096 if (!check_valid_pointer(s
, page
, object
)) {
1097 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1101 if (on_freelist(s
, page
, object
)) {
1102 object_err(s
, page
, object
, "Object already free");
1106 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1109 if (unlikely(s
!= page
->slab_cache
)) {
1110 if (!PageSlab(page
)) {
1111 slab_err(s
, page
, "Attempt to free object(0x%p) "
1112 "outside of slab", object
);
1113 } else if (!page
->slab_cache
) {
1115 "SLUB <none>: no slab for object 0x%p.\n",
1119 object_err(s
, page
, object
,
1120 "page slab pointer corrupt.");
1124 if (s
->flags
& SLAB_STORE_USER
)
1125 set_track(s
, object
, TRACK_FREE
, addr
);
1126 trace(s
, page
, object
, 0);
1127 init_object(s
, object
, SLUB_RED_INACTIVE
);
1131 * Keep node_lock to preserve integrity
1132 * until the object is actually freed
1138 spin_unlock_irqrestore(&n
->list_lock
, *flags
);
1139 slab_fix(s
, "Object at 0x%p not freed", object
);
1143 static int __init
setup_slub_debug(char *str
)
1145 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1146 if (*str
++ != '=' || !*str
)
1148 * No options specified. Switch on full debugging.
1154 * No options but restriction on slabs. This means full
1155 * debugging for slabs matching a pattern.
1159 if (tolower(*str
) == 'o') {
1161 * Avoid enabling debugging on caches if its minimum order
1162 * would increase as a result.
1164 disable_higher_order_debug
= 1;
1171 * Switch off all debugging measures.
1176 * Determine which debug features should be switched on
1178 for (; *str
&& *str
!= ','; str
++) {
1179 switch (tolower(*str
)) {
1181 slub_debug
|= SLAB_DEBUG_FREE
;
1184 slub_debug
|= SLAB_RED_ZONE
;
1187 slub_debug
|= SLAB_POISON
;
1190 slub_debug
|= SLAB_STORE_USER
;
1193 slub_debug
|= SLAB_TRACE
;
1196 slub_debug
|= SLAB_FAILSLAB
;
1199 printk(KERN_ERR
"slub_debug option '%c' "
1200 "unknown. skipped\n", *str
);
1206 slub_debug_slabs
= str
+ 1;
1211 __setup("slub_debug", setup_slub_debug
);
1213 static unsigned long kmem_cache_flags(unsigned long object_size
,
1214 unsigned long flags
, const char *name
,
1215 void (*ctor
)(void *))
1218 * Enable debugging if selected on the kernel commandline.
1220 if (slub_debug
&& (!slub_debug_slabs
||
1221 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1222 flags
|= slub_debug
;
1227 static inline void setup_object_debug(struct kmem_cache
*s
,
1228 struct page
*page
, void *object
) {}
1230 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1231 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1233 static inline struct kmem_cache_node
*free_debug_processing(
1234 struct kmem_cache
*s
, struct page
*page
, void *object
,
1235 unsigned long addr
, unsigned long *flags
) { return NULL
; }
1237 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1239 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1240 void *object
, u8 val
) { return 1; }
1241 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1242 struct page
*page
) {}
1243 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1244 static inline unsigned long kmem_cache_flags(unsigned long object_size
,
1245 unsigned long flags
, const char *name
,
1246 void (*ctor
)(void *))
1250 #define slub_debug 0
1252 #define disable_higher_order_debug 0
1254 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1256 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1258 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1260 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1263 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1266 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1269 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1271 #endif /* CONFIG_SLUB_DEBUG */
1274 * Slab allocation and freeing
1276 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1277 struct kmem_cache_order_objects oo
)
1279 int order
= oo_order(oo
);
1281 flags
|= __GFP_NOTRACK
;
1283 if (node
== NUMA_NO_NODE
)
1284 return alloc_pages(flags
, order
);
1286 return alloc_pages_exact_node(node
, flags
, order
);
1289 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1292 struct kmem_cache_order_objects oo
= s
->oo
;
1295 flags
&= gfp_allowed_mask
;
1297 if (flags
& __GFP_WAIT
)
1300 flags
|= s
->allocflags
;
1303 * Let the initial higher-order allocation fail under memory pressure
1304 * so we fall-back to the minimum order allocation.
1306 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1308 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1309 if (unlikely(!page
)) {
1312 * Allocation may have failed due to fragmentation.
1313 * Try a lower order alloc if possible
1315 page
= alloc_slab_page(flags
, node
, oo
);
1318 stat(s
, ORDER_FALLBACK
);
1321 if (kmemcheck_enabled
&& page
1322 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1323 int pages
= 1 << oo_order(oo
);
1325 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1328 * Objects from caches that have a constructor don't get
1329 * cleared when they're allocated, so we need to do it here.
1332 kmemcheck_mark_uninitialized_pages(page
, pages
);
1334 kmemcheck_mark_unallocated_pages(page
, pages
);
1337 if (flags
& __GFP_WAIT
)
1338 local_irq_disable();
1342 page
->objects
= oo_objects(oo
);
1343 mod_zone_page_state(page_zone(page
),
1344 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1345 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1351 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1354 setup_object_debug(s
, page
, object
);
1355 if (unlikely(s
->ctor
))
1359 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1367 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1369 page
= allocate_slab(s
,
1370 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1374 order
= compound_order(page
);
1375 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1376 memcg_bind_pages(s
, order
);
1377 page
->slab_cache
= s
;
1378 __SetPageSlab(page
);
1379 if (page
->pfmemalloc
)
1380 SetPageSlabPfmemalloc(page
);
1382 start
= page_address(page
);
1384 if (unlikely(s
->flags
& SLAB_POISON
))
1385 memset(start
, POISON_INUSE
, PAGE_SIZE
<< order
);
1388 for_each_object(p
, s
, start
, page
->objects
) {
1389 setup_object(s
, page
, last
);
1390 set_freepointer(s
, last
, p
);
1393 setup_object(s
, page
, last
);
1394 set_freepointer(s
, last
, NULL
);
1396 page
->freelist
= start
;
1397 page
->inuse
= page
->objects
;
1403 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1405 int order
= compound_order(page
);
1406 int pages
= 1 << order
;
1408 if (kmem_cache_debug(s
)) {
1411 slab_pad_check(s
, page
);
1412 for_each_object(p
, s
, page_address(page
),
1414 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1417 kmemcheck_free_shadow(page
, compound_order(page
));
1419 mod_zone_page_state(page_zone(page
),
1420 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1421 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1424 __ClearPageSlabPfmemalloc(page
);
1425 __ClearPageSlab(page
);
1427 memcg_release_pages(s
, order
);
1428 page_mapcount_reset(page
);
1429 if (current
->reclaim_state
)
1430 current
->reclaim_state
->reclaimed_slab
+= pages
;
1431 __free_memcg_kmem_pages(page
, order
);
1434 #define need_reserve_slab_rcu \
1435 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1437 static void rcu_free_slab(struct rcu_head
*h
)
1441 if (need_reserve_slab_rcu
)
1442 page
= virt_to_head_page(h
);
1444 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1446 __free_slab(page
->slab_cache
, page
);
1449 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1451 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1452 struct rcu_head
*head
;
1454 if (need_reserve_slab_rcu
) {
1455 int order
= compound_order(page
);
1456 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1458 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1459 head
= page_address(page
) + offset
;
1462 * RCU free overloads the RCU head over the LRU
1464 head
= (void *)&page
->lru
;
1467 call_rcu(head
, rcu_free_slab
);
1469 __free_slab(s
, page
);
1472 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1474 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1479 * Management of partially allocated slabs.
1481 * list_lock must be held.
1483 static inline void add_partial(struct kmem_cache_node
*n
,
1484 struct page
*page
, int tail
)
1487 if (tail
== DEACTIVATE_TO_TAIL
)
1488 list_add_tail(&page
->lru
, &n
->partial
);
1490 list_add(&page
->lru
, &n
->partial
);
1494 * list_lock must be held.
1496 static inline void remove_partial(struct kmem_cache_node
*n
,
1499 list_del(&page
->lru
);
1504 * Remove slab from the partial list, freeze it and
1505 * return the pointer to the freelist.
1507 * Returns a list of objects or NULL if it fails.
1509 * Must hold list_lock since we modify the partial list.
1511 static inline void *acquire_slab(struct kmem_cache
*s
,
1512 struct kmem_cache_node
*n
, struct page
*page
,
1513 int mode
, int *objects
)
1516 unsigned long counters
;
1520 * Zap the freelist and set the frozen bit.
1521 * The old freelist is the list of objects for the
1522 * per cpu allocation list.
1524 freelist
= page
->freelist
;
1525 counters
= page
->counters
;
1526 new.counters
= counters
;
1527 *objects
= new.objects
- new.inuse
;
1529 new.inuse
= page
->objects
;
1530 new.freelist
= NULL
;
1532 new.freelist
= freelist
;
1535 VM_BUG_ON(new.frozen
);
1538 if (!__cmpxchg_double_slab(s
, page
,
1540 new.freelist
, new.counters
,
1544 remove_partial(n
, page
);
1549 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1550 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
);
1553 * Try to allocate a partial slab from a specific node.
1555 static void *get_partial_node(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1556 struct kmem_cache_cpu
*c
, gfp_t flags
)
1558 struct page
*page
, *page2
;
1559 void *object
= NULL
;
1564 * Racy check. If we mistakenly see no partial slabs then we
1565 * just allocate an empty slab. If we mistakenly try to get a
1566 * partial slab and there is none available then get_partials()
1569 if (!n
|| !n
->nr_partial
)
1572 spin_lock(&n
->list_lock
);
1573 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1576 if (!pfmemalloc_match(page
, flags
))
1579 t
= acquire_slab(s
, n
, page
, object
== NULL
, &objects
);
1583 available
+= objects
;
1586 stat(s
, ALLOC_FROM_PARTIAL
);
1589 put_cpu_partial(s
, page
, 0);
1590 stat(s
, CPU_PARTIAL_NODE
);
1592 if (!kmem_cache_has_cpu_partial(s
)
1593 || available
> s
->cpu_partial
/ 2)
1597 spin_unlock(&n
->list_lock
);
1602 * Get a page from somewhere. Search in increasing NUMA distances.
1604 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1605 struct kmem_cache_cpu
*c
)
1608 struct zonelist
*zonelist
;
1611 enum zone_type high_zoneidx
= gfp_zone(flags
);
1613 unsigned int cpuset_mems_cookie
;
1616 * The defrag ratio allows a configuration of the tradeoffs between
1617 * inter node defragmentation and node local allocations. A lower
1618 * defrag_ratio increases the tendency to do local allocations
1619 * instead of attempting to obtain partial slabs from other nodes.
1621 * If the defrag_ratio is set to 0 then kmalloc() always
1622 * returns node local objects. If the ratio is higher then kmalloc()
1623 * may return off node objects because partial slabs are obtained
1624 * from other nodes and filled up.
1626 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1627 * defrag_ratio = 1000) then every (well almost) allocation will
1628 * first attempt to defrag slab caches on other nodes. This means
1629 * scanning over all nodes to look for partial slabs which may be
1630 * expensive if we do it every time we are trying to find a slab
1631 * with available objects.
1633 if (!s
->remote_node_defrag_ratio
||
1634 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1638 cpuset_mems_cookie
= get_mems_allowed();
1639 zonelist
= node_zonelist(slab_node(), flags
);
1640 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1641 struct kmem_cache_node
*n
;
1643 n
= get_node(s
, zone_to_nid(zone
));
1645 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1646 n
->nr_partial
> s
->min_partial
) {
1647 object
= get_partial_node(s
, n
, c
, flags
);
1650 * Return the object even if
1651 * put_mems_allowed indicated that
1652 * the cpuset mems_allowed was
1653 * updated in parallel. It's a
1654 * harmless race between the alloc
1655 * and the cpuset update.
1657 put_mems_allowed(cpuset_mems_cookie
);
1662 } while (!put_mems_allowed(cpuset_mems_cookie
));
1668 * Get a partial page, lock it and return it.
1670 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1671 struct kmem_cache_cpu
*c
)
1674 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1676 object
= get_partial_node(s
, get_node(s
, searchnode
), c
, flags
);
1677 if (object
|| node
!= NUMA_NO_NODE
)
1680 return get_any_partial(s
, flags
, c
);
1683 #ifdef CONFIG_PREEMPT
1685 * Calculate the next globally unique transaction for disambiguiation
1686 * during cmpxchg. The transactions start with the cpu number and are then
1687 * incremented by CONFIG_NR_CPUS.
1689 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1692 * No preemption supported therefore also no need to check for
1698 static inline unsigned long next_tid(unsigned long tid
)
1700 return tid
+ TID_STEP
;
1703 static inline unsigned int tid_to_cpu(unsigned long tid
)
1705 return tid
% TID_STEP
;
1708 static inline unsigned long tid_to_event(unsigned long tid
)
1710 return tid
/ TID_STEP
;
1713 static inline unsigned int init_tid(int cpu
)
1718 static inline void note_cmpxchg_failure(const char *n
,
1719 const struct kmem_cache
*s
, unsigned long tid
)
1721 #ifdef SLUB_DEBUG_CMPXCHG
1722 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1724 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1726 #ifdef CONFIG_PREEMPT
1727 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1728 printk("due to cpu change %d -> %d\n",
1729 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1732 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1733 printk("due to cpu running other code. Event %ld->%ld\n",
1734 tid_to_event(tid
), tid_to_event(actual_tid
));
1736 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1737 actual_tid
, tid
, next_tid(tid
));
1739 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1742 static void init_kmem_cache_cpus(struct kmem_cache
*s
)
1746 for_each_possible_cpu(cpu
)
1747 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1751 * Remove the cpu slab
1753 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
,
1756 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1757 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1759 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1761 int tail
= DEACTIVATE_TO_HEAD
;
1765 if (page
->freelist
) {
1766 stat(s
, DEACTIVATE_REMOTE_FREES
);
1767 tail
= DEACTIVATE_TO_TAIL
;
1771 * Stage one: Free all available per cpu objects back
1772 * to the page freelist while it is still frozen. Leave the
1775 * There is no need to take the list->lock because the page
1778 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1780 unsigned long counters
;
1783 prior
= page
->freelist
;
1784 counters
= page
->counters
;
1785 set_freepointer(s
, freelist
, prior
);
1786 new.counters
= counters
;
1788 VM_BUG_ON(!new.frozen
);
1790 } while (!__cmpxchg_double_slab(s
, page
,
1792 freelist
, new.counters
,
1793 "drain percpu freelist"));
1795 freelist
= nextfree
;
1799 * Stage two: Ensure that the page is unfrozen while the
1800 * list presence reflects the actual number of objects
1803 * We setup the list membership and then perform a cmpxchg
1804 * with the count. If there is a mismatch then the page
1805 * is not unfrozen but the page is on the wrong list.
1807 * Then we restart the process which may have to remove
1808 * the page from the list that we just put it on again
1809 * because the number of objects in the slab may have
1814 old
.freelist
= page
->freelist
;
1815 old
.counters
= page
->counters
;
1816 VM_BUG_ON(!old
.frozen
);
1818 /* Determine target state of the slab */
1819 new.counters
= old
.counters
;
1822 set_freepointer(s
, freelist
, old
.freelist
);
1823 new.freelist
= freelist
;
1825 new.freelist
= old
.freelist
;
1829 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1831 else if (new.freelist
) {
1836 * Taking the spinlock removes the possiblity
1837 * that acquire_slab() will see a slab page that
1840 spin_lock(&n
->list_lock
);
1844 if (kmem_cache_debug(s
) && !lock
) {
1847 * This also ensures that the scanning of full
1848 * slabs from diagnostic functions will not see
1851 spin_lock(&n
->list_lock
);
1859 remove_partial(n
, page
);
1861 else if (l
== M_FULL
)
1863 remove_full(s
, page
);
1865 if (m
== M_PARTIAL
) {
1867 add_partial(n
, page
, tail
);
1870 } else if (m
== M_FULL
) {
1872 stat(s
, DEACTIVATE_FULL
);
1873 add_full(s
, n
, page
);
1879 if (!__cmpxchg_double_slab(s
, page
,
1880 old
.freelist
, old
.counters
,
1881 new.freelist
, new.counters
,
1886 spin_unlock(&n
->list_lock
);
1889 stat(s
, DEACTIVATE_EMPTY
);
1890 discard_slab(s
, page
);
1896 * Unfreeze all the cpu partial slabs.
1898 * This function must be called with interrupts disabled
1899 * for the cpu using c (or some other guarantee must be there
1900 * to guarantee no concurrent accesses).
1902 static void unfreeze_partials(struct kmem_cache
*s
,
1903 struct kmem_cache_cpu
*c
)
1905 #ifdef CONFIG_SLUB_CPU_PARTIAL
1906 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1907 struct page
*page
, *discard_page
= NULL
;
1909 while ((page
= c
->partial
)) {
1913 c
->partial
= page
->next
;
1915 n2
= get_node(s
, page_to_nid(page
));
1918 spin_unlock(&n
->list_lock
);
1921 spin_lock(&n
->list_lock
);
1926 old
.freelist
= page
->freelist
;
1927 old
.counters
= page
->counters
;
1928 VM_BUG_ON(!old
.frozen
);
1930 new.counters
= old
.counters
;
1931 new.freelist
= old
.freelist
;
1935 } while (!__cmpxchg_double_slab(s
, page
,
1936 old
.freelist
, old
.counters
,
1937 new.freelist
, new.counters
,
1938 "unfreezing slab"));
1940 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
)) {
1941 page
->next
= discard_page
;
1942 discard_page
= page
;
1944 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1945 stat(s
, FREE_ADD_PARTIAL
);
1950 spin_unlock(&n
->list_lock
);
1952 while (discard_page
) {
1953 page
= discard_page
;
1954 discard_page
= discard_page
->next
;
1956 stat(s
, DEACTIVATE_EMPTY
);
1957 discard_slab(s
, page
);
1964 * Put a page that was just frozen (in __slab_free) into a partial page
1965 * slot if available. This is done without interrupts disabled and without
1966 * preemption disabled. The cmpxchg is racy and may put the partial page
1967 * onto a random cpus partial slot.
1969 * If we did not find a slot then simply move all the partials to the
1970 * per node partial list.
1972 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1974 #ifdef CONFIG_SLUB_CPU_PARTIAL
1975 struct page
*oldpage
;
1979 if (!s
->cpu_partial
)
1985 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1988 pobjects
= oldpage
->pobjects
;
1989 pages
= oldpage
->pages
;
1990 if (drain
&& pobjects
> s
->cpu_partial
) {
1991 unsigned long flags
;
1993 * partial array is full. Move the existing
1994 * set to the per node partial list.
1996 local_irq_save(flags
);
1997 unfreeze_partials(s
, this_cpu_ptr(s
->cpu_slab
));
1998 local_irq_restore(flags
);
2002 stat(s
, CPU_PARTIAL_DRAIN
);
2007 pobjects
+= page
->objects
- page
->inuse
;
2009 page
->pages
= pages
;
2010 page
->pobjects
= pobjects
;
2011 page
->next
= oldpage
;
2013 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
)
2018 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2020 stat(s
, CPUSLAB_FLUSH
);
2021 deactivate_slab(s
, c
->page
, c
->freelist
);
2023 c
->tid
= next_tid(c
->tid
);
2031 * Called from IPI handler with interrupts disabled.
2033 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2035 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2041 unfreeze_partials(s
, c
);
2045 static void flush_cpu_slab(void *d
)
2047 struct kmem_cache
*s
= d
;
2049 __flush_cpu_slab(s
, smp_processor_id());
2052 static bool has_cpu_slab(int cpu
, void *info
)
2054 struct kmem_cache
*s
= info
;
2055 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2057 return c
->page
|| c
->partial
;
2060 static void flush_all(struct kmem_cache
*s
)
2062 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2066 * Check if the objects in a per cpu structure fit numa
2067 * locality expectations.
2069 static inline int node_match(struct page
*page
, int node
)
2072 if (!page
|| (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
))
2078 static int count_free(struct page
*page
)
2080 return page
->objects
- page
->inuse
;
2083 static unsigned long count_partial(struct kmem_cache_node
*n
,
2084 int (*get_count
)(struct page
*))
2086 unsigned long flags
;
2087 unsigned long x
= 0;
2090 spin_lock_irqsave(&n
->list_lock
, flags
);
2091 list_for_each_entry(page
, &n
->partial
, lru
)
2092 x
+= get_count(page
);
2093 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2097 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2099 #ifdef CONFIG_SLUB_DEBUG
2100 return atomic_long_read(&n
->total_objects
);
2106 static noinline
void
2107 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2112 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2114 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2115 "default order: %d, min order: %d\n", s
->name
, s
->object_size
,
2116 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2118 if (oo_order(s
->min
) > get_order(s
->object_size
))
2119 printk(KERN_WARNING
" %s debugging increased min order, use "
2120 "slub_debug=O to disable.\n", s
->name
);
2122 for_each_online_node(node
) {
2123 struct kmem_cache_node
*n
= get_node(s
, node
);
2124 unsigned long nr_slabs
;
2125 unsigned long nr_objs
;
2126 unsigned long nr_free
;
2131 nr_free
= count_partial(n
, count_free
);
2132 nr_slabs
= node_nr_slabs(n
);
2133 nr_objs
= node_nr_objs(n
);
2136 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2137 node
, nr_slabs
, nr_objs
, nr_free
);
2141 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2142 int node
, struct kmem_cache_cpu
**pc
)
2145 struct kmem_cache_cpu
*c
= *pc
;
2148 freelist
= get_partial(s
, flags
, node
, c
);
2153 page
= new_slab(s
, flags
, node
);
2155 c
= __this_cpu_ptr(s
->cpu_slab
);
2160 * No other reference to the page yet so we can
2161 * muck around with it freely without cmpxchg
2163 freelist
= page
->freelist
;
2164 page
->freelist
= NULL
;
2166 stat(s
, ALLOC_SLAB
);
2175 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
)
2177 if (unlikely(PageSlabPfmemalloc(page
)))
2178 return gfp_pfmemalloc_allowed(gfpflags
);
2184 * Check the page->freelist of a page and either transfer the freelist to the
2185 * per cpu freelist or deactivate the page.
2187 * The page is still frozen if the return value is not NULL.
2189 * If this function returns NULL then the page has been unfrozen.
2191 * This function must be called with interrupt disabled.
2193 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2196 unsigned long counters
;
2200 freelist
= page
->freelist
;
2201 counters
= page
->counters
;
2203 new.counters
= counters
;
2204 VM_BUG_ON(!new.frozen
);
2206 new.inuse
= page
->objects
;
2207 new.frozen
= freelist
!= NULL
;
2209 } while (!__cmpxchg_double_slab(s
, page
,
2218 * Slow path. The lockless freelist is empty or we need to perform
2221 * Processing is still very fast if new objects have been freed to the
2222 * regular freelist. In that case we simply take over the regular freelist
2223 * as the lockless freelist and zap the regular freelist.
2225 * If that is not working then we fall back to the partial lists. We take the
2226 * first element of the freelist as the object to allocate now and move the
2227 * rest of the freelist to the lockless freelist.
2229 * And if we were unable to get a new slab from the partial slab lists then
2230 * we need to allocate a new slab. This is the slowest path since it involves
2231 * a call to the page allocator and the setup of a new slab.
2233 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2234 unsigned long addr
, struct kmem_cache_cpu
*c
)
2238 unsigned long flags
;
2240 local_irq_save(flags
);
2241 #ifdef CONFIG_PREEMPT
2243 * We may have been preempted and rescheduled on a different
2244 * cpu before disabling interrupts. Need to reload cpu area
2247 c
= this_cpu_ptr(s
->cpu_slab
);
2255 if (unlikely(!node_match(page
, node
))) {
2256 stat(s
, ALLOC_NODE_MISMATCH
);
2257 deactivate_slab(s
, page
, c
->freelist
);
2264 * By rights, we should be searching for a slab page that was
2265 * PFMEMALLOC but right now, we are losing the pfmemalloc
2266 * information when the page leaves the per-cpu allocator
2268 if (unlikely(!pfmemalloc_match(page
, gfpflags
))) {
2269 deactivate_slab(s
, page
, c
->freelist
);
2275 /* must check again c->freelist in case of cpu migration or IRQ */
2276 freelist
= c
->freelist
;
2280 stat(s
, ALLOC_SLOWPATH
);
2282 freelist
= get_freelist(s
, page
);
2286 stat(s
, DEACTIVATE_BYPASS
);
2290 stat(s
, ALLOC_REFILL
);
2294 * freelist is pointing to the list of objects to be used.
2295 * page is pointing to the page from which the objects are obtained.
2296 * That page must be frozen for per cpu allocations to work.
2298 VM_BUG_ON(!c
->page
->frozen
);
2299 c
->freelist
= get_freepointer(s
, freelist
);
2300 c
->tid
= next_tid(c
->tid
);
2301 local_irq_restore(flags
);
2307 page
= c
->page
= c
->partial
;
2308 c
->partial
= page
->next
;
2309 stat(s
, CPU_PARTIAL_ALLOC
);
2314 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2316 if (unlikely(!freelist
)) {
2317 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2318 slab_out_of_memory(s
, gfpflags
, node
);
2320 local_irq_restore(flags
);
2325 if (likely(!kmem_cache_debug(s
) && pfmemalloc_match(page
, gfpflags
)))
2328 /* Only entered in the debug case */
2329 if (kmem_cache_debug(s
) &&
2330 !alloc_debug_processing(s
, page
, freelist
, addr
))
2331 goto new_slab
; /* Slab failed checks. Next slab needed */
2333 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2336 local_irq_restore(flags
);
2341 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2342 * have the fastpath folded into their functions. So no function call
2343 * overhead for requests that can be satisfied on the fastpath.
2345 * The fastpath works by first checking if the lockless freelist can be used.
2346 * If not then __slab_alloc is called for slow processing.
2348 * Otherwise we can simply pick the next object from the lockless free list.
2350 static __always_inline
void *slab_alloc_node(struct kmem_cache
*s
,
2351 gfp_t gfpflags
, int node
, unsigned long addr
)
2354 struct kmem_cache_cpu
*c
;
2358 if (slab_pre_alloc_hook(s
, gfpflags
))
2361 s
= memcg_kmem_get_cache(s
, gfpflags
);
2364 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2365 * enabled. We may switch back and forth between cpus while
2366 * reading from one cpu area. That does not matter as long
2367 * as we end up on the original cpu again when doing the cmpxchg.
2369 * Preemption is disabled for the retrieval of the tid because that
2370 * must occur from the current processor. We cannot allow rescheduling
2371 * on a different processor between the determination of the pointer
2372 * and the retrieval of the tid.
2375 c
= __this_cpu_ptr(s
->cpu_slab
);
2378 * The transaction ids are globally unique per cpu and per operation on
2379 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2380 * occurs on the right processor and that there was no operation on the
2381 * linked list in between.
2386 object
= c
->freelist
;
2388 if (unlikely(!object
|| !node_match(page
, node
)))
2389 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2392 void *next_object
= get_freepointer_safe(s
, object
);
2395 * The cmpxchg will only match if there was no additional
2396 * operation and if we are on the right processor.
2398 * The cmpxchg does the following atomically (without lock
2400 * 1. Relocate first pointer to the current per cpu area.
2401 * 2. Verify that tid and freelist have not been changed
2402 * 3. If they were not changed replace tid and freelist
2404 * Since this is without lock semantics the protection is only
2405 * against code executing on this cpu *not* from access by
2408 if (unlikely(!this_cpu_cmpxchg_double(
2409 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2411 next_object
, next_tid(tid
)))) {
2413 note_cmpxchg_failure("slab_alloc", s
, tid
);
2416 prefetch_freepointer(s
, next_object
);
2417 stat(s
, ALLOC_FASTPATH
);
2420 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2421 memset(object
, 0, s
->object_size
);
2423 slab_post_alloc_hook(s
, gfpflags
, object
);
2428 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2429 gfp_t gfpflags
, unsigned long addr
)
2431 return slab_alloc_node(s
, gfpflags
, NUMA_NO_NODE
, addr
);
2434 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2436 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2438 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
,
2443 EXPORT_SYMBOL(kmem_cache_alloc
);
2445 #ifdef CONFIG_TRACING
2446 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2448 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2449 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2452 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2454 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2456 void *ret
= kmalloc_order(size
, flags
, order
);
2457 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2460 EXPORT_SYMBOL(kmalloc_order_trace
);
2464 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2466 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2468 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2469 s
->object_size
, s
->size
, gfpflags
, node
);
2473 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2475 #ifdef CONFIG_TRACING
2476 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2478 int node
, size_t size
)
2480 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2482 trace_kmalloc_node(_RET_IP_
, ret
,
2483 size
, s
->size
, gfpflags
, node
);
2486 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2491 * Slow patch handling. This may still be called frequently since objects
2492 * have a longer lifetime than the cpu slabs in most processing loads.
2494 * So we still attempt to reduce cache line usage. Just take the slab
2495 * lock and free the item. If there is no additional partial page
2496 * handling required then we can return immediately.
2498 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2499 void *x
, unsigned long addr
)
2502 void **object
= (void *)x
;
2505 unsigned long counters
;
2506 struct kmem_cache_node
*n
= NULL
;
2507 unsigned long uninitialized_var(flags
);
2509 stat(s
, FREE_SLOWPATH
);
2511 if (kmem_cache_debug(s
) &&
2512 !(n
= free_debug_processing(s
, page
, x
, addr
, &flags
)))
2517 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2520 prior
= page
->freelist
;
2521 counters
= page
->counters
;
2522 set_freepointer(s
, object
, prior
);
2523 new.counters
= counters
;
2524 was_frozen
= new.frozen
;
2526 if ((!new.inuse
|| !prior
) && !was_frozen
) {
2528 if (kmem_cache_has_cpu_partial(s
) && !prior
)
2531 * Slab was on no list before and will be
2533 * We can defer the list move and instead
2538 else { /* Needs to be taken off a list */
2540 n
= get_node(s
, page_to_nid(page
));
2542 * Speculatively acquire the list_lock.
2543 * If the cmpxchg does not succeed then we may
2544 * drop the list_lock without any processing.
2546 * Otherwise the list_lock will synchronize with
2547 * other processors updating the list of slabs.
2549 spin_lock_irqsave(&n
->list_lock
, flags
);
2554 } while (!cmpxchg_double_slab(s
, page
,
2556 object
, new.counters
,
2562 * If we just froze the page then put it onto the
2563 * per cpu partial list.
2565 if (new.frozen
&& !was_frozen
) {
2566 put_cpu_partial(s
, page
, 1);
2567 stat(s
, CPU_PARTIAL_FREE
);
2570 * The list lock was not taken therefore no list
2571 * activity can be necessary.
2574 stat(s
, FREE_FROZEN
);
2578 if (unlikely(!new.inuse
&& n
->nr_partial
> s
->min_partial
))
2582 * Objects left in the slab. If it was not on the partial list before
2585 if (!kmem_cache_has_cpu_partial(s
) && unlikely(!prior
)) {
2586 if (kmem_cache_debug(s
))
2587 remove_full(s
, page
);
2588 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2589 stat(s
, FREE_ADD_PARTIAL
);
2591 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2597 * Slab on the partial list.
2599 remove_partial(n
, page
);
2600 stat(s
, FREE_REMOVE_PARTIAL
);
2602 /* Slab must be on the full list */
2603 remove_full(s
, page
);
2605 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2607 discard_slab(s
, page
);
2611 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2612 * can perform fastpath freeing without additional function calls.
2614 * The fastpath is only possible if we are freeing to the current cpu slab
2615 * of this processor. This typically the case if we have just allocated
2618 * If fastpath is not possible then fall back to __slab_free where we deal
2619 * with all sorts of special processing.
2621 static __always_inline
void slab_free(struct kmem_cache
*s
,
2622 struct page
*page
, void *x
, unsigned long addr
)
2624 void **object
= (void *)x
;
2625 struct kmem_cache_cpu
*c
;
2628 slab_free_hook(s
, x
);
2632 * Determine the currently cpus per cpu slab.
2633 * The cpu may change afterward. However that does not matter since
2634 * data is retrieved via this pointer. If we are on the same cpu
2635 * during the cmpxchg then the free will succedd.
2638 c
= __this_cpu_ptr(s
->cpu_slab
);
2643 if (likely(page
== c
->page
)) {
2644 set_freepointer(s
, object
, c
->freelist
);
2646 if (unlikely(!this_cpu_cmpxchg_double(
2647 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2649 object
, next_tid(tid
)))) {
2651 note_cmpxchg_failure("slab_free", s
, tid
);
2654 stat(s
, FREE_FASTPATH
);
2656 __slab_free(s
, page
, x
, addr
);
2660 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2662 s
= cache_from_obj(s
, x
);
2665 slab_free(s
, virt_to_head_page(x
), x
, _RET_IP_
);
2666 trace_kmem_cache_free(_RET_IP_
, x
);
2668 EXPORT_SYMBOL(kmem_cache_free
);
2671 * Object placement in a slab is made very easy because we always start at
2672 * offset 0. If we tune the size of the object to the alignment then we can
2673 * get the required alignment by putting one properly sized object after
2676 * Notice that the allocation order determines the sizes of the per cpu
2677 * caches. Each processor has always one slab available for allocations.
2678 * Increasing the allocation order reduces the number of times that slabs
2679 * must be moved on and off the partial lists and is therefore a factor in
2684 * Mininum / Maximum order of slab pages. This influences locking overhead
2685 * and slab fragmentation. A higher order reduces the number of partial slabs
2686 * and increases the number of allocations possible without having to
2687 * take the list_lock.
2689 static int slub_min_order
;
2690 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2691 static int slub_min_objects
;
2694 * Merge control. If this is set then no merging of slab caches will occur.
2695 * (Could be removed. This was introduced to pacify the merge skeptics.)
2697 static int slub_nomerge
;
2700 * Calculate the order of allocation given an slab object size.
2702 * The order of allocation has significant impact on performance and other
2703 * system components. Generally order 0 allocations should be preferred since
2704 * order 0 does not cause fragmentation in the page allocator. Larger objects
2705 * be problematic to put into order 0 slabs because there may be too much
2706 * unused space left. We go to a higher order if more than 1/16th of the slab
2709 * In order to reach satisfactory performance we must ensure that a minimum
2710 * number of objects is in one slab. Otherwise we may generate too much
2711 * activity on the partial lists which requires taking the list_lock. This is
2712 * less a concern for large slabs though which are rarely used.
2714 * slub_max_order specifies the order where we begin to stop considering the
2715 * number of objects in a slab as critical. If we reach slub_max_order then
2716 * we try to keep the page order as low as possible. So we accept more waste
2717 * of space in favor of a small page order.
2719 * Higher order allocations also allow the placement of more objects in a
2720 * slab and thereby reduce object handling overhead. If the user has
2721 * requested a higher mininum order then we start with that one instead of
2722 * the smallest order which will fit the object.
2724 static inline int slab_order(int size
, int min_objects
,
2725 int max_order
, int fract_leftover
, int reserved
)
2729 int min_order
= slub_min_order
;
2731 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2732 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2734 for (order
= max(min_order
,
2735 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2736 order
<= max_order
; order
++) {
2738 unsigned long slab_size
= PAGE_SIZE
<< order
;
2740 if (slab_size
< min_objects
* size
+ reserved
)
2743 rem
= (slab_size
- reserved
) % size
;
2745 if (rem
<= slab_size
/ fract_leftover
)
2753 static inline int calculate_order(int size
, int reserved
)
2761 * Attempt to find best configuration for a slab. This
2762 * works by first attempting to generate a layout with
2763 * the best configuration and backing off gradually.
2765 * First we reduce the acceptable waste in a slab. Then
2766 * we reduce the minimum objects required in a slab.
2768 min_objects
= slub_min_objects
;
2770 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2771 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2772 min_objects
= min(min_objects
, max_objects
);
2774 while (min_objects
> 1) {
2776 while (fraction
>= 4) {
2777 order
= slab_order(size
, min_objects
,
2778 slub_max_order
, fraction
, reserved
);
2779 if (order
<= slub_max_order
)
2787 * We were unable to place multiple objects in a slab. Now
2788 * lets see if we can place a single object there.
2790 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2791 if (order
<= slub_max_order
)
2795 * Doh this slab cannot be placed using slub_max_order.
2797 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2798 if (order
< MAX_ORDER
)
2804 init_kmem_cache_node(struct kmem_cache_node
*n
)
2807 spin_lock_init(&n
->list_lock
);
2808 INIT_LIST_HEAD(&n
->partial
);
2809 #ifdef CONFIG_SLUB_DEBUG
2810 atomic_long_set(&n
->nr_slabs
, 0);
2811 atomic_long_set(&n
->total_objects
, 0);
2812 INIT_LIST_HEAD(&n
->full
);
2816 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2818 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2819 KMALLOC_SHIFT_HIGH
* sizeof(struct kmem_cache_cpu
));
2822 * Must align to double word boundary for the double cmpxchg
2823 * instructions to work; see __pcpu_double_call_return_bool().
2825 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2826 2 * sizeof(void *));
2831 init_kmem_cache_cpus(s
);
2836 static struct kmem_cache
*kmem_cache_node
;
2839 * No kmalloc_node yet so do it by hand. We know that this is the first
2840 * slab on the node for this slabcache. There are no concurrent accesses
2843 * Note that this function only works on the kmalloc_node_cache
2844 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2845 * memory on a fresh node that has no slab structures yet.
2847 static void early_kmem_cache_node_alloc(int node
)
2850 struct kmem_cache_node
*n
;
2852 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2854 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2857 if (page_to_nid(page
) != node
) {
2858 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2860 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2861 "in order to be able to continue\n");
2866 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2869 kmem_cache_node
->node
[node
] = n
;
2870 #ifdef CONFIG_SLUB_DEBUG
2871 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2872 init_tracking(kmem_cache_node
, n
);
2874 init_kmem_cache_node(n
);
2875 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2877 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2880 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2884 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2885 struct kmem_cache_node
*n
= s
->node
[node
];
2888 kmem_cache_free(kmem_cache_node
, n
);
2890 s
->node
[node
] = NULL
;
2894 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2898 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2899 struct kmem_cache_node
*n
;
2901 if (slab_state
== DOWN
) {
2902 early_kmem_cache_node_alloc(node
);
2905 n
= kmem_cache_alloc_node(kmem_cache_node
,
2909 free_kmem_cache_nodes(s
);
2914 init_kmem_cache_node(n
);
2919 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2921 if (min
< MIN_PARTIAL
)
2923 else if (min
> MAX_PARTIAL
)
2925 s
->min_partial
= min
;
2929 * calculate_sizes() determines the order and the distribution of data within
2932 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2934 unsigned long flags
= s
->flags
;
2935 unsigned long size
= s
->object_size
;
2939 * Round up object size to the next word boundary. We can only
2940 * place the free pointer at word boundaries and this determines
2941 * the possible location of the free pointer.
2943 size
= ALIGN(size
, sizeof(void *));
2945 #ifdef CONFIG_SLUB_DEBUG
2947 * Determine if we can poison the object itself. If the user of
2948 * the slab may touch the object after free or before allocation
2949 * then we should never poison the object itself.
2951 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2953 s
->flags
|= __OBJECT_POISON
;
2955 s
->flags
&= ~__OBJECT_POISON
;
2959 * If we are Redzoning then check if there is some space between the
2960 * end of the object and the free pointer. If not then add an
2961 * additional word to have some bytes to store Redzone information.
2963 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
2964 size
+= sizeof(void *);
2968 * With that we have determined the number of bytes in actual use
2969 * by the object. This is the potential offset to the free pointer.
2973 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2976 * Relocate free pointer after the object if it is not
2977 * permitted to overwrite the first word of the object on
2980 * This is the case if we do RCU, have a constructor or
2981 * destructor or are poisoning the objects.
2984 size
+= sizeof(void *);
2987 #ifdef CONFIG_SLUB_DEBUG
2988 if (flags
& SLAB_STORE_USER
)
2990 * Need to store information about allocs and frees after
2993 size
+= 2 * sizeof(struct track
);
2995 if (flags
& SLAB_RED_ZONE
)
2997 * Add some empty padding so that we can catch
2998 * overwrites from earlier objects rather than let
2999 * tracking information or the free pointer be
3000 * corrupted if a user writes before the start
3003 size
+= sizeof(void *);
3007 * SLUB stores one object immediately after another beginning from
3008 * offset 0. In order to align the objects we have to simply size
3009 * each object to conform to the alignment.
3011 size
= ALIGN(size
, s
->align
);
3013 if (forced_order
>= 0)
3014 order
= forced_order
;
3016 order
= calculate_order(size
, s
->reserved
);
3023 s
->allocflags
|= __GFP_COMP
;
3025 if (s
->flags
& SLAB_CACHE_DMA
)
3026 s
->allocflags
|= GFP_DMA
;
3028 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3029 s
->allocflags
|= __GFP_RECLAIMABLE
;
3032 * Determine the number of objects per slab
3034 s
->oo
= oo_make(order
, size
, s
->reserved
);
3035 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
3036 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3039 return !!oo_objects(s
->oo
);
3042 static int kmem_cache_open(struct kmem_cache
*s
, unsigned long flags
)
3044 s
->flags
= kmem_cache_flags(s
->size
, flags
, s
->name
, s
->ctor
);
3047 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3048 s
->reserved
= sizeof(struct rcu_head
);
3050 if (!calculate_sizes(s
, -1))
3052 if (disable_higher_order_debug
) {
3054 * Disable debugging flags that store metadata if the min slab
3057 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3058 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3060 if (!calculate_sizes(s
, -1))
3065 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3066 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3067 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3068 /* Enable fast mode */
3069 s
->flags
|= __CMPXCHG_DOUBLE
;
3073 * The larger the object size is, the more pages we want on the partial
3074 * list to avoid pounding the page allocator excessively.
3076 set_min_partial(s
, ilog2(s
->size
) / 2);
3079 * cpu_partial determined the maximum number of objects kept in the
3080 * per cpu partial lists of a processor.
3082 * Per cpu partial lists mainly contain slabs that just have one
3083 * object freed. If they are used for allocation then they can be
3084 * filled up again with minimal effort. The slab will never hit the
3085 * per node partial lists and therefore no locking will be required.
3087 * This setting also determines
3089 * A) The number of objects from per cpu partial slabs dumped to the
3090 * per node list when we reach the limit.
3091 * B) The number of objects in cpu partial slabs to extract from the
3092 * per node list when we run out of per cpu objects. We only fetch
3093 * 50% to keep some capacity around for frees.
3095 if (!kmem_cache_has_cpu_partial(s
))
3097 else if (s
->size
>= PAGE_SIZE
)
3099 else if (s
->size
>= 1024)
3101 else if (s
->size
>= 256)
3102 s
->cpu_partial
= 13;
3104 s
->cpu_partial
= 30;
3107 s
->remote_node_defrag_ratio
= 1000;
3109 if (!init_kmem_cache_nodes(s
))
3112 if (alloc_kmem_cache_cpus(s
))
3115 free_kmem_cache_nodes(s
);
3117 if (flags
& SLAB_PANIC
)
3118 panic("Cannot create slab %s size=%lu realsize=%u "
3119 "order=%u offset=%u flags=%lx\n",
3120 s
->name
, (unsigned long)s
->size
, s
->size
,
3121 oo_order(s
->oo
), s
->offset
, flags
);
3125 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3128 #ifdef CONFIG_SLUB_DEBUG
3129 void *addr
= page_address(page
);
3131 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3132 sizeof(long), GFP_ATOMIC
);
3135 slab_err(s
, page
, text
, s
->name
);
3138 get_map(s
, page
, map
);
3139 for_each_object(p
, s
, addr
, page
->objects
) {
3141 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3142 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3144 print_tracking(s
, p
);
3153 * Attempt to free all partial slabs on a node.
3154 * This is called from kmem_cache_close(). We must be the last thread
3155 * using the cache and therefore we do not need to lock anymore.
3157 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3159 struct page
*page
, *h
;
3161 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3163 remove_partial(n
, page
);
3164 discard_slab(s
, page
);
3166 list_slab_objects(s
, page
,
3167 "Objects remaining in %s on kmem_cache_close()");
3173 * Release all resources used by a slab cache.
3175 static inline int kmem_cache_close(struct kmem_cache
*s
)
3180 /* Attempt to free all objects */
3181 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3182 struct kmem_cache_node
*n
= get_node(s
, node
);
3185 if (n
->nr_partial
|| slabs_node(s
, node
))
3188 free_percpu(s
->cpu_slab
);
3189 free_kmem_cache_nodes(s
);
3193 int __kmem_cache_shutdown(struct kmem_cache
*s
)
3195 int rc
= kmem_cache_close(s
);
3199 * We do the same lock strategy around sysfs_slab_add, see
3200 * __kmem_cache_create. Because this is pretty much the last
3201 * operation we do and the lock will be released shortly after
3202 * that in slab_common.c, we could just move sysfs_slab_remove
3203 * to a later point in common code. We should do that when we
3204 * have a common sysfs framework for all allocators.
3206 mutex_unlock(&slab_mutex
);
3207 sysfs_slab_remove(s
);
3208 mutex_lock(&slab_mutex
);
3214 /********************************************************************
3216 *******************************************************************/
3218 static int __init
setup_slub_min_order(char *str
)
3220 get_option(&str
, &slub_min_order
);
3225 __setup("slub_min_order=", setup_slub_min_order
);
3227 static int __init
setup_slub_max_order(char *str
)
3229 get_option(&str
, &slub_max_order
);
3230 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3235 __setup("slub_max_order=", setup_slub_max_order
);
3237 static int __init
setup_slub_min_objects(char *str
)
3239 get_option(&str
, &slub_min_objects
);
3244 __setup("slub_min_objects=", setup_slub_min_objects
);
3246 static int __init
setup_slub_nomerge(char *str
)
3252 __setup("slub_nomerge", setup_slub_nomerge
);
3254 void *__kmalloc(size_t size
, gfp_t flags
)
3256 struct kmem_cache
*s
;
3259 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3260 return kmalloc_large(size
, flags
);
3262 s
= kmalloc_slab(size
, flags
);
3264 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3267 ret
= slab_alloc(s
, flags
, _RET_IP_
);
3269 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3273 EXPORT_SYMBOL(__kmalloc
);
3276 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3281 flags
|= __GFP_COMP
| __GFP_NOTRACK
| __GFP_KMEMCG
;
3282 page
= alloc_pages_node(node
, flags
, get_order(size
));
3284 ptr
= page_address(page
);
3286 kmemleak_alloc(ptr
, size
, 1, flags
);
3290 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3292 struct kmem_cache
*s
;
3295 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3296 ret
= kmalloc_large_node(size
, flags
, node
);
3298 trace_kmalloc_node(_RET_IP_
, ret
,
3299 size
, PAGE_SIZE
<< get_order(size
),
3305 s
= kmalloc_slab(size
, flags
);
3307 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3310 ret
= slab_alloc_node(s
, flags
, node
, _RET_IP_
);
3312 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3316 EXPORT_SYMBOL(__kmalloc_node
);
3319 size_t ksize(const void *object
)
3323 if (unlikely(object
== ZERO_SIZE_PTR
))
3326 page
= virt_to_head_page(object
);
3328 if (unlikely(!PageSlab(page
))) {
3329 WARN_ON(!PageCompound(page
));
3330 return PAGE_SIZE
<< compound_order(page
);
3333 return slab_ksize(page
->slab_cache
);
3335 EXPORT_SYMBOL(ksize
);
3337 #ifdef CONFIG_SLUB_DEBUG
3338 bool verify_mem_not_deleted(const void *x
)
3341 void *object
= (void *)x
;
3342 unsigned long flags
;
3345 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3348 local_irq_save(flags
);
3350 page
= virt_to_head_page(x
);
3351 if (unlikely(!PageSlab(page
))) {
3352 /* maybe it was from stack? */
3358 if (on_freelist(page
->slab_cache
, page
, object
)) {
3359 object_err(page
->slab_cache
, page
, object
,
3360 "Object is on free-list");
3368 local_irq_restore(flags
);
3371 EXPORT_SYMBOL(verify_mem_not_deleted
);
3374 void kfree(const void *x
)
3377 void *object
= (void *)x
;
3379 trace_kfree(_RET_IP_
, x
);
3381 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3384 page
= virt_to_head_page(x
);
3385 if (unlikely(!PageSlab(page
))) {
3386 BUG_ON(!PageCompound(page
));
3388 __free_memcg_kmem_pages(page
, compound_order(page
));
3391 slab_free(page
->slab_cache
, page
, object
, _RET_IP_
);
3393 EXPORT_SYMBOL(kfree
);
3396 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3397 * the remaining slabs by the number of items in use. The slabs with the
3398 * most items in use come first. New allocations will then fill those up
3399 * and thus they can be removed from the partial lists.
3401 * The slabs with the least items are placed last. This results in them
3402 * being allocated from last increasing the chance that the last objects
3403 * are freed in them.
3405 int kmem_cache_shrink(struct kmem_cache
*s
)
3409 struct kmem_cache_node
*n
;
3412 int objects
= oo_objects(s
->max
);
3413 struct list_head
*slabs_by_inuse
=
3414 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3415 unsigned long flags
;
3417 if (!slabs_by_inuse
)
3421 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3422 n
= get_node(s
, node
);
3427 for (i
= 0; i
< objects
; i
++)
3428 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3430 spin_lock_irqsave(&n
->list_lock
, flags
);
3433 * Build lists indexed by the items in use in each slab.
3435 * Note that concurrent frees may occur while we hold the
3436 * list_lock. page->inuse here is the upper limit.
3438 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3439 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3445 * Rebuild the partial list with the slabs filled up most
3446 * first and the least used slabs at the end.
3448 for (i
= objects
- 1; i
> 0; i
--)
3449 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3451 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3453 /* Release empty slabs */
3454 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3455 discard_slab(s
, page
);
3458 kfree(slabs_by_inuse
);
3461 EXPORT_SYMBOL(kmem_cache_shrink
);
3463 static int slab_mem_going_offline_callback(void *arg
)
3465 struct kmem_cache
*s
;
3467 mutex_lock(&slab_mutex
);
3468 list_for_each_entry(s
, &slab_caches
, list
)
3469 kmem_cache_shrink(s
);
3470 mutex_unlock(&slab_mutex
);
3475 static void slab_mem_offline_callback(void *arg
)
3477 struct kmem_cache_node
*n
;
3478 struct kmem_cache
*s
;
3479 struct memory_notify
*marg
= arg
;
3482 offline_node
= marg
->status_change_nid_normal
;
3485 * If the node still has available memory. we need kmem_cache_node
3488 if (offline_node
< 0)
3491 mutex_lock(&slab_mutex
);
3492 list_for_each_entry(s
, &slab_caches
, list
) {
3493 n
= get_node(s
, offline_node
);
3496 * if n->nr_slabs > 0, slabs still exist on the node
3497 * that is going down. We were unable to free them,
3498 * and offline_pages() function shouldn't call this
3499 * callback. So, we must fail.
3501 BUG_ON(slabs_node(s
, offline_node
));
3503 s
->node
[offline_node
] = NULL
;
3504 kmem_cache_free(kmem_cache_node
, n
);
3507 mutex_unlock(&slab_mutex
);
3510 static int slab_mem_going_online_callback(void *arg
)
3512 struct kmem_cache_node
*n
;
3513 struct kmem_cache
*s
;
3514 struct memory_notify
*marg
= arg
;
3515 int nid
= marg
->status_change_nid_normal
;
3519 * If the node's memory is already available, then kmem_cache_node is
3520 * already created. Nothing to do.
3526 * We are bringing a node online. No memory is available yet. We must
3527 * allocate a kmem_cache_node structure in order to bring the node
3530 mutex_lock(&slab_mutex
);
3531 list_for_each_entry(s
, &slab_caches
, list
) {
3533 * XXX: kmem_cache_alloc_node will fallback to other nodes
3534 * since memory is not yet available from the node that
3537 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3542 init_kmem_cache_node(n
);
3546 mutex_unlock(&slab_mutex
);
3550 static int slab_memory_callback(struct notifier_block
*self
,
3551 unsigned long action
, void *arg
)
3556 case MEM_GOING_ONLINE
:
3557 ret
= slab_mem_going_online_callback(arg
);
3559 case MEM_GOING_OFFLINE
:
3560 ret
= slab_mem_going_offline_callback(arg
);
3563 case MEM_CANCEL_ONLINE
:
3564 slab_mem_offline_callback(arg
);
3567 case MEM_CANCEL_OFFLINE
:
3571 ret
= notifier_from_errno(ret
);
3577 static struct notifier_block slab_memory_callback_nb
= {
3578 .notifier_call
= slab_memory_callback
,
3579 .priority
= SLAB_CALLBACK_PRI
,
3582 /********************************************************************
3583 * Basic setup of slabs
3584 *******************************************************************/
3587 * Used for early kmem_cache structures that were allocated using
3588 * the page allocator. Allocate them properly then fix up the pointers
3589 * that may be pointing to the wrong kmem_cache structure.
3592 static struct kmem_cache
* __init
bootstrap(struct kmem_cache
*static_cache
)
3595 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
3597 memcpy(s
, static_cache
, kmem_cache
->object_size
);
3600 * This runs very early, and only the boot processor is supposed to be
3601 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3604 __flush_cpu_slab(s
, smp_processor_id());
3605 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3606 struct kmem_cache_node
*n
= get_node(s
, node
);
3610 list_for_each_entry(p
, &n
->partial
, lru
)
3613 #ifdef CONFIG_SLUB_DEBUG
3614 list_for_each_entry(p
, &n
->full
, lru
)
3619 list_add(&s
->list
, &slab_caches
);
3623 void __init
kmem_cache_init(void)
3625 static __initdata
struct kmem_cache boot_kmem_cache
,
3626 boot_kmem_cache_node
;
3628 if (debug_guardpage_minorder())
3631 kmem_cache_node
= &boot_kmem_cache_node
;
3632 kmem_cache
= &boot_kmem_cache
;
3634 create_boot_cache(kmem_cache_node
, "kmem_cache_node",
3635 sizeof(struct kmem_cache_node
), SLAB_HWCACHE_ALIGN
);
3637 register_hotmemory_notifier(&slab_memory_callback_nb
);
3639 /* Able to allocate the per node structures */
3640 slab_state
= PARTIAL
;
3642 create_boot_cache(kmem_cache
, "kmem_cache",
3643 offsetof(struct kmem_cache
, node
) +
3644 nr_node_ids
* sizeof(struct kmem_cache_node
*),
3645 SLAB_HWCACHE_ALIGN
);
3647 kmem_cache
= bootstrap(&boot_kmem_cache
);
3650 * Allocate kmem_cache_node properly from the kmem_cache slab.
3651 * kmem_cache_node is separately allocated so no need to
3652 * update any list pointers.
3654 kmem_cache_node
= bootstrap(&boot_kmem_cache_node
);
3656 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3657 create_kmalloc_caches(0);
3660 register_cpu_notifier(&slab_notifier
);
3664 "SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d,"
3665 " CPUs=%d, Nodes=%d\n",
3667 slub_min_order
, slub_max_order
, slub_min_objects
,
3668 nr_cpu_ids
, nr_node_ids
);
3671 void __init
kmem_cache_init_late(void)
3676 * Find a mergeable slab cache
3678 static int slab_unmergeable(struct kmem_cache
*s
)
3680 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3687 * We may have set a slab to be unmergeable during bootstrap.
3689 if (s
->refcount
< 0)
3695 static struct kmem_cache
*find_mergeable(struct mem_cgroup
*memcg
, size_t size
,
3696 size_t align
, unsigned long flags
, const char *name
,
3697 void (*ctor
)(void *))
3699 struct kmem_cache
*s
;
3701 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3707 size
= ALIGN(size
, sizeof(void *));
3708 align
= calculate_alignment(flags
, align
, size
);
3709 size
= ALIGN(size
, align
);
3710 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3712 list_for_each_entry(s
, &slab_caches
, list
) {
3713 if (slab_unmergeable(s
))
3719 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3722 * Check if alignment is compatible.
3723 * Courtesy of Adrian Drzewiecki
3725 if ((s
->size
& ~(align
- 1)) != s
->size
)
3728 if (s
->size
- size
>= sizeof(void *))
3731 if (!cache_match_memcg(s
, memcg
))
3740 __kmem_cache_alias(struct mem_cgroup
*memcg
, const char *name
, size_t size
,
3741 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3743 struct kmem_cache
*s
;
3745 s
= find_mergeable(memcg
, size
, align
, flags
, name
, ctor
);
3749 * Adjust the object sizes so that we clear
3750 * the complete object on kzalloc.
3752 s
->object_size
= max(s
->object_size
, (int)size
);
3753 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3755 if (sysfs_slab_alias(s
, name
)) {
3764 int __kmem_cache_create(struct kmem_cache
*s
, unsigned long flags
)
3768 err
= kmem_cache_open(s
, flags
);
3772 /* Mutex is not taken during early boot */
3773 if (slab_state
<= UP
)
3776 memcg_propagate_slab_attrs(s
);
3777 mutex_unlock(&slab_mutex
);
3778 err
= sysfs_slab_add(s
);
3779 mutex_lock(&slab_mutex
);
3782 kmem_cache_close(s
);
3789 * Use the cpu notifier to insure that the cpu slabs are flushed when
3792 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3793 unsigned long action
, void *hcpu
)
3795 long cpu
= (long)hcpu
;
3796 struct kmem_cache
*s
;
3797 unsigned long flags
;
3800 case CPU_UP_CANCELED
:
3801 case CPU_UP_CANCELED_FROZEN
:
3803 case CPU_DEAD_FROZEN
:
3804 mutex_lock(&slab_mutex
);
3805 list_for_each_entry(s
, &slab_caches
, list
) {
3806 local_irq_save(flags
);
3807 __flush_cpu_slab(s
, cpu
);
3808 local_irq_restore(flags
);
3810 mutex_unlock(&slab_mutex
);
3818 static struct notifier_block __cpuinitdata slab_notifier
= {
3819 .notifier_call
= slab_cpuup_callback
3824 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3826 struct kmem_cache
*s
;
3829 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3830 return kmalloc_large(size
, gfpflags
);
3832 s
= kmalloc_slab(size
, gfpflags
);
3834 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3837 ret
= slab_alloc(s
, gfpflags
, caller
);
3839 /* Honor the call site pointer we received. */
3840 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3846 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3847 int node
, unsigned long caller
)
3849 struct kmem_cache
*s
;
3852 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3853 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3855 trace_kmalloc_node(caller
, ret
,
3856 size
, PAGE_SIZE
<< get_order(size
),
3862 s
= kmalloc_slab(size
, gfpflags
);
3864 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3867 ret
= slab_alloc_node(s
, gfpflags
, node
, caller
);
3869 /* Honor the call site pointer we received. */
3870 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3877 static int count_inuse(struct page
*page
)
3882 static int count_total(struct page
*page
)
3884 return page
->objects
;
3888 #ifdef CONFIG_SLUB_DEBUG
3889 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3893 void *addr
= page_address(page
);
3895 if (!check_slab(s
, page
) ||
3896 !on_freelist(s
, page
, NULL
))
3899 /* Now we know that a valid freelist exists */
3900 bitmap_zero(map
, page
->objects
);
3902 get_map(s
, page
, map
);
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_INACTIVE
))
3909 for_each_object(p
, s
, addr
, page
->objects
)
3910 if (!test_bit(slab_index(p
, s
, addr
), map
))
3911 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
3916 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3920 validate_slab(s
, page
, map
);
3924 static int validate_slab_node(struct kmem_cache
*s
,
3925 struct kmem_cache_node
*n
, unsigned long *map
)
3927 unsigned long count
= 0;
3929 unsigned long flags
;
3931 spin_lock_irqsave(&n
->list_lock
, flags
);
3933 list_for_each_entry(page
, &n
->partial
, lru
) {
3934 validate_slab_slab(s
, page
, map
);
3937 if (count
!= n
->nr_partial
)
3938 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3939 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3941 if (!(s
->flags
& SLAB_STORE_USER
))
3944 list_for_each_entry(page
, &n
->full
, lru
) {
3945 validate_slab_slab(s
, page
, map
);
3948 if (count
!= atomic_long_read(&n
->nr_slabs
))
3949 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3950 "counter=%ld\n", s
->name
, count
,
3951 atomic_long_read(&n
->nr_slabs
));
3954 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3958 static long validate_slab_cache(struct kmem_cache
*s
)
3961 unsigned long count
= 0;
3962 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3963 sizeof(unsigned long), GFP_KERNEL
);
3969 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3970 struct kmem_cache_node
*n
= get_node(s
, node
);
3972 count
+= validate_slab_node(s
, n
, map
);
3978 * Generate lists of code addresses where slabcache objects are allocated
3983 unsigned long count
;
3990 DECLARE_BITMAP(cpus
, NR_CPUS
);
3996 unsigned long count
;
3997 struct location
*loc
;
4000 static void free_loc_track(struct loc_track
*t
)
4003 free_pages((unsigned long)t
->loc
,
4004 get_order(sizeof(struct location
) * t
->max
));
4007 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4012 order
= get_order(sizeof(struct location
) * max
);
4014 l
= (void *)__get_free_pages(flags
, order
);
4019 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4027 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4028 const struct track
*track
)
4030 long start
, end
, pos
;
4032 unsigned long caddr
;
4033 unsigned long age
= jiffies
- track
->when
;
4039 pos
= start
+ (end
- start
+ 1) / 2;
4042 * There is nothing at "end". If we end up there
4043 * we need to add something to before end.
4048 caddr
= t
->loc
[pos
].addr
;
4049 if (track
->addr
== caddr
) {
4055 if (age
< l
->min_time
)
4057 if (age
> l
->max_time
)
4060 if (track
->pid
< l
->min_pid
)
4061 l
->min_pid
= track
->pid
;
4062 if (track
->pid
> l
->max_pid
)
4063 l
->max_pid
= track
->pid
;
4065 cpumask_set_cpu(track
->cpu
,
4066 to_cpumask(l
->cpus
));
4068 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4072 if (track
->addr
< caddr
)
4079 * Not found. Insert new tracking element.
4081 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4087 (t
->count
- pos
) * sizeof(struct location
));
4090 l
->addr
= track
->addr
;
4094 l
->min_pid
= track
->pid
;
4095 l
->max_pid
= track
->pid
;
4096 cpumask_clear(to_cpumask(l
->cpus
));
4097 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4098 nodes_clear(l
->nodes
);
4099 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4103 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4104 struct page
*page
, enum track_item alloc
,
4107 void *addr
= page_address(page
);
4110 bitmap_zero(map
, page
->objects
);
4111 get_map(s
, page
, map
);
4113 for_each_object(p
, s
, addr
, page
->objects
)
4114 if (!test_bit(slab_index(p
, s
, addr
), map
))
4115 add_location(t
, s
, get_track(s
, p
, alloc
));
4118 static int list_locations(struct kmem_cache
*s
, char *buf
,
4119 enum track_item alloc
)
4123 struct loc_track t
= { 0, 0, NULL
};
4125 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4126 sizeof(unsigned long), GFP_KERNEL
);
4128 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4131 return sprintf(buf
, "Out of memory\n");
4133 /* Push back cpu slabs */
4136 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4137 struct kmem_cache_node
*n
= get_node(s
, node
);
4138 unsigned long flags
;
4141 if (!atomic_long_read(&n
->nr_slabs
))
4144 spin_lock_irqsave(&n
->list_lock
, flags
);
4145 list_for_each_entry(page
, &n
->partial
, lru
)
4146 process_slab(&t
, s
, page
, alloc
, map
);
4147 list_for_each_entry(page
, &n
->full
, lru
)
4148 process_slab(&t
, s
, page
, alloc
, map
);
4149 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4152 for (i
= 0; i
< t
.count
; i
++) {
4153 struct location
*l
= &t
.loc
[i
];
4155 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4157 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4160 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4162 len
+= sprintf(buf
+ len
, "<not-available>");
4164 if (l
->sum_time
!= l
->min_time
) {
4165 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4167 (long)div_u64(l
->sum_time
, l
->count
),
4170 len
+= sprintf(buf
+ len
, " age=%ld",
4173 if (l
->min_pid
!= l
->max_pid
)
4174 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4175 l
->min_pid
, l
->max_pid
);
4177 len
+= sprintf(buf
+ len
, " pid=%ld",
4180 if (num_online_cpus() > 1 &&
4181 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4182 len
< PAGE_SIZE
- 60) {
4183 len
+= sprintf(buf
+ len
, " cpus=");
4184 len
+= cpulist_scnprintf(buf
+ len
,
4185 PAGE_SIZE
- len
- 50,
4186 to_cpumask(l
->cpus
));
4189 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4190 len
< PAGE_SIZE
- 60) {
4191 len
+= sprintf(buf
+ len
, " nodes=");
4192 len
+= nodelist_scnprintf(buf
+ len
,
4193 PAGE_SIZE
- len
- 50,
4197 len
+= sprintf(buf
+ len
, "\n");
4203 len
+= sprintf(buf
, "No data\n");
4208 #ifdef SLUB_RESILIENCY_TEST
4209 static void resiliency_test(void)
4213 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || KMALLOC_SHIFT_HIGH
< 10);
4215 printk(KERN_ERR
"SLUB resiliency testing\n");
4216 printk(KERN_ERR
"-----------------------\n");
4217 printk(KERN_ERR
"A. Corruption after allocation\n");
4219 p
= kzalloc(16, GFP_KERNEL
);
4221 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4222 " 0x12->0x%p\n\n", p
+ 16);
4224 validate_slab_cache(kmalloc_caches
[4]);
4226 /* Hmmm... The next two are dangerous */
4227 p
= kzalloc(32, GFP_KERNEL
);
4228 p
[32 + sizeof(void *)] = 0x34;
4229 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4230 " 0x34 -> -0x%p\n", p
);
4232 "If allocated object is overwritten then not detectable\n\n");
4234 validate_slab_cache(kmalloc_caches
[5]);
4235 p
= kzalloc(64, GFP_KERNEL
);
4236 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4238 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4241 "If allocated object is overwritten then not detectable\n\n");
4242 validate_slab_cache(kmalloc_caches
[6]);
4244 printk(KERN_ERR
"\nB. Corruption after free\n");
4245 p
= kzalloc(128, GFP_KERNEL
);
4248 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4249 validate_slab_cache(kmalloc_caches
[7]);
4251 p
= kzalloc(256, GFP_KERNEL
);
4254 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4256 validate_slab_cache(kmalloc_caches
[8]);
4258 p
= kzalloc(512, GFP_KERNEL
);
4261 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4262 validate_slab_cache(kmalloc_caches
[9]);
4266 static void resiliency_test(void) {};
4271 enum slab_stat_type
{
4272 SL_ALL
, /* All slabs */
4273 SL_PARTIAL
, /* Only partially allocated slabs */
4274 SL_CPU
, /* Only slabs used for cpu caches */
4275 SL_OBJECTS
, /* Determine allocated objects not slabs */
4276 SL_TOTAL
/* Determine object capacity not slabs */
4279 #define SO_ALL (1 << SL_ALL)
4280 #define SO_PARTIAL (1 << SL_PARTIAL)
4281 #define SO_CPU (1 << SL_CPU)
4282 #define SO_OBJECTS (1 << SL_OBJECTS)
4283 #define SO_TOTAL (1 << SL_TOTAL)
4285 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4286 char *buf
, unsigned long flags
)
4288 unsigned long total
= 0;
4291 unsigned long *nodes
;
4293 nodes
= kzalloc(sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4297 if (flags
& SO_CPU
) {
4300 for_each_possible_cpu(cpu
) {
4301 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
,
4306 page
= ACCESS_ONCE(c
->page
);
4310 node
= page_to_nid(page
);
4311 if (flags
& SO_TOTAL
)
4313 else if (flags
& SO_OBJECTS
)
4321 page
= ACCESS_ONCE(c
->partial
);
4330 lock_memory_hotplug();
4331 #ifdef CONFIG_SLUB_DEBUG
4332 if (flags
& SO_ALL
) {
4333 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4334 struct kmem_cache_node
*n
= get_node(s
, node
);
4336 if (flags
& SO_TOTAL
)
4337 x
= atomic_long_read(&n
->total_objects
);
4338 else if (flags
& SO_OBJECTS
)
4339 x
= atomic_long_read(&n
->total_objects
) -
4340 count_partial(n
, count_free
);
4342 x
= atomic_long_read(&n
->nr_slabs
);
4349 if (flags
& SO_PARTIAL
) {
4350 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4351 struct kmem_cache_node
*n
= get_node(s
, node
);
4353 if (flags
& SO_TOTAL
)
4354 x
= count_partial(n
, count_total
);
4355 else if (flags
& SO_OBJECTS
)
4356 x
= count_partial(n
, count_inuse
);
4363 x
= sprintf(buf
, "%lu", total
);
4365 for_each_node_state(node
, N_NORMAL_MEMORY
)
4367 x
+= sprintf(buf
+ x
, " N%d=%lu",
4370 unlock_memory_hotplug();
4372 return x
+ sprintf(buf
+ x
, "\n");
4375 #ifdef CONFIG_SLUB_DEBUG
4376 static int any_slab_objects(struct kmem_cache
*s
)
4380 for_each_online_node(node
) {
4381 struct kmem_cache_node
*n
= get_node(s
, node
);
4386 if (atomic_long_read(&n
->total_objects
))
4393 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4394 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4396 struct slab_attribute
{
4397 struct attribute attr
;
4398 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4399 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4402 #define SLAB_ATTR_RO(_name) \
4403 static struct slab_attribute _name##_attr = \
4404 __ATTR(_name, 0400, _name##_show, NULL)
4406 #define SLAB_ATTR(_name) \
4407 static struct slab_attribute _name##_attr = \
4408 __ATTR(_name, 0600, _name##_show, _name##_store)
4410 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4412 return sprintf(buf
, "%d\n", s
->size
);
4414 SLAB_ATTR_RO(slab_size
);
4416 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4418 return sprintf(buf
, "%d\n", s
->align
);
4420 SLAB_ATTR_RO(align
);
4422 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4424 return sprintf(buf
, "%d\n", s
->object_size
);
4426 SLAB_ATTR_RO(object_size
);
4428 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4430 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4432 SLAB_ATTR_RO(objs_per_slab
);
4434 static ssize_t
order_store(struct kmem_cache
*s
,
4435 const char *buf
, size_t length
)
4437 unsigned long order
;
4440 err
= strict_strtoul(buf
, 10, &order
);
4444 if (order
> slub_max_order
|| order
< slub_min_order
)
4447 calculate_sizes(s
, order
);
4451 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4453 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4457 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4459 return sprintf(buf
, "%lu\n", s
->min_partial
);
4462 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4468 err
= strict_strtoul(buf
, 10, &min
);
4472 set_min_partial(s
, min
);
4475 SLAB_ATTR(min_partial
);
4477 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4479 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4482 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4485 unsigned long objects
;
4488 err
= strict_strtoul(buf
, 10, &objects
);
4491 if (objects
&& !kmem_cache_has_cpu_partial(s
))
4494 s
->cpu_partial
= objects
;
4498 SLAB_ATTR(cpu_partial
);
4500 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4504 return sprintf(buf
, "%pS\n", s
->ctor
);
4508 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4510 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4512 SLAB_ATTR_RO(aliases
);
4514 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4516 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4518 SLAB_ATTR_RO(partial
);
4520 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4522 return show_slab_objects(s
, buf
, SO_CPU
);
4524 SLAB_ATTR_RO(cpu_slabs
);
4526 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4528 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4530 SLAB_ATTR_RO(objects
);
4532 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4534 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4536 SLAB_ATTR_RO(objects_partial
);
4538 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4545 for_each_online_cpu(cpu
) {
4546 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4549 pages
+= page
->pages
;
4550 objects
+= page
->pobjects
;
4554 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4557 for_each_online_cpu(cpu
) {
4558 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4560 if (page
&& len
< PAGE_SIZE
- 20)
4561 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4562 page
->pobjects
, page
->pages
);
4565 return len
+ sprintf(buf
+ len
, "\n");
4567 SLAB_ATTR_RO(slabs_cpu_partial
);
4569 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4571 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4574 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4575 const char *buf
, size_t length
)
4577 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4579 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4582 SLAB_ATTR(reclaim_account
);
4584 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4586 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4588 SLAB_ATTR_RO(hwcache_align
);
4590 #ifdef CONFIG_ZONE_DMA
4591 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4593 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4595 SLAB_ATTR_RO(cache_dma
);
4598 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4600 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4602 SLAB_ATTR_RO(destroy_by_rcu
);
4604 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4606 return sprintf(buf
, "%d\n", s
->reserved
);
4608 SLAB_ATTR_RO(reserved
);
4610 #ifdef CONFIG_SLUB_DEBUG
4611 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4613 return show_slab_objects(s
, buf
, SO_ALL
);
4615 SLAB_ATTR_RO(slabs
);
4617 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4619 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4621 SLAB_ATTR_RO(total_objects
);
4623 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4625 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4628 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4629 const char *buf
, size_t length
)
4631 s
->flags
&= ~SLAB_DEBUG_FREE
;
4632 if (buf
[0] == '1') {
4633 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4634 s
->flags
|= SLAB_DEBUG_FREE
;
4638 SLAB_ATTR(sanity_checks
);
4640 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4642 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4645 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4648 s
->flags
&= ~SLAB_TRACE
;
4649 if (buf
[0] == '1') {
4650 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4651 s
->flags
|= SLAB_TRACE
;
4657 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4659 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4662 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4663 const char *buf
, size_t length
)
4665 if (any_slab_objects(s
))
4668 s
->flags
&= ~SLAB_RED_ZONE
;
4669 if (buf
[0] == '1') {
4670 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4671 s
->flags
|= SLAB_RED_ZONE
;
4673 calculate_sizes(s
, -1);
4676 SLAB_ATTR(red_zone
);
4678 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4680 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4683 static ssize_t
poison_store(struct kmem_cache
*s
,
4684 const char *buf
, size_t length
)
4686 if (any_slab_objects(s
))
4689 s
->flags
&= ~SLAB_POISON
;
4690 if (buf
[0] == '1') {
4691 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4692 s
->flags
|= SLAB_POISON
;
4694 calculate_sizes(s
, -1);
4699 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4701 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4704 static ssize_t
store_user_store(struct kmem_cache
*s
,
4705 const char *buf
, size_t length
)
4707 if (any_slab_objects(s
))
4710 s
->flags
&= ~SLAB_STORE_USER
;
4711 if (buf
[0] == '1') {
4712 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4713 s
->flags
|= SLAB_STORE_USER
;
4715 calculate_sizes(s
, -1);
4718 SLAB_ATTR(store_user
);
4720 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4725 static ssize_t
validate_store(struct kmem_cache
*s
,
4726 const char *buf
, size_t length
)
4730 if (buf
[0] == '1') {
4731 ret
= validate_slab_cache(s
);
4737 SLAB_ATTR(validate
);
4739 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4741 if (!(s
->flags
& SLAB_STORE_USER
))
4743 return list_locations(s
, buf
, TRACK_ALLOC
);
4745 SLAB_ATTR_RO(alloc_calls
);
4747 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4749 if (!(s
->flags
& SLAB_STORE_USER
))
4751 return list_locations(s
, buf
, TRACK_FREE
);
4753 SLAB_ATTR_RO(free_calls
);
4754 #endif /* CONFIG_SLUB_DEBUG */
4756 #ifdef CONFIG_FAILSLAB
4757 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4759 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4762 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4765 s
->flags
&= ~SLAB_FAILSLAB
;
4767 s
->flags
|= SLAB_FAILSLAB
;
4770 SLAB_ATTR(failslab
);
4773 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4778 static ssize_t
shrink_store(struct kmem_cache
*s
,
4779 const char *buf
, size_t length
)
4781 if (buf
[0] == '1') {
4782 int rc
= kmem_cache_shrink(s
);
4793 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4795 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4798 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4799 const char *buf
, size_t length
)
4801 unsigned long ratio
;
4804 err
= strict_strtoul(buf
, 10, &ratio
);
4809 s
->remote_node_defrag_ratio
= ratio
* 10;
4813 SLAB_ATTR(remote_node_defrag_ratio
);
4816 #ifdef CONFIG_SLUB_STATS
4817 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4819 unsigned long sum
= 0;
4822 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4827 for_each_online_cpu(cpu
) {
4828 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4834 len
= sprintf(buf
, "%lu", sum
);
4837 for_each_online_cpu(cpu
) {
4838 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4839 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4843 return len
+ sprintf(buf
+ len
, "\n");
4846 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4850 for_each_online_cpu(cpu
)
4851 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4854 #define STAT_ATTR(si, text) \
4855 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4857 return show_stat(s, buf, si); \
4859 static ssize_t text##_store(struct kmem_cache *s, \
4860 const char *buf, size_t length) \
4862 if (buf[0] != '0') \
4864 clear_stat(s, si); \
4869 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4870 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4871 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4872 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4873 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4874 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4875 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4876 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4877 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4878 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4879 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
4880 STAT_ATTR(FREE_SLAB
, free_slab
);
4881 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4882 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4883 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4884 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4885 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4886 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4887 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
4888 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4889 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
4890 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
4891 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
4892 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
4893 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
4894 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
4897 static struct attribute
*slab_attrs
[] = {
4898 &slab_size_attr
.attr
,
4899 &object_size_attr
.attr
,
4900 &objs_per_slab_attr
.attr
,
4902 &min_partial_attr
.attr
,
4903 &cpu_partial_attr
.attr
,
4905 &objects_partial_attr
.attr
,
4907 &cpu_slabs_attr
.attr
,
4911 &hwcache_align_attr
.attr
,
4912 &reclaim_account_attr
.attr
,
4913 &destroy_by_rcu_attr
.attr
,
4915 &reserved_attr
.attr
,
4916 &slabs_cpu_partial_attr
.attr
,
4917 #ifdef CONFIG_SLUB_DEBUG
4918 &total_objects_attr
.attr
,
4920 &sanity_checks_attr
.attr
,
4922 &red_zone_attr
.attr
,
4924 &store_user_attr
.attr
,
4925 &validate_attr
.attr
,
4926 &alloc_calls_attr
.attr
,
4927 &free_calls_attr
.attr
,
4929 #ifdef CONFIG_ZONE_DMA
4930 &cache_dma_attr
.attr
,
4933 &remote_node_defrag_ratio_attr
.attr
,
4935 #ifdef CONFIG_SLUB_STATS
4936 &alloc_fastpath_attr
.attr
,
4937 &alloc_slowpath_attr
.attr
,
4938 &free_fastpath_attr
.attr
,
4939 &free_slowpath_attr
.attr
,
4940 &free_frozen_attr
.attr
,
4941 &free_add_partial_attr
.attr
,
4942 &free_remove_partial_attr
.attr
,
4943 &alloc_from_partial_attr
.attr
,
4944 &alloc_slab_attr
.attr
,
4945 &alloc_refill_attr
.attr
,
4946 &alloc_node_mismatch_attr
.attr
,
4947 &free_slab_attr
.attr
,
4948 &cpuslab_flush_attr
.attr
,
4949 &deactivate_full_attr
.attr
,
4950 &deactivate_empty_attr
.attr
,
4951 &deactivate_to_head_attr
.attr
,
4952 &deactivate_to_tail_attr
.attr
,
4953 &deactivate_remote_frees_attr
.attr
,
4954 &deactivate_bypass_attr
.attr
,
4955 &order_fallback_attr
.attr
,
4956 &cmpxchg_double_fail_attr
.attr
,
4957 &cmpxchg_double_cpu_fail_attr
.attr
,
4958 &cpu_partial_alloc_attr
.attr
,
4959 &cpu_partial_free_attr
.attr
,
4960 &cpu_partial_node_attr
.attr
,
4961 &cpu_partial_drain_attr
.attr
,
4963 #ifdef CONFIG_FAILSLAB
4964 &failslab_attr
.attr
,
4970 static struct attribute_group slab_attr_group
= {
4971 .attrs
= slab_attrs
,
4974 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4975 struct attribute
*attr
,
4978 struct slab_attribute
*attribute
;
4979 struct kmem_cache
*s
;
4982 attribute
= to_slab_attr(attr
);
4985 if (!attribute
->show
)
4988 err
= attribute
->show(s
, buf
);
4993 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4994 struct attribute
*attr
,
4995 const char *buf
, size_t len
)
4997 struct slab_attribute
*attribute
;
4998 struct kmem_cache
*s
;
5001 attribute
= to_slab_attr(attr
);
5004 if (!attribute
->store
)
5007 err
= attribute
->store(s
, buf
, len
);
5008 #ifdef CONFIG_MEMCG_KMEM
5009 if (slab_state
>= FULL
&& err
>= 0 && is_root_cache(s
)) {
5012 mutex_lock(&slab_mutex
);
5013 if (s
->max_attr_size
< len
)
5014 s
->max_attr_size
= len
;
5017 * This is a best effort propagation, so this function's return
5018 * value will be determined by the parent cache only. This is
5019 * basically because not all attributes will have a well
5020 * defined semantics for rollbacks - most of the actions will
5021 * have permanent effects.
5023 * Returning the error value of any of the children that fail
5024 * is not 100 % defined, in the sense that users seeing the
5025 * error code won't be able to know anything about the state of
5028 * Only returning the error code for the parent cache at least
5029 * has well defined semantics. The cache being written to
5030 * directly either failed or succeeded, in which case we loop
5031 * through the descendants with best-effort propagation.
5033 for_each_memcg_cache_index(i
) {
5034 struct kmem_cache
*c
= cache_from_memcg(s
, i
);
5036 attribute
->store(c
, buf
, len
);
5038 mutex_unlock(&slab_mutex
);
5044 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
)
5046 #ifdef CONFIG_MEMCG_KMEM
5048 char *buffer
= NULL
;
5050 if (!is_root_cache(s
))
5054 * This mean this cache had no attribute written. Therefore, no point
5055 * in copying default values around
5057 if (!s
->max_attr_size
)
5060 for (i
= 0; i
< ARRAY_SIZE(slab_attrs
); i
++) {
5063 struct slab_attribute
*attr
= to_slab_attr(slab_attrs
[i
]);
5065 if (!attr
|| !attr
->store
|| !attr
->show
)
5069 * It is really bad that we have to allocate here, so we will
5070 * do it only as a fallback. If we actually allocate, though,
5071 * we can just use the allocated buffer until the end.
5073 * Most of the slub attributes will tend to be very small in
5074 * size, but sysfs allows buffers up to a page, so they can
5075 * theoretically happen.
5079 else if (s
->max_attr_size
< ARRAY_SIZE(mbuf
))
5082 buffer
= (char *) get_zeroed_page(GFP_KERNEL
);
5083 if (WARN_ON(!buffer
))
5088 attr
->show(s
->memcg_params
->root_cache
, buf
);
5089 attr
->store(s
, buf
, strlen(buf
));
5093 free_page((unsigned long)buffer
);
5097 static const struct sysfs_ops slab_sysfs_ops
= {
5098 .show
= slab_attr_show
,
5099 .store
= slab_attr_store
,
5102 static struct kobj_type slab_ktype
= {
5103 .sysfs_ops
= &slab_sysfs_ops
,
5106 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5108 struct kobj_type
*ktype
= get_ktype(kobj
);
5110 if (ktype
== &slab_ktype
)
5115 static const struct kset_uevent_ops slab_uevent_ops
= {
5116 .filter
= uevent_filter
,
5119 static struct kset
*slab_kset
;
5121 #define ID_STR_LENGTH 64
5123 /* Create a unique string id for a slab cache:
5125 * Format :[flags-]size
5127 static char *create_unique_id(struct kmem_cache
*s
)
5129 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5136 * First flags affecting slabcache operations. We will only
5137 * get here for aliasable slabs so we do not need to support
5138 * too many flags. The flags here must cover all flags that
5139 * are matched during merging to guarantee that the id is
5142 if (s
->flags
& SLAB_CACHE_DMA
)
5144 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5146 if (s
->flags
& SLAB_DEBUG_FREE
)
5148 if (!(s
->flags
& SLAB_NOTRACK
))
5152 p
+= sprintf(p
, "%07d", s
->size
);
5154 #ifdef CONFIG_MEMCG_KMEM
5155 if (!is_root_cache(s
))
5156 p
+= sprintf(p
, "-%08d",
5157 memcg_cache_id(s
->memcg_params
->memcg
));
5160 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5164 static int sysfs_slab_add(struct kmem_cache
*s
)
5168 int unmergeable
= slab_unmergeable(s
);
5172 * Slabcache can never be merged so we can use the name proper.
5173 * This is typically the case for debug situations. In that
5174 * case we can catch duplicate names easily.
5176 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5180 * Create a unique name for the slab as a target
5183 name
= create_unique_id(s
);
5186 s
->kobj
.kset
= slab_kset
;
5187 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5189 kobject_put(&s
->kobj
);
5193 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5195 kobject_del(&s
->kobj
);
5196 kobject_put(&s
->kobj
);
5199 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5201 /* Setup first alias */
5202 sysfs_slab_alias(s
, s
->name
);
5208 static void sysfs_slab_remove(struct kmem_cache
*s
)
5210 if (slab_state
< FULL
)
5212 * Sysfs has not been setup yet so no need to remove the
5217 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5218 kobject_del(&s
->kobj
);
5219 kobject_put(&s
->kobj
);
5223 * Need to buffer aliases during bootup until sysfs becomes
5224 * available lest we lose that information.
5226 struct saved_alias
{
5227 struct kmem_cache
*s
;
5229 struct saved_alias
*next
;
5232 static struct saved_alias
*alias_list
;
5234 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5236 struct saved_alias
*al
;
5238 if (slab_state
== FULL
) {
5240 * If we have a leftover link then remove it.
5242 sysfs_remove_link(&slab_kset
->kobj
, name
);
5243 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5246 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5252 al
->next
= alias_list
;
5257 static int __init
slab_sysfs_init(void)
5259 struct kmem_cache
*s
;
5262 mutex_lock(&slab_mutex
);
5264 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5266 mutex_unlock(&slab_mutex
);
5267 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5273 list_for_each_entry(s
, &slab_caches
, list
) {
5274 err
= sysfs_slab_add(s
);
5276 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5277 " to sysfs\n", s
->name
);
5280 while (alias_list
) {
5281 struct saved_alias
*al
= alias_list
;
5283 alias_list
= alias_list
->next
;
5284 err
= sysfs_slab_alias(al
->s
, al
->name
);
5286 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5287 " %s to sysfs\n", al
->name
);
5291 mutex_unlock(&slab_mutex
);
5296 __initcall(slab_sysfs_init
);
5297 #endif /* CONFIG_SYSFS */
5300 * The /proc/slabinfo ABI
5302 #ifdef CONFIG_SLABINFO
5303 void get_slabinfo(struct kmem_cache
*s
, struct slabinfo
*sinfo
)
5305 unsigned long nr_slabs
= 0;
5306 unsigned long nr_objs
= 0;
5307 unsigned long nr_free
= 0;
5310 for_each_online_node(node
) {
5311 struct kmem_cache_node
*n
= get_node(s
, node
);
5316 nr_slabs
+= node_nr_slabs(n
);
5317 nr_objs
+= node_nr_objs(n
);
5318 nr_free
+= count_partial(n
, count_free
);
5321 sinfo
->active_objs
= nr_objs
- nr_free
;
5322 sinfo
->num_objs
= nr_objs
;
5323 sinfo
->active_slabs
= nr_slabs
;
5324 sinfo
->num_slabs
= nr_slabs
;
5325 sinfo
->objects_per_slab
= oo_objects(s
->oo
);
5326 sinfo
->cache_order
= oo_order(s
->oo
);
5329 void slabinfo_show_stats(struct seq_file
*m
, struct kmem_cache
*s
)
5333 ssize_t
slabinfo_write(struct file
*file
, const char __user
*buffer
,
5334 size_t count
, loff_t
*ppos
)
5338 #endif /* CONFIG_SLABINFO */