2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
20 #include <linux/proc_fs.h>
21 #include <linux/notifier.h>
22 #include <linux/seq_file.h>
23 #include <linux/kmemcheck.h>
24 #include <linux/cpu.h>
25 #include <linux/cpuset.h>
26 #include <linux/mempolicy.h>
27 #include <linux/ctype.h>
28 #include <linux/debugobjects.h>
29 #include <linux/kallsyms.h>
30 #include <linux/memory.h>
31 #include <linux/math64.h>
32 #include <linux/fault-inject.h>
33 #include <linux/stacktrace.h>
34 #include <linux/prefetch.h>
35 #include <linux/memcontrol.h>
37 #include <trace/events/kmem.h>
43 * 1. slab_mutex (Global Mutex)
45 * 3. slab_lock(page) (Only on some arches and for debugging)
49 * The role of the slab_mutex is to protect the list of all the slabs
50 * and to synchronize major metadata changes to slab cache structures.
52 * The slab_lock is only used for debugging and on arches that do not
53 * have the ability to do a cmpxchg_double. It only protects the second
54 * double word in the page struct. Meaning
55 * A. page->freelist -> List of object free in a page
56 * B. page->counters -> Counters of objects
57 * C. page->frozen -> frozen state
59 * If a slab is frozen then it is exempt from list management. It is not
60 * on any list. The processor that froze the slab is the one who can
61 * perform list operations on the page. Other processors may put objects
62 * onto the freelist but the processor that froze the slab is the only
63 * one that can retrieve the objects from the page's freelist.
65 * The list_lock protects the partial and full list on each node and
66 * the partial slab counter. If taken then no new slabs may be added or
67 * removed from the lists nor make the number of partial slabs be modified.
68 * (Note that the total number of slabs is an atomic value that may be
69 * modified without taking the list lock).
71 * The list_lock is a centralized lock and thus we avoid taking it as
72 * much as possible. As long as SLUB does not have to handle partial
73 * slabs, operations can continue without any centralized lock. F.e.
74 * allocating a long series of objects that fill up slabs does not require
76 * Interrupts are disabled during allocation and deallocation in order to
77 * make the slab allocator safe to use in the context of an irq. In addition
78 * interrupts are disabled to ensure that the processor does not change
79 * while handling per_cpu slabs, due to kernel preemption.
81 * SLUB assigns one slab for allocation to each processor.
82 * Allocations only occur from these slabs called cpu slabs.
84 * Slabs with free elements are kept on a partial list and during regular
85 * operations no list for full slabs is used. If an object in a full slab is
86 * freed then the slab will show up again on the partial lists.
87 * We track full slabs for debugging purposes though because otherwise we
88 * cannot scan all objects.
90 * Slabs are freed when they become empty. Teardown and setup is
91 * minimal so we rely on the page allocators per cpu caches for
92 * fast frees and allocs.
94 * Overloading of page flags that are otherwise used for LRU management.
96 * PageActive The slab is frozen and exempt from list processing.
97 * This means that the slab is dedicated to a purpose
98 * such as satisfying allocations for a specific
99 * processor. Objects may be freed in the slab while
100 * it is frozen but slab_free will then skip the usual
101 * list operations. It is up to the processor holding
102 * the slab to integrate the slab into the slab lists
103 * when the slab is no longer needed.
105 * One use of this flag is to mark slabs that are
106 * used for allocations. Then such a slab becomes a cpu
107 * slab. The cpu slab may be equipped with an additional
108 * freelist that allows lockless access to
109 * free objects in addition to the regular freelist
110 * that requires the slab lock.
112 * PageError Slab requires special handling due to debug
113 * options set. This moves slab handling out of
114 * the fast path and disables lockless freelists.
117 static inline int kmem_cache_debug(struct kmem_cache
*s
)
119 #ifdef CONFIG_SLUB_DEBUG
120 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
126 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache
*s
)
128 #ifdef CONFIG_SLUB_CPU_PARTIAL
129 return !kmem_cache_debug(s
);
136 * Issues still to be resolved:
138 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
140 * - Variable sizing of the per node arrays
143 /* Enable to test recovery from slab corruption on boot */
144 #undef SLUB_RESILIENCY_TEST
146 /* Enable to log cmpxchg failures */
147 #undef SLUB_DEBUG_CMPXCHG
150 * Mininum number of partial slabs. These will be left on the partial
151 * lists even if they are empty. kmem_cache_shrink may reclaim them.
153 #define MIN_PARTIAL 5
156 * Maximum number of desirable partial slabs.
157 * The existence of more partial slabs makes kmem_cache_shrink
158 * sort the partial list by the number of objects in use.
160 #define MAX_PARTIAL 10
162 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
163 SLAB_POISON | SLAB_STORE_USER)
166 * Debugging flags that require metadata to be stored in the slab. These get
167 * disabled when slub_debug=O is used and a cache's min order increases with
170 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
173 * Set of flags that will prevent slab merging
175 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
176 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
179 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
180 SLAB_CACHE_DMA | SLAB_NOTRACK)
183 #define OO_MASK ((1 << OO_SHIFT) - 1)
184 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
186 /* Internal SLUB flags */
187 #define __OBJECT_POISON 0x80000000UL /* Poison object */
188 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
191 static struct notifier_block slab_notifier
;
195 * Tracking user of a slab.
197 #define TRACK_ADDRS_COUNT 16
199 unsigned long addr
; /* Called from address */
200 #ifdef CONFIG_STACKTRACE
201 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
203 int cpu
; /* Was running on cpu */
204 int pid
; /* Pid context */
205 unsigned long when
; /* When did the operation occur */
208 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
211 static int sysfs_slab_add(struct kmem_cache
*);
212 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
213 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
);
215 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
216 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
218 static inline void memcg_propagate_slab_attrs(struct kmem_cache
*s
) { }
221 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
223 #ifdef CONFIG_SLUB_STATS
225 * The rmw is racy on a preemptible kernel but this is acceptable, so
226 * avoid this_cpu_add()'s irq-disable overhead.
228 raw_cpu_inc(s
->cpu_slab
->stat
[si
]);
232 /********************************************************************
233 * Core slab cache functions
234 *******************************************************************/
236 /* Verify that a pointer has an address that is valid within a slab page */
237 static inline int check_valid_pointer(struct kmem_cache
*s
,
238 struct page
*page
, const void *object
)
245 base
= page_address(page
);
246 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
247 (object
- base
) % s
->size
) {
254 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
256 return *(void **)(object
+ s
->offset
);
259 static void prefetch_freepointer(const struct kmem_cache
*s
, void *object
)
261 prefetch(object
+ s
->offset
);
264 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
268 #ifdef CONFIG_DEBUG_PAGEALLOC
269 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
271 p
= get_freepointer(s
, object
);
276 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
278 *(void **)(object
+ s
->offset
) = fp
;
281 /* Loop over all objects in a slab */
282 #define for_each_object(__p, __s, __addr, __objects) \
283 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
286 /* Determine object index from a given position */
287 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
289 return (p
- addr
) / s
->size
;
292 static inline size_t slab_ksize(const struct kmem_cache
*s
)
294 #ifdef CONFIG_SLUB_DEBUG
296 * Debugging requires use of the padding between object
297 * and whatever may come after it.
299 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
300 return s
->object_size
;
304 * If we have the need to store the freelist pointer
305 * back there or track user information then we can
306 * only use the space before that information.
308 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
311 * Else we can use all the padding etc for the allocation
316 static inline int order_objects(int order
, unsigned long size
, int reserved
)
318 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
321 static inline struct kmem_cache_order_objects
oo_make(int order
,
322 unsigned long size
, int reserved
)
324 struct kmem_cache_order_objects x
= {
325 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
331 static inline int oo_order(struct kmem_cache_order_objects x
)
333 return x
.x
>> OO_SHIFT
;
336 static inline int oo_objects(struct kmem_cache_order_objects x
)
338 return x
.x
& OO_MASK
;
342 * Per slab locking using the pagelock
344 static __always_inline
void slab_lock(struct page
*page
)
346 bit_spin_lock(PG_locked
, &page
->flags
);
349 static __always_inline
void slab_unlock(struct page
*page
)
351 __bit_spin_unlock(PG_locked
, &page
->flags
);
354 static inline void set_page_slub_counters(struct page
*page
, unsigned long counters_new
)
357 tmp
.counters
= counters_new
;
359 * page->counters can cover frozen/inuse/objects as well
360 * as page->_count. If we assign to ->counters directly
361 * we run the risk of losing updates to page->_count, so
362 * be careful and only assign to the fields we need.
364 page
->frozen
= tmp
.frozen
;
365 page
->inuse
= tmp
.inuse
;
366 page
->objects
= tmp
.objects
;
369 /* Interrupts must be disabled (for the fallback code to work right) */
370 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
371 void *freelist_old
, unsigned long counters_old
,
372 void *freelist_new
, unsigned long counters_new
,
375 VM_BUG_ON(!irqs_disabled());
376 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
377 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
378 if (s
->flags
& __CMPXCHG_DOUBLE
) {
379 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
380 freelist_old
, counters_old
,
381 freelist_new
, counters_new
))
387 if (page
->freelist
== freelist_old
&&
388 page
->counters
== counters_old
) {
389 page
->freelist
= freelist_new
;
390 set_page_slub_counters(page
, counters_new
);
398 stat(s
, CMPXCHG_DOUBLE_FAIL
);
400 #ifdef SLUB_DEBUG_CMPXCHG
401 pr_info("%s %s: cmpxchg double redo ", n
, s
->name
);
407 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
408 void *freelist_old
, unsigned long counters_old
,
409 void *freelist_new
, unsigned long counters_new
,
412 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
413 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
414 if (s
->flags
& __CMPXCHG_DOUBLE
) {
415 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
416 freelist_old
, counters_old
,
417 freelist_new
, counters_new
))
424 local_irq_save(flags
);
426 if (page
->freelist
== freelist_old
&&
427 page
->counters
== counters_old
) {
428 page
->freelist
= freelist_new
;
429 set_page_slub_counters(page
, counters_new
);
431 local_irq_restore(flags
);
435 local_irq_restore(flags
);
439 stat(s
, CMPXCHG_DOUBLE_FAIL
);
441 #ifdef SLUB_DEBUG_CMPXCHG
442 pr_info("%s %s: cmpxchg double redo ", n
, s
->name
);
448 #ifdef CONFIG_SLUB_DEBUG
450 * Determine a map of object in use on a page.
452 * Node listlock must be held to guarantee that the page does
453 * not vanish from under us.
455 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
458 void *addr
= page_address(page
);
460 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
461 set_bit(slab_index(p
, s
, addr
), map
);
467 #ifdef CONFIG_SLUB_DEBUG_ON
468 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
470 static int slub_debug
;
473 static char *slub_debug_slabs
;
474 static int disable_higher_order_debug
;
479 static void print_section(char *text
, u8
*addr
, unsigned int length
)
481 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
485 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
486 enum track_item alloc
)
491 p
= object
+ s
->offset
+ sizeof(void *);
493 p
= object
+ s
->inuse
;
498 static void set_track(struct kmem_cache
*s
, void *object
,
499 enum track_item alloc
, unsigned long addr
)
501 struct track
*p
= get_track(s
, object
, alloc
);
504 #ifdef CONFIG_STACKTRACE
505 struct stack_trace trace
;
508 trace
.nr_entries
= 0;
509 trace
.max_entries
= TRACK_ADDRS_COUNT
;
510 trace
.entries
= p
->addrs
;
512 save_stack_trace(&trace
);
514 /* See rant in lockdep.c */
515 if (trace
.nr_entries
!= 0 &&
516 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
519 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
523 p
->cpu
= smp_processor_id();
524 p
->pid
= current
->pid
;
527 memset(p
, 0, sizeof(struct track
));
530 static void init_tracking(struct kmem_cache
*s
, void *object
)
532 if (!(s
->flags
& SLAB_STORE_USER
))
535 set_track(s
, object
, TRACK_FREE
, 0UL);
536 set_track(s
, object
, TRACK_ALLOC
, 0UL);
539 static void print_track(const char *s
, struct track
*t
)
544 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
545 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
546 #ifdef CONFIG_STACKTRACE
549 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
551 pr_err("\t%pS\n", (void *)t
->addrs
[i
]);
558 static void print_tracking(struct kmem_cache
*s
, void *object
)
560 if (!(s
->flags
& SLAB_STORE_USER
))
563 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
564 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
567 static void print_page_info(struct page
*page
)
569 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
570 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
574 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
576 struct va_format vaf
;
582 pr_err("=============================================================================\n");
583 pr_err("BUG %s (%s): %pV\n", s
->name
, print_tainted(), &vaf
);
584 pr_err("-----------------------------------------------------------------------------\n\n");
586 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
590 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
592 struct va_format vaf
;
598 pr_err("FIX %s: %pV\n", s
->name
, &vaf
);
602 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
604 unsigned int off
; /* Offset of last byte */
605 u8
*addr
= page_address(page
);
607 print_tracking(s
, p
);
609 print_page_info(page
);
611 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
612 p
, p
- addr
, get_freepointer(s
, p
));
615 print_section("Bytes b4 ", p
- 16, 16);
617 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
619 if (s
->flags
& SLAB_RED_ZONE
)
620 print_section("Redzone ", p
+ s
->object_size
,
621 s
->inuse
- s
->object_size
);
624 off
= s
->offset
+ sizeof(void *);
628 if (s
->flags
& SLAB_STORE_USER
)
629 off
+= 2 * sizeof(struct track
);
632 /* Beginning of the filler is the free pointer */
633 print_section("Padding ", p
+ off
, s
->size
- off
);
638 static void object_err(struct kmem_cache
*s
, struct page
*page
,
639 u8
*object
, char *reason
)
641 slab_bug(s
, "%s", reason
);
642 print_trailer(s
, page
, object
);
645 static void slab_err(struct kmem_cache
*s
, struct page
*page
,
646 const char *fmt
, ...)
652 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
654 slab_bug(s
, "%s", buf
);
655 print_page_info(page
);
659 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
663 if (s
->flags
& __OBJECT_POISON
) {
664 memset(p
, POISON_FREE
, s
->object_size
- 1);
665 p
[s
->object_size
- 1] = POISON_END
;
668 if (s
->flags
& SLAB_RED_ZONE
)
669 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
672 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
673 void *from
, void *to
)
675 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
676 memset(from
, data
, to
- from
);
679 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
680 u8
*object
, char *what
,
681 u8
*start
, unsigned int value
, unsigned int bytes
)
686 fault
= memchr_inv(start
, value
, bytes
);
691 while (end
> fault
&& end
[-1] == value
)
694 slab_bug(s
, "%s overwritten", what
);
695 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
696 fault
, end
- 1, fault
[0], value
);
697 print_trailer(s
, page
, object
);
699 restore_bytes(s
, what
, value
, fault
, end
);
707 * Bytes of the object to be managed.
708 * If the freepointer may overlay the object then the free
709 * pointer is the first word of the object.
711 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
714 * object + s->object_size
715 * Padding to reach word boundary. This is also used for Redzoning.
716 * Padding is extended by another word if Redzoning is enabled and
717 * object_size == inuse.
719 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
720 * 0xcc (RED_ACTIVE) for objects in use.
723 * Meta data starts here.
725 * A. Free pointer (if we cannot overwrite object on free)
726 * B. Tracking data for SLAB_STORE_USER
727 * C. Padding to reach required alignment boundary or at mininum
728 * one word if debugging is on to be able to detect writes
729 * before the word boundary.
731 * Padding is done using 0x5a (POISON_INUSE)
734 * Nothing is used beyond s->size.
736 * If slabcaches are merged then the object_size and inuse boundaries are mostly
737 * ignored. And therefore no slab options that rely on these boundaries
738 * may be used with merged slabcaches.
741 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
743 unsigned long off
= s
->inuse
; /* The end of info */
746 /* Freepointer is placed after the object. */
747 off
+= sizeof(void *);
749 if (s
->flags
& SLAB_STORE_USER
)
750 /* We also have user information there */
751 off
+= 2 * sizeof(struct track
);
756 return check_bytes_and_report(s
, page
, p
, "Object padding",
757 p
+ off
, POISON_INUSE
, s
->size
- off
);
760 /* Check the pad bytes at the end of a slab page */
761 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
769 if (!(s
->flags
& SLAB_POISON
))
772 start
= page_address(page
);
773 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
774 end
= start
+ length
;
775 remainder
= length
% s
->size
;
779 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
782 while (end
> fault
&& end
[-1] == POISON_INUSE
)
785 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
786 print_section("Padding ", end
- remainder
, remainder
);
788 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
792 static int check_object(struct kmem_cache
*s
, struct page
*page
,
793 void *object
, u8 val
)
796 u8
*endobject
= object
+ s
->object_size
;
798 if (s
->flags
& SLAB_RED_ZONE
) {
799 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
800 endobject
, val
, s
->inuse
- s
->object_size
))
803 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
804 check_bytes_and_report(s
, page
, p
, "Alignment padding",
805 endobject
, POISON_INUSE
,
806 s
->inuse
- s
->object_size
);
810 if (s
->flags
& SLAB_POISON
) {
811 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
812 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
813 POISON_FREE
, s
->object_size
- 1) ||
814 !check_bytes_and_report(s
, page
, p
, "Poison",
815 p
+ s
->object_size
- 1, POISON_END
, 1)))
818 * check_pad_bytes cleans up on its own.
820 check_pad_bytes(s
, page
, p
);
823 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
825 * Object and freepointer overlap. Cannot check
826 * freepointer while object is allocated.
830 /* Check free pointer validity */
831 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
832 object_err(s
, page
, p
, "Freepointer corrupt");
834 * No choice but to zap it and thus lose the remainder
835 * of the free objects in this slab. May cause
836 * another error because the object count is now wrong.
838 set_freepointer(s
, p
, NULL
);
844 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
848 VM_BUG_ON(!irqs_disabled());
850 if (!PageSlab(page
)) {
851 slab_err(s
, page
, "Not a valid slab page");
855 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
856 if (page
->objects
> maxobj
) {
857 slab_err(s
, page
, "objects %u > max %u",
858 s
->name
, page
->objects
, maxobj
);
861 if (page
->inuse
> page
->objects
) {
862 slab_err(s
, page
, "inuse %u > max %u",
863 s
->name
, page
->inuse
, page
->objects
);
866 /* Slab_pad_check fixes things up after itself */
867 slab_pad_check(s
, page
);
872 * Determine if a certain object on a page is on the freelist. Must hold the
873 * slab lock to guarantee that the chains are in a consistent state.
875 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
880 unsigned long max_objects
;
883 while (fp
&& nr
<= page
->objects
) {
886 if (!check_valid_pointer(s
, page
, fp
)) {
888 object_err(s
, page
, object
,
889 "Freechain corrupt");
890 set_freepointer(s
, object
, NULL
);
892 slab_err(s
, page
, "Freepointer corrupt");
893 page
->freelist
= NULL
;
894 page
->inuse
= page
->objects
;
895 slab_fix(s
, "Freelist cleared");
901 fp
= get_freepointer(s
, object
);
905 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
906 if (max_objects
> MAX_OBJS_PER_PAGE
)
907 max_objects
= MAX_OBJS_PER_PAGE
;
909 if (page
->objects
!= max_objects
) {
910 slab_err(s
, page
, "Wrong number of objects. Found %d but "
911 "should be %d", page
->objects
, max_objects
);
912 page
->objects
= max_objects
;
913 slab_fix(s
, "Number of objects adjusted.");
915 if (page
->inuse
!= page
->objects
- nr
) {
916 slab_err(s
, page
, "Wrong object count. Counter is %d but "
917 "counted were %d", page
->inuse
, page
->objects
- nr
);
918 page
->inuse
= page
->objects
- nr
;
919 slab_fix(s
, "Object count adjusted.");
921 return search
== NULL
;
924 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
927 if (s
->flags
& SLAB_TRACE
) {
928 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
930 alloc
? "alloc" : "free",
935 print_section("Object ", (void *)object
,
943 * Hooks for other subsystems that check memory allocations. In a typical
944 * production configuration these hooks all should produce no code at all.
946 static inline void kmalloc_large_node_hook(void *ptr
, size_t size
, gfp_t flags
)
948 kmemleak_alloc(ptr
, size
, 1, flags
);
951 static inline void kfree_hook(const void *x
)
956 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
958 flags
&= gfp_allowed_mask
;
959 lockdep_trace_alloc(flags
);
960 might_sleep_if(flags
& __GFP_WAIT
);
962 return should_failslab(s
->object_size
, flags
, s
->flags
);
965 static inline void slab_post_alloc_hook(struct kmem_cache
*s
,
966 gfp_t flags
, void *object
)
968 flags
&= gfp_allowed_mask
;
969 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
970 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
973 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
975 kmemleak_free_recursive(x
, s
->flags
);
978 * Trouble is that we may no longer disable interrupts in the fast path
979 * So in order to make the debug calls that expect irqs to be
980 * disabled we need to disable interrupts temporarily.
982 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
986 local_irq_save(flags
);
987 kmemcheck_slab_free(s
, x
, s
->object_size
);
988 debug_check_no_locks_freed(x
, s
->object_size
);
989 local_irq_restore(flags
);
992 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
993 debug_check_no_obj_freed(x
, s
->object_size
);
997 * Tracking of fully allocated slabs for debugging purposes.
999 static void add_full(struct kmem_cache
*s
,
1000 struct kmem_cache_node
*n
, struct page
*page
)
1002 if (!(s
->flags
& SLAB_STORE_USER
))
1005 lockdep_assert_held(&n
->list_lock
);
1006 list_add(&page
->lru
, &n
->full
);
1009 static void remove_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
, struct page
*page
)
1011 if (!(s
->flags
& SLAB_STORE_USER
))
1014 lockdep_assert_held(&n
->list_lock
);
1015 list_del(&page
->lru
);
1018 /* Tracking of the number of slabs for debugging purposes */
1019 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1021 struct kmem_cache_node
*n
= get_node(s
, node
);
1023 return atomic_long_read(&n
->nr_slabs
);
1026 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1028 return atomic_long_read(&n
->nr_slabs
);
1031 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1033 struct kmem_cache_node
*n
= get_node(s
, node
);
1036 * May be called early in order to allocate a slab for the
1037 * kmem_cache_node structure. Solve the chicken-egg
1038 * dilemma by deferring the increment of the count during
1039 * bootstrap (see early_kmem_cache_node_alloc).
1042 atomic_long_inc(&n
->nr_slabs
);
1043 atomic_long_add(objects
, &n
->total_objects
);
1046 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1048 struct kmem_cache_node
*n
= get_node(s
, node
);
1050 atomic_long_dec(&n
->nr_slabs
);
1051 atomic_long_sub(objects
, &n
->total_objects
);
1054 /* Object debug checks for alloc/free paths */
1055 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1058 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1061 init_object(s
, object
, SLUB_RED_INACTIVE
);
1062 init_tracking(s
, object
);
1065 static noinline
int alloc_debug_processing(struct kmem_cache
*s
,
1067 void *object
, unsigned long addr
)
1069 if (!check_slab(s
, page
))
1072 if (!check_valid_pointer(s
, page
, object
)) {
1073 object_err(s
, page
, object
, "Freelist Pointer check fails");
1077 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1080 /* Success perform special debug activities for allocs */
1081 if (s
->flags
& SLAB_STORE_USER
)
1082 set_track(s
, object
, TRACK_ALLOC
, addr
);
1083 trace(s
, page
, object
, 1);
1084 init_object(s
, object
, SLUB_RED_ACTIVE
);
1088 if (PageSlab(page
)) {
1090 * If this is a slab page then lets do the best we can
1091 * to avoid issues in the future. Marking all objects
1092 * as used avoids touching the remaining objects.
1094 slab_fix(s
, "Marking all objects used");
1095 page
->inuse
= page
->objects
;
1096 page
->freelist
= NULL
;
1101 static noinline
struct kmem_cache_node
*free_debug_processing(
1102 struct kmem_cache
*s
, struct page
*page
, void *object
,
1103 unsigned long addr
, unsigned long *flags
)
1105 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1107 spin_lock_irqsave(&n
->list_lock
, *flags
);
1110 if (!check_slab(s
, page
))
1113 if (!check_valid_pointer(s
, page
, object
)) {
1114 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1118 if (on_freelist(s
, page
, object
)) {
1119 object_err(s
, page
, object
, "Object already free");
1123 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1126 if (unlikely(s
!= page
->slab_cache
)) {
1127 if (!PageSlab(page
)) {
1128 slab_err(s
, page
, "Attempt to free object(0x%p) "
1129 "outside of slab", object
);
1130 } else if (!page
->slab_cache
) {
1131 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1135 object_err(s
, page
, object
,
1136 "page slab pointer corrupt.");
1140 if (s
->flags
& SLAB_STORE_USER
)
1141 set_track(s
, object
, TRACK_FREE
, addr
);
1142 trace(s
, page
, object
, 0);
1143 init_object(s
, object
, SLUB_RED_INACTIVE
);
1147 * Keep node_lock to preserve integrity
1148 * until the object is actually freed
1154 spin_unlock_irqrestore(&n
->list_lock
, *flags
);
1155 slab_fix(s
, "Object at 0x%p not freed", object
);
1159 static int __init
setup_slub_debug(char *str
)
1161 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1162 if (*str
++ != '=' || !*str
)
1164 * No options specified. Switch on full debugging.
1170 * No options but restriction on slabs. This means full
1171 * debugging for slabs matching a pattern.
1175 if (tolower(*str
) == 'o') {
1177 * Avoid enabling debugging on caches if its minimum order
1178 * would increase as a result.
1180 disable_higher_order_debug
= 1;
1187 * Switch off all debugging measures.
1192 * Determine which debug features should be switched on
1194 for (; *str
&& *str
!= ','; str
++) {
1195 switch (tolower(*str
)) {
1197 slub_debug
|= SLAB_DEBUG_FREE
;
1200 slub_debug
|= SLAB_RED_ZONE
;
1203 slub_debug
|= SLAB_POISON
;
1206 slub_debug
|= SLAB_STORE_USER
;
1209 slub_debug
|= SLAB_TRACE
;
1212 slub_debug
|= SLAB_FAILSLAB
;
1215 pr_err("slub_debug option '%c' unknown. skipped\n",
1222 slub_debug_slabs
= str
+ 1;
1227 __setup("slub_debug", setup_slub_debug
);
1229 static unsigned long kmem_cache_flags(unsigned long object_size
,
1230 unsigned long flags
, const char *name
,
1231 void (*ctor
)(void *))
1234 * Enable debugging if selected on the kernel commandline.
1236 if (slub_debug
&& (!slub_debug_slabs
|| (name
&&
1237 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)))))
1238 flags
|= slub_debug
;
1243 static inline void setup_object_debug(struct kmem_cache
*s
,
1244 struct page
*page
, void *object
) {}
1246 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1247 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1249 static inline struct kmem_cache_node
*free_debug_processing(
1250 struct kmem_cache
*s
, struct page
*page
, void *object
,
1251 unsigned long addr
, unsigned long *flags
) { return NULL
; }
1253 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1255 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1256 void *object
, u8 val
) { return 1; }
1257 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1258 struct page
*page
) {}
1259 static inline void remove_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1260 struct page
*page
) {}
1261 static inline unsigned long kmem_cache_flags(unsigned long object_size
,
1262 unsigned long flags
, const char *name
,
1263 void (*ctor
)(void *))
1267 #define slub_debug 0
1269 #define disable_higher_order_debug 0
1271 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1273 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1275 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1277 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1280 static inline void kmalloc_large_node_hook(void *ptr
, size_t size
, gfp_t flags
)
1282 kmemleak_alloc(ptr
, size
, 1, flags
);
1285 static inline void kfree_hook(const void *x
)
1290 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1293 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1296 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
,
1297 flags
& gfp_allowed_mask
);
1300 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
1302 kmemleak_free_recursive(x
, s
->flags
);
1305 #endif /* CONFIG_SLUB_DEBUG */
1308 * Slab allocation and freeing
1310 static inline struct page
*alloc_slab_page(struct kmem_cache
*s
,
1311 gfp_t flags
, int node
, struct kmem_cache_order_objects oo
)
1314 int order
= oo_order(oo
);
1316 flags
|= __GFP_NOTRACK
;
1318 if (memcg_charge_slab(s
, flags
, order
))
1321 if (node
== NUMA_NO_NODE
)
1322 page
= alloc_pages(flags
, order
);
1324 page
= alloc_pages_exact_node(node
, flags
, order
);
1327 memcg_uncharge_slab(s
, order
);
1332 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1335 struct kmem_cache_order_objects oo
= s
->oo
;
1338 flags
&= gfp_allowed_mask
;
1340 if (flags
& __GFP_WAIT
)
1343 flags
|= s
->allocflags
;
1346 * Let the initial higher-order allocation fail under memory pressure
1347 * so we fall-back to the minimum order allocation.
1349 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1351 page
= alloc_slab_page(s
, alloc_gfp
, node
, oo
);
1352 if (unlikely(!page
)) {
1356 * Allocation may have failed due to fragmentation.
1357 * Try a lower order alloc if possible
1359 page
= alloc_slab_page(s
, alloc_gfp
, node
, oo
);
1362 stat(s
, ORDER_FALLBACK
);
1365 if (kmemcheck_enabled
&& page
1366 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1367 int pages
= 1 << oo_order(oo
);
1369 kmemcheck_alloc_shadow(page
, oo_order(oo
), alloc_gfp
, node
);
1372 * Objects from caches that have a constructor don't get
1373 * cleared when they're allocated, so we need to do it here.
1376 kmemcheck_mark_uninitialized_pages(page
, pages
);
1378 kmemcheck_mark_unallocated_pages(page
, pages
);
1381 if (flags
& __GFP_WAIT
)
1382 local_irq_disable();
1386 page
->objects
= oo_objects(oo
);
1387 mod_zone_page_state(page_zone(page
),
1388 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1389 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1395 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1398 setup_object_debug(s
, page
, object
);
1399 if (unlikely(s
->ctor
))
1403 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1411 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1413 page
= allocate_slab(s
,
1414 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1418 order
= compound_order(page
);
1419 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1420 page
->slab_cache
= s
;
1421 __SetPageSlab(page
);
1422 if (page
->pfmemalloc
)
1423 SetPageSlabPfmemalloc(page
);
1425 start
= page_address(page
);
1427 if (unlikely(s
->flags
& SLAB_POISON
))
1428 memset(start
, POISON_INUSE
, PAGE_SIZE
<< order
);
1431 for_each_object(p
, s
, start
, page
->objects
) {
1432 setup_object(s
, page
, last
);
1433 set_freepointer(s
, last
, p
);
1436 setup_object(s
, page
, last
);
1437 set_freepointer(s
, last
, NULL
);
1439 page
->freelist
= start
;
1440 page
->inuse
= page
->objects
;
1446 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1448 int order
= compound_order(page
);
1449 int pages
= 1 << order
;
1451 if (kmem_cache_debug(s
)) {
1454 slab_pad_check(s
, page
);
1455 for_each_object(p
, s
, page_address(page
),
1457 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1460 kmemcheck_free_shadow(page
, compound_order(page
));
1462 mod_zone_page_state(page_zone(page
),
1463 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1464 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1467 __ClearPageSlabPfmemalloc(page
);
1468 __ClearPageSlab(page
);
1470 page_mapcount_reset(page
);
1471 if (current
->reclaim_state
)
1472 current
->reclaim_state
->reclaimed_slab
+= pages
;
1473 __free_pages(page
, order
);
1474 memcg_uncharge_slab(s
, order
);
1477 #define need_reserve_slab_rcu \
1478 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1480 static void rcu_free_slab(struct rcu_head
*h
)
1484 if (need_reserve_slab_rcu
)
1485 page
= virt_to_head_page(h
);
1487 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1489 __free_slab(page
->slab_cache
, page
);
1492 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1494 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1495 struct rcu_head
*head
;
1497 if (need_reserve_slab_rcu
) {
1498 int order
= compound_order(page
);
1499 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1501 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1502 head
= page_address(page
) + offset
;
1505 * RCU free overloads the RCU head over the LRU
1507 head
= (void *)&page
->lru
;
1510 call_rcu(head
, rcu_free_slab
);
1512 __free_slab(s
, page
);
1515 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1517 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1522 * Management of partially allocated slabs.
1525 __add_partial(struct kmem_cache_node
*n
, struct page
*page
, int tail
)
1528 if (tail
== DEACTIVATE_TO_TAIL
)
1529 list_add_tail(&page
->lru
, &n
->partial
);
1531 list_add(&page
->lru
, &n
->partial
);
1534 static inline void add_partial(struct kmem_cache_node
*n
,
1535 struct page
*page
, int tail
)
1537 lockdep_assert_held(&n
->list_lock
);
1538 __add_partial(n
, page
, tail
);
1542 __remove_partial(struct kmem_cache_node
*n
, struct page
*page
)
1544 list_del(&page
->lru
);
1548 static inline void remove_partial(struct kmem_cache_node
*n
,
1551 lockdep_assert_held(&n
->list_lock
);
1552 __remove_partial(n
, page
);
1556 * Remove slab from the partial list, freeze it and
1557 * return the pointer to the freelist.
1559 * Returns a list of objects or NULL if it fails.
1561 static inline void *acquire_slab(struct kmem_cache
*s
,
1562 struct kmem_cache_node
*n
, struct page
*page
,
1563 int mode
, int *objects
)
1566 unsigned long counters
;
1569 lockdep_assert_held(&n
->list_lock
);
1572 * Zap the freelist and set the frozen bit.
1573 * The old freelist is the list of objects for the
1574 * per cpu allocation list.
1576 freelist
= page
->freelist
;
1577 counters
= page
->counters
;
1578 new.counters
= counters
;
1579 *objects
= new.objects
- new.inuse
;
1581 new.inuse
= page
->objects
;
1582 new.freelist
= NULL
;
1584 new.freelist
= freelist
;
1587 VM_BUG_ON(new.frozen
);
1590 if (!__cmpxchg_double_slab(s
, page
,
1592 new.freelist
, new.counters
,
1596 remove_partial(n
, page
);
1601 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1602 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
);
1605 * Try to allocate a partial slab from a specific node.
1607 static void *get_partial_node(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1608 struct kmem_cache_cpu
*c
, gfp_t flags
)
1610 struct page
*page
, *page2
;
1611 void *object
= NULL
;
1616 * Racy check. If we mistakenly see no partial slabs then we
1617 * just allocate an empty slab. If we mistakenly try to get a
1618 * partial slab and there is none available then get_partials()
1621 if (!n
|| !n
->nr_partial
)
1624 spin_lock(&n
->list_lock
);
1625 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1628 if (!pfmemalloc_match(page
, flags
))
1631 t
= acquire_slab(s
, n
, page
, object
== NULL
, &objects
);
1635 available
+= objects
;
1638 stat(s
, ALLOC_FROM_PARTIAL
);
1641 put_cpu_partial(s
, page
, 0);
1642 stat(s
, CPU_PARTIAL_NODE
);
1644 if (!kmem_cache_has_cpu_partial(s
)
1645 || available
> s
->cpu_partial
/ 2)
1649 spin_unlock(&n
->list_lock
);
1654 * Get a page from somewhere. Search in increasing NUMA distances.
1656 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1657 struct kmem_cache_cpu
*c
)
1660 struct zonelist
*zonelist
;
1663 enum zone_type high_zoneidx
= gfp_zone(flags
);
1665 unsigned int cpuset_mems_cookie
;
1668 * The defrag ratio allows a configuration of the tradeoffs between
1669 * inter node defragmentation and node local allocations. A lower
1670 * defrag_ratio increases the tendency to do local allocations
1671 * instead of attempting to obtain partial slabs from other nodes.
1673 * If the defrag_ratio is set to 0 then kmalloc() always
1674 * returns node local objects. If the ratio is higher then kmalloc()
1675 * may return off node objects because partial slabs are obtained
1676 * from other nodes and filled up.
1678 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1679 * defrag_ratio = 1000) then every (well almost) allocation will
1680 * first attempt to defrag slab caches on other nodes. This means
1681 * scanning over all nodes to look for partial slabs which may be
1682 * expensive if we do it every time we are trying to find a slab
1683 * with available objects.
1685 if (!s
->remote_node_defrag_ratio
||
1686 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1690 cpuset_mems_cookie
= read_mems_allowed_begin();
1691 zonelist
= node_zonelist(mempolicy_slab_node(), flags
);
1692 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1693 struct kmem_cache_node
*n
;
1695 n
= get_node(s
, zone_to_nid(zone
));
1697 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1698 n
->nr_partial
> s
->min_partial
) {
1699 object
= get_partial_node(s
, n
, c
, flags
);
1702 * Don't check read_mems_allowed_retry()
1703 * here - if mems_allowed was updated in
1704 * parallel, that was a harmless race
1705 * between allocation and the cpuset
1712 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
1718 * Get a partial page, lock it and return it.
1720 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1721 struct kmem_cache_cpu
*c
)
1724 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_mem_id() : node
;
1726 object
= get_partial_node(s
, get_node(s
, searchnode
), c
, flags
);
1727 if (object
|| node
!= NUMA_NO_NODE
)
1730 return get_any_partial(s
, flags
, c
);
1733 #ifdef CONFIG_PREEMPT
1735 * Calculate the next globally unique transaction for disambiguiation
1736 * during cmpxchg. The transactions start with the cpu number and are then
1737 * incremented by CONFIG_NR_CPUS.
1739 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1742 * No preemption supported therefore also no need to check for
1748 static inline unsigned long next_tid(unsigned long tid
)
1750 return tid
+ TID_STEP
;
1753 static inline unsigned int tid_to_cpu(unsigned long tid
)
1755 return tid
% TID_STEP
;
1758 static inline unsigned long tid_to_event(unsigned long tid
)
1760 return tid
/ TID_STEP
;
1763 static inline unsigned int init_tid(int cpu
)
1768 static inline void note_cmpxchg_failure(const char *n
,
1769 const struct kmem_cache
*s
, unsigned long tid
)
1771 #ifdef SLUB_DEBUG_CMPXCHG
1772 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1774 pr_info("%s %s: cmpxchg redo ", n
, s
->name
);
1776 #ifdef CONFIG_PREEMPT
1777 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1778 pr_warn("due to cpu change %d -> %d\n",
1779 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1782 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1783 pr_warn("due to cpu running other code. Event %ld->%ld\n",
1784 tid_to_event(tid
), tid_to_event(actual_tid
));
1786 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
1787 actual_tid
, tid
, next_tid(tid
));
1789 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1792 static void init_kmem_cache_cpus(struct kmem_cache
*s
)
1796 for_each_possible_cpu(cpu
)
1797 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1801 * Remove the cpu slab
1803 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
,
1806 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1807 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1809 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1811 int tail
= DEACTIVATE_TO_HEAD
;
1815 if (page
->freelist
) {
1816 stat(s
, DEACTIVATE_REMOTE_FREES
);
1817 tail
= DEACTIVATE_TO_TAIL
;
1821 * Stage one: Free all available per cpu objects back
1822 * to the page freelist while it is still frozen. Leave the
1825 * There is no need to take the list->lock because the page
1828 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1830 unsigned long counters
;
1833 prior
= page
->freelist
;
1834 counters
= page
->counters
;
1835 set_freepointer(s
, freelist
, prior
);
1836 new.counters
= counters
;
1838 VM_BUG_ON(!new.frozen
);
1840 } while (!__cmpxchg_double_slab(s
, page
,
1842 freelist
, new.counters
,
1843 "drain percpu freelist"));
1845 freelist
= nextfree
;
1849 * Stage two: Ensure that the page is unfrozen while the
1850 * list presence reflects the actual number of objects
1853 * We setup the list membership and then perform a cmpxchg
1854 * with the count. If there is a mismatch then the page
1855 * is not unfrozen but the page is on the wrong list.
1857 * Then we restart the process which may have to remove
1858 * the page from the list that we just put it on again
1859 * because the number of objects in the slab may have
1864 old
.freelist
= page
->freelist
;
1865 old
.counters
= page
->counters
;
1866 VM_BUG_ON(!old
.frozen
);
1868 /* Determine target state of the slab */
1869 new.counters
= old
.counters
;
1872 set_freepointer(s
, freelist
, old
.freelist
);
1873 new.freelist
= freelist
;
1875 new.freelist
= old
.freelist
;
1879 if (!new.inuse
&& n
->nr_partial
>= s
->min_partial
)
1881 else if (new.freelist
) {
1886 * Taking the spinlock removes the possiblity
1887 * that acquire_slab() will see a slab page that
1890 spin_lock(&n
->list_lock
);
1894 if (kmem_cache_debug(s
) && !lock
) {
1897 * This also ensures that the scanning of full
1898 * slabs from diagnostic functions will not see
1901 spin_lock(&n
->list_lock
);
1909 remove_partial(n
, page
);
1911 else if (l
== M_FULL
)
1913 remove_full(s
, n
, page
);
1915 if (m
== M_PARTIAL
) {
1917 add_partial(n
, page
, tail
);
1920 } else if (m
== M_FULL
) {
1922 stat(s
, DEACTIVATE_FULL
);
1923 add_full(s
, n
, page
);
1929 if (!__cmpxchg_double_slab(s
, page
,
1930 old
.freelist
, old
.counters
,
1931 new.freelist
, new.counters
,
1936 spin_unlock(&n
->list_lock
);
1939 stat(s
, DEACTIVATE_EMPTY
);
1940 discard_slab(s
, page
);
1946 * Unfreeze all the cpu partial slabs.
1948 * This function must be called with interrupts disabled
1949 * for the cpu using c (or some other guarantee must be there
1950 * to guarantee no concurrent accesses).
1952 static void unfreeze_partials(struct kmem_cache
*s
,
1953 struct kmem_cache_cpu
*c
)
1955 #ifdef CONFIG_SLUB_CPU_PARTIAL
1956 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1957 struct page
*page
, *discard_page
= NULL
;
1959 while ((page
= c
->partial
)) {
1963 c
->partial
= page
->next
;
1965 n2
= get_node(s
, page_to_nid(page
));
1968 spin_unlock(&n
->list_lock
);
1971 spin_lock(&n
->list_lock
);
1976 old
.freelist
= page
->freelist
;
1977 old
.counters
= page
->counters
;
1978 VM_BUG_ON(!old
.frozen
);
1980 new.counters
= old
.counters
;
1981 new.freelist
= old
.freelist
;
1985 } while (!__cmpxchg_double_slab(s
, page
,
1986 old
.freelist
, old
.counters
,
1987 new.freelist
, new.counters
,
1988 "unfreezing slab"));
1990 if (unlikely(!new.inuse
&& n
->nr_partial
>= s
->min_partial
)) {
1991 page
->next
= discard_page
;
1992 discard_page
= page
;
1994 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1995 stat(s
, FREE_ADD_PARTIAL
);
2000 spin_unlock(&n
->list_lock
);
2002 while (discard_page
) {
2003 page
= discard_page
;
2004 discard_page
= discard_page
->next
;
2006 stat(s
, DEACTIVATE_EMPTY
);
2007 discard_slab(s
, page
);
2014 * Put a page that was just frozen (in __slab_free) into a partial page
2015 * slot if available. This is done without interrupts disabled and without
2016 * preemption disabled. The cmpxchg is racy and may put the partial page
2017 * onto a random cpus partial slot.
2019 * If we did not find a slot then simply move all the partials to the
2020 * per node partial list.
2022 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
2024 #ifdef CONFIG_SLUB_CPU_PARTIAL
2025 struct page
*oldpage
;
2032 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
2035 pobjects
= oldpage
->pobjects
;
2036 pages
= oldpage
->pages
;
2037 if (drain
&& pobjects
> s
->cpu_partial
) {
2038 unsigned long flags
;
2040 * partial array is full. Move the existing
2041 * set to the per node partial list.
2043 local_irq_save(flags
);
2044 unfreeze_partials(s
, this_cpu_ptr(s
->cpu_slab
));
2045 local_irq_restore(flags
);
2049 stat(s
, CPU_PARTIAL_DRAIN
);
2054 pobjects
+= page
->objects
- page
->inuse
;
2056 page
->pages
= pages
;
2057 page
->pobjects
= pobjects
;
2058 page
->next
= oldpage
;
2060 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
)
2065 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2067 stat(s
, CPUSLAB_FLUSH
);
2068 deactivate_slab(s
, c
->page
, c
->freelist
);
2070 c
->tid
= next_tid(c
->tid
);
2078 * Called from IPI handler with interrupts disabled.
2080 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2082 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2088 unfreeze_partials(s
, c
);
2092 static void flush_cpu_slab(void *d
)
2094 struct kmem_cache
*s
= d
;
2096 __flush_cpu_slab(s
, smp_processor_id());
2099 static bool has_cpu_slab(int cpu
, void *info
)
2101 struct kmem_cache
*s
= info
;
2102 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2104 return c
->page
|| c
->partial
;
2107 static void flush_all(struct kmem_cache
*s
)
2109 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2113 * Check if the objects in a per cpu structure fit numa
2114 * locality expectations.
2116 static inline int node_match(struct page
*page
, int node
)
2119 if (!page
|| (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
))
2125 #ifdef CONFIG_SLUB_DEBUG
2126 static int count_free(struct page
*page
)
2128 return page
->objects
- page
->inuse
;
2131 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2133 return atomic_long_read(&n
->total_objects
);
2135 #endif /* CONFIG_SLUB_DEBUG */
2137 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2138 static unsigned long count_partial(struct kmem_cache_node
*n
,
2139 int (*get_count
)(struct page
*))
2141 unsigned long flags
;
2142 unsigned long x
= 0;
2145 spin_lock_irqsave(&n
->list_lock
, flags
);
2146 list_for_each_entry(page
, &n
->partial
, lru
)
2147 x
+= get_count(page
);
2148 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2151 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2153 static noinline
void
2154 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2156 #ifdef CONFIG_SLUB_DEBUG
2157 static DEFINE_RATELIMIT_STATE(slub_oom_rs
, DEFAULT_RATELIMIT_INTERVAL
,
2158 DEFAULT_RATELIMIT_BURST
);
2160 struct kmem_cache_node
*n
;
2162 if ((gfpflags
& __GFP_NOWARN
) || !__ratelimit(&slub_oom_rs
))
2165 pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2167 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2168 s
->name
, s
->object_size
, s
->size
, oo_order(s
->oo
),
2171 if (oo_order(s
->min
) > get_order(s
->object_size
))
2172 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2175 for_each_kmem_cache_node(s
, node
, n
) {
2176 unsigned long nr_slabs
;
2177 unsigned long nr_objs
;
2178 unsigned long nr_free
;
2180 nr_free
= count_partial(n
, count_free
);
2181 nr_slabs
= node_nr_slabs(n
);
2182 nr_objs
= node_nr_objs(n
);
2184 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
2185 node
, nr_slabs
, nr_objs
, nr_free
);
2190 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2191 int node
, struct kmem_cache_cpu
**pc
)
2194 struct kmem_cache_cpu
*c
= *pc
;
2197 freelist
= get_partial(s
, flags
, node
, c
);
2202 page
= new_slab(s
, flags
, node
);
2204 c
= raw_cpu_ptr(s
->cpu_slab
);
2209 * No other reference to the page yet so we can
2210 * muck around with it freely without cmpxchg
2212 freelist
= page
->freelist
;
2213 page
->freelist
= NULL
;
2215 stat(s
, ALLOC_SLAB
);
2224 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
)
2226 if (unlikely(PageSlabPfmemalloc(page
)))
2227 return gfp_pfmemalloc_allowed(gfpflags
);
2233 * Check the page->freelist of a page and either transfer the freelist to the
2234 * per cpu freelist or deactivate the page.
2236 * The page is still frozen if the return value is not NULL.
2238 * If this function returns NULL then the page has been unfrozen.
2240 * This function must be called with interrupt disabled.
2242 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2245 unsigned long counters
;
2249 freelist
= page
->freelist
;
2250 counters
= page
->counters
;
2252 new.counters
= counters
;
2253 VM_BUG_ON(!new.frozen
);
2255 new.inuse
= page
->objects
;
2256 new.frozen
= freelist
!= NULL
;
2258 } while (!__cmpxchg_double_slab(s
, page
,
2267 * Slow path. The lockless freelist is empty or we need to perform
2270 * Processing is still very fast if new objects have been freed to the
2271 * regular freelist. In that case we simply take over the regular freelist
2272 * as the lockless freelist and zap the regular freelist.
2274 * If that is not working then we fall back to the partial lists. We take the
2275 * first element of the freelist as the object to allocate now and move the
2276 * rest of the freelist to the lockless freelist.
2278 * And if we were unable to get a new slab from the partial slab lists then
2279 * we need to allocate a new slab. This is the slowest path since it involves
2280 * a call to the page allocator and the setup of a new slab.
2282 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2283 unsigned long addr
, struct kmem_cache_cpu
*c
)
2287 unsigned long flags
;
2289 local_irq_save(flags
);
2290 #ifdef CONFIG_PREEMPT
2292 * We may have been preempted and rescheduled on a different
2293 * cpu before disabling interrupts. Need to reload cpu area
2296 c
= this_cpu_ptr(s
->cpu_slab
);
2304 if (unlikely(!node_match(page
, node
))) {
2305 stat(s
, ALLOC_NODE_MISMATCH
);
2306 deactivate_slab(s
, page
, c
->freelist
);
2313 * By rights, we should be searching for a slab page that was
2314 * PFMEMALLOC but right now, we are losing the pfmemalloc
2315 * information when the page leaves the per-cpu allocator
2317 if (unlikely(!pfmemalloc_match(page
, gfpflags
))) {
2318 deactivate_slab(s
, page
, c
->freelist
);
2324 /* must check again c->freelist in case of cpu migration or IRQ */
2325 freelist
= c
->freelist
;
2329 freelist
= get_freelist(s
, page
);
2333 stat(s
, DEACTIVATE_BYPASS
);
2337 stat(s
, ALLOC_REFILL
);
2341 * freelist is pointing to the list of objects to be used.
2342 * page is pointing to the page from which the objects are obtained.
2343 * That page must be frozen for per cpu allocations to work.
2345 VM_BUG_ON(!c
->page
->frozen
);
2346 c
->freelist
= get_freepointer(s
, freelist
);
2347 c
->tid
= next_tid(c
->tid
);
2348 local_irq_restore(flags
);
2354 page
= c
->page
= c
->partial
;
2355 c
->partial
= page
->next
;
2356 stat(s
, CPU_PARTIAL_ALLOC
);
2361 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2363 if (unlikely(!freelist
)) {
2364 slab_out_of_memory(s
, gfpflags
, node
);
2365 local_irq_restore(flags
);
2370 if (likely(!kmem_cache_debug(s
) && pfmemalloc_match(page
, gfpflags
)))
2373 /* Only entered in the debug case */
2374 if (kmem_cache_debug(s
) &&
2375 !alloc_debug_processing(s
, page
, freelist
, addr
))
2376 goto new_slab
; /* Slab failed checks. Next slab needed */
2378 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2381 local_irq_restore(flags
);
2386 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2387 * have the fastpath folded into their functions. So no function call
2388 * overhead for requests that can be satisfied on the fastpath.
2390 * The fastpath works by first checking if the lockless freelist can be used.
2391 * If not then __slab_alloc is called for slow processing.
2393 * Otherwise we can simply pick the next object from the lockless free list.
2395 static __always_inline
void *slab_alloc_node(struct kmem_cache
*s
,
2396 gfp_t gfpflags
, int node
, unsigned long addr
)
2399 struct kmem_cache_cpu
*c
;
2403 if (slab_pre_alloc_hook(s
, gfpflags
))
2406 s
= memcg_kmem_get_cache(s
, gfpflags
);
2409 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2410 * enabled. We may switch back and forth between cpus while
2411 * reading from one cpu area. That does not matter as long
2412 * as we end up on the original cpu again when doing the cmpxchg.
2414 * Preemption is disabled for the retrieval of the tid because that
2415 * must occur from the current processor. We cannot allow rescheduling
2416 * on a different processor between the determination of the pointer
2417 * and the retrieval of the tid.
2420 c
= this_cpu_ptr(s
->cpu_slab
);
2423 * The transaction ids are globally unique per cpu and per operation on
2424 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2425 * occurs on the right processor and that there was no operation on the
2426 * linked list in between.
2431 object
= c
->freelist
;
2433 if (unlikely(!object
|| !node_match(page
, node
))) {
2434 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2435 stat(s
, ALLOC_SLOWPATH
);
2437 void *next_object
= get_freepointer_safe(s
, object
);
2440 * The cmpxchg will only match if there was no additional
2441 * operation and if we are on the right processor.
2443 * The cmpxchg does the following atomically (without lock
2445 * 1. Relocate first pointer to the current per cpu area.
2446 * 2. Verify that tid and freelist have not been changed
2447 * 3. If they were not changed replace tid and freelist
2449 * Since this is without lock semantics the protection is only
2450 * against code executing on this cpu *not* from access by
2453 if (unlikely(!this_cpu_cmpxchg_double(
2454 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2456 next_object
, next_tid(tid
)))) {
2458 note_cmpxchg_failure("slab_alloc", s
, tid
);
2461 prefetch_freepointer(s
, next_object
);
2462 stat(s
, ALLOC_FASTPATH
);
2465 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2466 memset(object
, 0, s
->object_size
);
2468 slab_post_alloc_hook(s
, gfpflags
, object
);
2473 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2474 gfp_t gfpflags
, unsigned long addr
)
2476 return slab_alloc_node(s
, gfpflags
, NUMA_NO_NODE
, addr
);
2479 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2481 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2483 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
,
2488 EXPORT_SYMBOL(kmem_cache_alloc
);
2490 #ifdef CONFIG_TRACING
2491 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2493 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2494 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2497 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2501 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2503 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2505 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2506 s
->object_size
, s
->size
, gfpflags
, node
);
2510 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2512 #ifdef CONFIG_TRACING
2513 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2515 int node
, size_t size
)
2517 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2519 trace_kmalloc_node(_RET_IP_
, ret
,
2520 size
, s
->size
, gfpflags
, node
);
2523 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2528 * Slow patch handling. This may still be called frequently since objects
2529 * have a longer lifetime than the cpu slabs in most processing loads.
2531 * So we still attempt to reduce cache line usage. Just take the slab
2532 * lock and free the item. If there is no additional partial page
2533 * handling required then we can return immediately.
2535 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2536 void *x
, unsigned long addr
)
2539 void **object
= (void *)x
;
2542 unsigned long counters
;
2543 struct kmem_cache_node
*n
= NULL
;
2544 unsigned long uninitialized_var(flags
);
2546 stat(s
, FREE_SLOWPATH
);
2548 if (kmem_cache_debug(s
) &&
2549 !(n
= free_debug_processing(s
, page
, x
, addr
, &flags
)))
2554 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2557 prior
= page
->freelist
;
2558 counters
= page
->counters
;
2559 set_freepointer(s
, object
, prior
);
2560 new.counters
= counters
;
2561 was_frozen
= new.frozen
;
2563 if ((!new.inuse
|| !prior
) && !was_frozen
) {
2565 if (kmem_cache_has_cpu_partial(s
) && !prior
) {
2568 * Slab was on no list before and will be
2570 * We can defer the list move and instead
2575 } else { /* Needs to be taken off a list */
2577 n
= get_node(s
, page_to_nid(page
));
2579 * Speculatively acquire the list_lock.
2580 * If the cmpxchg does not succeed then we may
2581 * drop the list_lock without any processing.
2583 * Otherwise the list_lock will synchronize with
2584 * other processors updating the list of slabs.
2586 spin_lock_irqsave(&n
->list_lock
, flags
);
2591 } while (!cmpxchg_double_slab(s
, page
,
2593 object
, new.counters
,
2599 * If we just froze the page then put it onto the
2600 * per cpu partial list.
2602 if (new.frozen
&& !was_frozen
) {
2603 put_cpu_partial(s
, page
, 1);
2604 stat(s
, CPU_PARTIAL_FREE
);
2607 * The list lock was not taken therefore no list
2608 * activity can be necessary.
2611 stat(s
, FREE_FROZEN
);
2615 if (unlikely(!new.inuse
&& n
->nr_partial
>= s
->min_partial
))
2619 * Objects left in the slab. If it was not on the partial list before
2622 if (!kmem_cache_has_cpu_partial(s
) && unlikely(!prior
)) {
2623 if (kmem_cache_debug(s
))
2624 remove_full(s
, n
, page
);
2625 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2626 stat(s
, FREE_ADD_PARTIAL
);
2628 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2634 * Slab on the partial list.
2636 remove_partial(n
, page
);
2637 stat(s
, FREE_REMOVE_PARTIAL
);
2639 /* Slab must be on the full list */
2640 remove_full(s
, n
, page
);
2643 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2645 discard_slab(s
, page
);
2649 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2650 * can perform fastpath freeing without additional function calls.
2652 * The fastpath is only possible if we are freeing to the current cpu slab
2653 * of this processor. This typically the case if we have just allocated
2656 * If fastpath is not possible then fall back to __slab_free where we deal
2657 * with all sorts of special processing.
2659 static __always_inline
void slab_free(struct kmem_cache
*s
,
2660 struct page
*page
, void *x
, unsigned long addr
)
2662 void **object
= (void *)x
;
2663 struct kmem_cache_cpu
*c
;
2666 slab_free_hook(s
, x
);
2670 * Determine the currently cpus per cpu slab.
2671 * The cpu may change afterward. However that does not matter since
2672 * data is retrieved via this pointer. If we are on the same cpu
2673 * during the cmpxchg then the free will succedd.
2676 c
= this_cpu_ptr(s
->cpu_slab
);
2681 if (likely(page
== c
->page
)) {
2682 set_freepointer(s
, object
, c
->freelist
);
2684 if (unlikely(!this_cpu_cmpxchg_double(
2685 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2687 object
, next_tid(tid
)))) {
2689 note_cmpxchg_failure("slab_free", s
, tid
);
2692 stat(s
, FREE_FASTPATH
);
2694 __slab_free(s
, page
, x
, addr
);
2698 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2700 s
= cache_from_obj(s
, x
);
2703 slab_free(s
, virt_to_head_page(x
), x
, _RET_IP_
);
2704 trace_kmem_cache_free(_RET_IP_
, x
);
2706 EXPORT_SYMBOL(kmem_cache_free
);
2709 * Object placement in a slab is made very easy because we always start at
2710 * offset 0. If we tune the size of the object to the alignment then we can
2711 * get the required alignment by putting one properly sized object after
2714 * Notice that the allocation order determines the sizes of the per cpu
2715 * caches. Each processor has always one slab available for allocations.
2716 * Increasing the allocation order reduces the number of times that slabs
2717 * must be moved on and off the partial lists and is therefore a factor in
2722 * Mininum / Maximum order of slab pages. This influences locking overhead
2723 * and slab fragmentation. A higher order reduces the number of partial slabs
2724 * and increases the number of allocations possible without having to
2725 * take the list_lock.
2727 static int slub_min_order
;
2728 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2729 static int slub_min_objects
;
2732 * Merge control. If this is set then no merging of slab caches will occur.
2733 * (Could be removed. This was introduced to pacify the merge skeptics.)
2735 static int slub_nomerge
;
2738 * Calculate the order of allocation given an slab object size.
2740 * The order of allocation has significant impact on performance and other
2741 * system components. Generally order 0 allocations should be preferred since
2742 * order 0 does not cause fragmentation in the page allocator. Larger objects
2743 * be problematic to put into order 0 slabs because there may be too much
2744 * unused space left. We go to a higher order if more than 1/16th of the slab
2747 * In order to reach satisfactory performance we must ensure that a minimum
2748 * number of objects is in one slab. Otherwise we may generate too much
2749 * activity on the partial lists which requires taking the list_lock. This is
2750 * less a concern for large slabs though which are rarely used.
2752 * slub_max_order specifies the order where we begin to stop considering the
2753 * number of objects in a slab as critical. If we reach slub_max_order then
2754 * we try to keep the page order as low as possible. So we accept more waste
2755 * of space in favor of a small page order.
2757 * Higher order allocations also allow the placement of more objects in a
2758 * slab and thereby reduce object handling overhead. If the user has
2759 * requested a higher mininum order then we start with that one instead of
2760 * the smallest order which will fit the object.
2762 static inline int slab_order(int size
, int min_objects
,
2763 int max_order
, int fract_leftover
, int reserved
)
2767 int min_order
= slub_min_order
;
2769 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2770 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2772 for (order
= max(min_order
,
2773 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2774 order
<= max_order
; order
++) {
2776 unsigned long slab_size
= PAGE_SIZE
<< order
;
2778 if (slab_size
< min_objects
* size
+ reserved
)
2781 rem
= (slab_size
- reserved
) % size
;
2783 if (rem
<= slab_size
/ fract_leftover
)
2791 static inline int calculate_order(int size
, int reserved
)
2799 * Attempt to find best configuration for a slab. This
2800 * works by first attempting to generate a layout with
2801 * the best configuration and backing off gradually.
2803 * First we reduce the acceptable waste in a slab. Then
2804 * we reduce the minimum objects required in a slab.
2806 min_objects
= slub_min_objects
;
2808 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2809 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2810 min_objects
= min(min_objects
, max_objects
);
2812 while (min_objects
> 1) {
2814 while (fraction
>= 4) {
2815 order
= slab_order(size
, min_objects
,
2816 slub_max_order
, fraction
, reserved
);
2817 if (order
<= slub_max_order
)
2825 * We were unable to place multiple objects in a slab. Now
2826 * lets see if we can place a single object there.
2828 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2829 if (order
<= slub_max_order
)
2833 * Doh this slab cannot be placed using slub_max_order.
2835 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2836 if (order
< MAX_ORDER
)
2842 init_kmem_cache_node(struct kmem_cache_node
*n
)
2845 spin_lock_init(&n
->list_lock
);
2846 INIT_LIST_HEAD(&n
->partial
);
2847 #ifdef CONFIG_SLUB_DEBUG
2848 atomic_long_set(&n
->nr_slabs
, 0);
2849 atomic_long_set(&n
->total_objects
, 0);
2850 INIT_LIST_HEAD(&n
->full
);
2854 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2856 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2857 KMALLOC_SHIFT_HIGH
* sizeof(struct kmem_cache_cpu
));
2860 * Must align to double word boundary for the double cmpxchg
2861 * instructions to work; see __pcpu_double_call_return_bool().
2863 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2864 2 * sizeof(void *));
2869 init_kmem_cache_cpus(s
);
2874 static struct kmem_cache
*kmem_cache_node
;
2877 * No kmalloc_node yet so do it by hand. We know that this is the first
2878 * slab on the node for this slabcache. There are no concurrent accesses
2881 * Note that this function only works on the kmem_cache_node
2882 * when allocating for the kmem_cache_node. This is used for bootstrapping
2883 * memory on a fresh node that has no slab structures yet.
2885 static void early_kmem_cache_node_alloc(int node
)
2888 struct kmem_cache_node
*n
;
2890 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2892 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2895 if (page_to_nid(page
) != node
) {
2896 pr_err("SLUB: Unable to allocate memory from node %d\n", node
);
2897 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
2902 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2905 kmem_cache_node
->node
[node
] = n
;
2906 #ifdef CONFIG_SLUB_DEBUG
2907 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2908 init_tracking(kmem_cache_node
, n
);
2910 init_kmem_cache_node(n
);
2911 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2914 * No locks need to be taken here as it has just been
2915 * initialized and there is no concurrent access.
2917 __add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2920 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2923 struct kmem_cache_node
*n
;
2925 for_each_kmem_cache_node(s
, node
, n
) {
2926 kmem_cache_free(kmem_cache_node
, n
);
2927 s
->node
[node
] = NULL
;
2931 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2935 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2936 struct kmem_cache_node
*n
;
2938 if (slab_state
== DOWN
) {
2939 early_kmem_cache_node_alloc(node
);
2942 n
= kmem_cache_alloc_node(kmem_cache_node
,
2946 free_kmem_cache_nodes(s
);
2951 init_kmem_cache_node(n
);
2956 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2958 if (min
< MIN_PARTIAL
)
2960 else if (min
> MAX_PARTIAL
)
2962 s
->min_partial
= min
;
2966 * calculate_sizes() determines the order and the distribution of data within
2969 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2971 unsigned long flags
= s
->flags
;
2972 unsigned long size
= s
->object_size
;
2976 * Round up object size to the next word boundary. We can only
2977 * place the free pointer at word boundaries and this determines
2978 * the possible location of the free pointer.
2980 size
= ALIGN(size
, sizeof(void *));
2982 #ifdef CONFIG_SLUB_DEBUG
2984 * Determine if we can poison the object itself. If the user of
2985 * the slab may touch the object after free or before allocation
2986 * then we should never poison the object itself.
2988 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2990 s
->flags
|= __OBJECT_POISON
;
2992 s
->flags
&= ~__OBJECT_POISON
;
2996 * If we are Redzoning then check if there is some space between the
2997 * end of the object and the free pointer. If not then add an
2998 * additional word to have some bytes to store Redzone information.
3000 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
3001 size
+= sizeof(void *);
3005 * With that we have determined the number of bytes in actual use
3006 * by the object. This is the potential offset to the free pointer.
3010 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
3013 * Relocate free pointer after the object if it is not
3014 * permitted to overwrite the first word of the object on
3017 * This is the case if we do RCU, have a constructor or
3018 * destructor or are poisoning the objects.
3021 size
+= sizeof(void *);
3024 #ifdef CONFIG_SLUB_DEBUG
3025 if (flags
& SLAB_STORE_USER
)
3027 * Need to store information about allocs and frees after
3030 size
+= 2 * sizeof(struct track
);
3032 if (flags
& SLAB_RED_ZONE
)
3034 * Add some empty padding so that we can catch
3035 * overwrites from earlier objects rather than let
3036 * tracking information or the free pointer be
3037 * corrupted if a user writes before the start
3040 size
+= sizeof(void *);
3044 * SLUB stores one object immediately after another beginning from
3045 * offset 0. In order to align the objects we have to simply size
3046 * each object to conform to the alignment.
3048 size
= ALIGN(size
, s
->align
);
3050 if (forced_order
>= 0)
3051 order
= forced_order
;
3053 order
= calculate_order(size
, s
->reserved
);
3060 s
->allocflags
|= __GFP_COMP
;
3062 if (s
->flags
& SLAB_CACHE_DMA
)
3063 s
->allocflags
|= GFP_DMA
;
3065 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3066 s
->allocflags
|= __GFP_RECLAIMABLE
;
3069 * Determine the number of objects per slab
3071 s
->oo
= oo_make(order
, size
, s
->reserved
);
3072 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
3073 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3076 return !!oo_objects(s
->oo
);
3079 static int kmem_cache_open(struct kmem_cache
*s
, unsigned long flags
)
3081 s
->flags
= kmem_cache_flags(s
->size
, flags
, s
->name
, s
->ctor
);
3084 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3085 s
->reserved
= sizeof(struct rcu_head
);
3087 if (!calculate_sizes(s
, -1))
3089 if (disable_higher_order_debug
) {
3091 * Disable debugging flags that store metadata if the min slab
3094 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3095 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3097 if (!calculate_sizes(s
, -1))
3102 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3103 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3104 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3105 /* Enable fast mode */
3106 s
->flags
|= __CMPXCHG_DOUBLE
;
3110 * The larger the object size is, the more pages we want on the partial
3111 * list to avoid pounding the page allocator excessively.
3113 set_min_partial(s
, ilog2(s
->size
) / 2);
3116 * cpu_partial determined the maximum number of objects kept in the
3117 * per cpu partial lists of a processor.
3119 * Per cpu partial lists mainly contain slabs that just have one
3120 * object freed. If they are used for allocation then they can be
3121 * filled up again with minimal effort. The slab will never hit the
3122 * per node partial lists and therefore no locking will be required.
3124 * This setting also determines
3126 * A) The number of objects from per cpu partial slabs dumped to the
3127 * per node list when we reach the limit.
3128 * B) The number of objects in cpu partial slabs to extract from the
3129 * per node list when we run out of per cpu objects. We only fetch
3130 * 50% to keep some capacity around for frees.
3132 if (!kmem_cache_has_cpu_partial(s
))
3134 else if (s
->size
>= PAGE_SIZE
)
3136 else if (s
->size
>= 1024)
3138 else if (s
->size
>= 256)
3139 s
->cpu_partial
= 13;
3141 s
->cpu_partial
= 30;
3144 s
->remote_node_defrag_ratio
= 1000;
3146 if (!init_kmem_cache_nodes(s
))
3149 if (alloc_kmem_cache_cpus(s
))
3152 free_kmem_cache_nodes(s
);
3154 if (flags
& SLAB_PANIC
)
3155 panic("Cannot create slab %s size=%lu realsize=%u "
3156 "order=%u offset=%u flags=%lx\n",
3157 s
->name
, (unsigned long)s
->size
, s
->size
,
3158 oo_order(s
->oo
), s
->offset
, flags
);
3162 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3165 #ifdef CONFIG_SLUB_DEBUG
3166 void *addr
= page_address(page
);
3168 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3169 sizeof(long), GFP_ATOMIC
);
3172 slab_err(s
, page
, text
, s
->name
);
3175 get_map(s
, page
, map
);
3176 for_each_object(p
, s
, addr
, page
->objects
) {
3178 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3179 pr_err("INFO: Object 0x%p @offset=%tu\n", p
, p
- addr
);
3180 print_tracking(s
, p
);
3189 * Attempt to free all partial slabs on a node.
3190 * This is called from kmem_cache_close(). We must be the last thread
3191 * using the cache and therefore we do not need to lock anymore.
3193 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3195 struct page
*page
, *h
;
3197 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3199 __remove_partial(n
, page
);
3200 discard_slab(s
, page
);
3202 list_slab_objects(s
, page
,
3203 "Objects remaining in %s on kmem_cache_close()");
3209 * Release all resources used by a slab cache.
3211 static inline int kmem_cache_close(struct kmem_cache
*s
)
3214 struct kmem_cache_node
*n
;
3217 /* Attempt to free all objects */
3218 for_each_kmem_cache_node(s
, node
, n
) {
3220 if (n
->nr_partial
|| slabs_node(s
, node
))
3223 free_percpu(s
->cpu_slab
);
3224 free_kmem_cache_nodes(s
);
3228 int __kmem_cache_shutdown(struct kmem_cache
*s
)
3230 return kmem_cache_close(s
);
3233 /********************************************************************
3235 *******************************************************************/
3237 static int __init
setup_slub_min_order(char *str
)
3239 get_option(&str
, &slub_min_order
);
3244 __setup("slub_min_order=", setup_slub_min_order
);
3246 static int __init
setup_slub_max_order(char *str
)
3248 get_option(&str
, &slub_max_order
);
3249 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3254 __setup("slub_max_order=", setup_slub_max_order
);
3256 static int __init
setup_slub_min_objects(char *str
)
3258 get_option(&str
, &slub_min_objects
);
3263 __setup("slub_min_objects=", setup_slub_min_objects
);
3265 static int __init
setup_slub_nomerge(char *str
)
3271 __setup("slub_nomerge", setup_slub_nomerge
);
3273 void *__kmalloc(size_t size
, gfp_t flags
)
3275 struct kmem_cache
*s
;
3278 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3279 return kmalloc_large(size
, flags
);
3281 s
= kmalloc_slab(size
, flags
);
3283 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3286 ret
= slab_alloc(s
, flags
, _RET_IP_
);
3288 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3292 EXPORT_SYMBOL(__kmalloc
);
3295 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3300 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3301 page
= alloc_kmem_pages_node(node
, flags
, get_order(size
));
3303 ptr
= page_address(page
);
3305 kmalloc_large_node_hook(ptr
, size
, flags
);
3309 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3311 struct kmem_cache
*s
;
3314 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3315 ret
= kmalloc_large_node(size
, flags
, node
);
3317 trace_kmalloc_node(_RET_IP_
, ret
,
3318 size
, PAGE_SIZE
<< get_order(size
),
3324 s
= kmalloc_slab(size
, flags
);
3326 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3329 ret
= slab_alloc_node(s
, flags
, node
, _RET_IP_
);
3331 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3335 EXPORT_SYMBOL(__kmalloc_node
);
3338 size_t ksize(const void *object
)
3342 if (unlikely(object
== ZERO_SIZE_PTR
))
3345 page
= virt_to_head_page(object
);
3347 if (unlikely(!PageSlab(page
))) {
3348 WARN_ON(!PageCompound(page
));
3349 return PAGE_SIZE
<< compound_order(page
);
3352 return slab_ksize(page
->slab_cache
);
3354 EXPORT_SYMBOL(ksize
);
3356 void kfree(const void *x
)
3359 void *object
= (void *)x
;
3361 trace_kfree(_RET_IP_
, x
);
3363 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3366 page
= virt_to_head_page(x
);
3367 if (unlikely(!PageSlab(page
))) {
3368 BUG_ON(!PageCompound(page
));
3370 __free_kmem_pages(page
, compound_order(page
));
3373 slab_free(page
->slab_cache
, page
, object
, _RET_IP_
);
3375 EXPORT_SYMBOL(kfree
);
3378 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3379 * the remaining slabs by the number of items in use. The slabs with the
3380 * most items in use come first. New allocations will then fill those up
3381 * and thus they can be removed from the partial lists.
3383 * The slabs with the least items are placed last. This results in them
3384 * being allocated from last increasing the chance that the last objects
3385 * are freed in them.
3387 int __kmem_cache_shrink(struct kmem_cache
*s
)
3391 struct kmem_cache_node
*n
;
3394 int objects
= oo_objects(s
->max
);
3395 struct list_head
*slabs_by_inuse
=
3396 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3397 unsigned long flags
;
3399 if (!slabs_by_inuse
)
3403 for_each_kmem_cache_node(s
, node
, n
) {
3407 for (i
= 0; i
< objects
; i
++)
3408 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3410 spin_lock_irqsave(&n
->list_lock
, flags
);
3413 * Build lists indexed by the items in use in each slab.
3415 * Note that concurrent frees may occur while we hold the
3416 * list_lock. page->inuse here is the upper limit.
3418 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3419 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3425 * Rebuild the partial list with the slabs filled up most
3426 * first and the least used slabs at the end.
3428 for (i
= objects
- 1; i
> 0; i
--)
3429 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3431 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3433 /* Release empty slabs */
3434 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3435 discard_slab(s
, page
);
3438 kfree(slabs_by_inuse
);
3442 static int slab_mem_going_offline_callback(void *arg
)
3444 struct kmem_cache
*s
;
3446 mutex_lock(&slab_mutex
);
3447 list_for_each_entry(s
, &slab_caches
, list
)
3448 __kmem_cache_shrink(s
);
3449 mutex_unlock(&slab_mutex
);
3454 static void slab_mem_offline_callback(void *arg
)
3456 struct kmem_cache_node
*n
;
3457 struct kmem_cache
*s
;
3458 struct memory_notify
*marg
= arg
;
3461 offline_node
= marg
->status_change_nid_normal
;
3464 * If the node still has available memory. we need kmem_cache_node
3467 if (offline_node
< 0)
3470 mutex_lock(&slab_mutex
);
3471 list_for_each_entry(s
, &slab_caches
, list
) {
3472 n
= get_node(s
, offline_node
);
3475 * if n->nr_slabs > 0, slabs still exist on the node
3476 * that is going down. We were unable to free them,
3477 * and offline_pages() function shouldn't call this
3478 * callback. So, we must fail.
3480 BUG_ON(slabs_node(s
, offline_node
));
3482 s
->node
[offline_node
] = NULL
;
3483 kmem_cache_free(kmem_cache_node
, n
);
3486 mutex_unlock(&slab_mutex
);
3489 static int slab_mem_going_online_callback(void *arg
)
3491 struct kmem_cache_node
*n
;
3492 struct kmem_cache
*s
;
3493 struct memory_notify
*marg
= arg
;
3494 int nid
= marg
->status_change_nid_normal
;
3498 * If the node's memory is already available, then kmem_cache_node is
3499 * already created. Nothing to do.
3505 * We are bringing a node online. No memory is available yet. We must
3506 * allocate a kmem_cache_node structure in order to bring the node
3509 mutex_lock(&slab_mutex
);
3510 list_for_each_entry(s
, &slab_caches
, list
) {
3512 * XXX: kmem_cache_alloc_node will fallback to other nodes
3513 * since memory is not yet available from the node that
3516 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3521 init_kmem_cache_node(n
);
3525 mutex_unlock(&slab_mutex
);
3529 static int slab_memory_callback(struct notifier_block
*self
,
3530 unsigned long action
, void *arg
)
3535 case MEM_GOING_ONLINE
:
3536 ret
= slab_mem_going_online_callback(arg
);
3538 case MEM_GOING_OFFLINE
:
3539 ret
= slab_mem_going_offline_callback(arg
);
3542 case MEM_CANCEL_ONLINE
:
3543 slab_mem_offline_callback(arg
);
3546 case MEM_CANCEL_OFFLINE
:
3550 ret
= notifier_from_errno(ret
);
3556 static struct notifier_block slab_memory_callback_nb
= {
3557 .notifier_call
= slab_memory_callback
,
3558 .priority
= SLAB_CALLBACK_PRI
,
3561 /********************************************************************
3562 * Basic setup of slabs
3563 *******************************************************************/
3566 * Used for early kmem_cache structures that were allocated using
3567 * the page allocator. Allocate them properly then fix up the pointers
3568 * that may be pointing to the wrong kmem_cache structure.
3571 static struct kmem_cache
* __init
bootstrap(struct kmem_cache
*static_cache
)
3574 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
3575 struct kmem_cache_node
*n
;
3577 memcpy(s
, static_cache
, kmem_cache
->object_size
);
3580 * This runs very early, and only the boot processor is supposed to be
3581 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3584 __flush_cpu_slab(s
, smp_processor_id());
3585 for_each_kmem_cache_node(s
, node
, n
) {
3588 list_for_each_entry(p
, &n
->partial
, lru
)
3591 #ifdef CONFIG_SLUB_DEBUG
3592 list_for_each_entry(p
, &n
->full
, lru
)
3596 list_add(&s
->list
, &slab_caches
);
3600 void __init
kmem_cache_init(void)
3602 static __initdata
struct kmem_cache boot_kmem_cache
,
3603 boot_kmem_cache_node
;
3605 if (debug_guardpage_minorder())
3608 kmem_cache_node
= &boot_kmem_cache_node
;
3609 kmem_cache
= &boot_kmem_cache
;
3611 create_boot_cache(kmem_cache_node
, "kmem_cache_node",
3612 sizeof(struct kmem_cache_node
), SLAB_HWCACHE_ALIGN
);
3614 register_hotmemory_notifier(&slab_memory_callback_nb
);
3616 /* Able to allocate the per node structures */
3617 slab_state
= PARTIAL
;
3619 create_boot_cache(kmem_cache
, "kmem_cache",
3620 offsetof(struct kmem_cache
, node
) +
3621 nr_node_ids
* sizeof(struct kmem_cache_node
*),
3622 SLAB_HWCACHE_ALIGN
);
3624 kmem_cache
= bootstrap(&boot_kmem_cache
);
3627 * Allocate kmem_cache_node properly from the kmem_cache slab.
3628 * kmem_cache_node is separately allocated so no need to
3629 * update any list pointers.
3631 kmem_cache_node
= bootstrap(&boot_kmem_cache_node
);
3633 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3634 create_kmalloc_caches(0);
3637 register_cpu_notifier(&slab_notifier
);
3640 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
3642 slub_min_order
, slub_max_order
, slub_min_objects
,
3643 nr_cpu_ids
, nr_node_ids
);
3646 void __init
kmem_cache_init_late(void)
3651 * Find a mergeable slab cache
3653 static int slab_unmergeable(struct kmem_cache
*s
)
3655 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3658 if (!is_root_cache(s
))
3665 * We may have set a slab to be unmergeable during bootstrap.
3667 if (s
->refcount
< 0)
3673 static struct kmem_cache
*find_mergeable(size_t size
, size_t align
,
3674 unsigned long flags
, const char *name
, void (*ctor
)(void *))
3676 struct kmem_cache
*s
;
3678 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3684 size
= ALIGN(size
, sizeof(void *));
3685 align
= calculate_alignment(flags
, align
, size
);
3686 size
= ALIGN(size
, align
);
3687 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3689 list_for_each_entry(s
, &slab_caches
, list
) {
3690 if (slab_unmergeable(s
))
3696 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3699 * Check if alignment is compatible.
3700 * Courtesy of Adrian Drzewiecki
3702 if ((s
->size
& ~(align
- 1)) != s
->size
)
3705 if (s
->size
- size
>= sizeof(void *))
3714 __kmem_cache_alias(const char *name
, size_t size
, size_t align
,
3715 unsigned long flags
, void (*ctor
)(void *))
3717 struct kmem_cache
*s
;
3719 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3722 struct kmem_cache
*c
;
3727 * Adjust the object sizes so that we clear
3728 * the complete object on kzalloc.
3730 s
->object_size
= max(s
->object_size
, (int)size
);
3731 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3733 for_each_memcg_cache_index(i
) {
3734 c
= cache_from_memcg_idx(s
, i
);
3737 c
->object_size
= s
->object_size
;
3738 c
->inuse
= max_t(int, c
->inuse
,
3739 ALIGN(size
, sizeof(void *)));
3742 if (sysfs_slab_alias(s
, name
)) {
3751 int __kmem_cache_create(struct kmem_cache
*s
, unsigned long flags
)
3755 err
= kmem_cache_open(s
, flags
);
3759 /* Mutex is not taken during early boot */
3760 if (slab_state
<= UP
)
3763 memcg_propagate_slab_attrs(s
);
3764 err
= sysfs_slab_add(s
);
3766 kmem_cache_close(s
);
3773 * Use the cpu notifier to insure that the cpu slabs are flushed when
3776 static int slab_cpuup_callback(struct notifier_block
*nfb
,
3777 unsigned long action
, void *hcpu
)
3779 long cpu
= (long)hcpu
;
3780 struct kmem_cache
*s
;
3781 unsigned long flags
;
3784 case CPU_UP_CANCELED
:
3785 case CPU_UP_CANCELED_FROZEN
:
3787 case CPU_DEAD_FROZEN
:
3788 mutex_lock(&slab_mutex
);
3789 list_for_each_entry(s
, &slab_caches
, list
) {
3790 local_irq_save(flags
);
3791 __flush_cpu_slab(s
, cpu
);
3792 local_irq_restore(flags
);
3794 mutex_unlock(&slab_mutex
);
3802 static struct notifier_block slab_notifier
= {
3803 .notifier_call
= slab_cpuup_callback
3808 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3810 struct kmem_cache
*s
;
3813 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3814 return kmalloc_large(size
, gfpflags
);
3816 s
= kmalloc_slab(size
, gfpflags
);
3818 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3821 ret
= slab_alloc(s
, gfpflags
, caller
);
3823 /* Honor the call site pointer we received. */
3824 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3830 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3831 int node
, unsigned long caller
)
3833 struct kmem_cache
*s
;
3836 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3837 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3839 trace_kmalloc_node(caller
, ret
,
3840 size
, PAGE_SIZE
<< get_order(size
),
3846 s
= kmalloc_slab(size
, gfpflags
);
3848 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3851 ret
= slab_alloc_node(s
, gfpflags
, node
, caller
);
3853 /* Honor the call site pointer we received. */
3854 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3861 static int count_inuse(struct page
*page
)
3866 static int count_total(struct page
*page
)
3868 return page
->objects
;
3872 #ifdef CONFIG_SLUB_DEBUG
3873 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3877 void *addr
= page_address(page
);
3879 if (!check_slab(s
, page
) ||
3880 !on_freelist(s
, page
, NULL
))
3883 /* Now we know that a valid freelist exists */
3884 bitmap_zero(map
, page
->objects
);
3886 get_map(s
, page
, map
);
3887 for_each_object(p
, s
, addr
, page
->objects
) {
3888 if (test_bit(slab_index(p
, s
, addr
), map
))
3889 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
3893 for_each_object(p
, s
, addr
, page
->objects
)
3894 if (!test_bit(slab_index(p
, s
, addr
), map
))
3895 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
3900 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3904 validate_slab(s
, page
, map
);
3908 static int validate_slab_node(struct kmem_cache
*s
,
3909 struct kmem_cache_node
*n
, unsigned long *map
)
3911 unsigned long count
= 0;
3913 unsigned long flags
;
3915 spin_lock_irqsave(&n
->list_lock
, flags
);
3917 list_for_each_entry(page
, &n
->partial
, lru
) {
3918 validate_slab_slab(s
, page
, map
);
3921 if (count
!= n
->nr_partial
)
3922 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
3923 s
->name
, count
, n
->nr_partial
);
3925 if (!(s
->flags
& SLAB_STORE_USER
))
3928 list_for_each_entry(page
, &n
->full
, lru
) {
3929 validate_slab_slab(s
, page
, map
);
3932 if (count
!= atomic_long_read(&n
->nr_slabs
))
3933 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
3934 s
->name
, count
, atomic_long_read(&n
->nr_slabs
));
3937 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3941 static long validate_slab_cache(struct kmem_cache
*s
)
3944 unsigned long count
= 0;
3945 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3946 sizeof(unsigned long), GFP_KERNEL
);
3947 struct kmem_cache_node
*n
;
3953 for_each_kmem_cache_node(s
, node
, n
)
3954 count
+= validate_slab_node(s
, n
, map
);
3959 * Generate lists of code addresses where slabcache objects are allocated
3964 unsigned long count
;
3971 DECLARE_BITMAP(cpus
, NR_CPUS
);
3977 unsigned long count
;
3978 struct location
*loc
;
3981 static void free_loc_track(struct loc_track
*t
)
3984 free_pages((unsigned long)t
->loc
,
3985 get_order(sizeof(struct location
) * t
->max
));
3988 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3993 order
= get_order(sizeof(struct location
) * max
);
3995 l
= (void *)__get_free_pages(flags
, order
);
4000 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4008 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4009 const struct track
*track
)
4011 long start
, end
, pos
;
4013 unsigned long caddr
;
4014 unsigned long age
= jiffies
- track
->when
;
4020 pos
= start
+ (end
- start
+ 1) / 2;
4023 * There is nothing at "end". If we end up there
4024 * we need to add something to before end.
4029 caddr
= t
->loc
[pos
].addr
;
4030 if (track
->addr
== caddr
) {
4036 if (age
< l
->min_time
)
4038 if (age
> l
->max_time
)
4041 if (track
->pid
< l
->min_pid
)
4042 l
->min_pid
= track
->pid
;
4043 if (track
->pid
> l
->max_pid
)
4044 l
->max_pid
= track
->pid
;
4046 cpumask_set_cpu(track
->cpu
,
4047 to_cpumask(l
->cpus
));
4049 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4053 if (track
->addr
< caddr
)
4060 * Not found. Insert new tracking element.
4062 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4068 (t
->count
- pos
) * sizeof(struct location
));
4071 l
->addr
= track
->addr
;
4075 l
->min_pid
= track
->pid
;
4076 l
->max_pid
= track
->pid
;
4077 cpumask_clear(to_cpumask(l
->cpus
));
4078 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4079 nodes_clear(l
->nodes
);
4080 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4084 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4085 struct page
*page
, enum track_item alloc
,
4088 void *addr
= page_address(page
);
4091 bitmap_zero(map
, page
->objects
);
4092 get_map(s
, page
, map
);
4094 for_each_object(p
, s
, addr
, page
->objects
)
4095 if (!test_bit(slab_index(p
, s
, addr
), map
))
4096 add_location(t
, s
, get_track(s
, p
, alloc
));
4099 static int list_locations(struct kmem_cache
*s
, char *buf
,
4100 enum track_item alloc
)
4104 struct loc_track t
= { 0, 0, NULL
};
4106 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4107 sizeof(unsigned long), GFP_KERNEL
);
4108 struct kmem_cache_node
*n
;
4110 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4113 return sprintf(buf
, "Out of memory\n");
4115 /* Push back cpu slabs */
4118 for_each_kmem_cache_node(s
, node
, n
) {
4119 unsigned long flags
;
4122 if (!atomic_long_read(&n
->nr_slabs
))
4125 spin_lock_irqsave(&n
->list_lock
, flags
);
4126 list_for_each_entry(page
, &n
->partial
, lru
)
4127 process_slab(&t
, s
, page
, alloc
, map
);
4128 list_for_each_entry(page
, &n
->full
, lru
)
4129 process_slab(&t
, s
, page
, alloc
, map
);
4130 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4133 for (i
= 0; i
< t
.count
; i
++) {
4134 struct location
*l
= &t
.loc
[i
];
4136 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4138 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4141 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4143 len
+= sprintf(buf
+ len
, "<not-available>");
4145 if (l
->sum_time
!= l
->min_time
) {
4146 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4148 (long)div_u64(l
->sum_time
, l
->count
),
4151 len
+= sprintf(buf
+ len
, " age=%ld",
4154 if (l
->min_pid
!= l
->max_pid
)
4155 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4156 l
->min_pid
, l
->max_pid
);
4158 len
+= sprintf(buf
+ len
, " pid=%ld",
4161 if (num_online_cpus() > 1 &&
4162 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4163 len
< PAGE_SIZE
- 60) {
4164 len
+= sprintf(buf
+ len
, " cpus=");
4165 len
+= cpulist_scnprintf(buf
+ len
,
4166 PAGE_SIZE
- len
- 50,
4167 to_cpumask(l
->cpus
));
4170 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4171 len
< PAGE_SIZE
- 60) {
4172 len
+= sprintf(buf
+ len
, " nodes=");
4173 len
+= nodelist_scnprintf(buf
+ len
,
4174 PAGE_SIZE
- len
- 50,
4178 len
+= sprintf(buf
+ len
, "\n");
4184 len
+= sprintf(buf
, "No data\n");
4189 #ifdef SLUB_RESILIENCY_TEST
4190 static void __init
resiliency_test(void)
4194 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || KMALLOC_SHIFT_HIGH
< 10);
4196 pr_err("SLUB resiliency testing\n");
4197 pr_err("-----------------------\n");
4198 pr_err("A. Corruption after allocation\n");
4200 p
= kzalloc(16, GFP_KERNEL
);
4202 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4205 validate_slab_cache(kmalloc_caches
[4]);
4207 /* Hmmm... The next two are dangerous */
4208 p
= kzalloc(32, GFP_KERNEL
);
4209 p
[32 + sizeof(void *)] = 0x34;
4210 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4212 pr_err("If allocated object is overwritten then not detectable\n\n");
4214 validate_slab_cache(kmalloc_caches
[5]);
4215 p
= kzalloc(64, GFP_KERNEL
);
4216 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4218 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4220 pr_err("If allocated object is overwritten then not detectable\n\n");
4221 validate_slab_cache(kmalloc_caches
[6]);
4223 pr_err("\nB. Corruption after free\n");
4224 p
= kzalloc(128, GFP_KERNEL
);
4227 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4228 validate_slab_cache(kmalloc_caches
[7]);
4230 p
= kzalloc(256, GFP_KERNEL
);
4233 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p
);
4234 validate_slab_cache(kmalloc_caches
[8]);
4236 p
= kzalloc(512, GFP_KERNEL
);
4239 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4240 validate_slab_cache(kmalloc_caches
[9]);
4244 static void resiliency_test(void) {};
4249 enum slab_stat_type
{
4250 SL_ALL
, /* All slabs */
4251 SL_PARTIAL
, /* Only partially allocated slabs */
4252 SL_CPU
, /* Only slabs used for cpu caches */
4253 SL_OBJECTS
, /* Determine allocated objects not slabs */
4254 SL_TOTAL
/* Determine object capacity not slabs */
4257 #define SO_ALL (1 << SL_ALL)
4258 #define SO_PARTIAL (1 << SL_PARTIAL)
4259 #define SO_CPU (1 << SL_CPU)
4260 #define SO_OBJECTS (1 << SL_OBJECTS)
4261 #define SO_TOTAL (1 << SL_TOTAL)
4263 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4264 char *buf
, unsigned long flags
)
4266 unsigned long total
= 0;
4269 unsigned long *nodes
;
4271 nodes
= kzalloc(sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4275 if (flags
& SO_CPU
) {
4278 for_each_possible_cpu(cpu
) {
4279 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
,
4284 page
= ACCESS_ONCE(c
->page
);
4288 node
= page_to_nid(page
);
4289 if (flags
& SO_TOTAL
)
4291 else if (flags
& SO_OBJECTS
)
4299 page
= ACCESS_ONCE(c
->partial
);
4301 node
= page_to_nid(page
);
4302 if (flags
& SO_TOTAL
)
4304 else if (flags
& SO_OBJECTS
)
4315 #ifdef CONFIG_SLUB_DEBUG
4316 if (flags
& SO_ALL
) {
4317 struct kmem_cache_node
*n
;
4319 for_each_kmem_cache_node(s
, node
, n
) {
4321 if (flags
& SO_TOTAL
)
4322 x
= atomic_long_read(&n
->total_objects
);
4323 else if (flags
& SO_OBJECTS
)
4324 x
= atomic_long_read(&n
->total_objects
) -
4325 count_partial(n
, count_free
);
4327 x
= atomic_long_read(&n
->nr_slabs
);
4334 if (flags
& SO_PARTIAL
) {
4335 struct kmem_cache_node
*n
;
4337 for_each_kmem_cache_node(s
, node
, n
) {
4338 if (flags
& SO_TOTAL
)
4339 x
= count_partial(n
, count_total
);
4340 else if (flags
& SO_OBJECTS
)
4341 x
= count_partial(n
, count_inuse
);
4348 x
= sprintf(buf
, "%lu", total
);
4350 for (node
= 0; node
< nr_node_ids
; node
++)
4352 x
+= sprintf(buf
+ x
, " N%d=%lu",
4357 return x
+ sprintf(buf
+ x
, "\n");
4360 #ifdef CONFIG_SLUB_DEBUG
4361 static int any_slab_objects(struct kmem_cache
*s
)
4364 struct kmem_cache_node
*n
;
4366 for_each_kmem_cache_node(s
, node
, n
)
4367 if (atomic_long_read(&n
->total_objects
))
4374 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4375 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4377 struct slab_attribute
{
4378 struct attribute attr
;
4379 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4380 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4383 #define SLAB_ATTR_RO(_name) \
4384 static struct slab_attribute _name##_attr = \
4385 __ATTR(_name, 0400, _name##_show, NULL)
4387 #define SLAB_ATTR(_name) \
4388 static struct slab_attribute _name##_attr = \
4389 __ATTR(_name, 0600, _name##_show, _name##_store)
4391 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4393 return sprintf(buf
, "%d\n", s
->size
);
4395 SLAB_ATTR_RO(slab_size
);
4397 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4399 return sprintf(buf
, "%d\n", s
->align
);
4401 SLAB_ATTR_RO(align
);
4403 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4405 return sprintf(buf
, "%d\n", s
->object_size
);
4407 SLAB_ATTR_RO(object_size
);
4409 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4411 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4413 SLAB_ATTR_RO(objs_per_slab
);
4415 static ssize_t
order_store(struct kmem_cache
*s
,
4416 const char *buf
, size_t length
)
4418 unsigned long order
;
4421 err
= kstrtoul(buf
, 10, &order
);
4425 if (order
> slub_max_order
|| order
< slub_min_order
)
4428 calculate_sizes(s
, order
);
4432 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4434 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4438 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4440 return sprintf(buf
, "%lu\n", s
->min_partial
);
4443 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4449 err
= kstrtoul(buf
, 10, &min
);
4453 set_min_partial(s
, min
);
4456 SLAB_ATTR(min_partial
);
4458 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4460 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4463 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4466 unsigned long objects
;
4469 err
= kstrtoul(buf
, 10, &objects
);
4472 if (objects
&& !kmem_cache_has_cpu_partial(s
))
4475 s
->cpu_partial
= objects
;
4479 SLAB_ATTR(cpu_partial
);
4481 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4485 return sprintf(buf
, "%pS\n", s
->ctor
);
4489 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4491 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4493 SLAB_ATTR_RO(aliases
);
4495 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4497 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4499 SLAB_ATTR_RO(partial
);
4501 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4503 return show_slab_objects(s
, buf
, SO_CPU
);
4505 SLAB_ATTR_RO(cpu_slabs
);
4507 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4509 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4511 SLAB_ATTR_RO(objects
);
4513 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4515 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4517 SLAB_ATTR_RO(objects_partial
);
4519 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4526 for_each_online_cpu(cpu
) {
4527 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4530 pages
+= page
->pages
;
4531 objects
+= page
->pobjects
;
4535 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4538 for_each_online_cpu(cpu
) {
4539 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4541 if (page
&& len
< PAGE_SIZE
- 20)
4542 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4543 page
->pobjects
, page
->pages
);
4546 return len
+ sprintf(buf
+ len
, "\n");
4548 SLAB_ATTR_RO(slabs_cpu_partial
);
4550 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4552 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4555 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4556 const char *buf
, size_t length
)
4558 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4560 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4563 SLAB_ATTR(reclaim_account
);
4565 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4567 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4569 SLAB_ATTR_RO(hwcache_align
);
4571 #ifdef CONFIG_ZONE_DMA
4572 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4574 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4576 SLAB_ATTR_RO(cache_dma
);
4579 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4581 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4583 SLAB_ATTR_RO(destroy_by_rcu
);
4585 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4587 return sprintf(buf
, "%d\n", s
->reserved
);
4589 SLAB_ATTR_RO(reserved
);
4591 #ifdef CONFIG_SLUB_DEBUG
4592 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4594 return show_slab_objects(s
, buf
, SO_ALL
);
4596 SLAB_ATTR_RO(slabs
);
4598 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4600 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4602 SLAB_ATTR_RO(total_objects
);
4604 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4606 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4609 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4610 const char *buf
, size_t length
)
4612 s
->flags
&= ~SLAB_DEBUG_FREE
;
4613 if (buf
[0] == '1') {
4614 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4615 s
->flags
|= SLAB_DEBUG_FREE
;
4619 SLAB_ATTR(sanity_checks
);
4621 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4623 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4626 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4629 s
->flags
&= ~SLAB_TRACE
;
4630 if (buf
[0] == '1') {
4631 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4632 s
->flags
|= SLAB_TRACE
;
4638 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4640 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4643 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4644 const char *buf
, size_t length
)
4646 if (any_slab_objects(s
))
4649 s
->flags
&= ~SLAB_RED_ZONE
;
4650 if (buf
[0] == '1') {
4651 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4652 s
->flags
|= SLAB_RED_ZONE
;
4654 calculate_sizes(s
, -1);
4657 SLAB_ATTR(red_zone
);
4659 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4661 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4664 static ssize_t
poison_store(struct kmem_cache
*s
,
4665 const char *buf
, size_t length
)
4667 if (any_slab_objects(s
))
4670 s
->flags
&= ~SLAB_POISON
;
4671 if (buf
[0] == '1') {
4672 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4673 s
->flags
|= SLAB_POISON
;
4675 calculate_sizes(s
, -1);
4680 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4682 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4685 static ssize_t
store_user_store(struct kmem_cache
*s
,
4686 const char *buf
, size_t length
)
4688 if (any_slab_objects(s
))
4691 s
->flags
&= ~SLAB_STORE_USER
;
4692 if (buf
[0] == '1') {
4693 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4694 s
->flags
|= SLAB_STORE_USER
;
4696 calculate_sizes(s
, -1);
4699 SLAB_ATTR(store_user
);
4701 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4706 static ssize_t
validate_store(struct kmem_cache
*s
,
4707 const char *buf
, size_t length
)
4711 if (buf
[0] == '1') {
4712 ret
= validate_slab_cache(s
);
4718 SLAB_ATTR(validate
);
4720 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4722 if (!(s
->flags
& SLAB_STORE_USER
))
4724 return list_locations(s
, buf
, TRACK_ALLOC
);
4726 SLAB_ATTR_RO(alloc_calls
);
4728 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4730 if (!(s
->flags
& SLAB_STORE_USER
))
4732 return list_locations(s
, buf
, TRACK_FREE
);
4734 SLAB_ATTR_RO(free_calls
);
4735 #endif /* CONFIG_SLUB_DEBUG */
4737 #ifdef CONFIG_FAILSLAB
4738 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4740 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4743 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4746 s
->flags
&= ~SLAB_FAILSLAB
;
4748 s
->flags
|= SLAB_FAILSLAB
;
4751 SLAB_ATTR(failslab
);
4754 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4759 static ssize_t
shrink_store(struct kmem_cache
*s
,
4760 const char *buf
, size_t length
)
4762 if (buf
[0] == '1') {
4763 int rc
= kmem_cache_shrink(s
);
4774 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4776 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4779 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4780 const char *buf
, size_t length
)
4782 unsigned long ratio
;
4785 err
= kstrtoul(buf
, 10, &ratio
);
4790 s
->remote_node_defrag_ratio
= ratio
* 10;
4794 SLAB_ATTR(remote_node_defrag_ratio
);
4797 #ifdef CONFIG_SLUB_STATS
4798 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4800 unsigned long sum
= 0;
4803 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4808 for_each_online_cpu(cpu
) {
4809 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4815 len
= sprintf(buf
, "%lu", sum
);
4818 for_each_online_cpu(cpu
) {
4819 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4820 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4824 return len
+ sprintf(buf
+ len
, "\n");
4827 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4831 for_each_online_cpu(cpu
)
4832 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4835 #define STAT_ATTR(si, text) \
4836 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4838 return show_stat(s, buf, si); \
4840 static ssize_t text##_store(struct kmem_cache *s, \
4841 const char *buf, size_t length) \
4843 if (buf[0] != '0') \
4845 clear_stat(s, si); \
4850 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4851 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4852 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4853 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4854 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4855 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4856 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4857 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4858 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4859 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4860 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
4861 STAT_ATTR(FREE_SLAB
, free_slab
);
4862 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4863 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4864 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4865 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4866 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4867 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4868 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
4869 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4870 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
4871 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
4872 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
4873 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
4874 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
4875 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
4878 static struct attribute
*slab_attrs
[] = {
4879 &slab_size_attr
.attr
,
4880 &object_size_attr
.attr
,
4881 &objs_per_slab_attr
.attr
,
4883 &min_partial_attr
.attr
,
4884 &cpu_partial_attr
.attr
,
4886 &objects_partial_attr
.attr
,
4888 &cpu_slabs_attr
.attr
,
4892 &hwcache_align_attr
.attr
,
4893 &reclaim_account_attr
.attr
,
4894 &destroy_by_rcu_attr
.attr
,
4896 &reserved_attr
.attr
,
4897 &slabs_cpu_partial_attr
.attr
,
4898 #ifdef CONFIG_SLUB_DEBUG
4899 &total_objects_attr
.attr
,
4901 &sanity_checks_attr
.attr
,
4903 &red_zone_attr
.attr
,
4905 &store_user_attr
.attr
,
4906 &validate_attr
.attr
,
4907 &alloc_calls_attr
.attr
,
4908 &free_calls_attr
.attr
,
4910 #ifdef CONFIG_ZONE_DMA
4911 &cache_dma_attr
.attr
,
4914 &remote_node_defrag_ratio_attr
.attr
,
4916 #ifdef CONFIG_SLUB_STATS
4917 &alloc_fastpath_attr
.attr
,
4918 &alloc_slowpath_attr
.attr
,
4919 &free_fastpath_attr
.attr
,
4920 &free_slowpath_attr
.attr
,
4921 &free_frozen_attr
.attr
,
4922 &free_add_partial_attr
.attr
,
4923 &free_remove_partial_attr
.attr
,
4924 &alloc_from_partial_attr
.attr
,
4925 &alloc_slab_attr
.attr
,
4926 &alloc_refill_attr
.attr
,
4927 &alloc_node_mismatch_attr
.attr
,
4928 &free_slab_attr
.attr
,
4929 &cpuslab_flush_attr
.attr
,
4930 &deactivate_full_attr
.attr
,
4931 &deactivate_empty_attr
.attr
,
4932 &deactivate_to_head_attr
.attr
,
4933 &deactivate_to_tail_attr
.attr
,
4934 &deactivate_remote_frees_attr
.attr
,
4935 &deactivate_bypass_attr
.attr
,
4936 &order_fallback_attr
.attr
,
4937 &cmpxchg_double_fail_attr
.attr
,
4938 &cmpxchg_double_cpu_fail_attr
.attr
,
4939 &cpu_partial_alloc_attr
.attr
,
4940 &cpu_partial_free_attr
.attr
,
4941 &cpu_partial_node_attr
.attr
,
4942 &cpu_partial_drain_attr
.attr
,
4944 #ifdef CONFIG_FAILSLAB
4945 &failslab_attr
.attr
,
4951 static struct attribute_group slab_attr_group
= {
4952 .attrs
= slab_attrs
,
4955 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4956 struct attribute
*attr
,
4959 struct slab_attribute
*attribute
;
4960 struct kmem_cache
*s
;
4963 attribute
= to_slab_attr(attr
);
4966 if (!attribute
->show
)
4969 err
= attribute
->show(s
, buf
);
4974 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4975 struct attribute
*attr
,
4976 const char *buf
, size_t len
)
4978 struct slab_attribute
*attribute
;
4979 struct kmem_cache
*s
;
4982 attribute
= to_slab_attr(attr
);
4985 if (!attribute
->store
)
4988 err
= attribute
->store(s
, buf
, len
);
4989 #ifdef CONFIG_MEMCG_KMEM
4990 if (slab_state
>= FULL
&& err
>= 0 && is_root_cache(s
)) {
4993 mutex_lock(&slab_mutex
);
4994 if (s
->max_attr_size
< len
)
4995 s
->max_attr_size
= len
;
4998 * This is a best effort propagation, so this function's return
4999 * value will be determined by the parent cache only. This is
5000 * basically because not all attributes will have a well
5001 * defined semantics for rollbacks - most of the actions will
5002 * have permanent effects.
5004 * Returning the error value of any of the children that fail
5005 * is not 100 % defined, in the sense that users seeing the
5006 * error code won't be able to know anything about the state of
5009 * Only returning the error code for the parent cache at least
5010 * has well defined semantics. The cache being written to
5011 * directly either failed or succeeded, in which case we loop
5012 * through the descendants with best-effort propagation.
5014 for_each_memcg_cache_index(i
) {
5015 struct kmem_cache
*c
= cache_from_memcg_idx(s
, i
);
5017 attribute
->store(c
, buf
, len
);
5019 mutex_unlock(&slab_mutex
);
5025 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
)
5027 #ifdef CONFIG_MEMCG_KMEM
5029 char *buffer
= NULL
;
5030 struct kmem_cache
*root_cache
;
5032 if (is_root_cache(s
))
5035 root_cache
= s
->memcg_params
->root_cache
;
5038 * This mean this cache had no attribute written. Therefore, no point
5039 * in copying default values around
5041 if (!root_cache
->max_attr_size
)
5044 for (i
= 0; i
< ARRAY_SIZE(slab_attrs
); i
++) {
5047 struct slab_attribute
*attr
= to_slab_attr(slab_attrs
[i
]);
5049 if (!attr
|| !attr
->store
|| !attr
->show
)
5053 * It is really bad that we have to allocate here, so we will
5054 * do it only as a fallback. If we actually allocate, though,
5055 * we can just use the allocated buffer until the end.
5057 * Most of the slub attributes will tend to be very small in
5058 * size, but sysfs allows buffers up to a page, so they can
5059 * theoretically happen.
5063 else if (root_cache
->max_attr_size
< ARRAY_SIZE(mbuf
))
5066 buffer
= (char *) get_zeroed_page(GFP_KERNEL
);
5067 if (WARN_ON(!buffer
))
5072 attr
->show(root_cache
, buf
);
5073 attr
->store(s
, buf
, strlen(buf
));
5077 free_page((unsigned long)buffer
);
5081 static void kmem_cache_release(struct kobject
*k
)
5083 slab_kmem_cache_release(to_slab(k
));
5086 static const struct sysfs_ops slab_sysfs_ops
= {
5087 .show
= slab_attr_show
,
5088 .store
= slab_attr_store
,
5091 static struct kobj_type slab_ktype
= {
5092 .sysfs_ops
= &slab_sysfs_ops
,
5093 .release
= kmem_cache_release
,
5096 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5098 struct kobj_type
*ktype
= get_ktype(kobj
);
5100 if (ktype
== &slab_ktype
)
5105 static const struct kset_uevent_ops slab_uevent_ops
= {
5106 .filter
= uevent_filter
,
5109 static struct kset
*slab_kset
;
5111 static inline struct kset
*cache_kset(struct kmem_cache
*s
)
5113 #ifdef CONFIG_MEMCG_KMEM
5114 if (!is_root_cache(s
))
5115 return s
->memcg_params
->root_cache
->memcg_kset
;
5120 #define ID_STR_LENGTH 64
5122 /* Create a unique string id for a slab cache:
5124 * Format :[flags-]size
5126 static char *create_unique_id(struct kmem_cache
*s
)
5128 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5135 * First flags affecting slabcache operations. We will only
5136 * get here for aliasable slabs so we do not need to support
5137 * too many flags. The flags here must cover all flags that
5138 * are matched during merging to guarantee that the id is
5141 if (s
->flags
& SLAB_CACHE_DMA
)
5143 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5145 if (s
->flags
& SLAB_DEBUG_FREE
)
5147 if (!(s
->flags
& SLAB_NOTRACK
))
5151 p
+= sprintf(p
, "%07d", s
->size
);
5153 #ifdef CONFIG_MEMCG_KMEM
5154 if (!is_root_cache(s
))
5155 p
+= sprintf(p
, "-%08d",
5156 memcg_cache_id(s
->memcg_params
->memcg
));
5159 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5163 static int sysfs_slab_add(struct kmem_cache
*s
)
5167 int unmergeable
= slab_unmergeable(s
);
5171 * Slabcache can never be merged so we can use the name proper.
5172 * This is typically the case for debug situations. In that
5173 * case we can catch duplicate names easily.
5175 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5179 * Create a unique name for the slab as a target
5182 name
= create_unique_id(s
);
5185 s
->kobj
.kset
= cache_kset(s
);
5186 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, "%s", name
);
5190 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5194 #ifdef CONFIG_MEMCG_KMEM
5195 if (is_root_cache(s
)) {
5196 s
->memcg_kset
= kset_create_and_add("cgroup", NULL
, &s
->kobj
);
5197 if (!s
->memcg_kset
) {
5204 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5206 /* Setup first alias */
5207 sysfs_slab_alias(s
, s
->name
);
5214 kobject_del(&s
->kobj
);
5216 kobject_put(&s
->kobj
);
5220 void sysfs_slab_remove(struct kmem_cache
*s
)
5222 if (slab_state
< FULL
)
5224 * Sysfs has not been setup yet so no need to remove the
5229 #ifdef CONFIG_MEMCG_KMEM
5230 kset_unregister(s
->memcg_kset
);
5232 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5233 kobject_del(&s
->kobj
);
5234 kobject_put(&s
->kobj
);
5238 * Need to buffer aliases during bootup until sysfs becomes
5239 * available lest we lose that information.
5241 struct saved_alias
{
5242 struct kmem_cache
*s
;
5244 struct saved_alias
*next
;
5247 static struct saved_alias
*alias_list
;
5249 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5251 struct saved_alias
*al
;
5253 if (slab_state
== FULL
) {
5255 * If we have a leftover link then remove it.
5257 sysfs_remove_link(&slab_kset
->kobj
, name
);
5258 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5261 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5267 al
->next
= alias_list
;
5272 static int __init
slab_sysfs_init(void)
5274 struct kmem_cache
*s
;
5277 mutex_lock(&slab_mutex
);
5279 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5281 mutex_unlock(&slab_mutex
);
5282 pr_err("Cannot register slab subsystem.\n");
5288 list_for_each_entry(s
, &slab_caches
, list
) {
5289 err
= sysfs_slab_add(s
);
5291 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5295 while (alias_list
) {
5296 struct saved_alias
*al
= alias_list
;
5298 alias_list
= alias_list
->next
;
5299 err
= sysfs_slab_alias(al
->s
, al
->name
);
5301 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5306 mutex_unlock(&slab_mutex
);
5311 __initcall(slab_sysfs_init
);
5312 #endif /* CONFIG_SYSFS */
5315 * The /proc/slabinfo ABI
5317 #ifdef CONFIG_SLABINFO
5318 void get_slabinfo(struct kmem_cache
*s
, struct slabinfo
*sinfo
)
5320 unsigned long nr_slabs
= 0;
5321 unsigned long nr_objs
= 0;
5322 unsigned long nr_free
= 0;
5324 struct kmem_cache_node
*n
;
5326 for_each_kmem_cache_node(s
, node
, n
) {
5327 nr_slabs
+= node_nr_slabs(n
);
5328 nr_objs
+= node_nr_objs(n
);
5329 nr_free
+= count_partial(n
, count_free
);
5332 sinfo
->active_objs
= nr_objs
- nr_free
;
5333 sinfo
->num_objs
= nr_objs
;
5334 sinfo
->active_slabs
= nr_slabs
;
5335 sinfo
->num_slabs
= nr_slabs
;
5336 sinfo
->objects_per_slab
= oo_objects(s
->oo
);
5337 sinfo
->cache_order
= oo_order(s
->oo
);
5340 void slabinfo_show_stats(struct seq_file
*m
, struct kmem_cache
*s
)
5344 ssize_t
slabinfo_write(struct file
*file
, const char __user
*buffer
,
5345 size_t count
, loff_t
*ppos
)
5349 #endif /* CONFIG_SLABINFO */