2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/kmemcheck.h>
22 #include <linux/cpu.h>
23 #include <linux/cpuset.h>
24 #include <linux/mempolicy.h>
25 #include <linux/ctype.h>
26 #include <linux/debugobjects.h>
27 #include <linux/kallsyms.h>
28 #include <linux/memory.h>
29 #include <linux/math64.h>
30 #include <linux/fault-inject.h>
31 #include <linux/stacktrace.h>
33 #include <trace/events/kmem.h>
37 * 1. slub_lock (Global Semaphore)
39 * 3. slab_lock(page) (Only on some arches and for debugging)
43 * The role of the slub_lock is to protect the list of all the slabs
44 * and to synchronize major metadata changes to slab cache structures.
46 * The slab_lock is only used for debugging and on arches that do not
47 * have the ability to do a cmpxchg_double. It only protects the second
48 * double word in the page struct. Meaning
49 * A. page->freelist -> List of object free in a page
50 * B. page->counters -> Counters of objects
51 * C. page->frozen -> frozen state
53 * If a slab is frozen then it is exempt from list management. It is not
54 * on any list. The processor that froze the slab is the one who can
55 * perform list operations on the page. Other processors may put objects
56 * onto the freelist but the processor that froze the slab is the only
57 * one that can retrieve the objects from the page's freelist.
59 * The list_lock protects the partial and full list on each node and
60 * the partial slab counter. If taken then no new slabs may be added or
61 * removed from the lists nor make the number of partial slabs be modified.
62 * (Note that the total number of slabs is an atomic value that may be
63 * modified without taking the list lock).
65 * The list_lock is a centralized lock and thus we avoid taking it as
66 * much as possible. As long as SLUB does not have to handle partial
67 * slabs, operations can continue without any centralized lock. F.e.
68 * allocating a long series of objects that fill up slabs does not require
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache
*s
)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
134 /* Enable to log cmpxchg failures */
135 #undef SLUB_DEBUG_CMPXCHG
138 * Mininum number of partial slabs. These will be left on the partial
139 * lists even if they are empty. kmem_cache_shrink may reclaim them.
141 #define MIN_PARTIAL 5
144 * Maximum number of desirable partial slabs.
145 * The existence of more partial slabs makes kmem_cache_shrink
146 * sort the partial list by the number of objects in the.
148 #define MAX_PARTIAL 10
150 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
151 SLAB_POISON | SLAB_STORE_USER)
154 * Debugging flags that require metadata to be stored in the slab. These get
155 * disabled when slub_debug=O is used and a cache's min order increases with
158 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
161 * Set of flags that will prevent slab merging
163 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
164 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
167 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
168 SLAB_CACHE_DMA | SLAB_NOTRACK)
171 #define OO_MASK ((1 << OO_SHIFT) - 1)
172 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
174 /* Internal SLUB flags */
175 #define __OBJECT_POISON 0x80000000UL /* Poison object */
176 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
178 static int kmem_size
= sizeof(struct kmem_cache
);
181 static struct notifier_block slab_notifier
;
185 DOWN
, /* No slab functionality available */
186 PARTIAL
, /* Kmem_cache_node works */
187 UP
, /* Everything works but does not show up in sysfs */
191 /* A list of all slab caches on the system */
192 static DECLARE_RWSEM(slub_lock
);
193 static LIST_HEAD(slab_caches
);
196 * Tracking user of a slab.
198 #define TRACK_ADDRS_COUNT 16
200 unsigned long addr
; /* Called from address */
201 #ifdef CONFIG_STACKTRACE
202 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
204 int cpu
; /* Was running on cpu */
205 int pid
; /* Pid context */
206 unsigned long when
; /* When did the operation occur */
209 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
212 static int sysfs_slab_add(struct kmem_cache
*);
213 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
214 static void sysfs_slab_remove(struct kmem_cache
*);
217 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
218 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
220 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
228 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
230 #ifdef CONFIG_SLUB_STATS
231 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
235 /********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
239 int slab_is_available(void)
241 return slab_state
>= UP
;
244 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
246 return s
->node
[node
];
249 /* Verify that a pointer has an address that is valid within a slab page */
250 static inline int check_valid_pointer(struct kmem_cache
*s
,
251 struct page
*page
, const void *object
)
258 base
= page_address(page
);
259 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
260 (object
- base
) % s
->size
) {
267 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
269 return *(void **)(object
+ s
->offset
);
272 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
276 #ifdef CONFIG_DEBUG_PAGEALLOC
277 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
279 p
= get_freepointer(s
, object
);
284 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
286 *(void **)(object
+ s
->offset
) = fp
;
289 /* Loop over all objects in a slab */
290 #define for_each_object(__p, __s, __addr, __objects) \
291 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
294 /* Determine object index from a given position */
295 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
297 return (p
- addr
) / s
->size
;
300 static inline size_t slab_ksize(const struct kmem_cache
*s
)
302 #ifdef CONFIG_SLUB_DEBUG
304 * Debugging requires use of the padding between object
305 * and whatever may come after it.
307 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
312 * If we have the need to store the freelist pointer
313 * back there or track user information then we can
314 * only use the space before that information.
316 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
319 * Else we can use all the padding etc for the allocation
324 static inline int order_objects(int order
, unsigned long size
, int reserved
)
326 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
329 static inline struct kmem_cache_order_objects
oo_make(int order
,
330 unsigned long size
, int reserved
)
332 struct kmem_cache_order_objects x
= {
333 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
339 static inline int oo_order(struct kmem_cache_order_objects x
)
341 return x
.x
>> OO_SHIFT
;
344 static inline int oo_objects(struct kmem_cache_order_objects x
)
346 return x
.x
& OO_MASK
;
350 * Per slab locking using the pagelock
352 static __always_inline
void slab_lock(struct page
*page
)
354 bit_spin_lock(PG_locked
, &page
->flags
);
357 static __always_inline
void slab_unlock(struct page
*page
)
359 __bit_spin_unlock(PG_locked
, &page
->flags
);
362 /* Interrupts must be disabled (for the fallback code to work right) */
363 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
364 void *freelist_old
, unsigned long counters_old
,
365 void *freelist_new
, unsigned long counters_new
,
368 VM_BUG_ON(!irqs_disabled());
369 #ifdef CONFIG_CMPXCHG_DOUBLE
370 if (s
->flags
& __CMPXCHG_DOUBLE
) {
371 if (cmpxchg_double(&page
->freelist
,
372 freelist_old
, counters_old
,
373 freelist_new
, counters_new
))
379 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
380 page
->freelist
= freelist_new
;
381 page
->counters
= counters_new
;
389 stat(s
, CMPXCHG_DOUBLE_FAIL
);
391 #ifdef SLUB_DEBUG_CMPXCHG
392 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
398 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
399 void *freelist_old
, unsigned long counters_old
,
400 void *freelist_new
, unsigned long counters_new
,
403 #ifdef CONFIG_CMPXCHG_DOUBLE
404 if (s
->flags
& __CMPXCHG_DOUBLE
) {
405 if (cmpxchg_double(&page
->freelist
,
406 freelist_old
, counters_old
,
407 freelist_new
, counters_new
))
414 local_irq_save(flags
);
416 if (page
->freelist
== freelist_old
&& page
->counters
== counters_old
) {
417 page
->freelist
= freelist_new
;
418 page
->counters
= counters_new
;
420 local_irq_restore(flags
);
424 local_irq_restore(flags
);
428 stat(s
, CMPXCHG_DOUBLE_FAIL
);
430 #ifdef SLUB_DEBUG_CMPXCHG
431 printk(KERN_INFO
"%s %s: cmpxchg double redo ", n
, s
->name
);
437 #ifdef CONFIG_SLUB_DEBUG
439 * Determine a map of object in use on a page.
441 * Node listlock must be held to guarantee that the page does
442 * not vanish from under us.
444 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
447 void *addr
= page_address(page
);
449 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
450 set_bit(slab_index(p
, s
, addr
), map
);
456 #ifdef CONFIG_SLUB_DEBUG_ON
457 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
459 static int slub_debug
;
462 static char *slub_debug_slabs
;
463 static int disable_higher_order_debug
;
468 static void print_section(char *text
, u8
*addr
, unsigned int length
)
470 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
474 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
475 enum track_item alloc
)
480 p
= object
+ s
->offset
+ sizeof(void *);
482 p
= object
+ s
->inuse
;
487 static void set_track(struct kmem_cache
*s
, void *object
,
488 enum track_item alloc
, unsigned long addr
)
490 struct track
*p
= get_track(s
, object
, alloc
);
493 #ifdef CONFIG_STACKTRACE
494 struct stack_trace trace
;
497 trace
.nr_entries
= 0;
498 trace
.max_entries
= TRACK_ADDRS_COUNT
;
499 trace
.entries
= p
->addrs
;
501 save_stack_trace(&trace
);
503 /* See rant in lockdep.c */
504 if (trace
.nr_entries
!= 0 &&
505 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
508 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
512 p
->cpu
= smp_processor_id();
513 p
->pid
= current
->pid
;
516 memset(p
, 0, sizeof(struct track
));
519 static void init_tracking(struct kmem_cache
*s
, void *object
)
521 if (!(s
->flags
& SLAB_STORE_USER
))
524 set_track(s
, object
, TRACK_FREE
, 0UL);
525 set_track(s
, object
, TRACK_ALLOC
, 0UL);
528 static void print_track(const char *s
, struct track
*t
)
533 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
534 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
535 #ifdef CONFIG_STACKTRACE
538 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
540 printk(KERN_ERR
"\t%pS\n", (void *)t
->addrs
[i
]);
547 static void print_tracking(struct kmem_cache
*s
, void *object
)
549 if (!(s
->flags
& SLAB_STORE_USER
))
552 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
553 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
556 static void print_page_info(struct page
*page
)
558 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
559 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
563 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
569 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
571 printk(KERN_ERR
"========================================"
572 "=====================================\n");
573 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
574 printk(KERN_ERR
"----------------------------------------"
575 "-------------------------------------\n\n");
578 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
584 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
586 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
589 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
591 unsigned int off
; /* Offset of last byte */
592 u8
*addr
= page_address(page
);
594 print_tracking(s
, p
);
596 print_page_info(page
);
598 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
599 p
, p
- addr
, get_freepointer(s
, p
));
602 print_section("Bytes b4 ", p
- 16, 16);
604 print_section("Object ", p
, min_t(unsigned long, s
->objsize
,
606 if (s
->flags
& SLAB_RED_ZONE
)
607 print_section("Redzone ", p
+ s
->objsize
,
608 s
->inuse
- s
->objsize
);
611 off
= s
->offset
+ sizeof(void *);
615 if (s
->flags
& SLAB_STORE_USER
)
616 off
+= 2 * sizeof(struct track
);
619 /* Beginning of the filler is the free pointer */
620 print_section("Padding ", p
+ off
, s
->size
- off
);
625 static void object_err(struct kmem_cache
*s
, struct page
*page
,
626 u8
*object
, char *reason
)
628 slab_bug(s
, "%s", reason
);
629 print_trailer(s
, page
, object
);
632 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
638 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
640 slab_bug(s
, "%s", buf
);
641 print_page_info(page
);
645 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
649 if (s
->flags
& __OBJECT_POISON
) {
650 memset(p
, POISON_FREE
, s
->objsize
- 1);
651 p
[s
->objsize
- 1] = POISON_END
;
654 if (s
->flags
& SLAB_RED_ZONE
)
655 memset(p
+ s
->objsize
, val
, s
->inuse
- s
->objsize
);
658 static u8
*check_bytes8(u8
*start
, u8 value
, unsigned int bytes
)
669 static u8
*check_bytes(u8
*start
, u8 value
, unsigned int bytes
)
672 unsigned int words
, prefix
;
675 return check_bytes8(start
, value
, bytes
);
677 value64
= value
| value
<< 8 | value
<< 16 | value
<< 24;
678 value64
= (value64
& 0xffffffff) | value64
<< 32;
679 prefix
= 8 - ((unsigned long)start
) % 8;
682 u8
*r
= check_bytes8(start
, value
, prefix
);
692 if (*(u64
*)start
!= value64
)
693 return check_bytes8(start
, value
, 8);
698 return check_bytes8(start
, value
, bytes
% 8);
701 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
702 void *from
, void *to
)
704 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
705 memset(from
, data
, to
- from
);
708 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
709 u8
*object
, char *what
,
710 u8
*start
, unsigned int value
, unsigned int bytes
)
715 fault
= check_bytes(start
, value
, bytes
);
720 while (end
> fault
&& end
[-1] == value
)
723 slab_bug(s
, "%s overwritten", what
);
724 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
725 fault
, end
- 1, fault
[0], value
);
726 print_trailer(s
, page
, object
);
728 restore_bytes(s
, what
, value
, fault
, end
);
736 * Bytes of the object to be managed.
737 * If the freepointer may overlay the object then the free
738 * pointer is the first word of the object.
740 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
743 * object + s->objsize
744 * Padding to reach word boundary. This is also used for Redzoning.
745 * Padding is extended by another word if Redzoning is enabled and
748 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
749 * 0xcc (RED_ACTIVE) for objects in use.
752 * Meta data starts here.
754 * A. Free pointer (if we cannot overwrite object on free)
755 * B. Tracking data for SLAB_STORE_USER
756 * C. Padding to reach required alignment boundary or at mininum
757 * one word if debugging is on to be able to detect writes
758 * before the word boundary.
760 * Padding is done using 0x5a (POISON_INUSE)
763 * Nothing is used beyond s->size.
765 * If slabcaches are merged then the objsize and inuse boundaries are mostly
766 * ignored. And therefore no slab options that rely on these boundaries
767 * may be used with merged slabcaches.
770 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
772 unsigned long off
= s
->inuse
; /* The end of info */
775 /* Freepointer is placed after the object. */
776 off
+= sizeof(void *);
778 if (s
->flags
& SLAB_STORE_USER
)
779 /* We also have user information there */
780 off
+= 2 * sizeof(struct track
);
785 return check_bytes_and_report(s
, page
, p
, "Object padding",
786 p
+ off
, POISON_INUSE
, s
->size
- off
);
789 /* Check the pad bytes at the end of a slab page */
790 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
798 if (!(s
->flags
& SLAB_POISON
))
801 start
= page_address(page
);
802 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
803 end
= start
+ length
;
804 remainder
= length
% s
->size
;
808 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
811 while (end
> fault
&& end
[-1] == POISON_INUSE
)
814 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
815 print_section("Padding ", end
- remainder
, remainder
);
817 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
821 static int check_object(struct kmem_cache
*s
, struct page
*page
,
822 void *object
, u8 val
)
825 u8
*endobject
= object
+ s
->objsize
;
827 if (s
->flags
& SLAB_RED_ZONE
) {
828 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
829 endobject
, val
, s
->inuse
- s
->objsize
))
832 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
833 check_bytes_and_report(s
, page
, p
, "Alignment padding",
834 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
838 if (s
->flags
& SLAB_POISON
) {
839 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
840 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
841 POISON_FREE
, s
->objsize
- 1) ||
842 !check_bytes_and_report(s
, page
, p
, "Poison",
843 p
+ s
->objsize
- 1, POISON_END
, 1)))
846 * check_pad_bytes cleans up on its own.
848 check_pad_bytes(s
, page
, p
);
851 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
853 * Object and freepointer overlap. Cannot check
854 * freepointer while object is allocated.
858 /* Check free pointer validity */
859 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
860 object_err(s
, page
, p
, "Freepointer corrupt");
862 * No choice but to zap it and thus lose the remainder
863 * of the free objects in this slab. May cause
864 * another error because the object count is now wrong.
866 set_freepointer(s
, p
, NULL
);
872 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
876 VM_BUG_ON(!irqs_disabled());
878 if (!PageSlab(page
)) {
879 slab_err(s
, page
, "Not a valid slab page");
883 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
884 if (page
->objects
> maxobj
) {
885 slab_err(s
, page
, "objects %u > max %u",
886 s
->name
, page
->objects
, maxobj
);
889 if (page
->inuse
> page
->objects
) {
890 slab_err(s
, page
, "inuse %u > max %u",
891 s
->name
, page
->inuse
, page
->objects
);
894 /* Slab_pad_check fixes things up after itself */
895 slab_pad_check(s
, page
);
900 * Determine if a certain object on a page is on the freelist. Must hold the
901 * slab lock to guarantee that the chains are in a consistent state.
903 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
908 unsigned long max_objects
;
911 while (fp
&& nr
<= page
->objects
) {
914 if (!check_valid_pointer(s
, page
, fp
)) {
916 object_err(s
, page
, object
,
917 "Freechain corrupt");
918 set_freepointer(s
, object
, NULL
);
921 slab_err(s
, page
, "Freepointer corrupt");
922 page
->freelist
= NULL
;
923 page
->inuse
= page
->objects
;
924 slab_fix(s
, "Freelist cleared");
930 fp
= get_freepointer(s
, object
);
934 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
935 if (max_objects
> MAX_OBJS_PER_PAGE
)
936 max_objects
= MAX_OBJS_PER_PAGE
;
938 if (page
->objects
!= max_objects
) {
939 slab_err(s
, page
, "Wrong number of objects. Found %d but "
940 "should be %d", page
->objects
, max_objects
);
941 page
->objects
= max_objects
;
942 slab_fix(s
, "Number of objects adjusted.");
944 if (page
->inuse
!= page
->objects
- nr
) {
945 slab_err(s
, page
, "Wrong object count. Counter is %d but "
946 "counted were %d", page
->inuse
, page
->objects
- nr
);
947 page
->inuse
= page
->objects
- nr
;
948 slab_fix(s
, "Object count adjusted.");
950 return search
== NULL
;
953 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
956 if (s
->flags
& SLAB_TRACE
) {
957 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
959 alloc
? "alloc" : "free",
964 print_section("Object ", (void *)object
, s
->objsize
);
971 * Hooks for other subsystems that check memory allocations. In a typical
972 * production configuration these hooks all should produce no code at all.
974 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
976 flags
&= gfp_allowed_mask
;
977 lockdep_trace_alloc(flags
);
978 might_sleep_if(flags
& __GFP_WAIT
);
980 return should_failslab(s
->objsize
, flags
, s
->flags
);
983 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
, void *object
)
985 flags
&= gfp_allowed_mask
;
986 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
987 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, flags
);
990 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
992 kmemleak_free_recursive(x
, s
->flags
);
995 * Trouble is that we may no longer disable interupts in the fast path
996 * So in order to make the debug calls that expect irqs to be
997 * disabled we need to disable interrupts temporarily.
999 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1001 unsigned long flags
;
1003 local_irq_save(flags
);
1004 kmemcheck_slab_free(s
, x
, s
->objsize
);
1005 debug_check_no_locks_freed(x
, s
->objsize
);
1006 local_irq_restore(flags
);
1009 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1010 debug_check_no_obj_freed(x
, s
->objsize
);
1014 * Tracking of fully allocated slabs for debugging purposes.
1016 * list_lock must be held.
1018 static void add_full(struct kmem_cache
*s
,
1019 struct kmem_cache_node
*n
, struct page
*page
)
1021 if (!(s
->flags
& SLAB_STORE_USER
))
1024 list_add(&page
->lru
, &n
->full
);
1028 * list_lock must be held.
1030 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
1032 if (!(s
->flags
& SLAB_STORE_USER
))
1035 list_del(&page
->lru
);
1038 /* Tracking of the number of slabs for debugging purposes */
1039 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1041 struct kmem_cache_node
*n
= get_node(s
, node
);
1043 return atomic_long_read(&n
->nr_slabs
);
1046 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1048 return atomic_long_read(&n
->nr_slabs
);
1051 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1053 struct kmem_cache_node
*n
= get_node(s
, node
);
1056 * May be called early in order to allocate a slab for the
1057 * kmem_cache_node structure. Solve the chicken-egg
1058 * dilemma by deferring the increment of the count during
1059 * bootstrap (see early_kmem_cache_node_alloc).
1062 atomic_long_inc(&n
->nr_slabs
);
1063 atomic_long_add(objects
, &n
->total_objects
);
1066 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
1068 struct kmem_cache_node
*n
= get_node(s
, node
);
1070 atomic_long_dec(&n
->nr_slabs
);
1071 atomic_long_sub(objects
, &n
->total_objects
);
1074 /* Object debug checks for alloc/free paths */
1075 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
1078 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1081 init_object(s
, object
, SLUB_RED_INACTIVE
);
1082 init_tracking(s
, object
);
1085 static noinline
int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
1086 void *object
, unsigned long addr
)
1088 if (!check_slab(s
, page
))
1091 if (!check_valid_pointer(s
, page
, object
)) {
1092 object_err(s
, page
, object
, "Freelist Pointer check fails");
1096 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1099 /* Success perform special debug activities for allocs */
1100 if (s
->flags
& SLAB_STORE_USER
)
1101 set_track(s
, object
, TRACK_ALLOC
, addr
);
1102 trace(s
, page
, object
, 1);
1103 init_object(s
, object
, SLUB_RED_ACTIVE
);
1107 if (PageSlab(page
)) {
1109 * If this is a slab page then lets do the best we can
1110 * to avoid issues in the future. Marking all objects
1111 * as used avoids touching the remaining objects.
1113 slab_fix(s
, "Marking all objects used");
1114 page
->inuse
= page
->objects
;
1115 page
->freelist
= NULL
;
1120 static noinline
int free_debug_processing(struct kmem_cache
*s
,
1121 struct page
*page
, void *object
, unsigned long addr
)
1123 unsigned long flags
;
1126 local_irq_save(flags
);
1129 if (!check_slab(s
, page
))
1132 if (!check_valid_pointer(s
, page
, object
)) {
1133 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1137 if (on_freelist(s
, page
, object
)) {
1138 object_err(s
, page
, object
, "Object already free");
1142 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1145 if (unlikely(s
!= page
->slab
)) {
1146 if (!PageSlab(page
)) {
1147 slab_err(s
, page
, "Attempt to free object(0x%p) "
1148 "outside of slab", object
);
1149 } else if (!page
->slab
) {
1151 "SLUB <none>: no slab for object 0x%p.\n",
1155 object_err(s
, page
, object
,
1156 "page slab pointer corrupt.");
1160 if (s
->flags
& SLAB_STORE_USER
)
1161 set_track(s
, object
, TRACK_FREE
, addr
);
1162 trace(s
, page
, object
, 0);
1163 init_object(s
, object
, SLUB_RED_INACTIVE
);
1167 local_irq_restore(flags
);
1171 slab_fix(s
, "Object at 0x%p not freed", object
);
1175 static int __init
setup_slub_debug(char *str
)
1177 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1178 if (*str
++ != '=' || !*str
)
1180 * No options specified. Switch on full debugging.
1186 * No options but restriction on slabs. This means full
1187 * debugging for slabs matching a pattern.
1191 if (tolower(*str
) == 'o') {
1193 * Avoid enabling debugging on caches if its minimum order
1194 * would increase as a result.
1196 disable_higher_order_debug
= 1;
1203 * Switch off all debugging measures.
1208 * Determine which debug features should be switched on
1210 for (; *str
&& *str
!= ','; str
++) {
1211 switch (tolower(*str
)) {
1213 slub_debug
|= SLAB_DEBUG_FREE
;
1216 slub_debug
|= SLAB_RED_ZONE
;
1219 slub_debug
|= SLAB_POISON
;
1222 slub_debug
|= SLAB_STORE_USER
;
1225 slub_debug
|= SLAB_TRACE
;
1228 slub_debug
|= SLAB_FAILSLAB
;
1231 printk(KERN_ERR
"slub_debug option '%c' "
1232 "unknown. skipped\n", *str
);
1238 slub_debug_slabs
= str
+ 1;
1243 __setup("slub_debug", setup_slub_debug
);
1245 static unsigned long kmem_cache_flags(unsigned long objsize
,
1246 unsigned long flags
, const char *name
,
1247 void (*ctor
)(void *))
1250 * Enable debugging if selected on the kernel commandline.
1252 if (slub_debug
&& (!slub_debug_slabs
||
1253 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1254 flags
|= slub_debug
;
1259 static inline void setup_object_debug(struct kmem_cache
*s
,
1260 struct page
*page
, void *object
) {}
1262 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1263 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1265 static inline int free_debug_processing(struct kmem_cache
*s
,
1266 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1268 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1270 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1271 void *object
, u8 val
) { return 1; }
1272 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1273 struct page
*page
) {}
1274 static inline void remove_full(struct kmem_cache
*s
, struct page
*page
) {}
1275 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1276 unsigned long flags
, const char *name
,
1277 void (*ctor
)(void *))
1281 #define slub_debug 0
1283 #define disable_higher_order_debug 0
1285 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1287 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1289 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1291 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1294 static inline int slab_pre_alloc_hook(struct kmem_cache
*s
, gfp_t flags
)
1297 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
1300 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
) {}
1302 #endif /* CONFIG_SLUB_DEBUG */
1305 * Slab allocation and freeing
1307 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1308 struct kmem_cache_order_objects oo
)
1310 int order
= oo_order(oo
);
1312 flags
|= __GFP_NOTRACK
;
1314 if (node
== NUMA_NO_NODE
)
1315 return alloc_pages(flags
, order
);
1317 return alloc_pages_exact_node(node
, flags
, order
);
1320 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1323 struct kmem_cache_order_objects oo
= s
->oo
;
1326 flags
&= gfp_allowed_mask
;
1328 if (flags
& __GFP_WAIT
)
1331 flags
|= s
->allocflags
;
1334 * Let the initial higher-order allocation fail under memory pressure
1335 * so we fall-back to the minimum order allocation.
1337 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1339 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1340 if (unlikely(!page
)) {
1343 * Allocation may have failed due to fragmentation.
1344 * Try a lower order alloc if possible
1346 page
= alloc_slab_page(flags
, node
, oo
);
1349 stat(s
, ORDER_FALLBACK
);
1352 if (flags
& __GFP_WAIT
)
1353 local_irq_disable();
1358 if (kmemcheck_enabled
1359 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1360 int pages
= 1 << oo_order(oo
);
1362 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1365 * Objects from caches that have a constructor don't get
1366 * cleared when they're allocated, so we need to do it here.
1369 kmemcheck_mark_uninitialized_pages(page
, pages
);
1371 kmemcheck_mark_unallocated_pages(page
, pages
);
1374 page
->objects
= oo_objects(oo
);
1375 mod_zone_page_state(page_zone(page
),
1376 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1377 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1383 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1386 setup_object_debug(s
, page
, object
);
1387 if (unlikely(s
->ctor
))
1391 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1398 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1400 page
= allocate_slab(s
,
1401 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1405 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1407 page
->flags
|= 1 << PG_slab
;
1409 start
= page_address(page
);
1411 if (unlikely(s
->flags
& SLAB_POISON
))
1412 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1415 for_each_object(p
, s
, start
, page
->objects
) {
1416 setup_object(s
, page
, last
);
1417 set_freepointer(s
, last
, p
);
1420 setup_object(s
, page
, last
);
1421 set_freepointer(s
, last
, NULL
);
1423 page
->freelist
= start
;
1424 page
->inuse
= page
->objects
;
1430 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1432 int order
= compound_order(page
);
1433 int pages
= 1 << order
;
1435 if (kmem_cache_debug(s
)) {
1438 slab_pad_check(s
, page
);
1439 for_each_object(p
, s
, page_address(page
),
1441 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1444 kmemcheck_free_shadow(page
, compound_order(page
));
1446 mod_zone_page_state(page_zone(page
),
1447 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1448 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1451 __ClearPageSlab(page
);
1452 reset_page_mapcount(page
);
1453 if (current
->reclaim_state
)
1454 current
->reclaim_state
->reclaimed_slab
+= pages
;
1455 __free_pages(page
, order
);
1458 #define need_reserve_slab_rcu \
1459 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1461 static void rcu_free_slab(struct rcu_head
*h
)
1465 if (need_reserve_slab_rcu
)
1466 page
= virt_to_head_page(h
);
1468 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1470 __free_slab(page
->slab
, page
);
1473 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1475 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1476 struct rcu_head
*head
;
1478 if (need_reserve_slab_rcu
) {
1479 int order
= compound_order(page
);
1480 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1482 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1483 head
= page_address(page
) + offset
;
1486 * RCU free overloads the RCU head over the LRU
1488 head
= (void *)&page
->lru
;
1491 call_rcu(head
, rcu_free_slab
);
1493 __free_slab(s
, page
);
1496 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1498 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1503 * Management of partially allocated slabs.
1505 * list_lock must be held.
1507 static inline void add_partial(struct kmem_cache_node
*n
,
1508 struct page
*page
, int tail
)
1511 if (tail
== DEACTIVATE_TO_TAIL
)
1512 list_add_tail(&page
->lru
, &n
->partial
);
1514 list_add(&page
->lru
, &n
->partial
);
1518 * list_lock must be held.
1520 static inline void remove_partial(struct kmem_cache_node
*n
,
1523 list_del(&page
->lru
);
1528 * Lock slab, remove from the partial list and put the object into the
1531 * Returns a list of objects or NULL if it fails.
1533 * Must hold list_lock.
1535 static inline void *acquire_slab(struct kmem_cache
*s
,
1536 struct kmem_cache_node
*n
, struct page
*page
,
1540 unsigned long counters
;
1544 * Zap the freelist and set the frozen bit.
1545 * The old freelist is the list of objects for the
1546 * per cpu allocation list.
1549 freelist
= page
->freelist
;
1550 counters
= page
->counters
;
1551 new.counters
= counters
;
1553 new.inuse
= page
->objects
;
1555 VM_BUG_ON(new.frozen
);
1558 } while (!__cmpxchg_double_slab(s
, page
,
1561 "lock and freeze"));
1563 remove_partial(n
, page
);
1567 static int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1570 * Try to allocate a partial slab from a specific node.
1572 static void *get_partial_node(struct kmem_cache
*s
,
1573 struct kmem_cache_node
*n
, struct kmem_cache_cpu
*c
)
1575 struct page
*page
, *page2
;
1576 void *object
= NULL
;
1579 * Racy check. If we mistakenly see no partial slabs then we
1580 * just allocate an empty slab. If we mistakenly try to get a
1581 * partial slab and there is none available then get_partials()
1584 if (!n
|| !n
->nr_partial
)
1587 spin_lock(&n
->list_lock
);
1588 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1589 void *t
= acquire_slab(s
, n
, page
, object
== NULL
);
1597 c
->node
= page_to_nid(page
);
1598 stat(s
, ALLOC_FROM_PARTIAL
);
1600 available
= page
->objects
- page
->inuse
;
1603 available
= put_cpu_partial(s
, page
, 0);
1605 if (kmem_cache_debug(s
) || available
> s
->cpu_partial
/ 2)
1609 spin_unlock(&n
->list_lock
);
1614 * Get a page from somewhere. Search in increasing NUMA distances.
1616 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1617 struct kmem_cache_cpu
*c
)
1620 struct zonelist
*zonelist
;
1623 enum zone_type high_zoneidx
= gfp_zone(flags
);
1627 * The defrag ratio allows a configuration of the tradeoffs between
1628 * inter node defragmentation and node local allocations. A lower
1629 * defrag_ratio increases the tendency to do local allocations
1630 * instead of attempting to obtain partial slabs from other nodes.
1632 * If the defrag_ratio is set to 0 then kmalloc() always
1633 * returns node local objects. If the ratio is higher then kmalloc()
1634 * may return off node objects because partial slabs are obtained
1635 * from other nodes and filled up.
1637 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1638 * defrag_ratio = 1000) then every (well almost) allocation will
1639 * first attempt to defrag slab caches on other nodes. This means
1640 * scanning over all nodes to look for partial slabs which may be
1641 * expensive if we do it every time we are trying to find a slab
1642 * with available objects.
1644 if (!s
->remote_node_defrag_ratio
||
1645 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1649 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1650 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1651 struct kmem_cache_node
*n
;
1653 n
= get_node(s
, zone_to_nid(zone
));
1655 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1656 n
->nr_partial
> s
->min_partial
) {
1657 object
= get_partial_node(s
, n
, c
);
1670 * Get a partial page, lock it and return it.
1672 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1673 struct kmem_cache_cpu
*c
)
1676 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1678 object
= get_partial_node(s
, get_node(s
, searchnode
), c
);
1679 if (object
|| node
!= NUMA_NO_NODE
)
1682 return get_any_partial(s
, flags
, c
);
1685 #ifdef CONFIG_PREEMPT
1687 * Calculate the next globally unique transaction for disambiguiation
1688 * during cmpxchg. The transactions start with the cpu number and are then
1689 * incremented by CONFIG_NR_CPUS.
1691 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1694 * No preemption supported therefore also no need to check for
1700 static inline unsigned long next_tid(unsigned long tid
)
1702 return tid
+ TID_STEP
;
1705 static inline unsigned int tid_to_cpu(unsigned long tid
)
1707 return tid
% TID_STEP
;
1710 static inline unsigned long tid_to_event(unsigned long tid
)
1712 return tid
/ TID_STEP
;
1715 static inline unsigned int init_tid(int cpu
)
1720 static inline void note_cmpxchg_failure(const char *n
,
1721 const struct kmem_cache
*s
, unsigned long tid
)
1723 #ifdef SLUB_DEBUG_CMPXCHG
1724 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1726 printk(KERN_INFO
"%s %s: cmpxchg redo ", n
, s
->name
);
1728 #ifdef CONFIG_PREEMPT
1729 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1730 printk("due to cpu change %d -> %d\n",
1731 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1734 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1735 printk("due to cpu running other code. Event %ld->%ld\n",
1736 tid_to_event(tid
), tid_to_event(actual_tid
));
1738 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1739 actual_tid
, tid
, next_tid(tid
));
1741 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1744 void init_kmem_cache_cpus(struct kmem_cache
*s
)
1748 for_each_possible_cpu(cpu
)
1749 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1753 * Remove the cpu slab
1755 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1757 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1758 struct page
*page
= c
->page
;
1759 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1761 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1764 int tail
= DEACTIVATE_TO_HEAD
;
1768 if (page
->freelist
) {
1769 stat(s
, DEACTIVATE_REMOTE_FREES
);
1770 tail
= DEACTIVATE_TO_TAIL
;
1773 c
->tid
= next_tid(c
->tid
);
1775 freelist
= c
->freelist
;
1779 * Stage one: Free all available per cpu objects back
1780 * to the page freelist while it is still frozen. Leave the
1783 * There is no need to take the list->lock because the page
1786 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1788 unsigned long counters
;
1791 prior
= page
->freelist
;
1792 counters
= page
->counters
;
1793 set_freepointer(s
, freelist
, prior
);
1794 new.counters
= counters
;
1796 VM_BUG_ON(!new.frozen
);
1798 } while (!__cmpxchg_double_slab(s
, page
,
1800 freelist
, new.counters
,
1801 "drain percpu freelist"));
1803 freelist
= nextfree
;
1807 * Stage two: Ensure that the page is unfrozen while the
1808 * list presence reflects the actual number of objects
1811 * We setup the list membership and then perform a cmpxchg
1812 * with the count. If there is a mismatch then the page
1813 * is not unfrozen but the page is on the wrong list.
1815 * Then we restart the process which may have to remove
1816 * the page from the list that we just put it on again
1817 * because the number of objects in the slab may have
1822 old
.freelist
= page
->freelist
;
1823 old
.counters
= page
->counters
;
1824 VM_BUG_ON(!old
.frozen
);
1826 /* Determine target state of the slab */
1827 new.counters
= old
.counters
;
1830 set_freepointer(s
, freelist
, old
.freelist
);
1831 new.freelist
= freelist
;
1833 new.freelist
= old
.freelist
;
1837 if (!new.inuse
&& n
->nr_partial
> s
->min_partial
)
1839 else if (new.freelist
) {
1844 * Taking the spinlock removes the possiblity
1845 * that acquire_slab() will see a slab page that
1848 spin_lock(&n
->list_lock
);
1852 if (kmem_cache_debug(s
) && !lock
) {
1855 * This also ensures that the scanning of full
1856 * slabs from diagnostic functions will not see
1859 spin_lock(&n
->list_lock
);
1867 remove_partial(n
, page
);
1869 else if (l
== M_FULL
)
1871 remove_full(s
, page
);
1873 if (m
== M_PARTIAL
) {
1875 add_partial(n
, page
, tail
);
1878 } else if (m
== M_FULL
) {
1880 stat(s
, DEACTIVATE_FULL
);
1881 add_full(s
, n
, page
);
1887 if (!__cmpxchg_double_slab(s
, page
,
1888 old
.freelist
, old
.counters
,
1889 new.freelist
, new.counters
,
1894 spin_unlock(&n
->list_lock
);
1897 stat(s
, DEACTIVATE_EMPTY
);
1898 discard_slab(s
, page
);
1903 /* Unfreeze all the cpu partial slabs */
1904 static void unfreeze_partials(struct kmem_cache
*s
)
1906 struct kmem_cache_node
*n
= NULL
;
1907 struct kmem_cache_cpu
*c
= this_cpu_ptr(s
->cpu_slab
);
1910 while ((page
= c
->partial
)) {
1911 enum slab_modes
{ M_PARTIAL
, M_FREE
};
1912 enum slab_modes l
, m
;
1916 c
->partial
= page
->next
;
1921 old
.freelist
= page
->freelist
;
1922 old
.counters
= page
->counters
;
1923 VM_BUG_ON(!old
.frozen
);
1925 new.counters
= old
.counters
;
1926 new.freelist
= old
.freelist
;
1930 if (!new.inuse
&& (!n
|| n
->nr_partial
> s
->min_partial
))
1933 struct kmem_cache_node
*n2
= get_node(s
,
1939 spin_unlock(&n
->list_lock
);
1942 spin_lock(&n
->list_lock
);
1948 remove_partial(n
, page
);
1950 add_partial(n
, page
, 1);
1955 } while (!cmpxchg_double_slab(s
, page
,
1956 old
.freelist
, old
.counters
,
1957 new.freelist
, new.counters
,
1958 "unfreezing slab"));
1961 stat(s
, DEACTIVATE_EMPTY
);
1962 discard_slab(s
, page
);
1968 spin_unlock(&n
->list_lock
);
1972 * Put a page that was just frozen (in __slab_free) into a partial page
1973 * slot if available. This is done without interrupts disabled and without
1974 * preemption disabled. The cmpxchg is racy and may put the partial page
1975 * onto a random cpus partial slot.
1977 * If we did not find a slot then simply move all the partials to the
1978 * per node partial list.
1980 int put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
1982 struct page
*oldpage
;
1989 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
1992 pobjects
= oldpage
->pobjects
;
1993 pages
= oldpage
->pages
;
1994 if (drain
&& pobjects
> s
->cpu_partial
) {
1995 unsigned long flags
;
1997 * partial array is full. Move the existing
1998 * set to the per node partial list.
2000 local_irq_save(flags
);
2001 unfreeze_partials(s
);
2002 local_irq_restore(flags
);
2009 pobjects
+= page
->objects
- page
->inuse
;
2011 page
->pages
= pages
;
2012 page
->pobjects
= pobjects
;
2013 page
->next
= oldpage
;
2015 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
) != oldpage
);
2016 stat(s
, CPU_PARTIAL_FREE
);
2020 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2022 stat(s
, CPUSLAB_FLUSH
);
2023 deactivate_slab(s
, c
);
2029 * Called from IPI handler with interrupts disabled.
2031 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2033 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2039 unfreeze_partials(s
);
2043 static void flush_cpu_slab(void *d
)
2045 struct kmem_cache
*s
= d
;
2047 __flush_cpu_slab(s
, smp_processor_id());
2050 static void flush_all(struct kmem_cache
*s
)
2052 on_each_cpu(flush_cpu_slab
, s
, 1);
2056 * Check if the objects in a per cpu structure fit numa
2057 * locality expectations.
2059 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
2062 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
2068 static int count_free(struct page
*page
)
2070 return page
->objects
- page
->inuse
;
2073 static unsigned long count_partial(struct kmem_cache_node
*n
,
2074 int (*get_count
)(struct page
*))
2076 unsigned long flags
;
2077 unsigned long x
= 0;
2080 spin_lock_irqsave(&n
->list_lock
, flags
);
2081 list_for_each_entry(page
, &n
->partial
, lru
)
2082 x
+= get_count(page
);
2083 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2087 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2089 #ifdef CONFIG_SLUB_DEBUG
2090 return atomic_long_read(&n
->total_objects
);
2096 static noinline
void
2097 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2102 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2104 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
2105 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
2106 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
2108 if (oo_order(s
->min
) > get_order(s
->objsize
))
2109 printk(KERN_WARNING
" %s debugging increased min order, use "
2110 "slub_debug=O to disable.\n", s
->name
);
2112 for_each_online_node(node
) {
2113 struct kmem_cache_node
*n
= get_node(s
, node
);
2114 unsigned long nr_slabs
;
2115 unsigned long nr_objs
;
2116 unsigned long nr_free
;
2121 nr_free
= count_partial(n
, count_free
);
2122 nr_slabs
= node_nr_slabs(n
);
2123 nr_objs
= node_nr_objs(n
);
2126 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
2127 node
, nr_slabs
, nr_objs
, nr_free
);
2131 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2132 int node
, struct kmem_cache_cpu
**pc
)
2135 struct kmem_cache_cpu
*c
;
2136 struct page
*page
= new_slab(s
, flags
, node
);
2139 c
= __this_cpu_ptr(s
->cpu_slab
);
2144 * No other reference to the page yet so we can
2145 * muck around with it freely without cmpxchg
2147 object
= page
->freelist
;
2148 page
->freelist
= NULL
;
2150 stat(s
, ALLOC_SLAB
);
2151 c
->node
= page_to_nid(page
);
2161 * Slow path. The lockless freelist is empty or we need to perform
2164 * Processing is still very fast if new objects have been freed to the
2165 * regular freelist. In that case we simply take over the regular freelist
2166 * as the lockless freelist and zap the regular freelist.
2168 * If that is not working then we fall back to the partial lists. We take the
2169 * first element of the freelist as the object to allocate now and move the
2170 * rest of the freelist to the lockless freelist.
2172 * And if we were unable to get a new slab from the partial slab lists then
2173 * we need to allocate a new slab. This is the slowest path since it involves
2174 * a call to the page allocator and the setup of a new slab.
2176 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2177 unsigned long addr
, struct kmem_cache_cpu
*c
)
2180 unsigned long flags
;
2182 unsigned long counters
;
2184 local_irq_save(flags
);
2185 #ifdef CONFIG_PREEMPT
2187 * We may have been preempted and rescheduled on a different
2188 * cpu before disabling interrupts. Need to reload cpu area
2191 c
= this_cpu_ptr(s
->cpu_slab
);
2197 if (unlikely(!node_match(c
, node
))) {
2198 stat(s
, ALLOC_NODE_MISMATCH
);
2199 deactivate_slab(s
, c
);
2203 stat(s
, ALLOC_SLOWPATH
);
2206 object
= c
->page
->freelist
;
2207 counters
= c
->page
->counters
;
2208 new.counters
= counters
;
2209 VM_BUG_ON(!new.frozen
);
2212 * If there is no object left then we use this loop to
2213 * deactivate the slab which is simple since no objects
2214 * are left in the slab and therefore we do not need to
2215 * put the page back onto the partial list.
2217 * If there are objects left then we retrieve them
2218 * and use them to refill the per cpu queue.
2221 new.inuse
= c
->page
->objects
;
2222 new.frozen
= object
!= NULL
;
2224 } while (!__cmpxchg_double_slab(s
, c
->page
,
2231 stat(s
, DEACTIVATE_BYPASS
);
2235 stat(s
, ALLOC_REFILL
);
2238 c
->freelist
= get_freepointer(s
, object
);
2239 c
->tid
= next_tid(c
->tid
);
2240 local_irq_restore(flags
);
2246 c
->page
= c
->partial
;
2247 c
->partial
= c
->page
->next
;
2248 c
->node
= page_to_nid(c
->page
);
2249 stat(s
, CPU_PARTIAL_ALLOC
);
2254 /* Then do expensive stuff like retrieving pages from the partial lists */
2255 object
= get_partial(s
, gfpflags
, node
, c
);
2257 if (unlikely(!object
)) {
2259 object
= new_slab_objects(s
, gfpflags
, node
, &c
);
2261 if (unlikely(!object
)) {
2262 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
2263 slab_out_of_memory(s
, gfpflags
, node
);
2265 local_irq_restore(flags
);
2270 if (likely(!kmem_cache_debug(s
)))
2273 /* Only entered in the debug case */
2274 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
2275 goto new_slab
; /* Slab failed checks. Next slab needed */
2277 c
->freelist
= get_freepointer(s
, object
);
2278 deactivate_slab(s
, c
);
2279 c
->node
= NUMA_NO_NODE
;
2280 local_irq_restore(flags
);
2285 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2286 * have the fastpath folded into their functions. So no function call
2287 * overhead for requests that can be satisfied on the fastpath.
2289 * The fastpath works by first checking if the lockless freelist can be used.
2290 * If not then __slab_alloc is called for slow processing.
2292 * Otherwise we can simply pick the next object from the lockless free list.
2294 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2295 gfp_t gfpflags
, int node
, unsigned long addr
)
2298 struct kmem_cache_cpu
*c
;
2301 if (slab_pre_alloc_hook(s
, gfpflags
))
2307 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2308 * enabled. We may switch back and forth between cpus while
2309 * reading from one cpu area. That does not matter as long
2310 * as we end up on the original cpu again when doing the cmpxchg.
2312 c
= __this_cpu_ptr(s
->cpu_slab
);
2315 * The transaction ids are globally unique per cpu and per operation on
2316 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2317 * occurs on the right processor and that there was no operation on the
2318 * linked list in between.
2323 object
= c
->freelist
;
2324 if (unlikely(!object
|| !node_match(c
, node
)))
2326 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2330 * The cmpxchg will only match if there was no additional
2331 * operation and if we are on the right processor.
2333 * The cmpxchg does the following atomically (without lock semantics!)
2334 * 1. Relocate first pointer to the current per cpu area.
2335 * 2. Verify that tid and freelist have not been changed
2336 * 3. If they were not changed replace tid and freelist
2338 * Since this is without lock semantics the protection is only against
2339 * code executing on this cpu *not* from access by other cpus.
2341 if (unlikely(!irqsafe_cpu_cmpxchg_double(
2342 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2344 get_freepointer_safe(s
, object
), next_tid(tid
)))) {
2346 note_cmpxchg_failure("slab_alloc", s
, tid
);
2349 stat(s
, ALLOC_FASTPATH
);
2352 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2353 memset(object
, 0, s
->objsize
);
2355 slab_post_alloc_hook(s
, gfpflags
, object
);
2360 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2362 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2364 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
2368 EXPORT_SYMBOL(kmem_cache_alloc
);
2370 #ifdef CONFIG_TRACING
2371 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2373 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
2374 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2377 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2379 void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
2381 void *ret
= kmalloc_order(size
, flags
, order
);
2382 trace_kmalloc(_RET_IP_
, ret
, size
, PAGE_SIZE
<< order
, flags
);
2385 EXPORT_SYMBOL(kmalloc_order_trace
);
2389 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2391 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2393 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2394 s
->objsize
, s
->size
, gfpflags
, node
);
2398 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2400 #ifdef CONFIG_TRACING
2401 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2403 int node
, size_t size
)
2405 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
2407 trace_kmalloc_node(_RET_IP_
, ret
,
2408 size
, s
->size
, gfpflags
, node
);
2411 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2416 * Slow patch handling. This may still be called frequently since objects
2417 * have a longer lifetime than the cpu slabs in most processing loads.
2419 * So we still attempt to reduce cache line usage. Just take the slab
2420 * lock and free the item. If there is no additional partial page
2421 * handling required then we can return immediately.
2423 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2424 void *x
, unsigned long addr
)
2427 void **object
= (void *)x
;
2431 unsigned long counters
;
2432 struct kmem_cache_node
*n
= NULL
;
2433 unsigned long uninitialized_var(flags
);
2435 stat(s
, FREE_SLOWPATH
);
2437 if (kmem_cache_debug(s
) && !free_debug_processing(s
, page
, x
, addr
))
2441 prior
= page
->freelist
;
2442 counters
= page
->counters
;
2443 set_freepointer(s
, object
, prior
);
2444 new.counters
= counters
;
2445 was_frozen
= new.frozen
;
2447 if ((!new.inuse
|| !prior
) && !was_frozen
&& !n
) {
2449 if (!kmem_cache_debug(s
) && !prior
)
2452 * Slab was on no list before and will be partially empty
2453 * We can defer the list move and instead freeze it.
2457 else { /* Needs to be taken off a list */
2459 n
= get_node(s
, page_to_nid(page
));
2461 * Speculatively acquire the list_lock.
2462 * If the cmpxchg does not succeed then we may
2463 * drop the list_lock without any processing.
2465 * Otherwise the list_lock will synchronize with
2466 * other processors updating the list of slabs.
2468 spin_lock_irqsave(&n
->list_lock
, flags
);
2474 } while (!cmpxchg_double_slab(s
, page
,
2476 object
, new.counters
,
2482 * If we just froze the page then put it onto the
2483 * per cpu partial list.
2485 if (new.frozen
&& !was_frozen
)
2486 put_cpu_partial(s
, page
, 1);
2489 * The list lock was not taken therefore no list
2490 * activity can be necessary.
2493 stat(s
, FREE_FROZEN
);
2498 * was_frozen may have been set after we acquired the list_lock in
2499 * an earlier loop. So we need to check it here again.
2502 stat(s
, FREE_FROZEN
);
2504 if (unlikely(!inuse
&& n
->nr_partial
> s
->min_partial
))
2508 * Objects left in the slab. If it was not on the partial list before
2511 if (unlikely(!prior
)) {
2512 remove_full(s
, page
);
2513 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2514 stat(s
, FREE_ADD_PARTIAL
);
2517 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2523 * Slab on the partial list.
2525 remove_partial(n
, page
);
2526 stat(s
, FREE_REMOVE_PARTIAL
);
2528 /* Slab must be on the full list */
2529 remove_full(s
, page
);
2531 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2533 discard_slab(s
, page
);
2537 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2538 * can perform fastpath freeing without additional function calls.
2540 * The fastpath is only possible if we are freeing to the current cpu slab
2541 * of this processor. This typically the case if we have just allocated
2544 * If fastpath is not possible then fall back to __slab_free where we deal
2545 * with all sorts of special processing.
2547 static __always_inline
void slab_free(struct kmem_cache
*s
,
2548 struct page
*page
, void *x
, unsigned long addr
)
2550 void **object
= (void *)x
;
2551 struct kmem_cache_cpu
*c
;
2554 slab_free_hook(s
, x
);
2558 * Determine the currently cpus per cpu slab.
2559 * The cpu may change afterward. However that does not matter since
2560 * data is retrieved via this pointer. If we are on the same cpu
2561 * during the cmpxchg then the free will succedd.
2563 c
= __this_cpu_ptr(s
->cpu_slab
);
2568 if (likely(page
== c
->page
)) {
2569 set_freepointer(s
, object
, c
->freelist
);
2571 if (unlikely(!irqsafe_cpu_cmpxchg_double(
2572 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2574 object
, next_tid(tid
)))) {
2576 note_cmpxchg_failure("slab_free", s
, tid
);
2579 stat(s
, FREE_FASTPATH
);
2581 __slab_free(s
, page
, x
, addr
);
2585 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2589 page
= virt_to_head_page(x
);
2591 slab_free(s
, page
, x
, _RET_IP_
);
2593 trace_kmem_cache_free(_RET_IP_
, x
);
2595 EXPORT_SYMBOL(kmem_cache_free
);
2598 * Object placement in a slab is made very easy because we always start at
2599 * offset 0. If we tune the size of the object to the alignment then we can
2600 * get the required alignment by putting one properly sized object after
2603 * Notice that the allocation order determines the sizes of the per cpu
2604 * caches. Each processor has always one slab available for allocations.
2605 * Increasing the allocation order reduces the number of times that slabs
2606 * must be moved on and off the partial lists and is therefore a factor in
2611 * Mininum / Maximum order of slab pages. This influences locking overhead
2612 * and slab fragmentation. A higher order reduces the number of partial slabs
2613 * and increases the number of allocations possible without having to
2614 * take the list_lock.
2616 static int slub_min_order
;
2617 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2618 static int slub_min_objects
;
2621 * Merge control. If this is set then no merging of slab caches will occur.
2622 * (Could be removed. This was introduced to pacify the merge skeptics.)
2624 static int slub_nomerge
;
2627 * Calculate the order of allocation given an slab object size.
2629 * The order of allocation has significant impact on performance and other
2630 * system components. Generally order 0 allocations should be preferred since
2631 * order 0 does not cause fragmentation in the page allocator. Larger objects
2632 * be problematic to put into order 0 slabs because there may be too much
2633 * unused space left. We go to a higher order if more than 1/16th of the slab
2636 * In order to reach satisfactory performance we must ensure that a minimum
2637 * number of objects is in one slab. Otherwise we may generate too much
2638 * activity on the partial lists which requires taking the list_lock. This is
2639 * less a concern for large slabs though which are rarely used.
2641 * slub_max_order specifies the order where we begin to stop considering the
2642 * number of objects in a slab as critical. If we reach slub_max_order then
2643 * we try to keep the page order as low as possible. So we accept more waste
2644 * of space in favor of a small page order.
2646 * Higher order allocations also allow the placement of more objects in a
2647 * slab and thereby reduce object handling overhead. If the user has
2648 * requested a higher mininum order then we start with that one instead of
2649 * the smallest order which will fit the object.
2651 static inline int slab_order(int size
, int min_objects
,
2652 int max_order
, int fract_leftover
, int reserved
)
2656 int min_order
= slub_min_order
;
2658 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2659 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2661 for (order
= max(min_order
,
2662 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2663 order
<= max_order
; order
++) {
2665 unsigned long slab_size
= PAGE_SIZE
<< order
;
2667 if (slab_size
< min_objects
* size
+ reserved
)
2670 rem
= (slab_size
- reserved
) % size
;
2672 if (rem
<= slab_size
/ fract_leftover
)
2680 static inline int calculate_order(int size
, int reserved
)
2688 * Attempt to find best configuration for a slab. This
2689 * works by first attempting to generate a layout with
2690 * the best configuration and backing off gradually.
2692 * First we reduce the acceptable waste in a slab. Then
2693 * we reduce the minimum objects required in a slab.
2695 min_objects
= slub_min_objects
;
2697 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2698 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2699 min_objects
= min(min_objects
, max_objects
);
2701 while (min_objects
> 1) {
2703 while (fraction
>= 4) {
2704 order
= slab_order(size
, min_objects
,
2705 slub_max_order
, fraction
, reserved
);
2706 if (order
<= slub_max_order
)
2714 * We were unable to place multiple objects in a slab. Now
2715 * lets see if we can place a single object there.
2717 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2718 if (order
<= slub_max_order
)
2722 * Doh this slab cannot be placed using slub_max_order.
2724 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2725 if (order
< MAX_ORDER
)
2731 * Figure out what the alignment of the objects will be.
2733 static unsigned long calculate_alignment(unsigned long flags
,
2734 unsigned long align
, unsigned long size
)
2737 * If the user wants hardware cache aligned objects then follow that
2738 * suggestion if the object is sufficiently large.
2740 * The hardware cache alignment cannot override the specified
2741 * alignment though. If that is greater then use it.
2743 if (flags
& SLAB_HWCACHE_ALIGN
) {
2744 unsigned long ralign
= cache_line_size();
2745 while (size
<= ralign
/ 2)
2747 align
= max(align
, ralign
);
2750 if (align
< ARCH_SLAB_MINALIGN
)
2751 align
= ARCH_SLAB_MINALIGN
;
2753 return ALIGN(align
, sizeof(void *));
2757 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2760 spin_lock_init(&n
->list_lock
);
2761 INIT_LIST_HEAD(&n
->partial
);
2762 #ifdef CONFIG_SLUB_DEBUG
2763 atomic_long_set(&n
->nr_slabs
, 0);
2764 atomic_long_set(&n
->total_objects
, 0);
2765 INIT_LIST_HEAD(&n
->full
);
2769 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2771 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2772 SLUB_PAGE_SHIFT
* sizeof(struct kmem_cache_cpu
));
2775 * Must align to double word boundary for the double cmpxchg
2776 * instructions to work; see __pcpu_double_call_return_bool().
2778 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2779 2 * sizeof(void *));
2784 init_kmem_cache_cpus(s
);
2789 static struct kmem_cache
*kmem_cache_node
;
2792 * No kmalloc_node yet so do it by hand. We know that this is the first
2793 * slab on the node for this slabcache. There are no concurrent accesses
2796 * Note that this function only works on the kmalloc_node_cache
2797 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2798 * memory on a fresh node that has no slab structures yet.
2800 static void early_kmem_cache_node_alloc(int node
)
2803 struct kmem_cache_node
*n
;
2805 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2807 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2810 if (page_to_nid(page
) != node
) {
2811 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2813 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2814 "in order to be able to continue\n");
2819 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2822 kmem_cache_node
->node
[node
] = n
;
2823 #ifdef CONFIG_SLUB_DEBUG
2824 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2825 init_tracking(kmem_cache_node
, n
);
2827 init_kmem_cache_node(n
, kmem_cache_node
);
2828 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2830 add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2833 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2837 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2838 struct kmem_cache_node
*n
= s
->node
[node
];
2841 kmem_cache_free(kmem_cache_node
, n
);
2843 s
->node
[node
] = NULL
;
2847 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2851 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2852 struct kmem_cache_node
*n
;
2854 if (slab_state
== DOWN
) {
2855 early_kmem_cache_node_alloc(node
);
2858 n
= kmem_cache_alloc_node(kmem_cache_node
,
2862 free_kmem_cache_nodes(s
);
2867 init_kmem_cache_node(n
, s
);
2872 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2874 if (min
< MIN_PARTIAL
)
2876 else if (min
> MAX_PARTIAL
)
2878 s
->min_partial
= min
;
2882 * calculate_sizes() determines the order and the distribution of data within
2885 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2887 unsigned long flags
= s
->flags
;
2888 unsigned long size
= s
->objsize
;
2889 unsigned long align
= s
->align
;
2893 * Round up object size to the next word boundary. We can only
2894 * place the free pointer at word boundaries and this determines
2895 * the possible location of the free pointer.
2897 size
= ALIGN(size
, sizeof(void *));
2899 #ifdef CONFIG_SLUB_DEBUG
2901 * Determine if we can poison the object itself. If the user of
2902 * the slab may touch the object after free or before allocation
2903 * then we should never poison the object itself.
2905 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2907 s
->flags
|= __OBJECT_POISON
;
2909 s
->flags
&= ~__OBJECT_POISON
;
2913 * If we are Redzoning then check if there is some space between the
2914 * end of the object and the free pointer. If not then add an
2915 * additional word to have some bytes to store Redzone information.
2917 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2918 size
+= sizeof(void *);
2922 * With that we have determined the number of bytes in actual use
2923 * by the object. This is the potential offset to the free pointer.
2927 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2930 * Relocate free pointer after the object if it is not
2931 * permitted to overwrite the first word of the object on
2934 * This is the case if we do RCU, have a constructor or
2935 * destructor or are poisoning the objects.
2938 size
+= sizeof(void *);
2941 #ifdef CONFIG_SLUB_DEBUG
2942 if (flags
& SLAB_STORE_USER
)
2944 * Need to store information about allocs and frees after
2947 size
+= 2 * sizeof(struct track
);
2949 if (flags
& SLAB_RED_ZONE
)
2951 * Add some empty padding so that we can catch
2952 * overwrites from earlier objects rather than let
2953 * tracking information or the free pointer be
2954 * corrupted if a user writes before the start
2957 size
+= sizeof(void *);
2961 * Determine the alignment based on various parameters that the
2962 * user specified and the dynamic determination of cache line size
2965 align
= calculate_alignment(flags
, align
, s
->objsize
);
2969 * SLUB stores one object immediately after another beginning from
2970 * offset 0. In order to align the objects we have to simply size
2971 * each object to conform to the alignment.
2973 size
= ALIGN(size
, align
);
2975 if (forced_order
>= 0)
2976 order
= forced_order
;
2978 order
= calculate_order(size
, s
->reserved
);
2985 s
->allocflags
|= __GFP_COMP
;
2987 if (s
->flags
& SLAB_CACHE_DMA
)
2988 s
->allocflags
|= SLUB_DMA
;
2990 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2991 s
->allocflags
|= __GFP_RECLAIMABLE
;
2994 * Determine the number of objects per slab
2996 s
->oo
= oo_make(order
, size
, s
->reserved
);
2997 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
2998 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3001 return !!oo_objects(s
->oo
);
3005 static int kmem_cache_open(struct kmem_cache
*s
,
3006 const char *name
, size_t size
,
3007 size_t align
, unsigned long flags
,
3008 void (*ctor
)(void *))
3010 memset(s
, 0, kmem_size
);
3015 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
3018 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3019 s
->reserved
= sizeof(struct rcu_head
);
3021 if (!calculate_sizes(s
, -1))
3023 if (disable_higher_order_debug
) {
3025 * Disable debugging flags that store metadata if the min slab
3028 if (get_order(s
->size
) > get_order(s
->objsize
)) {
3029 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3031 if (!calculate_sizes(s
, -1))
3036 #ifdef CONFIG_CMPXCHG_DOUBLE
3037 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3038 /* Enable fast mode */
3039 s
->flags
|= __CMPXCHG_DOUBLE
;
3043 * The larger the object size is, the more pages we want on the partial
3044 * list to avoid pounding the page allocator excessively.
3046 set_min_partial(s
, ilog2(s
->size
) / 2);
3049 * cpu_partial determined the maximum number of objects kept in the
3050 * per cpu partial lists of a processor.
3052 * Per cpu partial lists mainly contain slabs that just have one
3053 * object freed. If they are used for allocation then they can be
3054 * filled up again with minimal effort. The slab will never hit the
3055 * per node partial lists and therefore no locking will be required.
3057 * This setting also determines
3059 * A) The number of objects from per cpu partial slabs dumped to the
3060 * per node list when we reach the limit.
3061 * B) The number of objects in cpu partial slabs to extract from the
3062 * per node list when we run out of per cpu objects. We only fetch 50%
3063 * to keep some capacity around for frees.
3065 if (s
->size
>= PAGE_SIZE
)
3067 else if (s
->size
>= 1024)
3069 else if (s
->size
>= 256)
3070 s
->cpu_partial
= 13;
3072 s
->cpu_partial
= 30;
3076 s
->remote_node_defrag_ratio
= 1000;
3078 if (!init_kmem_cache_nodes(s
))
3081 if (alloc_kmem_cache_cpus(s
))
3084 free_kmem_cache_nodes(s
);
3086 if (flags
& SLAB_PANIC
)
3087 panic("Cannot create slab %s size=%lu realsize=%u "
3088 "order=%u offset=%u flags=%lx\n",
3089 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
3095 * Determine the size of a slab object
3097 unsigned int kmem_cache_size(struct kmem_cache
*s
)
3101 EXPORT_SYMBOL(kmem_cache_size
);
3103 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3106 #ifdef CONFIG_SLUB_DEBUG
3107 void *addr
= page_address(page
);
3109 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3110 sizeof(long), GFP_ATOMIC
);
3113 slab_err(s
, page
, "%s", text
);
3116 get_map(s
, page
, map
);
3117 for_each_object(p
, s
, addr
, page
->objects
) {
3119 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3120 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
3122 print_tracking(s
, p
);
3131 * Attempt to free all partial slabs on a node.
3132 * This is called from kmem_cache_close(). We must be the last thread
3133 * using the cache and therefore we do not need to lock anymore.
3135 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3137 struct page
*page
, *h
;
3139 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3141 remove_partial(n
, page
);
3142 discard_slab(s
, page
);
3144 list_slab_objects(s
, page
,
3145 "Objects remaining on kmem_cache_close()");
3151 * Release all resources used by a slab cache.
3153 static inline int kmem_cache_close(struct kmem_cache
*s
)
3158 free_percpu(s
->cpu_slab
);
3159 /* Attempt to free all objects */
3160 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3161 struct kmem_cache_node
*n
= get_node(s
, node
);
3164 if (n
->nr_partial
|| slabs_node(s
, node
))
3167 free_kmem_cache_nodes(s
);
3172 * Close a cache and release the kmem_cache structure
3173 * (must be used for caches created using kmem_cache_create)
3175 void kmem_cache_destroy(struct kmem_cache
*s
)
3177 down_write(&slub_lock
);
3181 up_write(&slub_lock
);
3182 if (kmem_cache_close(s
)) {
3183 printk(KERN_ERR
"SLUB %s: %s called for cache that "
3184 "still has objects.\n", s
->name
, __func__
);
3187 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
3189 sysfs_slab_remove(s
);
3191 up_write(&slub_lock
);
3193 EXPORT_SYMBOL(kmem_cache_destroy
);
3195 /********************************************************************
3197 *******************************************************************/
3199 struct kmem_cache
*kmalloc_caches
[SLUB_PAGE_SHIFT
];
3200 EXPORT_SYMBOL(kmalloc_caches
);
3202 static struct kmem_cache
*kmem_cache
;
3204 #ifdef CONFIG_ZONE_DMA
3205 static struct kmem_cache
*kmalloc_dma_caches
[SLUB_PAGE_SHIFT
];
3208 static int __init
setup_slub_min_order(char *str
)
3210 get_option(&str
, &slub_min_order
);
3215 __setup("slub_min_order=", setup_slub_min_order
);
3217 static int __init
setup_slub_max_order(char *str
)
3219 get_option(&str
, &slub_max_order
);
3220 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3225 __setup("slub_max_order=", setup_slub_max_order
);
3227 static int __init
setup_slub_min_objects(char *str
)
3229 get_option(&str
, &slub_min_objects
);
3234 __setup("slub_min_objects=", setup_slub_min_objects
);
3236 static int __init
setup_slub_nomerge(char *str
)
3242 __setup("slub_nomerge", setup_slub_nomerge
);
3244 static struct kmem_cache
*__init
create_kmalloc_cache(const char *name
,
3245 int size
, unsigned int flags
)
3247 struct kmem_cache
*s
;
3249 s
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3252 * This function is called with IRQs disabled during early-boot on
3253 * single CPU so there's no need to take slub_lock here.
3255 if (!kmem_cache_open(s
, name
, size
, ARCH_KMALLOC_MINALIGN
,
3259 list_add(&s
->list
, &slab_caches
);
3263 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
3268 * Conversion table for small slabs sizes / 8 to the index in the
3269 * kmalloc array. This is necessary for slabs < 192 since we have non power
3270 * of two cache sizes there. The size of larger slabs can be determined using
3273 static s8 size_index
[24] = {
3300 static inline int size_index_elem(size_t bytes
)
3302 return (bytes
- 1) / 8;
3305 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
3311 return ZERO_SIZE_PTR
;
3313 index
= size_index
[size_index_elem(size
)];
3315 index
= fls(size
- 1);
3317 #ifdef CONFIG_ZONE_DMA
3318 if (unlikely((flags
& SLUB_DMA
)))
3319 return kmalloc_dma_caches
[index
];
3322 return kmalloc_caches
[index
];
3325 void *__kmalloc(size_t size
, gfp_t flags
)
3327 struct kmem_cache
*s
;
3330 if (unlikely(size
> SLUB_MAX_SIZE
))
3331 return kmalloc_large(size
, flags
);
3333 s
= get_slab(size
, flags
);
3335 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3338 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
3340 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3344 EXPORT_SYMBOL(__kmalloc
);
3347 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3352 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3353 page
= alloc_pages_node(node
, flags
, get_order(size
));
3355 ptr
= page_address(page
);
3357 kmemleak_alloc(ptr
, size
, 1, flags
);
3361 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3363 struct kmem_cache
*s
;
3366 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3367 ret
= kmalloc_large_node(size
, flags
, node
);
3369 trace_kmalloc_node(_RET_IP_
, ret
,
3370 size
, PAGE_SIZE
<< get_order(size
),
3376 s
= get_slab(size
, flags
);
3378 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3381 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
3383 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3387 EXPORT_SYMBOL(__kmalloc_node
);
3390 size_t ksize(const void *object
)
3394 if (unlikely(object
== ZERO_SIZE_PTR
))
3397 page
= virt_to_head_page(object
);
3399 if (unlikely(!PageSlab(page
))) {
3400 WARN_ON(!PageCompound(page
));
3401 return PAGE_SIZE
<< compound_order(page
);
3404 return slab_ksize(page
->slab
);
3406 EXPORT_SYMBOL(ksize
);
3408 #ifdef CONFIG_SLUB_DEBUG
3409 bool verify_mem_not_deleted(const void *x
)
3412 void *object
= (void *)x
;
3413 unsigned long flags
;
3416 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3419 local_irq_save(flags
);
3421 page
= virt_to_head_page(x
);
3422 if (unlikely(!PageSlab(page
))) {
3423 /* maybe it was from stack? */
3429 if (on_freelist(page
->slab
, page
, object
)) {
3430 object_err(page
->slab
, page
, object
, "Object is on free-list");
3438 local_irq_restore(flags
);
3441 EXPORT_SYMBOL(verify_mem_not_deleted
);
3444 void kfree(const void *x
)
3447 void *object
= (void *)x
;
3449 trace_kfree(_RET_IP_
, x
);
3451 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3454 page
= virt_to_head_page(x
);
3455 if (unlikely(!PageSlab(page
))) {
3456 BUG_ON(!PageCompound(page
));
3461 slab_free(page
->slab
, page
, object
, _RET_IP_
);
3463 EXPORT_SYMBOL(kfree
);
3466 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3467 * the remaining slabs by the number of items in use. The slabs with the
3468 * most items in use come first. New allocations will then fill those up
3469 * and thus they can be removed from the partial lists.
3471 * The slabs with the least items are placed last. This results in them
3472 * being allocated from last increasing the chance that the last objects
3473 * are freed in them.
3475 int kmem_cache_shrink(struct kmem_cache
*s
)
3479 struct kmem_cache_node
*n
;
3482 int objects
= oo_objects(s
->max
);
3483 struct list_head
*slabs_by_inuse
=
3484 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3485 unsigned long flags
;
3487 if (!slabs_by_inuse
)
3491 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3492 n
= get_node(s
, node
);
3497 for (i
= 0; i
< objects
; i
++)
3498 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3500 spin_lock_irqsave(&n
->list_lock
, flags
);
3503 * Build lists indexed by the items in use in each slab.
3505 * Note that concurrent frees may occur while we hold the
3506 * list_lock. page->inuse here is the upper limit.
3508 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3509 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3515 * Rebuild the partial list with the slabs filled up most
3516 * first and the least used slabs at the end.
3518 for (i
= objects
- 1; i
> 0; i
--)
3519 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3521 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3523 /* Release empty slabs */
3524 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3525 discard_slab(s
, page
);
3528 kfree(slabs_by_inuse
);
3531 EXPORT_SYMBOL(kmem_cache_shrink
);
3533 #if defined(CONFIG_MEMORY_HOTPLUG)
3534 static int slab_mem_going_offline_callback(void *arg
)
3536 struct kmem_cache
*s
;
3538 down_read(&slub_lock
);
3539 list_for_each_entry(s
, &slab_caches
, list
)
3540 kmem_cache_shrink(s
);
3541 up_read(&slub_lock
);
3546 static void slab_mem_offline_callback(void *arg
)
3548 struct kmem_cache_node
*n
;
3549 struct kmem_cache
*s
;
3550 struct memory_notify
*marg
= arg
;
3553 offline_node
= marg
->status_change_nid
;
3556 * If the node still has available memory. we need kmem_cache_node
3559 if (offline_node
< 0)
3562 down_read(&slub_lock
);
3563 list_for_each_entry(s
, &slab_caches
, list
) {
3564 n
= get_node(s
, offline_node
);
3567 * if n->nr_slabs > 0, slabs still exist on the node
3568 * that is going down. We were unable to free them,
3569 * and offline_pages() function shouldn't call this
3570 * callback. So, we must fail.
3572 BUG_ON(slabs_node(s
, offline_node
));
3574 s
->node
[offline_node
] = NULL
;
3575 kmem_cache_free(kmem_cache_node
, n
);
3578 up_read(&slub_lock
);
3581 static int slab_mem_going_online_callback(void *arg
)
3583 struct kmem_cache_node
*n
;
3584 struct kmem_cache
*s
;
3585 struct memory_notify
*marg
= arg
;
3586 int nid
= marg
->status_change_nid
;
3590 * If the node's memory is already available, then kmem_cache_node is
3591 * already created. Nothing to do.
3597 * We are bringing a node online. No memory is available yet. We must
3598 * allocate a kmem_cache_node structure in order to bring the node
3601 down_read(&slub_lock
);
3602 list_for_each_entry(s
, &slab_caches
, list
) {
3604 * XXX: kmem_cache_alloc_node will fallback to other nodes
3605 * since memory is not yet available from the node that
3608 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3613 init_kmem_cache_node(n
, s
);
3617 up_read(&slub_lock
);
3621 static int slab_memory_callback(struct notifier_block
*self
,
3622 unsigned long action
, void *arg
)
3627 case MEM_GOING_ONLINE
:
3628 ret
= slab_mem_going_online_callback(arg
);
3630 case MEM_GOING_OFFLINE
:
3631 ret
= slab_mem_going_offline_callback(arg
);
3634 case MEM_CANCEL_ONLINE
:
3635 slab_mem_offline_callback(arg
);
3638 case MEM_CANCEL_OFFLINE
:
3642 ret
= notifier_from_errno(ret
);
3648 #endif /* CONFIG_MEMORY_HOTPLUG */
3650 /********************************************************************
3651 * Basic setup of slabs
3652 *******************************************************************/
3655 * Used for early kmem_cache structures that were allocated using
3656 * the page allocator
3659 static void __init
kmem_cache_bootstrap_fixup(struct kmem_cache
*s
)
3663 list_add(&s
->list
, &slab_caches
);
3666 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3667 struct kmem_cache_node
*n
= get_node(s
, node
);
3671 list_for_each_entry(p
, &n
->partial
, lru
)
3674 #ifdef CONFIG_SLUB_DEBUG
3675 list_for_each_entry(p
, &n
->full
, lru
)
3682 void __init
kmem_cache_init(void)
3686 struct kmem_cache
*temp_kmem_cache
;
3688 struct kmem_cache
*temp_kmem_cache_node
;
3689 unsigned long kmalloc_size
;
3691 kmem_size
= offsetof(struct kmem_cache
, node
) +
3692 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3694 /* Allocate two kmem_caches from the page allocator */
3695 kmalloc_size
= ALIGN(kmem_size
, cache_line_size());
3696 order
= get_order(2 * kmalloc_size
);
3697 kmem_cache
= (void *)__get_free_pages(GFP_NOWAIT
, order
);
3700 * Must first have the slab cache available for the allocations of the
3701 * struct kmem_cache_node's. There is special bootstrap code in
3702 * kmem_cache_open for slab_state == DOWN.
3704 kmem_cache_node
= (void *)kmem_cache
+ kmalloc_size
;
3706 kmem_cache_open(kmem_cache_node
, "kmem_cache_node",
3707 sizeof(struct kmem_cache_node
),
3708 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3710 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3712 /* Able to allocate the per node structures */
3713 slab_state
= PARTIAL
;
3715 temp_kmem_cache
= kmem_cache
;
3716 kmem_cache_open(kmem_cache
, "kmem_cache", kmem_size
,
3717 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3718 kmem_cache
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3719 memcpy(kmem_cache
, temp_kmem_cache
, kmem_size
);
3722 * Allocate kmem_cache_node properly from the kmem_cache slab.
3723 * kmem_cache_node is separately allocated so no need to
3724 * update any list pointers.
3726 temp_kmem_cache_node
= kmem_cache_node
;
3728 kmem_cache_node
= kmem_cache_alloc(kmem_cache
, GFP_NOWAIT
);
3729 memcpy(kmem_cache_node
, temp_kmem_cache_node
, kmem_size
);
3731 kmem_cache_bootstrap_fixup(kmem_cache_node
);
3734 kmem_cache_bootstrap_fixup(kmem_cache
);
3736 /* Free temporary boot structure */
3737 free_pages((unsigned long)temp_kmem_cache
, order
);
3739 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3742 * Patch up the size_index table if we have strange large alignment
3743 * requirements for the kmalloc array. This is only the case for
3744 * MIPS it seems. The standard arches will not generate any code here.
3746 * Largest permitted alignment is 256 bytes due to the way we
3747 * handle the index determination for the smaller caches.
3749 * Make sure that nothing crazy happens if someone starts tinkering
3750 * around with ARCH_KMALLOC_MINALIGN
3752 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3753 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3755 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3756 int elem
= size_index_elem(i
);
3757 if (elem
>= ARRAY_SIZE(size_index
))
3759 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3762 if (KMALLOC_MIN_SIZE
== 64) {
3764 * The 96 byte size cache is not used if the alignment
3767 for (i
= 64 + 8; i
<= 96; i
+= 8)
3768 size_index
[size_index_elem(i
)] = 7;
3769 } else if (KMALLOC_MIN_SIZE
== 128) {
3771 * The 192 byte sized cache is not used if the alignment
3772 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3775 for (i
= 128 + 8; i
<= 192; i
+= 8)
3776 size_index
[size_index_elem(i
)] = 8;
3779 /* Caches that are not of the two-to-the-power-of size */
3780 if (KMALLOC_MIN_SIZE
<= 32) {
3781 kmalloc_caches
[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3785 if (KMALLOC_MIN_SIZE
<= 64) {
3786 kmalloc_caches
[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3790 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3791 kmalloc_caches
[i
] = create_kmalloc_cache("kmalloc", 1 << i
, 0);
3797 /* Provide the correct kmalloc names now that the caches are up */
3798 if (KMALLOC_MIN_SIZE
<= 32) {
3799 kmalloc_caches
[1]->name
= kstrdup(kmalloc_caches
[1]->name
, GFP_NOWAIT
);
3800 BUG_ON(!kmalloc_caches
[1]->name
);
3803 if (KMALLOC_MIN_SIZE
<= 64) {
3804 kmalloc_caches
[2]->name
= kstrdup(kmalloc_caches
[2]->name
, GFP_NOWAIT
);
3805 BUG_ON(!kmalloc_caches
[2]->name
);
3808 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3809 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3812 kmalloc_caches
[i
]->name
= s
;
3816 register_cpu_notifier(&slab_notifier
);
3819 #ifdef CONFIG_ZONE_DMA
3820 for (i
= 0; i
< SLUB_PAGE_SHIFT
; i
++) {
3821 struct kmem_cache
*s
= kmalloc_caches
[i
];
3824 char *name
= kasprintf(GFP_NOWAIT
,
3825 "dma-kmalloc-%d", s
->objsize
);
3828 kmalloc_dma_caches
[i
] = create_kmalloc_cache(name
,
3829 s
->objsize
, SLAB_CACHE_DMA
);
3834 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3835 " CPUs=%d, Nodes=%d\n",
3836 caches
, cache_line_size(),
3837 slub_min_order
, slub_max_order
, slub_min_objects
,
3838 nr_cpu_ids
, nr_node_ids
);
3841 void __init
kmem_cache_init_late(void)
3846 * Find a mergeable slab cache
3848 static int slab_unmergeable(struct kmem_cache
*s
)
3850 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3857 * We may have set a slab to be unmergeable during bootstrap.
3859 if (s
->refcount
< 0)
3865 static struct kmem_cache
*find_mergeable(size_t size
,
3866 size_t align
, unsigned long flags
, const char *name
,
3867 void (*ctor
)(void *))
3869 struct kmem_cache
*s
;
3871 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3877 size
= ALIGN(size
, sizeof(void *));
3878 align
= calculate_alignment(flags
, align
, size
);
3879 size
= ALIGN(size
, align
);
3880 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3882 list_for_each_entry(s
, &slab_caches
, list
) {
3883 if (slab_unmergeable(s
))
3889 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3892 * Check if alignment is compatible.
3893 * Courtesy of Adrian Drzewiecki
3895 if ((s
->size
& ~(align
- 1)) != s
->size
)
3898 if (s
->size
- size
>= sizeof(void *))
3906 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3907 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3909 struct kmem_cache
*s
;
3915 down_write(&slub_lock
);
3916 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3920 * Adjust the object sizes so that we clear
3921 * the complete object on kzalloc.
3923 s
->objsize
= max(s
->objsize
, (int)size
);
3924 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3926 if (sysfs_slab_alias(s
, name
)) {
3930 up_write(&slub_lock
);
3934 n
= kstrdup(name
, GFP_KERNEL
);
3938 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3940 if (kmem_cache_open(s
, n
,
3941 size
, align
, flags
, ctor
)) {
3942 list_add(&s
->list
, &slab_caches
);
3943 if (sysfs_slab_add(s
)) {
3949 up_write(&slub_lock
);
3956 up_write(&slub_lock
);
3958 if (flags
& SLAB_PANIC
)
3959 panic("Cannot create slabcache %s\n", name
);
3964 EXPORT_SYMBOL(kmem_cache_create
);
3968 * Use the cpu notifier to insure that the cpu slabs are flushed when
3971 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3972 unsigned long action
, void *hcpu
)
3974 long cpu
= (long)hcpu
;
3975 struct kmem_cache
*s
;
3976 unsigned long flags
;
3979 case CPU_UP_CANCELED
:
3980 case CPU_UP_CANCELED_FROZEN
:
3982 case CPU_DEAD_FROZEN
:
3983 down_read(&slub_lock
);
3984 list_for_each_entry(s
, &slab_caches
, list
) {
3985 local_irq_save(flags
);
3986 __flush_cpu_slab(s
, cpu
);
3987 local_irq_restore(flags
);
3989 up_read(&slub_lock
);
3997 static struct notifier_block __cpuinitdata slab_notifier
= {
3998 .notifier_call
= slab_cpuup_callback
4003 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
4005 struct kmem_cache
*s
;
4008 if (unlikely(size
> SLUB_MAX_SIZE
))
4009 return kmalloc_large(size
, gfpflags
);
4011 s
= get_slab(size
, gfpflags
);
4013 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4016 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
4018 /* Honor the call site pointer we received. */
4019 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
4025 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
4026 int node
, unsigned long caller
)
4028 struct kmem_cache
*s
;
4031 if (unlikely(size
> SLUB_MAX_SIZE
)) {
4032 ret
= kmalloc_large_node(size
, gfpflags
, node
);
4034 trace_kmalloc_node(caller
, ret
,
4035 size
, PAGE_SIZE
<< get_order(size
),
4041 s
= get_slab(size
, gfpflags
);
4043 if (unlikely(ZERO_OR_NULL_PTR(s
)))
4046 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
4048 /* Honor the call site pointer we received. */
4049 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
4056 static int count_inuse(struct page
*page
)
4061 static int count_total(struct page
*page
)
4063 return page
->objects
;
4067 #ifdef CONFIG_SLUB_DEBUG
4068 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
4072 void *addr
= page_address(page
);
4074 if (!check_slab(s
, page
) ||
4075 !on_freelist(s
, page
, NULL
))
4078 /* Now we know that a valid freelist exists */
4079 bitmap_zero(map
, page
->objects
);
4081 get_map(s
, page
, map
);
4082 for_each_object(p
, s
, addr
, page
->objects
) {
4083 if (test_bit(slab_index(p
, s
, addr
), map
))
4084 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
4088 for_each_object(p
, s
, addr
, page
->objects
)
4089 if (!test_bit(slab_index(p
, s
, addr
), map
))
4090 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
4095 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
4099 validate_slab(s
, page
, map
);
4103 static int validate_slab_node(struct kmem_cache
*s
,
4104 struct kmem_cache_node
*n
, unsigned long *map
)
4106 unsigned long count
= 0;
4108 unsigned long flags
;
4110 spin_lock_irqsave(&n
->list_lock
, flags
);
4112 list_for_each_entry(page
, &n
->partial
, lru
) {
4113 validate_slab_slab(s
, page
, map
);
4116 if (count
!= n
->nr_partial
)
4117 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
4118 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
4120 if (!(s
->flags
& SLAB_STORE_USER
))
4123 list_for_each_entry(page
, &n
->full
, lru
) {
4124 validate_slab_slab(s
, page
, map
);
4127 if (count
!= atomic_long_read(&n
->nr_slabs
))
4128 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
4129 "counter=%ld\n", s
->name
, count
,
4130 atomic_long_read(&n
->nr_slabs
));
4133 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4137 static long validate_slab_cache(struct kmem_cache
*s
)
4140 unsigned long count
= 0;
4141 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4142 sizeof(unsigned long), GFP_KERNEL
);
4148 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4149 struct kmem_cache_node
*n
= get_node(s
, node
);
4151 count
+= validate_slab_node(s
, n
, map
);
4157 * Generate lists of code addresses where slabcache objects are allocated
4162 unsigned long count
;
4169 DECLARE_BITMAP(cpus
, NR_CPUS
);
4175 unsigned long count
;
4176 struct location
*loc
;
4179 static void free_loc_track(struct loc_track
*t
)
4182 free_pages((unsigned long)t
->loc
,
4183 get_order(sizeof(struct location
) * t
->max
));
4186 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
4191 order
= get_order(sizeof(struct location
) * max
);
4193 l
= (void *)__get_free_pages(flags
, order
);
4198 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
4206 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
4207 const struct track
*track
)
4209 long start
, end
, pos
;
4211 unsigned long caddr
;
4212 unsigned long age
= jiffies
- track
->when
;
4218 pos
= start
+ (end
- start
+ 1) / 2;
4221 * There is nothing at "end". If we end up there
4222 * we need to add something to before end.
4227 caddr
= t
->loc
[pos
].addr
;
4228 if (track
->addr
== caddr
) {
4234 if (age
< l
->min_time
)
4236 if (age
> l
->max_time
)
4239 if (track
->pid
< l
->min_pid
)
4240 l
->min_pid
= track
->pid
;
4241 if (track
->pid
> l
->max_pid
)
4242 l
->max_pid
= track
->pid
;
4244 cpumask_set_cpu(track
->cpu
,
4245 to_cpumask(l
->cpus
));
4247 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4251 if (track
->addr
< caddr
)
4258 * Not found. Insert new tracking element.
4260 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4266 (t
->count
- pos
) * sizeof(struct location
));
4269 l
->addr
= track
->addr
;
4273 l
->min_pid
= track
->pid
;
4274 l
->max_pid
= track
->pid
;
4275 cpumask_clear(to_cpumask(l
->cpus
));
4276 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4277 nodes_clear(l
->nodes
);
4278 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4282 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4283 struct page
*page
, enum track_item alloc
,
4286 void *addr
= page_address(page
);
4289 bitmap_zero(map
, page
->objects
);
4290 get_map(s
, page
, map
);
4292 for_each_object(p
, s
, addr
, page
->objects
)
4293 if (!test_bit(slab_index(p
, s
, addr
), map
))
4294 add_location(t
, s
, get_track(s
, p
, alloc
));
4297 static int list_locations(struct kmem_cache
*s
, char *buf
,
4298 enum track_item alloc
)
4302 struct loc_track t
= { 0, 0, NULL
};
4304 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4305 sizeof(unsigned long), GFP_KERNEL
);
4307 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4310 return sprintf(buf
, "Out of memory\n");
4312 /* Push back cpu slabs */
4315 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4316 struct kmem_cache_node
*n
= get_node(s
, node
);
4317 unsigned long flags
;
4320 if (!atomic_long_read(&n
->nr_slabs
))
4323 spin_lock_irqsave(&n
->list_lock
, flags
);
4324 list_for_each_entry(page
, &n
->partial
, lru
)
4325 process_slab(&t
, s
, page
, alloc
, map
);
4326 list_for_each_entry(page
, &n
->full
, lru
)
4327 process_slab(&t
, s
, page
, alloc
, map
);
4328 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4331 for (i
= 0; i
< t
.count
; i
++) {
4332 struct location
*l
= &t
.loc
[i
];
4334 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4336 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4339 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4341 len
+= sprintf(buf
+ len
, "<not-available>");
4343 if (l
->sum_time
!= l
->min_time
) {
4344 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4346 (long)div_u64(l
->sum_time
, l
->count
),
4349 len
+= sprintf(buf
+ len
, " age=%ld",
4352 if (l
->min_pid
!= l
->max_pid
)
4353 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4354 l
->min_pid
, l
->max_pid
);
4356 len
+= sprintf(buf
+ len
, " pid=%ld",
4359 if (num_online_cpus() > 1 &&
4360 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4361 len
< PAGE_SIZE
- 60) {
4362 len
+= sprintf(buf
+ len
, " cpus=");
4363 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4364 to_cpumask(l
->cpus
));
4367 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4368 len
< PAGE_SIZE
- 60) {
4369 len
+= sprintf(buf
+ len
, " nodes=");
4370 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
4374 len
+= sprintf(buf
+ len
, "\n");
4380 len
+= sprintf(buf
, "No data\n");
4385 #ifdef SLUB_RESILIENCY_TEST
4386 static void resiliency_test(void)
4390 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || SLUB_PAGE_SHIFT
< 10);
4392 printk(KERN_ERR
"SLUB resiliency testing\n");
4393 printk(KERN_ERR
"-----------------------\n");
4394 printk(KERN_ERR
"A. Corruption after allocation\n");
4396 p
= kzalloc(16, GFP_KERNEL
);
4398 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
4399 " 0x12->0x%p\n\n", p
+ 16);
4401 validate_slab_cache(kmalloc_caches
[4]);
4403 /* Hmmm... The next two are dangerous */
4404 p
= kzalloc(32, GFP_KERNEL
);
4405 p
[32 + sizeof(void *)] = 0x34;
4406 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
4407 " 0x34 -> -0x%p\n", p
);
4409 "If allocated object is overwritten then not detectable\n\n");
4411 validate_slab_cache(kmalloc_caches
[5]);
4412 p
= kzalloc(64, GFP_KERNEL
);
4413 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4415 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4418 "If allocated object is overwritten then not detectable\n\n");
4419 validate_slab_cache(kmalloc_caches
[6]);
4421 printk(KERN_ERR
"\nB. Corruption after free\n");
4422 p
= kzalloc(128, GFP_KERNEL
);
4425 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4426 validate_slab_cache(kmalloc_caches
[7]);
4428 p
= kzalloc(256, GFP_KERNEL
);
4431 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
4433 validate_slab_cache(kmalloc_caches
[8]);
4435 p
= kzalloc(512, GFP_KERNEL
);
4438 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4439 validate_slab_cache(kmalloc_caches
[9]);
4443 static void resiliency_test(void) {};
4448 enum slab_stat_type
{
4449 SL_ALL
, /* All slabs */
4450 SL_PARTIAL
, /* Only partially allocated slabs */
4451 SL_CPU
, /* Only slabs used for cpu caches */
4452 SL_OBJECTS
, /* Determine allocated objects not slabs */
4453 SL_TOTAL
/* Determine object capacity not slabs */
4456 #define SO_ALL (1 << SL_ALL)
4457 #define SO_PARTIAL (1 << SL_PARTIAL)
4458 #define SO_CPU (1 << SL_CPU)
4459 #define SO_OBJECTS (1 << SL_OBJECTS)
4460 #define SO_TOTAL (1 << SL_TOTAL)
4462 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4463 char *buf
, unsigned long flags
)
4465 unsigned long total
= 0;
4468 unsigned long *nodes
;
4469 unsigned long *per_cpu
;
4471 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4474 per_cpu
= nodes
+ nr_node_ids
;
4476 if (flags
& SO_CPU
) {
4479 for_each_possible_cpu(cpu
) {
4480 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
4483 if (!c
|| c
->node
< 0)
4487 if (flags
& SO_TOTAL
)
4488 x
= c
->page
->objects
;
4489 else if (flags
& SO_OBJECTS
)
4495 nodes
[c
->node
] += x
;
4502 nodes
[c
->node
] += x
;
4508 lock_memory_hotplug();
4509 #ifdef CONFIG_SLUB_DEBUG
4510 if (flags
& SO_ALL
) {
4511 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4512 struct kmem_cache_node
*n
= get_node(s
, node
);
4514 if (flags
& SO_TOTAL
)
4515 x
= atomic_long_read(&n
->total_objects
);
4516 else if (flags
& SO_OBJECTS
)
4517 x
= atomic_long_read(&n
->total_objects
) -
4518 count_partial(n
, count_free
);
4521 x
= atomic_long_read(&n
->nr_slabs
);
4528 if (flags
& SO_PARTIAL
) {
4529 for_each_node_state(node
, N_NORMAL_MEMORY
) {
4530 struct kmem_cache_node
*n
= get_node(s
, node
);
4532 if (flags
& SO_TOTAL
)
4533 x
= count_partial(n
, count_total
);
4534 else if (flags
& SO_OBJECTS
)
4535 x
= count_partial(n
, count_inuse
);
4542 x
= sprintf(buf
, "%lu", total
);
4544 for_each_node_state(node
, N_NORMAL_MEMORY
)
4546 x
+= sprintf(buf
+ x
, " N%d=%lu",
4549 unlock_memory_hotplug();
4551 return x
+ sprintf(buf
+ x
, "\n");
4554 #ifdef CONFIG_SLUB_DEBUG
4555 static int any_slab_objects(struct kmem_cache
*s
)
4559 for_each_online_node(node
) {
4560 struct kmem_cache_node
*n
= get_node(s
, node
);
4565 if (atomic_long_read(&n
->total_objects
))
4572 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4573 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4575 struct slab_attribute
{
4576 struct attribute attr
;
4577 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4578 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4581 #define SLAB_ATTR_RO(_name) \
4582 static struct slab_attribute _name##_attr = \
4583 __ATTR(_name, 0400, _name##_show, NULL)
4585 #define SLAB_ATTR(_name) \
4586 static struct slab_attribute _name##_attr = \
4587 __ATTR(_name, 0600, _name##_show, _name##_store)
4589 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4591 return sprintf(buf
, "%d\n", s
->size
);
4593 SLAB_ATTR_RO(slab_size
);
4595 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4597 return sprintf(buf
, "%d\n", s
->align
);
4599 SLAB_ATTR_RO(align
);
4601 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4603 return sprintf(buf
, "%d\n", s
->objsize
);
4605 SLAB_ATTR_RO(object_size
);
4607 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4609 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4611 SLAB_ATTR_RO(objs_per_slab
);
4613 static ssize_t
order_store(struct kmem_cache
*s
,
4614 const char *buf
, size_t length
)
4616 unsigned long order
;
4619 err
= strict_strtoul(buf
, 10, &order
);
4623 if (order
> slub_max_order
|| order
< slub_min_order
)
4626 calculate_sizes(s
, order
);
4630 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4632 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4636 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4638 return sprintf(buf
, "%lu\n", s
->min_partial
);
4641 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4647 err
= strict_strtoul(buf
, 10, &min
);
4651 set_min_partial(s
, min
);
4654 SLAB_ATTR(min_partial
);
4656 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4658 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4661 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4664 unsigned long objects
;
4667 err
= strict_strtoul(buf
, 10, &objects
);
4671 s
->cpu_partial
= objects
;
4675 SLAB_ATTR(cpu_partial
);
4677 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4681 return sprintf(buf
, "%pS\n", s
->ctor
);
4685 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4687 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4689 SLAB_ATTR_RO(aliases
);
4691 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4693 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4695 SLAB_ATTR_RO(partial
);
4697 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4699 return show_slab_objects(s
, buf
, SO_CPU
);
4701 SLAB_ATTR_RO(cpu_slabs
);
4703 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4705 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4707 SLAB_ATTR_RO(objects
);
4709 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4711 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4713 SLAB_ATTR_RO(objects_partial
);
4715 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4722 for_each_online_cpu(cpu
) {
4723 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4726 pages
+= page
->pages
;
4727 objects
+= page
->pobjects
;
4731 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4734 for_each_online_cpu(cpu
) {
4735 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4737 if (page
&& len
< PAGE_SIZE
- 20)
4738 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4739 page
->pobjects
, page
->pages
);
4742 return len
+ sprintf(buf
+ len
, "\n");
4744 SLAB_ATTR_RO(slabs_cpu_partial
);
4746 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4748 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4751 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4752 const char *buf
, size_t length
)
4754 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4756 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4759 SLAB_ATTR(reclaim_account
);
4761 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4763 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4765 SLAB_ATTR_RO(hwcache_align
);
4767 #ifdef CONFIG_ZONE_DMA
4768 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4770 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4772 SLAB_ATTR_RO(cache_dma
);
4775 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4777 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4779 SLAB_ATTR_RO(destroy_by_rcu
);
4781 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4783 return sprintf(buf
, "%d\n", s
->reserved
);
4785 SLAB_ATTR_RO(reserved
);
4787 #ifdef CONFIG_SLUB_DEBUG
4788 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4790 return show_slab_objects(s
, buf
, SO_ALL
);
4792 SLAB_ATTR_RO(slabs
);
4794 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4796 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4798 SLAB_ATTR_RO(total_objects
);
4800 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4802 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4805 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4806 const char *buf
, size_t length
)
4808 s
->flags
&= ~SLAB_DEBUG_FREE
;
4809 if (buf
[0] == '1') {
4810 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4811 s
->flags
|= SLAB_DEBUG_FREE
;
4815 SLAB_ATTR(sanity_checks
);
4817 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4819 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4822 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4825 s
->flags
&= ~SLAB_TRACE
;
4826 if (buf
[0] == '1') {
4827 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4828 s
->flags
|= SLAB_TRACE
;
4834 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4836 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4839 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4840 const char *buf
, size_t length
)
4842 if (any_slab_objects(s
))
4845 s
->flags
&= ~SLAB_RED_ZONE
;
4846 if (buf
[0] == '1') {
4847 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4848 s
->flags
|= SLAB_RED_ZONE
;
4850 calculate_sizes(s
, -1);
4853 SLAB_ATTR(red_zone
);
4855 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4857 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4860 static ssize_t
poison_store(struct kmem_cache
*s
,
4861 const char *buf
, size_t length
)
4863 if (any_slab_objects(s
))
4866 s
->flags
&= ~SLAB_POISON
;
4867 if (buf
[0] == '1') {
4868 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4869 s
->flags
|= SLAB_POISON
;
4871 calculate_sizes(s
, -1);
4876 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4878 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4881 static ssize_t
store_user_store(struct kmem_cache
*s
,
4882 const char *buf
, size_t length
)
4884 if (any_slab_objects(s
))
4887 s
->flags
&= ~SLAB_STORE_USER
;
4888 if (buf
[0] == '1') {
4889 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4890 s
->flags
|= SLAB_STORE_USER
;
4892 calculate_sizes(s
, -1);
4895 SLAB_ATTR(store_user
);
4897 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4902 static ssize_t
validate_store(struct kmem_cache
*s
,
4903 const char *buf
, size_t length
)
4907 if (buf
[0] == '1') {
4908 ret
= validate_slab_cache(s
);
4914 SLAB_ATTR(validate
);
4916 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4918 if (!(s
->flags
& SLAB_STORE_USER
))
4920 return list_locations(s
, buf
, TRACK_ALLOC
);
4922 SLAB_ATTR_RO(alloc_calls
);
4924 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4926 if (!(s
->flags
& SLAB_STORE_USER
))
4928 return list_locations(s
, buf
, TRACK_FREE
);
4930 SLAB_ATTR_RO(free_calls
);
4931 #endif /* CONFIG_SLUB_DEBUG */
4933 #ifdef CONFIG_FAILSLAB
4934 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4936 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4939 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4942 s
->flags
&= ~SLAB_FAILSLAB
;
4944 s
->flags
|= SLAB_FAILSLAB
;
4947 SLAB_ATTR(failslab
);
4950 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4955 static ssize_t
shrink_store(struct kmem_cache
*s
,
4956 const char *buf
, size_t length
)
4958 if (buf
[0] == '1') {
4959 int rc
= kmem_cache_shrink(s
);
4970 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4972 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4975 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4976 const char *buf
, size_t length
)
4978 unsigned long ratio
;
4981 err
= strict_strtoul(buf
, 10, &ratio
);
4986 s
->remote_node_defrag_ratio
= ratio
* 10;
4990 SLAB_ATTR(remote_node_defrag_ratio
);
4993 #ifdef CONFIG_SLUB_STATS
4994 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4996 unsigned long sum
= 0;
4999 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
5004 for_each_online_cpu(cpu
) {
5005 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
5011 len
= sprintf(buf
, "%lu", sum
);
5014 for_each_online_cpu(cpu
) {
5015 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
5016 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
5020 return len
+ sprintf(buf
+ len
, "\n");
5023 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
5027 for_each_online_cpu(cpu
)
5028 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
5031 #define STAT_ATTR(si, text) \
5032 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5034 return show_stat(s, buf, si); \
5036 static ssize_t text##_store(struct kmem_cache *s, \
5037 const char *buf, size_t length) \
5039 if (buf[0] != '0') \
5041 clear_stat(s, si); \
5046 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5047 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
5048 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
5049 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
5050 STAT_ATTR(FREE_FROZEN
, free_frozen
);
5051 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
5052 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
5053 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
5054 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
5055 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
5056 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
5057 STAT_ATTR(FREE_SLAB
, free_slab
);
5058 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
5059 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
5060 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
5061 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
5062 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
5063 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
5064 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
5065 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
5066 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
5067 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
5068 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
5069 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
5072 static struct attribute
*slab_attrs
[] = {
5073 &slab_size_attr
.attr
,
5074 &object_size_attr
.attr
,
5075 &objs_per_slab_attr
.attr
,
5077 &min_partial_attr
.attr
,
5078 &cpu_partial_attr
.attr
,
5080 &objects_partial_attr
.attr
,
5082 &cpu_slabs_attr
.attr
,
5086 &hwcache_align_attr
.attr
,
5087 &reclaim_account_attr
.attr
,
5088 &destroy_by_rcu_attr
.attr
,
5090 &reserved_attr
.attr
,
5091 &slabs_cpu_partial_attr
.attr
,
5092 #ifdef CONFIG_SLUB_DEBUG
5093 &total_objects_attr
.attr
,
5095 &sanity_checks_attr
.attr
,
5097 &red_zone_attr
.attr
,
5099 &store_user_attr
.attr
,
5100 &validate_attr
.attr
,
5101 &alloc_calls_attr
.attr
,
5102 &free_calls_attr
.attr
,
5104 #ifdef CONFIG_ZONE_DMA
5105 &cache_dma_attr
.attr
,
5108 &remote_node_defrag_ratio_attr
.attr
,
5110 #ifdef CONFIG_SLUB_STATS
5111 &alloc_fastpath_attr
.attr
,
5112 &alloc_slowpath_attr
.attr
,
5113 &free_fastpath_attr
.attr
,
5114 &free_slowpath_attr
.attr
,
5115 &free_frozen_attr
.attr
,
5116 &free_add_partial_attr
.attr
,
5117 &free_remove_partial_attr
.attr
,
5118 &alloc_from_partial_attr
.attr
,
5119 &alloc_slab_attr
.attr
,
5120 &alloc_refill_attr
.attr
,
5121 &alloc_node_mismatch_attr
.attr
,
5122 &free_slab_attr
.attr
,
5123 &cpuslab_flush_attr
.attr
,
5124 &deactivate_full_attr
.attr
,
5125 &deactivate_empty_attr
.attr
,
5126 &deactivate_to_head_attr
.attr
,
5127 &deactivate_to_tail_attr
.attr
,
5128 &deactivate_remote_frees_attr
.attr
,
5129 &deactivate_bypass_attr
.attr
,
5130 &order_fallback_attr
.attr
,
5131 &cmpxchg_double_fail_attr
.attr
,
5132 &cmpxchg_double_cpu_fail_attr
.attr
,
5133 &cpu_partial_alloc_attr
.attr
,
5134 &cpu_partial_free_attr
.attr
,
5136 #ifdef CONFIG_FAILSLAB
5137 &failslab_attr
.attr
,
5143 static struct attribute_group slab_attr_group
= {
5144 .attrs
= slab_attrs
,
5147 static ssize_t
slab_attr_show(struct kobject
*kobj
,
5148 struct attribute
*attr
,
5151 struct slab_attribute
*attribute
;
5152 struct kmem_cache
*s
;
5155 attribute
= to_slab_attr(attr
);
5158 if (!attribute
->show
)
5161 err
= attribute
->show(s
, buf
);
5166 static ssize_t
slab_attr_store(struct kobject
*kobj
,
5167 struct attribute
*attr
,
5168 const char *buf
, size_t len
)
5170 struct slab_attribute
*attribute
;
5171 struct kmem_cache
*s
;
5174 attribute
= to_slab_attr(attr
);
5177 if (!attribute
->store
)
5180 err
= attribute
->store(s
, buf
, len
);
5185 static void kmem_cache_release(struct kobject
*kobj
)
5187 struct kmem_cache
*s
= to_slab(kobj
);
5193 static const struct sysfs_ops slab_sysfs_ops
= {
5194 .show
= slab_attr_show
,
5195 .store
= slab_attr_store
,
5198 static struct kobj_type slab_ktype
= {
5199 .sysfs_ops
= &slab_sysfs_ops
,
5200 .release
= kmem_cache_release
5203 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5205 struct kobj_type
*ktype
= get_ktype(kobj
);
5207 if (ktype
== &slab_ktype
)
5212 static const struct kset_uevent_ops slab_uevent_ops
= {
5213 .filter
= uevent_filter
,
5216 static struct kset
*slab_kset
;
5218 #define ID_STR_LENGTH 64
5220 /* Create a unique string id for a slab cache:
5222 * Format :[flags-]size
5224 static char *create_unique_id(struct kmem_cache
*s
)
5226 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5233 * First flags affecting slabcache operations. We will only
5234 * get here for aliasable slabs so we do not need to support
5235 * too many flags. The flags here must cover all flags that
5236 * are matched during merging to guarantee that the id is
5239 if (s
->flags
& SLAB_CACHE_DMA
)
5241 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5243 if (s
->flags
& SLAB_DEBUG_FREE
)
5245 if (!(s
->flags
& SLAB_NOTRACK
))
5249 p
+= sprintf(p
, "%07d", s
->size
);
5250 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5254 static int sysfs_slab_add(struct kmem_cache
*s
)
5260 if (slab_state
< SYSFS
)
5261 /* Defer until later */
5264 unmergeable
= slab_unmergeable(s
);
5267 * Slabcache can never be merged so we can use the name proper.
5268 * This is typically the case for debug situations. In that
5269 * case we can catch duplicate names easily.
5271 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5275 * Create a unique name for the slab as a target
5278 name
= create_unique_id(s
);
5281 s
->kobj
.kset
= slab_kset
;
5282 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
5284 kobject_put(&s
->kobj
);
5288 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5290 kobject_del(&s
->kobj
);
5291 kobject_put(&s
->kobj
);
5294 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5296 /* Setup first alias */
5297 sysfs_slab_alias(s
, s
->name
);
5303 static void sysfs_slab_remove(struct kmem_cache
*s
)
5305 if (slab_state
< SYSFS
)
5307 * Sysfs has not been setup yet so no need to remove the
5312 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5313 kobject_del(&s
->kobj
);
5314 kobject_put(&s
->kobj
);
5318 * Need to buffer aliases during bootup until sysfs becomes
5319 * available lest we lose that information.
5321 struct saved_alias
{
5322 struct kmem_cache
*s
;
5324 struct saved_alias
*next
;
5327 static struct saved_alias
*alias_list
;
5329 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5331 struct saved_alias
*al
;
5333 if (slab_state
== SYSFS
) {
5335 * If we have a leftover link then remove it.
5337 sysfs_remove_link(&slab_kset
->kobj
, name
);
5338 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5341 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5347 al
->next
= alias_list
;
5352 static int __init
slab_sysfs_init(void)
5354 struct kmem_cache
*s
;
5357 down_write(&slub_lock
);
5359 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5361 up_write(&slub_lock
);
5362 printk(KERN_ERR
"Cannot register slab subsystem.\n");
5368 list_for_each_entry(s
, &slab_caches
, list
) {
5369 err
= sysfs_slab_add(s
);
5371 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
5372 " to sysfs\n", s
->name
);
5375 while (alias_list
) {
5376 struct saved_alias
*al
= alias_list
;
5378 alias_list
= alias_list
->next
;
5379 err
= sysfs_slab_alias(al
->s
, al
->name
);
5381 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
5382 " %s to sysfs\n", s
->name
);
5386 up_write(&slub_lock
);
5391 __initcall(slab_sysfs_init
);
5392 #endif /* CONFIG_SYSFS */
5395 * The /proc/slabinfo ABI
5397 #ifdef CONFIG_SLABINFO
5398 static void print_slabinfo_header(struct seq_file
*m
)
5400 seq_puts(m
, "slabinfo - version: 2.1\n");
5401 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
5402 "<objperslab> <pagesperslab>");
5403 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
5404 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
5408 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
5412 down_read(&slub_lock
);
5414 print_slabinfo_header(m
);
5416 return seq_list_start(&slab_caches
, *pos
);
5419 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
5421 return seq_list_next(p
, &slab_caches
, pos
);
5424 static void s_stop(struct seq_file
*m
, void *p
)
5426 up_read(&slub_lock
);
5429 static int s_show(struct seq_file
*m
, void *p
)
5431 unsigned long nr_partials
= 0;
5432 unsigned long nr_slabs
= 0;
5433 unsigned long nr_inuse
= 0;
5434 unsigned long nr_objs
= 0;
5435 unsigned long nr_free
= 0;
5436 struct kmem_cache
*s
;
5439 s
= list_entry(p
, struct kmem_cache
, list
);
5441 for_each_online_node(node
) {
5442 struct kmem_cache_node
*n
= get_node(s
, node
);
5447 nr_partials
+= n
->nr_partial
;
5448 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
5449 nr_objs
+= atomic_long_read(&n
->total_objects
);
5450 nr_free
+= count_partial(n
, count_free
);
5453 nr_inuse
= nr_objs
- nr_free
;
5455 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
5456 nr_objs
, s
->size
, oo_objects(s
->oo
),
5457 (1 << oo_order(s
->oo
)));
5458 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
5459 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
5465 static const struct seq_operations slabinfo_op
= {
5472 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
5474 return seq_open(file
, &slabinfo_op
);
5477 static const struct file_operations proc_slabinfo_operations
= {
5478 .open
= slabinfo_open
,
5480 .llseek
= seq_lseek
,
5481 .release
= seq_release
,
5484 static int __init
slab_proc_init(void)
5486 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
5489 module_init(slab_proc_init
);
5490 #endif /* CONFIG_SLABINFO */