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 and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemtrace.h>
21 #include <linux/kmemcheck.h>
22 #include <linux/cpu.h>
23 #include <linux/cpuset.h>
24 #include <linux/kmemleak.h>
25 #include <linux/mempolicy.h>
26 #include <linux/ctype.h>
27 #include <linux/debugobjects.h>
28 #include <linux/kallsyms.h>
29 #include <linux/memory.h>
30 #include <linux/math64.h>
31 #include <linux/fault-inject.h>
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
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 #ifdef CONFIG_SLUB_DEBUG
118 * Issues still to be resolved:
120 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
122 * - Variable sizing of the per node arrays
125 /* Enable to test recovery from slab corruption on boot */
126 #undef SLUB_RESILIENCY_TEST
129 * Mininum number of partial slabs. These will be left on the partial
130 * lists even if they are empty. kmem_cache_shrink may reclaim them.
132 #define MIN_PARTIAL 5
135 * Maximum number of desirable partial slabs.
136 * The existence of more partial slabs makes kmem_cache_shrink
137 * sort the partial list by the number of objects in the.
139 #define MAX_PARTIAL 10
141 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
142 SLAB_POISON | SLAB_STORE_USER)
145 * Set of flags that will prevent slab merging
147 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
148 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE)
150 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
151 SLAB_CACHE_DMA | SLAB_NOTRACK)
153 #ifndef ARCH_KMALLOC_MINALIGN
154 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
157 #ifndef ARCH_SLAB_MINALIGN
158 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
162 #define OO_MASK ((1 << OO_SHIFT) - 1)
163 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
165 /* Internal SLUB flags */
166 #define __OBJECT_POISON 0x80000000 /* Poison object */
167 #define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
169 static int kmem_size
= sizeof(struct kmem_cache
);
172 static struct notifier_block slab_notifier
;
176 DOWN
, /* No slab functionality available */
177 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
178 UP
, /* Everything works but does not show up in sysfs */
182 /* A list of all slab caches on the system */
183 static DECLARE_RWSEM(slub_lock
);
184 static LIST_HEAD(slab_caches
);
187 * Tracking user of a slab.
190 unsigned long addr
; /* Called from address */
191 int cpu
; /* Was running on cpu */
192 int pid
; /* Pid context */
193 unsigned long when
; /* When did the operation occur */
196 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
198 #ifdef CONFIG_SLUB_DEBUG
199 static int sysfs_slab_add(struct kmem_cache
*);
200 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
201 static void sysfs_slab_remove(struct kmem_cache
*);
204 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
205 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
207 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
214 static inline void stat(struct kmem_cache_cpu
*c
, enum stat_item si
)
216 #ifdef CONFIG_SLUB_STATS
221 /********************************************************************
222 * Core slab cache functions
223 *******************************************************************/
225 int slab_is_available(void)
227 return slab_state
>= UP
;
230 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
233 return s
->node
[node
];
235 return &s
->local_node
;
239 static inline struct kmem_cache_cpu
*get_cpu_slab(struct kmem_cache
*s
, int cpu
)
242 return s
->cpu_slab
[cpu
];
248 /* Verify that a pointer has an address that is valid within a slab page */
249 static inline int check_valid_pointer(struct kmem_cache
*s
,
250 struct page
*page
, const void *object
)
257 base
= page_address(page
);
258 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
259 (object
- base
) % s
->size
) {
267 * Slow version of get and set free pointer.
269 * This version requires touching the cache lines of kmem_cache which
270 * we avoid to do in the fast alloc free paths. There we obtain the offset
271 * from the page struct.
273 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
275 return *(void **)(object
+ s
->offset
);
278 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
280 *(void **)(object
+ s
->offset
) = fp
;
283 /* Loop over all objects in a slab */
284 #define for_each_object(__p, __s, __addr, __objects) \
285 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
289 #define for_each_free_object(__p, __s, __free) \
290 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
292 /* Determine object index from a given position */
293 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
295 return (p
- addr
) / s
->size
;
298 static inline struct kmem_cache_order_objects
oo_make(int order
,
301 struct kmem_cache_order_objects x
= {
302 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
308 static inline int oo_order(struct kmem_cache_order_objects x
)
310 return x
.x
>> OO_SHIFT
;
313 static inline int oo_objects(struct kmem_cache_order_objects x
)
315 return x
.x
& OO_MASK
;
318 #ifdef CONFIG_SLUB_DEBUG
322 #ifdef CONFIG_SLUB_DEBUG_ON
323 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
325 static int slub_debug
;
328 static char *slub_debug_slabs
;
333 static void print_section(char *text
, u8
*addr
, unsigned int length
)
341 for (i
= 0; i
< length
; i
++) {
343 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
346 printk(KERN_CONT
" %02x", addr
[i
]);
348 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
350 printk(KERN_CONT
" %s\n", ascii
);
357 printk(KERN_CONT
" ");
361 printk(KERN_CONT
" %s\n", ascii
);
365 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
366 enum track_item alloc
)
371 p
= object
+ s
->offset
+ sizeof(void *);
373 p
= object
+ s
->inuse
;
378 static void set_track(struct kmem_cache
*s
, void *object
,
379 enum track_item alloc
, unsigned long addr
)
381 struct track
*p
= get_track(s
, object
, alloc
);
385 p
->cpu
= smp_processor_id();
386 p
->pid
= current
->pid
;
389 memset(p
, 0, sizeof(struct track
));
392 static void init_tracking(struct kmem_cache
*s
, void *object
)
394 if (!(s
->flags
& SLAB_STORE_USER
))
397 set_track(s
, object
, TRACK_FREE
, 0UL);
398 set_track(s
, object
, TRACK_ALLOC
, 0UL);
401 static void print_track(const char *s
, struct track
*t
)
406 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
407 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
410 static void print_tracking(struct kmem_cache
*s
, void *object
)
412 if (!(s
->flags
& SLAB_STORE_USER
))
415 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
416 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
419 static void print_page_info(struct page
*page
)
421 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
422 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
426 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
432 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
434 printk(KERN_ERR
"========================================"
435 "=====================================\n");
436 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
437 printk(KERN_ERR
"----------------------------------------"
438 "-------------------------------------\n\n");
441 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
447 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
449 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
452 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
454 unsigned int off
; /* Offset of last byte */
455 u8
*addr
= page_address(page
);
457 print_tracking(s
, p
);
459 print_page_info(page
);
461 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
462 p
, p
- addr
, get_freepointer(s
, p
));
465 print_section("Bytes b4", p
- 16, 16);
467 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
469 if (s
->flags
& SLAB_RED_ZONE
)
470 print_section("Redzone", p
+ s
->objsize
,
471 s
->inuse
- s
->objsize
);
474 off
= s
->offset
+ sizeof(void *);
478 if (s
->flags
& SLAB_STORE_USER
)
479 off
+= 2 * sizeof(struct track
);
482 /* Beginning of the filler is the free pointer */
483 print_section("Padding", p
+ off
, s
->size
- off
);
488 static void object_err(struct kmem_cache
*s
, struct page
*page
,
489 u8
*object
, char *reason
)
491 slab_bug(s
, "%s", reason
);
492 print_trailer(s
, page
, object
);
495 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
501 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
503 slab_bug(s
, "%s", buf
);
504 print_page_info(page
);
508 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
512 if (s
->flags
& __OBJECT_POISON
) {
513 memset(p
, POISON_FREE
, s
->objsize
- 1);
514 p
[s
->objsize
- 1] = POISON_END
;
517 if (s
->flags
& SLAB_RED_ZONE
)
518 memset(p
+ s
->objsize
,
519 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
520 s
->inuse
- s
->objsize
);
523 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
526 if (*start
!= (u8
)value
)
534 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
535 void *from
, void *to
)
537 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
538 memset(from
, data
, to
- from
);
541 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
542 u8
*object
, char *what
,
543 u8
*start
, unsigned int value
, unsigned int bytes
)
548 fault
= check_bytes(start
, value
, bytes
);
553 while (end
> fault
&& end
[-1] == value
)
556 slab_bug(s
, "%s overwritten", what
);
557 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
558 fault
, end
- 1, fault
[0], value
);
559 print_trailer(s
, page
, object
);
561 restore_bytes(s
, what
, value
, fault
, end
);
569 * Bytes of the object to be managed.
570 * If the freepointer may overlay the object then the free
571 * pointer is the first word of the object.
573 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
576 * object + s->objsize
577 * Padding to reach word boundary. This is also used for Redzoning.
578 * Padding is extended by another word if Redzoning is enabled and
581 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
582 * 0xcc (RED_ACTIVE) for objects in use.
585 * Meta data starts here.
587 * A. Free pointer (if we cannot overwrite object on free)
588 * B. Tracking data for SLAB_STORE_USER
589 * C. Padding to reach required alignment boundary or at mininum
590 * one word if debugging is on to be able to detect writes
591 * before the word boundary.
593 * Padding is done using 0x5a (POISON_INUSE)
596 * Nothing is used beyond s->size.
598 * If slabcaches are merged then the objsize and inuse boundaries are mostly
599 * ignored. And therefore no slab options that rely on these boundaries
600 * may be used with merged slabcaches.
603 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
605 unsigned long off
= s
->inuse
; /* The end of info */
608 /* Freepointer is placed after the object. */
609 off
+= sizeof(void *);
611 if (s
->flags
& SLAB_STORE_USER
)
612 /* We also have user information there */
613 off
+= 2 * sizeof(struct track
);
618 return check_bytes_and_report(s
, page
, p
, "Object padding",
619 p
+ off
, POISON_INUSE
, s
->size
- off
);
622 /* Check the pad bytes at the end of a slab page */
623 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
631 if (!(s
->flags
& SLAB_POISON
))
634 start
= page_address(page
);
635 length
= (PAGE_SIZE
<< compound_order(page
));
636 end
= start
+ length
;
637 remainder
= length
% s
->size
;
641 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
644 while (end
> fault
&& end
[-1] == POISON_INUSE
)
647 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
648 print_section("Padding", end
- remainder
, remainder
);
650 restore_bytes(s
, "slab padding", POISON_INUSE
, start
, end
);
654 static int check_object(struct kmem_cache
*s
, struct page
*page
,
655 void *object
, int active
)
658 u8
*endobject
= object
+ s
->objsize
;
660 if (s
->flags
& SLAB_RED_ZONE
) {
662 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
664 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
665 endobject
, red
, s
->inuse
- s
->objsize
))
668 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
669 check_bytes_and_report(s
, page
, p
, "Alignment padding",
670 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
674 if (s
->flags
& SLAB_POISON
) {
675 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
676 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
677 POISON_FREE
, s
->objsize
- 1) ||
678 !check_bytes_and_report(s
, page
, p
, "Poison",
679 p
+ s
->objsize
- 1, POISON_END
, 1)))
682 * check_pad_bytes cleans up on its own.
684 check_pad_bytes(s
, page
, p
);
687 if (!s
->offset
&& active
)
689 * Object and freepointer overlap. Cannot check
690 * freepointer while object is allocated.
694 /* Check free pointer validity */
695 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
696 object_err(s
, page
, p
, "Freepointer corrupt");
698 * No choice but to zap it and thus lose the remainder
699 * of the free objects in this slab. May cause
700 * another error because the object count is now wrong.
702 set_freepointer(s
, p
, NULL
);
708 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
712 VM_BUG_ON(!irqs_disabled());
714 if (!PageSlab(page
)) {
715 slab_err(s
, page
, "Not a valid slab page");
719 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
720 if (page
->objects
> maxobj
) {
721 slab_err(s
, page
, "objects %u > max %u",
722 s
->name
, page
->objects
, maxobj
);
725 if (page
->inuse
> page
->objects
) {
726 slab_err(s
, page
, "inuse %u > max %u",
727 s
->name
, page
->inuse
, page
->objects
);
730 /* Slab_pad_check fixes things up after itself */
731 slab_pad_check(s
, page
);
736 * Determine if a certain object on a page is on the freelist. Must hold the
737 * slab lock to guarantee that the chains are in a consistent state.
739 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
742 void *fp
= page
->freelist
;
744 unsigned long max_objects
;
746 while (fp
&& nr
<= page
->objects
) {
749 if (!check_valid_pointer(s
, page
, fp
)) {
751 object_err(s
, page
, object
,
752 "Freechain corrupt");
753 set_freepointer(s
, object
, NULL
);
756 slab_err(s
, page
, "Freepointer corrupt");
757 page
->freelist
= NULL
;
758 page
->inuse
= page
->objects
;
759 slab_fix(s
, "Freelist cleared");
765 fp
= get_freepointer(s
, object
);
769 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
770 if (max_objects
> MAX_OBJS_PER_PAGE
)
771 max_objects
= MAX_OBJS_PER_PAGE
;
773 if (page
->objects
!= max_objects
) {
774 slab_err(s
, page
, "Wrong number of objects. Found %d but "
775 "should be %d", page
->objects
, max_objects
);
776 page
->objects
= max_objects
;
777 slab_fix(s
, "Number of objects adjusted.");
779 if (page
->inuse
!= page
->objects
- nr
) {
780 slab_err(s
, page
, "Wrong object count. Counter is %d but "
781 "counted were %d", page
->inuse
, page
->objects
- nr
);
782 page
->inuse
= page
->objects
- nr
;
783 slab_fix(s
, "Object count adjusted.");
785 return search
== NULL
;
788 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
791 if (s
->flags
& SLAB_TRACE
) {
792 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
794 alloc
? "alloc" : "free",
799 print_section("Object", (void *)object
, s
->objsize
);
806 * Tracking of fully allocated slabs for debugging purposes.
808 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
810 spin_lock(&n
->list_lock
);
811 list_add(&page
->lru
, &n
->full
);
812 spin_unlock(&n
->list_lock
);
815 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
817 struct kmem_cache_node
*n
;
819 if (!(s
->flags
& SLAB_STORE_USER
))
822 n
= get_node(s
, page_to_nid(page
));
824 spin_lock(&n
->list_lock
);
825 list_del(&page
->lru
);
826 spin_unlock(&n
->list_lock
);
829 /* Tracking of the number of slabs for debugging purposes */
830 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
832 struct kmem_cache_node
*n
= get_node(s
, node
);
834 return atomic_long_read(&n
->nr_slabs
);
837 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
839 struct kmem_cache_node
*n
= get_node(s
, node
);
842 * May be called early in order to allocate a slab for the
843 * kmem_cache_node structure. Solve the chicken-egg
844 * dilemma by deferring the increment of the count during
845 * bootstrap (see early_kmem_cache_node_alloc).
847 if (!NUMA_BUILD
|| n
) {
848 atomic_long_inc(&n
->nr_slabs
);
849 atomic_long_add(objects
, &n
->total_objects
);
852 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
854 struct kmem_cache_node
*n
= get_node(s
, node
);
856 atomic_long_dec(&n
->nr_slabs
);
857 atomic_long_sub(objects
, &n
->total_objects
);
860 /* Object debug checks for alloc/free paths */
861 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
864 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
867 init_object(s
, object
, 0);
868 init_tracking(s
, object
);
871 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
872 void *object
, unsigned long addr
)
874 if (!check_slab(s
, page
))
877 if (!on_freelist(s
, page
, object
)) {
878 object_err(s
, page
, object
, "Object already allocated");
882 if (!check_valid_pointer(s
, page
, object
)) {
883 object_err(s
, page
, object
, "Freelist Pointer check fails");
887 if (!check_object(s
, page
, object
, 0))
890 /* Success perform special debug activities for allocs */
891 if (s
->flags
& SLAB_STORE_USER
)
892 set_track(s
, object
, TRACK_ALLOC
, addr
);
893 trace(s
, page
, object
, 1);
894 init_object(s
, object
, 1);
898 if (PageSlab(page
)) {
900 * If this is a slab page then lets do the best we can
901 * to avoid issues in the future. Marking all objects
902 * as used avoids touching the remaining objects.
904 slab_fix(s
, "Marking all objects used");
905 page
->inuse
= page
->objects
;
906 page
->freelist
= NULL
;
911 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
912 void *object
, unsigned long addr
)
914 if (!check_slab(s
, page
))
917 if (!check_valid_pointer(s
, page
, object
)) {
918 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
922 if (on_freelist(s
, page
, object
)) {
923 object_err(s
, page
, object
, "Object already free");
927 if (!check_object(s
, page
, object
, 1))
930 if (unlikely(s
!= page
->slab
)) {
931 if (!PageSlab(page
)) {
932 slab_err(s
, page
, "Attempt to free object(0x%p) "
933 "outside of slab", object
);
934 } else if (!page
->slab
) {
936 "SLUB <none>: no slab for object 0x%p.\n",
940 object_err(s
, page
, object
,
941 "page slab pointer corrupt.");
945 /* Special debug activities for freeing objects */
946 if (!PageSlubFrozen(page
) && !page
->freelist
)
947 remove_full(s
, page
);
948 if (s
->flags
& SLAB_STORE_USER
)
949 set_track(s
, object
, TRACK_FREE
, addr
);
950 trace(s
, page
, object
, 0);
951 init_object(s
, object
, 0);
955 slab_fix(s
, "Object at 0x%p not freed", object
);
959 static int __init
setup_slub_debug(char *str
)
961 slub_debug
= DEBUG_DEFAULT_FLAGS
;
962 if (*str
++ != '=' || !*str
)
964 * No options specified. Switch on full debugging.
970 * No options but restriction on slabs. This means full
971 * debugging for slabs matching a pattern.
978 * Switch off all debugging measures.
983 * Determine which debug features should be switched on
985 for (; *str
&& *str
!= ','; str
++) {
986 switch (tolower(*str
)) {
988 slub_debug
|= SLAB_DEBUG_FREE
;
991 slub_debug
|= SLAB_RED_ZONE
;
994 slub_debug
|= SLAB_POISON
;
997 slub_debug
|= SLAB_STORE_USER
;
1000 slub_debug
|= SLAB_TRACE
;
1003 printk(KERN_ERR
"slub_debug option '%c' "
1004 "unknown. skipped\n", *str
);
1010 slub_debug_slabs
= str
+ 1;
1015 __setup("slub_debug", setup_slub_debug
);
1017 static unsigned long kmem_cache_flags(unsigned long objsize
,
1018 unsigned long flags
, const char *name
,
1019 void (*ctor
)(void *))
1022 * Enable debugging if selected on the kernel commandline.
1024 if (slub_debug
&& (!slub_debug_slabs
||
1025 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1026 flags
|= slub_debug
;
1031 static inline void setup_object_debug(struct kmem_cache
*s
,
1032 struct page
*page
, void *object
) {}
1034 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1035 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1037 static inline int free_debug_processing(struct kmem_cache
*s
,
1038 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1040 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1042 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1043 void *object
, int active
) { return 1; }
1044 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1045 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1046 unsigned long flags
, const char *name
,
1047 void (*ctor
)(void *))
1051 #define slub_debug 0
1053 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1055 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1057 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1062 * Slab allocation and freeing
1064 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1065 struct kmem_cache_order_objects oo
)
1067 int order
= oo_order(oo
);
1070 return alloc_pages(flags
, order
);
1072 return alloc_pages_node(node
, flags
, order
);
1075 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1078 struct kmem_cache_order_objects oo
= s
->oo
;
1080 flags
|= s
->allocflags
;
1082 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1084 if (unlikely(!page
)) {
1087 * Allocation may have failed due to fragmentation.
1088 * Try a lower order alloc if possible
1090 page
= alloc_slab_page(flags
, node
, oo
);
1094 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1097 if (kmemcheck_enabled
1098 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
)))
1100 kmemcheck_alloc_shadow(s
, flags
, node
, page
, compound_order(page
));
1103 page
->objects
= oo_objects(oo
);
1104 mod_zone_page_state(page_zone(page
),
1105 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1106 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1112 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1115 setup_object_debug(s
, page
, object
);
1116 if (unlikely(s
->ctor
))
1120 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1127 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1129 page
= allocate_slab(s
,
1130 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1134 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1136 page
->flags
|= 1 << PG_slab
;
1137 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1138 SLAB_STORE_USER
| SLAB_TRACE
))
1139 __SetPageSlubDebug(page
);
1141 start
= page_address(page
);
1143 if (unlikely(s
->flags
& SLAB_POISON
))
1144 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1147 for_each_object(p
, s
, start
, page
->objects
) {
1148 setup_object(s
, page
, last
);
1149 set_freepointer(s
, last
, p
);
1152 setup_object(s
, page
, last
);
1153 set_freepointer(s
, last
, NULL
);
1155 page
->freelist
= start
;
1161 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1163 int order
= compound_order(page
);
1164 int pages
= 1 << order
;
1166 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1169 slab_pad_check(s
, page
);
1170 for_each_object(p
, s
, page_address(page
),
1172 check_object(s
, page
, p
, 0);
1173 __ClearPageSlubDebug(page
);
1176 if (kmemcheck_page_is_tracked(page
))
1177 kmemcheck_free_shadow(s
, page
, compound_order(page
));
1179 mod_zone_page_state(page_zone(page
),
1180 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1181 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1184 __ClearPageSlab(page
);
1185 reset_page_mapcount(page
);
1186 if (current
->reclaim_state
)
1187 current
->reclaim_state
->reclaimed_slab
+= pages
;
1188 __free_pages(page
, order
);
1191 static void rcu_free_slab(struct rcu_head
*h
)
1195 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1196 __free_slab(page
->slab
, page
);
1199 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1201 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1203 * RCU free overloads the RCU head over the LRU
1205 struct rcu_head
*head
= (void *)&page
->lru
;
1207 call_rcu(head
, rcu_free_slab
);
1209 __free_slab(s
, page
);
1212 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1214 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1219 * Per slab locking using the pagelock
1221 static __always_inline
void slab_lock(struct page
*page
)
1223 bit_spin_lock(PG_locked
, &page
->flags
);
1226 static __always_inline
void slab_unlock(struct page
*page
)
1228 __bit_spin_unlock(PG_locked
, &page
->flags
);
1231 static __always_inline
int slab_trylock(struct page
*page
)
1235 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1240 * Management of partially allocated slabs
1242 static void add_partial(struct kmem_cache_node
*n
,
1243 struct page
*page
, int tail
)
1245 spin_lock(&n
->list_lock
);
1248 list_add_tail(&page
->lru
, &n
->partial
);
1250 list_add(&page
->lru
, &n
->partial
);
1251 spin_unlock(&n
->list_lock
);
1254 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1256 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1258 spin_lock(&n
->list_lock
);
1259 list_del(&page
->lru
);
1261 spin_unlock(&n
->list_lock
);
1265 * Lock slab and remove from the partial list.
1267 * Must hold list_lock.
1269 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1272 if (slab_trylock(page
)) {
1273 list_del(&page
->lru
);
1275 __SetPageSlubFrozen(page
);
1282 * Try to allocate a partial slab from a specific node.
1284 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1289 * Racy check. If we mistakenly see no partial slabs then we
1290 * just allocate an empty slab. If we mistakenly try to get a
1291 * partial slab and there is none available then get_partials()
1294 if (!n
|| !n
->nr_partial
)
1297 spin_lock(&n
->list_lock
);
1298 list_for_each_entry(page
, &n
->partial
, lru
)
1299 if (lock_and_freeze_slab(n
, page
))
1303 spin_unlock(&n
->list_lock
);
1308 * Get a page from somewhere. Search in increasing NUMA distances.
1310 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1313 struct zonelist
*zonelist
;
1316 enum zone_type high_zoneidx
= gfp_zone(flags
);
1320 * The defrag ratio allows a configuration of the tradeoffs between
1321 * inter node defragmentation and node local allocations. A lower
1322 * defrag_ratio increases the tendency to do local allocations
1323 * instead of attempting to obtain partial slabs from other nodes.
1325 * If the defrag_ratio is set to 0 then kmalloc() always
1326 * returns node local objects. If the ratio is higher then kmalloc()
1327 * may return off node objects because partial slabs are obtained
1328 * from other nodes and filled up.
1330 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1331 * defrag_ratio = 1000) then every (well almost) allocation will
1332 * first attempt to defrag slab caches on other nodes. This means
1333 * scanning over all nodes to look for partial slabs which may be
1334 * expensive if we do it every time we are trying to find a slab
1335 * with available objects.
1337 if (!s
->remote_node_defrag_ratio
||
1338 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1341 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1342 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1343 struct kmem_cache_node
*n
;
1345 n
= get_node(s
, zone_to_nid(zone
));
1347 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1348 n
->nr_partial
> s
->min_partial
) {
1349 page
= get_partial_node(n
);
1359 * Get a partial page, lock it and return it.
1361 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1364 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1366 page
= get_partial_node(get_node(s
, searchnode
));
1367 if (page
|| (flags
& __GFP_THISNODE
))
1370 return get_any_partial(s
, flags
);
1374 * Move a page back to the lists.
1376 * Must be called with the slab lock held.
1378 * On exit the slab lock will have been dropped.
1380 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1382 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1383 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1385 __ClearPageSlubFrozen(page
);
1388 if (page
->freelist
) {
1389 add_partial(n
, page
, tail
);
1390 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1392 stat(c
, DEACTIVATE_FULL
);
1393 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1394 (s
->flags
& SLAB_STORE_USER
))
1399 stat(c
, DEACTIVATE_EMPTY
);
1400 if (n
->nr_partial
< s
->min_partial
) {
1402 * Adding an empty slab to the partial slabs in order
1403 * to avoid page allocator overhead. This slab needs
1404 * to come after the other slabs with objects in
1405 * so that the others get filled first. That way the
1406 * size of the partial list stays small.
1408 * kmem_cache_shrink can reclaim any empty slabs from
1411 add_partial(n
, page
, 1);
1415 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1416 discard_slab(s
, page
);
1422 * Remove the cpu slab
1424 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1426 struct page
*page
= c
->page
;
1430 stat(c
, DEACTIVATE_REMOTE_FREES
);
1432 * Merge cpu freelist into slab freelist. Typically we get here
1433 * because both freelists are empty. So this is unlikely
1436 while (unlikely(c
->freelist
)) {
1439 tail
= 0; /* Hot objects. Put the slab first */
1441 /* Retrieve object from cpu_freelist */
1442 object
= c
->freelist
;
1443 c
->freelist
= c
->freelist
[c
->offset
];
1445 /* And put onto the regular freelist */
1446 object
[c
->offset
] = page
->freelist
;
1447 page
->freelist
= object
;
1451 unfreeze_slab(s
, page
, tail
);
1454 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1456 stat(c
, CPUSLAB_FLUSH
);
1458 deactivate_slab(s
, c
);
1464 * Called from IPI handler with interrupts disabled.
1466 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1468 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1470 if (likely(c
&& c
->page
))
1474 static void flush_cpu_slab(void *d
)
1476 struct kmem_cache
*s
= d
;
1478 __flush_cpu_slab(s
, smp_processor_id());
1481 static void flush_all(struct kmem_cache
*s
)
1483 on_each_cpu(flush_cpu_slab
, s
, 1);
1487 * Check if the objects in a per cpu structure fit numa
1488 * locality expectations.
1490 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1493 if (node
!= -1 && c
->node
!= node
)
1500 * Slow path. The lockless freelist is empty or we need to perform
1503 * Interrupts are disabled.
1505 * Processing is still very fast if new objects have been freed to the
1506 * regular freelist. In that case we simply take over the regular freelist
1507 * as the lockless freelist and zap the regular freelist.
1509 * If that is not working then we fall back to the partial lists. We take the
1510 * first element of the freelist as the object to allocate now and move the
1511 * rest of the freelist to the lockless freelist.
1513 * And if we were unable to get a new slab from the partial slab lists then
1514 * we need to allocate a new slab. This is the slowest path since it involves
1515 * a call to the page allocator and the setup of a new slab.
1517 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1518 unsigned long addr
, struct kmem_cache_cpu
*c
)
1523 /* We handle __GFP_ZERO in the caller */
1524 gfpflags
&= ~__GFP_ZERO
;
1530 if (unlikely(!node_match(c
, node
)))
1533 stat(c
, ALLOC_REFILL
);
1536 object
= c
->page
->freelist
;
1537 if (unlikely(!object
))
1539 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1542 c
->freelist
= object
[c
->offset
];
1543 c
->page
->inuse
= c
->page
->objects
;
1544 c
->page
->freelist
= NULL
;
1545 c
->node
= page_to_nid(c
->page
);
1547 slab_unlock(c
->page
);
1548 stat(c
, ALLOC_SLOWPATH
);
1552 deactivate_slab(s
, c
);
1555 new = get_partial(s
, gfpflags
, node
);
1558 stat(c
, ALLOC_FROM_PARTIAL
);
1562 if (gfpflags
& __GFP_WAIT
)
1565 new = new_slab(s
, gfpflags
, node
);
1567 if (gfpflags
& __GFP_WAIT
)
1568 local_irq_disable();
1571 c
= get_cpu_slab(s
, smp_processor_id());
1572 stat(c
, ALLOC_SLAB
);
1576 __SetPageSlubFrozen(new);
1582 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1586 c
->page
->freelist
= object
[c
->offset
];
1592 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1593 * have the fastpath folded into their functions. So no function call
1594 * overhead for requests that can be satisfied on the fastpath.
1596 * The fastpath works by first checking if the lockless freelist can be used.
1597 * If not then __slab_alloc is called for slow processing.
1599 * Otherwise we can simply pick the next object from the lockless free list.
1601 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1602 gfp_t gfpflags
, int node
, unsigned long addr
)
1605 struct kmem_cache_cpu
*c
;
1606 unsigned long flags
;
1607 unsigned int objsize
;
1609 lockdep_trace_alloc(gfpflags
);
1610 might_sleep_if(gfpflags
& __GFP_WAIT
);
1612 if (should_failslab(s
->objsize
, gfpflags
))
1615 local_irq_save(flags
);
1616 c
= get_cpu_slab(s
, smp_processor_id());
1617 objsize
= c
->objsize
;
1618 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1620 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1623 object
= c
->freelist
;
1624 c
->freelist
= object
[c
->offset
];
1625 stat(c
, ALLOC_FASTPATH
);
1627 local_irq_restore(flags
);
1629 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1630 memset(object
, 0, objsize
);
1632 kmemcheck_slab_alloc(s
, gfpflags
, object
, c
->objsize
);
1633 kmemleak_alloc_recursive(object
, objsize
, 1, s
->flags
, gfpflags
);
1638 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1640 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1642 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1646 EXPORT_SYMBOL(kmem_cache_alloc
);
1648 #ifdef CONFIG_KMEMTRACE
1649 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1651 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1653 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1657 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1659 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1661 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1662 s
->objsize
, s
->size
, gfpflags
, node
);
1666 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1669 #ifdef CONFIG_KMEMTRACE
1670 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1674 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1676 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1680 * Slow patch handling. This may still be called frequently since objects
1681 * have a longer lifetime than the cpu slabs in most processing loads.
1683 * So we still attempt to reduce cache line usage. Just take the slab
1684 * lock and free the item. If there is no additional partial page
1685 * handling required then we can return immediately.
1687 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1688 void *x
, unsigned long addr
, unsigned int offset
)
1691 void **object
= (void *)x
;
1692 struct kmem_cache_cpu
*c
;
1694 c
= get_cpu_slab(s
, raw_smp_processor_id());
1695 stat(c
, FREE_SLOWPATH
);
1698 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1702 prior
= object
[offset
] = page
->freelist
;
1703 page
->freelist
= object
;
1706 if (unlikely(PageSlubFrozen(page
))) {
1707 stat(c
, FREE_FROZEN
);
1711 if (unlikely(!page
->inuse
))
1715 * Objects left in the slab. If it was not on the partial list before
1718 if (unlikely(!prior
)) {
1719 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1720 stat(c
, FREE_ADD_PARTIAL
);
1730 * Slab still on the partial list.
1732 remove_partial(s
, page
);
1733 stat(c
, FREE_REMOVE_PARTIAL
);
1737 discard_slab(s
, page
);
1741 if (!free_debug_processing(s
, page
, x
, addr
))
1747 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1748 * can perform fastpath freeing without additional function calls.
1750 * The fastpath is only possible if we are freeing to the current cpu slab
1751 * of this processor. This typically the case if we have just allocated
1754 * If fastpath is not possible then fall back to __slab_free where we deal
1755 * with all sorts of special processing.
1757 static __always_inline
void slab_free(struct kmem_cache
*s
,
1758 struct page
*page
, void *x
, unsigned long addr
)
1760 void **object
= (void *)x
;
1761 struct kmem_cache_cpu
*c
;
1762 unsigned long flags
;
1764 kmemleak_free_recursive(x
, s
->flags
);
1765 local_irq_save(flags
);
1766 c
= get_cpu_slab(s
, smp_processor_id());
1767 kmemcheck_slab_free(s
, object
, c
->objsize
);
1768 debug_check_no_locks_freed(object
, c
->objsize
);
1769 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1770 debug_check_no_obj_freed(object
, c
->objsize
);
1771 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1772 object
[c
->offset
] = c
->freelist
;
1773 c
->freelist
= object
;
1774 stat(c
, FREE_FASTPATH
);
1776 __slab_free(s
, page
, x
, addr
, c
->offset
);
1778 local_irq_restore(flags
);
1781 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1785 page
= virt_to_head_page(x
);
1787 slab_free(s
, page
, x
, _RET_IP_
);
1789 trace_kmem_cache_free(_RET_IP_
, x
);
1791 EXPORT_SYMBOL(kmem_cache_free
);
1793 /* Figure out on which slab page the object resides */
1794 static struct page
*get_object_page(const void *x
)
1796 struct page
*page
= virt_to_head_page(x
);
1798 if (!PageSlab(page
))
1805 * Object placement in a slab is made very easy because we always start at
1806 * offset 0. If we tune the size of the object to the alignment then we can
1807 * get the required alignment by putting one properly sized object after
1810 * Notice that the allocation order determines the sizes of the per cpu
1811 * caches. Each processor has always one slab available for allocations.
1812 * Increasing the allocation order reduces the number of times that slabs
1813 * must be moved on and off the partial lists and is therefore a factor in
1818 * Mininum / Maximum order of slab pages. This influences locking overhead
1819 * and slab fragmentation. A higher order reduces the number of partial slabs
1820 * and increases the number of allocations possible without having to
1821 * take the list_lock.
1823 static int slub_min_order
;
1824 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1825 static int slub_min_objects
;
1828 * Merge control. If this is set then no merging of slab caches will occur.
1829 * (Could be removed. This was introduced to pacify the merge skeptics.)
1831 static int slub_nomerge
;
1834 * Calculate the order of allocation given an slab object size.
1836 * The order of allocation has significant impact on performance and other
1837 * system components. Generally order 0 allocations should be preferred since
1838 * order 0 does not cause fragmentation in the page allocator. Larger objects
1839 * be problematic to put into order 0 slabs because there may be too much
1840 * unused space left. We go to a higher order if more than 1/16th of the slab
1843 * In order to reach satisfactory performance we must ensure that a minimum
1844 * number of objects is in one slab. Otherwise we may generate too much
1845 * activity on the partial lists which requires taking the list_lock. This is
1846 * less a concern for large slabs though which are rarely used.
1848 * slub_max_order specifies the order where we begin to stop considering the
1849 * number of objects in a slab as critical. If we reach slub_max_order then
1850 * we try to keep the page order as low as possible. So we accept more waste
1851 * of space in favor of a small page order.
1853 * Higher order allocations also allow the placement of more objects in a
1854 * slab and thereby reduce object handling overhead. If the user has
1855 * requested a higher mininum order then we start with that one instead of
1856 * the smallest order which will fit the object.
1858 static inline int slab_order(int size
, int min_objects
,
1859 int max_order
, int fract_leftover
)
1863 int min_order
= slub_min_order
;
1865 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1866 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1868 for (order
= max(min_order
,
1869 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1870 order
<= max_order
; order
++) {
1872 unsigned long slab_size
= PAGE_SIZE
<< order
;
1874 if (slab_size
< min_objects
* size
)
1877 rem
= slab_size
% size
;
1879 if (rem
<= slab_size
/ fract_leftover
)
1887 static inline int calculate_order(int size
)
1895 * Attempt to find best configuration for a slab. This
1896 * works by first attempting to generate a layout with
1897 * the best configuration and backing off gradually.
1899 * First we reduce the acceptable waste in a slab. Then
1900 * we reduce the minimum objects required in a slab.
1902 min_objects
= slub_min_objects
;
1904 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1905 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1906 min_objects
= min(min_objects
, max_objects
);
1908 while (min_objects
> 1) {
1910 while (fraction
>= 4) {
1911 order
= slab_order(size
, min_objects
,
1912 slub_max_order
, fraction
);
1913 if (order
<= slub_max_order
)
1921 * We were unable to place multiple objects in a slab. Now
1922 * lets see if we can place a single object there.
1924 order
= slab_order(size
, 1, slub_max_order
, 1);
1925 if (order
<= slub_max_order
)
1929 * Doh this slab cannot be placed using slub_max_order.
1931 order
= slab_order(size
, 1, MAX_ORDER
, 1);
1932 if (order
< MAX_ORDER
)
1938 * Figure out what the alignment of the objects will be.
1940 static unsigned long calculate_alignment(unsigned long flags
,
1941 unsigned long align
, unsigned long size
)
1944 * If the user wants hardware cache aligned objects then follow that
1945 * suggestion if the object is sufficiently large.
1947 * The hardware cache alignment cannot override the specified
1948 * alignment though. If that is greater then use it.
1950 if (flags
& SLAB_HWCACHE_ALIGN
) {
1951 unsigned long ralign
= cache_line_size();
1952 while (size
<= ralign
/ 2)
1954 align
= max(align
, ralign
);
1957 if (align
< ARCH_SLAB_MINALIGN
)
1958 align
= ARCH_SLAB_MINALIGN
;
1960 return ALIGN(align
, sizeof(void *));
1963 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
1964 struct kmem_cache_cpu
*c
)
1969 c
->offset
= s
->offset
/ sizeof(void *);
1970 c
->objsize
= s
->objsize
;
1971 #ifdef CONFIG_SLUB_STATS
1972 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
1977 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
1980 spin_lock_init(&n
->list_lock
);
1981 INIT_LIST_HEAD(&n
->partial
);
1982 #ifdef CONFIG_SLUB_DEBUG
1983 atomic_long_set(&n
->nr_slabs
, 0);
1984 atomic_long_set(&n
->total_objects
, 0);
1985 INIT_LIST_HEAD(&n
->full
);
1991 * Per cpu array for per cpu structures.
1993 * The per cpu array places all kmem_cache_cpu structures from one processor
1994 * close together meaning that it becomes possible that multiple per cpu
1995 * structures are contained in one cacheline. This may be particularly
1996 * beneficial for the kmalloc caches.
1998 * A desktop system typically has around 60-80 slabs. With 100 here we are
1999 * likely able to get per cpu structures for all caches from the array defined
2000 * here. We must be able to cover all kmalloc caches during bootstrap.
2002 * If the per cpu array is exhausted then fall back to kmalloc
2003 * of individual cachelines. No sharing is possible then.
2005 #define NR_KMEM_CACHE_CPU 100
2007 static DEFINE_PER_CPU(struct kmem_cache_cpu
,
2008 kmem_cache_cpu
)[NR_KMEM_CACHE_CPU
];
2010 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
2011 static DECLARE_BITMAP(kmem_cach_cpu_free_init_once
, CONFIG_NR_CPUS
);
2013 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
2014 int cpu
, gfp_t flags
)
2016 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
2019 per_cpu(kmem_cache_cpu_free
, cpu
) =
2020 (void *)c
->freelist
;
2022 /* Table overflow: So allocate ourselves */
2024 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2025 flags
, cpu_to_node(cpu
));
2030 init_kmem_cache_cpu(s
, c
);
2034 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2036 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2037 c
>= per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2041 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2042 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2045 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2049 for_each_online_cpu(cpu
) {
2050 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2053 s
->cpu_slab
[cpu
] = NULL
;
2054 free_kmem_cache_cpu(c
, cpu
);
2059 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2063 for_each_online_cpu(cpu
) {
2064 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2069 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2071 free_kmem_cache_cpus(s
);
2074 s
->cpu_slab
[cpu
] = c
;
2080 * Initialize the per cpu array.
2082 static void init_alloc_cpu_cpu(int cpu
)
2086 if (cpumask_test_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
)))
2089 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2090 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2092 cpumask_set_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
));
2095 static void __init
init_alloc_cpu(void)
2099 for_each_online_cpu(cpu
)
2100 init_alloc_cpu_cpu(cpu
);
2104 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2105 static inline void init_alloc_cpu(void) {}
2107 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2109 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2116 * No kmalloc_node yet so do it by hand. We know that this is the first
2117 * slab on the node for this slabcache. There are no concurrent accesses
2120 * Note that this function only works on the kmalloc_node_cache
2121 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2122 * memory on a fresh node that has no slab structures yet.
2124 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2127 struct kmem_cache_node
*n
;
2128 unsigned long flags
;
2130 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2132 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2135 if (page_to_nid(page
) != node
) {
2136 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2138 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2139 "in order to be able to continue\n");
2144 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2146 kmalloc_caches
->node
[node
] = n
;
2147 #ifdef CONFIG_SLUB_DEBUG
2148 init_object(kmalloc_caches
, n
, 1);
2149 init_tracking(kmalloc_caches
, n
);
2151 init_kmem_cache_node(n
, kmalloc_caches
);
2152 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2155 * lockdep requires consistent irq usage for each lock
2156 * so even though there cannot be a race this early in
2157 * the boot sequence, we still disable irqs.
2159 local_irq_save(flags
);
2160 add_partial(n
, page
, 0);
2161 local_irq_restore(flags
);
2164 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2168 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2169 struct kmem_cache_node
*n
= s
->node
[node
];
2170 if (n
&& n
!= &s
->local_node
)
2171 kmem_cache_free(kmalloc_caches
, n
);
2172 s
->node
[node
] = NULL
;
2176 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2181 if (slab_state
>= UP
)
2182 local_node
= page_to_nid(virt_to_page(s
));
2186 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2187 struct kmem_cache_node
*n
;
2189 if (local_node
== node
)
2192 if (slab_state
== DOWN
) {
2193 early_kmem_cache_node_alloc(gfpflags
, node
);
2196 n
= kmem_cache_alloc_node(kmalloc_caches
,
2200 free_kmem_cache_nodes(s
);
2206 init_kmem_cache_node(n
, s
);
2211 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2215 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2217 init_kmem_cache_node(&s
->local_node
, s
);
2222 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2224 if (min
< MIN_PARTIAL
)
2226 else if (min
> MAX_PARTIAL
)
2228 s
->min_partial
= min
;
2232 * calculate_sizes() determines the order and the distribution of data within
2235 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2237 unsigned long flags
= s
->flags
;
2238 unsigned long size
= s
->objsize
;
2239 unsigned long align
= s
->align
;
2243 * Round up object size to the next word boundary. We can only
2244 * place the free pointer at word boundaries and this determines
2245 * the possible location of the free pointer.
2247 size
= ALIGN(size
, sizeof(void *));
2249 #ifdef CONFIG_SLUB_DEBUG
2251 * Determine if we can poison the object itself. If the user of
2252 * the slab may touch the object after free or before allocation
2253 * then we should never poison the object itself.
2255 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2257 s
->flags
|= __OBJECT_POISON
;
2259 s
->flags
&= ~__OBJECT_POISON
;
2263 * If we are Redzoning then check if there is some space between the
2264 * end of the object and the free pointer. If not then add an
2265 * additional word to have some bytes to store Redzone information.
2267 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2268 size
+= sizeof(void *);
2272 * With that we have determined the number of bytes in actual use
2273 * by the object. This is the potential offset to the free pointer.
2277 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2280 * Relocate free pointer after the object if it is not
2281 * permitted to overwrite the first word of the object on
2284 * This is the case if we do RCU, have a constructor or
2285 * destructor or are poisoning the objects.
2288 size
+= sizeof(void *);
2291 #ifdef CONFIG_SLUB_DEBUG
2292 if (flags
& SLAB_STORE_USER
)
2294 * Need to store information about allocs and frees after
2297 size
+= 2 * sizeof(struct track
);
2299 if (flags
& SLAB_RED_ZONE
)
2301 * Add some empty padding so that we can catch
2302 * overwrites from earlier objects rather than let
2303 * tracking information or the free pointer be
2304 * corrupted if a user writes before the start
2307 size
+= sizeof(void *);
2311 * Determine the alignment based on various parameters that the
2312 * user specified and the dynamic determination of cache line size
2315 align
= calculate_alignment(flags
, align
, s
->objsize
);
2318 * SLUB stores one object immediately after another beginning from
2319 * offset 0. In order to align the objects we have to simply size
2320 * each object to conform to the alignment.
2322 size
= ALIGN(size
, align
);
2324 if (forced_order
>= 0)
2325 order
= forced_order
;
2327 order
= calculate_order(size
);
2334 s
->allocflags
|= __GFP_COMP
;
2336 if (s
->flags
& SLAB_CACHE_DMA
)
2337 s
->allocflags
|= SLUB_DMA
;
2339 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2340 s
->allocflags
|= __GFP_RECLAIMABLE
;
2343 * Determine the number of objects per slab
2345 s
->oo
= oo_make(order
, size
);
2346 s
->min
= oo_make(get_order(size
), size
);
2347 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2350 return !!oo_objects(s
->oo
);
2354 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2355 const char *name
, size_t size
,
2356 size_t align
, unsigned long flags
,
2357 void (*ctor
)(void *))
2359 memset(s
, 0, kmem_size
);
2364 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2366 if (!calculate_sizes(s
, -1))
2370 * The larger the object size is, the more pages we want on the partial
2371 * list to avoid pounding the page allocator excessively.
2373 set_min_partial(s
, ilog2(s
->size
));
2376 s
->remote_node_defrag_ratio
= 1000;
2378 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2381 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2383 free_kmem_cache_nodes(s
);
2385 if (flags
& SLAB_PANIC
)
2386 panic("Cannot create slab %s size=%lu realsize=%u "
2387 "order=%u offset=%u flags=%lx\n",
2388 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2394 * Check if a given pointer is valid
2396 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2400 page
= get_object_page(object
);
2402 if (!page
|| s
!= page
->slab
)
2403 /* No slab or wrong slab */
2406 if (!check_valid_pointer(s
, page
, object
))
2410 * We could also check if the object is on the slabs freelist.
2411 * But this would be too expensive and it seems that the main
2412 * purpose of kmem_ptr_valid() is to check if the object belongs
2413 * to a certain slab.
2417 EXPORT_SYMBOL(kmem_ptr_validate
);
2420 * Determine the size of a slab object
2422 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2426 EXPORT_SYMBOL(kmem_cache_size
);
2428 const char *kmem_cache_name(struct kmem_cache
*s
)
2432 EXPORT_SYMBOL(kmem_cache_name
);
2434 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2437 #ifdef CONFIG_SLUB_DEBUG
2438 void *addr
= page_address(page
);
2440 DECLARE_BITMAP(map
, page
->objects
);
2442 bitmap_zero(map
, page
->objects
);
2443 slab_err(s
, page
, "%s", text
);
2445 for_each_free_object(p
, s
, page
->freelist
)
2446 set_bit(slab_index(p
, s
, addr
), map
);
2448 for_each_object(p
, s
, addr
, page
->objects
) {
2450 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2451 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2453 print_tracking(s
, p
);
2461 * Attempt to free all partial slabs on a node.
2463 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2465 unsigned long flags
;
2466 struct page
*page
, *h
;
2468 spin_lock_irqsave(&n
->list_lock
, flags
);
2469 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2471 list_del(&page
->lru
);
2472 discard_slab(s
, page
);
2475 list_slab_objects(s
, page
,
2476 "Objects remaining on kmem_cache_close()");
2479 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2483 * Release all resources used by a slab cache.
2485 static inline int kmem_cache_close(struct kmem_cache
*s
)
2491 /* Attempt to free all objects */
2492 free_kmem_cache_cpus(s
);
2493 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2494 struct kmem_cache_node
*n
= get_node(s
, node
);
2497 if (n
->nr_partial
|| slabs_node(s
, node
))
2500 free_kmem_cache_nodes(s
);
2505 * Close a cache and release the kmem_cache structure
2506 * (must be used for caches created using kmem_cache_create)
2508 void kmem_cache_destroy(struct kmem_cache
*s
)
2510 down_write(&slub_lock
);
2514 up_write(&slub_lock
);
2515 if (kmem_cache_close(s
)) {
2516 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2517 "still has objects.\n", s
->name
, __func__
);
2520 sysfs_slab_remove(s
);
2522 up_write(&slub_lock
);
2524 EXPORT_SYMBOL(kmem_cache_destroy
);
2526 /********************************************************************
2528 *******************************************************************/
2530 struct kmem_cache kmalloc_caches
[SLUB_PAGE_SHIFT
] __cacheline_aligned
;
2531 EXPORT_SYMBOL(kmalloc_caches
);
2533 static int __init
setup_slub_min_order(char *str
)
2535 get_option(&str
, &slub_min_order
);
2540 __setup("slub_min_order=", setup_slub_min_order
);
2542 static int __init
setup_slub_max_order(char *str
)
2544 get_option(&str
, &slub_max_order
);
2545 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2550 __setup("slub_max_order=", setup_slub_max_order
);
2552 static int __init
setup_slub_min_objects(char *str
)
2554 get_option(&str
, &slub_min_objects
);
2559 __setup("slub_min_objects=", setup_slub_min_objects
);
2561 static int __init
setup_slub_nomerge(char *str
)
2567 __setup("slub_nomerge", setup_slub_nomerge
);
2569 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2570 const char *name
, int size
, gfp_t gfp_flags
)
2572 unsigned int flags
= 0;
2574 if (gfp_flags
& SLUB_DMA
)
2575 flags
= SLAB_CACHE_DMA
;
2578 * This function is called with IRQs disabled during early-boot on
2579 * single CPU so there's no need to take slub_lock here.
2581 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2585 list_add(&s
->list
, &slab_caches
);
2587 if (sysfs_slab_add(s
))
2592 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2595 #ifdef CONFIG_ZONE_DMA
2596 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2598 static void sysfs_add_func(struct work_struct
*w
)
2600 struct kmem_cache
*s
;
2602 down_write(&slub_lock
);
2603 list_for_each_entry(s
, &slab_caches
, list
) {
2604 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2605 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2609 up_write(&slub_lock
);
2612 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2614 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2616 struct kmem_cache
*s
;
2620 s
= kmalloc_caches_dma
[index
];
2624 /* Dynamically create dma cache */
2625 if (flags
& __GFP_WAIT
)
2626 down_write(&slub_lock
);
2628 if (!down_write_trylock(&slub_lock
))
2632 if (kmalloc_caches_dma
[index
])
2635 realsize
= kmalloc_caches
[index
].objsize
;
2636 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2637 (unsigned int)realsize
);
2638 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2640 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2641 realsize
, ARCH_KMALLOC_MINALIGN
,
2642 SLAB_CACHE_DMA
|SLAB_NOTRACK
|__SYSFS_ADD_DEFERRED
,
2649 list_add(&s
->list
, &slab_caches
);
2650 kmalloc_caches_dma
[index
] = s
;
2652 schedule_work(&sysfs_add_work
);
2655 up_write(&slub_lock
);
2657 return kmalloc_caches_dma
[index
];
2662 * Conversion table for small slabs sizes / 8 to the index in the
2663 * kmalloc array. This is necessary for slabs < 192 since we have non power
2664 * of two cache sizes there. The size of larger slabs can be determined using
2667 static s8 size_index
[24] = {
2694 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2700 return ZERO_SIZE_PTR
;
2702 index
= size_index
[(size
- 1) / 8];
2704 index
= fls(size
- 1);
2706 #ifdef CONFIG_ZONE_DMA
2707 if (unlikely((flags
& SLUB_DMA
)))
2708 return dma_kmalloc_cache(index
, flags
);
2711 return &kmalloc_caches
[index
];
2714 void *__kmalloc(size_t size
, gfp_t flags
)
2716 struct kmem_cache
*s
;
2719 if (unlikely(size
> SLUB_MAX_SIZE
))
2720 return kmalloc_large(size
, flags
);
2722 s
= get_slab(size
, flags
);
2724 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2727 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2729 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2733 EXPORT_SYMBOL(__kmalloc
);
2735 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2737 struct page
*page
= alloc_pages_node(node
, flags
| __GFP_COMP
,
2741 return page_address(page
);
2747 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2749 struct kmem_cache
*s
;
2752 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2753 ret
= kmalloc_large_node(size
, flags
, node
);
2755 trace_kmalloc_node(_RET_IP_
, ret
,
2756 size
, PAGE_SIZE
<< get_order(size
),
2762 s
= get_slab(size
, flags
);
2764 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2767 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2769 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2773 EXPORT_SYMBOL(__kmalloc_node
);
2776 size_t ksize(const void *object
)
2779 struct kmem_cache
*s
;
2781 if (unlikely(object
== ZERO_SIZE_PTR
))
2784 page
= virt_to_head_page(object
);
2786 if (unlikely(!PageSlab(page
))) {
2787 WARN_ON(!PageCompound(page
));
2788 return PAGE_SIZE
<< compound_order(page
);
2792 #ifdef CONFIG_SLUB_DEBUG
2794 * Debugging requires use of the padding between object
2795 * and whatever may come after it.
2797 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2802 * If we have the need to store the freelist pointer
2803 * back there or track user information then we can
2804 * only use the space before that information.
2806 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2809 * Else we can use all the padding etc for the allocation
2813 EXPORT_SYMBOL(ksize
);
2815 void kfree(const void *x
)
2818 void *object
= (void *)x
;
2820 trace_kfree(_RET_IP_
, x
);
2822 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2825 page
= virt_to_head_page(x
);
2826 if (unlikely(!PageSlab(page
))) {
2827 BUG_ON(!PageCompound(page
));
2831 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2833 EXPORT_SYMBOL(kfree
);
2836 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2837 * the remaining slabs by the number of items in use. The slabs with the
2838 * most items in use come first. New allocations will then fill those up
2839 * and thus they can be removed from the partial lists.
2841 * The slabs with the least items are placed last. This results in them
2842 * being allocated from last increasing the chance that the last objects
2843 * are freed in them.
2845 int kmem_cache_shrink(struct kmem_cache
*s
)
2849 struct kmem_cache_node
*n
;
2852 int objects
= oo_objects(s
->max
);
2853 struct list_head
*slabs_by_inuse
=
2854 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2855 unsigned long flags
;
2857 if (!slabs_by_inuse
)
2861 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2862 n
= get_node(s
, node
);
2867 for (i
= 0; i
< objects
; i
++)
2868 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2870 spin_lock_irqsave(&n
->list_lock
, flags
);
2873 * Build lists indexed by the items in use in each slab.
2875 * Note that concurrent frees may occur while we hold the
2876 * list_lock. page->inuse here is the upper limit.
2878 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2879 if (!page
->inuse
&& slab_trylock(page
)) {
2881 * Must hold slab lock here because slab_free
2882 * may have freed the last object and be
2883 * waiting to release the slab.
2885 list_del(&page
->lru
);
2888 discard_slab(s
, page
);
2890 list_move(&page
->lru
,
2891 slabs_by_inuse
+ page
->inuse
);
2896 * Rebuild the partial list with the slabs filled up most
2897 * first and the least used slabs at the end.
2899 for (i
= objects
- 1; i
>= 0; i
--)
2900 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2902 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2905 kfree(slabs_by_inuse
);
2908 EXPORT_SYMBOL(kmem_cache_shrink
);
2910 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2911 static int slab_mem_going_offline_callback(void *arg
)
2913 struct kmem_cache
*s
;
2915 down_read(&slub_lock
);
2916 list_for_each_entry(s
, &slab_caches
, list
)
2917 kmem_cache_shrink(s
);
2918 up_read(&slub_lock
);
2923 static void slab_mem_offline_callback(void *arg
)
2925 struct kmem_cache_node
*n
;
2926 struct kmem_cache
*s
;
2927 struct memory_notify
*marg
= arg
;
2930 offline_node
= marg
->status_change_nid
;
2933 * If the node still has available memory. we need kmem_cache_node
2936 if (offline_node
< 0)
2939 down_read(&slub_lock
);
2940 list_for_each_entry(s
, &slab_caches
, list
) {
2941 n
= get_node(s
, offline_node
);
2944 * if n->nr_slabs > 0, slabs still exist on the node
2945 * that is going down. We were unable to free them,
2946 * and offline_pages() function shoudn't call this
2947 * callback. So, we must fail.
2949 BUG_ON(slabs_node(s
, offline_node
));
2951 s
->node
[offline_node
] = NULL
;
2952 kmem_cache_free(kmalloc_caches
, n
);
2955 up_read(&slub_lock
);
2958 static int slab_mem_going_online_callback(void *arg
)
2960 struct kmem_cache_node
*n
;
2961 struct kmem_cache
*s
;
2962 struct memory_notify
*marg
= arg
;
2963 int nid
= marg
->status_change_nid
;
2967 * If the node's memory is already available, then kmem_cache_node is
2968 * already created. Nothing to do.
2974 * We are bringing a node online. No memory is available yet. We must
2975 * allocate a kmem_cache_node structure in order to bring the node
2978 down_read(&slub_lock
);
2979 list_for_each_entry(s
, &slab_caches
, list
) {
2981 * XXX: kmem_cache_alloc_node will fallback to other nodes
2982 * since memory is not yet available from the node that
2985 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2990 init_kmem_cache_node(n
, s
);
2994 up_read(&slub_lock
);
2998 static int slab_memory_callback(struct notifier_block
*self
,
2999 unsigned long action
, void *arg
)
3004 case MEM_GOING_ONLINE
:
3005 ret
= slab_mem_going_online_callback(arg
);
3007 case MEM_GOING_OFFLINE
:
3008 ret
= slab_mem_going_offline_callback(arg
);
3011 case MEM_CANCEL_ONLINE
:
3012 slab_mem_offline_callback(arg
);
3015 case MEM_CANCEL_OFFLINE
:
3019 ret
= notifier_from_errno(ret
);
3025 #endif /* CONFIG_MEMORY_HOTPLUG */
3027 /********************************************************************
3028 * Basic setup of slabs
3029 *******************************************************************/
3031 void __init
kmem_cache_init(void)
3040 * Must first have the slab cache available for the allocations of the
3041 * struct kmem_cache_node's. There is special bootstrap code in
3042 * kmem_cache_open for slab_state == DOWN.
3044 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3045 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3046 kmalloc_caches
[0].refcount
= -1;
3049 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3052 /* Able to allocate the per node structures */
3053 slab_state
= PARTIAL
;
3055 /* Caches that are not of the two-to-the-power-of size */
3056 if (KMALLOC_MIN_SIZE
<= 64) {
3057 create_kmalloc_cache(&kmalloc_caches
[1],
3058 "kmalloc-96", 96, GFP_NOWAIT
);
3060 create_kmalloc_cache(&kmalloc_caches
[2],
3061 "kmalloc-192", 192, GFP_NOWAIT
);
3065 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3066 create_kmalloc_cache(&kmalloc_caches
[i
],
3067 "kmalloc", 1 << i
, GFP_NOWAIT
);
3073 * Patch up the size_index table if we have strange large alignment
3074 * requirements for the kmalloc array. This is only the case for
3075 * MIPS it seems. The standard arches will not generate any code here.
3077 * Largest permitted alignment is 256 bytes due to the way we
3078 * handle the index determination for the smaller caches.
3080 * Make sure that nothing crazy happens if someone starts tinkering
3081 * around with ARCH_KMALLOC_MINALIGN
3083 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3084 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3086 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3087 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3089 if (KMALLOC_MIN_SIZE
== 128) {
3091 * The 192 byte sized cache is not used if the alignment
3092 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3095 for (i
= 128 + 8; i
<= 192; i
+= 8)
3096 size_index
[(i
- 1) / 8] = 8;
3101 /* Provide the correct kmalloc names now that the caches are up */
3102 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3103 kmalloc_caches
[i
]. name
=
3104 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3107 register_cpu_notifier(&slab_notifier
);
3108 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3109 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3111 kmem_size
= sizeof(struct kmem_cache
);
3115 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3116 " CPUs=%d, Nodes=%d\n",
3117 caches
, cache_line_size(),
3118 slub_min_order
, slub_max_order
, slub_min_objects
,
3119 nr_cpu_ids
, nr_node_ids
);
3123 * Find a mergeable slab cache
3125 static int slab_unmergeable(struct kmem_cache
*s
)
3127 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3134 * We may have set a slab to be unmergeable during bootstrap.
3136 if (s
->refcount
< 0)
3142 static struct kmem_cache
*find_mergeable(size_t size
,
3143 size_t align
, unsigned long flags
, const char *name
,
3144 void (*ctor
)(void *))
3146 struct kmem_cache
*s
;
3148 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3154 size
= ALIGN(size
, sizeof(void *));
3155 align
= calculate_alignment(flags
, align
, size
);
3156 size
= ALIGN(size
, align
);
3157 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3159 list_for_each_entry(s
, &slab_caches
, list
) {
3160 if (slab_unmergeable(s
))
3166 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3169 * Check if alignment is compatible.
3170 * Courtesy of Adrian Drzewiecki
3172 if ((s
->size
& ~(align
- 1)) != s
->size
)
3175 if (s
->size
- size
>= sizeof(void *))
3183 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3184 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3186 struct kmem_cache
*s
;
3188 down_write(&slub_lock
);
3189 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3195 * Adjust the object sizes so that we clear
3196 * the complete object on kzalloc.
3198 s
->objsize
= max(s
->objsize
, (int)size
);
3201 * And then we need to update the object size in the
3202 * per cpu structures
3204 for_each_online_cpu(cpu
)
3205 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3207 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3208 up_write(&slub_lock
);
3210 if (sysfs_slab_alias(s
, name
)) {
3211 down_write(&slub_lock
);
3213 up_write(&slub_lock
);
3219 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3221 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3222 size
, align
, flags
, ctor
)) {
3223 list_add(&s
->list
, &slab_caches
);
3224 up_write(&slub_lock
);
3225 if (sysfs_slab_add(s
)) {
3226 down_write(&slub_lock
);
3228 up_write(&slub_lock
);
3236 up_write(&slub_lock
);
3239 if (flags
& SLAB_PANIC
)
3240 panic("Cannot create slabcache %s\n", name
);
3245 EXPORT_SYMBOL(kmem_cache_create
);
3249 * Use the cpu notifier to insure that the cpu slabs are flushed when
3252 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3253 unsigned long action
, void *hcpu
)
3255 long cpu
= (long)hcpu
;
3256 struct kmem_cache
*s
;
3257 unsigned long flags
;
3260 case CPU_UP_PREPARE
:
3261 case CPU_UP_PREPARE_FROZEN
:
3262 init_alloc_cpu_cpu(cpu
);
3263 down_read(&slub_lock
);
3264 list_for_each_entry(s
, &slab_caches
, list
)
3265 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3267 up_read(&slub_lock
);
3270 case CPU_UP_CANCELED
:
3271 case CPU_UP_CANCELED_FROZEN
:
3273 case CPU_DEAD_FROZEN
:
3274 down_read(&slub_lock
);
3275 list_for_each_entry(s
, &slab_caches
, list
) {
3276 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3278 local_irq_save(flags
);
3279 __flush_cpu_slab(s
, cpu
);
3280 local_irq_restore(flags
);
3281 free_kmem_cache_cpu(c
, cpu
);
3282 s
->cpu_slab
[cpu
] = NULL
;
3284 up_read(&slub_lock
);
3292 static struct notifier_block __cpuinitdata slab_notifier
= {
3293 .notifier_call
= slab_cpuup_callback
3298 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3300 struct kmem_cache
*s
;
3303 if (unlikely(size
> SLUB_MAX_SIZE
))
3304 return kmalloc_large(size
, gfpflags
);
3306 s
= get_slab(size
, gfpflags
);
3308 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3311 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3313 /* Honor the call site pointer we recieved. */
3314 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3319 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3320 int node
, unsigned long caller
)
3322 struct kmem_cache
*s
;
3325 if (unlikely(size
> SLUB_MAX_SIZE
))
3326 return kmalloc_large_node(size
, gfpflags
, node
);
3328 s
= get_slab(size
, gfpflags
);
3330 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3333 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3335 /* Honor the call site pointer we recieved. */
3336 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3341 #ifdef CONFIG_SLUB_DEBUG
3342 static unsigned long count_partial(struct kmem_cache_node
*n
,
3343 int (*get_count
)(struct page
*))
3345 unsigned long flags
;
3346 unsigned long x
= 0;
3349 spin_lock_irqsave(&n
->list_lock
, flags
);
3350 list_for_each_entry(page
, &n
->partial
, lru
)
3351 x
+= get_count(page
);
3352 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3356 static int count_inuse(struct page
*page
)
3361 static int count_total(struct page
*page
)
3363 return page
->objects
;
3366 static int count_free(struct page
*page
)
3368 return page
->objects
- page
->inuse
;
3371 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3375 void *addr
= page_address(page
);
3377 if (!check_slab(s
, page
) ||
3378 !on_freelist(s
, page
, NULL
))
3381 /* Now we know that a valid freelist exists */
3382 bitmap_zero(map
, page
->objects
);
3384 for_each_free_object(p
, s
, page
->freelist
) {
3385 set_bit(slab_index(p
, s
, addr
), map
);
3386 if (!check_object(s
, page
, p
, 0))
3390 for_each_object(p
, s
, addr
, page
->objects
)
3391 if (!test_bit(slab_index(p
, s
, addr
), map
))
3392 if (!check_object(s
, page
, p
, 1))
3397 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3400 if (slab_trylock(page
)) {
3401 validate_slab(s
, page
, map
);
3404 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3407 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3408 if (!PageSlubDebug(page
))
3409 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3410 "on slab 0x%p\n", s
->name
, page
);
3412 if (PageSlubDebug(page
))
3413 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3414 "slab 0x%p\n", s
->name
, page
);
3418 static int validate_slab_node(struct kmem_cache
*s
,
3419 struct kmem_cache_node
*n
, unsigned long *map
)
3421 unsigned long count
= 0;
3423 unsigned long flags
;
3425 spin_lock_irqsave(&n
->list_lock
, flags
);
3427 list_for_each_entry(page
, &n
->partial
, lru
) {
3428 validate_slab_slab(s
, page
, map
);
3431 if (count
!= n
->nr_partial
)
3432 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3433 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3435 if (!(s
->flags
& SLAB_STORE_USER
))
3438 list_for_each_entry(page
, &n
->full
, lru
) {
3439 validate_slab_slab(s
, page
, map
);
3442 if (count
!= atomic_long_read(&n
->nr_slabs
))
3443 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3444 "counter=%ld\n", s
->name
, count
,
3445 atomic_long_read(&n
->nr_slabs
));
3448 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3452 static long validate_slab_cache(struct kmem_cache
*s
)
3455 unsigned long count
= 0;
3456 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3457 sizeof(unsigned long), GFP_KERNEL
);
3463 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3464 struct kmem_cache_node
*n
= get_node(s
, node
);
3466 count
+= validate_slab_node(s
, n
, map
);
3472 #ifdef SLUB_RESILIENCY_TEST
3473 static void resiliency_test(void)
3477 printk(KERN_ERR
"SLUB resiliency testing\n");
3478 printk(KERN_ERR
"-----------------------\n");
3479 printk(KERN_ERR
"A. Corruption after allocation\n");
3481 p
= kzalloc(16, GFP_KERNEL
);
3483 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3484 " 0x12->0x%p\n\n", p
+ 16);
3486 validate_slab_cache(kmalloc_caches
+ 4);
3488 /* Hmmm... The next two are dangerous */
3489 p
= kzalloc(32, GFP_KERNEL
);
3490 p
[32 + sizeof(void *)] = 0x34;
3491 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3492 " 0x34 -> -0x%p\n", p
);
3494 "If allocated object is overwritten then not detectable\n\n");
3496 validate_slab_cache(kmalloc_caches
+ 5);
3497 p
= kzalloc(64, GFP_KERNEL
);
3498 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3500 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3503 "If allocated object is overwritten then not detectable\n\n");
3504 validate_slab_cache(kmalloc_caches
+ 6);
3506 printk(KERN_ERR
"\nB. Corruption after free\n");
3507 p
= kzalloc(128, GFP_KERNEL
);
3510 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3511 validate_slab_cache(kmalloc_caches
+ 7);
3513 p
= kzalloc(256, GFP_KERNEL
);
3516 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3518 validate_slab_cache(kmalloc_caches
+ 8);
3520 p
= kzalloc(512, GFP_KERNEL
);
3523 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3524 validate_slab_cache(kmalloc_caches
+ 9);
3527 static void resiliency_test(void) {};
3531 * Generate lists of code addresses where slabcache objects are allocated
3536 unsigned long count
;
3543 DECLARE_BITMAP(cpus
, NR_CPUS
);
3549 unsigned long count
;
3550 struct location
*loc
;
3553 static void free_loc_track(struct loc_track
*t
)
3556 free_pages((unsigned long)t
->loc
,
3557 get_order(sizeof(struct location
) * t
->max
));
3560 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3565 order
= get_order(sizeof(struct location
) * max
);
3567 l
= (void *)__get_free_pages(flags
, order
);
3572 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3580 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3581 const struct track
*track
)
3583 long start
, end
, pos
;
3585 unsigned long caddr
;
3586 unsigned long age
= jiffies
- track
->when
;
3592 pos
= start
+ (end
- start
+ 1) / 2;
3595 * There is nothing at "end". If we end up there
3596 * we need to add something to before end.
3601 caddr
= t
->loc
[pos
].addr
;
3602 if (track
->addr
== caddr
) {
3608 if (age
< l
->min_time
)
3610 if (age
> l
->max_time
)
3613 if (track
->pid
< l
->min_pid
)
3614 l
->min_pid
= track
->pid
;
3615 if (track
->pid
> l
->max_pid
)
3616 l
->max_pid
= track
->pid
;
3618 cpumask_set_cpu(track
->cpu
,
3619 to_cpumask(l
->cpus
));
3621 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3625 if (track
->addr
< caddr
)
3632 * Not found. Insert new tracking element.
3634 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3640 (t
->count
- pos
) * sizeof(struct location
));
3643 l
->addr
= track
->addr
;
3647 l
->min_pid
= track
->pid
;
3648 l
->max_pid
= track
->pid
;
3649 cpumask_clear(to_cpumask(l
->cpus
));
3650 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3651 nodes_clear(l
->nodes
);
3652 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3656 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3657 struct page
*page
, enum track_item alloc
)
3659 void *addr
= page_address(page
);
3660 DECLARE_BITMAP(map
, page
->objects
);
3663 bitmap_zero(map
, page
->objects
);
3664 for_each_free_object(p
, s
, page
->freelist
)
3665 set_bit(slab_index(p
, s
, addr
), map
);
3667 for_each_object(p
, s
, addr
, page
->objects
)
3668 if (!test_bit(slab_index(p
, s
, addr
), map
))
3669 add_location(t
, s
, get_track(s
, p
, alloc
));
3672 static int list_locations(struct kmem_cache
*s
, char *buf
,
3673 enum track_item alloc
)
3677 struct loc_track t
= { 0, 0, NULL
};
3680 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3682 return sprintf(buf
, "Out of memory\n");
3684 /* Push back cpu slabs */
3687 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3688 struct kmem_cache_node
*n
= get_node(s
, node
);
3689 unsigned long flags
;
3692 if (!atomic_long_read(&n
->nr_slabs
))
3695 spin_lock_irqsave(&n
->list_lock
, flags
);
3696 list_for_each_entry(page
, &n
->partial
, lru
)
3697 process_slab(&t
, s
, page
, alloc
);
3698 list_for_each_entry(page
, &n
->full
, lru
)
3699 process_slab(&t
, s
, page
, alloc
);
3700 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3703 for (i
= 0; i
< t
.count
; i
++) {
3704 struct location
*l
= &t
.loc
[i
];
3706 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3708 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3711 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3713 len
+= sprintf(buf
+ len
, "<not-available>");
3715 if (l
->sum_time
!= l
->min_time
) {
3716 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3718 (long)div_u64(l
->sum_time
, l
->count
),
3721 len
+= sprintf(buf
+ len
, " age=%ld",
3724 if (l
->min_pid
!= l
->max_pid
)
3725 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3726 l
->min_pid
, l
->max_pid
);
3728 len
+= sprintf(buf
+ len
, " pid=%ld",
3731 if (num_online_cpus() > 1 &&
3732 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3733 len
< PAGE_SIZE
- 60) {
3734 len
+= sprintf(buf
+ len
, " cpus=");
3735 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3736 to_cpumask(l
->cpus
));
3739 if (num_online_nodes() > 1 && !nodes_empty(l
->nodes
) &&
3740 len
< PAGE_SIZE
- 60) {
3741 len
+= sprintf(buf
+ len
, " nodes=");
3742 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3746 len
+= sprintf(buf
+ len
, "\n");
3751 len
+= sprintf(buf
, "No data\n");
3755 enum slab_stat_type
{
3756 SL_ALL
, /* All slabs */
3757 SL_PARTIAL
, /* Only partially allocated slabs */
3758 SL_CPU
, /* Only slabs used for cpu caches */
3759 SL_OBJECTS
, /* Determine allocated objects not slabs */
3760 SL_TOTAL
/* Determine object capacity not slabs */
3763 #define SO_ALL (1 << SL_ALL)
3764 #define SO_PARTIAL (1 << SL_PARTIAL)
3765 #define SO_CPU (1 << SL_CPU)
3766 #define SO_OBJECTS (1 << SL_OBJECTS)
3767 #define SO_TOTAL (1 << SL_TOTAL)
3769 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3770 char *buf
, unsigned long flags
)
3772 unsigned long total
= 0;
3775 unsigned long *nodes
;
3776 unsigned long *per_cpu
;
3778 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3781 per_cpu
= nodes
+ nr_node_ids
;
3783 if (flags
& SO_CPU
) {
3786 for_each_possible_cpu(cpu
) {
3787 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3789 if (!c
|| c
->node
< 0)
3793 if (flags
& SO_TOTAL
)
3794 x
= c
->page
->objects
;
3795 else if (flags
& SO_OBJECTS
)
3801 nodes
[c
->node
] += x
;
3807 if (flags
& SO_ALL
) {
3808 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3809 struct kmem_cache_node
*n
= get_node(s
, node
);
3811 if (flags
& SO_TOTAL
)
3812 x
= atomic_long_read(&n
->total_objects
);
3813 else if (flags
& SO_OBJECTS
)
3814 x
= atomic_long_read(&n
->total_objects
) -
3815 count_partial(n
, count_free
);
3818 x
= atomic_long_read(&n
->nr_slabs
);
3823 } else if (flags
& SO_PARTIAL
) {
3824 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3825 struct kmem_cache_node
*n
= get_node(s
, node
);
3827 if (flags
& SO_TOTAL
)
3828 x
= count_partial(n
, count_total
);
3829 else if (flags
& SO_OBJECTS
)
3830 x
= count_partial(n
, count_inuse
);
3837 x
= sprintf(buf
, "%lu", total
);
3839 for_each_node_state(node
, N_NORMAL_MEMORY
)
3841 x
+= sprintf(buf
+ x
, " N%d=%lu",
3845 return x
+ sprintf(buf
+ x
, "\n");
3848 static int any_slab_objects(struct kmem_cache
*s
)
3852 for_each_online_node(node
) {
3853 struct kmem_cache_node
*n
= get_node(s
, node
);
3858 if (atomic_long_read(&n
->total_objects
))
3864 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3865 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3867 struct slab_attribute
{
3868 struct attribute attr
;
3869 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3870 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3873 #define SLAB_ATTR_RO(_name) \
3874 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3876 #define SLAB_ATTR(_name) \
3877 static struct slab_attribute _name##_attr = \
3878 __ATTR(_name, 0644, _name##_show, _name##_store)
3880 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3882 return sprintf(buf
, "%d\n", s
->size
);
3884 SLAB_ATTR_RO(slab_size
);
3886 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3888 return sprintf(buf
, "%d\n", s
->align
);
3890 SLAB_ATTR_RO(align
);
3892 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3894 return sprintf(buf
, "%d\n", s
->objsize
);
3896 SLAB_ATTR_RO(object_size
);
3898 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3900 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3902 SLAB_ATTR_RO(objs_per_slab
);
3904 static ssize_t
order_store(struct kmem_cache
*s
,
3905 const char *buf
, size_t length
)
3907 unsigned long order
;
3910 err
= strict_strtoul(buf
, 10, &order
);
3914 if (order
> slub_max_order
|| order
< slub_min_order
)
3917 calculate_sizes(s
, order
);
3921 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3923 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3927 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3929 return sprintf(buf
, "%lu\n", s
->min_partial
);
3932 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3938 err
= strict_strtoul(buf
, 10, &min
);
3942 set_min_partial(s
, min
);
3945 SLAB_ATTR(min_partial
);
3947 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3950 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3952 return n
+ sprintf(buf
+ n
, "\n");
3958 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3960 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3962 SLAB_ATTR_RO(aliases
);
3964 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3966 return show_slab_objects(s
, buf
, SO_ALL
);
3968 SLAB_ATTR_RO(slabs
);
3970 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3972 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3974 SLAB_ATTR_RO(partial
);
3976 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3978 return show_slab_objects(s
, buf
, SO_CPU
);
3980 SLAB_ATTR_RO(cpu_slabs
);
3982 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3984 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3986 SLAB_ATTR_RO(objects
);
3988 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3990 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3992 SLAB_ATTR_RO(objects_partial
);
3994 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3996 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3998 SLAB_ATTR_RO(total_objects
);
4000 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4002 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4005 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4006 const char *buf
, size_t length
)
4008 s
->flags
&= ~SLAB_DEBUG_FREE
;
4010 s
->flags
|= SLAB_DEBUG_FREE
;
4013 SLAB_ATTR(sanity_checks
);
4015 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4017 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4020 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4023 s
->flags
&= ~SLAB_TRACE
;
4025 s
->flags
|= SLAB_TRACE
;
4030 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4032 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4035 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4036 const char *buf
, size_t length
)
4038 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4040 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4043 SLAB_ATTR(reclaim_account
);
4045 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4047 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4049 SLAB_ATTR_RO(hwcache_align
);
4051 #ifdef CONFIG_ZONE_DMA
4052 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4054 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4056 SLAB_ATTR_RO(cache_dma
);
4059 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4061 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4063 SLAB_ATTR_RO(destroy_by_rcu
);
4065 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4067 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4070 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4071 const char *buf
, size_t length
)
4073 if (any_slab_objects(s
))
4076 s
->flags
&= ~SLAB_RED_ZONE
;
4078 s
->flags
|= SLAB_RED_ZONE
;
4079 calculate_sizes(s
, -1);
4082 SLAB_ATTR(red_zone
);
4084 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4086 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4089 static ssize_t
poison_store(struct kmem_cache
*s
,
4090 const char *buf
, size_t length
)
4092 if (any_slab_objects(s
))
4095 s
->flags
&= ~SLAB_POISON
;
4097 s
->flags
|= SLAB_POISON
;
4098 calculate_sizes(s
, -1);
4103 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4105 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4108 static ssize_t
store_user_store(struct kmem_cache
*s
,
4109 const char *buf
, size_t length
)
4111 if (any_slab_objects(s
))
4114 s
->flags
&= ~SLAB_STORE_USER
;
4116 s
->flags
|= SLAB_STORE_USER
;
4117 calculate_sizes(s
, -1);
4120 SLAB_ATTR(store_user
);
4122 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4127 static ssize_t
validate_store(struct kmem_cache
*s
,
4128 const char *buf
, size_t length
)
4132 if (buf
[0] == '1') {
4133 ret
= validate_slab_cache(s
);
4139 SLAB_ATTR(validate
);
4141 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4146 static ssize_t
shrink_store(struct kmem_cache
*s
,
4147 const char *buf
, size_t length
)
4149 if (buf
[0] == '1') {
4150 int rc
= kmem_cache_shrink(s
);
4160 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4162 if (!(s
->flags
& SLAB_STORE_USER
))
4164 return list_locations(s
, buf
, TRACK_ALLOC
);
4166 SLAB_ATTR_RO(alloc_calls
);
4168 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4170 if (!(s
->flags
& SLAB_STORE_USER
))
4172 return list_locations(s
, buf
, TRACK_FREE
);
4174 SLAB_ATTR_RO(free_calls
);
4177 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4179 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4182 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4183 const char *buf
, size_t length
)
4185 unsigned long ratio
;
4188 err
= strict_strtoul(buf
, 10, &ratio
);
4193 s
->remote_node_defrag_ratio
= ratio
* 10;
4197 SLAB_ATTR(remote_node_defrag_ratio
);
4200 #ifdef CONFIG_SLUB_STATS
4201 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4203 unsigned long sum
= 0;
4206 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4211 for_each_online_cpu(cpu
) {
4212 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4218 len
= sprintf(buf
, "%lu", sum
);
4221 for_each_online_cpu(cpu
) {
4222 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4223 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4227 return len
+ sprintf(buf
+ len
, "\n");
4230 #define STAT_ATTR(si, text) \
4231 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4233 return show_stat(s, buf, si); \
4235 SLAB_ATTR_RO(text); \
4237 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4238 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4239 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4240 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4241 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4242 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4243 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4244 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4245 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4246 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4247 STAT_ATTR(FREE_SLAB
, free_slab
);
4248 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4249 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4250 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4251 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4252 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4253 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4254 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4257 static struct attribute
*slab_attrs
[] = {
4258 &slab_size_attr
.attr
,
4259 &object_size_attr
.attr
,
4260 &objs_per_slab_attr
.attr
,
4262 &min_partial_attr
.attr
,
4264 &objects_partial_attr
.attr
,
4265 &total_objects_attr
.attr
,
4268 &cpu_slabs_attr
.attr
,
4272 &sanity_checks_attr
.attr
,
4274 &hwcache_align_attr
.attr
,
4275 &reclaim_account_attr
.attr
,
4276 &destroy_by_rcu_attr
.attr
,
4277 &red_zone_attr
.attr
,
4279 &store_user_attr
.attr
,
4280 &validate_attr
.attr
,
4282 &alloc_calls_attr
.attr
,
4283 &free_calls_attr
.attr
,
4284 #ifdef CONFIG_ZONE_DMA
4285 &cache_dma_attr
.attr
,
4288 &remote_node_defrag_ratio_attr
.attr
,
4290 #ifdef CONFIG_SLUB_STATS
4291 &alloc_fastpath_attr
.attr
,
4292 &alloc_slowpath_attr
.attr
,
4293 &free_fastpath_attr
.attr
,
4294 &free_slowpath_attr
.attr
,
4295 &free_frozen_attr
.attr
,
4296 &free_add_partial_attr
.attr
,
4297 &free_remove_partial_attr
.attr
,
4298 &alloc_from_partial_attr
.attr
,
4299 &alloc_slab_attr
.attr
,
4300 &alloc_refill_attr
.attr
,
4301 &free_slab_attr
.attr
,
4302 &cpuslab_flush_attr
.attr
,
4303 &deactivate_full_attr
.attr
,
4304 &deactivate_empty_attr
.attr
,
4305 &deactivate_to_head_attr
.attr
,
4306 &deactivate_to_tail_attr
.attr
,
4307 &deactivate_remote_frees_attr
.attr
,
4308 &order_fallback_attr
.attr
,
4313 static struct attribute_group slab_attr_group
= {
4314 .attrs
= slab_attrs
,
4317 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4318 struct attribute
*attr
,
4321 struct slab_attribute
*attribute
;
4322 struct kmem_cache
*s
;
4325 attribute
= to_slab_attr(attr
);
4328 if (!attribute
->show
)
4331 err
= attribute
->show(s
, buf
);
4336 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4337 struct attribute
*attr
,
4338 const char *buf
, size_t len
)
4340 struct slab_attribute
*attribute
;
4341 struct kmem_cache
*s
;
4344 attribute
= to_slab_attr(attr
);
4347 if (!attribute
->store
)
4350 err
= attribute
->store(s
, buf
, len
);
4355 static void kmem_cache_release(struct kobject
*kobj
)
4357 struct kmem_cache
*s
= to_slab(kobj
);
4362 static struct sysfs_ops slab_sysfs_ops
= {
4363 .show
= slab_attr_show
,
4364 .store
= slab_attr_store
,
4367 static struct kobj_type slab_ktype
= {
4368 .sysfs_ops
= &slab_sysfs_ops
,
4369 .release
= kmem_cache_release
4372 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4374 struct kobj_type
*ktype
= get_ktype(kobj
);
4376 if (ktype
== &slab_ktype
)
4381 static struct kset_uevent_ops slab_uevent_ops
= {
4382 .filter
= uevent_filter
,
4385 static struct kset
*slab_kset
;
4387 #define ID_STR_LENGTH 64
4389 /* Create a unique string id for a slab cache:
4391 * Format :[flags-]size
4393 static char *create_unique_id(struct kmem_cache
*s
)
4395 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4402 * First flags affecting slabcache operations. We will only
4403 * get here for aliasable slabs so we do not need to support
4404 * too many flags. The flags here must cover all flags that
4405 * are matched during merging to guarantee that the id is
4408 if (s
->flags
& SLAB_CACHE_DMA
)
4410 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4412 if (s
->flags
& SLAB_DEBUG_FREE
)
4414 if (!(s
->flags
& SLAB_NOTRACK
))
4418 p
+= sprintf(p
, "%07d", s
->size
);
4419 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4423 static int sysfs_slab_add(struct kmem_cache
*s
)
4429 if (slab_state
< SYSFS
)
4430 /* Defer until later */
4433 unmergeable
= slab_unmergeable(s
);
4436 * Slabcache can never be merged so we can use the name proper.
4437 * This is typically the case for debug situations. In that
4438 * case we can catch duplicate names easily.
4440 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4444 * Create a unique name for the slab as a target
4447 name
= create_unique_id(s
);
4450 s
->kobj
.kset
= slab_kset
;
4451 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4453 kobject_put(&s
->kobj
);
4457 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4460 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4462 /* Setup first alias */
4463 sysfs_slab_alias(s
, s
->name
);
4469 static void sysfs_slab_remove(struct kmem_cache
*s
)
4471 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4472 kobject_del(&s
->kobj
);
4473 kobject_put(&s
->kobj
);
4477 * Need to buffer aliases during bootup until sysfs becomes
4478 * available lest we lose that information.
4480 struct saved_alias
{
4481 struct kmem_cache
*s
;
4483 struct saved_alias
*next
;
4486 static struct saved_alias
*alias_list
;
4488 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4490 struct saved_alias
*al
;
4492 if (slab_state
== SYSFS
) {
4494 * If we have a leftover link then remove it.
4496 sysfs_remove_link(&slab_kset
->kobj
, name
);
4497 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4500 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4506 al
->next
= alias_list
;
4511 static int __init
slab_sysfs_init(void)
4513 struct kmem_cache
*s
;
4516 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4518 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4524 list_for_each_entry(s
, &slab_caches
, list
) {
4525 err
= sysfs_slab_add(s
);
4527 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4528 " to sysfs\n", s
->name
);
4531 while (alias_list
) {
4532 struct saved_alias
*al
= alias_list
;
4534 alias_list
= alias_list
->next
;
4535 err
= sysfs_slab_alias(al
->s
, al
->name
);
4537 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4538 " %s to sysfs\n", s
->name
);
4546 __initcall(slab_sysfs_init
);
4550 * The /proc/slabinfo ABI
4552 #ifdef CONFIG_SLABINFO
4553 static void print_slabinfo_header(struct seq_file
*m
)
4555 seq_puts(m
, "slabinfo - version: 2.1\n");
4556 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4557 "<objperslab> <pagesperslab>");
4558 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4559 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4563 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4567 down_read(&slub_lock
);
4569 print_slabinfo_header(m
);
4571 return seq_list_start(&slab_caches
, *pos
);
4574 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4576 return seq_list_next(p
, &slab_caches
, pos
);
4579 static void s_stop(struct seq_file
*m
, void *p
)
4581 up_read(&slub_lock
);
4584 static int s_show(struct seq_file
*m
, void *p
)
4586 unsigned long nr_partials
= 0;
4587 unsigned long nr_slabs
= 0;
4588 unsigned long nr_inuse
= 0;
4589 unsigned long nr_objs
= 0;
4590 unsigned long nr_free
= 0;
4591 struct kmem_cache
*s
;
4594 s
= list_entry(p
, struct kmem_cache
, list
);
4596 for_each_online_node(node
) {
4597 struct kmem_cache_node
*n
= get_node(s
, node
);
4602 nr_partials
+= n
->nr_partial
;
4603 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4604 nr_objs
+= atomic_long_read(&n
->total_objects
);
4605 nr_free
+= count_partial(n
, count_free
);
4608 nr_inuse
= nr_objs
- nr_free
;
4610 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4611 nr_objs
, s
->size
, oo_objects(s
->oo
),
4612 (1 << oo_order(s
->oo
)));
4613 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4614 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4620 static const struct seq_operations slabinfo_op
= {
4627 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4629 return seq_open(file
, &slabinfo_op
);
4632 static const struct file_operations proc_slabinfo_operations
= {
4633 .open
= slabinfo_open
,
4635 .llseek
= seq_lseek
,
4636 .release
= seq_release
,
4639 static int __init
slab_proc_init(void)
4641 proc_create("slabinfo",S_IWUSR
|S_IRUGO
,NULL
,&proc_slabinfo_operations
);
4644 module_init(slab_proc_init
);
4645 #endif /* CONFIG_SLABINFO */