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 unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
839 return atomic_long_read(&n
->nr_slabs
);
842 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
844 struct kmem_cache_node
*n
= get_node(s
, node
);
847 * May be called early in order to allocate a slab for the
848 * kmem_cache_node structure. Solve the chicken-egg
849 * dilemma by deferring the increment of the count during
850 * bootstrap (see early_kmem_cache_node_alloc).
852 if (!NUMA_BUILD
|| n
) {
853 atomic_long_inc(&n
->nr_slabs
);
854 atomic_long_add(objects
, &n
->total_objects
);
857 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
859 struct kmem_cache_node
*n
= get_node(s
, node
);
861 atomic_long_dec(&n
->nr_slabs
);
862 atomic_long_sub(objects
, &n
->total_objects
);
865 /* Object debug checks for alloc/free paths */
866 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
869 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
872 init_object(s
, object
, 0);
873 init_tracking(s
, object
);
876 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
877 void *object
, unsigned long addr
)
879 if (!check_slab(s
, page
))
882 if (!on_freelist(s
, page
, object
)) {
883 object_err(s
, page
, object
, "Object already allocated");
887 if (!check_valid_pointer(s
, page
, object
)) {
888 object_err(s
, page
, object
, "Freelist Pointer check fails");
892 if (!check_object(s
, page
, object
, 0))
895 /* Success perform special debug activities for allocs */
896 if (s
->flags
& SLAB_STORE_USER
)
897 set_track(s
, object
, TRACK_ALLOC
, addr
);
898 trace(s
, page
, object
, 1);
899 init_object(s
, object
, 1);
903 if (PageSlab(page
)) {
905 * If this is a slab page then lets do the best we can
906 * to avoid issues in the future. Marking all objects
907 * as used avoids touching the remaining objects.
909 slab_fix(s
, "Marking all objects used");
910 page
->inuse
= page
->objects
;
911 page
->freelist
= NULL
;
916 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
917 void *object
, unsigned long addr
)
919 if (!check_slab(s
, page
))
922 if (!check_valid_pointer(s
, page
, object
)) {
923 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
927 if (on_freelist(s
, page
, object
)) {
928 object_err(s
, page
, object
, "Object already free");
932 if (!check_object(s
, page
, object
, 1))
935 if (unlikely(s
!= page
->slab
)) {
936 if (!PageSlab(page
)) {
937 slab_err(s
, page
, "Attempt to free object(0x%p) "
938 "outside of slab", object
);
939 } else if (!page
->slab
) {
941 "SLUB <none>: no slab for object 0x%p.\n",
945 object_err(s
, page
, object
,
946 "page slab pointer corrupt.");
950 /* Special debug activities for freeing objects */
951 if (!PageSlubFrozen(page
) && !page
->freelist
)
952 remove_full(s
, page
);
953 if (s
->flags
& SLAB_STORE_USER
)
954 set_track(s
, object
, TRACK_FREE
, addr
);
955 trace(s
, page
, object
, 0);
956 init_object(s
, object
, 0);
960 slab_fix(s
, "Object at 0x%p not freed", object
);
964 static int __init
setup_slub_debug(char *str
)
966 slub_debug
= DEBUG_DEFAULT_FLAGS
;
967 if (*str
++ != '=' || !*str
)
969 * No options specified. Switch on full debugging.
975 * No options but restriction on slabs. This means full
976 * debugging for slabs matching a pattern.
983 * Switch off all debugging measures.
988 * Determine which debug features should be switched on
990 for (; *str
&& *str
!= ','; str
++) {
991 switch (tolower(*str
)) {
993 slub_debug
|= SLAB_DEBUG_FREE
;
996 slub_debug
|= SLAB_RED_ZONE
;
999 slub_debug
|= SLAB_POISON
;
1002 slub_debug
|= SLAB_STORE_USER
;
1005 slub_debug
|= SLAB_TRACE
;
1008 printk(KERN_ERR
"slub_debug option '%c' "
1009 "unknown. skipped\n", *str
);
1015 slub_debug_slabs
= str
+ 1;
1020 __setup("slub_debug", setup_slub_debug
);
1022 static unsigned long kmem_cache_flags(unsigned long objsize
,
1023 unsigned long flags
, const char *name
,
1024 void (*ctor
)(void *))
1027 * Enable debugging if selected on the kernel commandline.
1029 if (slub_debug
&& (!slub_debug_slabs
||
1030 strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)) == 0))
1031 flags
|= slub_debug
;
1036 static inline void setup_object_debug(struct kmem_cache
*s
,
1037 struct page
*page
, void *object
) {}
1039 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1040 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1042 static inline int free_debug_processing(struct kmem_cache
*s
,
1043 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1045 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1047 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1048 void *object
, int active
) { return 1; }
1049 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1050 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1051 unsigned long flags
, const char *name
,
1052 void (*ctor
)(void *))
1056 #define slub_debug 0
1058 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1060 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1062 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1064 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1069 * Slab allocation and freeing
1071 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1072 struct kmem_cache_order_objects oo
)
1074 int order
= oo_order(oo
);
1076 flags
|= __GFP_NOTRACK
;
1079 return alloc_pages(flags
, order
);
1081 return alloc_pages_node(node
, flags
, order
);
1084 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1087 struct kmem_cache_order_objects oo
= s
->oo
;
1089 flags
|= s
->allocflags
;
1091 page
= alloc_slab_page(flags
| __GFP_NOWARN
| __GFP_NORETRY
, node
,
1093 if (unlikely(!page
)) {
1096 * Allocation may have failed due to fragmentation.
1097 * Try a lower order alloc if possible
1099 page
= alloc_slab_page(flags
, node
, oo
);
1103 stat(get_cpu_slab(s
, raw_smp_processor_id()), ORDER_FALLBACK
);
1106 if (kmemcheck_enabled
1107 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
)))
1109 int pages
= 1 << oo_order(oo
);
1111 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1114 * Objects from caches that have a constructor don't get
1115 * cleared when they're allocated, so we need to do it here.
1118 kmemcheck_mark_uninitialized_pages(page
, pages
);
1120 kmemcheck_mark_unallocated_pages(page
, pages
);
1123 page
->objects
= oo_objects(oo
);
1124 mod_zone_page_state(page_zone(page
),
1125 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1126 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1132 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1135 setup_object_debug(s
, page
, object
);
1136 if (unlikely(s
->ctor
))
1140 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1147 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1149 page
= allocate_slab(s
,
1150 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1154 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1156 page
->flags
|= 1 << PG_slab
;
1157 if (s
->flags
& (SLAB_DEBUG_FREE
| SLAB_RED_ZONE
| SLAB_POISON
|
1158 SLAB_STORE_USER
| SLAB_TRACE
))
1159 __SetPageSlubDebug(page
);
1161 start
= page_address(page
);
1163 if (unlikely(s
->flags
& SLAB_POISON
))
1164 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1167 for_each_object(p
, s
, start
, page
->objects
) {
1168 setup_object(s
, page
, last
);
1169 set_freepointer(s
, last
, p
);
1172 setup_object(s
, page
, last
);
1173 set_freepointer(s
, last
, NULL
);
1175 page
->freelist
= start
;
1181 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1183 int order
= compound_order(page
);
1184 int pages
= 1 << order
;
1186 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
))) {
1189 slab_pad_check(s
, page
);
1190 for_each_object(p
, s
, page_address(page
),
1192 check_object(s
, page
, p
, 0);
1193 __ClearPageSlubDebug(page
);
1196 kmemcheck_free_shadow(page
, compound_order(page
));
1198 mod_zone_page_state(page_zone(page
),
1199 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1200 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1203 __ClearPageSlab(page
);
1204 reset_page_mapcount(page
);
1205 if (current
->reclaim_state
)
1206 current
->reclaim_state
->reclaimed_slab
+= pages
;
1207 __free_pages(page
, order
);
1210 static void rcu_free_slab(struct rcu_head
*h
)
1214 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1215 __free_slab(page
->slab
, page
);
1218 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1220 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1222 * RCU free overloads the RCU head over the LRU
1224 struct rcu_head
*head
= (void *)&page
->lru
;
1226 call_rcu(head
, rcu_free_slab
);
1228 __free_slab(s
, page
);
1231 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1233 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1238 * Per slab locking using the pagelock
1240 static __always_inline
void slab_lock(struct page
*page
)
1242 bit_spin_lock(PG_locked
, &page
->flags
);
1245 static __always_inline
void slab_unlock(struct page
*page
)
1247 __bit_spin_unlock(PG_locked
, &page
->flags
);
1250 static __always_inline
int slab_trylock(struct page
*page
)
1254 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1259 * Management of partially allocated slabs
1261 static void add_partial(struct kmem_cache_node
*n
,
1262 struct page
*page
, int tail
)
1264 spin_lock(&n
->list_lock
);
1267 list_add_tail(&page
->lru
, &n
->partial
);
1269 list_add(&page
->lru
, &n
->partial
);
1270 spin_unlock(&n
->list_lock
);
1273 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1275 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1277 spin_lock(&n
->list_lock
);
1278 list_del(&page
->lru
);
1280 spin_unlock(&n
->list_lock
);
1284 * Lock slab and remove from the partial list.
1286 * Must hold list_lock.
1288 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1291 if (slab_trylock(page
)) {
1292 list_del(&page
->lru
);
1294 __SetPageSlubFrozen(page
);
1301 * Try to allocate a partial slab from a specific node.
1303 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1308 * Racy check. If we mistakenly see no partial slabs then we
1309 * just allocate an empty slab. If we mistakenly try to get a
1310 * partial slab and there is none available then get_partials()
1313 if (!n
|| !n
->nr_partial
)
1316 spin_lock(&n
->list_lock
);
1317 list_for_each_entry(page
, &n
->partial
, lru
)
1318 if (lock_and_freeze_slab(n
, page
))
1322 spin_unlock(&n
->list_lock
);
1327 * Get a page from somewhere. Search in increasing NUMA distances.
1329 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1332 struct zonelist
*zonelist
;
1335 enum zone_type high_zoneidx
= gfp_zone(flags
);
1339 * The defrag ratio allows a configuration of the tradeoffs between
1340 * inter node defragmentation and node local allocations. A lower
1341 * defrag_ratio increases the tendency to do local allocations
1342 * instead of attempting to obtain partial slabs from other nodes.
1344 * If the defrag_ratio is set to 0 then kmalloc() always
1345 * returns node local objects. If the ratio is higher then kmalloc()
1346 * may return off node objects because partial slabs are obtained
1347 * from other nodes and filled up.
1349 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1350 * defrag_ratio = 1000) then every (well almost) allocation will
1351 * first attempt to defrag slab caches on other nodes. This means
1352 * scanning over all nodes to look for partial slabs which may be
1353 * expensive if we do it every time we are trying to find a slab
1354 * with available objects.
1356 if (!s
->remote_node_defrag_ratio
||
1357 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1360 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1361 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1362 struct kmem_cache_node
*n
;
1364 n
= get_node(s
, zone_to_nid(zone
));
1366 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1367 n
->nr_partial
> s
->min_partial
) {
1368 page
= get_partial_node(n
);
1378 * Get a partial page, lock it and return it.
1380 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1383 int searchnode
= (node
== -1) ? numa_node_id() : node
;
1385 page
= get_partial_node(get_node(s
, searchnode
));
1386 if (page
|| (flags
& __GFP_THISNODE
))
1389 return get_any_partial(s
, flags
);
1393 * Move a page back to the lists.
1395 * Must be called with the slab lock held.
1397 * On exit the slab lock will have been dropped.
1399 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1401 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1402 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, smp_processor_id());
1404 __ClearPageSlubFrozen(page
);
1407 if (page
->freelist
) {
1408 add_partial(n
, page
, tail
);
1409 stat(c
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1411 stat(c
, DEACTIVATE_FULL
);
1412 if (SLABDEBUG
&& PageSlubDebug(page
) &&
1413 (s
->flags
& SLAB_STORE_USER
))
1418 stat(c
, DEACTIVATE_EMPTY
);
1419 if (n
->nr_partial
< s
->min_partial
) {
1421 * Adding an empty slab to the partial slabs in order
1422 * to avoid page allocator overhead. This slab needs
1423 * to come after the other slabs with objects in
1424 * so that the others get filled first. That way the
1425 * size of the partial list stays small.
1427 * kmem_cache_shrink can reclaim any empty slabs from
1430 add_partial(n
, page
, 1);
1434 stat(get_cpu_slab(s
, raw_smp_processor_id()), FREE_SLAB
);
1435 discard_slab(s
, page
);
1441 * Remove the cpu slab
1443 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1445 struct page
*page
= c
->page
;
1449 stat(c
, DEACTIVATE_REMOTE_FREES
);
1451 * Merge cpu freelist into slab freelist. Typically we get here
1452 * because both freelists are empty. So this is unlikely
1455 while (unlikely(c
->freelist
)) {
1458 tail
= 0; /* Hot objects. Put the slab first */
1460 /* Retrieve object from cpu_freelist */
1461 object
= c
->freelist
;
1462 c
->freelist
= c
->freelist
[c
->offset
];
1464 /* And put onto the regular freelist */
1465 object
[c
->offset
] = page
->freelist
;
1466 page
->freelist
= object
;
1470 unfreeze_slab(s
, page
, tail
);
1473 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1475 stat(c
, CPUSLAB_FLUSH
);
1477 deactivate_slab(s
, c
);
1483 * Called from IPI handler with interrupts disabled.
1485 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1487 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
1489 if (likely(c
&& c
->page
))
1493 static void flush_cpu_slab(void *d
)
1495 struct kmem_cache
*s
= d
;
1497 __flush_cpu_slab(s
, smp_processor_id());
1500 static void flush_all(struct kmem_cache
*s
)
1502 on_each_cpu(flush_cpu_slab
, s
, 1);
1506 * Check if the objects in a per cpu structure fit numa
1507 * locality expectations.
1509 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1512 if (node
!= -1 && c
->node
!= node
)
1518 static int count_free(struct page
*page
)
1520 return page
->objects
- page
->inuse
;
1523 static unsigned long count_partial(struct kmem_cache_node
*n
,
1524 int (*get_count
)(struct page
*))
1526 unsigned long flags
;
1527 unsigned long x
= 0;
1530 spin_lock_irqsave(&n
->list_lock
, flags
);
1531 list_for_each_entry(page
, &n
->partial
, lru
)
1532 x
+= get_count(page
);
1533 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1537 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1539 #ifdef CONFIG_SLUB_DEBUG
1540 return atomic_long_read(&n
->total_objects
);
1546 static noinline
void
1547 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1552 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1554 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1555 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1556 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1558 for_each_online_node(node
) {
1559 struct kmem_cache_node
*n
= get_node(s
, node
);
1560 unsigned long nr_slabs
;
1561 unsigned long nr_objs
;
1562 unsigned long nr_free
;
1567 nr_free
= count_partial(n
, count_free
);
1568 nr_slabs
= node_nr_slabs(n
);
1569 nr_objs
= node_nr_objs(n
);
1572 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1573 node
, nr_slabs
, nr_objs
, nr_free
);
1578 * Slow path. The lockless freelist is empty or we need to perform
1581 * Interrupts are disabled.
1583 * Processing is still very fast if new objects have been freed to the
1584 * regular freelist. In that case we simply take over the regular freelist
1585 * as the lockless freelist and zap the regular freelist.
1587 * If that is not working then we fall back to the partial lists. We take the
1588 * first element of the freelist as the object to allocate now and move the
1589 * rest of the freelist to the lockless freelist.
1591 * And if we were unable to get a new slab from the partial slab lists then
1592 * we need to allocate a new slab. This is the slowest path since it involves
1593 * a call to the page allocator and the setup of a new slab.
1595 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1596 unsigned long addr
, struct kmem_cache_cpu
*c
)
1601 /* We handle __GFP_ZERO in the caller */
1602 gfpflags
&= ~__GFP_ZERO
;
1608 if (unlikely(!node_match(c
, node
)))
1611 stat(c
, ALLOC_REFILL
);
1614 object
= c
->page
->freelist
;
1615 if (unlikely(!object
))
1617 if (unlikely(SLABDEBUG
&& PageSlubDebug(c
->page
)))
1620 c
->freelist
= object
[c
->offset
];
1621 c
->page
->inuse
= c
->page
->objects
;
1622 c
->page
->freelist
= NULL
;
1623 c
->node
= page_to_nid(c
->page
);
1625 slab_unlock(c
->page
);
1626 stat(c
, ALLOC_SLOWPATH
);
1630 deactivate_slab(s
, c
);
1633 new = get_partial(s
, gfpflags
, node
);
1636 stat(c
, ALLOC_FROM_PARTIAL
);
1640 if (gfpflags
& __GFP_WAIT
)
1643 new = new_slab(s
, gfpflags
, node
);
1645 if (gfpflags
& __GFP_WAIT
)
1646 local_irq_disable();
1649 c
= get_cpu_slab(s
, smp_processor_id());
1650 stat(c
, ALLOC_SLAB
);
1654 __SetPageSlubFrozen(new);
1658 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1659 slab_out_of_memory(s
, gfpflags
, node
);
1662 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1666 c
->page
->freelist
= object
[c
->offset
];
1672 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1673 * have the fastpath folded into their functions. So no function call
1674 * overhead for requests that can be satisfied on the fastpath.
1676 * The fastpath works by first checking if the lockless freelist can be used.
1677 * If not then __slab_alloc is called for slow processing.
1679 * Otherwise we can simply pick the next object from the lockless free list.
1681 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1682 gfp_t gfpflags
, int node
, unsigned long addr
)
1685 struct kmem_cache_cpu
*c
;
1686 unsigned long flags
;
1687 unsigned int objsize
;
1689 gfpflags
&= gfp_allowed_mask
;
1691 lockdep_trace_alloc(gfpflags
);
1692 might_sleep_if(gfpflags
& __GFP_WAIT
);
1694 if (should_failslab(s
->objsize
, gfpflags
))
1697 local_irq_save(flags
);
1698 c
= get_cpu_slab(s
, smp_processor_id());
1699 objsize
= c
->objsize
;
1700 if (unlikely(!c
->freelist
|| !node_match(c
, node
)))
1702 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1705 object
= c
->freelist
;
1706 c
->freelist
= object
[c
->offset
];
1707 stat(c
, ALLOC_FASTPATH
);
1709 local_irq_restore(flags
);
1711 if (unlikely((gfpflags
& __GFP_ZERO
) && object
))
1712 memset(object
, 0, objsize
);
1714 kmemcheck_slab_alloc(s
, gfpflags
, object
, c
->objsize
);
1715 kmemleak_alloc_recursive(object
, objsize
, 1, s
->flags
, gfpflags
);
1720 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1722 void *ret
= slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1724 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1728 EXPORT_SYMBOL(kmem_cache_alloc
);
1730 #ifdef CONFIG_KMEMTRACE
1731 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1733 return slab_alloc(s
, gfpflags
, -1, _RET_IP_
);
1735 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1739 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1741 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1743 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1744 s
->objsize
, s
->size
, gfpflags
, node
);
1748 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1751 #ifdef CONFIG_KMEMTRACE
1752 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1756 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1758 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1762 * Slow patch handling. This may still be called frequently since objects
1763 * have a longer lifetime than the cpu slabs in most processing loads.
1765 * So we still attempt to reduce cache line usage. Just take the slab
1766 * lock and free the item. If there is no additional partial page
1767 * handling required then we can return immediately.
1769 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
1770 void *x
, unsigned long addr
, unsigned int offset
)
1773 void **object
= (void *)x
;
1774 struct kmem_cache_cpu
*c
;
1776 c
= get_cpu_slab(s
, raw_smp_processor_id());
1777 stat(c
, FREE_SLOWPATH
);
1780 if (unlikely(SLABDEBUG
&& PageSlubDebug(page
)))
1784 prior
= object
[offset
] = page
->freelist
;
1785 page
->freelist
= object
;
1788 if (unlikely(PageSlubFrozen(page
))) {
1789 stat(c
, FREE_FROZEN
);
1793 if (unlikely(!page
->inuse
))
1797 * Objects left in the slab. If it was not on the partial list before
1800 if (unlikely(!prior
)) {
1801 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1802 stat(c
, FREE_ADD_PARTIAL
);
1812 * Slab still on the partial list.
1814 remove_partial(s
, page
);
1815 stat(c
, FREE_REMOVE_PARTIAL
);
1819 discard_slab(s
, page
);
1823 if (!free_debug_processing(s
, page
, x
, addr
))
1829 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1830 * can perform fastpath freeing without additional function calls.
1832 * The fastpath is only possible if we are freeing to the current cpu slab
1833 * of this processor. This typically the case if we have just allocated
1836 * If fastpath is not possible then fall back to __slab_free where we deal
1837 * with all sorts of special processing.
1839 static __always_inline
void slab_free(struct kmem_cache
*s
,
1840 struct page
*page
, void *x
, unsigned long addr
)
1842 void **object
= (void *)x
;
1843 struct kmem_cache_cpu
*c
;
1844 unsigned long flags
;
1846 kmemleak_free_recursive(x
, s
->flags
);
1847 local_irq_save(flags
);
1848 c
= get_cpu_slab(s
, smp_processor_id());
1849 kmemcheck_slab_free(s
, object
, c
->objsize
);
1850 debug_check_no_locks_freed(object
, c
->objsize
);
1851 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1852 debug_check_no_obj_freed(object
, c
->objsize
);
1853 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1854 object
[c
->offset
] = c
->freelist
;
1855 c
->freelist
= object
;
1856 stat(c
, FREE_FASTPATH
);
1858 __slab_free(s
, page
, x
, addr
, c
->offset
);
1860 local_irq_restore(flags
);
1863 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
1867 page
= virt_to_head_page(x
);
1869 slab_free(s
, page
, x
, _RET_IP_
);
1871 trace_kmem_cache_free(_RET_IP_
, x
);
1873 EXPORT_SYMBOL(kmem_cache_free
);
1875 /* Figure out on which slab page the object resides */
1876 static struct page
*get_object_page(const void *x
)
1878 struct page
*page
= virt_to_head_page(x
);
1880 if (!PageSlab(page
))
1887 * Object placement in a slab is made very easy because we always start at
1888 * offset 0. If we tune the size of the object to the alignment then we can
1889 * get the required alignment by putting one properly sized object after
1892 * Notice that the allocation order determines the sizes of the per cpu
1893 * caches. Each processor has always one slab available for allocations.
1894 * Increasing the allocation order reduces the number of times that slabs
1895 * must be moved on and off the partial lists and is therefore a factor in
1900 * Mininum / Maximum order of slab pages. This influences locking overhead
1901 * and slab fragmentation. A higher order reduces the number of partial slabs
1902 * and increases the number of allocations possible without having to
1903 * take the list_lock.
1905 static int slub_min_order
;
1906 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1907 static int slub_min_objects
;
1910 * Merge control. If this is set then no merging of slab caches will occur.
1911 * (Could be removed. This was introduced to pacify the merge skeptics.)
1913 static int slub_nomerge
;
1916 * Calculate the order of allocation given an slab object size.
1918 * The order of allocation has significant impact on performance and other
1919 * system components. Generally order 0 allocations should be preferred since
1920 * order 0 does not cause fragmentation in the page allocator. Larger objects
1921 * be problematic to put into order 0 slabs because there may be too much
1922 * unused space left. We go to a higher order if more than 1/16th of the slab
1925 * In order to reach satisfactory performance we must ensure that a minimum
1926 * number of objects is in one slab. Otherwise we may generate too much
1927 * activity on the partial lists which requires taking the list_lock. This is
1928 * less a concern for large slabs though which are rarely used.
1930 * slub_max_order specifies the order where we begin to stop considering the
1931 * number of objects in a slab as critical. If we reach slub_max_order then
1932 * we try to keep the page order as low as possible. So we accept more waste
1933 * of space in favor of a small page order.
1935 * Higher order allocations also allow the placement of more objects in a
1936 * slab and thereby reduce object handling overhead. If the user has
1937 * requested a higher mininum order then we start with that one instead of
1938 * the smallest order which will fit the object.
1940 static inline int slab_order(int size
, int min_objects
,
1941 int max_order
, int fract_leftover
)
1945 int min_order
= slub_min_order
;
1947 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1948 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1950 for (order
= max(min_order
,
1951 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1952 order
<= max_order
; order
++) {
1954 unsigned long slab_size
= PAGE_SIZE
<< order
;
1956 if (slab_size
< min_objects
* size
)
1959 rem
= slab_size
% size
;
1961 if (rem
<= slab_size
/ fract_leftover
)
1969 static inline int calculate_order(int size
)
1977 * Attempt to find best configuration for a slab. This
1978 * works by first attempting to generate a layout with
1979 * the best configuration and backing off gradually.
1981 * First we reduce the acceptable waste in a slab. Then
1982 * we reduce the minimum objects required in a slab.
1984 min_objects
= slub_min_objects
;
1986 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1987 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1988 min_objects
= min(min_objects
, max_objects
);
1990 while (min_objects
> 1) {
1992 while (fraction
>= 4) {
1993 order
= slab_order(size
, min_objects
,
1994 slub_max_order
, fraction
);
1995 if (order
<= slub_max_order
)
2003 * We were unable to place multiple objects in a slab. Now
2004 * lets see if we can place a single object there.
2006 order
= slab_order(size
, 1, slub_max_order
, 1);
2007 if (order
<= slub_max_order
)
2011 * Doh this slab cannot be placed using slub_max_order.
2013 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2014 if (order
< MAX_ORDER
)
2020 * Figure out what the alignment of the objects will be.
2022 static unsigned long calculate_alignment(unsigned long flags
,
2023 unsigned long align
, unsigned long size
)
2026 * If the user wants hardware cache aligned objects then follow that
2027 * suggestion if the object is sufficiently large.
2029 * The hardware cache alignment cannot override the specified
2030 * alignment though. If that is greater then use it.
2032 if (flags
& SLAB_HWCACHE_ALIGN
) {
2033 unsigned long ralign
= cache_line_size();
2034 while (size
<= ralign
/ 2)
2036 align
= max(align
, ralign
);
2039 if (align
< ARCH_SLAB_MINALIGN
)
2040 align
= ARCH_SLAB_MINALIGN
;
2042 return ALIGN(align
, sizeof(void *));
2045 static void init_kmem_cache_cpu(struct kmem_cache
*s
,
2046 struct kmem_cache_cpu
*c
)
2051 c
->offset
= s
->offset
/ sizeof(void *);
2052 c
->objsize
= s
->objsize
;
2053 #ifdef CONFIG_SLUB_STATS
2054 memset(c
->stat
, 0, NR_SLUB_STAT_ITEMS
* sizeof(unsigned));
2059 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2062 spin_lock_init(&n
->list_lock
);
2063 INIT_LIST_HEAD(&n
->partial
);
2064 #ifdef CONFIG_SLUB_DEBUG
2065 atomic_long_set(&n
->nr_slabs
, 0);
2066 atomic_long_set(&n
->total_objects
, 0);
2067 INIT_LIST_HEAD(&n
->full
);
2073 * Per cpu array for per cpu structures.
2075 * The per cpu array places all kmem_cache_cpu structures from one processor
2076 * close together meaning that it becomes possible that multiple per cpu
2077 * structures are contained in one cacheline. This may be particularly
2078 * beneficial for the kmalloc caches.
2080 * A desktop system typically has around 60-80 slabs. With 100 here we are
2081 * likely able to get per cpu structures for all caches from the array defined
2082 * here. We must be able to cover all kmalloc caches during bootstrap.
2084 * If the per cpu array is exhausted then fall back to kmalloc
2085 * of individual cachelines. No sharing is possible then.
2087 #define NR_KMEM_CACHE_CPU 100
2089 static DEFINE_PER_CPU(struct kmem_cache_cpu
[NR_KMEM_CACHE_CPU
],
2092 static DEFINE_PER_CPU(struct kmem_cache_cpu
*, kmem_cache_cpu_free
);
2093 static DECLARE_BITMAP(kmem_cach_cpu_free_init_once
, CONFIG_NR_CPUS
);
2095 static struct kmem_cache_cpu
*alloc_kmem_cache_cpu(struct kmem_cache
*s
,
2096 int cpu
, gfp_t flags
)
2098 struct kmem_cache_cpu
*c
= per_cpu(kmem_cache_cpu_free
, cpu
);
2101 per_cpu(kmem_cache_cpu_free
, cpu
) =
2102 (void *)c
->freelist
;
2104 /* Table overflow: So allocate ourselves */
2106 ALIGN(sizeof(struct kmem_cache_cpu
), cache_line_size()),
2107 flags
, cpu_to_node(cpu
));
2112 init_kmem_cache_cpu(s
, c
);
2116 static void free_kmem_cache_cpu(struct kmem_cache_cpu
*c
, int cpu
)
2118 if (c
< per_cpu(kmem_cache_cpu
, cpu
) ||
2119 c
>= per_cpu(kmem_cache_cpu
, cpu
) + NR_KMEM_CACHE_CPU
) {
2123 c
->freelist
= (void *)per_cpu(kmem_cache_cpu_free
, cpu
);
2124 per_cpu(kmem_cache_cpu_free
, cpu
) = c
;
2127 static void free_kmem_cache_cpus(struct kmem_cache
*s
)
2131 for_each_online_cpu(cpu
) {
2132 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2135 s
->cpu_slab
[cpu
] = NULL
;
2136 free_kmem_cache_cpu(c
, cpu
);
2141 static int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2145 for_each_online_cpu(cpu
) {
2146 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
2151 c
= alloc_kmem_cache_cpu(s
, cpu
, flags
);
2153 free_kmem_cache_cpus(s
);
2156 s
->cpu_slab
[cpu
] = c
;
2162 * Initialize the per cpu array.
2164 static void init_alloc_cpu_cpu(int cpu
)
2168 if (cpumask_test_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
)))
2171 for (i
= NR_KMEM_CACHE_CPU
- 1; i
>= 0; i
--)
2172 free_kmem_cache_cpu(&per_cpu(kmem_cache_cpu
, cpu
)[i
], cpu
);
2174 cpumask_set_cpu(cpu
, to_cpumask(kmem_cach_cpu_free_init_once
));
2177 static void __init
init_alloc_cpu(void)
2181 for_each_online_cpu(cpu
)
2182 init_alloc_cpu_cpu(cpu
);
2186 static inline void free_kmem_cache_cpus(struct kmem_cache
*s
) {}
2187 static inline void init_alloc_cpu(void) {}
2189 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2191 init_kmem_cache_cpu(s
, &s
->cpu_slab
);
2198 * No kmalloc_node yet so do it by hand. We know that this is the first
2199 * slab on the node for this slabcache. There are no concurrent accesses
2202 * Note that this function only works on the kmalloc_node_cache
2203 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2204 * memory on a fresh node that has no slab structures yet.
2206 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2209 struct kmem_cache_node
*n
;
2210 unsigned long flags
;
2212 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2214 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2217 if (page_to_nid(page
) != node
) {
2218 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2220 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2221 "in order to be able to continue\n");
2226 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2228 kmalloc_caches
->node
[node
] = n
;
2229 #ifdef CONFIG_SLUB_DEBUG
2230 init_object(kmalloc_caches
, n
, 1);
2231 init_tracking(kmalloc_caches
, n
);
2233 init_kmem_cache_node(n
, kmalloc_caches
);
2234 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2237 * lockdep requires consistent irq usage for each lock
2238 * so even though there cannot be a race this early in
2239 * the boot sequence, we still disable irqs.
2241 local_irq_save(flags
);
2242 add_partial(n
, page
, 0);
2243 local_irq_restore(flags
);
2246 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2250 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2251 struct kmem_cache_node
*n
= s
->node
[node
];
2252 if (n
&& n
!= &s
->local_node
)
2253 kmem_cache_free(kmalloc_caches
, n
);
2254 s
->node
[node
] = NULL
;
2258 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2263 if (slab_state
>= UP
)
2264 local_node
= page_to_nid(virt_to_page(s
));
2268 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2269 struct kmem_cache_node
*n
;
2271 if (local_node
== node
)
2274 if (slab_state
== DOWN
) {
2275 early_kmem_cache_node_alloc(gfpflags
, node
);
2278 n
= kmem_cache_alloc_node(kmalloc_caches
,
2282 free_kmem_cache_nodes(s
);
2288 init_kmem_cache_node(n
, s
);
2293 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2297 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2299 init_kmem_cache_node(&s
->local_node
, s
);
2304 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2306 if (min
< MIN_PARTIAL
)
2308 else if (min
> MAX_PARTIAL
)
2310 s
->min_partial
= min
;
2314 * calculate_sizes() determines the order and the distribution of data within
2317 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2319 unsigned long flags
= s
->flags
;
2320 unsigned long size
= s
->objsize
;
2321 unsigned long align
= s
->align
;
2325 * Round up object size to the next word boundary. We can only
2326 * place the free pointer at word boundaries and this determines
2327 * the possible location of the free pointer.
2329 size
= ALIGN(size
, sizeof(void *));
2331 #ifdef CONFIG_SLUB_DEBUG
2333 * Determine if we can poison the object itself. If the user of
2334 * the slab may touch the object after free or before allocation
2335 * then we should never poison the object itself.
2337 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2339 s
->flags
|= __OBJECT_POISON
;
2341 s
->flags
&= ~__OBJECT_POISON
;
2345 * If we are Redzoning then check if there is some space between the
2346 * end of the object and the free pointer. If not then add an
2347 * additional word to have some bytes to store Redzone information.
2349 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2350 size
+= sizeof(void *);
2354 * With that we have determined the number of bytes in actual use
2355 * by the object. This is the potential offset to the free pointer.
2359 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2362 * Relocate free pointer after the object if it is not
2363 * permitted to overwrite the first word of the object on
2366 * This is the case if we do RCU, have a constructor or
2367 * destructor or are poisoning the objects.
2370 size
+= sizeof(void *);
2373 #ifdef CONFIG_SLUB_DEBUG
2374 if (flags
& SLAB_STORE_USER
)
2376 * Need to store information about allocs and frees after
2379 size
+= 2 * sizeof(struct track
);
2381 if (flags
& SLAB_RED_ZONE
)
2383 * Add some empty padding so that we can catch
2384 * overwrites from earlier objects rather than let
2385 * tracking information or the free pointer be
2386 * corrupted if a user writes before the start
2389 size
+= sizeof(void *);
2393 * Determine the alignment based on various parameters that the
2394 * user specified and the dynamic determination of cache line size
2397 align
= calculate_alignment(flags
, align
, s
->objsize
);
2400 * SLUB stores one object immediately after another beginning from
2401 * offset 0. In order to align the objects we have to simply size
2402 * each object to conform to the alignment.
2404 size
= ALIGN(size
, align
);
2406 if (forced_order
>= 0)
2407 order
= forced_order
;
2409 order
= calculate_order(size
);
2416 s
->allocflags
|= __GFP_COMP
;
2418 if (s
->flags
& SLAB_CACHE_DMA
)
2419 s
->allocflags
|= SLUB_DMA
;
2421 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2422 s
->allocflags
|= __GFP_RECLAIMABLE
;
2425 * Determine the number of objects per slab
2427 s
->oo
= oo_make(order
, size
);
2428 s
->min
= oo_make(get_order(size
), size
);
2429 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2432 return !!oo_objects(s
->oo
);
2436 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2437 const char *name
, size_t size
,
2438 size_t align
, unsigned long flags
,
2439 void (*ctor
)(void *))
2441 memset(s
, 0, kmem_size
);
2446 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2448 if (!calculate_sizes(s
, -1))
2452 * The larger the object size is, the more pages we want on the partial
2453 * list to avoid pounding the page allocator excessively.
2455 set_min_partial(s
, ilog2(s
->size
));
2458 s
->remote_node_defrag_ratio
= 1000;
2460 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2463 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2465 free_kmem_cache_nodes(s
);
2467 if (flags
& SLAB_PANIC
)
2468 panic("Cannot create slab %s size=%lu realsize=%u "
2469 "order=%u offset=%u flags=%lx\n",
2470 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2476 * Check if a given pointer is valid
2478 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2482 page
= get_object_page(object
);
2484 if (!page
|| s
!= page
->slab
)
2485 /* No slab or wrong slab */
2488 if (!check_valid_pointer(s
, page
, object
))
2492 * We could also check if the object is on the slabs freelist.
2493 * But this would be too expensive and it seems that the main
2494 * purpose of kmem_ptr_valid() is to check if the object belongs
2495 * to a certain slab.
2499 EXPORT_SYMBOL(kmem_ptr_validate
);
2502 * Determine the size of a slab object
2504 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2508 EXPORT_SYMBOL(kmem_cache_size
);
2510 const char *kmem_cache_name(struct kmem_cache
*s
)
2514 EXPORT_SYMBOL(kmem_cache_name
);
2516 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2519 #ifdef CONFIG_SLUB_DEBUG
2520 void *addr
= page_address(page
);
2522 DECLARE_BITMAP(map
, page
->objects
);
2524 bitmap_zero(map
, page
->objects
);
2525 slab_err(s
, page
, "%s", text
);
2527 for_each_free_object(p
, s
, page
->freelist
)
2528 set_bit(slab_index(p
, s
, addr
), map
);
2530 for_each_object(p
, s
, addr
, page
->objects
) {
2532 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2533 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2535 print_tracking(s
, p
);
2543 * Attempt to free all partial slabs on a node.
2545 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2547 unsigned long flags
;
2548 struct page
*page
, *h
;
2550 spin_lock_irqsave(&n
->list_lock
, flags
);
2551 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2553 list_del(&page
->lru
);
2554 discard_slab(s
, page
);
2557 list_slab_objects(s
, page
,
2558 "Objects remaining on kmem_cache_close()");
2561 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2565 * Release all resources used by a slab cache.
2567 static inline int kmem_cache_close(struct kmem_cache
*s
)
2573 /* Attempt to free all objects */
2574 free_kmem_cache_cpus(s
);
2575 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2576 struct kmem_cache_node
*n
= get_node(s
, node
);
2579 if (n
->nr_partial
|| slabs_node(s
, node
))
2582 free_kmem_cache_nodes(s
);
2587 * Close a cache and release the kmem_cache structure
2588 * (must be used for caches created using kmem_cache_create)
2590 void kmem_cache_destroy(struct kmem_cache
*s
)
2592 down_write(&slub_lock
);
2596 up_write(&slub_lock
);
2597 if (kmem_cache_close(s
)) {
2598 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2599 "still has objects.\n", s
->name
, __func__
);
2602 sysfs_slab_remove(s
);
2604 up_write(&slub_lock
);
2606 EXPORT_SYMBOL(kmem_cache_destroy
);
2608 /********************************************************************
2610 *******************************************************************/
2612 struct kmem_cache kmalloc_caches
[SLUB_PAGE_SHIFT
] __cacheline_aligned
;
2613 EXPORT_SYMBOL(kmalloc_caches
);
2615 static int __init
setup_slub_min_order(char *str
)
2617 get_option(&str
, &slub_min_order
);
2622 __setup("slub_min_order=", setup_slub_min_order
);
2624 static int __init
setup_slub_max_order(char *str
)
2626 get_option(&str
, &slub_max_order
);
2627 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2632 __setup("slub_max_order=", setup_slub_max_order
);
2634 static int __init
setup_slub_min_objects(char *str
)
2636 get_option(&str
, &slub_min_objects
);
2641 __setup("slub_min_objects=", setup_slub_min_objects
);
2643 static int __init
setup_slub_nomerge(char *str
)
2649 __setup("slub_nomerge", setup_slub_nomerge
);
2651 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2652 const char *name
, int size
, gfp_t gfp_flags
)
2654 unsigned int flags
= 0;
2656 if (gfp_flags
& SLUB_DMA
)
2657 flags
= SLAB_CACHE_DMA
;
2660 * This function is called with IRQs disabled during early-boot on
2661 * single CPU so there's no need to take slub_lock here.
2663 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2667 list_add(&s
->list
, &slab_caches
);
2669 if (sysfs_slab_add(s
))
2674 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2677 #ifdef CONFIG_ZONE_DMA
2678 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2680 static void sysfs_add_func(struct work_struct
*w
)
2682 struct kmem_cache
*s
;
2684 down_write(&slub_lock
);
2685 list_for_each_entry(s
, &slab_caches
, list
) {
2686 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2687 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2691 up_write(&slub_lock
);
2694 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2696 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2698 struct kmem_cache
*s
;
2701 unsigned long slabflags
;
2703 s
= kmalloc_caches_dma
[index
];
2707 /* Dynamically create dma cache */
2708 if (flags
& __GFP_WAIT
)
2709 down_write(&slub_lock
);
2711 if (!down_write_trylock(&slub_lock
))
2715 if (kmalloc_caches_dma
[index
])
2718 realsize
= kmalloc_caches
[index
].objsize
;
2719 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2720 (unsigned int)realsize
);
2721 s
= kmalloc(kmem_size
, flags
& ~SLUB_DMA
);
2724 * Must defer sysfs creation to a workqueue because we don't know
2725 * what context we are called from. Before sysfs comes up, we don't
2726 * need to do anything because our sysfs initcall will start by
2727 * adding all existing slabs to sysfs.
2729 slabflags
= SLAB_CACHE_DMA
|SLAB_NOTRACK
;
2730 if (slab_state
>= SYSFS
)
2731 slabflags
|= __SYSFS_ADD_DEFERRED
;
2733 if (!s
|| !text
|| !kmem_cache_open(s
, flags
, text
,
2734 realsize
, ARCH_KMALLOC_MINALIGN
, slabflags
, NULL
)) {
2740 list_add(&s
->list
, &slab_caches
);
2741 kmalloc_caches_dma
[index
] = s
;
2743 if (slab_state
>= SYSFS
)
2744 schedule_work(&sysfs_add_work
);
2747 up_write(&slub_lock
);
2749 return kmalloc_caches_dma
[index
];
2754 * Conversion table for small slabs sizes / 8 to the index in the
2755 * kmalloc array. This is necessary for slabs < 192 since we have non power
2756 * of two cache sizes there. The size of larger slabs can be determined using
2759 static s8 size_index
[24] = {
2786 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2792 return ZERO_SIZE_PTR
;
2794 index
= size_index
[(size
- 1) / 8];
2796 index
= fls(size
- 1);
2798 #ifdef CONFIG_ZONE_DMA
2799 if (unlikely((flags
& SLUB_DMA
)))
2800 return dma_kmalloc_cache(index
, flags
);
2803 return &kmalloc_caches
[index
];
2806 void *__kmalloc(size_t size
, gfp_t flags
)
2808 struct kmem_cache
*s
;
2811 if (unlikely(size
> SLUB_MAX_SIZE
))
2812 return kmalloc_large(size
, flags
);
2814 s
= get_slab(size
, flags
);
2816 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2819 ret
= slab_alloc(s
, flags
, -1, _RET_IP_
);
2821 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2825 EXPORT_SYMBOL(__kmalloc
);
2827 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2831 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2832 page
= alloc_pages_node(node
, flags
, get_order(size
));
2834 return page_address(page
);
2840 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2842 struct kmem_cache
*s
;
2845 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2846 ret
= kmalloc_large_node(size
, flags
, node
);
2848 trace_kmalloc_node(_RET_IP_
, ret
,
2849 size
, PAGE_SIZE
<< get_order(size
),
2855 s
= get_slab(size
, flags
);
2857 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2860 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2862 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2866 EXPORT_SYMBOL(__kmalloc_node
);
2869 size_t ksize(const void *object
)
2872 struct kmem_cache
*s
;
2874 if (unlikely(object
== ZERO_SIZE_PTR
))
2877 page
= virt_to_head_page(object
);
2879 if (unlikely(!PageSlab(page
))) {
2880 WARN_ON(!PageCompound(page
));
2881 return PAGE_SIZE
<< compound_order(page
);
2885 #ifdef CONFIG_SLUB_DEBUG
2887 * Debugging requires use of the padding between object
2888 * and whatever may come after it.
2890 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2895 * If we have the need to store the freelist pointer
2896 * back there or track user information then we can
2897 * only use the space before that information.
2899 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2902 * Else we can use all the padding etc for the allocation
2906 EXPORT_SYMBOL(ksize
);
2908 void kfree(const void *x
)
2911 void *object
= (void *)x
;
2913 trace_kfree(_RET_IP_
, x
);
2915 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2918 page
= virt_to_head_page(x
);
2919 if (unlikely(!PageSlab(page
))) {
2920 BUG_ON(!PageCompound(page
));
2924 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2926 EXPORT_SYMBOL(kfree
);
2929 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2930 * the remaining slabs by the number of items in use. The slabs with the
2931 * most items in use come first. New allocations will then fill those up
2932 * and thus they can be removed from the partial lists.
2934 * The slabs with the least items are placed last. This results in them
2935 * being allocated from last increasing the chance that the last objects
2936 * are freed in them.
2938 int kmem_cache_shrink(struct kmem_cache
*s
)
2942 struct kmem_cache_node
*n
;
2945 int objects
= oo_objects(s
->max
);
2946 struct list_head
*slabs_by_inuse
=
2947 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2948 unsigned long flags
;
2950 if (!slabs_by_inuse
)
2954 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2955 n
= get_node(s
, node
);
2960 for (i
= 0; i
< objects
; i
++)
2961 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2963 spin_lock_irqsave(&n
->list_lock
, flags
);
2966 * Build lists indexed by the items in use in each slab.
2968 * Note that concurrent frees may occur while we hold the
2969 * list_lock. page->inuse here is the upper limit.
2971 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2972 if (!page
->inuse
&& slab_trylock(page
)) {
2974 * Must hold slab lock here because slab_free
2975 * may have freed the last object and be
2976 * waiting to release the slab.
2978 list_del(&page
->lru
);
2981 discard_slab(s
, page
);
2983 list_move(&page
->lru
,
2984 slabs_by_inuse
+ page
->inuse
);
2989 * Rebuild the partial list with the slabs filled up most
2990 * first and the least used slabs at the end.
2992 for (i
= objects
- 1; i
>= 0; i
--)
2993 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2995 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2998 kfree(slabs_by_inuse
);
3001 EXPORT_SYMBOL(kmem_cache_shrink
);
3003 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
3004 static int slab_mem_going_offline_callback(void *arg
)
3006 struct kmem_cache
*s
;
3008 down_read(&slub_lock
);
3009 list_for_each_entry(s
, &slab_caches
, list
)
3010 kmem_cache_shrink(s
);
3011 up_read(&slub_lock
);
3016 static void slab_mem_offline_callback(void *arg
)
3018 struct kmem_cache_node
*n
;
3019 struct kmem_cache
*s
;
3020 struct memory_notify
*marg
= arg
;
3023 offline_node
= marg
->status_change_nid
;
3026 * If the node still has available memory. we need kmem_cache_node
3029 if (offline_node
< 0)
3032 down_read(&slub_lock
);
3033 list_for_each_entry(s
, &slab_caches
, list
) {
3034 n
= get_node(s
, offline_node
);
3037 * if n->nr_slabs > 0, slabs still exist on the node
3038 * that is going down. We were unable to free them,
3039 * and offline_pages() function shoudn't call this
3040 * callback. So, we must fail.
3042 BUG_ON(slabs_node(s
, offline_node
));
3044 s
->node
[offline_node
] = NULL
;
3045 kmem_cache_free(kmalloc_caches
, n
);
3048 up_read(&slub_lock
);
3051 static int slab_mem_going_online_callback(void *arg
)
3053 struct kmem_cache_node
*n
;
3054 struct kmem_cache
*s
;
3055 struct memory_notify
*marg
= arg
;
3056 int nid
= marg
->status_change_nid
;
3060 * If the node's memory is already available, then kmem_cache_node is
3061 * already created. Nothing to do.
3067 * We are bringing a node online. No memory is available yet. We must
3068 * allocate a kmem_cache_node structure in order to bring the node
3071 down_read(&slub_lock
);
3072 list_for_each_entry(s
, &slab_caches
, list
) {
3074 * XXX: kmem_cache_alloc_node will fallback to other nodes
3075 * since memory is not yet available from the node that
3078 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
3083 init_kmem_cache_node(n
, s
);
3087 up_read(&slub_lock
);
3091 static int slab_memory_callback(struct notifier_block
*self
,
3092 unsigned long action
, void *arg
)
3097 case MEM_GOING_ONLINE
:
3098 ret
= slab_mem_going_online_callback(arg
);
3100 case MEM_GOING_OFFLINE
:
3101 ret
= slab_mem_going_offline_callback(arg
);
3104 case MEM_CANCEL_ONLINE
:
3105 slab_mem_offline_callback(arg
);
3108 case MEM_CANCEL_OFFLINE
:
3112 ret
= notifier_from_errno(ret
);
3118 #endif /* CONFIG_MEMORY_HOTPLUG */
3120 /********************************************************************
3121 * Basic setup of slabs
3122 *******************************************************************/
3124 void __init
kmem_cache_init(void)
3133 * Must first have the slab cache available for the allocations of the
3134 * struct kmem_cache_node's. There is special bootstrap code in
3135 * kmem_cache_open for slab_state == DOWN.
3137 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3138 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3139 kmalloc_caches
[0].refcount
= -1;
3142 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3145 /* Able to allocate the per node structures */
3146 slab_state
= PARTIAL
;
3148 /* Caches that are not of the two-to-the-power-of size */
3149 if (KMALLOC_MIN_SIZE
<= 64) {
3150 create_kmalloc_cache(&kmalloc_caches
[1],
3151 "kmalloc-96", 96, GFP_NOWAIT
);
3153 create_kmalloc_cache(&kmalloc_caches
[2],
3154 "kmalloc-192", 192, GFP_NOWAIT
);
3158 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3159 create_kmalloc_cache(&kmalloc_caches
[i
],
3160 "kmalloc", 1 << i
, GFP_NOWAIT
);
3166 * Patch up the size_index table if we have strange large alignment
3167 * requirements for the kmalloc array. This is only the case for
3168 * MIPS it seems. The standard arches will not generate any code here.
3170 * Largest permitted alignment is 256 bytes due to the way we
3171 * handle the index determination for the smaller caches.
3173 * Make sure that nothing crazy happens if someone starts tinkering
3174 * around with ARCH_KMALLOC_MINALIGN
3176 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3177 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3179 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8)
3180 size_index
[(i
- 1) / 8] = KMALLOC_SHIFT_LOW
;
3182 if (KMALLOC_MIN_SIZE
== 128) {
3184 * The 192 byte sized cache is not used if the alignment
3185 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3188 for (i
= 128 + 8; i
<= 192; i
+= 8)
3189 size_index
[(i
- 1) / 8] = 8;
3194 /* Provide the correct kmalloc names now that the caches are up */
3195 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++)
3196 kmalloc_caches
[i
]. name
=
3197 kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3200 register_cpu_notifier(&slab_notifier
);
3201 kmem_size
= offsetof(struct kmem_cache
, cpu_slab
) +
3202 nr_cpu_ids
* sizeof(struct kmem_cache_cpu
*);
3204 kmem_size
= sizeof(struct kmem_cache
);
3208 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3209 " CPUs=%d, Nodes=%d\n",
3210 caches
, cache_line_size(),
3211 slub_min_order
, slub_max_order
, slub_min_objects
,
3212 nr_cpu_ids
, nr_node_ids
);
3215 void __init
kmem_cache_init_late(void)
3220 * Find a mergeable slab cache
3222 static int slab_unmergeable(struct kmem_cache
*s
)
3224 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3231 * We may have set a slab to be unmergeable during bootstrap.
3233 if (s
->refcount
< 0)
3239 static struct kmem_cache
*find_mergeable(size_t size
,
3240 size_t align
, unsigned long flags
, const char *name
,
3241 void (*ctor
)(void *))
3243 struct kmem_cache
*s
;
3245 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3251 size
= ALIGN(size
, sizeof(void *));
3252 align
= calculate_alignment(flags
, align
, size
);
3253 size
= ALIGN(size
, align
);
3254 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3256 list_for_each_entry(s
, &slab_caches
, list
) {
3257 if (slab_unmergeable(s
))
3263 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3266 * Check if alignment is compatible.
3267 * Courtesy of Adrian Drzewiecki
3269 if ((s
->size
& ~(align
- 1)) != s
->size
)
3272 if (s
->size
- size
>= sizeof(void *))
3280 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3281 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3283 struct kmem_cache
*s
;
3285 down_write(&slub_lock
);
3286 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3292 * Adjust the object sizes so that we clear
3293 * the complete object on kzalloc.
3295 s
->objsize
= max(s
->objsize
, (int)size
);
3298 * And then we need to update the object size in the
3299 * per cpu structures
3301 for_each_online_cpu(cpu
)
3302 get_cpu_slab(s
, cpu
)->objsize
= s
->objsize
;
3304 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3305 up_write(&slub_lock
);
3307 if (sysfs_slab_alias(s
, name
)) {
3308 down_write(&slub_lock
);
3310 up_write(&slub_lock
);
3316 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3318 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3319 size
, align
, flags
, ctor
)) {
3320 list_add(&s
->list
, &slab_caches
);
3321 up_write(&slub_lock
);
3322 if (sysfs_slab_add(s
)) {
3323 down_write(&slub_lock
);
3325 up_write(&slub_lock
);
3333 up_write(&slub_lock
);
3336 if (flags
& SLAB_PANIC
)
3337 panic("Cannot create slabcache %s\n", name
);
3342 EXPORT_SYMBOL(kmem_cache_create
);
3346 * Use the cpu notifier to insure that the cpu slabs are flushed when
3349 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3350 unsigned long action
, void *hcpu
)
3352 long cpu
= (long)hcpu
;
3353 struct kmem_cache
*s
;
3354 unsigned long flags
;
3357 case CPU_UP_PREPARE
:
3358 case CPU_UP_PREPARE_FROZEN
:
3359 init_alloc_cpu_cpu(cpu
);
3360 down_read(&slub_lock
);
3361 list_for_each_entry(s
, &slab_caches
, list
)
3362 s
->cpu_slab
[cpu
] = alloc_kmem_cache_cpu(s
, cpu
,
3364 up_read(&slub_lock
);
3367 case CPU_UP_CANCELED
:
3368 case CPU_UP_CANCELED_FROZEN
:
3370 case CPU_DEAD_FROZEN
:
3371 down_read(&slub_lock
);
3372 list_for_each_entry(s
, &slab_caches
, list
) {
3373 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3375 local_irq_save(flags
);
3376 __flush_cpu_slab(s
, cpu
);
3377 local_irq_restore(flags
);
3378 free_kmem_cache_cpu(c
, cpu
);
3379 s
->cpu_slab
[cpu
] = NULL
;
3381 up_read(&slub_lock
);
3389 static struct notifier_block __cpuinitdata slab_notifier
= {
3390 .notifier_call
= slab_cpuup_callback
3395 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3397 struct kmem_cache
*s
;
3400 if (unlikely(size
> SLUB_MAX_SIZE
))
3401 return kmalloc_large(size
, gfpflags
);
3403 s
= get_slab(size
, gfpflags
);
3405 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3408 ret
= slab_alloc(s
, gfpflags
, -1, caller
);
3410 /* Honor the call site pointer we recieved. */
3411 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3416 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3417 int node
, unsigned long caller
)
3419 struct kmem_cache
*s
;
3422 if (unlikely(size
> SLUB_MAX_SIZE
))
3423 return kmalloc_large_node(size
, gfpflags
, node
);
3425 s
= get_slab(size
, gfpflags
);
3427 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3430 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3432 /* Honor the call site pointer we recieved. */
3433 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3438 #ifdef CONFIG_SLUB_DEBUG
3439 static int count_inuse(struct page
*page
)
3444 static int count_total(struct page
*page
)
3446 return page
->objects
;
3449 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3453 void *addr
= page_address(page
);
3455 if (!check_slab(s
, page
) ||
3456 !on_freelist(s
, page
, NULL
))
3459 /* Now we know that a valid freelist exists */
3460 bitmap_zero(map
, page
->objects
);
3462 for_each_free_object(p
, s
, page
->freelist
) {
3463 set_bit(slab_index(p
, s
, addr
), map
);
3464 if (!check_object(s
, page
, p
, 0))
3468 for_each_object(p
, s
, addr
, page
->objects
)
3469 if (!test_bit(slab_index(p
, s
, addr
), map
))
3470 if (!check_object(s
, page
, p
, 1))
3475 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3478 if (slab_trylock(page
)) {
3479 validate_slab(s
, page
, map
);
3482 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3485 if (s
->flags
& DEBUG_DEFAULT_FLAGS
) {
3486 if (!PageSlubDebug(page
))
3487 printk(KERN_ERR
"SLUB %s: SlubDebug not set "
3488 "on slab 0x%p\n", s
->name
, page
);
3490 if (PageSlubDebug(page
))
3491 printk(KERN_ERR
"SLUB %s: SlubDebug set on "
3492 "slab 0x%p\n", s
->name
, page
);
3496 static int validate_slab_node(struct kmem_cache
*s
,
3497 struct kmem_cache_node
*n
, unsigned long *map
)
3499 unsigned long count
= 0;
3501 unsigned long flags
;
3503 spin_lock_irqsave(&n
->list_lock
, flags
);
3505 list_for_each_entry(page
, &n
->partial
, lru
) {
3506 validate_slab_slab(s
, page
, map
);
3509 if (count
!= n
->nr_partial
)
3510 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3511 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3513 if (!(s
->flags
& SLAB_STORE_USER
))
3516 list_for_each_entry(page
, &n
->full
, lru
) {
3517 validate_slab_slab(s
, page
, map
);
3520 if (count
!= atomic_long_read(&n
->nr_slabs
))
3521 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3522 "counter=%ld\n", s
->name
, count
,
3523 atomic_long_read(&n
->nr_slabs
));
3526 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3530 static long validate_slab_cache(struct kmem_cache
*s
)
3533 unsigned long count
= 0;
3534 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3535 sizeof(unsigned long), GFP_KERNEL
);
3541 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3542 struct kmem_cache_node
*n
= get_node(s
, node
);
3544 count
+= validate_slab_node(s
, n
, map
);
3550 #ifdef SLUB_RESILIENCY_TEST
3551 static void resiliency_test(void)
3555 printk(KERN_ERR
"SLUB resiliency testing\n");
3556 printk(KERN_ERR
"-----------------------\n");
3557 printk(KERN_ERR
"A. Corruption after allocation\n");
3559 p
= kzalloc(16, GFP_KERNEL
);
3561 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3562 " 0x12->0x%p\n\n", p
+ 16);
3564 validate_slab_cache(kmalloc_caches
+ 4);
3566 /* Hmmm... The next two are dangerous */
3567 p
= kzalloc(32, GFP_KERNEL
);
3568 p
[32 + sizeof(void *)] = 0x34;
3569 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3570 " 0x34 -> -0x%p\n", p
);
3572 "If allocated object is overwritten then not detectable\n\n");
3574 validate_slab_cache(kmalloc_caches
+ 5);
3575 p
= kzalloc(64, GFP_KERNEL
);
3576 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3578 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3581 "If allocated object is overwritten then not detectable\n\n");
3582 validate_slab_cache(kmalloc_caches
+ 6);
3584 printk(KERN_ERR
"\nB. Corruption after free\n");
3585 p
= kzalloc(128, GFP_KERNEL
);
3588 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3589 validate_slab_cache(kmalloc_caches
+ 7);
3591 p
= kzalloc(256, GFP_KERNEL
);
3594 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3596 validate_slab_cache(kmalloc_caches
+ 8);
3598 p
= kzalloc(512, GFP_KERNEL
);
3601 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3602 validate_slab_cache(kmalloc_caches
+ 9);
3605 static void resiliency_test(void) {};
3609 * Generate lists of code addresses where slabcache objects are allocated
3614 unsigned long count
;
3621 DECLARE_BITMAP(cpus
, NR_CPUS
);
3627 unsigned long count
;
3628 struct location
*loc
;
3631 static void free_loc_track(struct loc_track
*t
)
3634 free_pages((unsigned long)t
->loc
,
3635 get_order(sizeof(struct location
) * t
->max
));
3638 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3643 order
= get_order(sizeof(struct location
) * max
);
3645 l
= (void *)__get_free_pages(flags
, order
);
3650 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3658 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3659 const struct track
*track
)
3661 long start
, end
, pos
;
3663 unsigned long caddr
;
3664 unsigned long age
= jiffies
- track
->when
;
3670 pos
= start
+ (end
- start
+ 1) / 2;
3673 * There is nothing at "end". If we end up there
3674 * we need to add something to before end.
3679 caddr
= t
->loc
[pos
].addr
;
3680 if (track
->addr
== caddr
) {
3686 if (age
< l
->min_time
)
3688 if (age
> l
->max_time
)
3691 if (track
->pid
< l
->min_pid
)
3692 l
->min_pid
= track
->pid
;
3693 if (track
->pid
> l
->max_pid
)
3694 l
->max_pid
= track
->pid
;
3696 cpumask_set_cpu(track
->cpu
,
3697 to_cpumask(l
->cpus
));
3699 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3703 if (track
->addr
< caddr
)
3710 * Not found. Insert new tracking element.
3712 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3718 (t
->count
- pos
) * sizeof(struct location
));
3721 l
->addr
= track
->addr
;
3725 l
->min_pid
= track
->pid
;
3726 l
->max_pid
= track
->pid
;
3727 cpumask_clear(to_cpumask(l
->cpus
));
3728 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3729 nodes_clear(l
->nodes
);
3730 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3734 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3735 struct page
*page
, enum track_item alloc
)
3737 void *addr
= page_address(page
);
3738 DECLARE_BITMAP(map
, page
->objects
);
3741 bitmap_zero(map
, page
->objects
);
3742 for_each_free_object(p
, s
, page
->freelist
)
3743 set_bit(slab_index(p
, s
, addr
), map
);
3745 for_each_object(p
, s
, addr
, page
->objects
)
3746 if (!test_bit(slab_index(p
, s
, addr
), map
))
3747 add_location(t
, s
, get_track(s
, p
, alloc
));
3750 static int list_locations(struct kmem_cache
*s
, char *buf
,
3751 enum track_item alloc
)
3755 struct loc_track t
= { 0, 0, NULL
};
3758 if (!alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3760 return sprintf(buf
, "Out of memory\n");
3762 /* Push back cpu slabs */
3765 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3766 struct kmem_cache_node
*n
= get_node(s
, node
);
3767 unsigned long flags
;
3770 if (!atomic_long_read(&n
->nr_slabs
))
3773 spin_lock_irqsave(&n
->list_lock
, flags
);
3774 list_for_each_entry(page
, &n
->partial
, lru
)
3775 process_slab(&t
, s
, page
, alloc
);
3776 list_for_each_entry(page
, &n
->full
, lru
)
3777 process_slab(&t
, s
, page
, alloc
);
3778 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3781 for (i
= 0; i
< t
.count
; i
++) {
3782 struct location
*l
= &t
.loc
[i
];
3784 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3786 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3789 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3791 len
+= sprintf(buf
+ len
, "<not-available>");
3793 if (l
->sum_time
!= l
->min_time
) {
3794 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3796 (long)div_u64(l
->sum_time
, l
->count
),
3799 len
+= sprintf(buf
+ len
, " age=%ld",
3802 if (l
->min_pid
!= l
->max_pid
)
3803 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3804 l
->min_pid
, l
->max_pid
);
3806 len
+= sprintf(buf
+ len
, " pid=%ld",
3809 if (num_online_cpus() > 1 &&
3810 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3811 len
< PAGE_SIZE
- 60) {
3812 len
+= sprintf(buf
+ len
, " cpus=");
3813 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3814 to_cpumask(l
->cpus
));
3817 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3818 len
< PAGE_SIZE
- 60) {
3819 len
+= sprintf(buf
+ len
, " nodes=");
3820 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3824 len
+= sprintf(buf
+ len
, "\n");
3829 len
+= sprintf(buf
, "No data\n");
3833 enum slab_stat_type
{
3834 SL_ALL
, /* All slabs */
3835 SL_PARTIAL
, /* Only partially allocated slabs */
3836 SL_CPU
, /* Only slabs used for cpu caches */
3837 SL_OBJECTS
, /* Determine allocated objects not slabs */
3838 SL_TOTAL
/* Determine object capacity not slabs */
3841 #define SO_ALL (1 << SL_ALL)
3842 #define SO_PARTIAL (1 << SL_PARTIAL)
3843 #define SO_CPU (1 << SL_CPU)
3844 #define SO_OBJECTS (1 << SL_OBJECTS)
3845 #define SO_TOTAL (1 << SL_TOTAL)
3847 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3848 char *buf
, unsigned long flags
)
3850 unsigned long total
= 0;
3853 unsigned long *nodes
;
3854 unsigned long *per_cpu
;
3856 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3859 per_cpu
= nodes
+ nr_node_ids
;
3861 if (flags
& SO_CPU
) {
3864 for_each_possible_cpu(cpu
) {
3865 struct kmem_cache_cpu
*c
= get_cpu_slab(s
, cpu
);
3867 if (!c
|| c
->node
< 0)
3871 if (flags
& SO_TOTAL
)
3872 x
= c
->page
->objects
;
3873 else if (flags
& SO_OBJECTS
)
3879 nodes
[c
->node
] += x
;
3885 if (flags
& SO_ALL
) {
3886 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3887 struct kmem_cache_node
*n
= get_node(s
, node
);
3889 if (flags
& SO_TOTAL
)
3890 x
= atomic_long_read(&n
->total_objects
);
3891 else if (flags
& SO_OBJECTS
)
3892 x
= atomic_long_read(&n
->total_objects
) -
3893 count_partial(n
, count_free
);
3896 x
= atomic_long_read(&n
->nr_slabs
);
3901 } else if (flags
& SO_PARTIAL
) {
3902 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3903 struct kmem_cache_node
*n
= get_node(s
, node
);
3905 if (flags
& SO_TOTAL
)
3906 x
= count_partial(n
, count_total
);
3907 else if (flags
& SO_OBJECTS
)
3908 x
= count_partial(n
, count_inuse
);
3915 x
= sprintf(buf
, "%lu", total
);
3917 for_each_node_state(node
, N_NORMAL_MEMORY
)
3919 x
+= sprintf(buf
+ x
, " N%d=%lu",
3923 return x
+ sprintf(buf
+ x
, "\n");
3926 static int any_slab_objects(struct kmem_cache
*s
)
3930 for_each_online_node(node
) {
3931 struct kmem_cache_node
*n
= get_node(s
, node
);
3936 if (atomic_long_read(&n
->total_objects
))
3942 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3943 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3945 struct slab_attribute
{
3946 struct attribute attr
;
3947 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3948 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3951 #define SLAB_ATTR_RO(_name) \
3952 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3954 #define SLAB_ATTR(_name) \
3955 static struct slab_attribute _name##_attr = \
3956 __ATTR(_name, 0644, _name##_show, _name##_store)
3958 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3960 return sprintf(buf
, "%d\n", s
->size
);
3962 SLAB_ATTR_RO(slab_size
);
3964 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3966 return sprintf(buf
, "%d\n", s
->align
);
3968 SLAB_ATTR_RO(align
);
3970 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3972 return sprintf(buf
, "%d\n", s
->objsize
);
3974 SLAB_ATTR_RO(object_size
);
3976 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3978 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3980 SLAB_ATTR_RO(objs_per_slab
);
3982 static ssize_t
order_store(struct kmem_cache
*s
,
3983 const char *buf
, size_t length
)
3985 unsigned long order
;
3988 err
= strict_strtoul(buf
, 10, &order
);
3992 if (order
> slub_max_order
|| order
< slub_min_order
)
3995 calculate_sizes(s
, order
);
3999 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4001 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4005 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4007 return sprintf(buf
, "%lu\n", s
->min_partial
);
4010 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4016 err
= strict_strtoul(buf
, 10, &min
);
4020 set_min_partial(s
, min
);
4023 SLAB_ATTR(min_partial
);
4025 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4028 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
4030 return n
+ sprintf(buf
+ n
, "\n");
4036 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4038 return sprintf(buf
, "%d\n", s
->refcount
- 1);
4040 SLAB_ATTR_RO(aliases
);
4042 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4044 return show_slab_objects(s
, buf
, SO_ALL
);
4046 SLAB_ATTR_RO(slabs
);
4048 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4050 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4052 SLAB_ATTR_RO(partial
);
4054 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4056 return show_slab_objects(s
, buf
, SO_CPU
);
4058 SLAB_ATTR_RO(cpu_slabs
);
4060 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4062 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4064 SLAB_ATTR_RO(objects
);
4066 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4068 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4070 SLAB_ATTR_RO(objects_partial
);
4072 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4074 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4076 SLAB_ATTR_RO(total_objects
);
4078 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4080 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4083 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4084 const char *buf
, size_t length
)
4086 s
->flags
&= ~SLAB_DEBUG_FREE
;
4088 s
->flags
|= SLAB_DEBUG_FREE
;
4091 SLAB_ATTR(sanity_checks
);
4093 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4095 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4098 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4101 s
->flags
&= ~SLAB_TRACE
;
4103 s
->flags
|= SLAB_TRACE
;
4108 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4110 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4113 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4114 const char *buf
, size_t length
)
4116 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4118 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4121 SLAB_ATTR(reclaim_account
);
4123 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4125 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4127 SLAB_ATTR_RO(hwcache_align
);
4129 #ifdef CONFIG_ZONE_DMA
4130 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4132 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4134 SLAB_ATTR_RO(cache_dma
);
4137 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4139 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4141 SLAB_ATTR_RO(destroy_by_rcu
);
4143 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4145 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4148 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4149 const char *buf
, size_t length
)
4151 if (any_slab_objects(s
))
4154 s
->flags
&= ~SLAB_RED_ZONE
;
4156 s
->flags
|= SLAB_RED_ZONE
;
4157 calculate_sizes(s
, -1);
4160 SLAB_ATTR(red_zone
);
4162 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4164 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4167 static ssize_t
poison_store(struct kmem_cache
*s
,
4168 const char *buf
, size_t length
)
4170 if (any_slab_objects(s
))
4173 s
->flags
&= ~SLAB_POISON
;
4175 s
->flags
|= SLAB_POISON
;
4176 calculate_sizes(s
, -1);
4181 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4183 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4186 static ssize_t
store_user_store(struct kmem_cache
*s
,
4187 const char *buf
, size_t length
)
4189 if (any_slab_objects(s
))
4192 s
->flags
&= ~SLAB_STORE_USER
;
4194 s
->flags
|= SLAB_STORE_USER
;
4195 calculate_sizes(s
, -1);
4198 SLAB_ATTR(store_user
);
4200 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4205 static ssize_t
validate_store(struct kmem_cache
*s
,
4206 const char *buf
, size_t length
)
4210 if (buf
[0] == '1') {
4211 ret
= validate_slab_cache(s
);
4217 SLAB_ATTR(validate
);
4219 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4224 static ssize_t
shrink_store(struct kmem_cache
*s
,
4225 const char *buf
, size_t length
)
4227 if (buf
[0] == '1') {
4228 int rc
= kmem_cache_shrink(s
);
4238 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4240 if (!(s
->flags
& SLAB_STORE_USER
))
4242 return list_locations(s
, buf
, TRACK_ALLOC
);
4244 SLAB_ATTR_RO(alloc_calls
);
4246 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4248 if (!(s
->flags
& SLAB_STORE_USER
))
4250 return list_locations(s
, buf
, TRACK_FREE
);
4252 SLAB_ATTR_RO(free_calls
);
4255 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4257 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4260 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4261 const char *buf
, size_t length
)
4263 unsigned long ratio
;
4266 err
= strict_strtoul(buf
, 10, &ratio
);
4271 s
->remote_node_defrag_ratio
= ratio
* 10;
4275 SLAB_ATTR(remote_node_defrag_ratio
);
4278 #ifdef CONFIG_SLUB_STATS
4279 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4281 unsigned long sum
= 0;
4284 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4289 for_each_online_cpu(cpu
) {
4290 unsigned x
= get_cpu_slab(s
, cpu
)->stat
[si
];
4296 len
= sprintf(buf
, "%lu", sum
);
4299 for_each_online_cpu(cpu
) {
4300 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4301 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4305 return len
+ sprintf(buf
+ len
, "\n");
4308 #define STAT_ATTR(si, text) \
4309 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4311 return show_stat(s, buf, si); \
4313 SLAB_ATTR_RO(text); \
4315 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4316 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4317 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4318 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4319 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4320 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4321 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4322 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4323 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4324 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4325 STAT_ATTR(FREE_SLAB
, free_slab
);
4326 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4327 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4328 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4329 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4330 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4331 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4332 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4335 static struct attribute
*slab_attrs
[] = {
4336 &slab_size_attr
.attr
,
4337 &object_size_attr
.attr
,
4338 &objs_per_slab_attr
.attr
,
4340 &min_partial_attr
.attr
,
4342 &objects_partial_attr
.attr
,
4343 &total_objects_attr
.attr
,
4346 &cpu_slabs_attr
.attr
,
4350 &sanity_checks_attr
.attr
,
4352 &hwcache_align_attr
.attr
,
4353 &reclaim_account_attr
.attr
,
4354 &destroy_by_rcu_attr
.attr
,
4355 &red_zone_attr
.attr
,
4357 &store_user_attr
.attr
,
4358 &validate_attr
.attr
,
4360 &alloc_calls_attr
.attr
,
4361 &free_calls_attr
.attr
,
4362 #ifdef CONFIG_ZONE_DMA
4363 &cache_dma_attr
.attr
,
4366 &remote_node_defrag_ratio_attr
.attr
,
4368 #ifdef CONFIG_SLUB_STATS
4369 &alloc_fastpath_attr
.attr
,
4370 &alloc_slowpath_attr
.attr
,
4371 &free_fastpath_attr
.attr
,
4372 &free_slowpath_attr
.attr
,
4373 &free_frozen_attr
.attr
,
4374 &free_add_partial_attr
.attr
,
4375 &free_remove_partial_attr
.attr
,
4376 &alloc_from_partial_attr
.attr
,
4377 &alloc_slab_attr
.attr
,
4378 &alloc_refill_attr
.attr
,
4379 &free_slab_attr
.attr
,
4380 &cpuslab_flush_attr
.attr
,
4381 &deactivate_full_attr
.attr
,
4382 &deactivate_empty_attr
.attr
,
4383 &deactivate_to_head_attr
.attr
,
4384 &deactivate_to_tail_attr
.attr
,
4385 &deactivate_remote_frees_attr
.attr
,
4386 &order_fallback_attr
.attr
,
4391 static struct attribute_group slab_attr_group
= {
4392 .attrs
= slab_attrs
,
4395 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4396 struct attribute
*attr
,
4399 struct slab_attribute
*attribute
;
4400 struct kmem_cache
*s
;
4403 attribute
= to_slab_attr(attr
);
4406 if (!attribute
->show
)
4409 err
= attribute
->show(s
, buf
);
4414 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4415 struct attribute
*attr
,
4416 const char *buf
, size_t len
)
4418 struct slab_attribute
*attribute
;
4419 struct kmem_cache
*s
;
4422 attribute
= to_slab_attr(attr
);
4425 if (!attribute
->store
)
4428 err
= attribute
->store(s
, buf
, len
);
4433 static void kmem_cache_release(struct kobject
*kobj
)
4435 struct kmem_cache
*s
= to_slab(kobj
);
4440 static struct sysfs_ops slab_sysfs_ops
= {
4441 .show
= slab_attr_show
,
4442 .store
= slab_attr_store
,
4445 static struct kobj_type slab_ktype
= {
4446 .sysfs_ops
= &slab_sysfs_ops
,
4447 .release
= kmem_cache_release
4450 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4452 struct kobj_type
*ktype
= get_ktype(kobj
);
4454 if (ktype
== &slab_ktype
)
4459 static struct kset_uevent_ops slab_uevent_ops
= {
4460 .filter
= uevent_filter
,
4463 static struct kset
*slab_kset
;
4465 #define ID_STR_LENGTH 64
4467 /* Create a unique string id for a slab cache:
4469 * Format :[flags-]size
4471 static char *create_unique_id(struct kmem_cache
*s
)
4473 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4480 * First flags affecting slabcache operations. We will only
4481 * get here for aliasable slabs so we do not need to support
4482 * too many flags. The flags here must cover all flags that
4483 * are matched during merging to guarantee that the id is
4486 if (s
->flags
& SLAB_CACHE_DMA
)
4488 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4490 if (s
->flags
& SLAB_DEBUG_FREE
)
4492 if (!(s
->flags
& SLAB_NOTRACK
))
4496 p
+= sprintf(p
, "%07d", s
->size
);
4497 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4501 static int sysfs_slab_add(struct kmem_cache
*s
)
4507 if (slab_state
< SYSFS
)
4508 /* Defer until later */
4511 unmergeable
= slab_unmergeable(s
);
4514 * Slabcache can never be merged so we can use the name proper.
4515 * This is typically the case for debug situations. In that
4516 * case we can catch duplicate names easily.
4518 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4522 * Create a unique name for the slab as a target
4525 name
= create_unique_id(s
);
4528 s
->kobj
.kset
= slab_kset
;
4529 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4531 kobject_put(&s
->kobj
);
4535 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4538 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4540 /* Setup first alias */
4541 sysfs_slab_alias(s
, s
->name
);
4547 static void sysfs_slab_remove(struct kmem_cache
*s
)
4549 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4550 kobject_del(&s
->kobj
);
4551 kobject_put(&s
->kobj
);
4555 * Need to buffer aliases during bootup until sysfs becomes
4556 * available lest we lose that information.
4558 struct saved_alias
{
4559 struct kmem_cache
*s
;
4561 struct saved_alias
*next
;
4564 static struct saved_alias
*alias_list
;
4566 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4568 struct saved_alias
*al
;
4570 if (slab_state
== SYSFS
) {
4572 * If we have a leftover link then remove it.
4574 sysfs_remove_link(&slab_kset
->kobj
, name
);
4575 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4578 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4584 al
->next
= alias_list
;
4589 static int __init
slab_sysfs_init(void)
4591 struct kmem_cache
*s
;
4594 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4596 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4602 list_for_each_entry(s
, &slab_caches
, list
) {
4603 err
= sysfs_slab_add(s
);
4605 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4606 " to sysfs\n", s
->name
);
4609 while (alias_list
) {
4610 struct saved_alias
*al
= alias_list
;
4612 alias_list
= alias_list
->next
;
4613 err
= sysfs_slab_alias(al
->s
, al
->name
);
4615 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4616 " %s to sysfs\n", s
->name
);
4624 __initcall(slab_sysfs_init
);
4628 * The /proc/slabinfo ABI
4630 #ifdef CONFIG_SLABINFO
4631 static void print_slabinfo_header(struct seq_file
*m
)
4633 seq_puts(m
, "slabinfo - version: 2.1\n");
4634 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4635 "<objperslab> <pagesperslab>");
4636 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4637 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4641 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4645 down_read(&slub_lock
);
4647 print_slabinfo_header(m
);
4649 return seq_list_start(&slab_caches
, *pos
);
4652 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4654 return seq_list_next(p
, &slab_caches
, pos
);
4657 static void s_stop(struct seq_file
*m
, void *p
)
4659 up_read(&slub_lock
);
4662 static int s_show(struct seq_file
*m
, void *p
)
4664 unsigned long nr_partials
= 0;
4665 unsigned long nr_slabs
= 0;
4666 unsigned long nr_inuse
= 0;
4667 unsigned long nr_objs
= 0;
4668 unsigned long nr_free
= 0;
4669 struct kmem_cache
*s
;
4672 s
= list_entry(p
, struct kmem_cache
, list
);
4674 for_each_online_node(node
) {
4675 struct kmem_cache_node
*n
= get_node(s
, node
);
4680 nr_partials
+= n
->nr_partial
;
4681 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4682 nr_objs
+= atomic_long_read(&n
->total_objects
);
4683 nr_free
+= count_partial(n
, count_free
);
4686 nr_inuse
= nr_objs
- nr_free
;
4688 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4689 nr_objs
, s
->size
, oo_objects(s
->oo
),
4690 (1 << oo_order(s
->oo
)));
4691 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4692 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4698 static const struct seq_operations slabinfo_op
= {
4705 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4707 return seq_open(file
, &slabinfo_op
);
4710 static const struct file_operations proc_slabinfo_operations
= {
4711 .open
= slabinfo_open
,
4713 .llseek
= seq_lseek
,
4714 .release
= seq_release
,
4717 static int __init
slab_proc_init(void)
4719 proc_create("slabinfo",S_IWUSR
|S_IRUGO
,NULL
,&proc_slabinfo_operations
);
4722 module_init(slab_proc_init
);
4723 #endif /* CONFIG_SLABINFO */