1 /* Modified by Broadcom Corp. Portions Copyright (c) Broadcom Corp, 2012. */
3 * SLUB: A slab allocator that limits cache line use instead of queuing
4 * objects in per cpu and per node lists.
6 * The allocator synchronizes using per slab locks and only
7 * uses a centralized lock to manage a pool of partial slabs.
9 * (C) 2007 SGI, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #include <linux/module.h>
15 #include <linux/bit_spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/bitops.h>
18 #include <linux/slab.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/kmemcheck.h>
22 #include <linux/cpu.h>
23 #include <linux/cpuset.h>
24 #include <linux/mempolicy.h>
25 #include <linux/ctype.h>
26 #include <linux/debugobjects.h>
27 #include <linux/kallsyms.h>
28 #include <linux/memory.h>
29 #include <linux/math64.h>
30 #include <linux/fault-inject.h>
40 * The slab_lock protects operations on the object of a particular
41 * slab and its metadata in the page struct. If the slab lock
42 * has been taken then no allocations nor frees can be performed
43 * on the objects in the slab nor can the slab be added or removed
44 * from the partial or full lists since this would mean modifying
45 * the page_struct of the slab.
47 * The list_lock protects the partial and full list on each node and
48 * the partial slab counter. If taken then no new slabs may be added or
49 * removed from the lists nor make the number of partial slabs be modified.
50 * (Note that the total number of slabs is an atomic value that may be
51 * modified without taking the list lock).
53 * The list_lock is a centralized lock and thus we avoid taking it as
54 * much as possible. As long as SLUB does not have to handle partial
55 * slabs, operations can continue without any centralized lock. F.e.
56 * allocating a long series of objects that fill up slabs does not require
59 * The lock order is sometimes inverted when we are trying to get a slab
60 * off a list. We take the list_lock and then look for a page on the list
61 * to use. While we do that objects in the slabs may be freed. We can
62 * only operate on the slab if we have also taken the slab_lock. So we use
63 * a slab_trylock() on the slab. If trylock was successful then no frees
64 * can occur anymore and we can use the slab for allocations etc. If the
65 * slab_trylock() does not succeed then frees are in progress in the slab and
66 * we must stay away from it for a while since we may cause a bouncing
67 * cacheline if we try to acquire the lock. So go onto the next slab.
68 * If all pages are busy then we may allocate a new slab instead of reusing
69 * a partial slab. A new slab has noone operating on it and thus there is
70 * no danger of cacheline contention.
72 * Interrupts are disabled during allocation and deallocation in order to
73 * make the slab allocator safe to use in the context of an irq. In addition
74 * interrupts are disabled to ensure that the processor does not change
75 * while handling per_cpu slabs, due to kernel preemption.
77 * SLUB assigns one slab for allocation to each processor.
78 * Allocations only occur from these slabs called cpu slabs.
80 * Slabs with free elements are kept on a partial list and during regular
81 * operations no list for full slabs is used. If an object in a full slab is
82 * freed then the slab will show up again on the partial lists.
83 * We track full slabs for debugging purposes though because otherwise we
84 * cannot scan all objects.
86 * Slabs are freed when they become empty. Teardown and setup is
87 * minimal so we rely on the page allocators per cpu caches for
88 * fast frees and allocs.
90 * Overloading of page flags that are otherwise used for LRU management.
92 * PageActive The slab is frozen and exempt from list processing.
93 * This means that the slab is dedicated to a purpose
94 * such as satisfying allocations for a specific
95 * processor. Objects may be freed in the slab while
96 * it is frozen but slab_free will then skip the usual
97 * list operations. It is up to the processor holding
98 * the slab to integrate the slab into the slab lists
99 * when the slab is no longer needed.
101 * One use of this flag is to mark slabs that are
102 * used for allocations. Then such a slab becomes a cpu
103 * slab. The cpu slab may be equipped with an additional
104 * freelist that allows lockless access to
105 * free objects in addition to the regular freelist
106 * that requires the slab lock.
108 * PageError Slab requires special handling due to debug
109 * options set. This moves slab handling out of
110 * the fast path and disables lockless freelists.
113 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
114 SLAB_TRACE | SLAB_DEBUG_FREE)
116 static inline int kmem_cache_debug(struct kmem_cache
*s
)
118 #ifdef CONFIG_SLUB_DEBUG
119 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
126 * Issues still to be resolved:
128 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
130 * - Variable sizing of the per node arrays
133 /* Enable to test recovery from slab corruption on boot */
134 #undef SLUB_RESILIENCY_TEST
137 * Mininum number of partial slabs. These will be left on the partial
138 * lists even if they are empty. kmem_cache_shrink may reclaim them.
140 #define MIN_PARTIAL 5
143 * Maximum number of desirable partial slabs.
144 * The existence of more partial slabs makes kmem_cache_shrink
145 * sort the partial list by the number of objects in the.
147 #define MAX_PARTIAL 10
149 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
150 SLAB_POISON | SLAB_STORE_USER)
153 * Debugging flags that require metadata to be stored in the slab. These get
154 * disabled when slub_debug=O is used and a cache's min order increases with
157 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
160 * Set of flags that will prevent slab merging
162 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
163 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
166 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
167 SLAB_CACHE_DMA | SLAB_NOTRACK)
170 #define OO_MASK ((1 << OO_SHIFT) - 1)
171 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
173 /* Internal SLUB flags */
174 #define __OBJECT_POISON 0x80000000UL /* Poison object */
175 #define __SYSFS_ADD_DEFERRED 0x40000000UL /* Not yet visible via sysfs */
177 static int kmem_size
= sizeof(struct kmem_cache
);
180 static struct notifier_block slab_notifier
;
184 DOWN
, /* No slab functionality available */
185 PARTIAL
, /* kmem_cache_open() works but kmalloc does not */
186 UP
, /* Everything works but does not show up in sysfs */
190 /* A list of all slab caches on the system */
191 static DECLARE_RWSEM(slub_lock
);
192 static LIST_HEAD(slab_caches
);
195 * Tracking user of a slab.
198 unsigned long addr
; /* Called from address */
199 int cpu
; /* Was running on cpu */
200 int pid
; /* Pid context */
201 unsigned long when
; /* When did the operation occur */
204 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
206 #ifdef CONFIG_SLUB_DEBUG
207 static int sysfs_slab_add(struct kmem_cache
*);
208 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
209 static void sysfs_slab_remove(struct kmem_cache
*);
212 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
213 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
215 static inline void sysfs_slab_remove(struct kmem_cache
*s
)
222 static inline void stat(struct kmem_cache
*s
, enum stat_item si
)
224 #ifdef CONFIG_SLUB_STATS
225 __this_cpu_inc(s
->cpu_slab
->stat
[si
]);
229 /********************************************************************
230 * Core slab cache functions
231 *******************************************************************/
233 int slab_is_available(void)
235 return slab_state
>= UP
;
238 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
241 return s
->node
[node
];
243 return &s
->local_node
;
247 /* Verify that a pointer has an address that is valid within a slab page */
248 static inline int check_valid_pointer(struct kmem_cache
*s
,
249 struct page
*page
, const void *object
)
256 base
= page_address(page
);
257 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
258 (object
- base
) % s
->size
) {
265 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
267 return *(void **)(object
+ s
->offset
);
270 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
272 *(void **)(object
+ s
->offset
) = fp
;
275 /* Loop over all objects in a slab */
276 #define for_each_object(__p, __s, __addr, __objects) \
277 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
281 #define for_each_free_object(__p, __s, __free) \
282 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
284 /* Determine object index from a given position */
285 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
287 return (p
- addr
) / s
->size
;
290 static inline struct kmem_cache_order_objects
oo_make(int order
,
293 struct kmem_cache_order_objects x
= {
294 (order
<< OO_SHIFT
) + (PAGE_SIZE
<< order
) / size
300 static inline int oo_order(struct kmem_cache_order_objects x
)
302 return x
.x
>> OO_SHIFT
;
305 static inline int oo_objects(struct kmem_cache_order_objects x
)
307 return x
.x
& OO_MASK
;
310 #ifdef CONFIG_SLUB_DEBUG
314 #ifdef CONFIG_SLUB_DEBUG_ON
315 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
317 static int slub_debug
;
320 static char *slub_debug_slabs
;
321 static int disable_higher_order_debug
;
326 static void print_section(char *text
, u8
*addr
, unsigned int length
)
334 for (i
= 0; i
< length
; i
++) {
336 printk(KERN_ERR
"%8s 0x%p: ", text
, addr
+ i
);
339 printk(KERN_CONT
" %02x", addr
[i
]);
341 ascii
[offset
] = isgraph(addr
[i
]) ? addr
[i
] : '.';
343 printk(KERN_CONT
" %s\n", ascii
);
350 printk(KERN_CONT
" ");
354 printk(KERN_CONT
" %s\n", ascii
);
358 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
359 enum track_item alloc
)
364 p
= object
+ s
->offset
+ sizeof(void *);
366 p
= object
+ s
->inuse
;
371 static void set_track(struct kmem_cache
*s
, void *object
,
372 enum track_item alloc
, unsigned long addr
)
374 struct track
*p
= get_track(s
, object
, alloc
);
378 p
->cpu
= smp_processor_id();
379 p
->pid
= current
->pid
;
382 memset(p
, 0, sizeof(struct track
));
385 static void init_tracking(struct kmem_cache
*s
, void *object
)
387 if (!(s
->flags
& SLAB_STORE_USER
))
390 set_track(s
, object
, TRACK_FREE
, 0UL);
391 set_track(s
, object
, TRACK_ALLOC
, 0UL);
394 static void print_track(const char *s
, struct track
*t
)
399 printk(KERN_ERR
"INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
400 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
403 static void print_tracking(struct kmem_cache
*s
, void *object
)
405 if (!(s
->flags
& SLAB_STORE_USER
))
408 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
409 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
412 static void print_page_info(struct page
*page
)
414 printk(KERN_ERR
"INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
415 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
419 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
425 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
427 printk(KERN_ERR
"========================================"
428 "=====================================\n");
429 printk(KERN_ERR
"BUG %s: %s\n", s
->name
, buf
);
430 printk(KERN_ERR
"----------------------------------------"
431 "-------------------------------------\n\n");
434 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
440 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
442 printk(KERN_ERR
"FIX %s: %s\n", s
->name
, buf
);
445 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
447 unsigned int off
; /* Offset of last byte */
448 u8
*addr
= page_address(page
);
450 print_tracking(s
, p
);
452 print_page_info(page
);
454 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
455 p
, p
- addr
, get_freepointer(s
, p
));
458 print_section("Bytes b4", p
- 16, 16);
460 print_section("Object", p
, min_t(unsigned long, s
->objsize
, PAGE_SIZE
));
462 if (s
->flags
& SLAB_RED_ZONE
)
463 print_section("Redzone", p
+ s
->objsize
,
464 s
->inuse
- s
->objsize
);
467 off
= s
->offset
+ sizeof(void *);
471 if (s
->flags
& SLAB_STORE_USER
)
472 off
+= 2 * sizeof(struct track
);
475 /* Beginning of the filler is the free pointer */
476 print_section("Padding", p
+ off
, s
->size
- off
);
481 static void object_err(struct kmem_cache
*s
, struct page
*page
,
482 u8
*object
, char *reason
)
484 slab_bug(s
, "%s", reason
);
485 print_trailer(s
, page
, object
);
488 static void slab_err(struct kmem_cache
*s
, struct page
*page
, char *fmt
, ...)
494 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
496 slab_bug(s
, "%s", buf
);
497 print_page_info(page
);
501 static void init_object(struct kmem_cache
*s
, void *object
, int active
)
505 if (s
->flags
& __OBJECT_POISON
) {
506 memset(p
, POISON_FREE
, s
->objsize
- 1);
507 p
[s
->objsize
- 1] = POISON_END
;
510 if (s
->flags
& SLAB_RED_ZONE
)
511 memset(p
+ s
->objsize
,
512 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
,
513 s
->inuse
- s
->objsize
);
516 static u8
*check_bytes(u8
*start
, unsigned int value
, unsigned int bytes
)
519 if (*start
!= (u8
)value
)
527 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
528 void *from
, void *to
)
530 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
531 memset(from
, data
, to
- from
);
534 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
535 u8
*object
, char *what
,
536 u8
*start
, unsigned int value
, unsigned int bytes
)
541 fault
= check_bytes(start
, value
, bytes
);
546 while (end
> fault
&& end
[-1] == value
)
549 slab_bug(s
, "%s overwritten", what
);
550 printk(KERN_ERR
"INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
551 fault
, end
- 1, fault
[0], value
);
552 print_trailer(s
, page
, object
);
554 restore_bytes(s
, what
, value
, fault
, end
);
562 * Bytes of the object to be managed.
563 * If the freepointer may overlay the object then the free
564 * pointer is the first word of the object.
566 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
569 * object + s->objsize
570 * Padding to reach word boundary. This is also used for Redzoning.
571 * Padding is extended by another word if Redzoning is enabled and
574 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
575 * 0xcc (RED_ACTIVE) for objects in use.
578 * Meta data starts here.
580 * A. Free pointer (if we cannot overwrite object on free)
581 * B. Tracking data for SLAB_STORE_USER
582 * C. Padding to reach required alignment boundary or at mininum
583 * one word if debugging is on to be able to detect writes
584 * before the word boundary.
586 * Padding is done using 0x5a (POISON_INUSE)
589 * Nothing is used beyond s->size.
591 * If slabcaches are merged then the objsize and inuse boundaries are mostly
592 * ignored. And therefore no slab options that rely on these boundaries
593 * may be used with merged slabcaches.
596 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
598 unsigned long off
= s
->inuse
; /* The end of info */
601 /* Freepointer is placed after the object. */
602 off
+= sizeof(void *);
604 if (s
->flags
& SLAB_STORE_USER
)
605 /* We also have user information there */
606 off
+= 2 * sizeof(struct track
);
611 return check_bytes_and_report(s
, page
, p
, "Object padding",
612 p
+ off
, POISON_INUSE
, s
->size
- off
);
615 /* Check the pad bytes at the end of a slab page */
616 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
624 if (!(s
->flags
& SLAB_POISON
))
627 start
= page_address(page
);
628 length
= (PAGE_SIZE
<< compound_order(page
));
629 end
= start
+ length
;
630 remainder
= length
% s
->size
;
634 fault
= check_bytes(end
- remainder
, POISON_INUSE
, remainder
);
637 while (end
> fault
&& end
[-1] == POISON_INUSE
)
640 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
641 print_section("Padding", end
- remainder
, remainder
);
643 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
647 static int check_object(struct kmem_cache
*s
, struct page
*page
,
648 void *object
, int active
)
651 u8
*endobject
= object
+ s
->objsize
;
653 if (s
->flags
& SLAB_RED_ZONE
) {
655 active
? SLUB_RED_ACTIVE
: SLUB_RED_INACTIVE
;
657 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
658 endobject
, red
, s
->inuse
- s
->objsize
))
661 if ((s
->flags
& SLAB_POISON
) && s
->objsize
< s
->inuse
) {
662 check_bytes_and_report(s
, page
, p
, "Alignment padding",
663 endobject
, POISON_INUSE
, s
->inuse
- s
->objsize
);
667 if (s
->flags
& SLAB_POISON
) {
668 if (!active
&& (s
->flags
& __OBJECT_POISON
) &&
669 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
670 POISON_FREE
, s
->objsize
- 1) ||
671 !check_bytes_and_report(s
, page
, p
, "Poison",
672 p
+ s
->objsize
- 1, POISON_END
, 1)))
675 * check_pad_bytes cleans up on its own.
677 check_pad_bytes(s
, page
, p
);
680 if (!s
->offset
&& active
)
682 * Object and freepointer overlap. Cannot check
683 * freepointer while object is allocated.
687 /* Check free pointer validity */
688 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
689 object_err(s
, page
, p
, "Freepointer corrupt");
691 * No choice but to zap it and thus lose the remainder
692 * of the free objects in this slab. May cause
693 * another error because the object count is now wrong.
695 set_freepointer(s
, p
, NULL
);
701 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
705 VM_BUG_ON(!irqs_disabled());
707 if (!PageSlab(page
)) {
708 slab_err(s
, page
, "Not a valid slab page");
712 maxobj
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
713 if (page
->objects
> maxobj
) {
714 slab_err(s
, page
, "objects %u > max %u",
715 s
->name
, page
->objects
, maxobj
);
718 if (page
->inuse
> page
->objects
) {
719 slab_err(s
, page
, "inuse %u > max %u",
720 s
->name
, page
->inuse
, page
->objects
);
723 /* Slab_pad_check fixes things up after itself */
724 slab_pad_check(s
, page
);
729 * Determine if a certain object on a page is on the freelist. Must hold the
730 * slab lock to guarantee that the chains are in a consistent state.
732 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
735 void *fp
= page
->freelist
;
737 unsigned long max_objects
;
739 while (fp
&& nr
<= page
->objects
) {
742 if (!check_valid_pointer(s
, page
, fp
)) {
744 object_err(s
, page
, object
,
745 "Freechain corrupt");
746 set_freepointer(s
, object
, NULL
);
749 slab_err(s
, page
, "Freepointer corrupt");
750 page
->freelist
= NULL
;
751 page
->inuse
= page
->objects
;
752 slab_fix(s
, "Freelist cleared");
758 fp
= get_freepointer(s
, object
);
762 max_objects
= (PAGE_SIZE
<< compound_order(page
)) / s
->size
;
763 if (max_objects
> MAX_OBJS_PER_PAGE
)
764 max_objects
= MAX_OBJS_PER_PAGE
;
766 if (page
->objects
!= max_objects
) {
767 slab_err(s
, page
, "Wrong number of objects. Found %d but "
768 "should be %d", page
->objects
, max_objects
);
769 page
->objects
= max_objects
;
770 slab_fix(s
, "Number of objects adjusted.");
772 if (page
->inuse
!= page
->objects
- nr
) {
773 slab_err(s
, page
, "Wrong object count. Counter is %d but "
774 "counted were %d", page
->inuse
, page
->objects
- nr
);
775 page
->inuse
= page
->objects
- nr
;
776 slab_fix(s
, "Object count adjusted.");
778 return search
== NULL
;
781 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
784 if (s
->flags
& SLAB_TRACE
) {
785 printk(KERN_INFO
"TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
787 alloc
? "alloc" : "free",
792 print_section("Object", (void *)object
, s
->objsize
);
799 * Tracking of fully allocated slabs for debugging purposes.
801 static void add_full(struct kmem_cache_node
*n
, struct page
*page
)
803 spin_lock(&n
->list_lock
);
804 list_add(&page
->lru
, &n
->full
);
805 spin_unlock(&n
->list_lock
);
808 static void remove_full(struct kmem_cache
*s
, struct page
*page
)
810 struct kmem_cache_node
*n
;
812 if (!(s
->flags
& SLAB_STORE_USER
))
815 n
= get_node(s
, page_to_nid(page
));
817 spin_lock(&n
->list_lock
);
818 list_del(&page
->lru
);
819 spin_unlock(&n
->list_lock
);
822 /* Tracking of the number of slabs for debugging purposes */
823 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
825 struct kmem_cache_node
*n
= get_node(s
, node
);
827 return atomic_long_read(&n
->nr_slabs
);
830 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
832 return atomic_long_read(&n
->nr_slabs
);
835 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
837 struct kmem_cache_node
*n
= get_node(s
, node
);
840 * May be called early in order to allocate a slab for the
841 * kmem_cache_node structure. Solve the chicken-egg
842 * dilemma by deferring the increment of the count during
843 * bootstrap (see early_kmem_cache_node_alloc).
845 if (!NUMA_BUILD
|| n
) {
846 atomic_long_inc(&n
->nr_slabs
);
847 atomic_long_add(objects
, &n
->total_objects
);
850 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
852 struct kmem_cache_node
*n
= get_node(s
, node
);
854 atomic_long_dec(&n
->nr_slabs
);
855 atomic_long_sub(objects
, &n
->total_objects
);
858 /* Object debug checks for alloc/free paths */
859 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
862 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
865 init_object(s
, object
, 0);
866 init_tracking(s
, object
);
869 static int alloc_debug_processing(struct kmem_cache
*s
, struct page
*page
,
870 void *object
, unsigned long addr
)
872 if (!check_slab(s
, page
))
875 if (!on_freelist(s
, page
, object
)) {
876 object_err(s
, page
, object
, "Object already allocated");
880 if (!check_valid_pointer(s
, page
, object
)) {
881 object_err(s
, page
, object
, "Freelist Pointer check fails");
885 if (!check_object(s
, page
, object
, 0))
888 /* Success perform special debug activities for allocs */
889 if (s
->flags
& SLAB_STORE_USER
)
890 set_track(s
, object
, TRACK_ALLOC
, addr
);
891 trace(s
, page
, object
, 1);
892 init_object(s
, object
, 1);
896 if (PageSlab(page
)) {
898 * If this is a slab page then lets do the best we can
899 * to avoid issues in the future. Marking all objects
900 * as used avoids touching the remaining objects.
902 slab_fix(s
, "Marking all objects used");
903 page
->inuse
= page
->objects
;
904 page
->freelist
= NULL
;
909 static int free_debug_processing(struct kmem_cache
*s
, struct page
*page
,
910 void *object
, unsigned long addr
)
912 if (!check_slab(s
, page
))
915 if (!check_valid_pointer(s
, page
, object
)) {
916 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
920 if (on_freelist(s
, page
, object
)) {
921 object_err(s
, page
, object
, "Object already free");
925 if (!check_object(s
, page
, object
, 1))
928 if (unlikely(s
!= page
->slab
)) {
929 if (!PageSlab(page
)) {
930 slab_err(s
, page
, "Attempt to free object(0x%p) "
931 "outside of slab", object
);
932 } else if (!page
->slab
) {
934 "SLUB <none>: no slab for object 0x%p.\n",
938 object_err(s
, page
, object
,
939 "page slab pointer corrupt.");
943 /* Special debug activities for freeing objects */
944 if (!PageSlubFrozen(page
) && !page
->freelist
)
945 remove_full(s
, page
);
946 if (s
->flags
& SLAB_STORE_USER
)
947 set_track(s
, object
, TRACK_FREE
, addr
);
948 trace(s
, page
, object
, 0);
949 init_object(s
, object
, 0);
953 slab_fix(s
, "Object at 0x%p not freed", object
);
957 static int __init
setup_slub_debug(char *str
)
959 slub_debug
= DEBUG_DEFAULT_FLAGS
;
960 if (*str
++ != '=' || !*str
)
962 * No options specified. Switch on full debugging.
968 * No options but restriction on slabs. This means full
969 * debugging for slabs matching a pattern.
973 if (tolower(*str
) == 'o') {
975 * Avoid enabling debugging on caches if its minimum order
976 * would increase as a result.
978 disable_higher_order_debug
= 1;
985 * Switch off all debugging measures.
990 * Determine which debug features should be switched on
992 for (; *str
&& *str
!= ','; str
++) {
993 switch (tolower(*str
)) {
995 slub_debug
|= SLAB_DEBUG_FREE
;
998 slub_debug
|= SLAB_RED_ZONE
;
1001 slub_debug
|= SLAB_POISON
;
1004 slub_debug
|= SLAB_STORE_USER
;
1007 slub_debug
|= SLAB_TRACE
;
1010 slub_debug
|= SLAB_FAILSLAB
;
1013 printk(KERN_ERR
"slub_debug option '%c' "
1014 "unknown. skipped\n", *str
);
1020 slub_debug_slabs
= str
+ 1;
1025 __setup("slub_debug", setup_slub_debug
);
1027 static unsigned long kmem_cache_flags(unsigned long objsize
,
1028 unsigned long flags
, const char *name
,
1029 void (*ctor
)(void *))
1032 * Enable debugging if selected on the kernel commandline.
1034 if (slub_debug
&& (!slub_debug_slabs
||
1035 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
))))
1036 flags
|= slub_debug
;
1041 static inline void setup_object_debug(struct kmem_cache
*s
,
1042 struct page
*page
, void *object
) {}
1044 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1045 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1047 static inline int free_debug_processing(struct kmem_cache
*s
,
1048 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1050 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1052 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1053 void *object
, int active
) { return 1; }
1054 static inline void add_full(struct kmem_cache_node
*n
, struct page
*page
) {}
1055 static inline unsigned long kmem_cache_flags(unsigned long objsize
,
1056 unsigned long flags
, const char *name
,
1057 void (*ctor
)(void *))
1061 #define slub_debug 0
1063 #define disable_higher_order_debug 0
1065 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1067 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1069 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1071 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1076 * Slab allocation and freeing
1078 static inline struct page
*alloc_slab_page(gfp_t flags
, int node
,
1079 struct kmem_cache_order_objects oo
)
1081 int order
= oo_order(oo
);
1083 flags
|= __GFP_NOTRACK
;
1085 if (node
== NUMA_NO_NODE
)
1086 return alloc_pages(flags
, order
);
1088 return alloc_pages_exact_node(node
, flags
, order
);
1091 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1094 struct kmem_cache_order_objects oo
= s
->oo
;
1097 flags
|= s
->allocflags
;
1100 * Let the initial higher-order allocation fail under memory pressure
1101 * so we fall-back to the minimum order allocation.
1103 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1105 page
= alloc_slab_page(alloc_gfp
, node
, oo
);
1106 if (unlikely(!page
)) {
1109 * Allocation may have failed due to fragmentation.
1110 * Try a lower order alloc if possible
1112 page
= alloc_slab_page(flags
, node
, oo
);
1116 stat(s
, ORDER_FALLBACK
);
1119 if (kmemcheck_enabled
1120 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1121 int pages
= 1 << oo_order(oo
);
1123 kmemcheck_alloc_shadow(page
, oo_order(oo
), flags
, node
);
1126 * Objects from caches that have a constructor don't get
1127 * cleared when they're allocated, so we need to do it here.
1130 kmemcheck_mark_uninitialized_pages(page
, pages
);
1132 kmemcheck_mark_unallocated_pages(page
, pages
);
1135 page
->objects
= oo_objects(oo
);
1136 mod_zone_page_state(page_zone(page
),
1137 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1138 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1144 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1147 setup_object_debug(s
, page
, object
);
1148 if (unlikely(s
->ctor
))
1152 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1159 BUG_ON(flags
& GFP_SLAB_BUG_MASK
);
1161 page
= allocate_slab(s
,
1162 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1166 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1168 page
->flags
|= 1 << PG_slab
;
1170 start
= page_address(page
);
1172 if (unlikely(s
->flags
& SLAB_POISON
))
1173 memset(start
, POISON_INUSE
, PAGE_SIZE
<< compound_order(page
));
1176 for_each_object(p
, s
, start
, page
->objects
) {
1177 setup_object(s
, page
, last
);
1178 set_freepointer(s
, last
, p
);
1181 setup_object(s
, page
, last
);
1182 set_freepointer(s
, last
, NULL
);
1184 page
->freelist
= start
;
1190 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1192 int order
= compound_order(page
);
1193 int pages
= 1 << order
;
1195 if (kmem_cache_debug(s
)) {
1198 slab_pad_check(s
, page
);
1199 for_each_object(p
, s
, page_address(page
),
1201 check_object(s
, page
, p
, 0);
1204 kmemcheck_free_shadow(page
, compound_order(page
));
1206 mod_zone_page_state(page_zone(page
),
1207 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1208 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1211 __ClearPageSlab(page
);
1212 reset_page_mapcount(page
);
1213 if (current
->reclaim_state
)
1214 current
->reclaim_state
->reclaimed_slab
+= pages
;
1215 __free_pages(page
, order
);
1218 static void rcu_free_slab(struct rcu_head
*h
)
1222 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1223 __free_slab(page
->slab
, page
);
1226 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1228 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1230 * RCU free overloads the RCU head over the LRU
1232 struct rcu_head
*head
= (void *)&page
->lru
;
1234 call_rcu(head
, rcu_free_slab
);
1236 __free_slab(s
, page
);
1239 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1241 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1246 * Per slab locking using the pagelock
1248 static __always_inline
void slab_lock(struct page
*page
)
1250 bit_spin_lock(PG_locked
, &page
->flags
);
1253 static __always_inline
void slab_unlock(struct page
*page
)
1255 __bit_spin_unlock(PG_locked
, &page
->flags
);
1258 static __always_inline
int slab_trylock(struct page
*page
)
1262 rc
= bit_spin_trylock(PG_locked
, &page
->flags
);
1267 * Management of partially allocated slabs
1269 static void add_partial(struct kmem_cache_node
*n
,
1270 struct page
*page
, int tail
)
1272 spin_lock(&n
->list_lock
);
1275 list_add_tail(&page
->lru
, &n
->partial
);
1277 list_add(&page
->lru
, &n
->partial
);
1278 spin_unlock(&n
->list_lock
);
1281 static void remove_partial(struct kmem_cache
*s
, struct page
*page
)
1283 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1285 spin_lock(&n
->list_lock
);
1286 list_del(&page
->lru
);
1288 spin_unlock(&n
->list_lock
);
1292 * Lock slab and remove from the partial list.
1294 * Must hold list_lock.
1296 static inline int lock_and_freeze_slab(struct kmem_cache_node
*n
,
1299 if (slab_trylock(page
)) {
1300 list_del(&page
->lru
);
1302 __SetPageSlubFrozen(page
);
1309 * Try to allocate a partial slab from a specific node.
1311 static struct page
*get_partial_node(struct kmem_cache_node
*n
)
1316 * Racy check. If we mistakenly see no partial slabs then we
1317 * just allocate an empty slab. If we mistakenly try to get a
1318 * partial slab and there is none available then get_partials()
1321 if (!n
|| !n
->nr_partial
)
1324 spin_lock(&n
->list_lock
);
1325 list_for_each_entry(page
, &n
->partial
, lru
)
1326 if (lock_and_freeze_slab(n
, page
))
1330 spin_unlock(&n
->list_lock
);
1335 * Get a page from somewhere. Search in increasing NUMA distances.
1337 static struct page
*get_any_partial(struct kmem_cache
*s
, gfp_t flags
)
1340 struct zonelist
*zonelist
;
1343 enum zone_type high_zoneidx
= gfp_zone(flags
);
1347 * The defrag ratio allows a configuration of the tradeoffs between
1348 * inter node defragmentation and node local allocations. A lower
1349 * defrag_ratio increases the tendency to do local allocations
1350 * instead of attempting to obtain partial slabs from other nodes.
1352 * If the defrag_ratio is set to 0 then kmalloc() always
1353 * returns node local objects. If the ratio is higher then kmalloc()
1354 * may return off node objects because partial slabs are obtained
1355 * from other nodes and filled up.
1357 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1358 * defrag_ratio = 1000) then every (well almost) allocation will
1359 * first attempt to defrag slab caches on other nodes. This means
1360 * scanning over all nodes to look for partial slabs which may be
1361 * expensive if we do it every time we are trying to find a slab
1362 * with available objects.
1364 if (!s
->remote_node_defrag_ratio
||
1365 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1369 zonelist
= node_zonelist(slab_node(current
->mempolicy
), flags
);
1370 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1371 struct kmem_cache_node
*n
;
1373 n
= get_node(s
, zone_to_nid(zone
));
1375 if (n
&& cpuset_zone_allowed_hardwall(zone
, flags
) &&
1376 n
->nr_partial
> s
->min_partial
) {
1377 page
= get_partial_node(n
);
1390 * Get a partial page, lock it and return it.
1392 static struct page
*get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
)
1395 int searchnode
= (node
== NUMA_NO_NODE
) ? numa_node_id() : node
;
1397 page
= get_partial_node(get_node(s
, searchnode
));
1398 if (page
|| node
!= -1)
1401 return get_any_partial(s
, flags
);
1405 * Move a page back to the lists.
1407 * Must be called with the slab lock held.
1409 * On exit the slab lock will have been dropped.
1411 static void unfreeze_slab(struct kmem_cache
*s
, struct page
*page
, int tail
)
1413 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1415 __ClearPageSlubFrozen(page
);
1418 if (page
->freelist
) {
1419 add_partial(n
, page
, tail
);
1420 stat(s
, tail
? DEACTIVATE_TO_TAIL
: DEACTIVATE_TO_HEAD
);
1422 stat(s
, DEACTIVATE_FULL
);
1423 if (kmem_cache_debug(s
) && (s
->flags
& SLAB_STORE_USER
))
1428 stat(s
, DEACTIVATE_EMPTY
);
1429 if (n
->nr_partial
< s
->min_partial
) {
1431 * Adding an empty slab to the partial slabs in order
1432 * to avoid page allocator overhead. This slab needs
1433 * to come after the other slabs with objects in
1434 * so that the others get filled first. That way the
1435 * size of the partial list stays small.
1437 * kmem_cache_shrink can reclaim any empty slabs from
1440 add_partial(n
, page
, 1);
1445 discard_slab(s
, page
);
1451 * Remove the cpu slab
1453 static void deactivate_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1455 struct page
*page
= c
->page
;
1459 stat(s
, DEACTIVATE_REMOTE_FREES
);
1461 * Merge cpu freelist into slab freelist. Typically we get here
1462 * because both freelists are empty. So this is unlikely
1465 while (unlikely(c
->freelist
)) {
1468 tail
= 0; /* Hot objects. Put the slab first */
1470 /* Retrieve object from cpu_freelist */
1471 object
= c
->freelist
;
1472 c
->freelist
= get_freepointer(s
, c
->freelist
);
1474 /* And put onto the regular freelist */
1475 set_freepointer(s
, object
, page
->freelist
);
1476 page
->freelist
= object
;
1480 unfreeze_slab(s
, page
, tail
);
1483 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
1485 stat(s
, CPUSLAB_FLUSH
);
1487 deactivate_slab(s
, c
);
1493 * Called from IPI handler with interrupts disabled.
1495 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
1497 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
1499 if (likely(c
&& c
->page
))
1503 static void flush_cpu_slab(void *d
)
1505 struct kmem_cache
*s
= d
;
1507 __flush_cpu_slab(s
, smp_processor_id());
1510 static void flush_all(struct kmem_cache
*s
)
1512 on_each_cpu(flush_cpu_slab
, s
, 1);
1516 * Check if the objects in a per cpu structure fit numa
1517 * locality expectations.
1519 static inline int node_match(struct kmem_cache_cpu
*c
, int node
)
1522 if (node
!= NUMA_NO_NODE
&& c
->node
!= node
)
1528 static int count_free(struct page
*page
)
1530 return page
->objects
- page
->inuse
;
1533 static unsigned long count_partial(struct kmem_cache_node
*n
,
1534 int (*get_count
)(struct page
*))
1536 unsigned long flags
;
1537 unsigned long x
= 0;
1540 spin_lock_irqsave(&n
->list_lock
, flags
);
1541 list_for_each_entry(page
, &n
->partial
, lru
)
1542 x
+= get_count(page
);
1543 spin_unlock_irqrestore(&n
->list_lock
, flags
);
1547 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
1549 #ifdef CONFIG_SLUB_DEBUG
1550 return atomic_long_read(&n
->total_objects
);
1556 static noinline
void
1557 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
1562 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1564 printk(KERN_WARNING
" cache: %s, object size: %d, buffer size: %d, "
1565 "default order: %d, min order: %d\n", s
->name
, s
->objsize
,
1566 s
->size
, oo_order(s
->oo
), oo_order(s
->min
));
1568 if (oo_order(s
->min
) > get_order(s
->objsize
))
1569 printk(KERN_WARNING
" %s debugging increased min order, use "
1570 "slub_debug=O to disable.\n", s
->name
);
1572 for_each_online_node(node
) {
1573 struct kmem_cache_node
*n
= get_node(s
, node
);
1574 unsigned long nr_slabs
;
1575 unsigned long nr_objs
;
1576 unsigned long nr_free
;
1581 nr_free
= count_partial(n
, count_free
);
1582 nr_slabs
= node_nr_slabs(n
);
1583 nr_objs
= node_nr_objs(n
);
1586 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1587 node
, nr_slabs
, nr_objs
, nr_free
);
1592 * Slow path. The lockless freelist is empty or we need to perform
1595 * Interrupts are disabled.
1597 * Processing is still very fast if new objects have been freed to the
1598 * regular freelist. In that case we simply take over the regular freelist
1599 * as the lockless freelist and zap the regular freelist.
1601 * If that is not working then we fall back to the partial lists. We take the
1602 * first element of the freelist as the object to allocate now and move the
1603 * rest of the freelist to the lockless freelist.
1605 * And if we were unable to get a new slab from the partial slab lists then
1606 * we need to allocate a new slab. This is the slowest path since it involves
1607 * a call to the page allocator and the setup of a new slab.
1609 static void * BCMFASTPATH_HOST
__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
1610 unsigned long addr
, struct kmem_cache_cpu
*c
)
1615 /* We handle __GFP_ZERO in the caller */
1616 gfpflags
&= ~__GFP_ZERO
;
1622 if (unlikely(!node_match(c
, node
)))
1625 stat(s
, ALLOC_REFILL
);
1628 object
= c
->page
->freelist
;
1629 if (unlikely(!object
))
1631 if (kmem_cache_debug(s
))
1634 c
->freelist
= get_freepointer(s
, object
);
1635 c
->page
->inuse
= c
->page
->objects
;
1636 c
->page
->freelist
= NULL
;
1637 c
->node
= page_to_nid(c
->page
);
1639 slab_unlock(c
->page
);
1640 stat(s
, ALLOC_SLOWPATH
);
1644 deactivate_slab(s
, c
);
1647 new = get_partial(s
, gfpflags
, node
);
1650 stat(s
, ALLOC_FROM_PARTIAL
);
1654 if (gfpflags
& __GFP_WAIT
)
1657 new = new_slab(s
, gfpflags
, node
);
1659 if (gfpflags
& __GFP_WAIT
)
1660 local_irq_disable();
1663 c
= __this_cpu_ptr(s
->cpu_slab
);
1664 stat(s
, ALLOC_SLAB
);
1668 __SetPageSlubFrozen(new);
1672 if (!(gfpflags
& __GFP_NOWARN
) && printk_ratelimit())
1673 slab_out_of_memory(s
, gfpflags
, node
);
1676 if (!alloc_debug_processing(s
, c
->page
, object
, addr
))
1680 c
->page
->freelist
= get_freepointer(s
, object
);
1686 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1687 * have the fastpath folded into their functions. So no function call
1688 * overhead for requests that can be satisfied on the fastpath.
1690 * The fastpath works by first checking if the lockless freelist can be used.
1691 * If not then __slab_alloc is called for slow processing.
1693 * Otherwise we can simply pick the next object from the lockless free list.
1695 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
1696 gfp_t gfpflags
, int node
, unsigned long addr
)
1699 struct kmem_cache_cpu
*c
;
1700 unsigned long flags
;
1702 gfpflags
&= gfp_allowed_mask
;
1704 lockdep_trace_alloc(gfpflags
);
1705 might_sleep_if(gfpflags
& __GFP_WAIT
);
1707 if (should_failslab(s
->objsize
, gfpflags
, s
->flags
))
1710 local_irq_save(flags
);
1711 c
= __this_cpu_ptr(s
->cpu_slab
);
1712 object
= c
->freelist
;
1713 if (unlikely(!object
|| !node_match(c
, node
)))
1715 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
1718 c
->freelist
= get_freepointer(s
, object
);
1719 stat(s
, ALLOC_FASTPATH
);
1721 local_irq_restore(flags
);
1723 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
1724 memset(object
, 0, s
->objsize
);
1726 kmemcheck_slab_alloc(s
, gfpflags
, object
, s
->objsize
);
1727 kmemleak_alloc_recursive(object
, s
->objsize
, 1, s
->flags
, gfpflags
);
1732 void * BCMFASTPATH_HOST
kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
1734 void *ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1736 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->objsize
, s
->size
, gfpflags
);
1740 EXPORT_SYMBOL(kmem_cache_alloc
);
1742 #ifdef CONFIG_TRACING
1743 void *kmem_cache_alloc_notrace(struct kmem_cache
*s
, gfp_t gfpflags
)
1745 return slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, _RET_IP_
);
1747 EXPORT_SYMBOL(kmem_cache_alloc_notrace
);
1751 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
1753 void *ret
= slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1755 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
1756 s
->objsize
, s
->size
, gfpflags
, node
);
1760 EXPORT_SYMBOL(kmem_cache_alloc_node
);
1763 #ifdef CONFIG_TRACING
1764 void *kmem_cache_alloc_node_notrace(struct kmem_cache
*s
,
1768 return slab_alloc(s
, gfpflags
, node
, _RET_IP_
);
1770 EXPORT_SYMBOL(kmem_cache_alloc_node_notrace
);
1774 * Slow patch handling. This may still be called frequently since objects
1775 * have a longer lifetime than the cpu slabs in most processing loads.
1777 * So we still attempt to reduce cache line usage. Just take the slab
1778 * lock and free the item. If there is no additional partial page
1779 * handling required then we can return immediately.
1781 static void BCMFASTPATH_HOST
__slab_free(struct kmem_cache
*s
, struct page
*page
,
1782 void *x
, unsigned long addr
)
1785 void **object
= (void *)x
;
1787 stat(s
, FREE_SLOWPATH
);
1790 if (kmem_cache_debug(s
))
1794 prior
= page
->freelist
;
1795 set_freepointer(s
, object
, prior
);
1796 page
->freelist
= object
;
1799 if (unlikely(PageSlubFrozen(page
))) {
1800 stat(s
, FREE_FROZEN
);
1804 if (unlikely(!page
->inuse
))
1808 * Objects left in the slab. If it was not on the partial list before
1811 if (unlikely(!prior
)) {
1812 add_partial(get_node(s
, page_to_nid(page
)), page
, 1);
1813 stat(s
, FREE_ADD_PARTIAL
);
1823 * Slab still on the partial list.
1825 remove_partial(s
, page
);
1826 stat(s
, FREE_REMOVE_PARTIAL
);
1830 discard_slab(s
, page
);
1834 if (!free_debug_processing(s
, page
, x
, addr
))
1840 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1841 * can perform fastpath freeing without additional function calls.
1843 * The fastpath is only possible if we are freeing to the current cpu slab
1844 * of this processor. This typically the case if we have just allocated
1847 * If fastpath is not possible then fall back to __slab_free where we deal
1848 * with all sorts of special processing.
1850 static __always_inline
void slab_free(struct kmem_cache
*s
,
1851 struct page
*page
, void *x
, unsigned long addr
)
1853 void **object
= (void *)x
;
1854 struct kmem_cache_cpu
*c
;
1855 unsigned long flags
;
1857 kmemleak_free_recursive(x
, s
->flags
);
1858 local_irq_save(flags
);
1859 c
= __this_cpu_ptr(s
->cpu_slab
);
1860 kmemcheck_slab_free(s
, object
, s
->objsize
);
1861 debug_check_no_locks_freed(object
, s
->objsize
);
1862 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1863 debug_check_no_obj_freed(object
, s
->objsize
);
1864 if (likely(page
== c
->page
&& c
->node
>= 0)) {
1865 set_freepointer(s
, object
, c
->freelist
);
1866 c
->freelist
= object
;
1867 stat(s
, FREE_FASTPATH
);
1869 __slab_free(s
, page
, x
, addr
);
1871 local_irq_restore(flags
);
1874 void BCMFASTPATH_HOST
kmem_cache_free(struct kmem_cache
*s
, void *x
)
1878 page
= virt_to_head_page(x
);
1880 slab_free(s
, page
, x
, _RET_IP_
);
1882 trace_kmem_cache_free(_RET_IP_
, x
);
1884 EXPORT_SYMBOL(kmem_cache_free
);
1886 /* Figure out on which slab page the object resides */
1887 static struct page
*get_object_page(const void *x
)
1889 struct page
*page
= virt_to_head_page(x
);
1891 if (!PageSlab(page
))
1898 * Object placement in a slab is made very easy because we always start at
1899 * offset 0. If we tune the size of the object to the alignment then we can
1900 * get the required alignment by putting one properly sized object after
1903 * Notice that the allocation order determines the sizes of the per cpu
1904 * caches. Each processor has always one slab available for allocations.
1905 * Increasing the allocation order reduces the number of times that slabs
1906 * must be moved on and off the partial lists and is therefore a factor in
1911 * Mininum / Maximum order of slab pages. This influences locking overhead
1912 * and slab fragmentation. A higher order reduces the number of partial slabs
1913 * and increases the number of allocations possible without having to
1914 * take the list_lock.
1916 static int slub_min_order
;
1917 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
1918 static int slub_min_objects
;
1921 * Merge control. If this is set then no merging of slab caches will occur.
1922 * (Could be removed. This was introduced to pacify the merge skeptics.)
1924 static int slub_nomerge
;
1927 * Calculate the order of allocation given an slab object size.
1929 * The order of allocation has significant impact on performance and other
1930 * system components. Generally order 0 allocations should be preferred since
1931 * order 0 does not cause fragmentation in the page allocator. Larger objects
1932 * be problematic to put into order 0 slabs because there may be too much
1933 * unused space left. We go to a higher order if more than 1/16th of the slab
1936 * In order to reach satisfactory performance we must ensure that a minimum
1937 * number of objects is in one slab. Otherwise we may generate too much
1938 * activity on the partial lists which requires taking the list_lock. This is
1939 * less a concern for large slabs though which are rarely used.
1941 * slub_max_order specifies the order where we begin to stop considering the
1942 * number of objects in a slab as critical. If we reach slub_max_order then
1943 * we try to keep the page order as low as possible. So we accept more waste
1944 * of space in favor of a small page order.
1946 * Higher order allocations also allow the placement of more objects in a
1947 * slab and thereby reduce object handling overhead. If the user has
1948 * requested a higher mininum order then we start with that one instead of
1949 * the smallest order which will fit the object.
1951 static inline int slab_order(int size
, int min_objects
,
1952 int max_order
, int fract_leftover
)
1956 int min_order
= slub_min_order
;
1958 if ((PAGE_SIZE
<< min_order
) / size
> MAX_OBJS_PER_PAGE
)
1959 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
1961 for (order
= max(min_order
,
1962 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
1963 order
<= max_order
; order
++) {
1965 unsigned long slab_size
= PAGE_SIZE
<< order
;
1967 if (slab_size
< min_objects
* size
)
1970 rem
= slab_size
% size
;
1972 if (rem
<= slab_size
/ fract_leftover
)
1980 static inline int calculate_order(int size
)
1988 * Attempt to find best configuration for a slab. This
1989 * works by first attempting to generate a layout with
1990 * the best configuration and backing off gradually.
1992 * First we reduce the acceptable waste in a slab. Then
1993 * we reduce the minimum objects required in a slab.
1995 min_objects
= slub_min_objects
;
1997 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
1998 max_objects
= (PAGE_SIZE
<< slub_max_order
)/size
;
1999 min_objects
= min(min_objects
, max_objects
);
2001 while (min_objects
> 1) {
2003 while (fraction
>= 4) {
2004 order
= slab_order(size
, min_objects
,
2005 slub_max_order
, fraction
);
2006 if (order
<= slub_max_order
)
2014 * We were unable to place multiple objects in a slab. Now
2015 * lets see if we can place a single object there.
2017 order
= slab_order(size
, 1, slub_max_order
, 1);
2018 if (order
<= slub_max_order
)
2022 * Doh this slab cannot be placed using slub_max_order.
2024 order
= slab_order(size
, 1, MAX_ORDER
, 1);
2025 if (order
< MAX_ORDER
)
2031 * Figure out what the alignment of the objects will be.
2033 static unsigned long calculate_alignment(unsigned long flags
,
2034 unsigned long align
, unsigned long size
)
2037 * If the user wants hardware cache aligned objects then follow that
2038 * suggestion if the object is sufficiently large.
2040 * The hardware cache alignment cannot override the specified
2041 * alignment though. If that is greater then use it.
2043 if (flags
& SLAB_HWCACHE_ALIGN
) {
2044 unsigned long ralign
= cache_line_size();
2045 while (size
<= ralign
/ 2)
2047 align
= max(align
, ralign
);
2050 if (align
< ARCH_SLAB_MINALIGN
)
2051 align
= ARCH_SLAB_MINALIGN
;
2053 return ALIGN(align
, sizeof(void *));
2057 init_kmem_cache_node(struct kmem_cache_node
*n
, struct kmem_cache
*s
)
2060 spin_lock_init(&n
->list_lock
);
2061 INIT_LIST_HEAD(&n
->partial
);
2062 #ifdef CONFIG_SLUB_DEBUG
2063 atomic_long_set(&n
->nr_slabs
, 0);
2064 atomic_long_set(&n
->total_objects
, 0);
2065 INIT_LIST_HEAD(&n
->full
);
2069 static DEFINE_PER_CPU(struct kmem_cache_cpu
, kmalloc_percpu
[KMALLOC_CACHES
]);
2071 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
, gfp_t flags
)
2073 if (s
< kmalloc_caches
+ KMALLOC_CACHES
&& s
>= kmalloc_caches
)
2075 * Boot time creation of the kmalloc array. Use static per cpu data
2076 * since the per cpu allocator is not available yet.
2078 s
->cpu_slab
= kmalloc_percpu
+ (s
- kmalloc_caches
);
2080 s
->cpu_slab
= alloc_percpu(struct kmem_cache_cpu
);
2090 * No kmalloc_node yet so do it by hand. We know that this is the first
2091 * slab on the node for this slabcache. There are no concurrent accesses
2094 * Note that this function only works on the kmalloc_node_cache
2095 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2096 * memory on a fresh node that has no slab structures yet.
2098 static void early_kmem_cache_node_alloc(gfp_t gfpflags
, int node
)
2101 struct kmem_cache_node
*n
;
2102 unsigned long flags
;
2104 BUG_ON(kmalloc_caches
->size
< sizeof(struct kmem_cache_node
));
2106 page
= new_slab(kmalloc_caches
, gfpflags
, node
);
2109 if (page_to_nid(page
) != node
) {
2110 printk(KERN_ERR
"SLUB: Unable to allocate memory from "
2112 printk(KERN_ERR
"SLUB: Allocating a useless per node structure "
2113 "in order to be able to continue\n");
2118 page
->freelist
= get_freepointer(kmalloc_caches
, n
);
2120 kmalloc_caches
->node
[node
] = n
;
2121 #ifdef CONFIG_SLUB_DEBUG
2122 init_object(kmalloc_caches
, n
, 1);
2123 init_tracking(kmalloc_caches
, n
);
2125 init_kmem_cache_node(n
, kmalloc_caches
);
2126 inc_slabs_node(kmalloc_caches
, node
, page
->objects
);
2129 * lockdep requires consistent irq usage for each lock
2130 * so even though there cannot be a race this early in
2131 * the boot sequence, we still disable irqs.
2133 local_irq_save(flags
);
2134 add_partial(n
, page
, 0);
2135 local_irq_restore(flags
);
2138 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2142 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2143 struct kmem_cache_node
*n
= s
->node
[node
];
2145 kmem_cache_free(kmalloc_caches
, n
);
2146 s
->node
[node
] = NULL
;
2150 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2154 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2155 struct kmem_cache_node
*n
;
2157 if (slab_state
== DOWN
) {
2158 early_kmem_cache_node_alloc(gfpflags
, node
);
2161 n
= kmem_cache_alloc_node(kmalloc_caches
,
2165 free_kmem_cache_nodes(s
);
2170 init_kmem_cache_node(n
, s
);
2175 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2179 static int init_kmem_cache_nodes(struct kmem_cache
*s
, gfp_t gfpflags
)
2181 init_kmem_cache_node(&s
->local_node
, s
);
2186 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2188 if (min
< MIN_PARTIAL
)
2190 else if (min
> MAX_PARTIAL
)
2192 s
->min_partial
= min
;
2196 * calculate_sizes() determines the order and the distribution of data within
2199 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2201 unsigned long flags
= s
->flags
;
2202 unsigned long size
= s
->objsize
;
2203 unsigned long align
= s
->align
;
2207 * Round up object size to the next word boundary. We can only
2208 * place the free pointer at word boundaries and this determines
2209 * the possible location of the free pointer.
2211 size
= ALIGN(size
, sizeof(void *));
2213 #ifdef CONFIG_SLUB_DEBUG
2215 * Determine if we can poison the object itself. If the user of
2216 * the slab may touch the object after free or before allocation
2217 * then we should never poison the object itself.
2219 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2221 s
->flags
|= __OBJECT_POISON
;
2223 s
->flags
&= ~__OBJECT_POISON
;
2227 * If we are Redzoning then check if there is some space between the
2228 * end of the object and the free pointer. If not then add an
2229 * additional word to have some bytes to store Redzone information.
2231 if ((flags
& SLAB_RED_ZONE
) && size
== s
->objsize
)
2232 size
+= sizeof(void *);
2236 * With that we have determined the number of bytes in actual use
2237 * by the object. This is the potential offset to the free pointer.
2241 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
2244 * Relocate free pointer after the object if it is not
2245 * permitted to overwrite the first word of the object on
2248 * This is the case if we do RCU, have a constructor or
2249 * destructor or are poisoning the objects.
2252 size
+= sizeof(void *);
2255 #ifdef CONFIG_SLUB_DEBUG
2256 if (flags
& SLAB_STORE_USER
)
2258 * Need to store information about allocs and frees after
2261 size
+= 2 * sizeof(struct track
);
2263 if (flags
& SLAB_RED_ZONE
)
2265 * Add some empty padding so that we can catch
2266 * overwrites from earlier objects rather than let
2267 * tracking information or the free pointer be
2268 * corrupted if a user writes before the start
2271 size
+= sizeof(void *);
2275 * Determine the alignment based on various parameters that the
2276 * user specified and the dynamic determination of cache line size
2279 align
= calculate_alignment(flags
, align
, s
->objsize
);
2283 * SLUB stores one object immediately after another beginning from
2284 * offset 0. In order to align the objects we have to simply size
2285 * each object to conform to the alignment.
2287 size
= ALIGN(size
, align
);
2289 if (forced_order
>= 0)
2290 order
= forced_order
;
2292 order
= calculate_order(size
);
2299 s
->allocflags
|= __GFP_COMP
;
2301 if (s
->flags
& SLAB_CACHE_DMA
)
2302 s
->allocflags
|= SLUB_DMA
;
2304 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
2305 s
->allocflags
|= __GFP_RECLAIMABLE
;
2308 * Determine the number of objects per slab
2310 s
->oo
= oo_make(order
, size
);
2311 s
->min
= oo_make(get_order(size
), size
);
2312 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
2315 return !!oo_objects(s
->oo
);
2319 static int kmem_cache_open(struct kmem_cache
*s
, gfp_t gfpflags
,
2320 const char *name
, size_t size
,
2321 size_t align
, unsigned long flags
,
2322 void (*ctor
)(void *))
2324 memset(s
, 0, kmem_size
);
2329 s
->flags
= kmem_cache_flags(size
, flags
, name
, ctor
);
2331 if (!calculate_sizes(s
, -1))
2333 if (disable_higher_order_debug
) {
2335 * Disable debugging flags that store metadata if the min slab
2338 if (get_order(s
->size
) > get_order(s
->objsize
)) {
2339 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
2341 if (!calculate_sizes(s
, -1))
2347 * The larger the object size is, the more pages we want on the partial
2348 * list to avoid pounding the page allocator excessively.
2350 set_min_partial(s
, ilog2(s
->size
));
2353 s
->remote_node_defrag_ratio
= 1000;
2355 if (!init_kmem_cache_nodes(s
, gfpflags
& ~SLUB_DMA
))
2358 if (alloc_kmem_cache_cpus(s
, gfpflags
& ~SLUB_DMA
))
2361 free_kmem_cache_nodes(s
);
2363 if (flags
& SLAB_PANIC
)
2364 panic("Cannot create slab %s size=%lu realsize=%u "
2365 "order=%u offset=%u flags=%lx\n",
2366 s
->name
, (unsigned long)size
, s
->size
, oo_order(s
->oo
),
2372 * Check if a given pointer is valid
2374 int kmem_ptr_validate(struct kmem_cache
*s
, const void *object
)
2378 if (!kern_ptr_validate(object
, s
->size
))
2381 page
= get_object_page(object
);
2383 if (!page
|| s
!= page
->slab
)
2384 /* No slab or wrong slab */
2387 if (!check_valid_pointer(s
, page
, object
))
2391 * We could also check if the object is on the slabs freelist.
2392 * But this would be too expensive and it seems that the main
2393 * purpose of kmem_ptr_valid() is to check if the object belongs
2394 * to a certain slab.
2398 EXPORT_SYMBOL(kmem_ptr_validate
);
2401 * Determine the size of a slab object
2403 unsigned int kmem_cache_size(struct kmem_cache
*s
)
2407 EXPORT_SYMBOL(kmem_cache_size
);
2409 const char *kmem_cache_name(struct kmem_cache
*s
)
2413 EXPORT_SYMBOL(kmem_cache_name
);
2415 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
2418 #ifdef CONFIG_SLUB_DEBUG
2419 void *addr
= page_address(page
);
2421 long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) * sizeof(long),
2426 slab_err(s
, page
, "%s", text
);
2428 for_each_free_object(p
, s
, page
->freelist
)
2429 set_bit(slab_index(p
, s
, addr
), map
);
2431 for_each_object(p
, s
, addr
, page
->objects
) {
2433 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
2434 printk(KERN_ERR
"INFO: Object 0x%p @offset=%tu\n",
2436 print_tracking(s
, p
);
2445 * Attempt to free all partial slabs on a node.
2447 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
2449 unsigned long flags
;
2450 struct page
*page
, *h
;
2452 spin_lock_irqsave(&n
->list_lock
, flags
);
2453 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
2455 list_del(&page
->lru
);
2456 discard_slab(s
, page
);
2459 list_slab_objects(s
, page
,
2460 "Objects remaining on kmem_cache_close()");
2463 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2467 * Release all resources used by a slab cache.
2469 static inline int kmem_cache_close(struct kmem_cache
*s
)
2474 free_percpu(s
->cpu_slab
);
2475 /* Attempt to free all objects */
2476 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2477 struct kmem_cache_node
*n
= get_node(s
, node
);
2480 if (n
->nr_partial
|| slabs_node(s
, node
))
2483 free_kmem_cache_nodes(s
);
2488 * Close a cache and release the kmem_cache structure
2489 * (must be used for caches created using kmem_cache_create)
2491 void kmem_cache_destroy(struct kmem_cache
*s
)
2493 down_write(&slub_lock
);
2497 if (kmem_cache_close(s
)) {
2498 printk(KERN_ERR
"SLUB %s: %s called for cache that "
2499 "still has objects.\n", s
->name
, __func__
);
2502 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
2504 sysfs_slab_remove(s
);
2506 up_write(&slub_lock
);
2508 EXPORT_SYMBOL(kmem_cache_destroy
);
2510 /********************************************************************
2512 *******************************************************************/
2514 struct kmem_cache kmalloc_caches
[KMALLOC_CACHES
] __cacheline_aligned
;
2515 EXPORT_SYMBOL(kmalloc_caches
);
2517 static int __init
setup_slub_min_order(char *str
)
2519 get_option(&str
, &slub_min_order
);
2524 __setup("slub_min_order=", setup_slub_min_order
);
2526 static int __init
setup_slub_max_order(char *str
)
2528 get_option(&str
, &slub_max_order
);
2529 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
2534 __setup("slub_max_order=", setup_slub_max_order
);
2536 static int __init
setup_slub_min_objects(char *str
)
2538 get_option(&str
, &slub_min_objects
);
2543 __setup("slub_min_objects=", setup_slub_min_objects
);
2545 static int __init
setup_slub_nomerge(char *str
)
2551 __setup("slub_nomerge", setup_slub_nomerge
);
2553 static struct kmem_cache
*create_kmalloc_cache(struct kmem_cache
*s
,
2554 const char *name
, int size
, gfp_t gfp_flags
)
2556 unsigned int flags
= 0;
2558 if (gfp_flags
& SLUB_DMA
)
2559 flags
= SLAB_CACHE_DMA
;
2562 * This function is called with IRQs disabled during early-boot on
2563 * single CPU so there's no need to take slub_lock here.
2565 if (!kmem_cache_open(s
, gfp_flags
, name
, size
, ARCH_KMALLOC_MINALIGN
,
2569 list_add(&s
->list
, &slab_caches
);
2571 if (sysfs_slab_add(s
))
2576 panic("Creation of kmalloc slab %s size=%d failed.\n", name
, size
);
2579 #ifdef CONFIG_ZONE_DMA
2580 static struct kmem_cache
*kmalloc_caches_dma
[SLUB_PAGE_SHIFT
];
2582 static void sysfs_add_func(struct work_struct
*w
)
2584 struct kmem_cache
*s
;
2586 down_write(&slub_lock
);
2587 list_for_each_entry(s
, &slab_caches
, list
) {
2588 if (s
->flags
& __SYSFS_ADD_DEFERRED
) {
2589 s
->flags
&= ~__SYSFS_ADD_DEFERRED
;
2593 up_write(&slub_lock
);
2596 static DECLARE_WORK(sysfs_add_work
, sysfs_add_func
);
2598 static noinline
struct kmem_cache
*dma_kmalloc_cache(int index
, gfp_t flags
)
2600 struct kmem_cache
*s
;
2603 unsigned long slabflags
;
2606 s
= kmalloc_caches_dma
[index
];
2610 /* Dynamically create dma cache */
2611 if (flags
& __GFP_WAIT
)
2612 down_write(&slub_lock
);
2614 if (!down_write_trylock(&slub_lock
))
2618 if (kmalloc_caches_dma
[index
])
2621 realsize
= kmalloc_caches
[index
].objsize
;
2622 text
= kasprintf(flags
& ~SLUB_DMA
, "kmalloc_dma-%d",
2623 (unsigned int)realsize
);
2626 for (i
= 0; i
< KMALLOC_CACHES
; i
++)
2627 if (!kmalloc_caches
[i
].size
)
2630 BUG_ON(i
>= KMALLOC_CACHES
);
2631 s
= kmalloc_caches
+ i
;
2634 * Must defer sysfs creation to a workqueue because we don't know
2635 * what context we are called from. Before sysfs comes up, we don't
2636 * need to do anything because our sysfs initcall will start by
2637 * adding all existing slabs to sysfs.
2639 slabflags
= SLAB_CACHE_DMA
|SLAB_NOTRACK
;
2640 if (slab_state
>= SYSFS
)
2641 slabflags
|= __SYSFS_ADD_DEFERRED
;
2643 if (!text
|| !kmem_cache_open(s
, flags
, text
,
2644 realsize
, ARCH_KMALLOC_MINALIGN
, slabflags
, NULL
)) {
2650 list_add(&s
->list
, &slab_caches
);
2651 kmalloc_caches_dma
[index
] = s
;
2653 if (slab_state
>= SYSFS
)
2654 schedule_work(&sysfs_add_work
);
2657 up_write(&slub_lock
);
2659 return kmalloc_caches_dma
[index
];
2664 * Conversion table for small slabs sizes / 8 to the index in the
2665 * kmalloc array. This is necessary for slabs < 192 since we have non power
2666 * of two cache sizes there. The size of larger slabs can be determined using
2669 static s8 size_index
[24] = {
2696 static inline int size_index_elem(size_t bytes
)
2698 return (bytes
- 1) / 8;
2701 static struct kmem_cache
*get_slab(size_t size
, gfp_t flags
)
2707 return ZERO_SIZE_PTR
;
2709 index
= size_index
[size_index_elem(size
)];
2711 index
= fls(size
- 1);
2713 #ifdef CONFIG_ZONE_DMA
2714 if (unlikely((flags
& SLUB_DMA
)))
2715 return dma_kmalloc_cache(index
, flags
);
2718 return &kmalloc_caches
[index
];
2721 void *__kmalloc(size_t size
, gfp_t flags
)
2723 struct kmem_cache
*s
;
2726 if (unlikely(size
> SLUB_MAX_SIZE
))
2727 return kmalloc_large(size
, flags
);
2729 s
= get_slab(size
, flags
);
2731 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2734 ret
= slab_alloc(s
, flags
, NUMA_NO_NODE
, _RET_IP_
);
2736 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
2740 EXPORT_SYMBOL(__kmalloc
);
2742 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
2747 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
2748 page
= alloc_pages_node(node
, flags
, get_order(size
));
2750 ptr
= page_address(page
);
2752 kmemleak_alloc(ptr
, size
, 1, flags
);
2757 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
2759 struct kmem_cache
*s
;
2762 if (unlikely(size
> SLUB_MAX_SIZE
)) {
2763 ret
= kmalloc_large_node(size
, flags
, node
);
2765 trace_kmalloc_node(_RET_IP_
, ret
,
2766 size
, PAGE_SIZE
<< get_order(size
),
2772 s
= get_slab(size
, flags
);
2774 if (unlikely(ZERO_OR_NULL_PTR(s
)))
2777 ret
= slab_alloc(s
, flags
, node
, _RET_IP_
);
2779 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
2783 EXPORT_SYMBOL(__kmalloc_node
);
2786 size_t ksize(const void *object
)
2789 struct kmem_cache
*s
;
2791 if (unlikely(object
== ZERO_SIZE_PTR
))
2794 page
= virt_to_head_page(object
);
2796 if (unlikely(!PageSlab(page
))) {
2797 WARN_ON(!PageCompound(page
));
2798 return PAGE_SIZE
<< compound_order(page
);
2802 #ifdef CONFIG_SLUB_DEBUG
2804 * Debugging requires use of the padding between object
2805 * and whatever may come after it.
2807 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
2812 * If we have the need to store the freelist pointer
2813 * back there or track user information then we can
2814 * only use the space before that information.
2816 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
2819 * Else we can use all the padding etc for the allocation
2823 EXPORT_SYMBOL(ksize
);
2825 void BCMFASTPATH_HOST
kfree(const void *x
)
2828 void *object
= (void *)x
;
2830 trace_kfree(_RET_IP_
, x
);
2832 if (unlikely(ZERO_OR_NULL_PTR(x
)))
2835 page
= virt_to_head_page(x
);
2836 if (unlikely(!PageSlab(page
))) {
2837 BUG_ON(!PageCompound(page
));
2842 slab_free(page
->slab
, page
, object
, _RET_IP_
);
2844 EXPORT_SYMBOL(kfree
);
2847 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2848 * the remaining slabs by the number of items in use. The slabs with the
2849 * most items in use come first. New allocations will then fill those up
2850 * and thus they can be removed from the partial lists.
2852 * The slabs with the least items are placed last. This results in them
2853 * being allocated from last increasing the chance that the last objects
2854 * are freed in them.
2856 int kmem_cache_shrink(struct kmem_cache
*s
)
2860 struct kmem_cache_node
*n
;
2863 int objects
= oo_objects(s
->max
);
2864 struct list_head
*slabs_by_inuse
=
2865 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
2866 unsigned long flags
;
2868 if (!slabs_by_inuse
)
2872 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2873 n
= get_node(s
, node
);
2878 for (i
= 0; i
< objects
; i
++)
2879 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
2881 spin_lock_irqsave(&n
->list_lock
, flags
);
2884 * Build lists indexed by the items in use in each slab.
2886 * Note that concurrent frees may occur while we hold the
2887 * list_lock. page->inuse here is the upper limit.
2889 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
2890 if (!page
->inuse
&& slab_trylock(page
)) {
2892 * Must hold slab lock here because slab_free
2893 * may have freed the last object and be
2894 * waiting to release the slab.
2896 list_del(&page
->lru
);
2899 discard_slab(s
, page
);
2901 list_move(&page
->lru
,
2902 slabs_by_inuse
+ page
->inuse
);
2907 * Rebuild the partial list with the slabs filled up most
2908 * first and the least used slabs at the end.
2910 for (i
= objects
- 1; i
>= 0; i
--)
2911 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
2913 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2916 kfree(slabs_by_inuse
);
2919 EXPORT_SYMBOL(kmem_cache_shrink
);
2921 #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2922 static int slab_mem_going_offline_callback(void *arg
)
2924 struct kmem_cache
*s
;
2926 down_read(&slub_lock
);
2927 list_for_each_entry(s
, &slab_caches
, list
)
2928 kmem_cache_shrink(s
);
2929 up_read(&slub_lock
);
2934 static void slab_mem_offline_callback(void *arg
)
2936 struct kmem_cache_node
*n
;
2937 struct kmem_cache
*s
;
2938 struct memory_notify
*marg
= arg
;
2941 offline_node
= marg
->status_change_nid
;
2944 * If the node still has available memory. we need kmem_cache_node
2947 if (offline_node
< 0)
2950 down_read(&slub_lock
);
2951 list_for_each_entry(s
, &slab_caches
, list
) {
2952 n
= get_node(s
, offline_node
);
2955 * if n->nr_slabs > 0, slabs still exist on the node
2956 * that is going down. We were unable to free them,
2957 * and offline_pages() function shouldn't call this
2958 * callback. So, we must fail.
2960 BUG_ON(slabs_node(s
, offline_node
));
2962 s
->node
[offline_node
] = NULL
;
2963 kmem_cache_free(kmalloc_caches
, n
);
2966 up_read(&slub_lock
);
2969 static int slab_mem_going_online_callback(void *arg
)
2971 struct kmem_cache_node
*n
;
2972 struct kmem_cache
*s
;
2973 struct memory_notify
*marg
= arg
;
2974 int nid
= marg
->status_change_nid
;
2978 * If the node's memory is already available, then kmem_cache_node is
2979 * already created. Nothing to do.
2985 * We are bringing a node online. No memory is available yet. We must
2986 * allocate a kmem_cache_node structure in order to bring the node
2989 down_read(&slub_lock
);
2990 list_for_each_entry(s
, &slab_caches
, list
) {
2991 n
= kmem_cache_alloc(kmalloc_caches
, GFP_KERNEL
);
2996 init_kmem_cache_node(n
, s
);
3000 up_read(&slub_lock
);
3004 static int slab_memory_callback(struct notifier_block
*self
,
3005 unsigned long action
, void *arg
)
3010 case MEM_GOING_ONLINE
:
3011 ret
= slab_mem_going_online_callback(arg
);
3013 case MEM_GOING_OFFLINE
:
3014 ret
= slab_mem_going_offline_callback(arg
);
3017 case MEM_CANCEL_ONLINE
:
3018 slab_mem_offline_callback(arg
);
3021 case MEM_CANCEL_OFFLINE
:
3025 ret
= notifier_from_errno(ret
);
3031 #endif /* CONFIG_MEMORY_HOTPLUG */
3033 /********************************************************************
3034 * Basic setup of slabs
3035 *******************************************************************/
3037 void __init
kmem_cache_init(void)
3044 * Must first have the slab cache available for the allocations of the
3045 * struct kmem_cache_node's. There is special bootstrap code in
3046 * kmem_cache_open for slab_state == DOWN.
3048 create_kmalloc_cache(&kmalloc_caches
[0], "kmem_cache_node",
3049 sizeof(struct kmem_cache_node
), GFP_NOWAIT
);
3050 kmalloc_caches
[0].refcount
= -1;
3053 hotplug_memory_notifier(slab_memory_callback
, SLAB_CALLBACK_PRI
);
3056 /* Able to allocate the per node structures */
3057 slab_state
= PARTIAL
;
3059 /* Caches that are not of the two-to-the-power-of size */
3060 if (KMALLOC_MIN_SIZE
<= 32) {
3061 create_kmalloc_cache(&kmalloc_caches
[1],
3062 "kmalloc-96", 96, GFP_NOWAIT
);
3065 if (KMALLOC_MIN_SIZE
<= 64) {
3066 create_kmalloc_cache(&kmalloc_caches
[2],
3067 "kmalloc-192", 192, GFP_NOWAIT
);
3071 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3072 create_kmalloc_cache(&kmalloc_caches
[i
],
3073 "kmalloc", 1 << i
, GFP_NOWAIT
);
3079 * Patch up the size_index table if we have strange large alignment
3080 * requirements for the kmalloc array. This is only the case for
3081 * MIPS it seems. The standard arches will not generate any code here.
3083 * Largest permitted alignment is 256 bytes due to the way we
3084 * handle the index determination for the smaller caches.
3086 * Make sure that nothing crazy happens if someone starts tinkering
3087 * around with ARCH_KMALLOC_MINALIGN
3089 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 256 ||
3090 (KMALLOC_MIN_SIZE
& (KMALLOC_MIN_SIZE
- 1)));
3092 for (i
= 8; i
< KMALLOC_MIN_SIZE
; i
+= 8) {
3093 int elem
= size_index_elem(i
);
3094 if (elem
>= ARRAY_SIZE(size_index
))
3096 size_index
[elem
] = KMALLOC_SHIFT_LOW
;
3099 if (KMALLOC_MIN_SIZE
== 64) {
3101 * The 96 byte size cache is not used if the alignment
3104 for (i
= 64 + 8; i
<= 96; i
+= 8)
3105 size_index
[size_index_elem(i
)] = 7;
3106 } else if (KMALLOC_MIN_SIZE
== 128) {
3108 * The 192 byte sized cache is not used if the alignment
3109 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3112 for (i
= 128 + 8; i
<= 192; i
+= 8)
3113 size_index
[size_index_elem(i
)] = 8;
3118 /* Provide the correct kmalloc names now that the caches are up */
3119 for (i
= KMALLOC_SHIFT_LOW
; i
< SLUB_PAGE_SHIFT
; i
++) {
3120 char *s
= kasprintf(GFP_NOWAIT
, "kmalloc-%d", 1 << i
);
3123 kmalloc_caches
[i
].name
= s
;
3127 register_cpu_notifier(&slab_notifier
);
3130 kmem_size
= offsetof(struct kmem_cache
, node
) +
3131 nr_node_ids
* sizeof(struct kmem_cache_node
*);
3133 kmem_size
= sizeof(struct kmem_cache
);
3137 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3138 " CPUs=%d, Nodes=%d\n",
3139 caches
, cache_line_size(),
3140 slub_min_order
, slub_max_order
, slub_min_objects
,
3141 nr_cpu_ids
, nr_node_ids
);
3144 void __init
kmem_cache_init_late(void)
3149 * Find a mergeable slab cache
3151 static int slab_unmergeable(struct kmem_cache
*s
)
3153 if (slub_nomerge
|| (s
->flags
& SLUB_NEVER_MERGE
))
3160 * We may have set a slab to be unmergeable during bootstrap.
3162 if (s
->refcount
< 0)
3168 static struct kmem_cache
*find_mergeable(size_t size
,
3169 size_t align
, unsigned long flags
, const char *name
,
3170 void (*ctor
)(void *))
3172 struct kmem_cache
*s
;
3174 if (slub_nomerge
|| (flags
& SLUB_NEVER_MERGE
))
3180 size
= ALIGN(size
, sizeof(void *));
3181 align
= calculate_alignment(flags
, align
, size
);
3182 size
= ALIGN(size
, align
);
3183 flags
= kmem_cache_flags(size
, flags
, name
, NULL
);
3185 list_for_each_entry(s
, &slab_caches
, list
) {
3186 if (slab_unmergeable(s
))
3192 if ((flags
& SLUB_MERGE_SAME
) != (s
->flags
& SLUB_MERGE_SAME
))
3195 * Check if alignment is compatible.
3196 * Courtesy of Adrian Drzewiecki
3198 if ((s
->size
& ~(align
- 1)) != s
->size
)
3201 if (s
->size
- size
>= sizeof(void *))
3209 struct kmem_cache
*kmem_cache_create(const char *name
, size_t size
,
3210 size_t align
, unsigned long flags
, void (*ctor
)(void *))
3212 struct kmem_cache
*s
;
3217 down_write(&slub_lock
);
3218 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3222 * Adjust the object sizes so that we clear
3223 * the complete object on kzalloc.
3225 s
->objsize
= max(s
->objsize
, (int)size
);
3226 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3228 if (sysfs_slab_alias(s
, name
)) {
3232 up_write(&slub_lock
);
3236 s
= kmalloc(kmem_size
, GFP_KERNEL
);
3238 if (kmem_cache_open(s
, GFP_KERNEL
, name
,
3239 size
, align
, flags
, ctor
)) {
3240 list_add(&s
->list
, &slab_caches
);
3241 if (sysfs_slab_add(s
)) {
3246 up_write(&slub_lock
);
3251 up_write(&slub_lock
);
3254 if (flags
& SLAB_PANIC
)
3255 panic("Cannot create slabcache %s\n", name
);
3260 EXPORT_SYMBOL(kmem_cache_create
);
3264 * Use the cpu notifier to insure that the cpu slabs are flushed when
3267 static int __cpuinit
slab_cpuup_callback(struct notifier_block
*nfb
,
3268 unsigned long action
, void *hcpu
)
3270 long cpu
= (long)hcpu
;
3271 struct kmem_cache
*s
;
3272 unsigned long flags
;
3275 case CPU_UP_CANCELED
:
3276 case CPU_UP_CANCELED_FROZEN
:
3278 case CPU_DEAD_FROZEN
:
3279 down_read(&slub_lock
);
3280 list_for_each_entry(s
, &slab_caches
, list
) {
3281 local_irq_save(flags
);
3282 __flush_cpu_slab(s
, cpu
);
3283 local_irq_restore(flags
);
3285 up_read(&slub_lock
);
3293 static struct notifier_block __cpuinitdata slab_notifier
= {
3294 .notifier_call
= slab_cpuup_callback
3299 void * BCMFASTPATH_HOST
__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3301 struct kmem_cache
*s
;
3304 if (unlikely(size
> SLUB_MAX_SIZE
))
3305 return kmalloc_large(size
, gfpflags
);
3307 s
= get_slab(size
, gfpflags
);
3309 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3312 ret
= slab_alloc(s
, gfpflags
, NUMA_NO_NODE
, caller
);
3314 /* Honor the call site pointer we recieved. */
3315 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3320 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3321 int node
, unsigned long caller
)
3323 struct kmem_cache
*s
;
3326 if (unlikely(size
> SLUB_MAX_SIZE
)) {
3327 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3329 trace_kmalloc_node(caller
, ret
,
3330 size
, PAGE_SIZE
<< get_order(size
),
3336 s
= get_slab(size
, gfpflags
);
3338 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3341 ret
= slab_alloc(s
, gfpflags
, node
, caller
);
3343 /* Honor the call site pointer we recieved. */
3344 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3349 #ifdef CONFIG_SLUB_DEBUG
3350 static int count_inuse(struct page
*page
)
3355 static int count_total(struct page
*page
)
3357 return page
->objects
;
3360 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3364 void *addr
= page_address(page
);
3366 if (!check_slab(s
, page
) ||
3367 !on_freelist(s
, page
, NULL
))
3370 /* Now we know that a valid freelist exists */
3371 bitmap_zero(map
, page
->objects
);
3373 for_each_free_object(p
, s
, page
->freelist
) {
3374 set_bit(slab_index(p
, s
, addr
), map
);
3375 if (!check_object(s
, page
, p
, 0))
3379 for_each_object(p
, s
, addr
, page
->objects
)
3380 if (!test_bit(slab_index(p
, s
, addr
), map
))
3381 if (!check_object(s
, page
, p
, 1))
3386 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3389 if (slab_trylock(page
)) {
3390 validate_slab(s
, page
, map
);
3393 printk(KERN_INFO
"SLUB %s: Skipped busy slab 0x%p\n",
3397 static int validate_slab_node(struct kmem_cache
*s
,
3398 struct kmem_cache_node
*n
, unsigned long *map
)
3400 unsigned long count
= 0;
3402 unsigned long flags
;
3404 spin_lock_irqsave(&n
->list_lock
, flags
);
3406 list_for_each_entry(page
, &n
->partial
, lru
) {
3407 validate_slab_slab(s
, page
, map
);
3410 if (count
!= n
->nr_partial
)
3411 printk(KERN_ERR
"SLUB %s: %ld partial slabs counted but "
3412 "counter=%ld\n", s
->name
, count
, n
->nr_partial
);
3414 if (!(s
->flags
& SLAB_STORE_USER
))
3417 list_for_each_entry(page
, &n
->full
, lru
) {
3418 validate_slab_slab(s
, page
, map
);
3421 if (count
!= atomic_long_read(&n
->nr_slabs
))
3422 printk(KERN_ERR
"SLUB: %s %ld slabs counted but "
3423 "counter=%ld\n", s
->name
, count
,
3424 atomic_long_read(&n
->nr_slabs
));
3427 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3431 static long validate_slab_cache(struct kmem_cache
*s
)
3434 unsigned long count
= 0;
3435 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3436 sizeof(unsigned long), GFP_KERNEL
);
3442 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3443 struct kmem_cache_node
*n
= get_node(s
, node
);
3445 count
+= validate_slab_node(s
, n
, map
);
3451 #ifdef SLUB_RESILIENCY_TEST
3452 static void resiliency_test(void)
3456 printk(KERN_ERR
"SLUB resiliency testing\n");
3457 printk(KERN_ERR
"-----------------------\n");
3458 printk(KERN_ERR
"A. Corruption after allocation\n");
3460 p
= kzalloc(16, GFP_KERNEL
);
3462 printk(KERN_ERR
"\n1. kmalloc-16: Clobber Redzone/next pointer"
3463 " 0x12->0x%p\n\n", p
+ 16);
3465 validate_slab_cache(kmalloc_caches
+ 4);
3467 /* Hmmm... The next two are dangerous */
3468 p
= kzalloc(32, GFP_KERNEL
);
3469 p
[32 + sizeof(void *)] = 0x34;
3470 printk(KERN_ERR
"\n2. kmalloc-32: Clobber next pointer/next slab"
3471 " 0x34 -> -0x%p\n", p
);
3473 "If allocated object is overwritten then not detectable\n\n");
3475 validate_slab_cache(kmalloc_caches
+ 5);
3476 p
= kzalloc(64, GFP_KERNEL
);
3477 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
3479 printk(KERN_ERR
"\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3482 "If allocated object is overwritten then not detectable\n\n");
3483 validate_slab_cache(kmalloc_caches
+ 6);
3485 printk(KERN_ERR
"\nB. Corruption after free\n");
3486 p
= kzalloc(128, GFP_KERNEL
);
3489 printk(KERN_ERR
"1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
3490 validate_slab_cache(kmalloc_caches
+ 7);
3492 p
= kzalloc(256, GFP_KERNEL
);
3495 printk(KERN_ERR
"\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3497 validate_slab_cache(kmalloc_caches
+ 8);
3499 p
= kzalloc(512, GFP_KERNEL
);
3502 printk(KERN_ERR
"\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
3503 validate_slab_cache(kmalloc_caches
+ 9);
3506 static void resiliency_test(void) {};
3510 * Generate lists of code addresses where slabcache objects are allocated
3515 unsigned long count
;
3522 DECLARE_BITMAP(cpus
, NR_CPUS
);
3528 unsigned long count
;
3529 struct location
*loc
;
3532 static void free_loc_track(struct loc_track
*t
)
3535 free_pages((unsigned long)t
->loc
,
3536 get_order(sizeof(struct location
) * t
->max
));
3539 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3544 order
= get_order(sizeof(struct location
) * max
);
3546 l
= (void *)__get_free_pages(flags
, order
);
3551 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3559 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3560 const struct track
*track
)
3562 long start
, end
, pos
;
3564 unsigned long caddr
;
3565 unsigned long age
= jiffies
- track
->when
;
3571 pos
= start
+ (end
- start
+ 1) / 2;
3574 * There is nothing at "end". If we end up there
3575 * we need to add something to before end.
3580 caddr
= t
->loc
[pos
].addr
;
3581 if (track
->addr
== caddr
) {
3587 if (age
< l
->min_time
)
3589 if (age
> l
->max_time
)
3592 if (track
->pid
< l
->min_pid
)
3593 l
->min_pid
= track
->pid
;
3594 if (track
->pid
> l
->max_pid
)
3595 l
->max_pid
= track
->pid
;
3597 cpumask_set_cpu(track
->cpu
,
3598 to_cpumask(l
->cpus
));
3600 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3604 if (track
->addr
< caddr
)
3611 * Not found. Insert new tracking element.
3613 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3619 (t
->count
- pos
) * sizeof(struct location
));
3622 l
->addr
= track
->addr
;
3626 l
->min_pid
= track
->pid
;
3627 l
->max_pid
= track
->pid
;
3628 cpumask_clear(to_cpumask(l
->cpus
));
3629 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
3630 nodes_clear(l
->nodes
);
3631 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3635 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
3636 struct page
*page
, enum track_item alloc
,
3639 void *addr
= page_address(page
);
3642 bitmap_zero(map
, page
->objects
);
3643 for_each_free_object(p
, s
, page
->freelist
)
3644 set_bit(slab_index(p
, s
, addr
), map
);
3646 for_each_object(p
, s
, addr
, page
->objects
)
3647 if (!test_bit(slab_index(p
, s
, addr
), map
))
3648 add_location(t
, s
, get_track(s
, p
, alloc
));
3651 static int list_locations(struct kmem_cache
*s
, char *buf
,
3652 enum track_item alloc
)
3656 struct loc_track t
= { 0, 0, NULL
};
3658 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3659 sizeof(unsigned long), GFP_KERNEL
);
3661 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
3664 return sprintf(buf
, "Out of memory\n");
3666 /* Push back cpu slabs */
3669 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3670 struct kmem_cache_node
*n
= get_node(s
, node
);
3671 unsigned long flags
;
3674 if (!atomic_long_read(&n
->nr_slabs
))
3677 spin_lock_irqsave(&n
->list_lock
, flags
);
3678 list_for_each_entry(page
, &n
->partial
, lru
)
3679 process_slab(&t
, s
, page
, alloc
, map
);
3680 list_for_each_entry(page
, &n
->full
, lru
)
3681 process_slab(&t
, s
, page
, alloc
, map
);
3682 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3685 for (i
= 0; i
< t
.count
; i
++) {
3686 struct location
*l
= &t
.loc
[i
];
3688 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
3690 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
3693 len
+= sprint_symbol(buf
+ len
, (unsigned long)l
->addr
);
3695 len
+= sprintf(buf
+ len
, "<not-available>");
3697 if (l
->sum_time
!= l
->min_time
) {
3698 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
3700 (long)div_u64(l
->sum_time
, l
->count
),
3703 len
+= sprintf(buf
+ len
, " age=%ld",
3706 if (l
->min_pid
!= l
->max_pid
)
3707 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
3708 l
->min_pid
, l
->max_pid
);
3710 len
+= sprintf(buf
+ len
, " pid=%ld",
3713 if (num_online_cpus() > 1 &&
3714 !cpumask_empty(to_cpumask(l
->cpus
)) &&
3715 len
< PAGE_SIZE
- 60) {
3716 len
+= sprintf(buf
+ len
, " cpus=");
3717 len
+= cpulist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3718 to_cpumask(l
->cpus
));
3721 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
3722 len
< PAGE_SIZE
- 60) {
3723 len
+= sprintf(buf
+ len
, " nodes=");
3724 len
+= nodelist_scnprintf(buf
+ len
, PAGE_SIZE
- len
- 50,
3728 len
+= sprintf(buf
+ len
, "\n");
3734 len
+= sprintf(buf
, "No data\n");
3738 enum slab_stat_type
{
3739 SL_ALL
, /* All slabs */
3740 SL_PARTIAL
, /* Only partially allocated slabs */
3741 SL_CPU
, /* Only slabs used for cpu caches */
3742 SL_OBJECTS
, /* Determine allocated objects not slabs */
3743 SL_TOTAL
/* Determine object capacity not slabs */
3746 #define SO_ALL (1 << SL_ALL)
3747 #define SO_PARTIAL (1 << SL_PARTIAL)
3748 #define SO_CPU (1 << SL_CPU)
3749 #define SO_OBJECTS (1 << SL_OBJECTS)
3750 #define SO_TOTAL (1 << SL_TOTAL)
3752 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
3753 char *buf
, unsigned long flags
)
3755 unsigned long total
= 0;
3758 unsigned long *nodes
;
3759 unsigned long *per_cpu
;
3761 nodes
= kzalloc(2 * sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
3764 per_cpu
= nodes
+ nr_node_ids
;
3766 if (flags
& SO_CPU
) {
3769 for_each_possible_cpu(cpu
) {
3770 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
3772 if (!c
|| c
->node
< 0)
3776 if (flags
& SO_TOTAL
)
3777 x
= c
->page
->objects
;
3778 else if (flags
& SO_OBJECTS
)
3784 nodes
[c
->node
] += x
;
3790 if (flags
& SO_ALL
) {
3791 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3792 struct kmem_cache_node
*n
= get_node(s
, node
);
3794 if (flags
& SO_TOTAL
)
3795 x
= atomic_long_read(&n
->total_objects
);
3796 else if (flags
& SO_OBJECTS
)
3797 x
= atomic_long_read(&n
->total_objects
) -
3798 count_partial(n
, count_free
);
3801 x
= atomic_long_read(&n
->nr_slabs
);
3806 } else if (flags
& SO_PARTIAL
) {
3807 for_each_node_state(node
, N_NORMAL_MEMORY
) {
3808 struct kmem_cache_node
*n
= get_node(s
, node
);
3810 if (flags
& SO_TOTAL
)
3811 x
= count_partial(n
, count_total
);
3812 else if (flags
& SO_OBJECTS
)
3813 x
= count_partial(n
, count_inuse
);
3820 x
= sprintf(buf
, "%lu", total
);
3822 for_each_node_state(node
, N_NORMAL_MEMORY
)
3824 x
+= sprintf(buf
+ x
, " N%d=%lu",
3828 return x
+ sprintf(buf
+ x
, "\n");
3831 static int any_slab_objects(struct kmem_cache
*s
)
3835 for_each_online_node(node
) {
3836 struct kmem_cache_node
*n
= get_node(s
, node
);
3841 if (atomic_long_read(&n
->total_objects
))
3847 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3848 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
3850 struct slab_attribute
{
3851 struct attribute attr
;
3852 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
3853 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
3856 #define SLAB_ATTR_RO(_name) \
3857 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3859 #define SLAB_ATTR(_name) \
3860 static struct slab_attribute _name##_attr = \
3861 __ATTR(_name, 0644, _name##_show, _name##_store)
3863 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
3865 return sprintf(buf
, "%d\n", s
->size
);
3867 SLAB_ATTR_RO(slab_size
);
3869 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
3871 return sprintf(buf
, "%d\n", s
->align
);
3873 SLAB_ATTR_RO(align
);
3875 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
3877 return sprintf(buf
, "%d\n", s
->objsize
);
3879 SLAB_ATTR_RO(object_size
);
3881 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
3883 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
3885 SLAB_ATTR_RO(objs_per_slab
);
3887 static ssize_t
order_store(struct kmem_cache
*s
,
3888 const char *buf
, size_t length
)
3890 unsigned long order
;
3893 err
= strict_strtoul(buf
, 10, &order
);
3897 if (order
> slub_max_order
|| order
< slub_min_order
)
3900 calculate_sizes(s
, order
);
3904 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
3906 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
3910 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
3912 return sprintf(buf
, "%lu\n", s
->min_partial
);
3915 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
3921 err
= strict_strtoul(buf
, 10, &min
);
3925 set_min_partial(s
, min
);
3928 SLAB_ATTR(min_partial
);
3930 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
3933 int n
= sprint_symbol(buf
, (unsigned long)s
->ctor
);
3935 return n
+ sprintf(buf
+ n
, "\n");
3941 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
3943 return sprintf(buf
, "%d\n", s
->refcount
- 1);
3945 SLAB_ATTR_RO(aliases
);
3947 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
3949 return show_slab_objects(s
, buf
, SO_ALL
);
3951 SLAB_ATTR_RO(slabs
);
3953 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
3955 return show_slab_objects(s
, buf
, SO_PARTIAL
);
3957 SLAB_ATTR_RO(partial
);
3959 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
3961 return show_slab_objects(s
, buf
, SO_CPU
);
3963 SLAB_ATTR_RO(cpu_slabs
);
3965 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
3967 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
3969 SLAB_ATTR_RO(objects
);
3971 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
3973 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
3975 SLAB_ATTR_RO(objects_partial
);
3977 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
3979 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
3981 SLAB_ATTR_RO(total_objects
);
3983 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
3985 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
3988 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
3989 const char *buf
, size_t length
)
3991 s
->flags
&= ~SLAB_DEBUG_FREE
;
3993 s
->flags
|= SLAB_DEBUG_FREE
;
3996 SLAB_ATTR(sanity_checks
);
3998 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4000 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4003 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4006 s
->flags
&= ~SLAB_TRACE
;
4008 s
->flags
|= SLAB_TRACE
;
4013 #ifdef CONFIG_FAILSLAB
4014 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4016 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4019 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4022 s
->flags
&= ~SLAB_FAILSLAB
;
4024 s
->flags
|= SLAB_FAILSLAB
;
4027 SLAB_ATTR(failslab
);
4030 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4032 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4035 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4036 const char *buf
, size_t length
)
4038 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4040 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4043 SLAB_ATTR(reclaim_account
);
4045 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4047 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4049 SLAB_ATTR_RO(hwcache_align
);
4051 #ifdef CONFIG_ZONE_DMA
4052 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4054 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4056 SLAB_ATTR_RO(cache_dma
);
4059 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4061 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4063 SLAB_ATTR_RO(destroy_by_rcu
);
4065 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4067 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4070 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4071 const char *buf
, size_t length
)
4073 if (any_slab_objects(s
))
4076 s
->flags
&= ~SLAB_RED_ZONE
;
4078 s
->flags
|= SLAB_RED_ZONE
;
4079 calculate_sizes(s
, -1);
4082 SLAB_ATTR(red_zone
);
4084 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4086 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4089 static ssize_t
poison_store(struct kmem_cache
*s
,
4090 const char *buf
, size_t length
)
4092 if (any_slab_objects(s
))
4095 s
->flags
&= ~SLAB_POISON
;
4097 s
->flags
|= SLAB_POISON
;
4098 calculate_sizes(s
, -1);
4103 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4105 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4108 static ssize_t
store_user_store(struct kmem_cache
*s
,
4109 const char *buf
, size_t length
)
4111 if (any_slab_objects(s
))
4114 s
->flags
&= ~SLAB_STORE_USER
;
4116 s
->flags
|= SLAB_STORE_USER
;
4117 calculate_sizes(s
, -1);
4120 SLAB_ATTR(store_user
);
4122 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4127 static ssize_t
validate_store(struct kmem_cache
*s
,
4128 const char *buf
, size_t length
)
4132 if (buf
[0] == '1') {
4133 ret
= validate_slab_cache(s
);
4139 SLAB_ATTR(validate
);
4141 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4146 static ssize_t
shrink_store(struct kmem_cache
*s
,
4147 const char *buf
, size_t length
)
4149 if (buf
[0] == '1') {
4150 int rc
= kmem_cache_shrink(s
);
4160 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4162 if (!(s
->flags
& SLAB_STORE_USER
))
4164 return list_locations(s
, buf
, TRACK_ALLOC
);
4166 SLAB_ATTR_RO(alloc_calls
);
4168 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4170 if (!(s
->flags
& SLAB_STORE_USER
))
4172 return list_locations(s
, buf
, TRACK_FREE
);
4174 SLAB_ATTR_RO(free_calls
);
4177 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4179 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4182 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4183 const char *buf
, size_t length
)
4185 unsigned long ratio
;
4188 err
= strict_strtoul(buf
, 10, &ratio
);
4193 s
->remote_node_defrag_ratio
= ratio
* 10;
4197 SLAB_ATTR(remote_node_defrag_ratio
);
4200 #ifdef CONFIG_SLUB_STATS
4201 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4203 unsigned long sum
= 0;
4206 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4211 for_each_online_cpu(cpu
) {
4212 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4218 len
= sprintf(buf
, "%lu", sum
);
4221 for_each_online_cpu(cpu
) {
4222 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4223 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4227 return len
+ sprintf(buf
+ len
, "\n");
4230 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4234 for_each_online_cpu(cpu
)
4235 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4238 #define STAT_ATTR(si, text) \
4239 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4241 return show_stat(s, buf, si); \
4243 static ssize_t text##_store(struct kmem_cache *s, \
4244 const char *buf, size_t length) \
4246 if (buf[0] != '0') \
4248 clear_stat(s, si); \
4253 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4254 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4255 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4256 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4257 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4258 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4259 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4260 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4261 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4262 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4263 STAT_ATTR(FREE_SLAB
, free_slab
);
4264 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4265 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4266 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4267 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4268 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4269 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4270 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4273 static struct attribute
*slab_attrs
[] = {
4274 &slab_size_attr
.attr
,
4275 &object_size_attr
.attr
,
4276 &objs_per_slab_attr
.attr
,
4278 &min_partial_attr
.attr
,
4280 &objects_partial_attr
.attr
,
4281 &total_objects_attr
.attr
,
4284 &cpu_slabs_attr
.attr
,
4288 &sanity_checks_attr
.attr
,
4290 &hwcache_align_attr
.attr
,
4291 &reclaim_account_attr
.attr
,
4292 &destroy_by_rcu_attr
.attr
,
4293 &red_zone_attr
.attr
,
4295 &store_user_attr
.attr
,
4296 &validate_attr
.attr
,
4298 &alloc_calls_attr
.attr
,
4299 &free_calls_attr
.attr
,
4300 #ifdef CONFIG_ZONE_DMA
4301 &cache_dma_attr
.attr
,
4304 &remote_node_defrag_ratio_attr
.attr
,
4306 #ifdef CONFIG_SLUB_STATS
4307 &alloc_fastpath_attr
.attr
,
4308 &alloc_slowpath_attr
.attr
,
4309 &free_fastpath_attr
.attr
,
4310 &free_slowpath_attr
.attr
,
4311 &free_frozen_attr
.attr
,
4312 &free_add_partial_attr
.attr
,
4313 &free_remove_partial_attr
.attr
,
4314 &alloc_from_partial_attr
.attr
,
4315 &alloc_slab_attr
.attr
,
4316 &alloc_refill_attr
.attr
,
4317 &free_slab_attr
.attr
,
4318 &cpuslab_flush_attr
.attr
,
4319 &deactivate_full_attr
.attr
,
4320 &deactivate_empty_attr
.attr
,
4321 &deactivate_to_head_attr
.attr
,
4322 &deactivate_to_tail_attr
.attr
,
4323 &deactivate_remote_frees_attr
.attr
,
4324 &order_fallback_attr
.attr
,
4326 #ifdef CONFIG_FAILSLAB
4327 &failslab_attr
.attr
,
4333 static struct attribute_group slab_attr_group
= {
4334 .attrs
= slab_attrs
,
4337 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4338 struct attribute
*attr
,
4341 struct slab_attribute
*attribute
;
4342 struct kmem_cache
*s
;
4345 attribute
= to_slab_attr(attr
);
4348 if (!attribute
->show
)
4351 err
= attribute
->show(s
, buf
);
4356 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4357 struct attribute
*attr
,
4358 const char *buf
, size_t len
)
4360 struct slab_attribute
*attribute
;
4361 struct kmem_cache
*s
;
4364 attribute
= to_slab_attr(attr
);
4367 if (!attribute
->store
)
4370 err
= attribute
->store(s
, buf
, len
);
4375 static void kmem_cache_release(struct kobject
*kobj
)
4377 struct kmem_cache
*s
= to_slab(kobj
);
4382 static const struct sysfs_ops slab_sysfs_ops
= {
4383 .show
= slab_attr_show
,
4384 .store
= slab_attr_store
,
4387 static struct kobj_type slab_ktype
= {
4388 .sysfs_ops
= &slab_sysfs_ops
,
4389 .release
= kmem_cache_release
4392 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
4394 struct kobj_type
*ktype
= get_ktype(kobj
);
4396 if (ktype
== &slab_ktype
)
4401 static const struct kset_uevent_ops slab_uevent_ops
= {
4402 .filter
= uevent_filter
,
4405 static struct kset
*slab_kset
;
4407 #define ID_STR_LENGTH 64
4409 /* Create a unique string id for a slab cache:
4411 * Format :[flags-]size
4413 static char *create_unique_id(struct kmem_cache
*s
)
4415 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
4422 * First flags affecting slabcache operations. We will only
4423 * get here for aliasable slabs so we do not need to support
4424 * too many flags. The flags here must cover all flags that
4425 * are matched during merging to guarantee that the id is
4428 if (s
->flags
& SLAB_CACHE_DMA
)
4430 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
4432 if (s
->flags
& SLAB_DEBUG_FREE
)
4434 if (!(s
->flags
& SLAB_NOTRACK
))
4438 p
+= sprintf(p
, "%07d", s
->size
);
4439 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
4443 static int sysfs_slab_add(struct kmem_cache
*s
)
4449 if (slab_state
< SYSFS
)
4450 /* Defer until later */
4453 unmergeable
= slab_unmergeable(s
);
4456 * Slabcache can never be merged so we can use the name proper.
4457 * This is typically the case for debug situations. In that
4458 * case we can catch duplicate names easily.
4460 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
4464 * Create a unique name for the slab as a target
4467 name
= create_unique_id(s
);
4470 s
->kobj
.kset
= slab_kset
;
4471 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, name
);
4473 kobject_put(&s
->kobj
);
4477 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
4479 kobject_del(&s
->kobj
);
4480 kobject_put(&s
->kobj
);
4483 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
4485 /* Setup first alias */
4486 sysfs_slab_alias(s
, s
->name
);
4492 static void sysfs_slab_remove(struct kmem_cache
*s
)
4494 if (slab_state
< SYSFS
)
4496 * Sysfs has not been setup yet so no need to remove the
4501 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
4502 kobject_del(&s
->kobj
);
4503 kobject_put(&s
->kobj
);
4507 * Need to buffer aliases during bootup until sysfs becomes
4508 * available lest we lose that information.
4510 struct saved_alias
{
4511 struct kmem_cache
*s
;
4513 struct saved_alias
*next
;
4516 static struct saved_alias
*alias_list
;
4518 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
4520 struct saved_alias
*al
;
4522 if (slab_state
== SYSFS
) {
4524 * If we have a leftover link then remove it.
4526 sysfs_remove_link(&slab_kset
->kobj
, name
);
4527 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
4530 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
4536 al
->next
= alias_list
;
4541 static int __init
slab_sysfs_init(void)
4543 struct kmem_cache
*s
;
4546 down_write(&slub_lock
);
4548 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
4550 up_write(&slub_lock
);
4551 printk(KERN_ERR
"Cannot register slab subsystem.\n");
4557 list_for_each_entry(s
, &slab_caches
, list
) {
4558 err
= sysfs_slab_add(s
);
4560 printk(KERN_ERR
"SLUB: Unable to add boot slab %s"
4561 " to sysfs\n", s
->name
);
4564 while (alias_list
) {
4565 struct saved_alias
*al
= alias_list
;
4567 alias_list
= alias_list
->next
;
4568 err
= sysfs_slab_alias(al
->s
, al
->name
);
4570 printk(KERN_ERR
"SLUB: Unable to add boot slab alias"
4571 " %s to sysfs\n", s
->name
);
4575 up_write(&slub_lock
);
4580 __initcall(slab_sysfs_init
);
4584 * The /proc/slabinfo ABI
4586 #ifdef CONFIG_SLABINFO
4587 static void print_slabinfo_header(struct seq_file
*m
)
4589 seq_puts(m
, "slabinfo - version: 2.1\n");
4590 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
4591 "<objperslab> <pagesperslab>");
4592 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
4593 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4597 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
4601 down_read(&slub_lock
);
4603 print_slabinfo_header(m
);
4605 return seq_list_start(&slab_caches
, *pos
);
4608 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
4610 return seq_list_next(p
, &slab_caches
, pos
);
4613 static void s_stop(struct seq_file
*m
, void *p
)
4615 up_read(&slub_lock
);
4618 static int s_show(struct seq_file
*m
, void *p
)
4620 unsigned long nr_partials
= 0;
4621 unsigned long nr_slabs
= 0;
4622 unsigned long nr_inuse
= 0;
4623 unsigned long nr_objs
= 0;
4624 unsigned long nr_free
= 0;
4625 struct kmem_cache
*s
;
4628 s
= list_entry(p
, struct kmem_cache
, list
);
4630 for_each_online_node(node
) {
4631 struct kmem_cache_node
*n
= get_node(s
, node
);
4636 nr_partials
+= n
->nr_partial
;
4637 nr_slabs
+= atomic_long_read(&n
->nr_slabs
);
4638 nr_objs
+= atomic_long_read(&n
->total_objects
);
4639 nr_free
+= count_partial(n
, count_free
);
4642 nr_inuse
= nr_objs
- nr_free
;
4644 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d", s
->name
, nr_inuse
,
4645 nr_objs
, s
->size
, oo_objects(s
->oo
),
4646 (1 << oo_order(s
->oo
)));
4647 seq_printf(m
, " : tunables %4u %4u %4u", 0, 0, 0);
4648 seq_printf(m
, " : slabdata %6lu %6lu %6lu", nr_slabs
, nr_slabs
,
4654 static const struct seq_operations slabinfo_op
= {
4661 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
4663 return seq_open(file
, &slabinfo_op
);
4666 static const struct file_operations proc_slabinfo_operations
= {
4667 .open
= slabinfo_open
,
4669 .llseek
= seq_lseek
,
4670 .release
= seq_release
,
4673 static int __init
slab_proc_init(void)
4675 proc_create("slabinfo", S_IRUGO
, NULL
, &proc_slabinfo_operations
);
4678 module_init(slab_proc_init
);
4679 #endif /* CONFIG_SLABINFO */