2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemcheck.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/mempolicy.h>
24 #include <linux/ctype.h>
25 #include <linux/debugobjects.h>
26 #include <linux/kallsyms.h>
27 #include <linux/memory.h>
28 #include <linux/math64.h>
29 #include <linux/fault-inject.h>
31 #include <trace/events/kmem.h>
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache *s)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
135 * Mininum number of partial slabs. These will be left on the partial
136 * lists even if they are empty. kmem_cache_shrink may reclaim them.
138 #define MIN_PARTIAL 5
141 * Maximum number of desirable partial slabs.
142 * The existence of more partial slabs makes kmem_cache_shrink
143 * sort the partial list by the number of objects in the.
145 #define MAX_PARTIAL 10
147 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
148 SLAB_POISON | SLAB_STORE_USER)
151 * Debugging flags that require metadata to be stored in the slab. These get
152 * disabled when slub_debug=O is used and a cache's min order increases with
155 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
158 * Set of flags that will prevent slab merging
160 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
161 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
164 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
165 SLAB_CACHE_DMA | SLAB_NOTRACK)
168 #define OO_MASK ((1 << OO_SHIFT) - 1)
169 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
171 /* Internal SLUB flags */
172 #define __OBJECT_POISON 0x80000000UL /* Poison object */
174 static int kmem_size = sizeof(struct kmem_cache);
177 static struct notifier_block slab_notifier;
181 DOWN, /* No slab functionality available */
182 PARTIAL, /* Kmem_cache_node works */
183 UP, /* Everything works but does not show up in sysfs */
187 /* A list of all slab caches on the system */
188 static DECLARE_RWSEM(slub_lock);
189 static LIST_HEAD(slab_caches);
192 * Tracking user of a slab.
195 unsigned long addr; /* Called from address */
196 int cpu; /* Was running on cpu */
197 int pid; /* Pid context */
198 unsigned long when; /* When did the operation occur */
201 enum track_item { TRACK_ALLOC, TRACK_FREE };
204 static int sysfs_slab_add(struct kmem_cache *);
205 static int sysfs_slab_alias(struct kmem_cache *, const char *);
206 static void sysfs_slab_remove(struct kmem_cache *);
209 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
210 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
212 static inline void sysfs_slab_remove(struct kmem_cache *s)
220 static inline void stat(struct kmem_cache *s, enum stat_item si)
222 #ifdef CONFIG_SLUB_STATS
223 __this_cpu_inc(s->cpu_slab->stat[si]);
227 /********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
231 int slab_is_available(void)
233 return slab_state >= UP;
236 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
238 return s->node[node];
241 /* Verify that a pointer has an address that is valid within a slab page */
242 static inline int check_valid_pointer(struct kmem_cache *s,
243 struct page *page, const void *object)
250 base = page_address(page);
251 if (object < base || object >= base + page->objects * s->size ||
252 (object - base) % s->size) {
259 static inline void *get_freepointer(struct kmem_cache *s, void *object)
261 return *(void **)(object + s->offset);
264 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
266 *(void **)(object + s->offset) = fp;
269 /* Loop over all objects in a slab */
270 #define for_each_object(__p, __s, __addr, __objects) \
271 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
275 #define for_each_free_object(__p, __s, __free) \
276 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
278 /* Determine object index from a given position */
279 static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
281 return (p - addr) / s->size;
284 static inline size_t slab_ksize(const struct kmem_cache *s)
286 #ifdef CONFIG_SLUB_DEBUG
288 * Debugging requires use of the padding between object
289 * and whatever may come after it.
291 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
296 * If we have the need to store the freelist pointer
297 * back there or track user information then we can
298 * only use the space before that information.
300 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
303 * Else we can use all the padding etc for the allocation
308 static inline int order_objects(int order, unsigned long size, int reserved)
310 return ((PAGE_SIZE << order) - reserved) / size;
313 static inline struct kmem_cache_order_objects oo_make(int order,
314 unsigned long size, int reserved)
316 struct kmem_cache_order_objects x = {
317 (order << OO_SHIFT) + order_objects(order, size, reserved)
323 static inline int oo_order(struct kmem_cache_order_objects x)
325 return x.x >> OO_SHIFT;
328 static inline int oo_objects(struct kmem_cache_order_objects x)
330 return x.x & OO_MASK;
333 #ifdef CONFIG_SLUB_DEBUG
337 #ifdef CONFIG_SLUB_DEBUG_ON
338 static int slub_debug = DEBUG_DEFAULT_FLAGS;
340 static int slub_debug;
343 static char *slub_debug_slabs;
344 static int disable_higher_order_debug;
349 static void print_section(char *text, u8 *addr, unsigned int length)
357 for (i = 0; i < length; i++) {
359 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
362 printk(KERN_CONT " %02x", addr[i]);
364 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
366 printk(KERN_CONT " %s\n", ascii);
373 printk(KERN_CONT " ");
377 printk(KERN_CONT " %s\n", ascii);
381 static struct track *get_track(struct kmem_cache *s, void *object,
382 enum track_item alloc)
387 p = object + s->offset + sizeof(void *);
389 p = object + s->inuse;
394 static void set_track(struct kmem_cache *s, void *object,
395 enum track_item alloc, unsigned long addr)
397 struct track *p = get_track(s, object, alloc);
401 p->cpu = smp_processor_id();
402 p->pid = current->pid;
405 memset(p, 0, sizeof(struct track));
408 static void init_tracking(struct kmem_cache *s, void *object)
410 if (!(s->flags & SLAB_STORE_USER))
413 set_track(s, object, TRACK_FREE, 0UL);
414 set_track(s, object, TRACK_ALLOC, 0UL);
417 static void print_track(const char *s, struct track *t)
422 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
423 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
426 static void print_tracking(struct kmem_cache *s, void *object)
428 if (!(s->flags & SLAB_STORE_USER))
431 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
432 print_track("Freed", get_track(s, object, TRACK_FREE));
435 static void print_page_info(struct page *page)
437 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
438 page, page->objects, page->inuse, page->freelist, page->flags);
442 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
448 vsnprintf(buf, sizeof(buf), fmt, args);
450 printk(KERN_ERR "========================================"
451 "=====================================\n");
452 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
453 printk(KERN_ERR "----------------------------------------"
454 "-------------------------------------\n\n");
457 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
463 vsnprintf(buf, sizeof(buf), fmt, args);
465 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
468 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
470 unsigned int off; /* Offset of last byte */
471 u8 *addr = page_address(page);
473 print_tracking(s, p);
475 print_page_info(page);
477 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
478 p, p - addr, get_freepointer(s, p));
481 print_section("Bytes b4", p - 16, 16);
483 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
485 if (s->flags & SLAB_RED_ZONE)
486 print_section("Redzone", p + s->objsize,
487 s->inuse - s->objsize);
490 off = s->offset + sizeof(void *);
494 if (s->flags & SLAB_STORE_USER)
495 off += 2 * sizeof(struct track);
498 /* Beginning of the filler is the free pointer */
499 print_section("Padding", p + off, s->size - off);
504 static void object_err(struct kmem_cache *s, struct page *page,
505 u8 *object, char *reason)
507 slab_bug(s, "%s", reason);
508 print_trailer(s, page, object);
511 static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
517 vsnprintf(buf, sizeof(buf), fmt, args);
519 slab_bug(s, "%s", buf);
520 print_page_info(page);
524 static void init_object(struct kmem_cache *s, void *object, u8 val)
528 if (s->flags & __OBJECT_POISON) {
529 memset(p, POISON_FREE, s->objsize - 1);
530 p[s->objsize - 1] = POISON_END;
533 if (s->flags & SLAB_RED_ZONE)
534 memset(p + s->objsize, val, s->inuse - s->objsize);
537 static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
540 if (*start != (u8)value)
548 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
549 void *from, void *to)
551 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
552 memset(from, data, to - from);
555 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
556 u8 *object, char *what,
557 u8 *start, unsigned int value, unsigned int bytes)
562 fault = check_bytes(start, value, bytes);
567 while (end > fault && end[-1] == value)
570 slab_bug(s, "%s overwritten", what);
571 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
572 fault, end - 1, fault[0], value);
573 print_trailer(s, page, object);
575 restore_bytes(s, what, value, fault, end);
583 * Bytes of the object to be managed.
584 * If the freepointer may overlay the object then the free
585 * pointer is the first word of the object.
587 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
590 * object + s->objsize
591 * Padding to reach word boundary. This is also used for Redzoning.
592 * Padding is extended by another word if Redzoning is enabled and
595 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
596 * 0xcc (RED_ACTIVE) for objects in use.
599 * Meta data starts here.
601 * A. Free pointer (if we cannot overwrite object on free)
602 * B. Tracking data for SLAB_STORE_USER
603 * C. Padding to reach required alignment boundary or at mininum
604 * one word if debugging is on to be able to detect writes
605 * before the word boundary.
607 * Padding is done using 0x5a (POISON_INUSE)
610 * Nothing is used beyond s->size.
612 * If slabcaches are merged then the objsize and inuse boundaries are mostly
613 * ignored. And therefore no slab options that rely on these boundaries
614 * may be used with merged slabcaches.
617 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
619 unsigned long off = s->inuse; /* The end of info */
622 /* Freepointer is placed after the object. */
623 off += sizeof(void *);
625 if (s->flags & SLAB_STORE_USER)
626 /* We also have user information there */
627 off += 2 * sizeof(struct track);
632 return check_bytes_and_report(s, page, p, "Object padding",
633 p + off, POISON_INUSE, s->size - off);
636 /* Check the pad bytes at the end of a slab page */
637 static int slab_pad_check(struct kmem_cache *s, struct page *page)
645 if (!(s->flags & SLAB_POISON))
648 start = page_address(page);
649 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
650 end = start + length;
651 remainder = length % s->size;
655 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
658 while (end > fault && end[-1] == POISON_INUSE)
661 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
662 print_section("Padding", end - remainder, remainder);
664 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
668 static int check_object(struct kmem_cache *s, struct page *page,
669 void *object, u8 val)
672 u8 *endobject = object + s->objsize;
674 if (s->flags & SLAB_RED_ZONE) {
675 if (!check_bytes_and_report(s, page, object, "Redzone",
676 endobject, val, s->inuse - s->objsize))
679 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
680 check_bytes_and_report(s, page, p, "Alignment padding",
681 endobject, POISON_INUSE, s->inuse - s->objsize);
685 if (s->flags & SLAB_POISON) {
686 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
687 (!check_bytes_and_report(s, page, p, "Poison", p,
688 POISON_FREE, s->objsize - 1) ||
689 !check_bytes_and_report(s, page, p, "Poison",
690 p + s->objsize - 1, POISON_END, 1)))
693 * check_pad_bytes cleans up on its own.
695 check_pad_bytes(s, page, p);
698 if (!s->offset && val == SLUB_RED_ACTIVE)
700 * Object and freepointer overlap. Cannot check
701 * freepointer while object is allocated.
705 /* Check free pointer validity */
706 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
707 object_err(s, page, p, "Freepointer corrupt");
709 * No choice but to zap it and thus lose the remainder
710 * of the free objects in this slab. May cause
711 * another error because the object count is now wrong.
713 set_freepointer(s, p, NULL);
719 static int check_slab(struct kmem_cache *s, struct page *page)
723 VM_BUG_ON(!irqs_disabled());
725 if (!PageSlab(page)) {
726 slab_err(s, page, "Not a valid slab page");
730 maxobj = order_objects(compound_order(page), s->size, s->reserved);
731 if (page->objects > maxobj) {
732 slab_err(s, page, "objects %u > max %u",
733 s->name, page->objects, maxobj);
736 if (page->inuse > page->objects) {
737 slab_err(s, page, "inuse %u > max %u",
738 s->name, page->inuse, page->objects);
741 /* Slab_pad_check fixes things up after itself */
742 slab_pad_check(s, page);
747 * Determine if a certain object on a page is on the freelist. Must hold the
748 * slab lock to guarantee that the chains are in a consistent state.
750 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
753 void *fp = page->freelist;
755 unsigned long max_objects;
757 while (fp && nr <= page->objects) {
760 if (!check_valid_pointer(s, page, fp)) {
762 object_err(s, page, object,
763 "Freechain corrupt");
764 set_freepointer(s, object, NULL);
767 slab_err(s, page, "Freepointer corrupt");
768 page->freelist = NULL;
769 page->inuse = page->objects;
770 slab_fix(s, "Freelist cleared");
776 fp = get_freepointer(s, object);
780 max_objects = order_objects(compound_order(page), s->size, s->reserved);
781 if (max_objects > MAX_OBJS_PER_PAGE)
782 max_objects = MAX_OBJS_PER_PAGE;
784 if (page->objects != max_objects) {
785 slab_err(s, page, "Wrong number of objects. Found %d but "
786 "should be %d", page->objects, max_objects);
787 page->objects = max_objects;
788 slab_fix(s, "Number of objects adjusted.");
790 if (page->inuse != page->objects - nr) {
791 slab_err(s, page, "Wrong object count. Counter is %d but "
792 "counted were %d", page->inuse, page->objects - nr);
793 page->inuse = page->objects - nr;
794 slab_fix(s, "Object count adjusted.");
796 return search == NULL;
799 static void trace(struct kmem_cache *s, struct page *page, void *object,
802 if (s->flags & SLAB_TRACE) {
803 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
805 alloc ? "alloc" : "free",
810 print_section("Object", (void *)object, s->objsize);
817 * Hooks for other subsystems that check memory allocations. In a typical
818 * production configuration these hooks all should produce no code at all.
820 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
822 flags &= gfp_allowed_mask;
823 lockdep_trace_alloc(flags);
824 might_sleep_if(flags & __GFP_WAIT);
826 return should_failslab(s->objsize, flags, s->flags);
829 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
831 flags &= gfp_allowed_mask;
832 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
833 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
836 static inline void slab_free_hook(struct kmem_cache *s, void *x)
838 kmemleak_free_recursive(x, s->flags);
841 * Trouble is that we may no longer disable interupts in the fast path
842 * So in order to make the debug calls that expect irqs to be
843 * disabled we need to disable interrupts temporarily.
845 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
849 local_irq_save(flags);
850 kmemcheck_slab_free(s, x, s->objsize);
851 debug_check_no_locks_freed(x, s->objsize);
852 if (!(s->flags & SLAB_DEBUG_OBJECTS))
853 debug_check_no_obj_freed(x, s->objsize);
854 local_irq_restore(flags);
860 * Tracking of fully allocated slabs for debugging purposes.
862 static void add_full(struct kmem_cache_node *n, struct page *page)
864 spin_lock(&n->list_lock);
865 list_add(&page->lru, &n->full);
866 spin_unlock(&n->list_lock);
869 static void remove_full(struct kmem_cache *s, struct page *page)
871 struct kmem_cache_node *n;
873 if (!(s->flags & SLAB_STORE_USER))
876 n = get_node(s, page_to_nid(page));
878 spin_lock(&n->list_lock);
879 list_del(&page->lru);
880 spin_unlock(&n->list_lock);
883 /* Tracking of the number of slabs for debugging purposes */
884 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
886 struct kmem_cache_node *n = get_node(s, node);
888 return atomic_long_read(&n->nr_slabs);
891 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
893 return atomic_long_read(&n->nr_slabs);
896 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
898 struct kmem_cache_node *n = get_node(s, node);
901 * May be called early in order to allocate a slab for the
902 * kmem_cache_node structure. Solve the chicken-egg
903 * dilemma by deferring the increment of the count during
904 * bootstrap (see early_kmem_cache_node_alloc).
907 atomic_long_inc(&n->nr_slabs);
908 atomic_long_add(objects, &n->total_objects);
911 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
913 struct kmem_cache_node *n = get_node(s, node);
915 atomic_long_dec(&n->nr_slabs);
916 atomic_long_sub(objects, &n->total_objects);
919 /* Object debug checks for alloc/free paths */
920 static void setup_object_debug(struct kmem_cache *s, struct page *page,
923 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
926 init_object(s, object, SLUB_RED_INACTIVE);
927 init_tracking(s, object);
930 static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
931 void *object, unsigned long addr)
933 if (!check_slab(s, page))
936 if (!on_freelist(s, page, object)) {
937 object_err(s, page, object, "Object already allocated");
941 if (!check_valid_pointer(s, page, object)) {
942 object_err(s, page, object, "Freelist Pointer check fails");
946 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
949 /* Success perform special debug activities for allocs */
950 if (s->flags & SLAB_STORE_USER)
951 set_track(s, object, TRACK_ALLOC, addr);
952 trace(s, page, object, 1);
953 init_object(s, object, SLUB_RED_ACTIVE);
957 if (PageSlab(page)) {
959 * If this is a slab page then lets do the best we can
960 * to avoid issues in the future. Marking all objects
961 * as used avoids touching the remaining objects.
963 slab_fix(s, "Marking all objects used");
964 page->inuse = page->objects;
965 page->freelist = NULL;
970 static noinline int free_debug_processing(struct kmem_cache *s,
971 struct page *page, void *object, unsigned long addr)
973 if (!check_slab(s, page))
976 if (!check_valid_pointer(s, page, object)) {
977 slab_err(s, page, "Invalid object pointer 0x%p", object);
981 if (on_freelist(s, page, object)) {
982 object_err(s, page, object, "Object already free");
986 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
989 if (unlikely(s != page->slab)) {
990 if (!PageSlab(page)) {
991 slab_err(s, page, "Attempt to free object(0x%p) "
992 "outside of slab", object);
993 } else if (!page->slab) {
995 "SLUB <none>: no slab for object 0x%p.\n",
999 object_err(s, page, object,
1000 "page slab pointer corrupt.");
1004 /* Special debug activities for freeing objects */
1005 if (!PageSlubFrozen(page) && !page->freelist)
1006 remove_full(s, page);
1007 if (s->flags & SLAB_STORE_USER)
1008 set_track(s, object, TRACK_FREE, addr);
1009 trace(s, page, object, 0);
1010 init_object(s, object, SLUB_RED_INACTIVE);
1014 slab_fix(s, "Object at 0x%p not freed", object);
1018 static int __init setup_slub_debug(char *str)
1020 slub_debug = DEBUG_DEFAULT_FLAGS;
1021 if (*str++ != '=' || !*str)
1023 * No options specified. Switch on full debugging.
1029 * No options but restriction on slabs. This means full
1030 * debugging for slabs matching a pattern.
1034 if (tolower(*str) == 'o') {
1036 * Avoid enabling debugging on caches if its minimum order
1037 * would increase as a result.
1039 disable_higher_order_debug = 1;
1046 * Switch off all debugging measures.
1051 * Determine which debug features should be switched on
1053 for (; *str && *str != ','; str++) {
1054 switch (tolower(*str)) {
1056 slub_debug |= SLAB_DEBUG_FREE;
1059 slub_debug |= SLAB_RED_ZONE;
1062 slub_debug |= SLAB_POISON;
1065 slub_debug |= SLAB_STORE_USER;
1068 slub_debug |= SLAB_TRACE;
1071 slub_debug |= SLAB_FAILSLAB;
1074 printk(KERN_ERR "slub_debug option '%c' "
1075 "unknown. skipped\n", *str);
1081 slub_debug_slabs = str + 1;
1086 __setup("slub_debug", setup_slub_debug);
1088 static unsigned long kmem_cache_flags(unsigned long objsize,
1089 unsigned long flags, const char *name,
1090 void (*ctor)(void *))
1093 * Enable debugging if selected on the kernel commandline.
1095 if (slub_debug && (!slub_debug_slabs ||
1096 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1097 flags |= slub_debug;
1102 static inline void setup_object_debug(struct kmem_cache *s,
1103 struct page *page, void *object) {}
1105 static inline int alloc_debug_processing(struct kmem_cache *s,
1106 struct page *page, void *object, unsigned long addr) { return 0; }
1108 static inline int free_debug_processing(struct kmem_cache *s,
1109 struct page *page, void *object, unsigned long addr) { return 0; }
1111 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1113 static inline int check_object(struct kmem_cache *s, struct page *page,
1114 void *object, u8 val) { return 1; }
1115 static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
1116 static inline unsigned long kmem_cache_flags(unsigned long objsize,
1117 unsigned long flags, const char *name,
1118 void (*ctor)(void *))
1122 #define slub_debug 0
1124 #define disable_higher_order_debug 0
1126 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1128 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1130 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1132 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1135 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1138 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1141 static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1143 #endif /* CONFIG_SLUB_DEBUG */
1146 * Slab allocation and freeing
1148 static inline struct page *alloc_slab_page(gfp_t flags, int node,
1149 struct kmem_cache_order_objects oo)
1151 int order = oo_order(oo);
1153 flags |= __GFP_NOTRACK;
1155 if (node == NUMA_NO_NODE)
1156 return alloc_pages(flags, order);
1158 return alloc_pages_exact_node(node, flags, order);
1161 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1164 struct kmem_cache_order_objects oo = s->oo;
1167 flags |= s->allocflags;
1170 * Let the initial higher-order allocation fail under memory pressure
1171 * so we fall-back to the minimum order allocation.
1173 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1175 page = alloc_slab_page(alloc_gfp, node, oo);
1176 if (unlikely(!page)) {
1179 * Allocation may have failed due to fragmentation.
1180 * Try a lower order alloc if possible
1182 page = alloc_slab_page(flags, node, oo);
1186 stat(s, ORDER_FALLBACK);
1189 if (kmemcheck_enabled
1190 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
1191 int pages = 1 << oo_order(oo);
1193 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1196 * Objects from caches that have a constructor don't get
1197 * cleared when they're allocated, so we need to do it here.
1200 kmemcheck_mark_uninitialized_pages(page, pages);
1202 kmemcheck_mark_unallocated_pages(page, pages);
1205 page->objects = oo_objects(oo);
1206 mod_zone_page_state(page_zone(page),
1207 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1208 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1214 static void setup_object(struct kmem_cache *s, struct page *page,
1217 setup_object_debug(s, page, object);
1218 if (unlikely(s->ctor))
1222 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1229 BUG_ON(flags & GFP_SLAB_BUG_MASK);
1231 page = allocate_slab(s,
1232 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1236 inc_slabs_node(s, page_to_nid(page), page->objects);
1238 page->flags |= 1 << PG_slab;
1240 start = page_address(page);
1242 if (unlikely(s->flags & SLAB_POISON))
1243 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
1246 for_each_object(p, s, start, page->objects) {
1247 setup_object(s, page, last);
1248 set_freepointer(s, last, p);
1251 setup_object(s, page, last);
1252 set_freepointer(s, last, NULL);
1254 page->freelist = start;
1260 static void __free_slab(struct kmem_cache *s, struct page *page)
1262 int order = compound_order(page);
1263 int pages = 1 << order;
1265 if (kmem_cache_debug(s)) {
1268 slab_pad_check(s, page);
1269 for_each_object(p, s, page_address(page),
1271 check_object(s, page, p, SLUB_RED_INACTIVE);
1274 kmemcheck_free_shadow(page, compound_order(page));
1276 mod_zone_page_state(page_zone(page),
1277 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1278 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1281 __ClearPageSlab(page);
1282 reset_page_mapcount(page);
1283 if (current->reclaim_state)
1284 current->reclaim_state->reclaimed_slab += pages;
1285 __free_pages(page, order);
1288 #define need_reserve_slab_rcu \
1289 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1291 static void rcu_free_slab(struct rcu_head *h)
1295 if (need_reserve_slab_rcu)
1296 page = virt_to_head_page(h);
1298 page = container_of((struct list_head *)h, struct page, lru);
1300 __free_slab(page->slab, page);
1303 static void free_slab(struct kmem_cache *s, struct page *page)
1305 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1306 struct rcu_head *head;
1308 if (need_reserve_slab_rcu) {
1309 int order = compound_order(page);
1310 int offset = (PAGE_SIZE << order) - s->reserved;
1312 VM_BUG_ON(s->reserved != sizeof(*head));
1313 head = page_address(page) + offset;
1316 * RCU free overloads the RCU head over the LRU
1318 head = (void *)&page->lru;
1321 call_rcu(head, rcu_free_slab);
1323 __free_slab(s, page);
1326 static void discard_slab(struct kmem_cache *s, struct page *page)
1328 dec_slabs_node(s, page_to_nid(page), page->objects);
1333 * Per slab locking using the pagelock
1335 static __always_inline void slab_lock(struct page *page)
1337 bit_spin_lock(PG_locked, &page->flags);
1340 static __always_inline void slab_unlock(struct page *page)
1342 __bit_spin_unlock(PG_locked, &page->flags);
1345 static __always_inline int slab_trylock(struct page *page)
1349 rc = bit_spin_trylock(PG_locked, &page->flags);
1354 * Management of partially allocated slabs
1356 static void add_partial(struct kmem_cache_node *n,
1357 struct page *page, int tail)
1359 spin_lock(&n->list_lock);
1362 list_add_tail(&page->lru, &n->partial);
1364 list_add(&page->lru, &n->partial);
1365 spin_unlock(&n->list_lock);
1368 static inline void __remove_partial(struct kmem_cache_node *n,
1371 list_del(&page->lru);
1375 static void remove_partial(struct kmem_cache *s, struct page *page)
1377 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1379 spin_lock(&n->list_lock);
1380 __remove_partial(n, page);
1381 spin_unlock(&n->list_lock);
1385 * Lock slab and remove from the partial list.
1387 * Must hold list_lock.
1389 static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1392 if (slab_trylock(page)) {
1393 __remove_partial(n, page);
1394 __SetPageSlubFrozen(page);
1401 * Try to allocate a partial slab from a specific node.
1403 static struct page *get_partial_node(struct kmem_cache_node *n)
1408 * Racy check. If we mistakenly see no partial slabs then we
1409 * just allocate an empty slab. If we mistakenly try to get a
1410 * partial slab and there is none available then get_partials()
1413 if (!n || !n->nr_partial)
1416 spin_lock(&n->list_lock);
1417 list_for_each_entry(page, &n->partial, lru)
1418 if (lock_and_freeze_slab(n, page))
1422 spin_unlock(&n->list_lock);
1427 * Get a page from somewhere. Search in increasing NUMA distances.
1429 static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1432 struct zonelist *zonelist;
1435 enum zone_type high_zoneidx = gfp_zone(flags);
1439 * The defrag ratio allows a configuration of the tradeoffs between
1440 * inter node defragmentation and node local allocations. A lower
1441 * defrag_ratio increases the tendency to do local allocations
1442 * instead of attempting to obtain partial slabs from other nodes.
1444 * If the defrag_ratio is set to 0 then kmalloc() always
1445 * returns node local objects. If the ratio is higher then kmalloc()
1446 * may return off node objects because partial slabs are obtained
1447 * from other nodes and filled up.
1449 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1450 * defrag_ratio = 1000) then every (well almost) allocation will
1451 * first attempt to defrag slab caches on other nodes. This means
1452 * scanning over all nodes to look for partial slabs which may be
1453 * expensive if we do it every time we are trying to find a slab
1454 * with available objects.
1456 if (!s->remote_node_defrag_ratio ||
1457 get_cycles() % 1024 > s->remote_node_defrag_ratio)
1461 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1462 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1463 struct kmem_cache_node *n;
1465 n = get_node(s, zone_to_nid(zone));
1467 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1468 n->nr_partial > s->min_partial) {
1469 page = get_partial_node(n);
1482 * Get a partial page, lock it and return it.
1484 static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1487 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
1489 page = get_partial_node(get_node(s, searchnode));
1490 if (page || node != -1)
1493 return get_any_partial(s, flags);
1497 * Move a page back to the lists.
1499 * Must be called with the slab lock held.
1501 * On exit the slab lock will have been dropped.
1503 static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
1506 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1508 __ClearPageSlubFrozen(page);
1511 if (page->freelist) {
1512 add_partial(n, page, tail);
1513 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1515 stat(s, DEACTIVATE_FULL);
1516 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
1521 stat(s, DEACTIVATE_EMPTY);
1522 if (n->nr_partial < s->min_partial) {
1524 * Adding an empty slab to the partial slabs in order
1525 * to avoid page allocator overhead. This slab needs
1526 * to come after the other slabs with objects in
1527 * so that the others get filled first. That way the
1528 * size of the partial list stays small.
1530 * kmem_cache_shrink can reclaim any empty slabs from
1533 add_partial(n, page, 1);
1538 discard_slab(s, page);
1543 #ifdef CONFIG_CMPXCHG_LOCAL
1544 #ifdef CONFIG_PREEMPT
1546 * Calculate the next globally unique transaction for disambiguiation
1547 * during cmpxchg. The transactions start with the cpu number and are then
1548 * incremented by CONFIG_NR_CPUS.
1550 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1553 * No preemption supported therefore also no need to check for
1559 static inline unsigned long next_tid(unsigned long tid)
1561 return tid + TID_STEP;
1564 static inline unsigned int tid_to_cpu(unsigned long tid)
1566 return tid % TID_STEP;
1569 static inline unsigned long tid_to_event(unsigned long tid)
1571 return tid / TID_STEP;
1574 static inline unsigned int init_tid(int cpu)
1579 static inline void note_cmpxchg_failure(const char *n,
1580 const struct kmem_cache *s, unsigned long tid)
1582 #ifdef SLUB_DEBUG_CMPXCHG
1583 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1585 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1587 #ifdef CONFIG_PREEMPT
1588 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1589 printk("due to cpu change %d -> %d\n",
1590 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1593 if (tid_to_event(tid) != tid_to_event(actual_tid))
1594 printk("due to cpu running other code. Event %ld->%ld\n",
1595 tid_to_event(tid), tid_to_event(actual_tid));
1597 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1598 actual_tid, tid, next_tid(tid));
1604 void init_kmem_cache_cpus(struct kmem_cache *s)
1606 #if defined(CONFIG_CMPXCHG_LOCAL) && defined(CONFIG_PREEMPT)
1609 for_each_possible_cpu(cpu)
1610 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
1615 * Remove the cpu slab
1617 static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1620 struct page *page = c->page;
1624 stat(s, DEACTIVATE_REMOTE_FREES);
1626 * Merge cpu freelist into slab freelist. Typically we get here
1627 * because both freelists are empty. So this is unlikely
1630 while (unlikely(c->freelist)) {
1633 tail = 0; /* Hot objects. Put the slab first */
1635 /* Retrieve object from cpu_freelist */
1636 object = c->freelist;
1637 c->freelist = get_freepointer(s, c->freelist);
1639 /* And put onto the regular freelist */
1640 set_freepointer(s, object, page->freelist);
1641 page->freelist = object;
1645 #ifdef CONFIG_CMPXCHG_LOCAL
1646 c->tid = next_tid(c->tid);
1648 unfreeze_slab(s, page, tail);
1651 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1653 stat(s, CPUSLAB_FLUSH);
1655 deactivate_slab(s, c);
1661 * Called from IPI handler with interrupts disabled.
1663 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
1665 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
1667 if (likely(c && c->page))
1671 static void flush_cpu_slab(void *d)
1673 struct kmem_cache *s = d;
1675 __flush_cpu_slab(s, smp_processor_id());
1678 static void flush_all(struct kmem_cache *s)
1680 on_each_cpu(flush_cpu_slab, s, 1);
1684 * Check if the objects in a per cpu structure fit numa
1685 * locality expectations.
1687 static inline int node_match(struct kmem_cache_cpu *c, int node)
1690 if (node != NUMA_NO_NODE && c->node != node)
1696 static int count_free(struct page *page)
1698 return page->objects - page->inuse;
1701 static unsigned long count_partial(struct kmem_cache_node *n,
1702 int (*get_count)(struct page *))
1704 unsigned long flags;
1705 unsigned long x = 0;
1708 spin_lock_irqsave(&n->list_lock, flags);
1709 list_for_each_entry(page, &n->partial, lru)
1710 x += get_count(page);
1711 spin_unlock_irqrestore(&n->list_lock, flags);
1715 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1717 #ifdef CONFIG_SLUB_DEBUG
1718 return atomic_long_read(&n->total_objects);
1724 static noinline void
1725 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1730 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1732 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1733 "default order: %d, min order: %d\n", s->name, s->objsize,
1734 s->size, oo_order(s->oo), oo_order(s->min));
1736 if (oo_order(s->min) > get_order(s->objsize))
1737 printk(KERN_WARNING " %s debugging increased min order, use "
1738 "slub_debug=O to disable.\n", s->name);
1740 for_each_online_node(node) {
1741 struct kmem_cache_node *n = get_node(s, node);
1742 unsigned long nr_slabs;
1743 unsigned long nr_objs;
1744 unsigned long nr_free;
1749 nr_free = count_partial(n, count_free);
1750 nr_slabs = node_nr_slabs(n);
1751 nr_objs = node_nr_objs(n);
1754 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1755 node, nr_slabs, nr_objs, nr_free);
1760 * Slow path. The lockless freelist is empty or we need to perform
1763 * Interrupts are disabled.
1765 * Processing is still very fast if new objects have been freed to the
1766 * regular freelist. In that case we simply take over the regular freelist
1767 * as the lockless freelist and zap the regular freelist.
1769 * If that is not working then we fall back to the partial lists. We take the
1770 * first element of the freelist as the object to allocate now and move the
1771 * rest of the freelist to the lockless freelist.
1773 * And if we were unable to get a new slab from the partial slab lists then
1774 * we need to allocate a new slab. This is the slowest path since it involves
1775 * a call to the page allocator and the setup of a new slab.
1777 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1778 unsigned long addr, struct kmem_cache_cpu *c)
1782 #ifdef CONFIG_CMPXCHG_LOCAL
1783 unsigned long flags;
1785 local_irq_save(flags);
1786 #ifdef CONFIG_PREEMPT
1788 * We may have been preempted and rescheduled on a different
1789 * cpu before disabling interrupts. Need to reload cpu area
1792 c = this_cpu_ptr(s->cpu_slab);
1796 /* We handle __GFP_ZERO in the caller */
1797 gfpflags &= ~__GFP_ZERO;
1803 if (unlikely(!node_match(c, node)))
1806 stat(s, ALLOC_REFILL);
1809 object = c->page->freelist;
1810 if (unlikely(!object))
1812 if (kmem_cache_debug(s))
1815 c->freelist = get_freepointer(s, object);
1816 c->page->inuse = c->page->objects;
1817 c->page->freelist = NULL;
1818 c->node = page_to_nid(c->page);
1820 slab_unlock(c->page);
1821 #ifdef CONFIG_CMPXCHG_LOCAL
1822 c->tid = next_tid(c->tid);
1823 local_irq_restore(flags);
1825 stat(s, ALLOC_SLOWPATH);
1829 deactivate_slab(s, c);
1832 new = get_partial(s, gfpflags, node);
1835 stat(s, ALLOC_FROM_PARTIAL);
1839 gfpflags &= gfp_allowed_mask;
1840 if (gfpflags & __GFP_WAIT)
1843 new = new_slab(s, gfpflags, node);
1845 if (gfpflags & __GFP_WAIT)
1846 local_irq_disable();
1849 c = __this_cpu_ptr(s->cpu_slab);
1850 stat(s, ALLOC_SLAB);
1854 __SetPageSlubFrozen(new);
1858 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1859 slab_out_of_memory(s, gfpflags, node);
1860 #ifdef CONFIG_CMPXCHG_LOCAL
1861 local_irq_restore(flags);
1865 if (!alloc_debug_processing(s, c->page, object, addr))
1869 c->page->freelist = get_freepointer(s, object);
1870 c->node = NUMA_NO_NODE;
1875 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1876 * have the fastpath folded into their functions. So no function call
1877 * overhead for requests that can be satisfied on the fastpath.
1879 * The fastpath works by first checking if the lockless freelist can be used.
1880 * If not then __slab_alloc is called for slow processing.
1882 * Otherwise we can simply pick the next object from the lockless free list.
1884 static __always_inline void *slab_alloc(struct kmem_cache *s,
1885 gfp_t gfpflags, int node, unsigned long addr)
1888 struct kmem_cache_cpu *c;
1889 #ifdef CONFIG_CMPXCHG_LOCAL
1892 unsigned long flags;
1895 if (slab_pre_alloc_hook(s, gfpflags))
1898 #ifndef CONFIG_CMPXCHG_LOCAL
1899 local_irq_save(flags);
1905 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1906 * enabled. We may switch back and forth between cpus while
1907 * reading from one cpu area. That does not matter as long
1908 * as we end up on the original cpu again when doing the cmpxchg.
1910 c = __this_cpu_ptr(s->cpu_slab);
1912 #ifdef CONFIG_CMPXCHG_LOCAL
1914 * The transaction ids are globally unique per cpu and per operation on
1915 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1916 * occurs on the right processor and that there was no operation on the
1917 * linked list in between.
1923 object = c->freelist;
1924 if (unlikely(!object || !node_match(c, node)))
1926 object = __slab_alloc(s, gfpflags, node, addr, c);
1929 #ifdef CONFIG_CMPXCHG_LOCAL
1931 * The cmpxchg will only match if there was no additonal
1932 * operation and if we are on the right processor.
1934 * The cmpxchg does the following atomically (without lock semantics!)
1935 * 1. Relocate first pointer to the current per cpu area.
1936 * 2. Verify that tid and freelist have not been changed
1937 * 3. If they were not changed replace tid and freelist
1939 * Since this is without lock semantics the protection is only against
1940 * code executing on this cpu *not* from access by other cpus.
1942 if (unlikely(!this_cpu_cmpxchg_double(
1943 s->cpu_slab->freelist, s->cpu_slab->tid,
1945 get_freepointer(s, object), next_tid(tid)))) {
1947 note_cmpxchg_failure("slab_alloc", s, tid);
1951 c->freelist = get_freepointer(s, object);
1953 stat(s, ALLOC_FASTPATH);
1956 #ifndef CONFIG_CMPXCHG_LOCAL
1957 local_irq_restore(flags);
1960 if (unlikely(gfpflags & __GFP_ZERO) && object)
1961 memset(object, 0, s->objsize);
1963 slab_post_alloc_hook(s, gfpflags, object);
1968 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1970 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1972 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
1976 EXPORT_SYMBOL(kmem_cache_alloc);
1978 #ifdef CONFIG_TRACING
1979 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1981 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1982 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1985 EXPORT_SYMBOL(kmem_cache_alloc_trace);
1987 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1989 void *ret = kmalloc_order(size, flags, order);
1990 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1993 EXPORT_SYMBOL(kmalloc_order_trace);
1997 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1999 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2001 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2002 s->objsize, s->size, gfpflags, node);
2006 EXPORT_SYMBOL(kmem_cache_alloc_node);
2008 #ifdef CONFIG_TRACING
2009 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2011 int node, size_t size)
2013 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2015 trace_kmalloc_node(_RET_IP_, ret,
2016 size, s->size, gfpflags, node);
2019 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2024 * Slow patch handling. This may still be called frequently since objects
2025 * have a longer lifetime than the cpu slabs in most processing loads.
2027 * So we still attempt to reduce cache line usage. Just take the slab
2028 * lock and free the item. If there is no additional partial page
2029 * handling required then we can return immediately.
2031 static void __slab_free(struct kmem_cache *s, struct page *page,
2032 void *x, unsigned long addr)
2035 void **object = (void *)x;
2036 #ifdef CONFIG_CMPXCHG_LOCAL
2037 unsigned long flags;
2039 local_irq_save(flags);
2042 stat(s, FREE_SLOWPATH);
2044 if (kmem_cache_debug(s))
2048 prior = page->freelist;
2049 set_freepointer(s, object, prior);
2050 page->freelist = object;
2053 if (unlikely(PageSlubFrozen(page))) {
2054 stat(s, FREE_FROZEN);
2058 if (unlikely(!page->inuse))
2062 * Objects left in the slab. If it was not on the partial list before
2065 if (unlikely(!prior)) {
2066 add_partial(get_node(s, page_to_nid(page)), page, 1);
2067 stat(s, FREE_ADD_PARTIAL);
2072 #ifdef CONFIG_CMPXCHG_LOCAL
2073 local_irq_restore(flags);
2080 * Slab still on the partial list.
2082 remove_partial(s, page);
2083 stat(s, FREE_REMOVE_PARTIAL);
2086 #ifdef CONFIG_CMPXCHG_LOCAL
2087 local_irq_restore(flags);
2090 discard_slab(s, page);
2094 if (!free_debug_processing(s, page, x, addr))
2100 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2101 * can perform fastpath freeing without additional function calls.
2103 * The fastpath is only possible if we are freeing to the current cpu slab
2104 * of this processor. This typically the case if we have just allocated
2107 * If fastpath is not possible then fall back to __slab_free where we deal
2108 * with all sorts of special processing.
2110 static __always_inline void slab_free(struct kmem_cache *s,
2111 struct page *page, void *x, unsigned long addr)
2113 void **object = (void *)x;
2114 struct kmem_cache_cpu *c;
2115 #ifdef CONFIG_CMPXCHG_LOCAL
2118 unsigned long flags;
2121 slab_free_hook(s, x);
2123 #ifndef CONFIG_CMPXCHG_LOCAL
2124 local_irq_save(flags);
2131 * Determine the currently cpus per cpu slab.
2132 * The cpu may change afterward. However that does not matter since
2133 * data is retrieved via this pointer. If we are on the same cpu
2134 * during the cmpxchg then the free will succedd.
2136 c = __this_cpu_ptr(s->cpu_slab);
2138 #ifdef CONFIG_CMPXCHG_LOCAL
2143 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
2144 set_freepointer(s, object, c->freelist);
2146 #ifdef CONFIG_CMPXCHG_LOCAL
2147 if (unlikely(!this_cpu_cmpxchg_double(
2148 s->cpu_slab->freelist, s->cpu_slab->tid,
2150 object, next_tid(tid)))) {
2152 note_cmpxchg_failure("slab_free", s, tid);
2156 c->freelist = object;
2158 stat(s, FREE_FASTPATH);
2160 __slab_free(s, page, x, addr);
2162 #ifndef CONFIG_CMPXCHG_LOCAL
2163 local_irq_restore(flags);
2167 void kmem_cache_free(struct kmem_cache *s, void *x)
2171 page = virt_to_head_page(x);
2173 slab_free(s, page, x, _RET_IP_);
2175 trace_kmem_cache_free(_RET_IP_, x);
2177 EXPORT_SYMBOL(kmem_cache_free);
2180 * Object placement in a slab is made very easy because we always start at
2181 * offset 0. If we tune the size of the object to the alignment then we can
2182 * get the required alignment by putting one properly sized object after
2185 * Notice that the allocation order determines the sizes of the per cpu
2186 * caches. Each processor has always one slab available for allocations.
2187 * Increasing the allocation order reduces the number of times that slabs
2188 * must be moved on and off the partial lists and is therefore a factor in
2193 * Mininum / Maximum order of slab pages. This influences locking overhead
2194 * and slab fragmentation. A higher order reduces the number of partial slabs
2195 * and increases the number of allocations possible without having to
2196 * take the list_lock.
2198 static int slub_min_order;
2199 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
2200 static int slub_min_objects;
2203 * Merge control. If this is set then no merging of slab caches will occur.
2204 * (Could be removed. This was introduced to pacify the merge skeptics.)
2206 static int slub_nomerge;
2209 * Calculate the order of allocation given an slab object size.
2211 * The order of allocation has significant impact on performance and other
2212 * system components. Generally order 0 allocations should be preferred since
2213 * order 0 does not cause fragmentation in the page allocator. Larger objects
2214 * be problematic to put into order 0 slabs because there may be too much
2215 * unused space left. We go to a higher order if more than 1/16th of the slab
2218 * In order to reach satisfactory performance we must ensure that a minimum
2219 * number of objects is in one slab. Otherwise we may generate too much
2220 * activity on the partial lists which requires taking the list_lock. This is
2221 * less a concern for large slabs though which are rarely used.
2223 * slub_max_order specifies the order where we begin to stop considering the
2224 * number of objects in a slab as critical. If we reach slub_max_order then
2225 * we try to keep the page order as low as possible. So we accept more waste
2226 * of space in favor of a small page order.
2228 * Higher order allocations also allow the placement of more objects in a
2229 * slab and thereby reduce object handling overhead. If the user has
2230 * requested a higher mininum order then we start with that one instead of
2231 * the smallest order which will fit the object.
2233 static inline int slab_order(int size, int min_objects,
2234 int max_order, int fract_leftover, int reserved)
2238 int min_order = slub_min_order;
2240 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
2241 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
2243 for (order = max(min_order,
2244 fls(min_objects * size - 1) - PAGE_SHIFT);
2245 order <= max_order; order++) {
2247 unsigned long slab_size = PAGE_SIZE << order;
2249 if (slab_size < min_objects * size + reserved)
2252 rem = (slab_size - reserved) % size;
2254 if (rem <= slab_size / fract_leftover)
2262 static inline int calculate_order(int size, int reserved)
2270 * Attempt to find best configuration for a slab. This
2271 * works by first attempting to generate a layout with
2272 * the best configuration and backing off gradually.
2274 * First we reduce the acceptable waste in a slab. Then
2275 * we reduce the minimum objects required in a slab.
2277 min_objects = slub_min_objects;
2279 min_objects = 4 * (fls(nr_cpu_ids) + 1);
2280 max_objects = order_objects(slub_max_order, size, reserved);
2281 min_objects = min(min_objects, max_objects);
2283 while (min_objects > 1) {
2285 while (fraction >= 4) {
2286 order = slab_order(size, min_objects,
2287 slub_max_order, fraction, reserved);
2288 if (order <= slub_max_order)
2296 * We were unable to place multiple objects in a slab. Now
2297 * lets see if we can place a single object there.
2299 order = slab_order(size, 1, slub_max_order, 1, reserved);
2300 if (order <= slub_max_order)
2304 * Doh this slab cannot be placed using slub_max_order.
2306 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
2307 if (order < MAX_ORDER)
2313 * Figure out what the alignment of the objects will be.
2315 static unsigned long calculate_alignment(unsigned long flags,
2316 unsigned long align, unsigned long size)
2319 * If the user wants hardware cache aligned objects then follow that
2320 * suggestion if the object is sufficiently large.
2322 * The hardware cache alignment cannot override the specified
2323 * alignment though. If that is greater then use it.
2325 if (flags & SLAB_HWCACHE_ALIGN) {
2326 unsigned long ralign = cache_line_size();
2327 while (size <= ralign / 2)
2329 align = max(align, ralign);
2332 if (align < ARCH_SLAB_MINALIGN)
2333 align = ARCH_SLAB_MINALIGN;
2335 return ALIGN(align, sizeof(void *));
2339 init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
2342 spin_lock_init(&n->list_lock);
2343 INIT_LIST_HEAD(&n->partial);
2344 #ifdef CONFIG_SLUB_DEBUG
2345 atomic_long_set(&n->nr_slabs, 0);
2346 atomic_long_set(&n->total_objects, 0);
2347 INIT_LIST_HEAD(&n->full);
2351 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
2353 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2354 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
2356 #ifdef CONFIG_CMPXCHG_LOCAL
2358 * Must align to double word boundary for the double cmpxchg instructions
2361 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
2363 /* Regular alignment is sufficient */
2364 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2370 init_kmem_cache_cpus(s);
2375 static struct kmem_cache *kmem_cache_node;
2378 * No kmalloc_node yet so do it by hand. We know that this is the first
2379 * slab on the node for this slabcache. There are no concurrent accesses
2382 * Note that this function only works on the kmalloc_node_cache
2383 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2384 * memory on a fresh node that has no slab structures yet.
2386 static void early_kmem_cache_node_alloc(int node)
2389 struct kmem_cache_node *n;
2390 unsigned long flags;
2392 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
2394 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
2397 if (page_to_nid(page) != node) {
2398 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2400 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2401 "in order to be able to continue\n");
2406 page->freelist = get_freepointer(kmem_cache_node, n);
2408 kmem_cache_node->node[node] = n;
2409 #ifdef CONFIG_SLUB_DEBUG
2410 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
2411 init_tracking(kmem_cache_node, n);
2413 init_kmem_cache_node(n, kmem_cache_node);
2414 inc_slabs_node(kmem_cache_node, node, page->objects);
2417 * lockdep requires consistent irq usage for each lock
2418 * so even though there cannot be a race this early in
2419 * the boot sequence, we still disable irqs.
2421 local_irq_save(flags);
2422 add_partial(n, page, 0);
2423 local_irq_restore(flags);
2426 static void free_kmem_cache_nodes(struct kmem_cache *s)
2430 for_each_node_state(node, N_NORMAL_MEMORY) {
2431 struct kmem_cache_node *n = s->node[node];
2434 kmem_cache_free(kmem_cache_node, n);
2436 s->node[node] = NULL;
2440 static int init_kmem_cache_nodes(struct kmem_cache *s)
2444 for_each_node_state(node, N_NORMAL_MEMORY) {
2445 struct kmem_cache_node *n;
2447 if (slab_state == DOWN) {
2448 early_kmem_cache_node_alloc(node);
2451 n = kmem_cache_alloc_node(kmem_cache_node,
2455 free_kmem_cache_nodes(s);
2460 init_kmem_cache_node(n, s);
2465 static void set_min_partial(struct kmem_cache *s, unsigned long min)
2467 if (min < MIN_PARTIAL)
2469 else if (min > MAX_PARTIAL)
2471 s->min_partial = min;
2475 * calculate_sizes() determines the order and the distribution of data within
2478 static int calculate_sizes(struct kmem_cache *s, int forced_order)
2480 unsigned long flags = s->flags;
2481 unsigned long size = s->objsize;
2482 unsigned long align = s->align;
2486 * Round up object size to the next word boundary. We can only
2487 * place the free pointer at word boundaries and this determines
2488 * the possible location of the free pointer.
2490 size = ALIGN(size, sizeof(void *));
2492 #ifdef CONFIG_SLUB_DEBUG
2494 * Determine if we can poison the object itself. If the user of
2495 * the slab may touch the object after free or before allocation
2496 * then we should never poison the object itself.
2498 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
2500 s->flags |= __OBJECT_POISON;
2502 s->flags &= ~__OBJECT_POISON;
2506 * If we are Redzoning then check if there is some space between the
2507 * end of the object and the free pointer. If not then add an
2508 * additional word to have some bytes to store Redzone information.
2510 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2511 size += sizeof(void *);
2515 * With that we have determined the number of bytes in actual use
2516 * by the object. This is the potential offset to the free pointer.
2520 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
2523 * Relocate free pointer after the object if it is not
2524 * permitted to overwrite the first word of the object on
2527 * This is the case if we do RCU, have a constructor or
2528 * destructor or are poisoning the objects.
2531 size += sizeof(void *);
2534 #ifdef CONFIG_SLUB_DEBUG
2535 if (flags & SLAB_STORE_USER)
2537 * Need to store information about allocs and frees after
2540 size += 2 * sizeof(struct track);
2542 if (flags & SLAB_RED_ZONE)
2544 * Add some empty padding so that we can catch
2545 * overwrites from earlier objects rather than let
2546 * tracking information or the free pointer be
2547 * corrupted if a user writes before the start
2550 size += sizeof(void *);
2554 * Determine the alignment based on various parameters that the
2555 * user specified and the dynamic determination of cache line size
2558 align = calculate_alignment(flags, align, s->objsize);
2562 * SLUB stores one object immediately after another beginning from
2563 * offset 0. In order to align the objects we have to simply size
2564 * each object to conform to the alignment.
2566 size = ALIGN(size, align);
2568 if (forced_order >= 0)
2569 order = forced_order;
2571 order = calculate_order(size, s->reserved);
2578 s->allocflags |= __GFP_COMP;
2580 if (s->flags & SLAB_CACHE_DMA)
2581 s->allocflags |= SLUB_DMA;
2583 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2584 s->allocflags |= __GFP_RECLAIMABLE;
2587 * Determine the number of objects per slab
2589 s->oo = oo_make(order, size, s->reserved);
2590 s->min = oo_make(get_order(size), size, s->reserved);
2591 if (oo_objects(s->oo) > oo_objects(s->max))
2594 return !!oo_objects(s->oo);
2598 static int kmem_cache_open(struct kmem_cache *s,
2599 const char *name, size_t size,
2600 size_t align, unsigned long flags,
2601 void (*ctor)(void *))
2603 memset(s, 0, kmem_size);
2608 s->flags = kmem_cache_flags(size, flags, name, ctor);
2611 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2612 s->reserved = sizeof(struct rcu_head);
2614 if (!calculate_sizes(s, -1))
2616 if (disable_higher_order_debug) {
2618 * Disable debugging flags that store metadata if the min slab
2621 if (get_order(s->size) > get_order(s->objsize)) {
2622 s->flags &= ~DEBUG_METADATA_FLAGS;
2624 if (!calculate_sizes(s, -1))
2630 * The larger the object size is, the more pages we want on the partial
2631 * list to avoid pounding the page allocator excessively.
2633 set_min_partial(s, ilog2(s->size));
2636 s->remote_node_defrag_ratio = 1000;
2638 if (!init_kmem_cache_nodes(s))
2641 if (alloc_kmem_cache_cpus(s))
2644 free_kmem_cache_nodes(s);
2646 if (flags & SLAB_PANIC)
2647 panic("Cannot create slab %s size=%lu realsize=%u "
2648 "order=%u offset=%u flags=%lx\n",
2649 s->name, (unsigned long)size, s->size, oo_order(s->oo),
2655 * Determine the size of a slab object
2657 unsigned int kmem_cache_size(struct kmem_cache *s)
2661 EXPORT_SYMBOL(kmem_cache_size);
2663 static void list_slab_objects(struct kmem_cache *s, struct page *page,
2666 #ifdef CONFIG_SLUB_DEBUG
2667 void *addr = page_address(page);
2669 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2670 sizeof(long), GFP_ATOMIC);
2673 slab_err(s, page, "%s", text);
2675 for_each_free_object(p, s, page->freelist)
2676 set_bit(slab_index(p, s, addr), map);
2678 for_each_object(p, s, addr, page->objects) {
2680 if (!test_bit(slab_index(p, s, addr), map)) {
2681 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2683 print_tracking(s, p);
2692 * Attempt to free all partial slabs on a node.
2694 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
2696 unsigned long flags;
2697 struct page *page, *h;
2699 spin_lock_irqsave(&n->list_lock, flags);
2700 list_for_each_entry_safe(page, h, &n->partial, lru) {
2702 __remove_partial(n, page);
2703 discard_slab(s, page);
2705 list_slab_objects(s, page,
2706 "Objects remaining on kmem_cache_close()");
2709 spin_unlock_irqrestore(&n->list_lock, flags);
2713 * Release all resources used by a slab cache.
2715 static inline int kmem_cache_close(struct kmem_cache *s)
2720 free_percpu(s->cpu_slab);
2721 /* Attempt to free all objects */
2722 for_each_node_state(node, N_NORMAL_MEMORY) {
2723 struct kmem_cache_node *n = get_node(s, node);
2726 if (n->nr_partial || slabs_node(s, node))
2729 free_kmem_cache_nodes(s);
2734 * Close a cache and release the kmem_cache structure
2735 * (must be used for caches created using kmem_cache_create)
2737 void kmem_cache_destroy(struct kmem_cache *s)
2739 down_write(&slub_lock);
2743 if (kmem_cache_close(s)) {
2744 printk(KERN_ERR "SLUB %s: %s called for cache that "
2745 "still has objects.\n", s->name, __func__);
2748 if (s->flags & SLAB_DESTROY_BY_RCU)
2750 sysfs_slab_remove(s);
2752 up_write(&slub_lock);
2754 EXPORT_SYMBOL(kmem_cache_destroy);
2756 /********************************************************************
2758 *******************************************************************/
2760 struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
2761 EXPORT_SYMBOL(kmalloc_caches);
2763 static struct kmem_cache *kmem_cache;
2765 #ifdef CONFIG_ZONE_DMA
2766 static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
2769 static int __init setup_slub_min_order(char *str)
2771 get_option(&str, &slub_min_order);
2776 __setup("slub_min_order=", setup_slub_min_order);
2778 static int __init setup_slub_max_order(char *str)
2780 get_option(&str, &slub_max_order);
2781 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
2786 __setup("slub_max_order=", setup_slub_max_order);
2788 static int __init setup_slub_min_objects(char *str)
2790 get_option(&str, &slub_min_objects);
2795 __setup("slub_min_objects=", setup_slub_min_objects);
2797 static int __init setup_slub_nomerge(char *str)
2803 __setup("slub_nomerge", setup_slub_nomerge);
2805 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2806 int size, unsigned int flags)
2808 struct kmem_cache *s;
2810 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2813 * This function is called with IRQs disabled during early-boot on
2814 * single CPU so there's no need to take slub_lock here.
2816 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
2820 list_add(&s->list, &slab_caches);
2824 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2829 * Conversion table for small slabs sizes / 8 to the index in the
2830 * kmalloc array. This is necessary for slabs < 192 since we have non power
2831 * of two cache sizes there. The size of larger slabs can be determined using
2834 static s8 size_index[24] = {
2861 static inline int size_index_elem(size_t bytes)
2863 return (bytes - 1) / 8;
2866 static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2872 return ZERO_SIZE_PTR;
2874 index = size_index[size_index_elem(size)];
2876 index = fls(size - 1);
2878 #ifdef CONFIG_ZONE_DMA
2879 if (unlikely((flags & SLUB_DMA)))
2880 return kmalloc_dma_caches[index];
2883 return kmalloc_caches[index];
2886 void *__kmalloc(size_t size, gfp_t flags)
2888 struct kmem_cache *s;
2891 if (unlikely(size > SLUB_MAX_SIZE))
2892 return kmalloc_large(size, flags);
2894 s = get_slab(size, flags);
2896 if (unlikely(ZERO_OR_NULL_PTR(s)))
2899 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
2901 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
2905 EXPORT_SYMBOL(__kmalloc);
2908 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2913 flags |= __GFP_COMP | __GFP_NOTRACK;
2914 page = alloc_pages_node(node, flags, get_order(size));
2916 ptr = page_address(page);
2918 kmemleak_alloc(ptr, size, 1, flags);
2922 void *__kmalloc_node(size_t size, gfp_t flags, int node)
2924 struct kmem_cache *s;
2927 if (unlikely(size > SLUB_MAX_SIZE)) {
2928 ret = kmalloc_large_node(size, flags, node);
2930 trace_kmalloc_node(_RET_IP_, ret,
2931 size, PAGE_SIZE << get_order(size),
2937 s = get_slab(size, flags);
2939 if (unlikely(ZERO_OR_NULL_PTR(s)))
2942 ret = slab_alloc(s, flags, node, _RET_IP_);
2944 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
2948 EXPORT_SYMBOL(__kmalloc_node);
2951 size_t ksize(const void *object)
2955 if (unlikely(object == ZERO_SIZE_PTR))
2958 page = virt_to_head_page(object);
2960 if (unlikely(!PageSlab(page))) {
2961 WARN_ON(!PageCompound(page));
2962 return PAGE_SIZE << compound_order(page);
2965 return slab_ksize(page->slab);
2967 EXPORT_SYMBOL(ksize);
2969 void kfree(const void *x)
2972 void *object = (void *)x;
2974 trace_kfree(_RET_IP_, x);
2976 if (unlikely(ZERO_OR_NULL_PTR(x)))
2979 page = virt_to_head_page(x);
2980 if (unlikely(!PageSlab(page))) {
2981 BUG_ON(!PageCompound(page));
2986 slab_free(page->slab, page, object, _RET_IP_);
2988 EXPORT_SYMBOL(kfree);
2991 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2992 * the remaining slabs by the number of items in use. The slabs with the
2993 * most items in use come first. New allocations will then fill those up
2994 * and thus they can be removed from the partial lists.
2996 * The slabs with the least items are placed last. This results in them
2997 * being allocated from last increasing the chance that the last objects
2998 * are freed in them.
3000 int kmem_cache_shrink(struct kmem_cache *s)
3004 struct kmem_cache_node *n;
3007 int objects = oo_objects(s->max);
3008 struct list_head *slabs_by_inuse =
3009 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
3010 unsigned long flags;
3012 if (!slabs_by_inuse)
3016 for_each_node_state(node, N_NORMAL_MEMORY) {
3017 n = get_node(s, node);
3022 for (i = 0; i < objects; i++)
3023 INIT_LIST_HEAD(slabs_by_inuse + i);
3025 spin_lock_irqsave(&n->list_lock, flags);
3028 * Build lists indexed by the items in use in each slab.
3030 * Note that concurrent frees may occur while we hold the
3031 * list_lock. page->inuse here is the upper limit.
3033 list_for_each_entry_safe(page, t, &n->partial, lru) {
3034 if (!page->inuse && slab_trylock(page)) {
3036 * Must hold slab lock here because slab_free
3037 * may have freed the last object and be
3038 * waiting to release the slab.
3040 __remove_partial(n, page);
3042 discard_slab(s, page);
3044 list_move(&page->lru,
3045 slabs_by_inuse + page->inuse);
3050 * Rebuild the partial list with the slabs filled up most
3051 * first and the least used slabs at the end.
3053 for (i = objects - 1; i >= 0; i--)
3054 list_splice(slabs_by_inuse + i, n->partial.prev);
3056 spin_unlock_irqrestore(&n->list_lock, flags);
3059 kfree(slabs_by_inuse);
3062 EXPORT_SYMBOL(kmem_cache_shrink);
3064 #if defined(CONFIG_MEMORY_HOTPLUG)
3065 static int slab_mem_going_offline_callback(void *arg)
3067 struct kmem_cache *s;
3069 down_read(&slub_lock);
3070 list_for_each_entry(s, &slab_caches, list)
3071 kmem_cache_shrink(s);
3072 up_read(&slub_lock);
3077 static void slab_mem_offline_callback(void *arg)
3079 struct kmem_cache_node *n;
3080 struct kmem_cache *s;
3081 struct memory_notify *marg = arg;
3084 offline_node = marg->status_change_nid;
3087 * If the node still has available memory. we need kmem_cache_node
3090 if (offline_node < 0)
3093 down_read(&slub_lock);
3094 list_for_each_entry(s, &slab_caches, list) {
3095 n = get_node(s, offline_node);
3098 * if n->nr_slabs > 0, slabs still exist on the node
3099 * that is going down. We were unable to free them,
3100 * and offline_pages() function shouldn't call this
3101 * callback. So, we must fail.
3103 BUG_ON(slabs_node(s, offline_node));
3105 s->node[offline_node] = NULL;
3106 kmem_cache_free(kmem_cache_node, n);
3109 up_read(&slub_lock);
3112 static int slab_mem_going_online_callback(void *arg)
3114 struct kmem_cache_node *n;
3115 struct kmem_cache *s;
3116 struct memory_notify *marg = arg;
3117 int nid = marg->status_change_nid;
3121 * If the node's memory is already available, then kmem_cache_node is
3122 * already created. Nothing to do.
3128 * We are bringing a node online. No memory is available yet. We must
3129 * allocate a kmem_cache_node structure in order to bring the node
3132 down_read(&slub_lock);
3133 list_for_each_entry(s, &slab_caches, list) {
3135 * XXX: kmem_cache_alloc_node will fallback to other nodes
3136 * since memory is not yet available from the node that
3139 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
3144 init_kmem_cache_node(n, s);
3148 up_read(&slub_lock);
3152 static int slab_memory_callback(struct notifier_block *self,
3153 unsigned long action, void *arg)
3158 case MEM_GOING_ONLINE:
3159 ret = slab_mem_going_online_callback(arg);
3161 case MEM_GOING_OFFLINE:
3162 ret = slab_mem_going_offline_callback(arg);
3165 case MEM_CANCEL_ONLINE:
3166 slab_mem_offline_callback(arg);
3169 case MEM_CANCEL_OFFLINE:
3173 ret = notifier_from_errno(ret);
3179 #endif /* CONFIG_MEMORY_HOTPLUG */
3181 /********************************************************************
3182 * Basic setup of slabs
3183 *******************************************************************/
3186 * Used for early kmem_cache structures that were allocated using
3187 * the page allocator
3190 static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3194 list_add(&s->list, &slab_caches);
3197 for_each_node_state(node, N_NORMAL_MEMORY) {
3198 struct kmem_cache_node *n = get_node(s, node);
3202 list_for_each_entry(p, &n->partial, lru)
3205 #ifdef CONFIG_SLAB_DEBUG
3206 list_for_each_entry(p, &n->full, lru)
3213 void __init kmem_cache_init(void)
3217 struct kmem_cache *temp_kmem_cache;
3219 struct kmem_cache *temp_kmem_cache_node;
3220 unsigned long kmalloc_size;
3222 kmem_size = offsetof(struct kmem_cache, node) +
3223 nr_node_ids * sizeof(struct kmem_cache_node *);
3225 /* Allocate two kmem_caches from the page allocator */
3226 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3227 order = get_order(2 * kmalloc_size);
3228 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3231 * Must first have the slab cache available for the allocations of the
3232 * struct kmem_cache_node's. There is special bootstrap code in
3233 * kmem_cache_open for slab_state == DOWN.
3235 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3237 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3238 sizeof(struct kmem_cache_node),
3239 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3241 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
3243 /* Able to allocate the per node structures */
3244 slab_state = PARTIAL;
3246 temp_kmem_cache = kmem_cache;
3247 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3248 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3249 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3250 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
3253 * Allocate kmem_cache_node properly from the kmem_cache slab.
3254 * kmem_cache_node is separately allocated so no need to
3255 * update any list pointers.
3257 temp_kmem_cache_node = kmem_cache_node;
3259 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3260 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3262 kmem_cache_bootstrap_fixup(kmem_cache_node);
3265 kmem_cache_bootstrap_fixup(kmem_cache);
3267 /* Free temporary boot structure */
3268 free_pages((unsigned long)temp_kmem_cache, order);
3270 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3273 * Patch up the size_index table if we have strange large alignment
3274 * requirements for the kmalloc array. This is only the case for
3275 * MIPS it seems. The standard arches will not generate any code here.
3277 * Largest permitted alignment is 256 bytes due to the way we
3278 * handle the index determination for the smaller caches.
3280 * Make sure that nothing crazy happens if someone starts tinkering
3281 * around with ARCH_KMALLOC_MINALIGN
3283 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3284 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3286 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3287 int elem = size_index_elem(i);
3288 if (elem >= ARRAY_SIZE(size_index))
3290 size_index[elem] = KMALLOC_SHIFT_LOW;
3293 if (KMALLOC_MIN_SIZE == 64) {
3295 * The 96 byte size cache is not used if the alignment
3298 for (i = 64 + 8; i <= 96; i += 8)
3299 size_index[size_index_elem(i)] = 7;
3300 } else if (KMALLOC_MIN_SIZE == 128) {
3302 * The 192 byte sized cache is not used if the alignment
3303 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3306 for (i = 128 + 8; i <= 192; i += 8)
3307 size_index[size_index_elem(i)] = 8;
3310 /* Caches that are not of the two-to-the-power-of size */
3311 if (KMALLOC_MIN_SIZE <= 32) {
3312 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3316 if (KMALLOC_MIN_SIZE <= 64) {
3317 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3321 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3322 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3328 /* Provide the correct kmalloc names now that the caches are up */
3329 if (KMALLOC_MIN_SIZE <= 32) {
3330 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3331 BUG_ON(!kmalloc_caches[1]->name);
3334 if (KMALLOC_MIN_SIZE <= 64) {
3335 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3336 BUG_ON(!kmalloc_caches[2]->name);
3339 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3340 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3343 kmalloc_caches[i]->name = s;
3347 register_cpu_notifier(&slab_notifier);
3350 #ifdef CONFIG_ZONE_DMA
3351 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3352 struct kmem_cache *s = kmalloc_caches[i];
3355 char *name = kasprintf(GFP_NOWAIT,
3356 "dma-kmalloc-%d", s->objsize);
3359 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3360 s->objsize, SLAB_CACHE_DMA);
3365 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3366 " CPUs=%d, Nodes=%d\n",
3367 caches, cache_line_size(),
3368 slub_min_order, slub_max_order, slub_min_objects,
3369 nr_cpu_ids, nr_node_ids);
3372 void __init kmem_cache_init_late(void)
3377 * Find a mergeable slab cache
3379 static int slab_unmergeable(struct kmem_cache *s)
3381 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3388 * We may have set a slab to be unmergeable during bootstrap.
3390 if (s->refcount < 0)
3396 static struct kmem_cache *find_mergeable(size_t size,
3397 size_t align, unsigned long flags, const char *name,
3398 void (*ctor)(void *))
3400 struct kmem_cache *s;
3402 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3408 size = ALIGN(size, sizeof(void *));
3409 align = calculate_alignment(flags, align, size);
3410 size = ALIGN(size, align);
3411 flags = kmem_cache_flags(size, flags, name, NULL);
3413 list_for_each_entry(s, &slab_caches, list) {
3414 if (slab_unmergeable(s))
3420 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
3423 * Check if alignment is compatible.
3424 * Courtesy of Adrian Drzewiecki
3426 if ((s->size & ~(align - 1)) != s->size)
3429 if (s->size - size >= sizeof(void *))
3437 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
3438 size_t align, unsigned long flags, void (*ctor)(void *))
3440 struct kmem_cache *s;
3446 down_write(&slub_lock);
3447 s = find_mergeable(size, align, flags, name, ctor);
3451 * Adjust the object sizes so that we clear
3452 * the complete object on kzalloc.
3454 s->objsize = max(s->objsize, (int)size);
3455 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
3457 if (sysfs_slab_alias(s, name)) {
3461 up_write(&slub_lock);
3465 n = kstrdup(name, GFP_KERNEL);
3469 s = kmalloc(kmem_size, GFP_KERNEL);
3471 if (kmem_cache_open(s, n,
3472 size, align, flags, ctor)) {
3473 list_add(&s->list, &slab_caches);
3474 if (sysfs_slab_add(s)) {
3480 up_write(&slub_lock);
3487 up_write(&slub_lock);
3489 if (flags & SLAB_PANIC)
3490 panic("Cannot create slabcache %s\n", name);
3495 EXPORT_SYMBOL(kmem_cache_create);
3499 * Use the cpu notifier to insure that the cpu slabs are flushed when
3502 static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3503 unsigned long action, void *hcpu)
3505 long cpu = (long)hcpu;
3506 struct kmem_cache *s;
3507 unsigned long flags;
3510 case CPU_UP_CANCELED:
3511 case CPU_UP_CANCELED_FROZEN:
3513 case CPU_DEAD_FROZEN:
3514 down_read(&slub_lock);
3515 list_for_each_entry(s, &slab_caches, list) {
3516 local_irq_save(flags);
3517 __flush_cpu_slab(s, cpu);
3518 local_irq_restore(flags);
3520 up_read(&slub_lock);
3528 static struct notifier_block __cpuinitdata slab_notifier = {
3529 .notifier_call = slab_cpuup_callback
3534 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
3536 struct kmem_cache *s;
3539 if (unlikely(size > SLUB_MAX_SIZE))
3540 return kmalloc_large(size, gfpflags);
3542 s = get_slab(size, gfpflags);
3544 if (unlikely(ZERO_OR_NULL_PTR(s)))
3547 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
3549 /* Honor the call site pointer we recieved. */
3550 trace_kmalloc(caller, ret, size, s->size, gfpflags);
3556 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
3557 int node, unsigned long caller)
3559 struct kmem_cache *s;
3562 if (unlikely(size > SLUB_MAX_SIZE)) {
3563 ret = kmalloc_large_node(size, gfpflags, node);
3565 trace_kmalloc_node(caller, ret,
3566 size, PAGE_SIZE << get_order(size),
3572 s = get_slab(size, gfpflags);
3574 if (unlikely(ZERO_OR_NULL_PTR(s)))
3577 ret = slab_alloc(s, gfpflags, node, caller);
3579 /* Honor the call site pointer we recieved. */
3580 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
3587 static int count_inuse(struct page *page)
3592 static int count_total(struct page *page)
3594 return page->objects;
3598 #ifdef CONFIG_SLUB_DEBUG
3599 static int validate_slab(struct kmem_cache *s, struct page *page,
3603 void *addr = page_address(page);
3605 if (!check_slab(s, page) ||
3606 !on_freelist(s, page, NULL))
3609 /* Now we know that a valid freelist exists */
3610 bitmap_zero(map, page->objects);
3612 for_each_free_object(p, s, page->freelist) {
3613 set_bit(slab_index(p, s, addr), map);
3614 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3618 for_each_object(p, s, addr, page->objects)
3619 if (!test_bit(slab_index(p, s, addr), map))
3620 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
3625 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3628 if (slab_trylock(page)) {
3629 validate_slab(s, page, map);
3632 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3636 static int validate_slab_node(struct kmem_cache *s,
3637 struct kmem_cache_node *n, unsigned long *map)
3639 unsigned long count = 0;
3641 unsigned long flags;
3643 spin_lock_irqsave(&n->list_lock, flags);
3645 list_for_each_entry(page, &n->partial, lru) {
3646 validate_slab_slab(s, page, map);
3649 if (count != n->nr_partial)
3650 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3651 "counter=%ld\n", s->name, count, n->nr_partial);
3653 if (!(s->flags & SLAB_STORE_USER))
3656 list_for_each_entry(page, &n->full, lru) {
3657 validate_slab_slab(s, page, map);
3660 if (count != atomic_long_read(&n->nr_slabs))
3661 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3662 "counter=%ld\n", s->name, count,
3663 atomic_long_read(&n->nr_slabs));
3666 spin_unlock_irqrestore(&n->list_lock, flags);
3670 static long validate_slab_cache(struct kmem_cache *s)
3673 unsigned long count = 0;
3674 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3675 sizeof(unsigned long), GFP_KERNEL);
3681 for_each_node_state(node, N_NORMAL_MEMORY) {
3682 struct kmem_cache_node *n = get_node(s, node);
3684 count += validate_slab_node(s, n, map);
3690 * Generate lists of code addresses where slabcache objects are allocated
3695 unsigned long count;
3702 DECLARE_BITMAP(cpus, NR_CPUS);
3708 unsigned long count;
3709 struct location *loc;
3712 static void free_loc_track(struct loc_track *t)
3715 free_pages((unsigned long)t->loc,
3716 get_order(sizeof(struct location) * t->max));
3719 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
3724 order = get_order(sizeof(struct location) * max);
3726 l = (void *)__get_free_pages(flags, order);
3731 memcpy(l, t->loc, sizeof(struct location) * t->count);
3739 static int add_location(struct loc_track *t, struct kmem_cache *s,
3740 const struct track *track)
3742 long start, end, pos;
3744 unsigned long caddr;
3745 unsigned long age = jiffies - track->when;
3751 pos = start + (end - start + 1) / 2;
3754 * There is nothing at "end". If we end up there
3755 * we need to add something to before end.
3760 caddr = t->loc[pos].addr;
3761 if (track->addr == caddr) {
3767 if (age < l->min_time)
3769 if (age > l->max_time)
3772 if (track->pid < l->min_pid)
3773 l->min_pid = track->pid;
3774 if (track->pid > l->max_pid)
3775 l->max_pid = track->pid;
3777 cpumask_set_cpu(track->cpu,
3778 to_cpumask(l->cpus));
3780 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3784 if (track->addr < caddr)
3791 * Not found. Insert new tracking element.
3793 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
3799 (t->count - pos) * sizeof(struct location));
3802 l->addr = track->addr;
3806 l->min_pid = track->pid;
3807 l->max_pid = track->pid;
3808 cpumask_clear(to_cpumask(l->cpus));
3809 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
3810 nodes_clear(l->nodes);
3811 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3815 static void process_slab(struct loc_track *t, struct kmem_cache *s,
3816 struct page *page, enum track_item alloc,
3819 void *addr = page_address(page);
3822 bitmap_zero(map, page->objects);
3823 for_each_free_object(p, s, page->freelist)
3824 set_bit(slab_index(p, s, addr), map);
3826 for_each_object(p, s, addr, page->objects)
3827 if (!test_bit(slab_index(p, s, addr), map))
3828 add_location(t, s, get_track(s, p, alloc));
3831 static int list_locations(struct kmem_cache *s, char *buf,
3832 enum track_item alloc)
3836 struct loc_track t = { 0, 0, NULL };
3838 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3839 sizeof(unsigned long), GFP_KERNEL);
3841 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3844 return sprintf(buf, "Out of memory\n");
3846 /* Push back cpu slabs */
3849 for_each_node_state(node, N_NORMAL_MEMORY) {
3850 struct kmem_cache_node *n = get_node(s, node);
3851 unsigned long flags;
3854 if (!atomic_long_read(&n->nr_slabs))
3857 spin_lock_irqsave(&n->list_lock, flags);
3858 list_for_each_entry(page, &n->partial, lru)
3859 process_slab(&t, s, page, alloc, map);
3860 list_for_each_entry(page, &n->full, lru)
3861 process_slab(&t, s, page, alloc, map);
3862 spin_unlock_irqrestore(&n->list_lock, flags);
3865 for (i = 0; i < t.count; i++) {
3866 struct location *l = &t.loc[i];
3868 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
3870 len += sprintf(buf + len, "%7ld ", l->count);
3873 len += sprintf(buf + len, "%pS", (void *)l->addr);
3875 len += sprintf(buf + len, "<not-available>");
3877 if (l->sum_time != l->min_time) {
3878 len += sprintf(buf + len, " age=%ld/%ld/%ld",
3880 (long)div_u64(l->sum_time, l->count),
3883 len += sprintf(buf + len, " age=%ld",
3886 if (l->min_pid != l->max_pid)
3887 len += sprintf(buf + len, " pid=%ld-%ld",
3888 l->min_pid, l->max_pid);
3890 len += sprintf(buf + len, " pid=%ld",
3893 if (num_online_cpus() > 1 &&
3894 !cpumask_empty(to_cpumask(l->cpus)) &&
3895 len < PAGE_SIZE - 60) {
3896 len += sprintf(buf + len, " cpus=");
3897 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3898 to_cpumask(l->cpus));
3901 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
3902 len < PAGE_SIZE - 60) {
3903 len += sprintf(buf + len, " nodes=");
3904 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3908 len += sprintf(buf + len, "\n");
3914 len += sprintf(buf, "No data\n");
3919 #ifdef SLUB_RESILIENCY_TEST
3920 static void resiliency_test(void)
3924 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3926 printk(KERN_ERR "SLUB resiliency testing\n");
3927 printk(KERN_ERR "-----------------------\n");
3928 printk(KERN_ERR "A. Corruption after allocation\n");
3930 p = kzalloc(16, GFP_KERNEL);
3932 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3933 " 0x12->0x%p\n\n", p + 16);
3935 validate_slab_cache(kmalloc_caches[4]);
3937 /* Hmmm... The next two are dangerous */
3938 p = kzalloc(32, GFP_KERNEL);
3939 p[32 + sizeof(void *)] = 0x34;
3940 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3941 " 0x34 -> -0x%p\n", p);
3943 "If allocated object is overwritten then not detectable\n\n");
3945 validate_slab_cache(kmalloc_caches[5]);
3946 p = kzalloc(64, GFP_KERNEL);
3947 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3949 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3952 "If allocated object is overwritten then not detectable\n\n");
3953 validate_slab_cache(kmalloc_caches[6]);
3955 printk(KERN_ERR "\nB. Corruption after free\n");
3956 p = kzalloc(128, GFP_KERNEL);
3959 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3960 validate_slab_cache(kmalloc_caches[7]);
3962 p = kzalloc(256, GFP_KERNEL);
3965 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3967 validate_slab_cache(kmalloc_caches[8]);
3969 p = kzalloc(512, GFP_KERNEL);
3972 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3973 validate_slab_cache(kmalloc_caches[9]);
3977 static void resiliency_test(void) {};
3982 enum slab_stat_type {
3983 SL_ALL, /* All slabs */
3984 SL_PARTIAL, /* Only partially allocated slabs */
3985 SL_CPU, /* Only slabs used for cpu caches */
3986 SL_OBJECTS, /* Determine allocated objects not slabs */
3987 SL_TOTAL /* Determine object capacity not slabs */
3990 #define SO_ALL (1 << SL_ALL)
3991 #define SO_PARTIAL (1 << SL_PARTIAL)
3992 #define SO_CPU (1 << SL_CPU)
3993 #define SO_OBJECTS (1 << SL_OBJECTS)
3994 #define SO_TOTAL (1 << SL_TOTAL)
3996 static ssize_t show_slab_objects(struct kmem_cache *s,
3997 char *buf, unsigned long flags)
3999 unsigned long total = 0;
4002 unsigned long *nodes;
4003 unsigned long *per_cpu;
4005 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
4008 per_cpu = nodes + nr_node_ids;
4010 if (flags & SO_CPU) {
4013 for_each_possible_cpu(cpu) {
4014 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
4016 if (!c || c->node < 0)
4020 if (flags & SO_TOTAL)
4021 x = c->page->objects;
4022 else if (flags & SO_OBJECTS)
4028 nodes[c->node] += x;
4034 lock_memory_hotplug();
4035 #ifdef CONFIG_SLUB_DEBUG
4036 if (flags & SO_ALL) {
4037 for_each_node_state(node, N_NORMAL_MEMORY) {
4038 struct kmem_cache_node *n = get_node(s, node);
4040 if (flags & SO_TOTAL)
4041 x = atomic_long_read(&n->total_objects);
4042 else if (flags & SO_OBJECTS)
4043 x = atomic_long_read(&n->total_objects) -
4044 count_partial(n, count_free);
4047 x = atomic_long_read(&n->nr_slabs);
4054 if (flags & SO_PARTIAL) {
4055 for_each_node_state(node, N_NORMAL_MEMORY) {
4056 struct kmem_cache_node *n = get_node(s, node);
4058 if (flags & SO_TOTAL)
4059 x = count_partial(n, count_total);
4060 else if (flags & SO_OBJECTS)
4061 x = count_partial(n, count_inuse);
4068 x = sprintf(buf, "%lu", total);
4070 for_each_node_state(node, N_NORMAL_MEMORY)
4072 x += sprintf(buf + x, " N%d=%lu",
4075 unlock_memory_hotplug();
4077 return x + sprintf(buf + x, "\n");
4080 #ifdef CONFIG_SLUB_DEBUG
4081 static int any_slab_objects(struct kmem_cache *s)
4085 for_each_online_node(node) {
4086 struct kmem_cache_node *n = get_node(s, node);
4091 if (atomic_long_read(&n->total_objects))
4098 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4099 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
4101 struct slab_attribute {
4102 struct attribute attr;
4103 ssize_t (*show)(struct kmem_cache *s, char *buf);
4104 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4107 #define SLAB_ATTR_RO(_name) \
4108 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4110 #define SLAB_ATTR(_name) \
4111 static struct slab_attribute _name##_attr = \
4112 __ATTR(_name, 0644, _name##_show, _name##_store)
4114 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4116 return sprintf(buf, "%d\n", s->size);
4118 SLAB_ATTR_RO(slab_size);
4120 static ssize_t align_show(struct kmem_cache *s, char *buf)
4122 return sprintf(buf, "%d\n", s->align);
4124 SLAB_ATTR_RO(align);
4126 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4128 return sprintf(buf, "%d\n", s->objsize);
4130 SLAB_ATTR_RO(object_size);
4132 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4134 return sprintf(buf, "%d\n", oo_objects(s->oo));
4136 SLAB_ATTR_RO(objs_per_slab);
4138 static ssize_t order_store(struct kmem_cache *s,
4139 const char *buf, size_t length)
4141 unsigned long order;
4144 err = strict_strtoul(buf, 10, &order);
4148 if (order > slub_max_order || order < slub_min_order)
4151 calculate_sizes(s, order);
4155 static ssize_t order_show(struct kmem_cache *s, char *buf)
4157 return sprintf(buf, "%d\n", oo_order(s->oo));
4161 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4163 return sprintf(buf, "%lu\n", s->min_partial);
4166 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4172 err = strict_strtoul(buf, 10, &min);
4176 set_min_partial(s, min);
4179 SLAB_ATTR(min_partial);
4181 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4185 return sprintf(buf, "%pS\n", s->ctor);
4189 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4191 return sprintf(buf, "%d\n", s->refcount - 1);
4193 SLAB_ATTR_RO(aliases);
4195 static ssize_t partial_show(struct kmem_cache *s, char *buf)
4197 return show_slab_objects(s, buf, SO_PARTIAL);
4199 SLAB_ATTR_RO(partial);
4201 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4203 return show_slab_objects(s, buf, SO_CPU);
4205 SLAB_ATTR_RO(cpu_slabs);
4207 static ssize_t objects_show(struct kmem_cache *s, char *buf)
4209 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
4211 SLAB_ATTR_RO(objects);
4213 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4215 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4217 SLAB_ATTR_RO(objects_partial);
4219 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4221 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4224 static ssize_t reclaim_account_store(struct kmem_cache *s,
4225 const char *buf, size_t length)
4227 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4229 s->flags |= SLAB_RECLAIM_ACCOUNT;
4232 SLAB_ATTR(reclaim_account);
4234 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4236 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
4238 SLAB_ATTR_RO(hwcache_align);
4240 #ifdef CONFIG_ZONE_DMA
4241 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4243 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4245 SLAB_ATTR_RO(cache_dma);
4248 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4250 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4252 SLAB_ATTR_RO(destroy_by_rcu);
4254 static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4256 return sprintf(buf, "%d\n", s->reserved);
4258 SLAB_ATTR_RO(reserved);
4260 #ifdef CONFIG_SLUB_DEBUG
4261 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4263 return show_slab_objects(s, buf, SO_ALL);
4265 SLAB_ATTR_RO(slabs);
4267 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4269 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4271 SLAB_ATTR_RO(total_objects);
4273 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4275 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4278 static ssize_t sanity_checks_store(struct kmem_cache *s,
4279 const char *buf, size_t length)
4281 s->flags &= ~SLAB_DEBUG_FREE;
4283 s->flags |= SLAB_DEBUG_FREE;
4286 SLAB_ATTR(sanity_checks);
4288 static ssize_t trace_show(struct kmem_cache *s, char *buf)
4290 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4293 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4296 s->flags &= ~SLAB_TRACE;
4298 s->flags |= SLAB_TRACE;
4303 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4305 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4308 static ssize_t red_zone_store(struct kmem_cache *s,
4309 const char *buf, size_t length)
4311 if (any_slab_objects(s))
4314 s->flags &= ~SLAB_RED_ZONE;
4316 s->flags |= SLAB_RED_ZONE;
4317 calculate_sizes(s, -1);
4320 SLAB_ATTR(red_zone);
4322 static ssize_t poison_show(struct kmem_cache *s, char *buf)
4324 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4327 static ssize_t poison_store(struct kmem_cache *s,
4328 const char *buf, size_t length)
4330 if (any_slab_objects(s))
4333 s->flags &= ~SLAB_POISON;
4335 s->flags |= SLAB_POISON;
4336 calculate_sizes(s, -1);
4341 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4343 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4346 static ssize_t store_user_store(struct kmem_cache *s,
4347 const char *buf, size_t length)
4349 if (any_slab_objects(s))
4352 s->flags &= ~SLAB_STORE_USER;
4354 s->flags |= SLAB_STORE_USER;
4355 calculate_sizes(s, -1);
4358 SLAB_ATTR(store_user);
4360 static ssize_t validate_show(struct kmem_cache *s, char *buf)
4365 static ssize_t validate_store(struct kmem_cache *s,
4366 const char *buf, size_t length)
4370 if (buf[0] == '1') {
4371 ret = validate_slab_cache(s);
4377 SLAB_ATTR(validate);
4379 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4381 if (!(s->flags & SLAB_STORE_USER))
4383 return list_locations(s, buf, TRACK_ALLOC);
4385 SLAB_ATTR_RO(alloc_calls);
4387 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4389 if (!(s->flags & SLAB_STORE_USER))
4391 return list_locations(s, buf, TRACK_FREE);
4393 SLAB_ATTR_RO(free_calls);
4394 #endif /* CONFIG_SLUB_DEBUG */
4396 #ifdef CONFIG_FAILSLAB
4397 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4399 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4402 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4405 s->flags &= ~SLAB_FAILSLAB;
4407 s->flags |= SLAB_FAILSLAB;
4410 SLAB_ATTR(failslab);
4413 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4418 static ssize_t shrink_store(struct kmem_cache *s,
4419 const char *buf, size_t length)
4421 if (buf[0] == '1') {
4422 int rc = kmem_cache_shrink(s);
4433 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
4435 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
4438 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
4439 const char *buf, size_t length)
4441 unsigned long ratio;
4444 err = strict_strtoul(buf, 10, &ratio);
4449 s->remote_node_defrag_ratio = ratio * 10;
4453 SLAB_ATTR(remote_node_defrag_ratio);
4456 #ifdef CONFIG_SLUB_STATS
4457 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4459 unsigned long sum = 0;
4462 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4467 for_each_online_cpu(cpu) {
4468 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
4474 len = sprintf(buf, "%lu", sum);
4477 for_each_online_cpu(cpu) {
4478 if (data[cpu] && len < PAGE_SIZE - 20)
4479 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
4483 return len + sprintf(buf + len, "\n");
4486 static void clear_stat(struct kmem_cache *s, enum stat_item si)
4490 for_each_online_cpu(cpu)
4491 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
4494 #define STAT_ATTR(si, text) \
4495 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4497 return show_stat(s, buf, si); \
4499 static ssize_t text##_store(struct kmem_cache *s, \
4500 const char *buf, size_t length) \
4502 if (buf[0] != '0') \
4504 clear_stat(s, si); \
4509 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4510 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4511 STAT_ATTR(FREE_FASTPATH, free_fastpath);
4512 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4513 STAT_ATTR(FREE_FROZEN, free_frozen);
4514 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4515 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4516 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4517 STAT_ATTR(ALLOC_SLAB, alloc_slab);
4518 STAT_ATTR(ALLOC_REFILL, alloc_refill);
4519 STAT_ATTR(FREE_SLAB, free_slab);
4520 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4521 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4522 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4523 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4524 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4525 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
4526 STAT_ATTR(ORDER_FALLBACK, order_fallback);
4529 static struct attribute *slab_attrs[] = {
4530 &slab_size_attr.attr,
4531 &object_size_attr.attr,
4532 &objs_per_slab_attr.attr,
4534 &min_partial_attr.attr,
4536 &objects_partial_attr.attr,
4538 &cpu_slabs_attr.attr,
4542 &hwcache_align_attr.attr,
4543 &reclaim_account_attr.attr,
4544 &destroy_by_rcu_attr.attr,
4546 &reserved_attr.attr,
4547 #ifdef CONFIG_SLUB_DEBUG
4548 &total_objects_attr.attr,
4550 &sanity_checks_attr.attr,
4552 &red_zone_attr.attr,
4554 &store_user_attr.attr,
4555 &validate_attr.attr,
4556 &alloc_calls_attr.attr,
4557 &free_calls_attr.attr,
4559 #ifdef CONFIG_ZONE_DMA
4560 &cache_dma_attr.attr,
4563 &remote_node_defrag_ratio_attr.attr,
4565 #ifdef CONFIG_SLUB_STATS
4566 &alloc_fastpath_attr.attr,
4567 &alloc_slowpath_attr.attr,
4568 &free_fastpath_attr.attr,
4569 &free_slowpath_attr.attr,
4570 &free_frozen_attr.attr,
4571 &free_add_partial_attr.attr,
4572 &free_remove_partial_attr.attr,
4573 &alloc_from_partial_attr.attr,
4574 &alloc_slab_attr.attr,
4575 &alloc_refill_attr.attr,
4576 &free_slab_attr.attr,
4577 &cpuslab_flush_attr.attr,
4578 &deactivate_full_attr.attr,
4579 &deactivate_empty_attr.attr,
4580 &deactivate_to_head_attr.attr,
4581 &deactivate_to_tail_attr.attr,
4582 &deactivate_remote_frees_attr.attr,
4583 &order_fallback_attr.attr,
4585 #ifdef CONFIG_FAILSLAB
4586 &failslab_attr.attr,
4592 static struct attribute_group slab_attr_group = {
4593 .attrs = slab_attrs,
4596 static ssize_t slab_attr_show(struct kobject *kobj,
4597 struct attribute *attr,
4600 struct slab_attribute *attribute;
4601 struct kmem_cache *s;
4604 attribute = to_slab_attr(attr);
4607 if (!attribute->show)
4610 err = attribute->show(s, buf);
4615 static ssize_t slab_attr_store(struct kobject *kobj,
4616 struct attribute *attr,
4617 const char *buf, size_t len)
4619 struct slab_attribute *attribute;
4620 struct kmem_cache *s;
4623 attribute = to_slab_attr(attr);
4626 if (!attribute->store)
4629 err = attribute->store(s, buf, len);
4634 static void kmem_cache_release(struct kobject *kobj)
4636 struct kmem_cache *s = to_slab(kobj);
4642 static const struct sysfs_ops slab_sysfs_ops = {
4643 .show = slab_attr_show,
4644 .store = slab_attr_store,
4647 static struct kobj_type slab_ktype = {
4648 .sysfs_ops = &slab_sysfs_ops,
4649 .release = kmem_cache_release
4652 static int uevent_filter(struct kset *kset, struct kobject *kobj)
4654 struct kobj_type *ktype = get_ktype(kobj);
4656 if (ktype == &slab_ktype)
4661 static const struct kset_uevent_ops slab_uevent_ops = {
4662 .filter = uevent_filter,
4665 static struct kset *slab_kset;
4667 #define ID_STR_LENGTH 64
4669 /* Create a unique string id for a slab cache:
4671 * Format :[flags-]size
4673 static char *create_unique_id(struct kmem_cache *s)
4675 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4682 * First flags affecting slabcache operations. We will only
4683 * get here for aliasable slabs so we do not need to support
4684 * too many flags. The flags here must cover all flags that
4685 * are matched during merging to guarantee that the id is
4688 if (s->flags & SLAB_CACHE_DMA)
4690 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4692 if (s->flags & SLAB_DEBUG_FREE)
4694 if (!(s->flags & SLAB_NOTRACK))
4698 p += sprintf(p, "%07d", s->size);
4699 BUG_ON(p > name + ID_STR_LENGTH - 1);
4703 static int sysfs_slab_add(struct kmem_cache *s)
4709 if (slab_state < SYSFS)
4710 /* Defer until later */
4713 unmergeable = slab_unmergeable(s);
4716 * Slabcache can never be merged so we can use the name proper.
4717 * This is typically the case for debug situations. In that
4718 * case we can catch duplicate names easily.
4720 sysfs_remove_link(&slab_kset->kobj, s->name);
4724 * Create a unique name for the slab as a target
4727 name = create_unique_id(s);
4730 s->kobj.kset = slab_kset;
4731 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4733 kobject_put(&s->kobj);
4737 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4739 kobject_del(&s->kobj);
4740 kobject_put(&s->kobj);
4743 kobject_uevent(&s->kobj, KOBJ_ADD);
4745 /* Setup first alias */
4746 sysfs_slab_alias(s, s->name);
4752 static void sysfs_slab_remove(struct kmem_cache *s)
4754 if (slab_state < SYSFS)
4756 * Sysfs has not been setup yet so no need to remove the
4761 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4762 kobject_del(&s->kobj);
4763 kobject_put(&s->kobj);
4767 * Need to buffer aliases during bootup until sysfs becomes
4768 * available lest we lose that information.
4770 struct saved_alias {
4771 struct kmem_cache *s;
4773 struct saved_alias *next;
4776 static struct saved_alias *alias_list;
4778 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4780 struct saved_alias *al;
4782 if (slab_state == SYSFS) {
4784 * If we have a leftover link then remove it.
4786 sysfs_remove_link(&slab_kset->kobj, name);
4787 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
4790 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4796 al->next = alias_list;
4801 static int __init slab_sysfs_init(void)
4803 struct kmem_cache *s;
4806 down_write(&slub_lock);
4808 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
4810 up_write(&slub_lock);
4811 printk(KERN_ERR "Cannot register slab subsystem.\n");
4817 list_for_each_entry(s, &slab_caches, list) {
4818 err = sysfs_slab_add(s);
4820 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4821 " to sysfs\n", s->name);
4824 while (alias_list) {
4825 struct saved_alias *al = alias_list;
4827 alias_list = alias_list->next;
4828 err = sysfs_slab_alias(al->s, al->name);
4830 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4831 " %s to sysfs\n", s->name);
4835 up_write(&slub_lock);
4840 __initcall(slab_sysfs_init);
4841 #endif /* CONFIG_SYSFS */
4844 * The /proc/slabinfo ABI
4846 #ifdef CONFIG_SLABINFO
4847 static void print_slabinfo_header(struct seq_file *m)
4849 seq_puts(m, "slabinfo - version: 2.1\n");
4850 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4851 "<objperslab> <pagesperslab>");
4852 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4853 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4857 static void *s_start(struct seq_file *m, loff_t *pos)
4861 down_read(&slub_lock);
4863 print_slabinfo_header(m);
4865 return seq_list_start(&slab_caches, *pos);
4868 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4870 return seq_list_next(p, &slab_caches, pos);
4873 static void s_stop(struct seq_file *m, void *p)
4875 up_read(&slub_lock);
4878 static int s_show(struct seq_file *m, void *p)
4880 unsigned long nr_partials = 0;
4881 unsigned long nr_slabs = 0;
4882 unsigned long nr_inuse = 0;
4883 unsigned long nr_objs = 0;
4884 unsigned long nr_free = 0;
4885 struct kmem_cache *s;
4888 s = list_entry(p, struct kmem_cache, list);
4890 for_each_online_node(node) {
4891 struct kmem_cache_node *n = get_node(s, node);
4896 nr_partials += n->nr_partial;
4897 nr_slabs += atomic_long_read(&n->nr_slabs);
4898 nr_objs += atomic_long_read(&n->total_objects);
4899 nr_free += count_partial(n, count_free);
4902 nr_inuse = nr_objs - nr_free;
4904 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
4905 nr_objs, s->size, oo_objects(s->oo),
4906 (1 << oo_order(s->oo)));
4907 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4908 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4914 static const struct seq_operations slabinfo_op = {
4921 static int slabinfo_open(struct inode *inode, struct file *file)
4923 return seq_open(file, &slabinfo_op);
4926 static const struct file_operations proc_slabinfo_operations = {
4927 .open = slabinfo_open,
4929 .llseek = seq_lseek,
4930 .release = seq_release,
4933 static int __init slab_proc_init(void)
4935 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
4938 module_init(slab_proc_init);
4939 #endif /* CONFIG_SLABINFO */