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(const 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 *get_freepointer_safe(struct kmem_cache *s, void *object)
268 #ifdef CONFIG_DEBUG_PAGEALLOC
269 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
271 p = get_freepointer(s, object);
276 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
278 *(void **)(object + s->offset) = fp;
281 /* Loop over all objects in a slab */
282 #define for_each_object(__p, __s, __addr, __objects) \
283 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
286 /* Determine object index from a given position */
287 static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
289 return (p - addr) / s->size;
292 static inline size_t slab_ksize(const struct kmem_cache *s)
294 #ifdef CONFIG_SLUB_DEBUG
296 * Debugging requires use of the padding between object
297 * and whatever may come after it.
299 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
304 * If we have the need to store the freelist pointer
305 * back there or track user information then we can
306 * only use the space before that information.
308 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
311 * Else we can use all the padding etc for the allocation
316 static inline int order_objects(int order, unsigned long size, int reserved)
318 return ((PAGE_SIZE << order) - reserved) / size;
321 static inline struct kmem_cache_order_objects oo_make(int order,
322 unsigned long size, int reserved)
324 struct kmem_cache_order_objects x = {
325 (order << OO_SHIFT) + order_objects(order, size, reserved)
331 static inline int oo_order(struct kmem_cache_order_objects x)
333 return x.x >> OO_SHIFT;
336 static inline int oo_objects(struct kmem_cache_order_objects x)
338 return x.x & OO_MASK;
341 #ifdef CONFIG_SLUB_DEBUG
343 * Determine a map of object in use on a page.
345 * Slab lock or node listlock must be held to guarantee that the page does
346 * not vanish from under us.
348 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
351 void *addr = page_address(page);
353 for (p = page->freelist; p; p = get_freepointer(s, p))
354 set_bit(slab_index(p, s, addr), map);
360 #ifdef CONFIG_SLUB_DEBUG_ON
361 static int slub_debug = DEBUG_DEFAULT_FLAGS;
363 static int slub_debug;
366 static char *slub_debug_slabs;
367 static int disable_higher_order_debug;
372 static void print_section(char *text, u8 *addr, unsigned int length)
380 for (i = 0; i < length; i++) {
382 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
385 printk(KERN_CONT " %02x", addr[i]);
387 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
389 printk(KERN_CONT " %s\n", ascii);
396 printk(KERN_CONT " ");
400 printk(KERN_CONT " %s\n", ascii);
404 static struct track *get_track(struct kmem_cache *s, void *object,
405 enum track_item alloc)
410 p = object + s->offset + sizeof(void *);
412 p = object + s->inuse;
417 static void set_track(struct kmem_cache *s, void *object,
418 enum track_item alloc, unsigned long addr)
420 struct track *p = get_track(s, object, alloc);
424 p->cpu = smp_processor_id();
425 p->pid = current->pid;
428 memset(p, 0, sizeof(struct track));
431 static void init_tracking(struct kmem_cache *s, void *object)
433 if (!(s->flags & SLAB_STORE_USER))
436 set_track(s, object, TRACK_FREE, 0UL);
437 set_track(s, object, TRACK_ALLOC, 0UL);
440 static void print_track(const char *s, struct track *t)
445 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
446 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
449 static void print_tracking(struct kmem_cache *s, void *object)
451 if (!(s->flags & SLAB_STORE_USER))
454 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
455 print_track("Freed", get_track(s, object, TRACK_FREE));
458 static void print_page_info(struct page *page)
460 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
461 page, page->objects, page->inuse, page->freelist, page->flags);
465 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
471 vsnprintf(buf, sizeof(buf), fmt, args);
473 printk(KERN_ERR "========================================"
474 "=====================================\n");
475 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
476 printk(KERN_ERR "----------------------------------------"
477 "-------------------------------------\n\n");
480 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
486 vsnprintf(buf, sizeof(buf), fmt, args);
488 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
491 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
493 unsigned int off; /* Offset of last byte */
494 u8 *addr = page_address(page);
496 print_tracking(s, p);
498 print_page_info(page);
500 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
501 p, p - addr, get_freepointer(s, p));
504 print_section("Bytes b4", p - 16, 16);
506 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
508 if (s->flags & SLAB_RED_ZONE)
509 print_section("Redzone", p + s->objsize,
510 s->inuse - s->objsize);
513 off = s->offset + sizeof(void *);
517 if (s->flags & SLAB_STORE_USER)
518 off += 2 * sizeof(struct track);
521 /* Beginning of the filler is the free pointer */
522 print_section("Padding", p + off, s->size - off);
527 static void object_err(struct kmem_cache *s, struct page *page,
528 u8 *object, char *reason)
530 slab_bug(s, "%s", reason);
531 print_trailer(s, page, object);
534 static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
540 vsnprintf(buf, sizeof(buf), fmt, args);
542 slab_bug(s, "%s", buf);
543 print_page_info(page);
547 static void init_object(struct kmem_cache *s, void *object, u8 val)
551 if (s->flags & __OBJECT_POISON) {
552 memset(p, POISON_FREE, s->objsize - 1);
553 p[s->objsize - 1] = POISON_END;
556 if (s->flags & SLAB_RED_ZONE)
557 memset(p + s->objsize, val, s->inuse - s->objsize);
560 static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
563 if (*start != (u8)value)
571 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
572 void *from, void *to)
574 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
575 memset(from, data, to - from);
578 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
579 u8 *object, char *what,
580 u8 *start, unsigned int value, unsigned int bytes)
585 fault = check_bytes(start, value, bytes);
590 while (end > fault && end[-1] == value)
593 slab_bug(s, "%s overwritten", what);
594 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
595 fault, end - 1, fault[0], value);
596 print_trailer(s, page, object);
598 restore_bytes(s, what, value, fault, end);
606 * Bytes of the object to be managed.
607 * If the freepointer may overlay the object then the free
608 * pointer is the first word of the object.
610 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
613 * object + s->objsize
614 * Padding to reach word boundary. This is also used for Redzoning.
615 * Padding is extended by another word if Redzoning is enabled and
618 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
619 * 0xcc (RED_ACTIVE) for objects in use.
622 * Meta data starts here.
624 * A. Free pointer (if we cannot overwrite object on free)
625 * B. Tracking data for SLAB_STORE_USER
626 * C. Padding to reach required alignment boundary or at mininum
627 * one word if debugging is on to be able to detect writes
628 * before the word boundary.
630 * Padding is done using 0x5a (POISON_INUSE)
633 * Nothing is used beyond s->size.
635 * If slabcaches are merged then the objsize and inuse boundaries are mostly
636 * ignored. And therefore no slab options that rely on these boundaries
637 * may be used with merged slabcaches.
640 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
642 unsigned long off = s->inuse; /* The end of info */
645 /* Freepointer is placed after the object. */
646 off += sizeof(void *);
648 if (s->flags & SLAB_STORE_USER)
649 /* We also have user information there */
650 off += 2 * sizeof(struct track);
655 return check_bytes_and_report(s, page, p, "Object padding",
656 p + off, POISON_INUSE, s->size - off);
659 /* Check the pad bytes at the end of a slab page */
660 static int slab_pad_check(struct kmem_cache *s, struct page *page)
668 if (!(s->flags & SLAB_POISON))
671 start = page_address(page);
672 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
673 end = start + length;
674 remainder = length % s->size;
678 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
681 while (end > fault && end[-1] == POISON_INUSE)
684 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
685 print_section("Padding", end - remainder, remainder);
687 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
691 static int check_object(struct kmem_cache *s, struct page *page,
692 void *object, u8 val)
695 u8 *endobject = object + s->objsize;
697 if (s->flags & SLAB_RED_ZONE) {
698 if (!check_bytes_and_report(s, page, object, "Redzone",
699 endobject, val, s->inuse - s->objsize))
702 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
703 check_bytes_and_report(s, page, p, "Alignment padding",
704 endobject, POISON_INUSE, s->inuse - s->objsize);
708 if (s->flags & SLAB_POISON) {
709 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
710 (!check_bytes_and_report(s, page, p, "Poison", p,
711 POISON_FREE, s->objsize - 1) ||
712 !check_bytes_and_report(s, page, p, "Poison",
713 p + s->objsize - 1, POISON_END, 1)))
716 * check_pad_bytes cleans up on its own.
718 check_pad_bytes(s, page, p);
721 if (!s->offset && val == SLUB_RED_ACTIVE)
723 * Object and freepointer overlap. Cannot check
724 * freepointer while object is allocated.
728 /* Check free pointer validity */
729 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
730 object_err(s, page, p, "Freepointer corrupt");
732 * No choice but to zap it and thus lose the remainder
733 * of the free objects in this slab. May cause
734 * another error because the object count is now wrong.
736 set_freepointer(s, p, NULL);
742 static int check_slab(struct kmem_cache *s, struct page *page)
746 VM_BUG_ON(!irqs_disabled());
748 if (!PageSlab(page)) {
749 slab_err(s, page, "Not a valid slab page");
753 maxobj = order_objects(compound_order(page), s->size, s->reserved);
754 if (page->objects > maxobj) {
755 slab_err(s, page, "objects %u > max %u",
756 s->name, page->objects, maxobj);
759 if (page->inuse > page->objects) {
760 slab_err(s, page, "inuse %u > max %u",
761 s->name, page->inuse, page->objects);
764 /* Slab_pad_check fixes things up after itself */
765 slab_pad_check(s, page);
770 * Determine if a certain object on a page is on the freelist. Must hold the
771 * slab lock to guarantee that the chains are in a consistent state.
773 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
776 void *fp = page->freelist;
778 unsigned long max_objects;
780 while (fp && nr <= page->objects) {
783 if (!check_valid_pointer(s, page, fp)) {
785 object_err(s, page, object,
786 "Freechain corrupt");
787 set_freepointer(s, object, NULL);
790 slab_err(s, page, "Freepointer corrupt");
791 page->freelist = NULL;
792 page->inuse = page->objects;
793 slab_fix(s, "Freelist cleared");
799 fp = get_freepointer(s, object);
803 max_objects = order_objects(compound_order(page), s->size, s->reserved);
804 if (max_objects > MAX_OBJS_PER_PAGE)
805 max_objects = MAX_OBJS_PER_PAGE;
807 if (page->objects != max_objects) {
808 slab_err(s, page, "Wrong number of objects. Found %d but "
809 "should be %d", page->objects, max_objects);
810 page->objects = max_objects;
811 slab_fix(s, "Number of objects adjusted.");
813 if (page->inuse != page->objects - nr) {
814 slab_err(s, page, "Wrong object count. Counter is %d but "
815 "counted were %d", page->inuse, page->objects - nr);
816 page->inuse = page->objects - nr;
817 slab_fix(s, "Object count adjusted.");
819 return search == NULL;
822 static void trace(struct kmem_cache *s, struct page *page, void *object,
825 if (s->flags & SLAB_TRACE) {
826 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
828 alloc ? "alloc" : "free",
833 print_section("Object", (void *)object, s->objsize);
840 * Hooks for other subsystems that check memory allocations. In a typical
841 * production configuration these hooks all should produce no code at all.
843 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
845 flags &= gfp_allowed_mask;
846 lockdep_trace_alloc(flags);
847 might_sleep_if(flags & __GFP_WAIT);
849 return should_failslab(s->objsize, flags, s->flags);
852 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
854 flags &= gfp_allowed_mask;
855 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
856 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
859 static inline void slab_free_hook(struct kmem_cache *s, void *x)
861 kmemleak_free_recursive(x, s->flags);
864 * Trouble is that we may no longer disable interupts in the fast path
865 * So in order to make the debug calls that expect irqs to be
866 * disabled we need to disable interrupts temporarily.
868 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
872 local_irq_save(flags);
873 kmemcheck_slab_free(s, x, s->objsize);
874 debug_check_no_locks_freed(x, s->objsize);
875 local_irq_restore(flags);
878 if (!(s->flags & SLAB_DEBUG_OBJECTS))
879 debug_check_no_obj_freed(x, s->objsize);
883 * Tracking of fully allocated slabs for debugging purposes.
885 static void add_full(struct kmem_cache_node *n, struct page *page)
887 spin_lock(&n->list_lock);
888 list_add(&page->lru, &n->full);
889 spin_unlock(&n->list_lock);
892 static void remove_full(struct kmem_cache *s, struct page *page)
894 struct kmem_cache_node *n;
896 if (!(s->flags & SLAB_STORE_USER))
899 n = get_node(s, page_to_nid(page));
901 spin_lock(&n->list_lock);
902 list_del(&page->lru);
903 spin_unlock(&n->list_lock);
906 /* Tracking of the number of slabs for debugging purposes */
907 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
909 struct kmem_cache_node *n = get_node(s, node);
911 return atomic_long_read(&n->nr_slabs);
914 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
916 return atomic_long_read(&n->nr_slabs);
919 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
921 struct kmem_cache_node *n = get_node(s, node);
924 * May be called early in order to allocate a slab for the
925 * kmem_cache_node structure. Solve the chicken-egg
926 * dilemma by deferring the increment of the count during
927 * bootstrap (see early_kmem_cache_node_alloc).
930 atomic_long_inc(&n->nr_slabs);
931 atomic_long_add(objects, &n->total_objects);
934 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
936 struct kmem_cache_node *n = get_node(s, node);
938 atomic_long_dec(&n->nr_slabs);
939 atomic_long_sub(objects, &n->total_objects);
942 /* Object debug checks for alloc/free paths */
943 static void setup_object_debug(struct kmem_cache *s, struct page *page,
946 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
949 init_object(s, object, SLUB_RED_INACTIVE);
950 init_tracking(s, object);
953 static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
954 void *object, unsigned long addr)
956 if (!check_slab(s, page))
959 if (!on_freelist(s, page, object)) {
960 object_err(s, page, object, "Object already allocated");
964 if (!check_valid_pointer(s, page, object)) {
965 object_err(s, page, object, "Freelist Pointer check fails");
969 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
972 /* Success perform special debug activities for allocs */
973 if (s->flags & SLAB_STORE_USER)
974 set_track(s, object, TRACK_ALLOC, addr);
975 trace(s, page, object, 1);
976 init_object(s, object, SLUB_RED_ACTIVE);
980 if (PageSlab(page)) {
982 * If this is a slab page then lets do the best we can
983 * to avoid issues in the future. Marking all objects
984 * as used avoids touching the remaining objects.
986 slab_fix(s, "Marking all objects used");
987 page->inuse = page->objects;
988 page->freelist = NULL;
993 static noinline int free_debug_processing(struct kmem_cache *s,
994 struct page *page, void *object, unsigned long addr)
996 if (!check_slab(s, page))
999 if (!check_valid_pointer(s, page, object)) {
1000 slab_err(s, page, "Invalid object pointer 0x%p", object);
1004 if (on_freelist(s, page, object)) {
1005 object_err(s, page, object, "Object already free");
1009 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1012 if (unlikely(s != page->slab)) {
1013 if (!PageSlab(page)) {
1014 slab_err(s, page, "Attempt to free object(0x%p) "
1015 "outside of slab", object);
1016 } else if (!page->slab) {
1018 "SLUB <none>: no slab for object 0x%p.\n",
1022 object_err(s, page, object,
1023 "page slab pointer corrupt.");
1027 /* Special debug activities for freeing objects */
1028 if (!PageSlubFrozen(page) && !page->freelist)
1029 remove_full(s, page);
1030 if (s->flags & SLAB_STORE_USER)
1031 set_track(s, object, TRACK_FREE, addr);
1032 trace(s, page, object, 0);
1033 init_object(s, object, SLUB_RED_INACTIVE);
1037 slab_fix(s, "Object at 0x%p not freed", object);
1041 static int __init setup_slub_debug(char *str)
1043 slub_debug = DEBUG_DEFAULT_FLAGS;
1044 if (*str++ != '=' || !*str)
1046 * No options specified. Switch on full debugging.
1052 * No options but restriction on slabs. This means full
1053 * debugging for slabs matching a pattern.
1057 if (tolower(*str) == 'o') {
1059 * Avoid enabling debugging on caches if its minimum order
1060 * would increase as a result.
1062 disable_higher_order_debug = 1;
1069 * Switch off all debugging measures.
1074 * Determine which debug features should be switched on
1076 for (; *str && *str != ','; str++) {
1077 switch (tolower(*str)) {
1079 slub_debug |= SLAB_DEBUG_FREE;
1082 slub_debug |= SLAB_RED_ZONE;
1085 slub_debug |= SLAB_POISON;
1088 slub_debug |= SLAB_STORE_USER;
1091 slub_debug |= SLAB_TRACE;
1094 slub_debug |= SLAB_FAILSLAB;
1097 printk(KERN_ERR "slub_debug option '%c' "
1098 "unknown. skipped\n", *str);
1104 slub_debug_slabs = str + 1;
1109 __setup("slub_debug", setup_slub_debug);
1111 static unsigned long kmem_cache_flags(unsigned long objsize,
1112 unsigned long flags, const char *name,
1113 void (*ctor)(void *))
1116 * Enable debugging if selected on the kernel commandline.
1118 if (slub_debug && (!slub_debug_slabs ||
1119 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1120 flags |= slub_debug;
1125 static inline void setup_object_debug(struct kmem_cache *s,
1126 struct page *page, void *object) {}
1128 static inline int alloc_debug_processing(struct kmem_cache *s,
1129 struct page *page, void *object, unsigned long addr) { return 0; }
1131 static inline int free_debug_processing(struct kmem_cache *s,
1132 struct page *page, void *object, unsigned long addr) { return 0; }
1134 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1136 static inline int check_object(struct kmem_cache *s, struct page *page,
1137 void *object, u8 val) { return 1; }
1138 static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
1139 static inline unsigned long kmem_cache_flags(unsigned long objsize,
1140 unsigned long flags, const char *name,
1141 void (*ctor)(void *))
1145 #define slub_debug 0
1147 #define disable_higher_order_debug 0
1149 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1151 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1153 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1155 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1158 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1161 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1164 static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1166 #endif /* CONFIG_SLUB_DEBUG */
1169 * Slab allocation and freeing
1171 static inline struct page *alloc_slab_page(gfp_t flags, int node,
1172 struct kmem_cache_order_objects oo)
1174 int order = oo_order(oo);
1176 flags |= __GFP_NOTRACK;
1178 if (node == NUMA_NO_NODE)
1179 return alloc_pages(flags, order);
1181 return alloc_pages_exact_node(node, flags, order);
1184 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1187 struct kmem_cache_order_objects oo = s->oo;
1190 flags |= s->allocflags;
1193 * Let the initial higher-order allocation fail under memory pressure
1194 * so we fall-back to the minimum order allocation.
1196 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1198 page = alloc_slab_page(alloc_gfp, node, oo);
1199 if (unlikely(!page)) {
1202 * Allocation may have failed due to fragmentation.
1203 * Try a lower order alloc if possible
1205 page = alloc_slab_page(flags, node, oo);
1209 stat(s, ORDER_FALLBACK);
1212 if (kmemcheck_enabled
1213 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
1214 int pages = 1 << oo_order(oo);
1216 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1219 * Objects from caches that have a constructor don't get
1220 * cleared when they're allocated, so we need to do it here.
1223 kmemcheck_mark_uninitialized_pages(page, pages);
1225 kmemcheck_mark_unallocated_pages(page, pages);
1228 page->objects = oo_objects(oo);
1229 mod_zone_page_state(page_zone(page),
1230 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1231 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1237 static void setup_object(struct kmem_cache *s, struct page *page,
1240 setup_object_debug(s, page, object);
1241 if (unlikely(s->ctor))
1245 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1252 BUG_ON(flags & GFP_SLAB_BUG_MASK);
1254 page = allocate_slab(s,
1255 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1259 inc_slabs_node(s, page_to_nid(page), page->objects);
1261 page->flags |= 1 << PG_slab;
1263 start = page_address(page);
1265 if (unlikely(s->flags & SLAB_POISON))
1266 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
1269 for_each_object(p, s, start, page->objects) {
1270 setup_object(s, page, last);
1271 set_freepointer(s, last, p);
1274 setup_object(s, page, last);
1275 set_freepointer(s, last, NULL);
1277 page->freelist = start;
1283 static void __free_slab(struct kmem_cache *s, struct page *page)
1285 int order = compound_order(page);
1286 int pages = 1 << order;
1288 if (kmem_cache_debug(s)) {
1291 slab_pad_check(s, page);
1292 for_each_object(p, s, page_address(page),
1294 check_object(s, page, p, SLUB_RED_INACTIVE);
1297 kmemcheck_free_shadow(page, compound_order(page));
1299 mod_zone_page_state(page_zone(page),
1300 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1301 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1304 __ClearPageSlab(page);
1305 reset_page_mapcount(page);
1306 if (current->reclaim_state)
1307 current->reclaim_state->reclaimed_slab += pages;
1308 __free_pages(page, order);
1311 #define need_reserve_slab_rcu \
1312 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1314 static void rcu_free_slab(struct rcu_head *h)
1318 if (need_reserve_slab_rcu)
1319 page = virt_to_head_page(h);
1321 page = container_of((struct list_head *)h, struct page, lru);
1323 __free_slab(page->slab, page);
1326 static void free_slab(struct kmem_cache *s, struct page *page)
1328 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1329 struct rcu_head *head;
1331 if (need_reserve_slab_rcu) {
1332 int order = compound_order(page);
1333 int offset = (PAGE_SIZE << order) - s->reserved;
1335 VM_BUG_ON(s->reserved != sizeof(*head));
1336 head = page_address(page) + offset;
1339 * RCU free overloads the RCU head over the LRU
1341 head = (void *)&page->lru;
1344 call_rcu(head, rcu_free_slab);
1346 __free_slab(s, page);
1349 static void discard_slab(struct kmem_cache *s, struct page *page)
1351 dec_slabs_node(s, page_to_nid(page), page->objects);
1356 * Per slab locking using the pagelock
1358 static __always_inline void slab_lock(struct page *page)
1360 bit_spin_lock(PG_locked, &page->flags);
1363 static __always_inline void slab_unlock(struct page *page)
1365 __bit_spin_unlock(PG_locked, &page->flags);
1368 static __always_inline int slab_trylock(struct page *page)
1372 rc = bit_spin_trylock(PG_locked, &page->flags);
1377 * Management of partially allocated slabs
1379 static void add_partial(struct kmem_cache_node *n,
1380 struct page *page, int tail)
1382 spin_lock(&n->list_lock);
1385 list_add_tail(&page->lru, &n->partial);
1387 list_add(&page->lru, &n->partial);
1388 spin_unlock(&n->list_lock);
1391 static inline void __remove_partial(struct kmem_cache_node *n,
1394 list_del(&page->lru);
1398 static void remove_partial(struct kmem_cache *s, struct page *page)
1400 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1402 spin_lock(&n->list_lock);
1403 __remove_partial(n, page);
1404 spin_unlock(&n->list_lock);
1408 * Lock slab and remove from the partial list.
1410 * Must hold list_lock.
1412 static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1415 if (slab_trylock(page)) {
1416 __remove_partial(n, page);
1417 __SetPageSlubFrozen(page);
1424 * Try to allocate a partial slab from a specific node.
1426 static struct page *get_partial_node(struct kmem_cache_node *n)
1431 * Racy check. If we mistakenly see no partial slabs then we
1432 * just allocate an empty slab. If we mistakenly try to get a
1433 * partial slab and there is none available then get_partials()
1436 if (!n || !n->nr_partial)
1439 spin_lock(&n->list_lock);
1440 list_for_each_entry(page, &n->partial, lru)
1441 if (lock_and_freeze_slab(n, page))
1445 spin_unlock(&n->list_lock);
1450 * Get a page from somewhere. Search in increasing NUMA distances.
1452 static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1455 struct zonelist *zonelist;
1458 enum zone_type high_zoneidx = gfp_zone(flags);
1462 * The defrag ratio allows a configuration of the tradeoffs between
1463 * inter node defragmentation and node local allocations. A lower
1464 * defrag_ratio increases the tendency to do local allocations
1465 * instead of attempting to obtain partial slabs from other nodes.
1467 * If the defrag_ratio is set to 0 then kmalloc() always
1468 * returns node local objects. If the ratio is higher then kmalloc()
1469 * may return off node objects because partial slabs are obtained
1470 * from other nodes and filled up.
1472 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1473 * defrag_ratio = 1000) then every (well almost) allocation will
1474 * first attempt to defrag slab caches on other nodes. This means
1475 * scanning over all nodes to look for partial slabs which may be
1476 * expensive if we do it every time we are trying to find a slab
1477 * with available objects.
1479 if (!s->remote_node_defrag_ratio ||
1480 get_cycles() % 1024 > s->remote_node_defrag_ratio)
1484 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1485 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1486 struct kmem_cache_node *n;
1488 n = get_node(s, zone_to_nid(zone));
1490 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1491 n->nr_partial > s->min_partial) {
1492 page = get_partial_node(n);
1505 * Get a partial page, lock it and return it.
1507 static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1510 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
1512 page = get_partial_node(get_node(s, searchnode));
1513 if (page || node != NUMA_NO_NODE)
1516 return get_any_partial(s, flags);
1520 * Move a page back to the lists.
1522 * Must be called with the slab lock held.
1524 * On exit the slab lock will have been dropped.
1526 static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
1529 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1531 __ClearPageSlubFrozen(page);
1534 if (page->freelist) {
1535 add_partial(n, page, tail);
1536 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1538 stat(s, DEACTIVATE_FULL);
1539 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
1544 stat(s, DEACTIVATE_EMPTY);
1545 if (n->nr_partial < s->min_partial) {
1547 * Adding an empty slab to the partial slabs in order
1548 * to avoid page allocator overhead. This slab needs
1549 * to come after the other slabs with objects in
1550 * so that the others get filled first. That way the
1551 * size of the partial list stays small.
1553 * kmem_cache_shrink can reclaim any empty slabs from
1556 add_partial(n, page, 1);
1561 discard_slab(s, page);
1566 #ifdef CONFIG_PREEMPT
1568 * Calculate the next globally unique transaction for disambiguiation
1569 * during cmpxchg. The transactions start with the cpu number and are then
1570 * incremented by CONFIG_NR_CPUS.
1572 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1575 * No preemption supported therefore also no need to check for
1581 static inline unsigned long next_tid(unsigned long tid)
1583 return tid + TID_STEP;
1586 static inline unsigned int tid_to_cpu(unsigned long tid)
1588 return tid % TID_STEP;
1591 static inline unsigned long tid_to_event(unsigned long tid)
1593 return tid / TID_STEP;
1596 static inline unsigned int init_tid(int cpu)
1601 static inline void note_cmpxchg_failure(const char *n,
1602 const struct kmem_cache *s, unsigned long tid)
1604 #ifdef SLUB_DEBUG_CMPXCHG
1605 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1607 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1609 #ifdef CONFIG_PREEMPT
1610 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1611 printk("due to cpu change %d -> %d\n",
1612 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1615 if (tid_to_event(tid) != tid_to_event(actual_tid))
1616 printk("due to cpu running other code. Event %ld->%ld\n",
1617 tid_to_event(tid), tid_to_event(actual_tid));
1619 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1620 actual_tid, tid, next_tid(tid));
1622 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
1625 void init_kmem_cache_cpus(struct kmem_cache *s)
1629 for_each_possible_cpu(cpu)
1630 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
1633 * Remove the cpu slab
1635 static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1638 struct page *page = c->page;
1642 stat(s, DEACTIVATE_REMOTE_FREES);
1644 * Merge cpu freelist into slab freelist. Typically we get here
1645 * because both freelists are empty. So this is unlikely
1648 while (unlikely(c->freelist)) {
1651 tail = 0; /* Hot objects. Put the slab first */
1653 /* Retrieve object from cpu_freelist */
1654 object = c->freelist;
1655 c->freelist = get_freepointer(s, c->freelist);
1657 /* And put onto the regular freelist */
1658 set_freepointer(s, object, page->freelist);
1659 page->freelist = object;
1663 c->tid = next_tid(c->tid);
1664 unfreeze_slab(s, page, tail);
1667 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1669 stat(s, CPUSLAB_FLUSH);
1671 deactivate_slab(s, c);
1677 * Called from IPI handler with interrupts disabled.
1679 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
1681 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
1683 if (likely(c && c->page))
1687 static void flush_cpu_slab(void *d)
1689 struct kmem_cache *s = d;
1691 __flush_cpu_slab(s, smp_processor_id());
1694 static void flush_all(struct kmem_cache *s)
1696 on_each_cpu(flush_cpu_slab, s, 1);
1700 * Check if the objects in a per cpu structure fit numa
1701 * locality expectations.
1703 static inline int node_match(struct kmem_cache_cpu *c, int node)
1706 if (node != NUMA_NO_NODE && c->node != node)
1712 static int count_free(struct page *page)
1714 return page->objects - page->inuse;
1717 static unsigned long count_partial(struct kmem_cache_node *n,
1718 int (*get_count)(struct page *))
1720 unsigned long flags;
1721 unsigned long x = 0;
1724 spin_lock_irqsave(&n->list_lock, flags);
1725 list_for_each_entry(page, &n->partial, lru)
1726 x += get_count(page);
1727 spin_unlock_irqrestore(&n->list_lock, flags);
1731 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1733 #ifdef CONFIG_SLUB_DEBUG
1734 return atomic_long_read(&n->total_objects);
1740 static noinline void
1741 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1746 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1748 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1749 "default order: %d, min order: %d\n", s->name, s->objsize,
1750 s->size, oo_order(s->oo), oo_order(s->min));
1752 if (oo_order(s->min) > get_order(s->objsize))
1753 printk(KERN_WARNING " %s debugging increased min order, use "
1754 "slub_debug=O to disable.\n", s->name);
1756 for_each_online_node(node) {
1757 struct kmem_cache_node *n = get_node(s, node);
1758 unsigned long nr_slabs;
1759 unsigned long nr_objs;
1760 unsigned long nr_free;
1765 nr_free = count_partial(n, count_free);
1766 nr_slabs = node_nr_slabs(n);
1767 nr_objs = node_nr_objs(n);
1770 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1771 node, nr_slabs, nr_objs, nr_free);
1776 * Slow path. The lockless freelist is empty or we need to perform
1779 * Interrupts are disabled.
1781 * Processing is still very fast if new objects have been freed to the
1782 * regular freelist. In that case we simply take over the regular freelist
1783 * as the lockless freelist and zap the regular freelist.
1785 * If that is not working then we fall back to the partial lists. We take the
1786 * first element of the freelist as the object to allocate now and move the
1787 * rest of the freelist to the lockless freelist.
1789 * And if we were unable to get a new slab from the partial slab lists then
1790 * we need to allocate a new slab. This is the slowest path since it involves
1791 * a call to the page allocator and the setup of a new slab.
1793 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1794 unsigned long addr, struct kmem_cache_cpu *c)
1798 unsigned long flags;
1800 local_irq_save(flags);
1801 #ifdef CONFIG_PREEMPT
1803 * We may have been preempted and rescheduled on a different
1804 * cpu before disabling interrupts. Need to reload cpu area
1807 c = this_cpu_ptr(s->cpu_slab);
1810 /* We handle __GFP_ZERO in the caller */
1811 gfpflags &= ~__GFP_ZERO;
1818 if (unlikely(!node_match(c, node)))
1821 stat(s, ALLOC_REFILL);
1824 object = page->freelist;
1825 if (unlikely(!object))
1827 if (kmem_cache_debug(s))
1830 c->freelist = get_freepointer(s, object);
1831 page->inuse = page->objects;
1832 page->freelist = NULL;
1836 c->tid = next_tid(c->tid);
1837 local_irq_restore(flags);
1838 stat(s, ALLOC_SLOWPATH);
1842 deactivate_slab(s, c);
1845 page = get_partial(s, gfpflags, node);
1847 stat(s, ALLOC_FROM_PARTIAL);
1849 c->node = page_to_nid(page);
1854 gfpflags &= gfp_allowed_mask;
1855 if (gfpflags & __GFP_WAIT)
1858 page = new_slab(s, gfpflags, node);
1860 if (gfpflags & __GFP_WAIT)
1861 local_irq_disable();
1864 c = __this_cpu_ptr(s->cpu_slab);
1865 stat(s, ALLOC_SLAB);
1870 __SetPageSlubFrozen(page);
1872 goto load_from_page;
1874 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1875 slab_out_of_memory(s, gfpflags, node);
1876 local_irq_restore(flags);
1879 if (!alloc_debug_processing(s, page, object, addr))
1883 page->freelist = get_freepointer(s, object);
1884 c->node = NUMA_NO_NODE;
1889 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1890 * have the fastpath folded into their functions. So no function call
1891 * overhead for requests that can be satisfied on the fastpath.
1893 * The fastpath works by first checking if the lockless freelist can be used.
1894 * If not then __slab_alloc is called for slow processing.
1896 * Otherwise we can simply pick the next object from the lockless free list.
1898 static __always_inline void *slab_alloc(struct kmem_cache *s,
1899 gfp_t gfpflags, int node, unsigned long addr)
1902 struct kmem_cache_cpu *c;
1905 if (slab_pre_alloc_hook(s, gfpflags))
1911 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1912 * enabled. We may switch back and forth between cpus while
1913 * reading from one cpu area. That does not matter as long
1914 * as we end up on the original cpu again when doing the cmpxchg.
1916 c = __this_cpu_ptr(s->cpu_slab);
1919 * The transaction ids are globally unique per cpu and per operation on
1920 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1921 * occurs on the right processor and that there was no operation on the
1922 * linked list in between.
1927 object = c->freelist;
1928 if (unlikely(!object || !node_match(c, node)))
1930 object = __slab_alloc(s, gfpflags, node, addr, c);
1934 * The cmpxchg will only match if there was no additonal
1935 * operation and if we are on the right processor.
1937 * The cmpxchg does the following atomically (without lock semantics!)
1938 * 1. Relocate first pointer to the current per cpu area.
1939 * 2. Verify that tid and freelist have not been changed
1940 * 3. If they were not changed replace tid and freelist
1942 * Since this is without lock semantics the protection is only against
1943 * code executing on this cpu *not* from access by other cpus.
1945 if (unlikely(!this_cpu_cmpxchg_double(
1946 s->cpu_slab->freelist, s->cpu_slab->tid,
1948 get_freepointer_safe(s, object), next_tid(tid)))) {
1950 note_cmpxchg_failure("slab_alloc", s, tid);
1953 stat(s, ALLOC_FASTPATH);
1956 if (unlikely(gfpflags & __GFP_ZERO) && object)
1957 memset(object, 0, s->objsize);
1959 slab_post_alloc_hook(s, gfpflags, object);
1964 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1966 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1968 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
1972 EXPORT_SYMBOL(kmem_cache_alloc);
1974 #ifdef CONFIG_TRACING
1975 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1977 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1978 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1981 EXPORT_SYMBOL(kmem_cache_alloc_trace);
1983 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1985 void *ret = kmalloc_order(size, flags, order);
1986 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1989 EXPORT_SYMBOL(kmalloc_order_trace);
1993 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1995 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1997 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1998 s->objsize, s->size, gfpflags, node);
2002 EXPORT_SYMBOL(kmem_cache_alloc_node);
2004 #ifdef CONFIG_TRACING
2005 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2007 int node, size_t size)
2009 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2011 trace_kmalloc_node(_RET_IP_, ret,
2012 size, s->size, gfpflags, node);
2015 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2020 * Slow patch handling. This may still be called frequently since objects
2021 * have a longer lifetime than the cpu slabs in most processing loads.
2023 * So we still attempt to reduce cache line usage. Just take the slab
2024 * lock and free the item. If there is no additional partial page
2025 * handling required then we can return immediately.
2027 static void __slab_free(struct kmem_cache *s, struct page *page,
2028 void *x, unsigned long addr)
2031 void **object = (void *)x;
2032 unsigned long flags;
2034 local_irq_save(flags);
2036 stat(s, FREE_SLOWPATH);
2038 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2041 prior = page->freelist;
2042 set_freepointer(s, object, prior);
2043 page->freelist = object;
2046 if (unlikely(PageSlubFrozen(page))) {
2047 stat(s, FREE_FROZEN);
2051 if (unlikely(!page->inuse))
2055 * Objects left in the slab. If it was not on the partial list before
2058 if (unlikely(!prior)) {
2059 add_partial(get_node(s, page_to_nid(page)), page, 1);
2060 stat(s, FREE_ADD_PARTIAL);
2065 local_irq_restore(flags);
2071 * Slab still on the partial list.
2073 remove_partial(s, page);
2074 stat(s, FREE_REMOVE_PARTIAL);
2077 local_irq_restore(flags);
2079 discard_slab(s, page);
2083 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2084 * can perform fastpath freeing without additional function calls.
2086 * The fastpath is only possible if we are freeing to the current cpu slab
2087 * of this processor. This typically the case if we have just allocated
2090 * If fastpath is not possible then fall back to __slab_free where we deal
2091 * with all sorts of special processing.
2093 static __always_inline void slab_free(struct kmem_cache *s,
2094 struct page *page, void *x, unsigned long addr)
2096 void **object = (void *)x;
2097 struct kmem_cache_cpu *c;
2100 slab_free_hook(s, x);
2105 * Determine the currently cpus per cpu slab.
2106 * The cpu may change afterward. However that does not matter since
2107 * data is retrieved via this pointer. If we are on the same cpu
2108 * during the cmpxchg then the free will succedd.
2110 c = __this_cpu_ptr(s->cpu_slab);
2115 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
2116 set_freepointer(s, object, c->freelist);
2118 if (unlikely(!this_cpu_cmpxchg_double(
2119 s->cpu_slab->freelist, s->cpu_slab->tid,
2121 object, next_tid(tid)))) {
2123 note_cmpxchg_failure("slab_free", s, tid);
2126 stat(s, FREE_FASTPATH);
2128 __slab_free(s, page, x, addr);
2132 void kmem_cache_free(struct kmem_cache *s, void *x)
2136 page = virt_to_head_page(x);
2138 slab_free(s, page, x, _RET_IP_);
2140 trace_kmem_cache_free(_RET_IP_, x);
2142 EXPORT_SYMBOL(kmem_cache_free);
2145 * Object placement in a slab is made very easy because we always start at
2146 * offset 0. If we tune the size of the object to the alignment then we can
2147 * get the required alignment by putting one properly sized object after
2150 * Notice that the allocation order determines the sizes of the per cpu
2151 * caches. Each processor has always one slab available for allocations.
2152 * Increasing the allocation order reduces the number of times that slabs
2153 * must be moved on and off the partial lists and is therefore a factor in
2158 * Mininum / Maximum order of slab pages. This influences locking overhead
2159 * and slab fragmentation. A higher order reduces the number of partial slabs
2160 * and increases the number of allocations possible without having to
2161 * take the list_lock.
2163 static int slub_min_order;
2164 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
2165 static int slub_min_objects;
2168 * Merge control. If this is set then no merging of slab caches will occur.
2169 * (Could be removed. This was introduced to pacify the merge skeptics.)
2171 static int slub_nomerge;
2174 * Calculate the order of allocation given an slab object size.
2176 * The order of allocation has significant impact on performance and other
2177 * system components. Generally order 0 allocations should be preferred since
2178 * order 0 does not cause fragmentation in the page allocator. Larger objects
2179 * be problematic to put into order 0 slabs because there may be too much
2180 * unused space left. We go to a higher order if more than 1/16th of the slab
2183 * In order to reach satisfactory performance we must ensure that a minimum
2184 * number of objects is in one slab. Otherwise we may generate too much
2185 * activity on the partial lists which requires taking the list_lock. This is
2186 * less a concern for large slabs though which are rarely used.
2188 * slub_max_order specifies the order where we begin to stop considering the
2189 * number of objects in a slab as critical. If we reach slub_max_order then
2190 * we try to keep the page order as low as possible. So we accept more waste
2191 * of space in favor of a small page order.
2193 * Higher order allocations also allow the placement of more objects in a
2194 * slab and thereby reduce object handling overhead. If the user has
2195 * requested a higher mininum order then we start with that one instead of
2196 * the smallest order which will fit the object.
2198 static inline int slab_order(int size, int min_objects,
2199 int max_order, int fract_leftover, int reserved)
2203 int min_order = slub_min_order;
2205 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
2206 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
2208 for (order = max(min_order,
2209 fls(min_objects * size - 1) - PAGE_SHIFT);
2210 order <= max_order; order++) {
2212 unsigned long slab_size = PAGE_SIZE << order;
2214 if (slab_size < min_objects * size + reserved)
2217 rem = (slab_size - reserved) % size;
2219 if (rem <= slab_size / fract_leftover)
2227 static inline int calculate_order(int size, int reserved)
2235 * Attempt to find best configuration for a slab. This
2236 * works by first attempting to generate a layout with
2237 * the best configuration and backing off gradually.
2239 * First we reduce the acceptable waste in a slab. Then
2240 * we reduce the minimum objects required in a slab.
2242 min_objects = slub_min_objects;
2244 min_objects = 4 * (fls(nr_cpu_ids) + 1);
2245 max_objects = order_objects(slub_max_order, size, reserved);
2246 min_objects = min(min_objects, max_objects);
2248 while (min_objects > 1) {
2250 while (fraction >= 4) {
2251 order = slab_order(size, min_objects,
2252 slub_max_order, fraction, reserved);
2253 if (order <= slub_max_order)
2261 * We were unable to place multiple objects in a slab. Now
2262 * lets see if we can place a single object there.
2264 order = slab_order(size, 1, slub_max_order, 1, reserved);
2265 if (order <= slub_max_order)
2269 * Doh this slab cannot be placed using slub_max_order.
2271 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
2272 if (order < MAX_ORDER)
2278 * Figure out what the alignment of the objects will be.
2280 static unsigned long calculate_alignment(unsigned long flags,
2281 unsigned long align, unsigned long size)
2284 * If the user wants hardware cache aligned objects then follow that
2285 * suggestion if the object is sufficiently large.
2287 * The hardware cache alignment cannot override the specified
2288 * alignment though. If that is greater then use it.
2290 if (flags & SLAB_HWCACHE_ALIGN) {
2291 unsigned long ralign = cache_line_size();
2292 while (size <= ralign / 2)
2294 align = max(align, ralign);
2297 if (align < ARCH_SLAB_MINALIGN)
2298 align = ARCH_SLAB_MINALIGN;
2300 return ALIGN(align, sizeof(void *));
2304 init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
2307 spin_lock_init(&n->list_lock);
2308 INIT_LIST_HEAD(&n->partial);
2309 #ifdef CONFIG_SLUB_DEBUG
2310 atomic_long_set(&n->nr_slabs, 0);
2311 atomic_long_set(&n->total_objects, 0);
2312 INIT_LIST_HEAD(&n->full);
2316 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
2318 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2319 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
2321 #ifdef CONFIG_CMPXCHG_LOCAL
2323 * Must align to double word boundary for the double cmpxchg instructions
2326 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
2328 /* Regular alignment is sufficient */
2329 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2335 init_kmem_cache_cpus(s);
2340 static struct kmem_cache *kmem_cache_node;
2343 * No kmalloc_node yet so do it by hand. We know that this is the first
2344 * slab on the node for this slabcache. There are no concurrent accesses
2347 * Note that this function only works on the kmalloc_node_cache
2348 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2349 * memory on a fresh node that has no slab structures yet.
2351 static void early_kmem_cache_node_alloc(int node)
2354 struct kmem_cache_node *n;
2355 unsigned long flags;
2357 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
2359 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
2362 if (page_to_nid(page) != node) {
2363 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2365 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2366 "in order to be able to continue\n");
2371 page->freelist = get_freepointer(kmem_cache_node, n);
2373 kmem_cache_node->node[node] = n;
2374 #ifdef CONFIG_SLUB_DEBUG
2375 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
2376 init_tracking(kmem_cache_node, n);
2378 init_kmem_cache_node(n, kmem_cache_node);
2379 inc_slabs_node(kmem_cache_node, node, page->objects);
2382 * lockdep requires consistent irq usage for each lock
2383 * so even though there cannot be a race this early in
2384 * the boot sequence, we still disable irqs.
2386 local_irq_save(flags);
2387 add_partial(n, page, 0);
2388 local_irq_restore(flags);
2391 static void free_kmem_cache_nodes(struct kmem_cache *s)
2395 for_each_node_state(node, N_NORMAL_MEMORY) {
2396 struct kmem_cache_node *n = s->node[node];
2399 kmem_cache_free(kmem_cache_node, n);
2401 s->node[node] = NULL;
2405 static int init_kmem_cache_nodes(struct kmem_cache *s)
2409 for_each_node_state(node, N_NORMAL_MEMORY) {
2410 struct kmem_cache_node *n;
2412 if (slab_state == DOWN) {
2413 early_kmem_cache_node_alloc(node);
2416 n = kmem_cache_alloc_node(kmem_cache_node,
2420 free_kmem_cache_nodes(s);
2425 init_kmem_cache_node(n, s);
2430 static void set_min_partial(struct kmem_cache *s, unsigned long min)
2432 if (min < MIN_PARTIAL)
2434 else if (min > MAX_PARTIAL)
2436 s->min_partial = min;
2440 * calculate_sizes() determines the order and the distribution of data within
2443 static int calculate_sizes(struct kmem_cache *s, int forced_order)
2445 unsigned long flags = s->flags;
2446 unsigned long size = s->objsize;
2447 unsigned long align = s->align;
2451 * Round up object size to the next word boundary. We can only
2452 * place the free pointer at word boundaries and this determines
2453 * the possible location of the free pointer.
2455 size = ALIGN(size, sizeof(void *));
2457 #ifdef CONFIG_SLUB_DEBUG
2459 * Determine if we can poison the object itself. If the user of
2460 * the slab may touch the object after free or before allocation
2461 * then we should never poison the object itself.
2463 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
2465 s->flags |= __OBJECT_POISON;
2467 s->flags &= ~__OBJECT_POISON;
2471 * If we are Redzoning then check if there is some space between the
2472 * end of the object and the free pointer. If not then add an
2473 * additional word to have some bytes to store Redzone information.
2475 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2476 size += sizeof(void *);
2480 * With that we have determined the number of bytes in actual use
2481 * by the object. This is the potential offset to the free pointer.
2485 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
2488 * Relocate free pointer after the object if it is not
2489 * permitted to overwrite the first word of the object on
2492 * This is the case if we do RCU, have a constructor or
2493 * destructor or are poisoning the objects.
2496 size += sizeof(void *);
2499 #ifdef CONFIG_SLUB_DEBUG
2500 if (flags & SLAB_STORE_USER)
2502 * Need to store information about allocs and frees after
2505 size += 2 * sizeof(struct track);
2507 if (flags & SLAB_RED_ZONE)
2509 * Add some empty padding so that we can catch
2510 * overwrites from earlier objects rather than let
2511 * tracking information or the free pointer be
2512 * corrupted if a user writes before the start
2515 size += sizeof(void *);
2519 * Determine the alignment based on various parameters that the
2520 * user specified and the dynamic determination of cache line size
2523 align = calculate_alignment(flags, align, s->objsize);
2527 * SLUB stores one object immediately after another beginning from
2528 * offset 0. In order to align the objects we have to simply size
2529 * each object to conform to the alignment.
2531 size = ALIGN(size, align);
2533 if (forced_order >= 0)
2534 order = forced_order;
2536 order = calculate_order(size, s->reserved);
2543 s->allocflags |= __GFP_COMP;
2545 if (s->flags & SLAB_CACHE_DMA)
2546 s->allocflags |= SLUB_DMA;
2548 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2549 s->allocflags |= __GFP_RECLAIMABLE;
2552 * Determine the number of objects per slab
2554 s->oo = oo_make(order, size, s->reserved);
2555 s->min = oo_make(get_order(size), size, s->reserved);
2556 if (oo_objects(s->oo) > oo_objects(s->max))
2559 return !!oo_objects(s->oo);
2563 static int kmem_cache_open(struct kmem_cache *s,
2564 const char *name, size_t size,
2565 size_t align, unsigned long flags,
2566 void (*ctor)(void *))
2568 memset(s, 0, kmem_size);
2573 s->flags = kmem_cache_flags(size, flags, name, ctor);
2576 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2577 s->reserved = sizeof(struct rcu_head);
2579 if (!calculate_sizes(s, -1))
2581 if (disable_higher_order_debug) {
2583 * Disable debugging flags that store metadata if the min slab
2586 if (get_order(s->size) > get_order(s->objsize)) {
2587 s->flags &= ~DEBUG_METADATA_FLAGS;
2589 if (!calculate_sizes(s, -1))
2595 * The larger the object size is, the more pages we want on the partial
2596 * list to avoid pounding the page allocator excessively.
2598 set_min_partial(s, ilog2(s->size));
2601 s->remote_node_defrag_ratio = 1000;
2603 if (!init_kmem_cache_nodes(s))
2606 if (alloc_kmem_cache_cpus(s))
2609 free_kmem_cache_nodes(s);
2611 if (flags & SLAB_PANIC)
2612 panic("Cannot create slab %s size=%lu realsize=%u "
2613 "order=%u offset=%u flags=%lx\n",
2614 s->name, (unsigned long)size, s->size, oo_order(s->oo),
2620 * Determine the size of a slab object
2622 unsigned int kmem_cache_size(struct kmem_cache *s)
2626 EXPORT_SYMBOL(kmem_cache_size);
2628 static void list_slab_objects(struct kmem_cache *s, struct page *page,
2631 #ifdef CONFIG_SLUB_DEBUG
2632 void *addr = page_address(page);
2634 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2635 sizeof(long), GFP_ATOMIC);
2638 slab_err(s, page, "%s", text);
2641 get_map(s, page, map);
2642 for_each_object(p, s, addr, page->objects) {
2644 if (!test_bit(slab_index(p, s, addr), map)) {
2645 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2647 print_tracking(s, p);
2656 * Attempt to free all partial slabs on a node.
2658 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
2660 unsigned long flags;
2661 struct page *page, *h;
2663 spin_lock_irqsave(&n->list_lock, flags);
2664 list_for_each_entry_safe(page, h, &n->partial, lru) {
2666 __remove_partial(n, page);
2667 discard_slab(s, page);
2669 list_slab_objects(s, page,
2670 "Objects remaining on kmem_cache_close()");
2673 spin_unlock_irqrestore(&n->list_lock, flags);
2677 * Release all resources used by a slab cache.
2679 static inline int kmem_cache_close(struct kmem_cache *s)
2684 free_percpu(s->cpu_slab);
2685 /* Attempt to free all objects */
2686 for_each_node_state(node, N_NORMAL_MEMORY) {
2687 struct kmem_cache_node *n = get_node(s, node);
2690 if (n->nr_partial || slabs_node(s, node))
2693 free_kmem_cache_nodes(s);
2698 * Close a cache and release the kmem_cache structure
2699 * (must be used for caches created using kmem_cache_create)
2701 void kmem_cache_destroy(struct kmem_cache *s)
2703 down_write(&slub_lock);
2707 if (kmem_cache_close(s)) {
2708 printk(KERN_ERR "SLUB %s: %s called for cache that "
2709 "still has objects.\n", s->name, __func__);
2712 if (s->flags & SLAB_DESTROY_BY_RCU)
2714 sysfs_slab_remove(s);
2716 up_write(&slub_lock);
2718 EXPORT_SYMBOL(kmem_cache_destroy);
2720 /********************************************************************
2722 *******************************************************************/
2724 struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
2725 EXPORT_SYMBOL(kmalloc_caches);
2727 static struct kmem_cache *kmem_cache;
2729 #ifdef CONFIG_ZONE_DMA
2730 static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
2733 static int __init setup_slub_min_order(char *str)
2735 get_option(&str, &slub_min_order);
2740 __setup("slub_min_order=", setup_slub_min_order);
2742 static int __init setup_slub_max_order(char *str)
2744 get_option(&str, &slub_max_order);
2745 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
2750 __setup("slub_max_order=", setup_slub_max_order);
2752 static int __init setup_slub_min_objects(char *str)
2754 get_option(&str, &slub_min_objects);
2759 __setup("slub_min_objects=", setup_slub_min_objects);
2761 static int __init setup_slub_nomerge(char *str)
2767 __setup("slub_nomerge", setup_slub_nomerge);
2769 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2770 int size, unsigned int flags)
2772 struct kmem_cache *s;
2774 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2777 * This function is called with IRQs disabled during early-boot on
2778 * single CPU so there's no need to take slub_lock here.
2780 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
2784 list_add(&s->list, &slab_caches);
2788 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2793 * Conversion table for small slabs sizes / 8 to the index in the
2794 * kmalloc array. This is necessary for slabs < 192 since we have non power
2795 * of two cache sizes there. The size of larger slabs can be determined using
2798 static s8 size_index[24] = {
2825 static inline int size_index_elem(size_t bytes)
2827 return (bytes - 1) / 8;
2830 static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2836 return ZERO_SIZE_PTR;
2838 index = size_index[size_index_elem(size)];
2840 index = fls(size - 1);
2842 #ifdef CONFIG_ZONE_DMA
2843 if (unlikely((flags & SLUB_DMA)))
2844 return kmalloc_dma_caches[index];
2847 return kmalloc_caches[index];
2850 void *__kmalloc(size_t size, gfp_t flags)
2852 struct kmem_cache *s;
2855 if (unlikely(size > SLUB_MAX_SIZE))
2856 return kmalloc_large(size, flags);
2858 s = get_slab(size, flags);
2860 if (unlikely(ZERO_OR_NULL_PTR(s)))
2863 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
2865 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
2869 EXPORT_SYMBOL(__kmalloc);
2872 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2877 flags |= __GFP_COMP | __GFP_NOTRACK;
2878 page = alloc_pages_node(node, flags, get_order(size));
2880 ptr = page_address(page);
2882 kmemleak_alloc(ptr, size, 1, flags);
2886 void *__kmalloc_node(size_t size, gfp_t flags, int node)
2888 struct kmem_cache *s;
2891 if (unlikely(size > SLUB_MAX_SIZE)) {
2892 ret = kmalloc_large_node(size, flags, node);
2894 trace_kmalloc_node(_RET_IP_, ret,
2895 size, PAGE_SIZE << get_order(size),
2901 s = get_slab(size, flags);
2903 if (unlikely(ZERO_OR_NULL_PTR(s)))
2906 ret = slab_alloc(s, flags, node, _RET_IP_);
2908 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
2912 EXPORT_SYMBOL(__kmalloc_node);
2915 size_t ksize(const void *object)
2919 if (unlikely(object == ZERO_SIZE_PTR))
2922 page = virt_to_head_page(object);
2924 if (unlikely(!PageSlab(page))) {
2925 WARN_ON(!PageCompound(page));
2926 return PAGE_SIZE << compound_order(page);
2929 return slab_ksize(page->slab);
2931 EXPORT_SYMBOL(ksize);
2933 void kfree(const void *x)
2936 void *object = (void *)x;
2938 trace_kfree(_RET_IP_, x);
2940 if (unlikely(ZERO_OR_NULL_PTR(x)))
2943 page = virt_to_head_page(x);
2944 if (unlikely(!PageSlab(page))) {
2945 BUG_ON(!PageCompound(page));
2950 slab_free(page->slab, page, object, _RET_IP_);
2952 EXPORT_SYMBOL(kfree);
2955 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2956 * the remaining slabs by the number of items in use. The slabs with the
2957 * most items in use come first. New allocations will then fill those up
2958 * and thus they can be removed from the partial lists.
2960 * The slabs with the least items are placed last. This results in them
2961 * being allocated from last increasing the chance that the last objects
2962 * are freed in them.
2964 int kmem_cache_shrink(struct kmem_cache *s)
2968 struct kmem_cache_node *n;
2971 int objects = oo_objects(s->max);
2972 struct list_head *slabs_by_inuse =
2973 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
2974 unsigned long flags;
2976 if (!slabs_by_inuse)
2980 for_each_node_state(node, N_NORMAL_MEMORY) {
2981 n = get_node(s, node);
2986 for (i = 0; i < objects; i++)
2987 INIT_LIST_HEAD(slabs_by_inuse + i);
2989 spin_lock_irqsave(&n->list_lock, flags);
2992 * Build lists indexed by the items in use in each slab.
2994 * Note that concurrent frees may occur while we hold the
2995 * list_lock. page->inuse here is the upper limit.
2997 list_for_each_entry_safe(page, t, &n->partial, lru) {
2998 if (!page->inuse && slab_trylock(page)) {
3000 * Must hold slab lock here because slab_free
3001 * may have freed the last object and be
3002 * waiting to release the slab.
3004 __remove_partial(n, page);
3006 discard_slab(s, page);
3008 list_move(&page->lru,
3009 slabs_by_inuse + page->inuse);
3014 * Rebuild the partial list with the slabs filled up most
3015 * first and the least used slabs at the end.
3017 for (i = objects - 1; i >= 0; i--)
3018 list_splice(slabs_by_inuse + i, n->partial.prev);
3020 spin_unlock_irqrestore(&n->list_lock, flags);
3023 kfree(slabs_by_inuse);
3026 EXPORT_SYMBOL(kmem_cache_shrink);
3028 #if defined(CONFIG_MEMORY_HOTPLUG)
3029 static int slab_mem_going_offline_callback(void *arg)
3031 struct kmem_cache *s;
3033 down_read(&slub_lock);
3034 list_for_each_entry(s, &slab_caches, list)
3035 kmem_cache_shrink(s);
3036 up_read(&slub_lock);
3041 static void slab_mem_offline_callback(void *arg)
3043 struct kmem_cache_node *n;
3044 struct kmem_cache *s;
3045 struct memory_notify *marg = arg;
3048 offline_node = marg->status_change_nid;
3051 * If the node still has available memory. we need kmem_cache_node
3054 if (offline_node < 0)
3057 down_read(&slub_lock);
3058 list_for_each_entry(s, &slab_caches, list) {
3059 n = get_node(s, offline_node);
3062 * if n->nr_slabs > 0, slabs still exist on the node
3063 * that is going down. We were unable to free them,
3064 * and offline_pages() function shouldn't call this
3065 * callback. So, we must fail.
3067 BUG_ON(slabs_node(s, offline_node));
3069 s->node[offline_node] = NULL;
3070 kmem_cache_free(kmem_cache_node, n);
3073 up_read(&slub_lock);
3076 static int slab_mem_going_online_callback(void *arg)
3078 struct kmem_cache_node *n;
3079 struct kmem_cache *s;
3080 struct memory_notify *marg = arg;
3081 int nid = marg->status_change_nid;
3085 * If the node's memory is already available, then kmem_cache_node is
3086 * already created. Nothing to do.
3092 * We are bringing a node online. No memory is available yet. We must
3093 * allocate a kmem_cache_node structure in order to bring the node
3096 down_read(&slub_lock);
3097 list_for_each_entry(s, &slab_caches, list) {
3099 * XXX: kmem_cache_alloc_node will fallback to other nodes
3100 * since memory is not yet available from the node that
3103 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
3108 init_kmem_cache_node(n, s);
3112 up_read(&slub_lock);
3116 static int slab_memory_callback(struct notifier_block *self,
3117 unsigned long action, void *arg)
3122 case MEM_GOING_ONLINE:
3123 ret = slab_mem_going_online_callback(arg);
3125 case MEM_GOING_OFFLINE:
3126 ret = slab_mem_going_offline_callback(arg);
3129 case MEM_CANCEL_ONLINE:
3130 slab_mem_offline_callback(arg);
3133 case MEM_CANCEL_OFFLINE:
3137 ret = notifier_from_errno(ret);
3143 #endif /* CONFIG_MEMORY_HOTPLUG */
3145 /********************************************************************
3146 * Basic setup of slabs
3147 *******************************************************************/
3150 * Used for early kmem_cache structures that were allocated using
3151 * the page allocator
3154 static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3158 list_add(&s->list, &slab_caches);
3161 for_each_node_state(node, N_NORMAL_MEMORY) {
3162 struct kmem_cache_node *n = get_node(s, node);
3166 list_for_each_entry(p, &n->partial, lru)
3169 #ifdef CONFIG_SLUB_DEBUG
3170 list_for_each_entry(p, &n->full, lru)
3177 void __init kmem_cache_init(void)
3181 struct kmem_cache *temp_kmem_cache;
3183 struct kmem_cache *temp_kmem_cache_node;
3184 unsigned long kmalloc_size;
3186 kmem_size = offsetof(struct kmem_cache, node) +
3187 nr_node_ids * sizeof(struct kmem_cache_node *);
3189 /* Allocate two kmem_caches from the page allocator */
3190 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3191 order = get_order(2 * kmalloc_size);
3192 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3195 * Must first have the slab cache available for the allocations of the
3196 * struct kmem_cache_node's. There is special bootstrap code in
3197 * kmem_cache_open for slab_state == DOWN.
3199 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3201 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3202 sizeof(struct kmem_cache_node),
3203 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3205 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
3207 /* Able to allocate the per node structures */
3208 slab_state = PARTIAL;
3210 temp_kmem_cache = kmem_cache;
3211 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3212 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3213 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3214 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
3217 * Allocate kmem_cache_node properly from the kmem_cache slab.
3218 * kmem_cache_node is separately allocated so no need to
3219 * update any list pointers.
3221 temp_kmem_cache_node = kmem_cache_node;
3223 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3224 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3226 kmem_cache_bootstrap_fixup(kmem_cache_node);
3229 kmem_cache_bootstrap_fixup(kmem_cache);
3231 /* Free temporary boot structure */
3232 free_pages((unsigned long)temp_kmem_cache, order);
3234 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3237 * Patch up the size_index table if we have strange large alignment
3238 * requirements for the kmalloc array. This is only the case for
3239 * MIPS it seems. The standard arches will not generate any code here.
3241 * Largest permitted alignment is 256 bytes due to the way we
3242 * handle the index determination for the smaller caches.
3244 * Make sure that nothing crazy happens if someone starts tinkering
3245 * around with ARCH_KMALLOC_MINALIGN
3247 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3248 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3250 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3251 int elem = size_index_elem(i);
3252 if (elem >= ARRAY_SIZE(size_index))
3254 size_index[elem] = KMALLOC_SHIFT_LOW;
3257 if (KMALLOC_MIN_SIZE == 64) {
3259 * The 96 byte size cache is not used if the alignment
3262 for (i = 64 + 8; i <= 96; i += 8)
3263 size_index[size_index_elem(i)] = 7;
3264 } else if (KMALLOC_MIN_SIZE == 128) {
3266 * The 192 byte sized cache is not used if the alignment
3267 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3270 for (i = 128 + 8; i <= 192; i += 8)
3271 size_index[size_index_elem(i)] = 8;
3274 /* Caches that are not of the two-to-the-power-of size */
3275 if (KMALLOC_MIN_SIZE <= 32) {
3276 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3280 if (KMALLOC_MIN_SIZE <= 64) {
3281 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3285 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3286 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3292 /* Provide the correct kmalloc names now that the caches are up */
3293 if (KMALLOC_MIN_SIZE <= 32) {
3294 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3295 BUG_ON(!kmalloc_caches[1]->name);
3298 if (KMALLOC_MIN_SIZE <= 64) {
3299 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3300 BUG_ON(!kmalloc_caches[2]->name);
3303 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3304 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3307 kmalloc_caches[i]->name = s;
3311 register_cpu_notifier(&slab_notifier);
3314 #ifdef CONFIG_ZONE_DMA
3315 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3316 struct kmem_cache *s = kmalloc_caches[i];
3319 char *name = kasprintf(GFP_NOWAIT,
3320 "dma-kmalloc-%d", s->objsize);
3323 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3324 s->objsize, SLAB_CACHE_DMA);
3329 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3330 " CPUs=%d, Nodes=%d\n",
3331 caches, cache_line_size(),
3332 slub_min_order, slub_max_order, slub_min_objects,
3333 nr_cpu_ids, nr_node_ids);
3336 void __init kmem_cache_init_late(void)
3341 * Find a mergeable slab cache
3343 static int slab_unmergeable(struct kmem_cache *s)
3345 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3352 * We may have set a slab to be unmergeable during bootstrap.
3354 if (s->refcount < 0)
3360 static struct kmem_cache *find_mergeable(size_t size,
3361 size_t align, unsigned long flags, const char *name,
3362 void (*ctor)(void *))
3364 struct kmem_cache *s;
3366 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3372 size = ALIGN(size, sizeof(void *));
3373 align = calculate_alignment(flags, align, size);
3374 size = ALIGN(size, align);
3375 flags = kmem_cache_flags(size, flags, name, NULL);
3377 list_for_each_entry(s, &slab_caches, list) {
3378 if (slab_unmergeable(s))
3384 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
3387 * Check if alignment is compatible.
3388 * Courtesy of Adrian Drzewiecki
3390 if ((s->size & ~(align - 1)) != s->size)
3393 if (s->size - size >= sizeof(void *))
3401 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
3402 size_t align, unsigned long flags, void (*ctor)(void *))
3404 struct kmem_cache *s;
3410 down_write(&slub_lock);
3411 s = find_mergeable(size, align, flags, name, ctor);
3415 * Adjust the object sizes so that we clear
3416 * the complete object on kzalloc.
3418 s->objsize = max(s->objsize, (int)size);
3419 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
3421 if (sysfs_slab_alias(s, name)) {
3425 up_write(&slub_lock);
3429 n = kstrdup(name, GFP_KERNEL);
3433 s = kmalloc(kmem_size, GFP_KERNEL);
3435 if (kmem_cache_open(s, n,
3436 size, align, flags, ctor)) {
3437 list_add(&s->list, &slab_caches);
3438 if (sysfs_slab_add(s)) {
3444 up_write(&slub_lock);
3451 up_write(&slub_lock);
3453 if (flags & SLAB_PANIC)
3454 panic("Cannot create slabcache %s\n", name);
3459 EXPORT_SYMBOL(kmem_cache_create);
3463 * Use the cpu notifier to insure that the cpu slabs are flushed when
3466 static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3467 unsigned long action, void *hcpu)
3469 long cpu = (long)hcpu;
3470 struct kmem_cache *s;
3471 unsigned long flags;
3474 case CPU_UP_CANCELED:
3475 case CPU_UP_CANCELED_FROZEN:
3477 case CPU_DEAD_FROZEN:
3478 down_read(&slub_lock);
3479 list_for_each_entry(s, &slab_caches, list) {
3480 local_irq_save(flags);
3481 __flush_cpu_slab(s, cpu);
3482 local_irq_restore(flags);
3484 up_read(&slub_lock);
3492 static struct notifier_block __cpuinitdata slab_notifier = {
3493 .notifier_call = slab_cpuup_callback
3498 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
3500 struct kmem_cache *s;
3503 if (unlikely(size > SLUB_MAX_SIZE))
3504 return kmalloc_large(size, gfpflags);
3506 s = get_slab(size, gfpflags);
3508 if (unlikely(ZERO_OR_NULL_PTR(s)))
3511 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
3513 /* Honor the call site pointer we recieved. */
3514 trace_kmalloc(caller, ret, size, s->size, gfpflags);
3520 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
3521 int node, unsigned long caller)
3523 struct kmem_cache *s;
3526 if (unlikely(size > SLUB_MAX_SIZE)) {
3527 ret = kmalloc_large_node(size, gfpflags, node);
3529 trace_kmalloc_node(caller, ret,
3530 size, PAGE_SIZE << get_order(size),
3536 s = get_slab(size, gfpflags);
3538 if (unlikely(ZERO_OR_NULL_PTR(s)))
3541 ret = slab_alloc(s, gfpflags, node, caller);
3543 /* Honor the call site pointer we recieved. */
3544 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
3551 static int count_inuse(struct page *page)
3556 static int count_total(struct page *page)
3558 return page->objects;
3562 #ifdef CONFIG_SLUB_DEBUG
3563 static int validate_slab(struct kmem_cache *s, struct page *page,
3567 void *addr = page_address(page);
3569 if (!check_slab(s, page) ||
3570 !on_freelist(s, page, NULL))
3573 /* Now we know that a valid freelist exists */
3574 bitmap_zero(map, page->objects);
3576 get_map(s, page, map);
3577 for_each_object(p, s, addr, page->objects) {
3578 if (test_bit(slab_index(p, s, addr), map))
3579 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3583 for_each_object(p, s, addr, page->objects)
3584 if (!test_bit(slab_index(p, s, addr), map))
3585 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
3590 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3593 if (slab_trylock(page)) {
3594 validate_slab(s, page, map);
3597 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3601 static int validate_slab_node(struct kmem_cache *s,
3602 struct kmem_cache_node *n, unsigned long *map)
3604 unsigned long count = 0;
3606 unsigned long flags;
3608 spin_lock_irqsave(&n->list_lock, flags);
3610 list_for_each_entry(page, &n->partial, lru) {
3611 validate_slab_slab(s, page, map);
3614 if (count != n->nr_partial)
3615 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3616 "counter=%ld\n", s->name, count, n->nr_partial);
3618 if (!(s->flags & SLAB_STORE_USER))
3621 list_for_each_entry(page, &n->full, lru) {
3622 validate_slab_slab(s, page, map);
3625 if (count != atomic_long_read(&n->nr_slabs))
3626 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3627 "counter=%ld\n", s->name, count,
3628 atomic_long_read(&n->nr_slabs));
3631 spin_unlock_irqrestore(&n->list_lock, flags);
3635 static long validate_slab_cache(struct kmem_cache *s)
3638 unsigned long count = 0;
3639 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3640 sizeof(unsigned long), GFP_KERNEL);
3646 for_each_node_state(node, N_NORMAL_MEMORY) {
3647 struct kmem_cache_node *n = get_node(s, node);
3649 count += validate_slab_node(s, n, map);
3655 * Generate lists of code addresses where slabcache objects are allocated
3660 unsigned long count;
3667 DECLARE_BITMAP(cpus, NR_CPUS);
3673 unsigned long count;
3674 struct location *loc;
3677 static void free_loc_track(struct loc_track *t)
3680 free_pages((unsigned long)t->loc,
3681 get_order(sizeof(struct location) * t->max));
3684 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
3689 order = get_order(sizeof(struct location) * max);
3691 l = (void *)__get_free_pages(flags, order);
3696 memcpy(l, t->loc, sizeof(struct location) * t->count);
3704 static int add_location(struct loc_track *t, struct kmem_cache *s,
3705 const struct track *track)
3707 long start, end, pos;
3709 unsigned long caddr;
3710 unsigned long age = jiffies - track->when;
3716 pos = start + (end - start + 1) / 2;
3719 * There is nothing at "end". If we end up there
3720 * we need to add something to before end.
3725 caddr = t->loc[pos].addr;
3726 if (track->addr == caddr) {
3732 if (age < l->min_time)
3734 if (age > l->max_time)
3737 if (track->pid < l->min_pid)
3738 l->min_pid = track->pid;
3739 if (track->pid > l->max_pid)
3740 l->max_pid = track->pid;
3742 cpumask_set_cpu(track->cpu,
3743 to_cpumask(l->cpus));
3745 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3749 if (track->addr < caddr)
3756 * Not found. Insert new tracking element.
3758 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
3764 (t->count - pos) * sizeof(struct location));
3767 l->addr = track->addr;
3771 l->min_pid = track->pid;
3772 l->max_pid = track->pid;
3773 cpumask_clear(to_cpumask(l->cpus));
3774 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
3775 nodes_clear(l->nodes);
3776 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3780 static void process_slab(struct loc_track *t, struct kmem_cache *s,
3781 struct page *page, enum track_item alloc,
3784 void *addr = page_address(page);
3787 bitmap_zero(map, page->objects);
3788 get_map(s, page, map);
3790 for_each_object(p, s, addr, page->objects)
3791 if (!test_bit(slab_index(p, s, addr), map))
3792 add_location(t, s, get_track(s, p, alloc));
3795 static int list_locations(struct kmem_cache *s, char *buf,
3796 enum track_item alloc)
3800 struct loc_track t = { 0, 0, NULL };
3802 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3803 sizeof(unsigned long), GFP_KERNEL);
3805 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3808 return sprintf(buf, "Out of memory\n");
3810 /* Push back cpu slabs */
3813 for_each_node_state(node, N_NORMAL_MEMORY) {
3814 struct kmem_cache_node *n = get_node(s, node);
3815 unsigned long flags;
3818 if (!atomic_long_read(&n->nr_slabs))
3821 spin_lock_irqsave(&n->list_lock, flags);
3822 list_for_each_entry(page, &n->partial, lru)
3823 process_slab(&t, s, page, alloc, map);
3824 list_for_each_entry(page, &n->full, lru)
3825 process_slab(&t, s, page, alloc, map);
3826 spin_unlock_irqrestore(&n->list_lock, flags);
3829 for (i = 0; i < t.count; i++) {
3830 struct location *l = &t.loc[i];
3832 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
3834 len += sprintf(buf + len, "%7ld ", l->count);
3837 len += sprintf(buf + len, "%pS", (void *)l->addr);
3839 len += sprintf(buf + len, "<not-available>");
3841 if (l->sum_time != l->min_time) {
3842 len += sprintf(buf + len, " age=%ld/%ld/%ld",
3844 (long)div_u64(l->sum_time, l->count),
3847 len += sprintf(buf + len, " age=%ld",
3850 if (l->min_pid != l->max_pid)
3851 len += sprintf(buf + len, " pid=%ld-%ld",
3852 l->min_pid, l->max_pid);
3854 len += sprintf(buf + len, " pid=%ld",
3857 if (num_online_cpus() > 1 &&
3858 !cpumask_empty(to_cpumask(l->cpus)) &&
3859 len < PAGE_SIZE - 60) {
3860 len += sprintf(buf + len, " cpus=");
3861 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3862 to_cpumask(l->cpus));
3865 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
3866 len < PAGE_SIZE - 60) {
3867 len += sprintf(buf + len, " nodes=");
3868 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3872 len += sprintf(buf + len, "\n");
3878 len += sprintf(buf, "No data\n");
3883 #ifdef SLUB_RESILIENCY_TEST
3884 static void resiliency_test(void)
3888 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3890 printk(KERN_ERR "SLUB resiliency testing\n");
3891 printk(KERN_ERR "-----------------------\n");
3892 printk(KERN_ERR "A. Corruption after allocation\n");
3894 p = kzalloc(16, GFP_KERNEL);
3896 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3897 " 0x12->0x%p\n\n", p + 16);
3899 validate_slab_cache(kmalloc_caches[4]);
3901 /* Hmmm... The next two are dangerous */
3902 p = kzalloc(32, GFP_KERNEL);
3903 p[32 + sizeof(void *)] = 0x34;
3904 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3905 " 0x34 -> -0x%p\n", p);
3907 "If allocated object is overwritten then not detectable\n\n");
3909 validate_slab_cache(kmalloc_caches[5]);
3910 p = kzalloc(64, GFP_KERNEL);
3911 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3913 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3916 "If allocated object is overwritten then not detectable\n\n");
3917 validate_slab_cache(kmalloc_caches[6]);
3919 printk(KERN_ERR "\nB. Corruption after free\n");
3920 p = kzalloc(128, GFP_KERNEL);
3923 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3924 validate_slab_cache(kmalloc_caches[7]);
3926 p = kzalloc(256, GFP_KERNEL);
3929 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3931 validate_slab_cache(kmalloc_caches[8]);
3933 p = kzalloc(512, GFP_KERNEL);
3936 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3937 validate_slab_cache(kmalloc_caches[9]);
3941 static void resiliency_test(void) {};
3946 enum slab_stat_type {
3947 SL_ALL, /* All slabs */
3948 SL_PARTIAL, /* Only partially allocated slabs */
3949 SL_CPU, /* Only slabs used for cpu caches */
3950 SL_OBJECTS, /* Determine allocated objects not slabs */
3951 SL_TOTAL /* Determine object capacity not slabs */
3954 #define SO_ALL (1 << SL_ALL)
3955 #define SO_PARTIAL (1 << SL_PARTIAL)
3956 #define SO_CPU (1 << SL_CPU)
3957 #define SO_OBJECTS (1 << SL_OBJECTS)
3958 #define SO_TOTAL (1 << SL_TOTAL)
3960 static ssize_t show_slab_objects(struct kmem_cache *s,
3961 char *buf, unsigned long flags)
3963 unsigned long total = 0;
3966 unsigned long *nodes;
3967 unsigned long *per_cpu;
3969 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
3972 per_cpu = nodes + nr_node_ids;
3974 if (flags & SO_CPU) {
3977 for_each_possible_cpu(cpu) {
3978 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3980 if (!c || c->node < 0)
3984 if (flags & SO_TOTAL)
3985 x = c->page->objects;
3986 else if (flags & SO_OBJECTS)
3992 nodes[c->node] += x;
3998 lock_memory_hotplug();
3999 #ifdef CONFIG_SLUB_DEBUG
4000 if (flags & SO_ALL) {
4001 for_each_node_state(node, N_NORMAL_MEMORY) {
4002 struct kmem_cache_node *n = get_node(s, node);
4004 if (flags & SO_TOTAL)
4005 x = atomic_long_read(&n->total_objects);
4006 else if (flags & SO_OBJECTS)
4007 x = atomic_long_read(&n->total_objects) -
4008 count_partial(n, count_free);
4011 x = atomic_long_read(&n->nr_slabs);
4018 if (flags & SO_PARTIAL) {
4019 for_each_node_state(node, N_NORMAL_MEMORY) {
4020 struct kmem_cache_node *n = get_node(s, node);
4022 if (flags & SO_TOTAL)
4023 x = count_partial(n, count_total);
4024 else if (flags & SO_OBJECTS)
4025 x = count_partial(n, count_inuse);
4032 x = sprintf(buf, "%lu", total);
4034 for_each_node_state(node, N_NORMAL_MEMORY)
4036 x += sprintf(buf + x, " N%d=%lu",
4039 unlock_memory_hotplug();
4041 return x + sprintf(buf + x, "\n");
4044 #ifdef CONFIG_SLUB_DEBUG
4045 static int any_slab_objects(struct kmem_cache *s)
4049 for_each_online_node(node) {
4050 struct kmem_cache_node *n = get_node(s, node);
4055 if (atomic_long_read(&n->total_objects))
4062 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4063 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
4065 struct slab_attribute {
4066 struct attribute attr;
4067 ssize_t (*show)(struct kmem_cache *s, char *buf);
4068 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4071 #define SLAB_ATTR_RO(_name) \
4072 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4074 #define SLAB_ATTR(_name) \
4075 static struct slab_attribute _name##_attr = \
4076 __ATTR(_name, 0644, _name##_show, _name##_store)
4078 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4080 return sprintf(buf, "%d\n", s->size);
4082 SLAB_ATTR_RO(slab_size);
4084 static ssize_t align_show(struct kmem_cache *s, char *buf)
4086 return sprintf(buf, "%d\n", s->align);
4088 SLAB_ATTR_RO(align);
4090 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4092 return sprintf(buf, "%d\n", s->objsize);
4094 SLAB_ATTR_RO(object_size);
4096 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4098 return sprintf(buf, "%d\n", oo_objects(s->oo));
4100 SLAB_ATTR_RO(objs_per_slab);
4102 static ssize_t order_store(struct kmem_cache *s,
4103 const char *buf, size_t length)
4105 unsigned long order;
4108 err = strict_strtoul(buf, 10, &order);
4112 if (order > slub_max_order || order < slub_min_order)
4115 calculate_sizes(s, order);
4119 static ssize_t order_show(struct kmem_cache *s, char *buf)
4121 return sprintf(buf, "%d\n", oo_order(s->oo));
4125 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4127 return sprintf(buf, "%lu\n", s->min_partial);
4130 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4136 err = strict_strtoul(buf, 10, &min);
4140 set_min_partial(s, min);
4143 SLAB_ATTR(min_partial);
4145 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4149 return sprintf(buf, "%pS\n", s->ctor);
4153 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4155 return sprintf(buf, "%d\n", s->refcount - 1);
4157 SLAB_ATTR_RO(aliases);
4159 static ssize_t partial_show(struct kmem_cache *s, char *buf)
4161 return show_slab_objects(s, buf, SO_PARTIAL);
4163 SLAB_ATTR_RO(partial);
4165 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4167 return show_slab_objects(s, buf, SO_CPU);
4169 SLAB_ATTR_RO(cpu_slabs);
4171 static ssize_t objects_show(struct kmem_cache *s, char *buf)
4173 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
4175 SLAB_ATTR_RO(objects);
4177 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4179 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4181 SLAB_ATTR_RO(objects_partial);
4183 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4185 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4188 static ssize_t reclaim_account_store(struct kmem_cache *s,
4189 const char *buf, size_t length)
4191 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4193 s->flags |= SLAB_RECLAIM_ACCOUNT;
4196 SLAB_ATTR(reclaim_account);
4198 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4200 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
4202 SLAB_ATTR_RO(hwcache_align);
4204 #ifdef CONFIG_ZONE_DMA
4205 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4207 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4209 SLAB_ATTR_RO(cache_dma);
4212 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4214 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4216 SLAB_ATTR_RO(destroy_by_rcu);
4218 static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4220 return sprintf(buf, "%d\n", s->reserved);
4222 SLAB_ATTR_RO(reserved);
4224 #ifdef CONFIG_SLUB_DEBUG
4225 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4227 return show_slab_objects(s, buf, SO_ALL);
4229 SLAB_ATTR_RO(slabs);
4231 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4233 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4235 SLAB_ATTR_RO(total_objects);
4237 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4239 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4242 static ssize_t sanity_checks_store(struct kmem_cache *s,
4243 const char *buf, size_t length)
4245 s->flags &= ~SLAB_DEBUG_FREE;
4247 s->flags |= SLAB_DEBUG_FREE;
4250 SLAB_ATTR(sanity_checks);
4252 static ssize_t trace_show(struct kmem_cache *s, char *buf)
4254 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4257 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4260 s->flags &= ~SLAB_TRACE;
4262 s->flags |= SLAB_TRACE;
4267 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4269 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4272 static ssize_t red_zone_store(struct kmem_cache *s,
4273 const char *buf, size_t length)
4275 if (any_slab_objects(s))
4278 s->flags &= ~SLAB_RED_ZONE;
4280 s->flags |= SLAB_RED_ZONE;
4281 calculate_sizes(s, -1);
4284 SLAB_ATTR(red_zone);
4286 static ssize_t poison_show(struct kmem_cache *s, char *buf)
4288 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4291 static ssize_t poison_store(struct kmem_cache *s,
4292 const char *buf, size_t length)
4294 if (any_slab_objects(s))
4297 s->flags &= ~SLAB_POISON;
4299 s->flags |= SLAB_POISON;
4300 calculate_sizes(s, -1);
4305 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4307 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4310 static ssize_t store_user_store(struct kmem_cache *s,
4311 const char *buf, size_t length)
4313 if (any_slab_objects(s))
4316 s->flags &= ~SLAB_STORE_USER;
4318 s->flags |= SLAB_STORE_USER;
4319 calculate_sizes(s, -1);
4322 SLAB_ATTR(store_user);
4324 static ssize_t validate_show(struct kmem_cache *s, char *buf)
4329 static ssize_t validate_store(struct kmem_cache *s,
4330 const char *buf, size_t length)
4334 if (buf[0] == '1') {
4335 ret = validate_slab_cache(s);
4341 SLAB_ATTR(validate);
4343 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4345 if (!(s->flags & SLAB_STORE_USER))
4347 return list_locations(s, buf, TRACK_ALLOC);
4349 SLAB_ATTR_RO(alloc_calls);
4351 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4353 if (!(s->flags & SLAB_STORE_USER))
4355 return list_locations(s, buf, TRACK_FREE);
4357 SLAB_ATTR_RO(free_calls);
4358 #endif /* CONFIG_SLUB_DEBUG */
4360 #ifdef CONFIG_FAILSLAB
4361 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4363 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4366 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4369 s->flags &= ~SLAB_FAILSLAB;
4371 s->flags |= SLAB_FAILSLAB;
4374 SLAB_ATTR(failslab);
4377 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4382 static ssize_t shrink_store(struct kmem_cache *s,
4383 const char *buf, size_t length)
4385 if (buf[0] == '1') {
4386 int rc = kmem_cache_shrink(s);
4397 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
4399 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
4402 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
4403 const char *buf, size_t length)
4405 unsigned long ratio;
4408 err = strict_strtoul(buf, 10, &ratio);
4413 s->remote_node_defrag_ratio = ratio * 10;
4417 SLAB_ATTR(remote_node_defrag_ratio);
4420 #ifdef CONFIG_SLUB_STATS
4421 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4423 unsigned long sum = 0;
4426 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4431 for_each_online_cpu(cpu) {
4432 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
4438 len = sprintf(buf, "%lu", sum);
4441 for_each_online_cpu(cpu) {
4442 if (data[cpu] && len < PAGE_SIZE - 20)
4443 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
4447 return len + sprintf(buf + len, "\n");
4450 static void clear_stat(struct kmem_cache *s, enum stat_item si)
4454 for_each_online_cpu(cpu)
4455 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
4458 #define STAT_ATTR(si, text) \
4459 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4461 return show_stat(s, buf, si); \
4463 static ssize_t text##_store(struct kmem_cache *s, \
4464 const char *buf, size_t length) \
4466 if (buf[0] != '0') \
4468 clear_stat(s, si); \
4473 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4474 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4475 STAT_ATTR(FREE_FASTPATH, free_fastpath);
4476 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4477 STAT_ATTR(FREE_FROZEN, free_frozen);
4478 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4479 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4480 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4481 STAT_ATTR(ALLOC_SLAB, alloc_slab);
4482 STAT_ATTR(ALLOC_REFILL, alloc_refill);
4483 STAT_ATTR(FREE_SLAB, free_slab);
4484 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4485 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4486 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4487 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4488 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4489 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
4490 STAT_ATTR(ORDER_FALLBACK, order_fallback);
4493 static struct attribute *slab_attrs[] = {
4494 &slab_size_attr.attr,
4495 &object_size_attr.attr,
4496 &objs_per_slab_attr.attr,
4498 &min_partial_attr.attr,
4500 &objects_partial_attr.attr,
4502 &cpu_slabs_attr.attr,
4506 &hwcache_align_attr.attr,
4507 &reclaim_account_attr.attr,
4508 &destroy_by_rcu_attr.attr,
4510 &reserved_attr.attr,
4511 #ifdef CONFIG_SLUB_DEBUG
4512 &total_objects_attr.attr,
4514 &sanity_checks_attr.attr,
4516 &red_zone_attr.attr,
4518 &store_user_attr.attr,
4519 &validate_attr.attr,
4520 &alloc_calls_attr.attr,
4521 &free_calls_attr.attr,
4523 #ifdef CONFIG_ZONE_DMA
4524 &cache_dma_attr.attr,
4527 &remote_node_defrag_ratio_attr.attr,
4529 #ifdef CONFIG_SLUB_STATS
4530 &alloc_fastpath_attr.attr,
4531 &alloc_slowpath_attr.attr,
4532 &free_fastpath_attr.attr,
4533 &free_slowpath_attr.attr,
4534 &free_frozen_attr.attr,
4535 &free_add_partial_attr.attr,
4536 &free_remove_partial_attr.attr,
4537 &alloc_from_partial_attr.attr,
4538 &alloc_slab_attr.attr,
4539 &alloc_refill_attr.attr,
4540 &free_slab_attr.attr,
4541 &cpuslab_flush_attr.attr,
4542 &deactivate_full_attr.attr,
4543 &deactivate_empty_attr.attr,
4544 &deactivate_to_head_attr.attr,
4545 &deactivate_to_tail_attr.attr,
4546 &deactivate_remote_frees_attr.attr,
4547 &order_fallback_attr.attr,
4549 #ifdef CONFIG_FAILSLAB
4550 &failslab_attr.attr,
4556 static struct attribute_group slab_attr_group = {
4557 .attrs = slab_attrs,
4560 static ssize_t slab_attr_show(struct kobject *kobj,
4561 struct attribute *attr,
4564 struct slab_attribute *attribute;
4565 struct kmem_cache *s;
4568 attribute = to_slab_attr(attr);
4571 if (!attribute->show)
4574 err = attribute->show(s, buf);
4579 static ssize_t slab_attr_store(struct kobject *kobj,
4580 struct attribute *attr,
4581 const char *buf, size_t len)
4583 struct slab_attribute *attribute;
4584 struct kmem_cache *s;
4587 attribute = to_slab_attr(attr);
4590 if (!attribute->store)
4593 err = attribute->store(s, buf, len);
4598 static void kmem_cache_release(struct kobject *kobj)
4600 struct kmem_cache *s = to_slab(kobj);
4606 static const struct sysfs_ops slab_sysfs_ops = {
4607 .show = slab_attr_show,
4608 .store = slab_attr_store,
4611 static struct kobj_type slab_ktype = {
4612 .sysfs_ops = &slab_sysfs_ops,
4613 .release = kmem_cache_release
4616 static int uevent_filter(struct kset *kset, struct kobject *kobj)
4618 struct kobj_type *ktype = get_ktype(kobj);
4620 if (ktype == &slab_ktype)
4625 static const struct kset_uevent_ops slab_uevent_ops = {
4626 .filter = uevent_filter,
4629 static struct kset *slab_kset;
4631 #define ID_STR_LENGTH 64
4633 /* Create a unique string id for a slab cache:
4635 * Format :[flags-]size
4637 static char *create_unique_id(struct kmem_cache *s)
4639 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4646 * First flags affecting slabcache operations. We will only
4647 * get here for aliasable slabs so we do not need to support
4648 * too many flags. The flags here must cover all flags that
4649 * are matched during merging to guarantee that the id is
4652 if (s->flags & SLAB_CACHE_DMA)
4654 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4656 if (s->flags & SLAB_DEBUG_FREE)
4658 if (!(s->flags & SLAB_NOTRACK))
4662 p += sprintf(p, "%07d", s->size);
4663 BUG_ON(p > name + ID_STR_LENGTH - 1);
4667 static int sysfs_slab_add(struct kmem_cache *s)
4673 if (slab_state < SYSFS)
4674 /* Defer until later */
4677 unmergeable = slab_unmergeable(s);
4680 * Slabcache can never be merged so we can use the name proper.
4681 * This is typically the case for debug situations. In that
4682 * case we can catch duplicate names easily.
4684 sysfs_remove_link(&slab_kset->kobj, s->name);
4688 * Create a unique name for the slab as a target
4691 name = create_unique_id(s);
4694 s->kobj.kset = slab_kset;
4695 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4697 kobject_put(&s->kobj);
4701 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4703 kobject_del(&s->kobj);
4704 kobject_put(&s->kobj);
4707 kobject_uevent(&s->kobj, KOBJ_ADD);
4709 /* Setup first alias */
4710 sysfs_slab_alias(s, s->name);
4716 static void sysfs_slab_remove(struct kmem_cache *s)
4718 if (slab_state < SYSFS)
4720 * Sysfs has not been setup yet so no need to remove the
4725 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4726 kobject_del(&s->kobj);
4727 kobject_put(&s->kobj);
4731 * Need to buffer aliases during bootup until sysfs becomes
4732 * available lest we lose that information.
4734 struct saved_alias {
4735 struct kmem_cache *s;
4737 struct saved_alias *next;
4740 static struct saved_alias *alias_list;
4742 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4744 struct saved_alias *al;
4746 if (slab_state == SYSFS) {
4748 * If we have a leftover link then remove it.
4750 sysfs_remove_link(&slab_kset->kobj, name);
4751 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
4754 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4760 al->next = alias_list;
4765 static int __init slab_sysfs_init(void)
4767 struct kmem_cache *s;
4770 down_write(&slub_lock);
4772 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
4774 up_write(&slub_lock);
4775 printk(KERN_ERR "Cannot register slab subsystem.\n");
4781 list_for_each_entry(s, &slab_caches, list) {
4782 err = sysfs_slab_add(s);
4784 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4785 " to sysfs\n", s->name);
4788 while (alias_list) {
4789 struct saved_alias *al = alias_list;
4791 alias_list = alias_list->next;
4792 err = sysfs_slab_alias(al->s, al->name);
4794 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4795 " %s to sysfs\n", s->name);
4799 up_write(&slub_lock);
4804 __initcall(slab_sysfs_init);
4805 #endif /* CONFIG_SYSFS */
4808 * The /proc/slabinfo ABI
4810 #ifdef CONFIG_SLABINFO
4811 static void print_slabinfo_header(struct seq_file *m)
4813 seq_puts(m, "slabinfo - version: 2.1\n");
4814 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4815 "<objperslab> <pagesperslab>");
4816 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4817 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4821 static void *s_start(struct seq_file *m, loff_t *pos)
4825 down_read(&slub_lock);
4827 print_slabinfo_header(m);
4829 return seq_list_start(&slab_caches, *pos);
4832 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4834 return seq_list_next(p, &slab_caches, pos);
4837 static void s_stop(struct seq_file *m, void *p)
4839 up_read(&slub_lock);
4842 static int s_show(struct seq_file *m, void *p)
4844 unsigned long nr_partials = 0;
4845 unsigned long nr_slabs = 0;
4846 unsigned long nr_inuse = 0;
4847 unsigned long nr_objs = 0;
4848 unsigned long nr_free = 0;
4849 struct kmem_cache *s;
4852 s = list_entry(p, struct kmem_cache, list);
4854 for_each_online_node(node) {
4855 struct kmem_cache_node *n = get_node(s, node);
4860 nr_partials += n->nr_partial;
4861 nr_slabs += atomic_long_read(&n->nr_slabs);
4862 nr_objs += atomic_long_read(&n->total_objects);
4863 nr_free += count_partial(n, count_free);
4866 nr_inuse = nr_objs - nr_free;
4868 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
4869 nr_objs, s->size, oo_objects(s->oo),
4870 (1 << oo_order(s->oo)));
4871 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4872 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4878 static const struct seq_operations slabinfo_op = {
4885 static int slabinfo_open(struct inode *inode, struct file *file)
4887 return seq_open(file, &slabinfo_op);
4890 static const struct file_operations proc_slabinfo_operations = {
4891 .open = slabinfo_open,
4893 .llseek = seq_lseek,
4894 .release = seq_release,
4897 static int __init slab_proc_init(void)
4899 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
4902 module_init(slab_proc_init);
4903 #endif /* CONFIG_SLABINFO */