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 no one 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);
1460 unsigned int cpuset_mems_cookie;
1463 * The defrag ratio allows a configuration of the tradeoffs between
1464 * inter node defragmentation and node local allocations. A lower
1465 * defrag_ratio increases the tendency to do local allocations
1466 * instead of attempting to obtain partial slabs from other nodes.
1468 * If the defrag_ratio is set to 0 then kmalloc() always
1469 * returns node local objects. If the ratio is higher then kmalloc()
1470 * may return off node objects because partial slabs are obtained
1471 * from other nodes and filled up.
1473 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1474 * defrag_ratio = 1000) then every (well almost) allocation will
1475 * first attempt to defrag slab caches on other nodes. This means
1476 * scanning over all nodes to look for partial slabs which may be
1477 * expensive if we do it every time we are trying to find a slab
1478 * with available objects.
1480 if (!s->remote_node_defrag_ratio ||
1481 get_cycles() % 1024 > s->remote_node_defrag_ratio)
1485 cpuset_mems_cookie = get_mems_allowed();
1486 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1487 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1488 struct kmem_cache_node *n;
1490 n = get_node(s, zone_to_nid(zone));
1492 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1493 n->nr_partial > s->min_partial) {
1494 page = get_partial_node(n);
1497 * Return the object even if
1498 * put_mems_allowed indicated that
1499 * the cpuset mems_allowed was
1500 * updated in parallel. It's a
1501 * harmless race between the alloc
1502 * and the cpuset update.
1504 put_mems_allowed(cpuset_mems_cookie);
1509 } while (!put_mems_allowed(cpuset_mems_cookie));
1515 * Get a partial page, lock it and return it.
1517 static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1520 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
1522 page = get_partial_node(get_node(s, searchnode));
1523 if (page || node != NUMA_NO_NODE)
1526 return get_any_partial(s, flags);
1530 * Move a page back to the lists.
1532 * Must be called with the slab lock held.
1534 * On exit the slab lock will have been dropped.
1536 static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
1539 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1541 __ClearPageSlubFrozen(page);
1544 if (page->freelist) {
1545 add_partial(n, page, tail);
1546 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1548 stat(s, DEACTIVATE_FULL);
1549 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
1554 stat(s, DEACTIVATE_EMPTY);
1555 if (n->nr_partial < s->min_partial) {
1557 * Adding an empty slab to the partial slabs in order
1558 * to avoid page allocator overhead. This slab needs
1559 * to come after the other slabs with objects in
1560 * so that the others get filled first. That way the
1561 * size of the partial list stays small.
1563 * kmem_cache_shrink can reclaim any empty slabs from
1566 add_partial(n, page, 1);
1571 discard_slab(s, page);
1576 #ifdef CONFIG_PREEMPT
1578 * Calculate the next globally unique transaction for disambiguiation
1579 * during cmpxchg. The transactions start with the cpu number and are then
1580 * incremented by CONFIG_NR_CPUS.
1582 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1585 * No preemption supported therefore also no need to check for
1591 static inline unsigned long next_tid(unsigned long tid)
1593 return tid + TID_STEP;
1596 static inline unsigned int tid_to_cpu(unsigned long tid)
1598 return tid % TID_STEP;
1601 static inline unsigned long tid_to_event(unsigned long tid)
1603 return tid / TID_STEP;
1606 static inline unsigned int init_tid(int cpu)
1611 static inline void note_cmpxchg_failure(const char *n,
1612 const struct kmem_cache *s, unsigned long tid)
1614 #ifdef SLUB_DEBUG_CMPXCHG
1615 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1617 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1619 #ifdef CONFIG_PREEMPT
1620 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1621 printk("due to cpu change %d -> %d\n",
1622 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1625 if (tid_to_event(tid) != tid_to_event(actual_tid))
1626 printk("due to cpu running other code. Event %ld->%ld\n",
1627 tid_to_event(tid), tid_to_event(actual_tid));
1629 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1630 actual_tid, tid, next_tid(tid));
1632 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
1635 void init_kmem_cache_cpus(struct kmem_cache *s)
1639 for_each_possible_cpu(cpu)
1640 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
1643 * Remove the cpu slab
1645 static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1648 struct page *page = c->page;
1652 stat(s, DEACTIVATE_REMOTE_FREES);
1654 * Merge cpu freelist into slab freelist. Typically we get here
1655 * because both freelists are empty. So this is unlikely
1658 while (unlikely(c->freelist)) {
1661 tail = 0; /* Hot objects. Put the slab first */
1663 /* Retrieve object from cpu_freelist */
1664 object = c->freelist;
1665 c->freelist = get_freepointer(s, c->freelist);
1667 /* And put onto the regular freelist */
1668 set_freepointer(s, object, page->freelist);
1669 page->freelist = object;
1673 c->tid = next_tid(c->tid);
1674 unfreeze_slab(s, page, tail);
1677 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1679 stat(s, CPUSLAB_FLUSH);
1681 deactivate_slab(s, c);
1687 * Called from IPI handler with interrupts disabled.
1689 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
1691 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
1693 if (likely(c && c->page))
1697 static void flush_cpu_slab(void *d)
1699 struct kmem_cache *s = d;
1701 __flush_cpu_slab(s, smp_processor_id());
1704 static void flush_all(struct kmem_cache *s)
1706 on_each_cpu(flush_cpu_slab, s, 1);
1710 * Check if the objects in a per cpu structure fit numa
1711 * locality expectations.
1713 static inline int node_match(struct kmem_cache_cpu *c, int node)
1716 if (node != NUMA_NO_NODE && c->node != node)
1722 static int count_free(struct page *page)
1724 return page->objects - page->inuse;
1727 static unsigned long count_partial(struct kmem_cache_node *n,
1728 int (*get_count)(struct page *))
1730 unsigned long flags;
1731 unsigned long x = 0;
1734 spin_lock_irqsave(&n->list_lock, flags);
1735 list_for_each_entry(page, &n->partial, lru)
1736 x += get_count(page);
1737 spin_unlock_irqrestore(&n->list_lock, flags);
1741 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1743 #ifdef CONFIG_SLUB_DEBUG
1744 return atomic_long_read(&n->total_objects);
1750 static noinline void
1751 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1756 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1758 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1759 "default order: %d, min order: %d\n", s->name, s->objsize,
1760 s->size, oo_order(s->oo), oo_order(s->min));
1762 if (oo_order(s->min) > get_order(s->objsize))
1763 printk(KERN_WARNING " %s debugging increased min order, use "
1764 "slub_debug=O to disable.\n", s->name);
1766 for_each_online_node(node) {
1767 struct kmem_cache_node *n = get_node(s, node);
1768 unsigned long nr_slabs;
1769 unsigned long nr_objs;
1770 unsigned long nr_free;
1775 nr_free = count_partial(n, count_free);
1776 nr_slabs = node_nr_slabs(n);
1777 nr_objs = node_nr_objs(n);
1780 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1781 node, nr_slabs, nr_objs, nr_free);
1786 * Slow path. The lockless freelist is empty or we need to perform
1789 * Interrupts are disabled.
1791 * Processing is still very fast if new objects have been freed to the
1792 * regular freelist. In that case we simply take over the regular freelist
1793 * as the lockless freelist and zap the regular freelist.
1795 * If that is not working then we fall back to the partial lists. We take the
1796 * first element of the freelist as the object to allocate now and move the
1797 * rest of the freelist to the lockless freelist.
1799 * And if we were unable to get a new slab from the partial slab lists then
1800 * we need to allocate a new slab. This is the slowest path since it involves
1801 * a call to the page allocator and the setup of a new slab.
1803 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1804 unsigned long addr, struct kmem_cache_cpu *c)
1808 unsigned long flags;
1810 local_irq_save(flags);
1811 #ifdef CONFIG_PREEMPT
1813 * We may have been preempted and rescheduled on a different
1814 * cpu before disabling interrupts. Need to reload cpu area
1817 c = this_cpu_ptr(s->cpu_slab);
1820 /* We handle __GFP_ZERO in the caller */
1821 gfpflags &= ~__GFP_ZERO;
1828 if (unlikely(!node_match(c, node)))
1831 /* must check again c->freelist in case of cpu migration or IRQ */
1832 object = c->freelist;
1834 goto update_freelist;
1836 stat(s, ALLOC_REFILL);
1839 object = page->freelist;
1840 if (unlikely(!object))
1842 if (kmem_cache_debug(s))
1846 c->freelist = get_freepointer(s, object);
1847 page->inuse = page->objects;
1848 page->freelist = NULL;
1851 c->tid = next_tid(c->tid);
1852 local_irq_restore(flags);
1853 stat(s, ALLOC_SLOWPATH);
1857 deactivate_slab(s, c);
1860 page = get_partial(s, gfpflags, node);
1862 stat(s, ALLOC_FROM_PARTIAL);
1863 c->node = page_to_nid(page);
1868 gfpflags &= gfp_allowed_mask;
1869 if (gfpflags & __GFP_WAIT)
1872 page = new_slab(s, gfpflags, node);
1874 if (gfpflags & __GFP_WAIT)
1875 local_irq_disable();
1878 c = __this_cpu_ptr(s->cpu_slab);
1879 stat(s, ALLOC_SLAB);
1884 __SetPageSlubFrozen(page);
1885 c->node = page_to_nid(page);
1889 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1890 slab_out_of_memory(s, gfpflags, node);
1891 local_irq_restore(flags);
1894 if (!alloc_debug_processing(s, page, object, addr))
1898 page->freelist = get_freepointer(s, object);
1899 deactivate_slab(s, c);
1901 c->node = NUMA_NO_NODE;
1902 local_irq_restore(flags);
1907 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1908 * have the fastpath folded into their functions. So no function call
1909 * overhead for requests that can be satisfied on the fastpath.
1911 * The fastpath works by first checking if the lockless freelist can be used.
1912 * If not then __slab_alloc is called for slow processing.
1914 * Otherwise we can simply pick the next object from the lockless free list.
1916 static __always_inline void *slab_alloc(struct kmem_cache *s,
1917 gfp_t gfpflags, int node, unsigned long addr)
1920 struct kmem_cache_cpu *c;
1923 if (slab_pre_alloc_hook(s, gfpflags))
1929 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1930 * enabled. We may switch back and forth between cpus while
1931 * reading from one cpu area. That does not matter as long
1932 * as we end up on the original cpu again when doing the cmpxchg.
1934 c = __this_cpu_ptr(s->cpu_slab);
1937 * The transaction ids are globally unique per cpu and per operation on
1938 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1939 * occurs on the right processor and that there was no operation on the
1940 * linked list in between.
1945 object = c->freelist;
1946 if (unlikely(!object || !node_match(c, node)))
1948 object = __slab_alloc(s, gfpflags, node, addr, c);
1952 * The cmpxchg will only match if there was no additional
1953 * operation and if we are on the right processor.
1955 * The cmpxchg does the following atomically (without lock semantics!)
1956 * 1. Relocate first pointer to the current per cpu area.
1957 * 2. Verify that tid and freelist have not been changed
1958 * 3. If they were not changed replace tid and freelist
1960 * Since this is without lock semantics the protection is only against
1961 * code executing on this cpu *not* from access by other cpus.
1963 if (unlikely(!irqsafe_cpu_cmpxchg_double(
1964 s->cpu_slab->freelist, s->cpu_slab->tid,
1966 get_freepointer_safe(s, object), next_tid(tid)))) {
1968 note_cmpxchg_failure("slab_alloc", s, tid);
1971 stat(s, ALLOC_FASTPATH);
1974 if (unlikely(gfpflags & __GFP_ZERO) && object)
1975 memset(object, 0, s->objsize);
1977 slab_post_alloc_hook(s, gfpflags, object);
1982 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1984 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1986 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
1990 EXPORT_SYMBOL(kmem_cache_alloc);
1992 #ifdef CONFIG_TRACING
1993 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1995 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1996 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1999 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2001 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
2003 void *ret = kmalloc_order(size, flags, order);
2004 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
2007 EXPORT_SYMBOL(kmalloc_order_trace);
2011 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2013 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2015 trace_kmem_cache_alloc_node(_RET_IP_, ret,
2016 s->objsize, s->size, gfpflags, node);
2020 EXPORT_SYMBOL(kmem_cache_alloc_node);
2022 #ifdef CONFIG_TRACING
2023 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2025 int node, size_t size)
2027 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
2029 trace_kmalloc_node(_RET_IP_, ret,
2030 size, s->size, gfpflags, node);
2033 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2038 * Slow patch handling. This may still be called frequently since objects
2039 * have a longer lifetime than the cpu slabs in most processing loads.
2041 * So we still attempt to reduce cache line usage. Just take the slab
2042 * lock and free the item. If there is no additional partial page
2043 * handling required then we can return immediately.
2045 static void __slab_free(struct kmem_cache *s, struct page *page,
2046 void *x, unsigned long addr)
2049 void **object = (void *)x;
2050 unsigned long flags;
2052 local_irq_save(flags);
2054 stat(s, FREE_SLOWPATH);
2056 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2059 prior = page->freelist;
2060 set_freepointer(s, object, prior);
2061 page->freelist = object;
2064 if (unlikely(PageSlubFrozen(page))) {
2065 stat(s, FREE_FROZEN);
2069 if (unlikely(!page->inuse))
2073 * Objects left in the slab. If it was not on the partial list before
2076 if (unlikely(!prior)) {
2077 add_partial(get_node(s, page_to_nid(page)), page, 1);
2078 stat(s, FREE_ADD_PARTIAL);
2083 local_irq_restore(flags);
2089 * Slab still on the partial list.
2091 remove_partial(s, page);
2092 stat(s, FREE_REMOVE_PARTIAL);
2095 local_irq_restore(flags);
2097 discard_slab(s, page);
2101 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2102 * can perform fastpath freeing without additional function calls.
2104 * The fastpath is only possible if we are freeing to the current cpu slab
2105 * of this processor. This typically the case if we have just allocated
2108 * If fastpath is not possible then fall back to __slab_free where we deal
2109 * with all sorts of special processing.
2111 static __always_inline void slab_free(struct kmem_cache *s,
2112 struct page *page, void *x, unsigned long addr)
2114 void **object = (void *)x;
2115 struct kmem_cache_cpu *c;
2118 slab_free_hook(s, x);
2123 * Determine the currently cpus per cpu slab.
2124 * The cpu may change afterward. However that does not matter since
2125 * data is retrieved via this pointer. If we are on the same cpu
2126 * during the cmpxchg then the free will succedd.
2128 c = __this_cpu_ptr(s->cpu_slab);
2133 if (likely(page == c->page)) {
2134 set_freepointer(s, object, c->freelist);
2136 if (unlikely(!irqsafe_cpu_cmpxchg_double(
2137 s->cpu_slab->freelist, s->cpu_slab->tid,
2139 object, next_tid(tid)))) {
2141 note_cmpxchg_failure("slab_free", s, tid);
2144 stat(s, FREE_FASTPATH);
2146 __slab_free(s, page, x, addr);
2150 void kmem_cache_free(struct kmem_cache *s, void *x)
2154 page = virt_to_head_page(x);
2156 slab_free(s, page, x, _RET_IP_);
2158 trace_kmem_cache_free(_RET_IP_, x);
2160 EXPORT_SYMBOL(kmem_cache_free);
2163 * Object placement in a slab is made very easy because we always start at
2164 * offset 0. If we tune the size of the object to the alignment then we can
2165 * get the required alignment by putting one properly sized object after
2168 * Notice that the allocation order determines the sizes of the per cpu
2169 * caches. Each processor has always one slab available for allocations.
2170 * Increasing the allocation order reduces the number of times that slabs
2171 * must be moved on and off the partial lists and is therefore a factor in
2176 * Mininum / Maximum order of slab pages. This influences locking overhead
2177 * and slab fragmentation. A higher order reduces the number of partial slabs
2178 * and increases the number of allocations possible without having to
2179 * take the list_lock.
2181 static int slub_min_order;
2182 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
2183 static int slub_min_objects;
2186 * Merge control. If this is set then no merging of slab caches will occur.
2187 * (Could be removed. This was introduced to pacify the merge skeptics.)
2189 static int slub_nomerge;
2192 * Calculate the order of allocation given an slab object size.
2194 * The order of allocation has significant impact on performance and other
2195 * system components. Generally order 0 allocations should be preferred since
2196 * order 0 does not cause fragmentation in the page allocator. Larger objects
2197 * be problematic to put into order 0 slabs because there may be too much
2198 * unused space left. We go to a higher order if more than 1/16th of the slab
2201 * In order to reach satisfactory performance we must ensure that a minimum
2202 * number of objects is in one slab. Otherwise we may generate too much
2203 * activity on the partial lists which requires taking the list_lock. This is
2204 * less a concern for large slabs though which are rarely used.
2206 * slub_max_order specifies the order where we begin to stop considering the
2207 * number of objects in a slab as critical. If we reach slub_max_order then
2208 * we try to keep the page order as low as possible. So we accept more waste
2209 * of space in favor of a small page order.
2211 * Higher order allocations also allow the placement of more objects in a
2212 * slab and thereby reduce object handling overhead. If the user has
2213 * requested a higher mininum order then we start with that one instead of
2214 * the smallest order which will fit the object.
2216 static inline int slab_order(int size, int min_objects,
2217 int max_order, int fract_leftover, int reserved)
2221 int min_order = slub_min_order;
2223 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
2224 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
2226 for (order = max(min_order,
2227 fls(min_objects * size - 1) - PAGE_SHIFT);
2228 order <= max_order; order++) {
2230 unsigned long slab_size = PAGE_SIZE << order;
2232 if (slab_size < min_objects * size + reserved)
2235 rem = (slab_size - reserved) % size;
2237 if (rem <= slab_size / fract_leftover)
2245 static inline int calculate_order(int size, int reserved)
2253 * Attempt to find best configuration for a slab. This
2254 * works by first attempting to generate a layout with
2255 * the best configuration and backing off gradually.
2257 * First we reduce the acceptable waste in a slab. Then
2258 * we reduce the minimum objects required in a slab.
2260 min_objects = slub_min_objects;
2262 min_objects = 4 * (fls(nr_cpu_ids) + 1);
2263 max_objects = order_objects(slub_max_order, size, reserved);
2264 min_objects = min(min_objects, max_objects);
2266 while (min_objects > 1) {
2268 while (fraction >= 4) {
2269 order = slab_order(size, min_objects,
2270 slub_max_order, fraction, reserved);
2271 if (order <= slub_max_order)
2279 * We were unable to place multiple objects in a slab. Now
2280 * lets see if we can place a single object there.
2282 order = slab_order(size, 1, slub_max_order, 1, reserved);
2283 if (order <= slub_max_order)
2287 * Doh this slab cannot be placed using slub_max_order.
2289 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
2290 if (order < MAX_ORDER)
2296 * Figure out what the alignment of the objects will be.
2298 static unsigned long calculate_alignment(unsigned long flags,
2299 unsigned long align, unsigned long size)
2302 * If the user wants hardware cache aligned objects then follow that
2303 * suggestion if the object is sufficiently large.
2305 * The hardware cache alignment cannot override the specified
2306 * alignment though. If that is greater then use it.
2308 if (flags & SLAB_HWCACHE_ALIGN) {
2309 unsigned long ralign = cache_line_size();
2310 while (size <= ralign / 2)
2312 align = max(align, ralign);
2315 if (align < ARCH_SLAB_MINALIGN)
2316 align = ARCH_SLAB_MINALIGN;
2318 return ALIGN(align, sizeof(void *));
2322 init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
2325 spin_lock_init(&n->list_lock);
2326 INIT_LIST_HEAD(&n->partial);
2327 #ifdef CONFIG_SLUB_DEBUG
2328 atomic_long_set(&n->nr_slabs, 0);
2329 atomic_long_set(&n->total_objects, 0);
2330 INIT_LIST_HEAD(&n->full);
2334 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
2336 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2337 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
2340 * Must align to double word boundary for the double cmpxchg
2341 * instructions to work; see __pcpu_double_call_return_bool().
2343 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2344 2 * sizeof(void *));
2349 init_kmem_cache_cpus(s);
2354 static struct kmem_cache *kmem_cache_node;
2357 * No kmalloc_node yet so do it by hand. We know that this is the first
2358 * slab on the node for this slabcache. There are no concurrent accesses
2361 * Note that this function only works on the kmalloc_node_cache
2362 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2363 * memory on a fresh node that has no slab structures yet.
2365 static void early_kmem_cache_node_alloc(int node)
2368 struct kmem_cache_node *n;
2369 unsigned long flags;
2371 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
2373 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
2376 if (page_to_nid(page) != node) {
2377 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2379 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2380 "in order to be able to continue\n");
2385 page->freelist = get_freepointer(kmem_cache_node, n);
2387 kmem_cache_node->node[node] = n;
2388 #ifdef CONFIG_SLUB_DEBUG
2389 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
2390 init_tracking(kmem_cache_node, n);
2392 init_kmem_cache_node(n, kmem_cache_node);
2393 inc_slabs_node(kmem_cache_node, node, page->objects);
2396 * lockdep requires consistent irq usage for each lock
2397 * so even though there cannot be a race this early in
2398 * the boot sequence, we still disable irqs.
2400 local_irq_save(flags);
2401 add_partial(n, page, 0);
2402 local_irq_restore(flags);
2405 static void free_kmem_cache_nodes(struct kmem_cache *s)
2409 for_each_node_state(node, N_NORMAL_MEMORY) {
2410 struct kmem_cache_node *n = s->node[node];
2413 kmem_cache_free(kmem_cache_node, n);
2415 s->node[node] = NULL;
2419 static int init_kmem_cache_nodes(struct kmem_cache *s)
2423 for_each_node_state(node, N_NORMAL_MEMORY) {
2424 struct kmem_cache_node *n;
2426 if (slab_state == DOWN) {
2427 early_kmem_cache_node_alloc(node);
2430 n = kmem_cache_alloc_node(kmem_cache_node,
2434 free_kmem_cache_nodes(s);
2439 init_kmem_cache_node(n, s);
2444 static void set_min_partial(struct kmem_cache *s, unsigned long min)
2446 if (min < MIN_PARTIAL)
2448 else if (min > MAX_PARTIAL)
2450 s->min_partial = min;
2454 * calculate_sizes() determines the order and the distribution of data within
2457 static int calculate_sizes(struct kmem_cache *s, int forced_order)
2459 unsigned long flags = s->flags;
2460 unsigned long size = s->objsize;
2461 unsigned long align = s->align;
2465 * Round up object size to the next word boundary. We can only
2466 * place the free pointer at word boundaries and this determines
2467 * the possible location of the free pointer.
2469 size = ALIGN(size, sizeof(void *));
2471 #ifdef CONFIG_SLUB_DEBUG
2473 * Determine if we can poison the object itself. If the user of
2474 * the slab may touch the object after free or before allocation
2475 * then we should never poison the object itself.
2477 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
2479 s->flags |= __OBJECT_POISON;
2481 s->flags &= ~__OBJECT_POISON;
2485 * If we are Redzoning then check if there is some space between the
2486 * end of the object and the free pointer. If not then add an
2487 * additional word to have some bytes to store Redzone information.
2489 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2490 size += sizeof(void *);
2494 * With that we have determined the number of bytes in actual use
2495 * by the object. This is the potential offset to the free pointer.
2499 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
2502 * Relocate free pointer after the object if it is not
2503 * permitted to overwrite the first word of the object on
2506 * This is the case if we do RCU, have a constructor or
2507 * destructor or are poisoning the objects.
2510 size += sizeof(void *);
2513 #ifdef CONFIG_SLUB_DEBUG
2514 if (flags & SLAB_STORE_USER)
2516 * Need to store information about allocs and frees after
2519 size += 2 * sizeof(struct track);
2521 if (flags & SLAB_RED_ZONE)
2523 * Add some empty padding so that we can catch
2524 * overwrites from earlier objects rather than let
2525 * tracking information or the free pointer be
2526 * corrupted if a user writes before the start
2529 size += sizeof(void *);
2533 * Determine the alignment based on various parameters that the
2534 * user specified and the dynamic determination of cache line size
2537 align = calculate_alignment(flags, align, s->objsize);
2541 * SLUB stores one object immediately after another beginning from
2542 * offset 0. In order to align the objects we have to simply size
2543 * each object to conform to the alignment.
2545 size = ALIGN(size, align);
2547 if (forced_order >= 0)
2548 order = forced_order;
2550 order = calculate_order(size, s->reserved);
2557 s->allocflags |= __GFP_COMP;
2559 if (s->flags & SLAB_CACHE_DMA)
2560 s->allocflags |= SLUB_DMA;
2562 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2563 s->allocflags |= __GFP_RECLAIMABLE;
2566 * Determine the number of objects per slab
2568 s->oo = oo_make(order, size, s->reserved);
2569 s->min = oo_make(get_order(size), size, s->reserved);
2570 if (oo_objects(s->oo) > oo_objects(s->max))
2573 return !!oo_objects(s->oo);
2577 static int kmem_cache_open(struct kmem_cache *s,
2578 const char *name, size_t size,
2579 size_t align, unsigned long flags,
2580 void (*ctor)(void *))
2582 memset(s, 0, kmem_size);
2587 s->flags = kmem_cache_flags(size, flags, name, ctor);
2590 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2591 s->reserved = sizeof(struct rcu_head);
2593 if (!calculate_sizes(s, -1))
2595 if (disable_higher_order_debug) {
2597 * Disable debugging flags that store metadata if the min slab
2600 if (get_order(s->size) > get_order(s->objsize)) {
2601 s->flags &= ~DEBUG_METADATA_FLAGS;
2603 if (!calculate_sizes(s, -1))
2609 * The larger the object size is, the more pages we want on the partial
2610 * list to avoid pounding the page allocator excessively.
2612 set_min_partial(s, ilog2(s->size));
2615 s->remote_node_defrag_ratio = 1000;
2617 if (!init_kmem_cache_nodes(s))
2620 if (alloc_kmem_cache_cpus(s))
2623 free_kmem_cache_nodes(s);
2625 if (flags & SLAB_PANIC)
2626 panic("Cannot create slab %s size=%lu realsize=%u "
2627 "order=%u offset=%u flags=%lx\n",
2628 s->name, (unsigned long)size, s->size, oo_order(s->oo),
2634 * Determine the size of a slab object
2636 unsigned int kmem_cache_size(struct kmem_cache *s)
2640 EXPORT_SYMBOL(kmem_cache_size);
2642 static void list_slab_objects(struct kmem_cache *s, struct page *page,
2645 #ifdef CONFIG_SLUB_DEBUG
2646 void *addr = page_address(page);
2648 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2649 sizeof(long), GFP_ATOMIC);
2652 slab_err(s, page, "%s", text);
2655 get_map(s, page, map);
2656 for_each_object(p, s, addr, page->objects) {
2658 if (!test_bit(slab_index(p, s, addr), map)) {
2659 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2661 print_tracking(s, p);
2670 * Attempt to free all partial slabs on a node.
2672 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
2674 unsigned long flags;
2675 struct page *page, *h;
2677 spin_lock_irqsave(&n->list_lock, flags);
2678 list_for_each_entry_safe(page, h, &n->partial, lru) {
2680 __remove_partial(n, page);
2681 discard_slab(s, page);
2683 list_slab_objects(s, page,
2684 "Objects remaining on kmem_cache_close()");
2687 spin_unlock_irqrestore(&n->list_lock, flags);
2691 * Release all resources used by a slab cache.
2693 static inline int kmem_cache_close(struct kmem_cache *s)
2698 free_percpu(s->cpu_slab);
2699 /* Attempt to free all objects */
2700 for_each_node_state(node, N_NORMAL_MEMORY) {
2701 struct kmem_cache_node *n = get_node(s, node);
2704 if (n->nr_partial || slabs_node(s, node))
2707 free_kmem_cache_nodes(s);
2712 * Close a cache and release the kmem_cache structure
2713 * (must be used for caches created using kmem_cache_create)
2715 void kmem_cache_destroy(struct kmem_cache *s)
2717 down_write(&slub_lock);
2721 if (kmem_cache_close(s)) {
2722 printk(KERN_ERR "SLUB %s: %s called for cache that "
2723 "still has objects.\n", s->name, __func__);
2726 if (s->flags & SLAB_DESTROY_BY_RCU)
2728 sysfs_slab_remove(s);
2730 up_write(&slub_lock);
2732 EXPORT_SYMBOL(kmem_cache_destroy);
2734 /********************************************************************
2736 *******************************************************************/
2738 struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
2739 EXPORT_SYMBOL(kmalloc_caches);
2741 static struct kmem_cache *kmem_cache;
2743 #ifdef CONFIG_ZONE_DMA
2744 static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
2747 static int __init setup_slub_min_order(char *str)
2749 get_option(&str, &slub_min_order);
2754 __setup("slub_min_order=", setup_slub_min_order);
2756 static int __init setup_slub_max_order(char *str)
2758 get_option(&str, &slub_max_order);
2759 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
2764 __setup("slub_max_order=", setup_slub_max_order);
2766 static int __init setup_slub_min_objects(char *str)
2768 get_option(&str, &slub_min_objects);
2773 __setup("slub_min_objects=", setup_slub_min_objects);
2775 static int __init setup_slub_nomerge(char *str)
2781 __setup("slub_nomerge", setup_slub_nomerge);
2783 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2784 int size, unsigned int flags)
2786 struct kmem_cache *s;
2788 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2791 * This function is called with IRQs disabled during early-boot on
2792 * single CPU so there's no need to take slub_lock here.
2794 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
2798 list_add(&s->list, &slab_caches);
2802 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2807 * Conversion table for small slabs sizes / 8 to the index in the
2808 * kmalloc array. This is necessary for slabs < 192 since we have non power
2809 * of two cache sizes there. The size of larger slabs can be determined using
2812 static s8 size_index[24] = {
2839 static inline int size_index_elem(size_t bytes)
2841 return (bytes - 1) / 8;
2844 static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2850 return ZERO_SIZE_PTR;
2852 index = size_index[size_index_elem(size)];
2854 index = fls(size - 1);
2856 #ifdef CONFIG_ZONE_DMA
2857 if (unlikely((flags & SLUB_DMA)))
2858 return kmalloc_dma_caches[index];
2861 return kmalloc_caches[index];
2864 void *__kmalloc(size_t size, gfp_t flags)
2866 struct kmem_cache *s;
2869 if (unlikely(size > SLUB_MAX_SIZE))
2870 return kmalloc_large(size, flags);
2872 s = get_slab(size, flags);
2874 if (unlikely(ZERO_OR_NULL_PTR(s)))
2877 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
2879 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
2883 EXPORT_SYMBOL(__kmalloc);
2886 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2891 flags |= __GFP_COMP | __GFP_NOTRACK;
2892 page = alloc_pages_node(node, flags, get_order(size));
2894 ptr = page_address(page);
2896 kmemleak_alloc(ptr, size, 1, flags);
2900 void *__kmalloc_node(size_t size, gfp_t flags, int node)
2902 struct kmem_cache *s;
2905 if (unlikely(size > SLUB_MAX_SIZE)) {
2906 ret = kmalloc_large_node(size, flags, node);
2908 trace_kmalloc_node(_RET_IP_, ret,
2909 size, PAGE_SIZE << get_order(size),
2915 s = get_slab(size, flags);
2917 if (unlikely(ZERO_OR_NULL_PTR(s)))
2920 ret = slab_alloc(s, flags, node, _RET_IP_);
2922 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
2926 EXPORT_SYMBOL(__kmalloc_node);
2929 size_t ksize(const void *object)
2933 if (unlikely(object == ZERO_SIZE_PTR))
2936 page = virt_to_head_page(object);
2938 if (unlikely(!PageSlab(page))) {
2939 WARN_ON(!PageCompound(page));
2940 return PAGE_SIZE << compound_order(page);
2943 return slab_ksize(page->slab);
2945 EXPORT_SYMBOL(ksize);
2947 void kfree(const void *x)
2950 void *object = (void *)x;
2952 trace_kfree(_RET_IP_, x);
2954 if (unlikely(ZERO_OR_NULL_PTR(x)))
2957 page = virt_to_head_page(x);
2958 if (unlikely(!PageSlab(page))) {
2959 BUG_ON(!PageCompound(page));
2964 slab_free(page->slab, page, object, _RET_IP_);
2966 EXPORT_SYMBOL(kfree);
2969 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2970 * the remaining slabs by the number of items in use. The slabs with the
2971 * most items in use come first. New allocations will then fill those up
2972 * and thus they can be removed from the partial lists.
2974 * The slabs with the least items are placed last. This results in them
2975 * being allocated from last increasing the chance that the last objects
2976 * are freed in them.
2978 int kmem_cache_shrink(struct kmem_cache *s)
2982 struct kmem_cache_node *n;
2985 int objects = oo_objects(s->max);
2986 struct list_head *slabs_by_inuse =
2987 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
2988 unsigned long flags;
2990 if (!slabs_by_inuse)
2994 for_each_node_state(node, N_NORMAL_MEMORY) {
2995 n = get_node(s, node);
3000 for (i = 0; i < objects; i++)
3001 INIT_LIST_HEAD(slabs_by_inuse + i);
3003 spin_lock_irqsave(&n->list_lock, flags);
3006 * Build lists indexed by the items in use in each slab.
3008 * Note that concurrent frees may occur while we hold the
3009 * list_lock. page->inuse here is the upper limit.
3011 list_for_each_entry_safe(page, t, &n->partial, lru) {
3012 if (!page->inuse && slab_trylock(page)) {
3014 * Must hold slab lock here because slab_free
3015 * may have freed the last object and be
3016 * waiting to release the slab.
3018 __remove_partial(n, page);
3020 discard_slab(s, page);
3022 list_move(&page->lru,
3023 slabs_by_inuse + page->inuse);
3028 * Rebuild the partial list with the slabs filled up most
3029 * first and the least used slabs at the end.
3031 for (i = objects - 1; i >= 0; i--)
3032 list_splice(slabs_by_inuse + i, n->partial.prev);
3034 spin_unlock_irqrestore(&n->list_lock, flags);
3037 kfree(slabs_by_inuse);
3040 EXPORT_SYMBOL(kmem_cache_shrink);
3042 #if defined(CONFIG_MEMORY_HOTPLUG)
3043 static int slab_mem_going_offline_callback(void *arg)
3045 struct kmem_cache *s;
3047 down_read(&slub_lock);
3048 list_for_each_entry(s, &slab_caches, list)
3049 kmem_cache_shrink(s);
3050 up_read(&slub_lock);
3055 static void slab_mem_offline_callback(void *arg)
3057 struct kmem_cache_node *n;
3058 struct kmem_cache *s;
3059 struct memory_notify *marg = arg;
3062 offline_node = marg->status_change_nid;
3065 * If the node still has available memory. we need kmem_cache_node
3068 if (offline_node < 0)
3071 down_read(&slub_lock);
3072 list_for_each_entry(s, &slab_caches, list) {
3073 n = get_node(s, offline_node);
3076 * if n->nr_slabs > 0, slabs still exist on the node
3077 * that is going down. We were unable to free them,
3078 * and offline_pages() function shouldn't call this
3079 * callback. So, we must fail.
3081 BUG_ON(slabs_node(s, offline_node));
3083 s->node[offline_node] = NULL;
3084 kmem_cache_free(kmem_cache_node, n);
3087 up_read(&slub_lock);
3090 static int slab_mem_going_online_callback(void *arg)
3092 struct kmem_cache_node *n;
3093 struct kmem_cache *s;
3094 struct memory_notify *marg = arg;
3095 int nid = marg->status_change_nid;
3099 * If the node's memory is already available, then kmem_cache_node is
3100 * already created. Nothing to do.
3106 * We are bringing a node online. No memory is available yet. We must
3107 * allocate a kmem_cache_node structure in order to bring the node
3110 down_read(&slub_lock);
3111 list_for_each_entry(s, &slab_caches, list) {
3113 * XXX: kmem_cache_alloc_node will fallback to other nodes
3114 * since memory is not yet available from the node that
3117 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
3122 init_kmem_cache_node(n, s);
3126 up_read(&slub_lock);
3130 static int slab_memory_callback(struct notifier_block *self,
3131 unsigned long action, void *arg)
3136 case MEM_GOING_ONLINE:
3137 ret = slab_mem_going_online_callback(arg);
3139 case MEM_GOING_OFFLINE:
3140 ret = slab_mem_going_offline_callback(arg);
3143 case MEM_CANCEL_ONLINE:
3144 slab_mem_offline_callback(arg);
3147 case MEM_CANCEL_OFFLINE:
3151 ret = notifier_from_errno(ret);
3157 #endif /* CONFIG_MEMORY_HOTPLUG */
3159 /********************************************************************
3160 * Basic setup of slabs
3161 *******************************************************************/
3164 * Used for early kmem_cache structures that were allocated using
3165 * the page allocator
3168 static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3172 list_add(&s->list, &slab_caches);
3175 for_each_node_state(node, N_NORMAL_MEMORY) {
3176 struct kmem_cache_node *n = get_node(s, node);
3180 list_for_each_entry(p, &n->partial, lru)
3183 #ifdef CONFIG_SLUB_DEBUG
3184 list_for_each_entry(p, &n->full, lru)
3191 void __init kmem_cache_init(void)
3195 struct kmem_cache *temp_kmem_cache;
3197 struct kmem_cache *temp_kmem_cache_node;
3198 unsigned long kmalloc_size;
3200 kmem_size = offsetof(struct kmem_cache, node) +
3201 nr_node_ids * sizeof(struct kmem_cache_node *);
3203 /* Allocate two kmem_caches from the page allocator */
3204 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3205 order = get_order(2 * kmalloc_size);
3206 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3209 * Must first have the slab cache available for the allocations of the
3210 * struct kmem_cache_node's. There is special bootstrap code in
3211 * kmem_cache_open for slab_state == DOWN.
3213 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3215 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3216 sizeof(struct kmem_cache_node),
3217 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3219 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
3221 /* Able to allocate the per node structures */
3222 slab_state = PARTIAL;
3224 temp_kmem_cache = kmem_cache;
3225 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3226 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3227 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3228 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
3231 * Allocate kmem_cache_node properly from the kmem_cache slab.
3232 * kmem_cache_node is separately allocated so no need to
3233 * update any list pointers.
3235 temp_kmem_cache_node = kmem_cache_node;
3237 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3238 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3240 kmem_cache_bootstrap_fixup(kmem_cache_node);
3243 kmem_cache_bootstrap_fixup(kmem_cache);
3245 /* Free temporary boot structure */
3246 free_pages((unsigned long)temp_kmem_cache, order);
3248 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3251 * Patch up the size_index table if we have strange large alignment
3252 * requirements for the kmalloc array. This is only the case for
3253 * MIPS it seems. The standard arches will not generate any code here.
3255 * Largest permitted alignment is 256 bytes due to the way we
3256 * handle the index determination for the smaller caches.
3258 * Make sure that nothing crazy happens if someone starts tinkering
3259 * around with ARCH_KMALLOC_MINALIGN
3261 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3262 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3264 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3265 int elem = size_index_elem(i);
3266 if (elem >= ARRAY_SIZE(size_index))
3268 size_index[elem] = KMALLOC_SHIFT_LOW;
3271 if (KMALLOC_MIN_SIZE == 64) {
3273 * The 96 byte size cache is not used if the alignment
3276 for (i = 64 + 8; i <= 96; i += 8)
3277 size_index[size_index_elem(i)] = 7;
3278 } else if (KMALLOC_MIN_SIZE == 128) {
3280 * The 192 byte sized cache is not used if the alignment
3281 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3284 for (i = 128 + 8; i <= 192; i += 8)
3285 size_index[size_index_elem(i)] = 8;
3288 /* Caches that are not of the two-to-the-power-of size */
3289 if (KMALLOC_MIN_SIZE <= 32) {
3290 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3294 if (KMALLOC_MIN_SIZE <= 64) {
3295 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3299 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3300 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3306 /* Provide the correct kmalloc names now that the caches are up */
3307 if (KMALLOC_MIN_SIZE <= 32) {
3308 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3309 BUG_ON(!kmalloc_caches[1]->name);
3312 if (KMALLOC_MIN_SIZE <= 64) {
3313 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3314 BUG_ON(!kmalloc_caches[2]->name);
3317 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3318 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3321 kmalloc_caches[i]->name = s;
3325 register_cpu_notifier(&slab_notifier);
3328 #ifdef CONFIG_ZONE_DMA
3329 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3330 struct kmem_cache *s = kmalloc_caches[i];
3333 char *name = kasprintf(GFP_NOWAIT,
3334 "dma-kmalloc-%d", s->objsize);
3337 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3338 s->objsize, SLAB_CACHE_DMA);
3343 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3344 " CPUs=%d, Nodes=%d\n",
3345 caches, cache_line_size(),
3346 slub_min_order, slub_max_order, slub_min_objects,
3347 nr_cpu_ids, nr_node_ids);
3350 void __init kmem_cache_init_late(void)
3355 * Find a mergeable slab cache
3357 static int slab_unmergeable(struct kmem_cache *s)
3359 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3366 * We may have set a slab to be unmergeable during bootstrap.
3368 if (s->refcount < 0)
3374 static struct kmem_cache *find_mergeable(size_t size,
3375 size_t align, unsigned long flags, const char *name,
3376 void (*ctor)(void *))
3378 struct kmem_cache *s;
3380 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3386 size = ALIGN(size, sizeof(void *));
3387 align = calculate_alignment(flags, align, size);
3388 size = ALIGN(size, align);
3389 flags = kmem_cache_flags(size, flags, name, NULL);
3391 list_for_each_entry(s, &slab_caches, list) {
3392 if (slab_unmergeable(s))
3398 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
3401 * Check if alignment is compatible.
3402 * Courtesy of Adrian Drzewiecki
3404 if ((s->size & ~(align - 1)) != s->size)
3407 if (s->size - size >= sizeof(void *))
3415 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
3416 size_t align, unsigned long flags, void (*ctor)(void *))
3418 struct kmem_cache *s;
3424 down_write(&slub_lock);
3425 s = find_mergeable(size, align, flags, name, ctor);
3429 * Adjust the object sizes so that we clear
3430 * the complete object on kzalloc.
3432 s->objsize = max(s->objsize, (int)size);
3433 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
3435 if (sysfs_slab_alias(s, name)) {
3439 up_write(&slub_lock);
3443 n = kstrdup(name, GFP_KERNEL);
3447 s = kmalloc(kmem_size, GFP_KERNEL);
3449 if (kmem_cache_open(s, n,
3450 size, align, flags, ctor)) {
3451 list_add(&s->list, &slab_caches);
3452 up_write(&slub_lock);
3453 if (sysfs_slab_add(s)) {
3454 down_write(&slub_lock);
3466 up_write(&slub_lock);
3468 if (flags & SLAB_PANIC)
3469 panic("Cannot create slabcache %s\n", name);
3474 EXPORT_SYMBOL(kmem_cache_create);
3478 * Use the cpu notifier to insure that the cpu slabs are flushed when
3481 static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3482 unsigned long action, void *hcpu)
3484 long cpu = (long)hcpu;
3485 struct kmem_cache *s;
3486 unsigned long flags;
3489 case CPU_UP_CANCELED:
3490 case CPU_UP_CANCELED_FROZEN:
3492 case CPU_DEAD_FROZEN:
3493 down_read(&slub_lock);
3494 list_for_each_entry(s, &slab_caches, list) {
3495 local_irq_save(flags);
3496 __flush_cpu_slab(s, cpu);
3497 local_irq_restore(flags);
3499 up_read(&slub_lock);
3507 static struct notifier_block __cpuinitdata slab_notifier = {
3508 .notifier_call = slab_cpuup_callback
3513 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
3515 struct kmem_cache *s;
3518 if (unlikely(size > SLUB_MAX_SIZE))
3519 return kmalloc_large(size, gfpflags);
3521 s = get_slab(size, gfpflags);
3523 if (unlikely(ZERO_OR_NULL_PTR(s)))
3526 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
3528 /* Honor the call site pointer we received. */
3529 trace_kmalloc(caller, ret, size, s->size, gfpflags);
3535 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
3536 int node, unsigned long caller)
3538 struct kmem_cache *s;
3541 if (unlikely(size > SLUB_MAX_SIZE)) {
3542 ret = kmalloc_large_node(size, gfpflags, node);
3544 trace_kmalloc_node(caller, ret,
3545 size, PAGE_SIZE << get_order(size),
3551 s = get_slab(size, gfpflags);
3553 if (unlikely(ZERO_OR_NULL_PTR(s)))
3556 ret = slab_alloc(s, gfpflags, node, caller);
3558 /* Honor the call site pointer we received. */
3559 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
3566 static int count_inuse(struct page *page)
3571 static int count_total(struct page *page)
3573 return page->objects;
3577 #ifdef CONFIG_SLUB_DEBUG
3578 static int validate_slab(struct kmem_cache *s, struct page *page,
3582 void *addr = page_address(page);
3584 if (!check_slab(s, page) ||
3585 !on_freelist(s, page, NULL))
3588 /* Now we know that a valid freelist exists */
3589 bitmap_zero(map, page->objects);
3591 get_map(s, page, map);
3592 for_each_object(p, s, addr, page->objects) {
3593 if (test_bit(slab_index(p, s, addr), map))
3594 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3598 for_each_object(p, s, addr, page->objects)
3599 if (!test_bit(slab_index(p, s, addr), map))
3600 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
3605 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3608 if (slab_trylock(page)) {
3609 validate_slab(s, page, map);
3612 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3616 static int validate_slab_node(struct kmem_cache *s,
3617 struct kmem_cache_node *n, unsigned long *map)
3619 unsigned long count = 0;
3621 unsigned long flags;
3623 spin_lock_irqsave(&n->list_lock, flags);
3625 list_for_each_entry(page, &n->partial, lru) {
3626 validate_slab_slab(s, page, map);
3629 if (count != n->nr_partial)
3630 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3631 "counter=%ld\n", s->name, count, n->nr_partial);
3633 if (!(s->flags & SLAB_STORE_USER))
3636 list_for_each_entry(page, &n->full, lru) {
3637 validate_slab_slab(s, page, map);
3640 if (count != atomic_long_read(&n->nr_slabs))
3641 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3642 "counter=%ld\n", s->name, count,
3643 atomic_long_read(&n->nr_slabs));
3646 spin_unlock_irqrestore(&n->list_lock, flags);
3650 static long validate_slab_cache(struct kmem_cache *s)
3653 unsigned long count = 0;
3654 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3655 sizeof(unsigned long), GFP_KERNEL);
3661 for_each_node_state(node, N_NORMAL_MEMORY) {
3662 struct kmem_cache_node *n = get_node(s, node);
3664 count += validate_slab_node(s, n, map);
3670 * Generate lists of code addresses where slabcache objects are allocated
3675 unsigned long count;
3682 DECLARE_BITMAP(cpus, NR_CPUS);
3688 unsigned long count;
3689 struct location *loc;
3692 static void free_loc_track(struct loc_track *t)
3695 free_pages((unsigned long)t->loc,
3696 get_order(sizeof(struct location) * t->max));
3699 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
3704 order = get_order(sizeof(struct location) * max);
3706 l = (void *)__get_free_pages(flags, order);
3711 memcpy(l, t->loc, sizeof(struct location) * t->count);
3719 static int add_location(struct loc_track *t, struct kmem_cache *s,
3720 const struct track *track)
3722 long start, end, pos;
3724 unsigned long caddr;
3725 unsigned long age = jiffies - track->when;
3731 pos = start + (end - start + 1) / 2;
3734 * There is nothing at "end". If we end up there
3735 * we need to add something to before end.
3740 caddr = t->loc[pos].addr;
3741 if (track->addr == caddr) {
3747 if (age < l->min_time)
3749 if (age > l->max_time)
3752 if (track->pid < l->min_pid)
3753 l->min_pid = track->pid;
3754 if (track->pid > l->max_pid)
3755 l->max_pid = track->pid;
3757 cpumask_set_cpu(track->cpu,
3758 to_cpumask(l->cpus));
3760 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3764 if (track->addr < caddr)
3771 * Not found. Insert new tracking element.
3773 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
3779 (t->count - pos) * sizeof(struct location));
3782 l->addr = track->addr;
3786 l->min_pid = track->pid;
3787 l->max_pid = track->pid;
3788 cpumask_clear(to_cpumask(l->cpus));
3789 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
3790 nodes_clear(l->nodes);
3791 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3795 static void process_slab(struct loc_track *t, struct kmem_cache *s,
3796 struct page *page, enum track_item alloc,
3799 void *addr = page_address(page);
3802 bitmap_zero(map, page->objects);
3803 get_map(s, page, map);
3805 for_each_object(p, s, addr, page->objects)
3806 if (!test_bit(slab_index(p, s, addr), map))
3807 add_location(t, s, get_track(s, p, alloc));
3810 static int list_locations(struct kmem_cache *s, char *buf,
3811 enum track_item alloc)
3815 struct loc_track t = { 0, 0, NULL };
3817 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3818 sizeof(unsigned long), GFP_KERNEL);
3820 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3823 return sprintf(buf, "Out of memory\n");
3825 /* Push back cpu slabs */
3828 for_each_node_state(node, N_NORMAL_MEMORY) {
3829 struct kmem_cache_node *n = get_node(s, node);
3830 unsigned long flags;
3833 if (!atomic_long_read(&n->nr_slabs))
3836 spin_lock_irqsave(&n->list_lock, flags);
3837 list_for_each_entry(page, &n->partial, lru)
3838 process_slab(&t, s, page, alloc, map);
3839 list_for_each_entry(page, &n->full, lru)
3840 process_slab(&t, s, page, alloc, map);
3841 spin_unlock_irqrestore(&n->list_lock, flags);
3844 for (i = 0; i < t.count; i++) {
3845 struct location *l = &t.loc[i];
3847 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
3849 len += sprintf(buf + len, "%7ld ", l->count);
3852 len += sprintf(buf + len, "%pS", (void *)l->addr);
3854 len += sprintf(buf + len, "<not-available>");
3856 if (l->sum_time != l->min_time) {
3857 len += sprintf(buf + len, " age=%ld/%ld/%ld",
3859 (long)div_u64(l->sum_time, l->count),
3862 len += sprintf(buf + len, " age=%ld",
3865 if (l->min_pid != l->max_pid)
3866 len += sprintf(buf + len, " pid=%ld-%ld",
3867 l->min_pid, l->max_pid);
3869 len += sprintf(buf + len, " pid=%ld",
3872 if (num_online_cpus() > 1 &&
3873 !cpumask_empty(to_cpumask(l->cpus)) &&
3874 len < PAGE_SIZE - 60) {
3875 len += sprintf(buf + len, " cpus=");
3876 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3877 to_cpumask(l->cpus));
3880 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
3881 len < PAGE_SIZE - 60) {
3882 len += sprintf(buf + len, " nodes=");
3883 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3887 len += sprintf(buf + len, "\n");
3893 len += sprintf(buf, "No data\n");
3898 #ifdef SLUB_RESILIENCY_TEST
3899 static void resiliency_test(void)
3903 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3905 printk(KERN_ERR "SLUB resiliency testing\n");
3906 printk(KERN_ERR "-----------------------\n");
3907 printk(KERN_ERR "A. Corruption after allocation\n");
3909 p = kzalloc(16, GFP_KERNEL);
3911 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3912 " 0x12->0x%p\n\n", p + 16);
3914 validate_slab_cache(kmalloc_caches[4]);
3916 /* Hmmm... The next two are dangerous */
3917 p = kzalloc(32, GFP_KERNEL);
3918 p[32 + sizeof(void *)] = 0x34;
3919 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3920 " 0x34 -> -0x%p\n", p);
3922 "If allocated object is overwritten then not detectable\n\n");
3924 validate_slab_cache(kmalloc_caches[5]);
3925 p = kzalloc(64, GFP_KERNEL);
3926 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3928 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3931 "If allocated object is overwritten then not detectable\n\n");
3932 validate_slab_cache(kmalloc_caches[6]);
3934 printk(KERN_ERR "\nB. Corruption after free\n");
3935 p = kzalloc(128, GFP_KERNEL);
3938 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3939 validate_slab_cache(kmalloc_caches[7]);
3941 p = kzalloc(256, GFP_KERNEL);
3944 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3946 validate_slab_cache(kmalloc_caches[8]);
3948 p = kzalloc(512, GFP_KERNEL);
3951 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3952 validate_slab_cache(kmalloc_caches[9]);
3956 static void resiliency_test(void) {};
3961 enum slab_stat_type {
3962 SL_ALL, /* All slabs */
3963 SL_PARTIAL, /* Only partially allocated slabs */
3964 SL_CPU, /* Only slabs used for cpu caches */
3965 SL_OBJECTS, /* Determine allocated objects not slabs */
3966 SL_TOTAL /* Determine object capacity not slabs */
3969 #define SO_ALL (1 << SL_ALL)
3970 #define SO_PARTIAL (1 << SL_PARTIAL)
3971 #define SO_CPU (1 << SL_CPU)
3972 #define SO_OBJECTS (1 << SL_OBJECTS)
3973 #define SO_TOTAL (1 << SL_TOTAL)
3975 static ssize_t show_slab_objects(struct kmem_cache *s,
3976 char *buf, unsigned long flags)
3978 unsigned long total = 0;
3981 unsigned long *nodes;
3982 unsigned long *per_cpu;
3984 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
3987 per_cpu = nodes + nr_node_ids;
3989 if (flags & SO_CPU) {
3992 for_each_possible_cpu(cpu) {
3993 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3995 if (!c || c->node < 0)
3999 if (flags & SO_TOTAL)
4000 x = c->page->objects;
4001 else if (flags & SO_OBJECTS)
4007 nodes[c->node] += x;
4013 lock_memory_hotplug();
4014 #ifdef CONFIG_SLUB_DEBUG
4015 if (flags & SO_ALL) {
4016 for_each_node_state(node, N_NORMAL_MEMORY) {
4017 struct kmem_cache_node *n = get_node(s, node);
4019 if (flags & SO_TOTAL)
4020 x = atomic_long_read(&n->total_objects);
4021 else if (flags & SO_OBJECTS)
4022 x = atomic_long_read(&n->total_objects) -
4023 count_partial(n, count_free);
4026 x = atomic_long_read(&n->nr_slabs);
4033 if (flags & SO_PARTIAL) {
4034 for_each_node_state(node, N_NORMAL_MEMORY) {
4035 struct kmem_cache_node *n = get_node(s, node);
4037 if (flags & SO_TOTAL)
4038 x = count_partial(n, count_total);
4039 else if (flags & SO_OBJECTS)
4040 x = count_partial(n, count_inuse);
4047 x = sprintf(buf, "%lu", total);
4049 for_each_node_state(node, N_NORMAL_MEMORY)
4051 x += sprintf(buf + x, " N%d=%lu",
4054 unlock_memory_hotplug();
4056 return x + sprintf(buf + x, "\n");
4059 #ifdef CONFIG_SLUB_DEBUG
4060 static int any_slab_objects(struct kmem_cache *s)
4064 for_each_online_node(node) {
4065 struct kmem_cache_node *n = get_node(s, node);
4070 if (atomic_long_read(&n->total_objects))
4077 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4078 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
4080 struct slab_attribute {
4081 struct attribute attr;
4082 ssize_t (*show)(struct kmem_cache *s, char *buf);
4083 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4086 #define SLAB_ATTR_RO(_name) \
4087 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4089 #define SLAB_ATTR(_name) \
4090 static struct slab_attribute _name##_attr = \
4091 __ATTR(_name, 0644, _name##_show, _name##_store)
4093 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4095 return sprintf(buf, "%d\n", s->size);
4097 SLAB_ATTR_RO(slab_size);
4099 static ssize_t align_show(struct kmem_cache *s, char *buf)
4101 return sprintf(buf, "%d\n", s->align);
4103 SLAB_ATTR_RO(align);
4105 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4107 return sprintf(buf, "%d\n", s->objsize);
4109 SLAB_ATTR_RO(object_size);
4111 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4113 return sprintf(buf, "%d\n", oo_objects(s->oo));
4115 SLAB_ATTR_RO(objs_per_slab);
4117 static ssize_t order_store(struct kmem_cache *s,
4118 const char *buf, size_t length)
4120 unsigned long order;
4123 err = strict_strtoul(buf, 10, &order);
4127 if (order > slub_max_order || order < slub_min_order)
4130 calculate_sizes(s, order);
4134 static ssize_t order_show(struct kmem_cache *s, char *buf)
4136 return sprintf(buf, "%d\n", oo_order(s->oo));
4140 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4142 return sprintf(buf, "%lu\n", s->min_partial);
4145 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4151 err = strict_strtoul(buf, 10, &min);
4155 set_min_partial(s, min);
4158 SLAB_ATTR(min_partial);
4160 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4164 return sprintf(buf, "%pS\n", s->ctor);
4168 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4170 return sprintf(buf, "%d\n", s->refcount - 1);
4172 SLAB_ATTR_RO(aliases);
4174 static ssize_t partial_show(struct kmem_cache *s, char *buf)
4176 return show_slab_objects(s, buf, SO_PARTIAL);
4178 SLAB_ATTR_RO(partial);
4180 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4182 return show_slab_objects(s, buf, SO_CPU);
4184 SLAB_ATTR_RO(cpu_slabs);
4186 static ssize_t objects_show(struct kmem_cache *s, char *buf)
4188 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
4190 SLAB_ATTR_RO(objects);
4192 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4194 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4196 SLAB_ATTR_RO(objects_partial);
4198 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4200 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4203 static ssize_t reclaim_account_store(struct kmem_cache *s,
4204 const char *buf, size_t length)
4206 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4208 s->flags |= SLAB_RECLAIM_ACCOUNT;
4211 SLAB_ATTR(reclaim_account);
4213 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4215 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
4217 SLAB_ATTR_RO(hwcache_align);
4219 #ifdef CONFIG_ZONE_DMA
4220 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4222 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4224 SLAB_ATTR_RO(cache_dma);
4227 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4229 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4231 SLAB_ATTR_RO(destroy_by_rcu);
4233 static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4235 return sprintf(buf, "%d\n", s->reserved);
4237 SLAB_ATTR_RO(reserved);
4239 #ifdef CONFIG_SLUB_DEBUG
4240 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4242 return show_slab_objects(s, buf, SO_ALL);
4244 SLAB_ATTR_RO(slabs);
4246 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4248 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4250 SLAB_ATTR_RO(total_objects);
4252 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4254 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4257 static ssize_t sanity_checks_store(struct kmem_cache *s,
4258 const char *buf, size_t length)
4260 s->flags &= ~SLAB_DEBUG_FREE;
4262 s->flags |= SLAB_DEBUG_FREE;
4265 SLAB_ATTR(sanity_checks);
4267 static ssize_t trace_show(struct kmem_cache *s, char *buf)
4269 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4272 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4275 s->flags &= ~SLAB_TRACE;
4277 s->flags |= SLAB_TRACE;
4282 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4284 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4287 static ssize_t red_zone_store(struct kmem_cache *s,
4288 const char *buf, size_t length)
4290 if (any_slab_objects(s))
4293 s->flags &= ~SLAB_RED_ZONE;
4295 s->flags |= SLAB_RED_ZONE;
4296 calculate_sizes(s, -1);
4299 SLAB_ATTR(red_zone);
4301 static ssize_t poison_show(struct kmem_cache *s, char *buf)
4303 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4306 static ssize_t poison_store(struct kmem_cache *s,
4307 const char *buf, size_t length)
4309 if (any_slab_objects(s))
4312 s->flags &= ~SLAB_POISON;
4314 s->flags |= SLAB_POISON;
4315 calculate_sizes(s, -1);
4320 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4322 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4325 static ssize_t store_user_store(struct kmem_cache *s,
4326 const char *buf, size_t length)
4328 if (any_slab_objects(s))
4331 s->flags &= ~SLAB_STORE_USER;
4333 s->flags |= SLAB_STORE_USER;
4334 calculate_sizes(s, -1);
4337 SLAB_ATTR(store_user);
4339 static ssize_t validate_show(struct kmem_cache *s, char *buf)
4344 static ssize_t validate_store(struct kmem_cache *s,
4345 const char *buf, size_t length)
4349 if (buf[0] == '1') {
4350 ret = validate_slab_cache(s);
4356 SLAB_ATTR(validate);
4358 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4360 if (!(s->flags & SLAB_STORE_USER))
4362 return list_locations(s, buf, TRACK_ALLOC);
4364 SLAB_ATTR_RO(alloc_calls);
4366 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4368 if (!(s->flags & SLAB_STORE_USER))
4370 return list_locations(s, buf, TRACK_FREE);
4372 SLAB_ATTR_RO(free_calls);
4373 #endif /* CONFIG_SLUB_DEBUG */
4375 #ifdef CONFIG_FAILSLAB
4376 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4378 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4381 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4384 s->flags &= ~SLAB_FAILSLAB;
4386 s->flags |= SLAB_FAILSLAB;
4389 SLAB_ATTR(failslab);
4392 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4397 static ssize_t shrink_store(struct kmem_cache *s,
4398 const char *buf, size_t length)
4400 if (buf[0] == '1') {
4401 int rc = kmem_cache_shrink(s);
4412 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
4414 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
4417 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
4418 const char *buf, size_t length)
4420 unsigned long ratio;
4423 err = strict_strtoul(buf, 10, &ratio);
4428 s->remote_node_defrag_ratio = ratio * 10;
4432 SLAB_ATTR(remote_node_defrag_ratio);
4435 #ifdef CONFIG_SLUB_STATS
4436 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4438 unsigned long sum = 0;
4441 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4446 for_each_online_cpu(cpu) {
4447 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
4453 len = sprintf(buf, "%lu", sum);
4456 for_each_online_cpu(cpu) {
4457 if (data[cpu] && len < PAGE_SIZE - 20)
4458 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
4462 return len + sprintf(buf + len, "\n");
4465 static void clear_stat(struct kmem_cache *s, enum stat_item si)
4469 for_each_online_cpu(cpu)
4470 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
4473 #define STAT_ATTR(si, text) \
4474 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4476 return show_stat(s, buf, si); \
4478 static ssize_t text##_store(struct kmem_cache *s, \
4479 const char *buf, size_t length) \
4481 if (buf[0] != '0') \
4483 clear_stat(s, si); \
4488 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4489 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4490 STAT_ATTR(FREE_FASTPATH, free_fastpath);
4491 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4492 STAT_ATTR(FREE_FROZEN, free_frozen);
4493 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4494 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4495 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4496 STAT_ATTR(ALLOC_SLAB, alloc_slab);
4497 STAT_ATTR(ALLOC_REFILL, alloc_refill);
4498 STAT_ATTR(FREE_SLAB, free_slab);
4499 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4500 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4501 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4502 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4503 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4504 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
4505 STAT_ATTR(ORDER_FALLBACK, order_fallback);
4508 static struct attribute *slab_attrs[] = {
4509 &slab_size_attr.attr,
4510 &object_size_attr.attr,
4511 &objs_per_slab_attr.attr,
4513 &min_partial_attr.attr,
4515 &objects_partial_attr.attr,
4517 &cpu_slabs_attr.attr,
4521 &hwcache_align_attr.attr,
4522 &reclaim_account_attr.attr,
4523 &destroy_by_rcu_attr.attr,
4525 &reserved_attr.attr,
4526 #ifdef CONFIG_SLUB_DEBUG
4527 &total_objects_attr.attr,
4529 &sanity_checks_attr.attr,
4531 &red_zone_attr.attr,
4533 &store_user_attr.attr,
4534 &validate_attr.attr,
4535 &alloc_calls_attr.attr,
4536 &free_calls_attr.attr,
4538 #ifdef CONFIG_ZONE_DMA
4539 &cache_dma_attr.attr,
4542 &remote_node_defrag_ratio_attr.attr,
4544 #ifdef CONFIG_SLUB_STATS
4545 &alloc_fastpath_attr.attr,
4546 &alloc_slowpath_attr.attr,
4547 &free_fastpath_attr.attr,
4548 &free_slowpath_attr.attr,
4549 &free_frozen_attr.attr,
4550 &free_add_partial_attr.attr,
4551 &free_remove_partial_attr.attr,
4552 &alloc_from_partial_attr.attr,
4553 &alloc_slab_attr.attr,
4554 &alloc_refill_attr.attr,
4555 &free_slab_attr.attr,
4556 &cpuslab_flush_attr.attr,
4557 &deactivate_full_attr.attr,
4558 &deactivate_empty_attr.attr,
4559 &deactivate_to_head_attr.attr,
4560 &deactivate_to_tail_attr.attr,
4561 &deactivate_remote_frees_attr.attr,
4562 &order_fallback_attr.attr,
4564 #ifdef CONFIG_FAILSLAB
4565 &failslab_attr.attr,
4571 static struct attribute_group slab_attr_group = {
4572 .attrs = slab_attrs,
4575 static ssize_t slab_attr_show(struct kobject *kobj,
4576 struct attribute *attr,
4579 struct slab_attribute *attribute;
4580 struct kmem_cache *s;
4583 attribute = to_slab_attr(attr);
4586 if (!attribute->show)
4589 err = attribute->show(s, buf);
4594 static ssize_t slab_attr_store(struct kobject *kobj,
4595 struct attribute *attr,
4596 const char *buf, size_t len)
4598 struct slab_attribute *attribute;
4599 struct kmem_cache *s;
4602 attribute = to_slab_attr(attr);
4605 if (!attribute->store)
4608 err = attribute->store(s, buf, len);
4613 static void kmem_cache_release(struct kobject *kobj)
4615 struct kmem_cache *s = to_slab(kobj);
4621 static const struct sysfs_ops slab_sysfs_ops = {
4622 .show = slab_attr_show,
4623 .store = slab_attr_store,
4626 static struct kobj_type slab_ktype = {
4627 .sysfs_ops = &slab_sysfs_ops,
4628 .release = kmem_cache_release
4631 static int uevent_filter(struct kset *kset, struct kobject *kobj)
4633 struct kobj_type *ktype = get_ktype(kobj);
4635 if (ktype == &slab_ktype)
4640 static const struct kset_uevent_ops slab_uevent_ops = {
4641 .filter = uevent_filter,
4644 static struct kset *slab_kset;
4646 #define ID_STR_LENGTH 64
4648 /* Create a unique string id for a slab cache:
4650 * Format :[flags-]size
4652 static char *create_unique_id(struct kmem_cache *s)
4654 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4661 * First flags affecting slabcache operations. We will only
4662 * get here for aliasable slabs so we do not need to support
4663 * too many flags. The flags here must cover all flags that
4664 * are matched during merging to guarantee that the id is
4667 if (s->flags & SLAB_CACHE_DMA)
4669 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4671 if (s->flags & SLAB_DEBUG_FREE)
4673 if (!(s->flags & SLAB_NOTRACK))
4677 p += sprintf(p, "%07d", s->size);
4678 BUG_ON(p > name + ID_STR_LENGTH - 1);
4682 static int sysfs_slab_add(struct kmem_cache *s)
4688 if (slab_state < SYSFS)
4689 /* Defer until later */
4692 unmergeable = slab_unmergeable(s);
4695 * Slabcache can never be merged so we can use the name proper.
4696 * This is typically the case for debug situations. In that
4697 * case we can catch duplicate names easily.
4699 sysfs_remove_link(&slab_kset->kobj, s->name);
4703 * Create a unique name for the slab as a target
4706 name = create_unique_id(s);
4709 s->kobj.kset = slab_kset;
4710 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4712 kobject_put(&s->kobj);
4716 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4718 kobject_del(&s->kobj);
4719 kobject_put(&s->kobj);
4722 kobject_uevent(&s->kobj, KOBJ_ADD);
4724 /* Setup first alias */
4725 sysfs_slab_alias(s, s->name);
4731 static void sysfs_slab_remove(struct kmem_cache *s)
4733 if (slab_state < SYSFS)
4735 * Sysfs has not been setup yet so no need to remove the
4740 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4741 kobject_del(&s->kobj);
4742 kobject_put(&s->kobj);
4746 * Need to buffer aliases during bootup until sysfs becomes
4747 * available lest we lose that information.
4749 struct saved_alias {
4750 struct kmem_cache *s;
4752 struct saved_alias *next;
4755 static struct saved_alias *alias_list;
4757 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4759 struct saved_alias *al;
4761 if (slab_state == SYSFS) {
4763 * If we have a leftover link then remove it.
4765 sysfs_remove_link(&slab_kset->kobj, name);
4766 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
4769 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4775 al->next = alias_list;
4780 static int __init slab_sysfs_init(void)
4782 struct kmem_cache *s;
4785 down_write(&slub_lock);
4787 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
4789 up_write(&slub_lock);
4790 printk(KERN_ERR "Cannot register slab subsystem.\n");
4796 list_for_each_entry(s, &slab_caches, list) {
4797 err = sysfs_slab_add(s);
4799 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4800 " to sysfs\n", s->name);
4803 while (alias_list) {
4804 struct saved_alias *al = alias_list;
4806 alias_list = alias_list->next;
4807 err = sysfs_slab_alias(al->s, al->name);
4809 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4810 " %s to sysfs\n", s->name);
4814 up_write(&slub_lock);
4819 __initcall(slab_sysfs_init);
4820 #endif /* CONFIG_SYSFS */
4823 * The /proc/slabinfo ABI
4825 #ifdef CONFIG_SLABINFO
4826 static void print_slabinfo_header(struct seq_file *m)
4828 seq_puts(m, "slabinfo - version: 2.1\n");
4829 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4830 "<objperslab> <pagesperslab>");
4831 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4832 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4836 static void *s_start(struct seq_file *m, loff_t *pos)
4840 down_read(&slub_lock);
4842 print_slabinfo_header(m);
4844 return seq_list_start(&slab_caches, *pos);
4847 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4849 return seq_list_next(p, &slab_caches, pos);
4852 static void s_stop(struct seq_file *m, void *p)
4854 up_read(&slub_lock);
4857 static int s_show(struct seq_file *m, void *p)
4859 unsigned long nr_partials = 0;
4860 unsigned long nr_slabs = 0;
4861 unsigned long nr_inuse = 0;
4862 unsigned long nr_objs = 0;
4863 unsigned long nr_free = 0;
4864 struct kmem_cache *s;
4867 s = list_entry(p, struct kmem_cache, list);
4869 for_each_online_node(node) {
4870 struct kmem_cache_node *n = get_node(s, node);
4875 nr_partials += n->nr_partial;
4876 nr_slabs += atomic_long_read(&n->nr_slabs);
4877 nr_objs += atomic_long_read(&n->total_objects);
4878 nr_free += count_partial(n, count_free);
4881 nr_inuse = nr_objs - nr_free;
4883 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
4884 nr_objs, s->size, oo_objects(s->oo),
4885 (1 << oo_order(s->oo)));
4886 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4887 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4893 static const struct seq_operations slabinfo_op = {
4900 static int slabinfo_open(struct inode *inode, struct file *file)
4902 return seq_open(file, &slabinfo_op);
4905 static const struct file_operations proc_slabinfo_operations = {
4906 .open = slabinfo_open,
4908 .llseek = seq_lseek,
4909 .release = seq_release,
4912 static int __init slab_proc_init(void)
4914 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
4917 module_init(slab_proc_init);
4918 #endif /* CONFIG_SLABINFO */