2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemcheck.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/mempolicy.h>
24 #include <linux/ctype.h>
25 #include <linux/debugobjects.h>
26 #include <linux/kallsyms.h>
27 #include <linux/memory.h>
28 #include <linux/math64.h>
29 #include <linux/fault-inject.h>
31 #include <trace/events/kmem.h>
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache *s)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
135 * Mininum number of partial slabs. These will be left on the partial
136 * lists even if they are empty. kmem_cache_shrink may reclaim them.
138 #define MIN_PARTIAL 5
141 * Maximum number of desirable partial slabs.
142 * The existence of more partial slabs makes kmem_cache_shrink
143 * sort the partial list by the number of objects in the.
145 #define MAX_PARTIAL 10
147 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
148 SLAB_POISON | SLAB_STORE_USER)
151 * Debugging flags that require metadata to be stored in the slab. These get
152 * disabled when slub_debug=O is used and a cache's min order increases with
155 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
158 * Set of flags that will prevent slab merging
160 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
161 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
164 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
165 SLAB_CACHE_DMA | SLAB_NOTRACK)
168 #define OO_MASK ((1 << OO_SHIFT) - 1)
169 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
171 /* Internal SLUB flags */
172 #define __OBJECT_POISON 0x80000000UL /* Poison object */
174 static int kmem_size = sizeof(struct kmem_cache);
177 static struct notifier_block slab_notifier;
181 DOWN, /* No slab functionality available */
182 PARTIAL, /* Kmem_cache_node works */
183 UP, /* Everything works but does not show up in sysfs */
187 /* A list of all slab caches on the system */
188 static DECLARE_RWSEM(slub_lock);
189 static LIST_HEAD(slab_caches);
192 * Tracking user of a slab.
195 unsigned long addr; /* Called from address */
196 int cpu; /* Was running on cpu */
197 int pid; /* Pid context */
198 unsigned long when; /* When did the operation occur */
201 enum track_item { TRACK_ALLOC, TRACK_FREE };
204 static int sysfs_slab_add(struct kmem_cache *);
205 static int sysfs_slab_alias(struct kmem_cache *, const char *);
206 static void sysfs_slab_remove(struct kmem_cache *);
209 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
210 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
212 static inline void sysfs_slab_remove(struct kmem_cache *s)
220 static inline void stat(const struct kmem_cache *s, enum stat_item si)
222 #ifdef CONFIG_SLUB_STATS
223 __this_cpu_inc(s->cpu_slab->stat[si]);
227 /********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
231 int slab_is_available(void)
233 return slab_state >= UP;
236 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
238 return s->node[node];
241 /* Verify that a pointer has an address that is valid within a slab page */
242 static inline int check_valid_pointer(struct kmem_cache *s,
243 struct page *page, const void *object)
250 base = page_address(page);
251 if (object < base || object >= base + page->objects * s->size ||
252 (object - base) % s->size) {
259 static inline void *get_freepointer(struct kmem_cache *s, void *object)
261 return *(void **)(object + s->offset);
264 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
266 *(void **)(object + s->offset) = fp;
269 /* Loop over all objects in a slab */
270 #define for_each_object(__p, __s, __addr, __objects) \
271 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
274 /* Determine object index from a given position */
275 static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
277 return (p - addr) / s->size;
280 static inline size_t slab_ksize(const struct kmem_cache *s)
282 #ifdef CONFIG_SLUB_DEBUG
284 * Debugging requires use of the padding between object
285 * and whatever may come after it.
287 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
292 * If we have the need to store the freelist pointer
293 * back there or track user information then we can
294 * only use the space before that information.
296 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
299 * Else we can use all the padding etc for the allocation
304 static inline int order_objects(int order, unsigned long size, int reserved)
306 return ((PAGE_SIZE << order) - reserved) / size;
309 static inline struct kmem_cache_order_objects oo_make(int order,
310 unsigned long size, int reserved)
312 struct kmem_cache_order_objects x = {
313 (order << OO_SHIFT) + order_objects(order, size, reserved)
319 static inline int oo_order(struct kmem_cache_order_objects x)
321 return x.x >> OO_SHIFT;
324 static inline int oo_objects(struct kmem_cache_order_objects x)
326 return x.x & OO_MASK;
330 * Determine a map of object in use on a page.
332 * Slab lock or node listlock must be held to guarantee that the page does
333 * not vanish from under us.
335 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
338 void *addr = page_address(page);
340 for (p = page->freelist; p; p = get_freepointer(s, p))
341 set_bit(slab_index(p, s, addr), map);
344 #ifdef CONFIG_SLUB_DEBUG
348 #ifdef CONFIG_SLUB_DEBUG_ON
349 static int slub_debug = DEBUG_DEFAULT_FLAGS;
351 static int slub_debug;
354 static char *slub_debug_slabs;
355 static int disable_higher_order_debug;
360 static void print_section(char *text, u8 *addr, unsigned int length)
368 for (i = 0; i < length; i++) {
370 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
373 printk(KERN_CONT " %02x", addr[i]);
375 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
377 printk(KERN_CONT " %s\n", ascii);
384 printk(KERN_CONT " ");
388 printk(KERN_CONT " %s\n", ascii);
392 static struct track *get_track(struct kmem_cache *s, void *object,
393 enum track_item alloc)
398 p = object + s->offset + sizeof(void *);
400 p = object + s->inuse;
405 static void set_track(struct kmem_cache *s, void *object,
406 enum track_item alloc, unsigned long addr)
408 struct track *p = get_track(s, object, alloc);
412 p->cpu = smp_processor_id();
413 p->pid = current->pid;
416 memset(p, 0, sizeof(struct track));
419 static void init_tracking(struct kmem_cache *s, void *object)
421 if (!(s->flags & SLAB_STORE_USER))
424 set_track(s, object, TRACK_FREE, 0UL);
425 set_track(s, object, TRACK_ALLOC, 0UL);
428 static void print_track(const char *s, struct track *t)
433 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
434 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
437 static void print_tracking(struct kmem_cache *s, void *object)
439 if (!(s->flags & SLAB_STORE_USER))
442 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
443 print_track("Freed", get_track(s, object, TRACK_FREE));
446 static void print_page_info(struct page *page)
448 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
449 page, page->objects, page->inuse, page->freelist, page->flags);
453 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
459 vsnprintf(buf, sizeof(buf), fmt, args);
461 printk(KERN_ERR "========================================"
462 "=====================================\n");
463 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
464 printk(KERN_ERR "----------------------------------------"
465 "-------------------------------------\n\n");
468 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
474 vsnprintf(buf, sizeof(buf), fmt, args);
476 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
479 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
481 unsigned int off; /* Offset of last byte */
482 u8 *addr = page_address(page);
484 print_tracking(s, p);
486 print_page_info(page);
488 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
489 p, p - addr, get_freepointer(s, p));
492 print_section("Bytes b4", p - 16, 16);
494 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
496 if (s->flags & SLAB_RED_ZONE)
497 print_section("Redzone", p + s->objsize,
498 s->inuse - s->objsize);
501 off = s->offset + sizeof(void *);
505 if (s->flags & SLAB_STORE_USER)
506 off += 2 * sizeof(struct track);
509 /* Beginning of the filler is the free pointer */
510 print_section("Padding", p + off, s->size - off);
515 static void object_err(struct kmem_cache *s, struct page *page,
516 u8 *object, char *reason)
518 slab_bug(s, "%s", reason);
519 print_trailer(s, page, object);
522 static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
528 vsnprintf(buf, sizeof(buf), fmt, args);
530 slab_bug(s, "%s", buf);
531 print_page_info(page);
535 static void init_object(struct kmem_cache *s, void *object, u8 val)
539 if (s->flags & __OBJECT_POISON) {
540 memset(p, POISON_FREE, s->objsize - 1);
541 p[s->objsize - 1] = POISON_END;
544 if (s->flags & SLAB_RED_ZONE)
545 memset(p + s->objsize, val, s->inuse - s->objsize);
548 static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
551 if (*start != (u8)value)
559 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
560 void *from, void *to)
562 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
563 memset(from, data, to - from);
566 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
567 u8 *object, char *what,
568 u8 *start, unsigned int value, unsigned int bytes)
573 fault = check_bytes(start, value, bytes);
578 while (end > fault && end[-1] == value)
581 slab_bug(s, "%s overwritten", what);
582 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
583 fault, end - 1, fault[0], value);
584 print_trailer(s, page, object);
586 restore_bytes(s, what, value, fault, end);
594 * Bytes of the object to be managed.
595 * If the freepointer may overlay the object then the free
596 * pointer is the first word of the object.
598 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
601 * object + s->objsize
602 * Padding to reach word boundary. This is also used for Redzoning.
603 * Padding is extended by another word if Redzoning is enabled and
606 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
607 * 0xcc (RED_ACTIVE) for objects in use.
610 * Meta data starts here.
612 * A. Free pointer (if we cannot overwrite object on free)
613 * B. Tracking data for SLAB_STORE_USER
614 * C. Padding to reach required alignment boundary or at mininum
615 * one word if debugging is on to be able to detect writes
616 * before the word boundary.
618 * Padding is done using 0x5a (POISON_INUSE)
621 * Nothing is used beyond s->size.
623 * If slabcaches are merged then the objsize and inuse boundaries are mostly
624 * ignored. And therefore no slab options that rely on these boundaries
625 * may be used with merged slabcaches.
628 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
630 unsigned long off = s->inuse; /* The end of info */
633 /* Freepointer is placed after the object. */
634 off += sizeof(void *);
636 if (s->flags & SLAB_STORE_USER)
637 /* We also have user information there */
638 off += 2 * sizeof(struct track);
643 return check_bytes_and_report(s, page, p, "Object padding",
644 p + off, POISON_INUSE, s->size - off);
647 /* Check the pad bytes at the end of a slab page */
648 static int slab_pad_check(struct kmem_cache *s, struct page *page)
656 if (!(s->flags & SLAB_POISON))
659 start = page_address(page);
660 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
661 end = start + length;
662 remainder = length % s->size;
666 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
669 while (end > fault && end[-1] == POISON_INUSE)
672 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
673 print_section("Padding", end - remainder, remainder);
675 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
679 static int check_object(struct kmem_cache *s, struct page *page,
680 void *object, u8 val)
683 u8 *endobject = object + s->objsize;
685 if (s->flags & SLAB_RED_ZONE) {
686 if (!check_bytes_and_report(s, page, object, "Redzone",
687 endobject, val, s->inuse - s->objsize))
690 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
691 check_bytes_and_report(s, page, p, "Alignment padding",
692 endobject, POISON_INUSE, s->inuse - s->objsize);
696 if (s->flags & SLAB_POISON) {
697 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
698 (!check_bytes_and_report(s, page, p, "Poison", p,
699 POISON_FREE, s->objsize - 1) ||
700 !check_bytes_and_report(s, page, p, "Poison",
701 p + s->objsize - 1, POISON_END, 1)))
704 * check_pad_bytes cleans up on its own.
706 check_pad_bytes(s, page, p);
709 if (!s->offset && val == SLUB_RED_ACTIVE)
711 * Object and freepointer overlap. Cannot check
712 * freepointer while object is allocated.
716 /* Check free pointer validity */
717 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
718 object_err(s, page, p, "Freepointer corrupt");
720 * No choice but to zap it and thus lose the remainder
721 * of the free objects in this slab. May cause
722 * another error because the object count is now wrong.
724 set_freepointer(s, p, NULL);
730 static int check_slab(struct kmem_cache *s, struct page *page)
734 VM_BUG_ON(!irqs_disabled());
736 if (!PageSlab(page)) {
737 slab_err(s, page, "Not a valid slab page");
741 maxobj = order_objects(compound_order(page), s->size, s->reserved);
742 if (page->objects > maxobj) {
743 slab_err(s, page, "objects %u > max %u",
744 s->name, page->objects, maxobj);
747 if (page->inuse > page->objects) {
748 slab_err(s, page, "inuse %u > max %u",
749 s->name, page->inuse, page->objects);
752 /* Slab_pad_check fixes things up after itself */
753 slab_pad_check(s, page);
758 * Determine if a certain object on a page is on the freelist. Must hold the
759 * slab lock to guarantee that the chains are in a consistent state.
761 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
764 void *fp = page->freelist;
766 unsigned long max_objects;
768 while (fp && nr <= page->objects) {
771 if (!check_valid_pointer(s, page, fp)) {
773 object_err(s, page, object,
774 "Freechain corrupt");
775 set_freepointer(s, object, NULL);
778 slab_err(s, page, "Freepointer corrupt");
779 page->freelist = NULL;
780 page->inuse = page->objects;
781 slab_fix(s, "Freelist cleared");
787 fp = get_freepointer(s, object);
791 max_objects = order_objects(compound_order(page), s->size, s->reserved);
792 if (max_objects > MAX_OBJS_PER_PAGE)
793 max_objects = MAX_OBJS_PER_PAGE;
795 if (page->objects != max_objects) {
796 slab_err(s, page, "Wrong number of objects. Found %d but "
797 "should be %d", page->objects, max_objects);
798 page->objects = max_objects;
799 slab_fix(s, "Number of objects adjusted.");
801 if (page->inuse != page->objects - nr) {
802 slab_err(s, page, "Wrong object count. Counter is %d but "
803 "counted were %d", page->inuse, page->objects - nr);
804 page->inuse = page->objects - nr;
805 slab_fix(s, "Object count adjusted.");
807 return search == NULL;
810 static void trace(struct kmem_cache *s, struct page *page, void *object,
813 if (s->flags & SLAB_TRACE) {
814 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
816 alloc ? "alloc" : "free",
821 print_section("Object", (void *)object, s->objsize);
828 * Hooks for other subsystems that check memory allocations. In a typical
829 * production configuration these hooks all should produce no code at all.
831 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
833 flags &= gfp_allowed_mask;
834 lockdep_trace_alloc(flags);
835 might_sleep_if(flags & __GFP_WAIT);
837 return should_failslab(s->objsize, flags, s->flags);
840 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
842 flags &= gfp_allowed_mask;
843 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
844 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
847 static inline void slab_free_hook(struct kmem_cache *s, void *x)
849 kmemleak_free_recursive(x, s->flags);
852 * Trouble is that we may no longer disable interupts in the fast path
853 * So in order to make the debug calls that expect irqs to be
854 * disabled we need to disable interrupts temporarily.
856 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
860 local_irq_save(flags);
861 kmemcheck_slab_free(s, x, s->objsize);
862 debug_check_no_locks_freed(x, s->objsize);
863 local_irq_restore(flags);
866 if (!(s->flags & SLAB_DEBUG_OBJECTS))
867 debug_check_no_obj_freed(x, s->objsize);
871 * Tracking of fully allocated slabs for debugging purposes.
873 static void add_full(struct kmem_cache_node *n, struct page *page)
875 spin_lock(&n->list_lock);
876 list_add(&page->lru, &n->full);
877 spin_unlock(&n->list_lock);
880 static void remove_full(struct kmem_cache *s, struct page *page)
882 struct kmem_cache_node *n;
884 if (!(s->flags & SLAB_STORE_USER))
887 n = get_node(s, page_to_nid(page));
889 spin_lock(&n->list_lock);
890 list_del(&page->lru);
891 spin_unlock(&n->list_lock);
894 /* Tracking of the number of slabs for debugging purposes */
895 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
897 struct kmem_cache_node *n = get_node(s, node);
899 return atomic_long_read(&n->nr_slabs);
902 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
904 return atomic_long_read(&n->nr_slabs);
907 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
909 struct kmem_cache_node *n = get_node(s, node);
912 * May be called early in order to allocate a slab for the
913 * kmem_cache_node structure. Solve the chicken-egg
914 * dilemma by deferring the increment of the count during
915 * bootstrap (see early_kmem_cache_node_alloc).
918 atomic_long_inc(&n->nr_slabs);
919 atomic_long_add(objects, &n->total_objects);
922 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
924 struct kmem_cache_node *n = get_node(s, node);
926 atomic_long_dec(&n->nr_slabs);
927 atomic_long_sub(objects, &n->total_objects);
930 /* Object debug checks for alloc/free paths */
931 static void setup_object_debug(struct kmem_cache *s, struct page *page,
934 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
937 init_object(s, object, SLUB_RED_INACTIVE);
938 init_tracking(s, object);
941 static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
942 void *object, unsigned long addr)
944 if (!check_slab(s, page))
947 if (!on_freelist(s, page, object)) {
948 object_err(s, page, object, "Object already allocated");
952 if (!check_valid_pointer(s, page, object)) {
953 object_err(s, page, object, "Freelist Pointer check fails");
957 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
960 /* Success perform special debug activities for allocs */
961 if (s->flags & SLAB_STORE_USER)
962 set_track(s, object, TRACK_ALLOC, addr);
963 trace(s, page, object, 1);
964 init_object(s, object, SLUB_RED_ACTIVE);
968 if (PageSlab(page)) {
970 * If this is a slab page then lets do the best we can
971 * to avoid issues in the future. Marking all objects
972 * as used avoids touching the remaining objects.
974 slab_fix(s, "Marking all objects used");
975 page->inuse = page->objects;
976 page->freelist = NULL;
981 static noinline int free_debug_processing(struct kmem_cache *s,
982 struct page *page, void *object, unsigned long addr)
984 if (!check_slab(s, page))
987 if (!check_valid_pointer(s, page, object)) {
988 slab_err(s, page, "Invalid object pointer 0x%p", object);
992 if (on_freelist(s, page, object)) {
993 object_err(s, page, object, "Object already free");
997 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1000 if (unlikely(s != page->slab)) {
1001 if (!PageSlab(page)) {
1002 slab_err(s, page, "Attempt to free object(0x%p) "
1003 "outside of slab", object);
1004 } else if (!page->slab) {
1006 "SLUB <none>: no slab for object 0x%p.\n",
1010 object_err(s, page, object,
1011 "page slab pointer corrupt.");
1015 /* Special debug activities for freeing objects */
1016 if (!PageSlubFrozen(page) && !page->freelist)
1017 remove_full(s, page);
1018 if (s->flags & SLAB_STORE_USER)
1019 set_track(s, object, TRACK_FREE, addr);
1020 trace(s, page, object, 0);
1021 init_object(s, object, SLUB_RED_INACTIVE);
1025 slab_fix(s, "Object at 0x%p not freed", object);
1029 static int __init setup_slub_debug(char *str)
1031 slub_debug = DEBUG_DEFAULT_FLAGS;
1032 if (*str++ != '=' || !*str)
1034 * No options specified. Switch on full debugging.
1040 * No options but restriction on slabs. This means full
1041 * debugging for slabs matching a pattern.
1045 if (tolower(*str) == 'o') {
1047 * Avoid enabling debugging on caches if its minimum order
1048 * would increase as a result.
1050 disable_higher_order_debug = 1;
1057 * Switch off all debugging measures.
1062 * Determine which debug features should be switched on
1064 for (; *str && *str != ','; str++) {
1065 switch (tolower(*str)) {
1067 slub_debug |= SLAB_DEBUG_FREE;
1070 slub_debug |= SLAB_RED_ZONE;
1073 slub_debug |= SLAB_POISON;
1076 slub_debug |= SLAB_STORE_USER;
1079 slub_debug |= SLAB_TRACE;
1082 slub_debug |= SLAB_FAILSLAB;
1085 printk(KERN_ERR "slub_debug option '%c' "
1086 "unknown. skipped\n", *str);
1092 slub_debug_slabs = str + 1;
1097 __setup("slub_debug", setup_slub_debug);
1099 static unsigned long kmem_cache_flags(unsigned long objsize,
1100 unsigned long flags, const char *name,
1101 void (*ctor)(void *))
1104 * Enable debugging if selected on the kernel commandline.
1106 if (slub_debug && (!slub_debug_slabs ||
1107 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1108 flags |= slub_debug;
1113 static inline void setup_object_debug(struct kmem_cache *s,
1114 struct page *page, void *object) {}
1116 static inline int alloc_debug_processing(struct kmem_cache *s,
1117 struct page *page, void *object, unsigned long addr) { return 0; }
1119 static inline int free_debug_processing(struct kmem_cache *s,
1120 struct page *page, void *object, unsigned long addr) { return 0; }
1122 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1124 static inline int check_object(struct kmem_cache *s, struct page *page,
1125 void *object, u8 val) { return 1; }
1126 static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
1127 static inline unsigned long kmem_cache_flags(unsigned long objsize,
1128 unsigned long flags, const char *name,
1129 void (*ctor)(void *))
1133 #define slub_debug 0
1135 #define disable_higher_order_debug 0
1137 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1139 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1141 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1143 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1146 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1149 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1152 static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1154 #endif /* CONFIG_SLUB_DEBUG */
1157 * Slab allocation and freeing
1159 static inline struct page *alloc_slab_page(gfp_t flags, int node,
1160 struct kmem_cache_order_objects oo)
1162 int order = oo_order(oo);
1164 flags |= __GFP_NOTRACK;
1166 if (node == NUMA_NO_NODE)
1167 return alloc_pages(flags, order);
1169 return alloc_pages_exact_node(node, flags, order);
1172 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1175 struct kmem_cache_order_objects oo = s->oo;
1178 flags |= s->allocflags;
1181 * Let the initial higher-order allocation fail under memory pressure
1182 * so we fall-back to the minimum order allocation.
1184 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1186 page = alloc_slab_page(alloc_gfp, node, oo);
1187 if (unlikely(!page)) {
1190 * Allocation may have failed due to fragmentation.
1191 * Try a lower order alloc if possible
1193 page = alloc_slab_page(flags, node, oo);
1197 stat(s, ORDER_FALLBACK);
1200 if (kmemcheck_enabled
1201 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
1202 int pages = 1 << oo_order(oo);
1204 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1207 * Objects from caches that have a constructor don't get
1208 * cleared when they're allocated, so we need to do it here.
1211 kmemcheck_mark_uninitialized_pages(page, pages);
1213 kmemcheck_mark_unallocated_pages(page, pages);
1216 page->objects = oo_objects(oo);
1217 mod_zone_page_state(page_zone(page),
1218 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1219 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1225 static void setup_object(struct kmem_cache *s, struct page *page,
1228 setup_object_debug(s, page, object);
1229 if (unlikely(s->ctor))
1233 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1240 BUG_ON(flags & GFP_SLAB_BUG_MASK);
1242 page = allocate_slab(s,
1243 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1247 inc_slabs_node(s, page_to_nid(page), page->objects);
1249 page->flags |= 1 << PG_slab;
1251 start = page_address(page);
1253 if (unlikely(s->flags & SLAB_POISON))
1254 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
1257 for_each_object(p, s, start, page->objects) {
1258 setup_object(s, page, last);
1259 set_freepointer(s, last, p);
1262 setup_object(s, page, last);
1263 set_freepointer(s, last, NULL);
1265 page->freelist = start;
1271 static void __free_slab(struct kmem_cache *s, struct page *page)
1273 int order = compound_order(page);
1274 int pages = 1 << order;
1276 if (kmem_cache_debug(s)) {
1279 slab_pad_check(s, page);
1280 for_each_object(p, s, page_address(page),
1282 check_object(s, page, p, SLUB_RED_INACTIVE);
1285 kmemcheck_free_shadow(page, compound_order(page));
1287 mod_zone_page_state(page_zone(page),
1288 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1289 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1292 __ClearPageSlab(page);
1293 reset_page_mapcount(page);
1294 if (current->reclaim_state)
1295 current->reclaim_state->reclaimed_slab += pages;
1296 __free_pages(page, order);
1299 #define need_reserve_slab_rcu \
1300 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1302 static void rcu_free_slab(struct rcu_head *h)
1306 if (need_reserve_slab_rcu)
1307 page = virt_to_head_page(h);
1309 page = container_of((struct list_head *)h, struct page, lru);
1311 __free_slab(page->slab, page);
1314 static void free_slab(struct kmem_cache *s, struct page *page)
1316 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1317 struct rcu_head *head;
1319 if (need_reserve_slab_rcu) {
1320 int order = compound_order(page);
1321 int offset = (PAGE_SIZE << order) - s->reserved;
1323 VM_BUG_ON(s->reserved != sizeof(*head));
1324 head = page_address(page) + offset;
1327 * RCU free overloads the RCU head over the LRU
1329 head = (void *)&page->lru;
1332 call_rcu(head, rcu_free_slab);
1334 __free_slab(s, page);
1337 static void discard_slab(struct kmem_cache *s, struct page *page)
1339 dec_slabs_node(s, page_to_nid(page), page->objects);
1344 * Per slab locking using the pagelock
1346 static __always_inline void slab_lock(struct page *page)
1348 bit_spin_lock(PG_locked, &page->flags);
1351 static __always_inline void slab_unlock(struct page *page)
1353 __bit_spin_unlock(PG_locked, &page->flags);
1356 static __always_inline int slab_trylock(struct page *page)
1360 rc = bit_spin_trylock(PG_locked, &page->flags);
1365 * Management of partially allocated slabs
1367 static void add_partial(struct kmem_cache_node *n,
1368 struct page *page, int tail)
1370 spin_lock(&n->list_lock);
1373 list_add_tail(&page->lru, &n->partial);
1375 list_add(&page->lru, &n->partial);
1376 spin_unlock(&n->list_lock);
1379 static inline void __remove_partial(struct kmem_cache_node *n,
1382 list_del(&page->lru);
1386 static void remove_partial(struct kmem_cache *s, struct page *page)
1388 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1390 spin_lock(&n->list_lock);
1391 __remove_partial(n, page);
1392 spin_unlock(&n->list_lock);
1396 * Lock slab and remove from the partial list.
1398 * Must hold list_lock.
1400 static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1403 if (slab_trylock(page)) {
1404 __remove_partial(n, page);
1405 __SetPageSlubFrozen(page);
1412 * Try to allocate a partial slab from a specific node.
1414 static struct page *get_partial_node(struct kmem_cache_node *n)
1419 * Racy check. If we mistakenly see no partial slabs then we
1420 * just allocate an empty slab. If we mistakenly try to get a
1421 * partial slab and there is none available then get_partials()
1424 if (!n || !n->nr_partial)
1427 spin_lock(&n->list_lock);
1428 list_for_each_entry(page, &n->partial, lru)
1429 if (lock_and_freeze_slab(n, page))
1433 spin_unlock(&n->list_lock);
1438 * Get a page from somewhere. Search in increasing NUMA distances.
1440 static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1443 struct zonelist *zonelist;
1446 enum zone_type high_zoneidx = gfp_zone(flags);
1450 * The defrag ratio allows a configuration of the tradeoffs between
1451 * inter node defragmentation and node local allocations. A lower
1452 * defrag_ratio increases the tendency to do local allocations
1453 * instead of attempting to obtain partial slabs from other nodes.
1455 * If the defrag_ratio is set to 0 then kmalloc() always
1456 * returns node local objects. If the ratio is higher then kmalloc()
1457 * may return off node objects because partial slabs are obtained
1458 * from other nodes and filled up.
1460 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1461 * defrag_ratio = 1000) then every (well almost) allocation will
1462 * first attempt to defrag slab caches on other nodes. This means
1463 * scanning over all nodes to look for partial slabs which may be
1464 * expensive if we do it every time we are trying to find a slab
1465 * with available objects.
1467 if (!s->remote_node_defrag_ratio ||
1468 get_cycles() % 1024 > s->remote_node_defrag_ratio)
1472 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1473 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1474 struct kmem_cache_node *n;
1476 n = get_node(s, zone_to_nid(zone));
1478 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1479 n->nr_partial > s->min_partial) {
1480 page = get_partial_node(n);
1493 * Get a partial page, lock it and return it.
1495 static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1498 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
1500 page = get_partial_node(get_node(s, searchnode));
1501 if (page || node != NUMA_NO_NODE)
1504 return get_any_partial(s, flags);
1508 * Move a page back to the lists.
1510 * Must be called with the slab lock held.
1512 * On exit the slab lock will have been dropped.
1514 static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
1517 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1519 __ClearPageSlubFrozen(page);
1522 if (page->freelist) {
1523 add_partial(n, page, tail);
1524 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1526 stat(s, DEACTIVATE_FULL);
1527 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
1532 stat(s, DEACTIVATE_EMPTY);
1533 if (n->nr_partial < s->min_partial) {
1535 * Adding an empty slab to the partial slabs in order
1536 * to avoid page allocator overhead. This slab needs
1537 * to come after the other slabs with objects in
1538 * so that the others get filled first. That way the
1539 * size of the partial list stays small.
1541 * kmem_cache_shrink can reclaim any empty slabs from
1544 add_partial(n, page, 1);
1549 discard_slab(s, page);
1554 #ifdef CONFIG_PREEMPT
1556 * Calculate the next globally unique transaction for disambiguiation
1557 * during cmpxchg. The transactions start with the cpu number and are then
1558 * incremented by CONFIG_NR_CPUS.
1560 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1563 * No preemption supported therefore also no need to check for
1569 static inline unsigned long next_tid(unsigned long tid)
1571 return tid + TID_STEP;
1574 static inline unsigned int tid_to_cpu(unsigned long tid)
1576 return tid % TID_STEP;
1579 static inline unsigned long tid_to_event(unsigned long tid)
1581 return tid / TID_STEP;
1584 static inline unsigned int init_tid(int cpu)
1589 static inline void note_cmpxchg_failure(const char *n,
1590 const struct kmem_cache *s, unsigned long tid)
1592 #ifdef SLUB_DEBUG_CMPXCHG
1593 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1595 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1597 #ifdef CONFIG_PREEMPT
1598 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1599 printk("due to cpu change %d -> %d\n",
1600 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1603 if (tid_to_event(tid) != tid_to_event(actual_tid))
1604 printk("due to cpu running other code. Event %ld->%ld\n",
1605 tid_to_event(tid), tid_to_event(actual_tid));
1607 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1608 actual_tid, tid, next_tid(tid));
1610 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
1613 void init_kmem_cache_cpus(struct kmem_cache *s)
1617 for_each_possible_cpu(cpu)
1618 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
1621 * Remove the cpu slab
1623 static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1626 struct page *page = c->page;
1630 stat(s, DEACTIVATE_REMOTE_FREES);
1632 * Merge cpu freelist into slab freelist. Typically we get here
1633 * because both freelists are empty. So this is unlikely
1636 while (unlikely(c->freelist)) {
1639 tail = 0; /* Hot objects. Put the slab first */
1641 /* Retrieve object from cpu_freelist */
1642 object = c->freelist;
1643 c->freelist = get_freepointer(s, c->freelist);
1645 /* And put onto the regular freelist */
1646 set_freepointer(s, object, page->freelist);
1647 page->freelist = object;
1651 c->tid = next_tid(c->tid);
1652 unfreeze_slab(s, page, tail);
1655 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1657 stat(s, CPUSLAB_FLUSH);
1659 deactivate_slab(s, c);
1665 * Called from IPI handler with interrupts disabled.
1667 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
1669 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
1671 if (likely(c && c->page))
1675 static void flush_cpu_slab(void *d)
1677 struct kmem_cache *s = d;
1679 __flush_cpu_slab(s, smp_processor_id());
1682 static void flush_all(struct kmem_cache *s)
1684 on_each_cpu(flush_cpu_slab, s, 1);
1688 * Check if the objects in a per cpu structure fit numa
1689 * locality expectations.
1691 static inline int node_match(struct kmem_cache_cpu *c, int node)
1694 if (node != NUMA_NO_NODE && c->node != node)
1700 static int count_free(struct page *page)
1702 return page->objects - page->inuse;
1705 static unsigned long count_partial(struct kmem_cache_node *n,
1706 int (*get_count)(struct page *))
1708 unsigned long flags;
1709 unsigned long x = 0;
1712 spin_lock_irqsave(&n->list_lock, flags);
1713 list_for_each_entry(page, &n->partial, lru)
1714 x += get_count(page);
1715 spin_unlock_irqrestore(&n->list_lock, flags);
1719 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1721 #ifdef CONFIG_SLUB_DEBUG
1722 return atomic_long_read(&n->total_objects);
1728 static noinline void
1729 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1734 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1736 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1737 "default order: %d, min order: %d\n", s->name, s->objsize,
1738 s->size, oo_order(s->oo), oo_order(s->min));
1740 if (oo_order(s->min) > get_order(s->objsize))
1741 printk(KERN_WARNING " %s debugging increased min order, use "
1742 "slub_debug=O to disable.\n", s->name);
1744 for_each_online_node(node) {
1745 struct kmem_cache_node *n = get_node(s, node);
1746 unsigned long nr_slabs;
1747 unsigned long nr_objs;
1748 unsigned long nr_free;
1753 nr_free = count_partial(n, count_free);
1754 nr_slabs = node_nr_slabs(n);
1755 nr_objs = node_nr_objs(n);
1758 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1759 node, nr_slabs, nr_objs, nr_free);
1764 * Slow path. The lockless freelist is empty or we need to perform
1767 * Interrupts are disabled.
1769 * Processing is still very fast if new objects have been freed to the
1770 * regular freelist. In that case we simply take over the regular freelist
1771 * as the lockless freelist and zap the regular freelist.
1773 * If that is not working then we fall back to the partial lists. We take the
1774 * first element of the freelist as the object to allocate now and move the
1775 * rest of the freelist to the lockless freelist.
1777 * And if we were unable to get a new slab from the partial slab lists then
1778 * we need to allocate a new slab. This is the slowest path since it involves
1779 * a call to the page allocator and the setup of a new slab.
1781 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1782 unsigned long addr, struct kmem_cache_cpu *c)
1786 unsigned long flags;
1788 local_irq_save(flags);
1789 #ifdef CONFIG_PREEMPT
1791 * We may have been preempted and rescheduled on a different
1792 * cpu before disabling interrupts. Need to reload cpu area
1795 c = this_cpu_ptr(s->cpu_slab);
1798 /* We handle __GFP_ZERO in the caller */
1799 gfpflags &= ~__GFP_ZERO;
1806 if (unlikely(!node_match(c, node)))
1809 stat(s, ALLOC_REFILL);
1812 object = page->freelist;
1813 if (unlikely(!object))
1815 if (kmem_cache_debug(s))
1818 c->freelist = get_freepointer(s, object);
1819 page->inuse = page->objects;
1820 page->freelist = NULL;
1824 c->tid = next_tid(c->tid);
1825 local_irq_restore(flags);
1826 stat(s, ALLOC_SLOWPATH);
1830 deactivate_slab(s, c);
1833 page = get_partial(s, gfpflags, node);
1835 stat(s, ALLOC_FROM_PARTIAL);
1837 c->node = page_to_nid(page);
1842 gfpflags &= gfp_allowed_mask;
1843 if (gfpflags & __GFP_WAIT)
1846 page = new_slab(s, gfpflags, node);
1848 if (gfpflags & __GFP_WAIT)
1849 local_irq_disable();
1852 c = __this_cpu_ptr(s->cpu_slab);
1853 stat(s, ALLOC_SLAB);
1858 __SetPageSlubFrozen(page);
1860 goto load_from_page;
1862 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1863 slab_out_of_memory(s, gfpflags, node);
1864 local_irq_restore(flags);
1867 if (!alloc_debug_processing(s, page, object, addr))
1871 page->freelist = get_freepointer(s, object);
1872 c->node = NUMA_NO_NODE;
1877 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1878 * have the fastpath folded into their functions. So no function call
1879 * overhead for requests that can be satisfied on the fastpath.
1881 * The fastpath works by first checking if the lockless freelist can be used.
1882 * If not then __slab_alloc is called for slow processing.
1884 * Otherwise we can simply pick the next object from the lockless free list.
1886 static __always_inline void *slab_alloc(struct kmem_cache *s,
1887 gfp_t gfpflags, int node, unsigned long addr)
1890 struct kmem_cache_cpu *c;
1893 if (slab_pre_alloc_hook(s, gfpflags))
1899 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1900 * enabled. We may switch back and forth between cpus while
1901 * reading from one cpu area. That does not matter as long
1902 * as we end up on the original cpu again when doing the cmpxchg.
1904 c = __this_cpu_ptr(s->cpu_slab);
1907 * The transaction ids are globally unique per cpu and per operation on
1908 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1909 * occurs on the right processor and that there was no operation on the
1910 * linked list in between.
1915 object = c->freelist;
1916 if (unlikely(!object || !node_match(c, node)))
1918 object = __slab_alloc(s, gfpflags, node, addr, c);
1922 * The cmpxchg will only match if there was no additonal
1923 * operation and if we are on the right processor.
1925 * The cmpxchg does the following atomically (without lock semantics!)
1926 * 1. Relocate first pointer to the current per cpu area.
1927 * 2. Verify that tid and freelist have not been changed
1928 * 3. If they were not changed replace tid and freelist
1930 * Since this is without lock semantics the protection is only against
1931 * code executing on this cpu *not* from access by other cpus.
1933 if (unlikely(!this_cpu_cmpxchg_double(
1934 s->cpu_slab->freelist, s->cpu_slab->tid,
1936 get_freepointer(s, object), next_tid(tid)))) {
1938 note_cmpxchg_failure("slab_alloc", s, tid);
1941 stat(s, ALLOC_FASTPATH);
1944 if (unlikely(gfpflags & __GFP_ZERO) && object)
1945 memset(object, 0, s->objsize);
1947 slab_post_alloc_hook(s, gfpflags, object);
1952 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1954 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1956 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
1960 EXPORT_SYMBOL(kmem_cache_alloc);
1962 #ifdef CONFIG_TRACING
1963 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1965 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1966 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1969 EXPORT_SYMBOL(kmem_cache_alloc_trace);
1971 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1973 void *ret = kmalloc_order(size, flags, order);
1974 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1977 EXPORT_SYMBOL(kmalloc_order_trace);
1981 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1983 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1985 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1986 s->objsize, s->size, gfpflags, node);
1990 EXPORT_SYMBOL(kmem_cache_alloc_node);
1992 #ifdef CONFIG_TRACING
1993 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
1995 int node, size_t size)
1997 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1999 trace_kmalloc_node(_RET_IP_, ret,
2000 size, s->size, gfpflags, node);
2003 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2008 * Slow patch handling. This may still be called frequently since objects
2009 * have a longer lifetime than the cpu slabs in most processing loads.
2011 * So we still attempt to reduce cache line usage. Just take the slab
2012 * lock and free the item. If there is no additional partial page
2013 * handling required then we can return immediately.
2015 static void __slab_free(struct kmem_cache *s, struct page *page,
2016 void *x, unsigned long addr)
2019 void **object = (void *)x;
2020 unsigned long flags;
2022 local_irq_save(flags);
2024 stat(s, FREE_SLOWPATH);
2026 if (kmem_cache_debug(s) && !free_debug_processing(s, page, x, addr))
2029 prior = page->freelist;
2030 set_freepointer(s, object, prior);
2031 page->freelist = object;
2034 if (unlikely(PageSlubFrozen(page))) {
2035 stat(s, FREE_FROZEN);
2039 if (unlikely(!page->inuse))
2043 * Objects left in the slab. If it was not on the partial list before
2046 if (unlikely(!prior)) {
2047 add_partial(get_node(s, page_to_nid(page)), page, 1);
2048 stat(s, FREE_ADD_PARTIAL);
2053 local_irq_restore(flags);
2059 * Slab still on the partial list.
2061 remove_partial(s, page);
2062 stat(s, FREE_REMOVE_PARTIAL);
2065 local_irq_restore(flags);
2067 discard_slab(s, page);
2071 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2072 * can perform fastpath freeing without additional function calls.
2074 * The fastpath is only possible if we are freeing to the current cpu slab
2075 * of this processor. This typically the case if we have just allocated
2078 * If fastpath is not possible then fall back to __slab_free where we deal
2079 * with all sorts of special processing.
2081 static __always_inline void slab_free(struct kmem_cache *s,
2082 struct page *page, void *x, unsigned long addr)
2084 void **object = (void *)x;
2085 struct kmem_cache_cpu *c;
2088 slab_free_hook(s, x);
2093 * Determine the currently cpus per cpu slab.
2094 * The cpu may change afterward. However that does not matter since
2095 * data is retrieved via this pointer. If we are on the same cpu
2096 * during the cmpxchg then the free will succedd.
2098 c = __this_cpu_ptr(s->cpu_slab);
2103 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
2104 set_freepointer(s, object, c->freelist);
2106 if (unlikely(!this_cpu_cmpxchg_double(
2107 s->cpu_slab->freelist, s->cpu_slab->tid,
2109 object, next_tid(tid)))) {
2111 note_cmpxchg_failure("slab_free", s, tid);
2114 stat(s, FREE_FASTPATH);
2116 __slab_free(s, page, x, addr);
2120 void kmem_cache_free(struct kmem_cache *s, void *x)
2124 page = virt_to_head_page(x);
2126 slab_free(s, page, x, _RET_IP_);
2128 trace_kmem_cache_free(_RET_IP_, x);
2130 EXPORT_SYMBOL(kmem_cache_free);
2133 * Object placement in a slab is made very easy because we always start at
2134 * offset 0. If we tune the size of the object to the alignment then we can
2135 * get the required alignment by putting one properly sized object after
2138 * Notice that the allocation order determines the sizes of the per cpu
2139 * caches. Each processor has always one slab available for allocations.
2140 * Increasing the allocation order reduces the number of times that slabs
2141 * must be moved on and off the partial lists and is therefore a factor in
2146 * Mininum / Maximum order of slab pages. This influences locking overhead
2147 * and slab fragmentation. A higher order reduces the number of partial slabs
2148 * and increases the number of allocations possible without having to
2149 * take the list_lock.
2151 static int slub_min_order;
2152 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
2153 static int slub_min_objects;
2156 * Merge control. If this is set then no merging of slab caches will occur.
2157 * (Could be removed. This was introduced to pacify the merge skeptics.)
2159 static int slub_nomerge;
2162 * Calculate the order of allocation given an slab object size.
2164 * The order of allocation has significant impact on performance and other
2165 * system components. Generally order 0 allocations should be preferred since
2166 * order 0 does not cause fragmentation in the page allocator. Larger objects
2167 * be problematic to put into order 0 slabs because there may be too much
2168 * unused space left. We go to a higher order if more than 1/16th of the slab
2171 * In order to reach satisfactory performance we must ensure that a minimum
2172 * number of objects is in one slab. Otherwise we may generate too much
2173 * activity on the partial lists which requires taking the list_lock. This is
2174 * less a concern for large slabs though which are rarely used.
2176 * slub_max_order specifies the order where we begin to stop considering the
2177 * number of objects in a slab as critical. If we reach slub_max_order then
2178 * we try to keep the page order as low as possible. So we accept more waste
2179 * of space in favor of a small page order.
2181 * Higher order allocations also allow the placement of more objects in a
2182 * slab and thereby reduce object handling overhead. If the user has
2183 * requested a higher mininum order then we start with that one instead of
2184 * the smallest order which will fit the object.
2186 static inline int slab_order(int size, int min_objects,
2187 int max_order, int fract_leftover, int reserved)
2191 int min_order = slub_min_order;
2193 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
2194 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
2196 for (order = max(min_order,
2197 fls(min_objects * size - 1) - PAGE_SHIFT);
2198 order <= max_order; order++) {
2200 unsigned long slab_size = PAGE_SIZE << order;
2202 if (slab_size < min_objects * size + reserved)
2205 rem = (slab_size - reserved) % size;
2207 if (rem <= slab_size / fract_leftover)
2215 static inline int calculate_order(int size, int reserved)
2223 * Attempt to find best configuration for a slab. This
2224 * works by first attempting to generate a layout with
2225 * the best configuration and backing off gradually.
2227 * First we reduce the acceptable waste in a slab. Then
2228 * we reduce the minimum objects required in a slab.
2230 min_objects = slub_min_objects;
2232 min_objects = 4 * (fls(nr_cpu_ids) + 1);
2233 max_objects = order_objects(slub_max_order, size, reserved);
2234 min_objects = min(min_objects, max_objects);
2236 while (min_objects > 1) {
2238 while (fraction >= 4) {
2239 order = slab_order(size, min_objects,
2240 slub_max_order, fraction, reserved);
2241 if (order <= slub_max_order)
2249 * We were unable to place multiple objects in a slab. Now
2250 * lets see if we can place a single object there.
2252 order = slab_order(size, 1, slub_max_order, 1, reserved);
2253 if (order <= slub_max_order)
2257 * Doh this slab cannot be placed using slub_max_order.
2259 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
2260 if (order < MAX_ORDER)
2266 * Figure out what the alignment of the objects will be.
2268 static unsigned long calculate_alignment(unsigned long flags,
2269 unsigned long align, unsigned long size)
2272 * If the user wants hardware cache aligned objects then follow that
2273 * suggestion if the object is sufficiently large.
2275 * The hardware cache alignment cannot override the specified
2276 * alignment though. If that is greater then use it.
2278 if (flags & SLAB_HWCACHE_ALIGN) {
2279 unsigned long ralign = cache_line_size();
2280 while (size <= ralign / 2)
2282 align = max(align, ralign);
2285 if (align < ARCH_SLAB_MINALIGN)
2286 align = ARCH_SLAB_MINALIGN;
2288 return ALIGN(align, sizeof(void *));
2292 init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
2295 spin_lock_init(&n->list_lock);
2296 INIT_LIST_HEAD(&n->partial);
2297 #ifdef CONFIG_SLUB_DEBUG
2298 atomic_long_set(&n->nr_slabs, 0);
2299 atomic_long_set(&n->total_objects, 0);
2300 INIT_LIST_HEAD(&n->full);
2304 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
2306 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2307 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
2309 #ifdef CONFIG_CMPXCHG_LOCAL
2311 * Must align to double word boundary for the double cmpxchg instructions
2314 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
2316 /* Regular alignment is sufficient */
2317 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2323 init_kmem_cache_cpus(s);
2328 static struct kmem_cache *kmem_cache_node;
2331 * No kmalloc_node yet so do it by hand. We know that this is the first
2332 * slab on the node for this slabcache. There are no concurrent accesses
2335 * Note that this function only works on the kmalloc_node_cache
2336 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2337 * memory on a fresh node that has no slab structures yet.
2339 static void early_kmem_cache_node_alloc(int node)
2342 struct kmem_cache_node *n;
2343 unsigned long flags;
2345 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
2347 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
2350 if (page_to_nid(page) != node) {
2351 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2353 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2354 "in order to be able to continue\n");
2359 page->freelist = get_freepointer(kmem_cache_node, n);
2361 kmem_cache_node->node[node] = n;
2362 #ifdef CONFIG_SLUB_DEBUG
2363 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
2364 init_tracking(kmem_cache_node, n);
2366 init_kmem_cache_node(n, kmem_cache_node);
2367 inc_slabs_node(kmem_cache_node, node, page->objects);
2370 * lockdep requires consistent irq usage for each lock
2371 * so even though there cannot be a race this early in
2372 * the boot sequence, we still disable irqs.
2374 local_irq_save(flags);
2375 add_partial(n, page, 0);
2376 local_irq_restore(flags);
2379 static void free_kmem_cache_nodes(struct kmem_cache *s)
2383 for_each_node_state(node, N_NORMAL_MEMORY) {
2384 struct kmem_cache_node *n = s->node[node];
2387 kmem_cache_free(kmem_cache_node, n);
2389 s->node[node] = NULL;
2393 static int init_kmem_cache_nodes(struct kmem_cache *s)
2397 for_each_node_state(node, N_NORMAL_MEMORY) {
2398 struct kmem_cache_node *n;
2400 if (slab_state == DOWN) {
2401 early_kmem_cache_node_alloc(node);
2404 n = kmem_cache_alloc_node(kmem_cache_node,
2408 free_kmem_cache_nodes(s);
2413 init_kmem_cache_node(n, s);
2418 static void set_min_partial(struct kmem_cache *s, unsigned long min)
2420 if (min < MIN_PARTIAL)
2422 else if (min > MAX_PARTIAL)
2424 s->min_partial = min;
2428 * calculate_sizes() determines the order and the distribution of data within
2431 static int calculate_sizes(struct kmem_cache *s, int forced_order)
2433 unsigned long flags = s->flags;
2434 unsigned long size = s->objsize;
2435 unsigned long align = s->align;
2439 * Round up object size to the next word boundary. We can only
2440 * place the free pointer at word boundaries and this determines
2441 * the possible location of the free pointer.
2443 size = ALIGN(size, sizeof(void *));
2445 #ifdef CONFIG_SLUB_DEBUG
2447 * Determine if we can poison the object itself. If the user of
2448 * the slab may touch the object after free or before allocation
2449 * then we should never poison the object itself.
2451 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
2453 s->flags |= __OBJECT_POISON;
2455 s->flags &= ~__OBJECT_POISON;
2459 * If we are Redzoning then check if there is some space between the
2460 * end of the object and the free pointer. If not then add an
2461 * additional word to have some bytes to store Redzone information.
2463 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2464 size += sizeof(void *);
2468 * With that we have determined the number of bytes in actual use
2469 * by the object. This is the potential offset to the free pointer.
2473 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
2476 * Relocate free pointer after the object if it is not
2477 * permitted to overwrite the first word of the object on
2480 * This is the case if we do RCU, have a constructor or
2481 * destructor or are poisoning the objects.
2484 size += sizeof(void *);
2487 #ifdef CONFIG_SLUB_DEBUG
2488 if (flags & SLAB_STORE_USER)
2490 * Need to store information about allocs and frees after
2493 size += 2 * sizeof(struct track);
2495 if (flags & SLAB_RED_ZONE)
2497 * Add some empty padding so that we can catch
2498 * overwrites from earlier objects rather than let
2499 * tracking information or the free pointer be
2500 * corrupted if a user writes before the start
2503 size += sizeof(void *);
2507 * Determine the alignment based on various parameters that the
2508 * user specified and the dynamic determination of cache line size
2511 align = calculate_alignment(flags, align, s->objsize);
2515 * SLUB stores one object immediately after another beginning from
2516 * offset 0. In order to align the objects we have to simply size
2517 * each object to conform to the alignment.
2519 size = ALIGN(size, align);
2521 if (forced_order >= 0)
2522 order = forced_order;
2524 order = calculate_order(size, s->reserved);
2531 s->allocflags |= __GFP_COMP;
2533 if (s->flags & SLAB_CACHE_DMA)
2534 s->allocflags |= SLUB_DMA;
2536 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2537 s->allocflags |= __GFP_RECLAIMABLE;
2540 * Determine the number of objects per slab
2542 s->oo = oo_make(order, size, s->reserved);
2543 s->min = oo_make(get_order(size), size, s->reserved);
2544 if (oo_objects(s->oo) > oo_objects(s->max))
2547 return !!oo_objects(s->oo);
2551 static int kmem_cache_open(struct kmem_cache *s,
2552 const char *name, size_t size,
2553 size_t align, unsigned long flags,
2554 void (*ctor)(void *))
2556 memset(s, 0, kmem_size);
2561 s->flags = kmem_cache_flags(size, flags, name, ctor);
2564 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
2565 s->reserved = sizeof(struct rcu_head);
2567 if (!calculate_sizes(s, -1))
2569 if (disable_higher_order_debug) {
2571 * Disable debugging flags that store metadata if the min slab
2574 if (get_order(s->size) > get_order(s->objsize)) {
2575 s->flags &= ~DEBUG_METADATA_FLAGS;
2577 if (!calculate_sizes(s, -1))
2583 * The larger the object size is, the more pages we want on the partial
2584 * list to avoid pounding the page allocator excessively.
2586 set_min_partial(s, ilog2(s->size));
2589 s->remote_node_defrag_ratio = 1000;
2591 if (!init_kmem_cache_nodes(s))
2594 if (alloc_kmem_cache_cpus(s))
2597 free_kmem_cache_nodes(s);
2599 if (flags & SLAB_PANIC)
2600 panic("Cannot create slab %s size=%lu realsize=%u "
2601 "order=%u offset=%u flags=%lx\n",
2602 s->name, (unsigned long)size, s->size, oo_order(s->oo),
2608 * Determine the size of a slab object
2610 unsigned int kmem_cache_size(struct kmem_cache *s)
2614 EXPORT_SYMBOL(kmem_cache_size);
2616 static void list_slab_objects(struct kmem_cache *s, struct page *page,
2619 #ifdef CONFIG_SLUB_DEBUG
2620 void *addr = page_address(page);
2622 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
2623 sizeof(long), GFP_ATOMIC);
2626 slab_err(s, page, "%s", text);
2629 get_map(s, page, map);
2630 for_each_object(p, s, addr, page->objects) {
2632 if (!test_bit(slab_index(p, s, addr), map)) {
2633 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2635 print_tracking(s, p);
2644 * Attempt to free all partial slabs on a node.
2646 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
2648 unsigned long flags;
2649 struct page *page, *h;
2651 spin_lock_irqsave(&n->list_lock, flags);
2652 list_for_each_entry_safe(page, h, &n->partial, lru) {
2654 __remove_partial(n, page);
2655 discard_slab(s, page);
2657 list_slab_objects(s, page,
2658 "Objects remaining on kmem_cache_close()");
2661 spin_unlock_irqrestore(&n->list_lock, flags);
2665 * Release all resources used by a slab cache.
2667 static inline int kmem_cache_close(struct kmem_cache *s)
2672 free_percpu(s->cpu_slab);
2673 /* Attempt to free all objects */
2674 for_each_node_state(node, N_NORMAL_MEMORY) {
2675 struct kmem_cache_node *n = get_node(s, node);
2678 if (n->nr_partial || slabs_node(s, node))
2681 free_kmem_cache_nodes(s);
2686 * Close a cache and release the kmem_cache structure
2687 * (must be used for caches created using kmem_cache_create)
2689 void kmem_cache_destroy(struct kmem_cache *s)
2691 down_write(&slub_lock);
2695 if (kmem_cache_close(s)) {
2696 printk(KERN_ERR "SLUB %s: %s called for cache that "
2697 "still has objects.\n", s->name, __func__);
2700 if (s->flags & SLAB_DESTROY_BY_RCU)
2702 sysfs_slab_remove(s);
2704 up_write(&slub_lock);
2706 EXPORT_SYMBOL(kmem_cache_destroy);
2708 /********************************************************************
2710 *******************************************************************/
2712 struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
2713 EXPORT_SYMBOL(kmalloc_caches);
2715 static struct kmem_cache *kmem_cache;
2717 #ifdef CONFIG_ZONE_DMA
2718 static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
2721 static int __init setup_slub_min_order(char *str)
2723 get_option(&str, &slub_min_order);
2728 __setup("slub_min_order=", setup_slub_min_order);
2730 static int __init setup_slub_max_order(char *str)
2732 get_option(&str, &slub_max_order);
2733 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
2738 __setup("slub_max_order=", setup_slub_max_order);
2740 static int __init setup_slub_min_objects(char *str)
2742 get_option(&str, &slub_min_objects);
2747 __setup("slub_min_objects=", setup_slub_min_objects);
2749 static int __init setup_slub_nomerge(char *str)
2755 __setup("slub_nomerge", setup_slub_nomerge);
2757 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2758 int size, unsigned int flags)
2760 struct kmem_cache *s;
2762 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2765 * This function is called with IRQs disabled during early-boot on
2766 * single CPU so there's no need to take slub_lock here.
2768 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
2772 list_add(&s->list, &slab_caches);
2776 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2781 * Conversion table for small slabs sizes / 8 to the index in the
2782 * kmalloc array. This is necessary for slabs < 192 since we have non power
2783 * of two cache sizes there. The size of larger slabs can be determined using
2786 static s8 size_index[24] = {
2813 static inline int size_index_elem(size_t bytes)
2815 return (bytes - 1) / 8;
2818 static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2824 return ZERO_SIZE_PTR;
2826 index = size_index[size_index_elem(size)];
2828 index = fls(size - 1);
2830 #ifdef CONFIG_ZONE_DMA
2831 if (unlikely((flags & SLUB_DMA)))
2832 return kmalloc_dma_caches[index];
2835 return kmalloc_caches[index];
2838 void *__kmalloc(size_t size, gfp_t flags)
2840 struct kmem_cache *s;
2843 if (unlikely(size > SLUB_MAX_SIZE))
2844 return kmalloc_large(size, flags);
2846 s = get_slab(size, flags);
2848 if (unlikely(ZERO_OR_NULL_PTR(s)))
2851 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
2853 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
2857 EXPORT_SYMBOL(__kmalloc);
2860 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2865 flags |= __GFP_COMP | __GFP_NOTRACK;
2866 page = alloc_pages_node(node, flags, get_order(size));
2868 ptr = page_address(page);
2870 kmemleak_alloc(ptr, size, 1, flags);
2874 void *__kmalloc_node(size_t size, gfp_t flags, int node)
2876 struct kmem_cache *s;
2879 if (unlikely(size > SLUB_MAX_SIZE)) {
2880 ret = kmalloc_large_node(size, flags, node);
2882 trace_kmalloc_node(_RET_IP_, ret,
2883 size, PAGE_SIZE << get_order(size),
2889 s = get_slab(size, flags);
2891 if (unlikely(ZERO_OR_NULL_PTR(s)))
2894 ret = slab_alloc(s, flags, node, _RET_IP_);
2896 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
2900 EXPORT_SYMBOL(__kmalloc_node);
2903 size_t ksize(const void *object)
2907 if (unlikely(object == ZERO_SIZE_PTR))
2910 page = virt_to_head_page(object);
2912 if (unlikely(!PageSlab(page))) {
2913 WARN_ON(!PageCompound(page));
2914 return PAGE_SIZE << compound_order(page);
2917 return slab_ksize(page->slab);
2919 EXPORT_SYMBOL(ksize);
2921 void kfree(const void *x)
2924 void *object = (void *)x;
2926 trace_kfree(_RET_IP_, x);
2928 if (unlikely(ZERO_OR_NULL_PTR(x)))
2931 page = virt_to_head_page(x);
2932 if (unlikely(!PageSlab(page))) {
2933 BUG_ON(!PageCompound(page));
2938 slab_free(page->slab, page, object, _RET_IP_);
2940 EXPORT_SYMBOL(kfree);
2943 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2944 * the remaining slabs by the number of items in use. The slabs with the
2945 * most items in use come first. New allocations will then fill those up
2946 * and thus they can be removed from the partial lists.
2948 * The slabs with the least items are placed last. This results in them
2949 * being allocated from last increasing the chance that the last objects
2950 * are freed in them.
2952 int kmem_cache_shrink(struct kmem_cache *s)
2956 struct kmem_cache_node *n;
2959 int objects = oo_objects(s->max);
2960 struct list_head *slabs_by_inuse =
2961 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
2962 unsigned long flags;
2964 if (!slabs_by_inuse)
2968 for_each_node_state(node, N_NORMAL_MEMORY) {
2969 n = get_node(s, node);
2974 for (i = 0; i < objects; i++)
2975 INIT_LIST_HEAD(slabs_by_inuse + i);
2977 spin_lock_irqsave(&n->list_lock, flags);
2980 * Build lists indexed by the items in use in each slab.
2982 * Note that concurrent frees may occur while we hold the
2983 * list_lock. page->inuse here is the upper limit.
2985 list_for_each_entry_safe(page, t, &n->partial, lru) {
2986 if (!page->inuse && slab_trylock(page)) {
2988 * Must hold slab lock here because slab_free
2989 * may have freed the last object and be
2990 * waiting to release the slab.
2992 __remove_partial(n, page);
2994 discard_slab(s, page);
2996 list_move(&page->lru,
2997 slabs_by_inuse + page->inuse);
3002 * Rebuild the partial list with the slabs filled up most
3003 * first and the least used slabs at the end.
3005 for (i = objects - 1; i >= 0; i--)
3006 list_splice(slabs_by_inuse + i, n->partial.prev);
3008 spin_unlock_irqrestore(&n->list_lock, flags);
3011 kfree(slabs_by_inuse);
3014 EXPORT_SYMBOL(kmem_cache_shrink);
3016 #if defined(CONFIG_MEMORY_HOTPLUG)
3017 static int slab_mem_going_offline_callback(void *arg)
3019 struct kmem_cache *s;
3021 down_read(&slub_lock);
3022 list_for_each_entry(s, &slab_caches, list)
3023 kmem_cache_shrink(s);
3024 up_read(&slub_lock);
3029 static void slab_mem_offline_callback(void *arg)
3031 struct kmem_cache_node *n;
3032 struct kmem_cache *s;
3033 struct memory_notify *marg = arg;
3036 offline_node = marg->status_change_nid;
3039 * If the node still has available memory. we need kmem_cache_node
3042 if (offline_node < 0)
3045 down_read(&slub_lock);
3046 list_for_each_entry(s, &slab_caches, list) {
3047 n = get_node(s, offline_node);
3050 * if n->nr_slabs > 0, slabs still exist on the node
3051 * that is going down. We were unable to free them,
3052 * and offline_pages() function shouldn't call this
3053 * callback. So, we must fail.
3055 BUG_ON(slabs_node(s, offline_node));
3057 s->node[offline_node] = NULL;
3058 kmem_cache_free(kmem_cache_node, n);
3061 up_read(&slub_lock);
3064 static int slab_mem_going_online_callback(void *arg)
3066 struct kmem_cache_node *n;
3067 struct kmem_cache *s;
3068 struct memory_notify *marg = arg;
3069 int nid = marg->status_change_nid;
3073 * If the node's memory is already available, then kmem_cache_node is
3074 * already created. Nothing to do.
3080 * We are bringing a node online. No memory is available yet. We must
3081 * allocate a kmem_cache_node structure in order to bring the node
3084 down_read(&slub_lock);
3085 list_for_each_entry(s, &slab_caches, list) {
3087 * XXX: kmem_cache_alloc_node will fallback to other nodes
3088 * since memory is not yet available from the node that
3091 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
3096 init_kmem_cache_node(n, s);
3100 up_read(&slub_lock);
3104 static int slab_memory_callback(struct notifier_block *self,
3105 unsigned long action, void *arg)
3110 case MEM_GOING_ONLINE:
3111 ret = slab_mem_going_online_callback(arg);
3113 case MEM_GOING_OFFLINE:
3114 ret = slab_mem_going_offline_callback(arg);
3117 case MEM_CANCEL_ONLINE:
3118 slab_mem_offline_callback(arg);
3121 case MEM_CANCEL_OFFLINE:
3125 ret = notifier_from_errno(ret);
3131 #endif /* CONFIG_MEMORY_HOTPLUG */
3133 /********************************************************************
3134 * Basic setup of slabs
3135 *******************************************************************/
3138 * Used for early kmem_cache structures that were allocated using
3139 * the page allocator
3142 static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3146 list_add(&s->list, &slab_caches);
3149 for_each_node_state(node, N_NORMAL_MEMORY) {
3150 struct kmem_cache_node *n = get_node(s, node);
3154 list_for_each_entry(p, &n->partial, lru)
3157 #ifdef CONFIG_SLUB_DEBUG
3158 list_for_each_entry(p, &n->full, lru)
3165 void __init kmem_cache_init(void)
3169 struct kmem_cache *temp_kmem_cache;
3171 struct kmem_cache *temp_kmem_cache_node;
3172 unsigned long kmalloc_size;
3174 kmem_size = offsetof(struct kmem_cache, node) +
3175 nr_node_ids * sizeof(struct kmem_cache_node *);
3177 /* Allocate two kmem_caches from the page allocator */
3178 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3179 order = get_order(2 * kmalloc_size);
3180 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3183 * Must first have the slab cache available for the allocations of the
3184 * struct kmem_cache_node's. There is special bootstrap code in
3185 * kmem_cache_open for slab_state == DOWN.
3187 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3189 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3190 sizeof(struct kmem_cache_node),
3191 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3193 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
3195 /* Able to allocate the per node structures */
3196 slab_state = PARTIAL;
3198 temp_kmem_cache = kmem_cache;
3199 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3200 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3201 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3202 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
3205 * Allocate kmem_cache_node properly from the kmem_cache slab.
3206 * kmem_cache_node is separately allocated so no need to
3207 * update any list pointers.
3209 temp_kmem_cache_node = kmem_cache_node;
3211 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3212 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3214 kmem_cache_bootstrap_fixup(kmem_cache_node);
3217 kmem_cache_bootstrap_fixup(kmem_cache);
3219 /* Free temporary boot structure */
3220 free_pages((unsigned long)temp_kmem_cache, order);
3222 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3225 * Patch up the size_index table if we have strange large alignment
3226 * requirements for the kmalloc array. This is only the case for
3227 * MIPS it seems. The standard arches will not generate any code here.
3229 * Largest permitted alignment is 256 bytes due to the way we
3230 * handle the index determination for the smaller caches.
3232 * Make sure that nothing crazy happens if someone starts tinkering
3233 * around with ARCH_KMALLOC_MINALIGN
3235 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3236 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3238 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3239 int elem = size_index_elem(i);
3240 if (elem >= ARRAY_SIZE(size_index))
3242 size_index[elem] = KMALLOC_SHIFT_LOW;
3245 if (KMALLOC_MIN_SIZE == 64) {
3247 * The 96 byte size cache is not used if the alignment
3250 for (i = 64 + 8; i <= 96; i += 8)
3251 size_index[size_index_elem(i)] = 7;
3252 } else if (KMALLOC_MIN_SIZE == 128) {
3254 * The 192 byte sized cache is not used if the alignment
3255 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3258 for (i = 128 + 8; i <= 192; i += 8)
3259 size_index[size_index_elem(i)] = 8;
3262 /* Caches that are not of the two-to-the-power-of size */
3263 if (KMALLOC_MIN_SIZE <= 32) {
3264 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3268 if (KMALLOC_MIN_SIZE <= 64) {
3269 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3273 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3274 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3280 /* Provide the correct kmalloc names now that the caches are up */
3281 if (KMALLOC_MIN_SIZE <= 32) {
3282 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3283 BUG_ON(!kmalloc_caches[1]->name);
3286 if (KMALLOC_MIN_SIZE <= 64) {
3287 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3288 BUG_ON(!kmalloc_caches[2]->name);
3291 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3292 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3295 kmalloc_caches[i]->name = s;
3299 register_cpu_notifier(&slab_notifier);
3302 #ifdef CONFIG_ZONE_DMA
3303 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3304 struct kmem_cache *s = kmalloc_caches[i];
3307 char *name = kasprintf(GFP_NOWAIT,
3308 "dma-kmalloc-%d", s->objsize);
3311 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3312 s->objsize, SLAB_CACHE_DMA);
3317 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3318 " CPUs=%d, Nodes=%d\n",
3319 caches, cache_line_size(),
3320 slub_min_order, slub_max_order, slub_min_objects,
3321 nr_cpu_ids, nr_node_ids);
3324 void __init kmem_cache_init_late(void)
3329 * Find a mergeable slab cache
3331 static int slab_unmergeable(struct kmem_cache *s)
3333 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3340 * We may have set a slab to be unmergeable during bootstrap.
3342 if (s->refcount < 0)
3348 static struct kmem_cache *find_mergeable(size_t size,
3349 size_t align, unsigned long flags, const char *name,
3350 void (*ctor)(void *))
3352 struct kmem_cache *s;
3354 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3360 size = ALIGN(size, sizeof(void *));
3361 align = calculate_alignment(flags, align, size);
3362 size = ALIGN(size, align);
3363 flags = kmem_cache_flags(size, flags, name, NULL);
3365 list_for_each_entry(s, &slab_caches, list) {
3366 if (slab_unmergeable(s))
3372 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
3375 * Check if alignment is compatible.
3376 * Courtesy of Adrian Drzewiecki
3378 if ((s->size & ~(align - 1)) != s->size)
3381 if (s->size - size >= sizeof(void *))
3389 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
3390 size_t align, unsigned long flags, void (*ctor)(void *))
3392 struct kmem_cache *s;
3398 down_write(&slub_lock);
3399 s = find_mergeable(size, align, flags, name, ctor);
3403 * Adjust the object sizes so that we clear
3404 * the complete object on kzalloc.
3406 s->objsize = max(s->objsize, (int)size);
3407 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
3409 if (sysfs_slab_alias(s, name)) {
3413 up_write(&slub_lock);
3417 n = kstrdup(name, GFP_KERNEL);
3421 s = kmalloc(kmem_size, GFP_KERNEL);
3423 if (kmem_cache_open(s, n,
3424 size, align, flags, ctor)) {
3425 list_add(&s->list, &slab_caches);
3426 if (sysfs_slab_add(s)) {
3432 up_write(&slub_lock);
3439 up_write(&slub_lock);
3441 if (flags & SLAB_PANIC)
3442 panic("Cannot create slabcache %s\n", name);
3447 EXPORT_SYMBOL(kmem_cache_create);
3451 * Use the cpu notifier to insure that the cpu slabs are flushed when
3454 static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3455 unsigned long action, void *hcpu)
3457 long cpu = (long)hcpu;
3458 struct kmem_cache *s;
3459 unsigned long flags;
3462 case CPU_UP_CANCELED:
3463 case CPU_UP_CANCELED_FROZEN:
3465 case CPU_DEAD_FROZEN:
3466 down_read(&slub_lock);
3467 list_for_each_entry(s, &slab_caches, list) {
3468 local_irq_save(flags);
3469 __flush_cpu_slab(s, cpu);
3470 local_irq_restore(flags);
3472 up_read(&slub_lock);
3480 static struct notifier_block __cpuinitdata slab_notifier = {
3481 .notifier_call = slab_cpuup_callback
3486 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
3488 struct kmem_cache *s;
3491 if (unlikely(size > SLUB_MAX_SIZE))
3492 return kmalloc_large(size, gfpflags);
3494 s = get_slab(size, gfpflags);
3496 if (unlikely(ZERO_OR_NULL_PTR(s)))
3499 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
3501 /* Honor the call site pointer we recieved. */
3502 trace_kmalloc(caller, ret, size, s->size, gfpflags);
3508 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
3509 int node, unsigned long caller)
3511 struct kmem_cache *s;
3514 if (unlikely(size > SLUB_MAX_SIZE)) {
3515 ret = kmalloc_large_node(size, gfpflags, node);
3517 trace_kmalloc_node(caller, ret,
3518 size, PAGE_SIZE << get_order(size),
3524 s = get_slab(size, gfpflags);
3526 if (unlikely(ZERO_OR_NULL_PTR(s)))
3529 ret = slab_alloc(s, gfpflags, node, caller);
3531 /* Honor the call site pointer we recieved. */
3532 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
3539 static int count_inuse(struct page *page)
3544 static int count_total(struct page *page)
3546 return page->objects;
3550 #ifdef CONFIG_SLUB_DEBUG
3551 static int validate_slab(struct kmem_cache *s, struct page *page,
3555 void *addr = page_address(page);
3557 if (!check_slab(s, page) ||
3558 !on_freelist(s, page, NULL))
3561 /* Now we know that a valid freelist exists */
3562 bitmap_zero(map, page->objects);
3564 get_map(s, page, map);
3565 for_each_object(p, s, addr, page->objects) {
3566 if (test_bit(slab_index(p, s, addr), map))
3567 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3571 for_each_object(p, s, addr, page->objects)
3572 if (!test_bit(slab_index(p, s, addr), map))
3573 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
3578 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3581 if (slab_trylock(page)) {
3582 validate_slab(s, page, map);
3585 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3589 static int validate_slab_node(struct kmem_cache *s,
3590 struct kmem_cache_node *n, unsigned long *map)
3592 unsigned long count = 0;
3594 unsigned long flags;
3596 spin_lock_irqsave(&n->list_lock, flags);
3598 list_for_each_entry(page, &n->partial, lru) {
3599 validate_slab_slab(s, page, map);
3602 if (count != n->nr_partial)
3603 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3604 "counter=%ld\n", s->name, count, n->nr_partial);
3606 if (!(s->flags & SLAB_STORE_USER))
3609 list_for_each_entry(page, &n->full, lru) {
3610 validate_slab_slab(s, page, map);
3613 if (count != atomic_long_read(&n->nr_slabs))
3614 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3615 "counter=%ld\n", s->name, count,
3616 atomic_long_read(&n->nr_slabs));
3619 spin_unlock_irqrestore(&n->list_lock, flags);
3623 static long validate_slab_cache(struct kmem_cache *s)
3626 unsigned long count = 0;
3627 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3628 sizeof(unsigned long), GFP_KERNEL);
3634 for_each_node_state(node, N_NORMAL_MEMORY) {
3635 struct kmem_cache_node *n = get_node(s, node);
3637 count += validate_slab_node(s, n, map);
3643 * Generate lists of code addresses where slabcache objects are allocated
3648 unsigned long count;
3655 DECLARE_BITMAP(cpus, NR_CPUS);
3661 unsigned long count;
3662 struct location *loc;
3665 static void free_loc_track(struct loc_track *t)
3668 free_pages((unsigned long)t->loc,
3669 get_order(sizeof(struct location) * t->max));
3672 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
3677 order = get_order(sizeof(struct location) * max);
3679 l = (void *)__get_free_pages(flags, order);
3684 memcpy(l, t->loc, sizeof(struct location) * t->count);
3692 static int add_location(struct loc_track *t, struct kmem_cache *s,
3693 const struct track *track)
3695 long start, end, pos;
3697 unsigned long caddr;
3698 unsigned long age = jiffies - track->when;
3704 pos = start + (end - start + 1) / 2;
3707 * There is nothing at "end". If we end up there
3708 * we need to add something to before end.
3713 caddr = t->loc[pos].addr;
3714 if (track->addr == caddr) {
3720 if (age < l->min_time)
3722 if (age > l->max_time)
3725 if (track->pid < l->min_pid)
3726 l->min_pid = track->pid;
3727 if (track->pid > l->max_pid)
3728 l->max_pid = track->pid;
3730 cpumask_set_cpu(track->cpu,
3731 to_cpumask(l->cpus));
3733 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3737 if (track->addr < caddr)
3744 * Not found. Insert new tracking element.
3746 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
3752 (t->count - pos) * sizeof(struct location));
3755 l->addr = track->addr;
3759 l->min_pid = track->pid;
3760 l->max_pid = track->pid;
3761 cpumask_clear(to_cpumask(l->cpus));
3762 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
3763 nodes_clear(l->nodes);
3764 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3768 static void process_slab(struct loc_track *t, struct kmem_cache *s,
3769 struct page *page, enum track_item alloc,
3772 void *addr = page_address(page);
3775 bitmap_zero(map, page->objects);
3776 get_map(s, page, map);
3778 for_each_object(p, s, addr, page->objects)
3779 if (!test_bit(slab_index(p, s, addr), map))
3780 add_location(t, s, get_track(s, p, alloc));
3783 static int list_locations(struct kmem_cache *s, char *buf,
3784 enum track_item alloc)
3788 struct loc_track t = { 0, 0, NULL };
3790 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3791 sizeof(unsigned long), GFP_KERNEL);
3793 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3796 return sprintf(buf, "Out of memory\n");
3798 /* Push back cpu slabs */
3801 for_each_node_state(node, N_NORMAL_MEMORY) {
3802 struct kmem_cache_node *n = get_node(s, node);
3803 unsigned long flags;
3806 if (!atomic_long_read(&n->nr_slabs))
3809 spin_lock_irqsave(&n->list_lock, flags);
3810 list_for_each_entry(page, &n->partial, lru)
3811 process_slab(&t, s, page, alloc, map);
3812 list_for_each_entry(page, &n->full, lru)
3813 process_slab(&t, s, page, alloc, map);
3814 spin_unlock_irqrestore(&n->list_lock, flags);
3817 for (i = 0; i < t.count; i++) {
3818 struct location *l = &t.loc[i];
3820 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
3822 len += sprintf(buf + len, "%7ld ", l->count);
3825 len += sprintf(buf + len, "%pS", (void *)l->addr);
3827 len += sprintf(buf + len, "<not-available>");
3829 if (l->sum_time != l->min_time) {
3830 len += sprintf(buf + len, " age=%ld/%ld/%ld",
3832 (long)div_u64(l->sum_time, l->count),
3835 len += sprintf(buf + len, " age=%ld",
3838 if (l->min_pid != l->max_pid)
3839 len += sprintf(buf + len, " pid=%ld-%ld",
3840 l->min_pid, l->max_pid);
3842 len += sprintf(buf + len, " pid=%ld",
3845 if (num_online_cpus() > 1 &&
3846 !cpumask_empty(to_cpumask(l->cpus)) &&
3847 len < PAGE_SIZE - 60) {
3848 len += sprintf(buf + len, " cpus=");
3849 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3850 to_cpumask(l->cpus));
3853 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
3854 len < PAGE_SIZE - 60) {
3855 len += sprintf(buf + len, " nodes=");
3856 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3860 len += sprintf(buf + len, "\n");
3866 len += sprintf(buf, "No data\n");
3871 #ifdef SLUB_RESILIENCY_TEST
3872 static void resiliency_test(void)
3876 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3878 printk(KERN_ERR "SLUB resiliency testing\n");
3879 printk(KERN_ERR "-----------------------\n");
3880 printk(KERN_ERR "A. Corruption after allocation\n");
3882 p = kzalloc(16, GFP_KERNEL);
3884 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3885 " 0x12->0x%p\n\n", p + 16);
3887 validate_slab_cache(kmalloc_caches[4]);
3889 /* Hmmm... The next two are dangerous */
3890 p = kzalloc(32, GFP_KERNEL);
3891 p[32 + sizeof(void *)] = 0x34;
3892 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3893 " 0x34 -> -0x%p\n", p);
3895 "If allocated object is overwritten then not detectable\n\n");
3897 validate_slab_cache(kmalloc_caches[5]);
3898 p = kzalloc(64, GFP_KERNEL);
3899 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3901 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3904 "If allocated object is overwritten then not detectable\n\n");
3905 validate_slab_cache(kmalloc_caches[6]);
3907 printk(KERN_ERR "\nB. Corruption after free\n");
3908 p = kzalloc(128, GFP_KERNEL);
3911 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3912 validate_slab_cache(kmalloc_caches[7]);
3914 p = kzalloc(256, GFP_KERNEL);
3917 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3919 validate_slab_cache(kmalloc_caches[8]);
3921 p = kzalloc(512, GFP_KERNEL);
3924 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3925 validate_slab_cache(kmalloc_caches[9]);
3929 static void resiliency_test(void) {};
3934 enum slab_stat_type {
3935 SL_ALL, /* All slabs */
3936 SL_PARTIAL, /* Only partially allocated slabs */
3937 SL_CPU, /* Only slabs used for cpu caches */
3938 SL_OBJECTS, /* Determine allocated objects not slabs */
3939 SL_TOTAL /* Determine object capacity not slabs */
3942 #define SO_ALL (1 << SL_ALL)
3943 #define SO_PARTIAL (1 << SL_PARTIAL)
3944 #define SO_CPU (1 << SL_CPU)
3945 #define SO_OBJECTS (1 << SL_OBJECTS)
3946 #define SO_TOTAL (1 << SL_TOTAL)
3948 static ssize_t show_slab_objects(struct kmem_cache *s,
3949 char *buf, unsigned long flags)
3951 unsigned long total = 0;
3954 unsigned long *nodes;
3955 unsigned long *per_cpu;
3957 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
3960 per_cpu = nodes + nr_node_ids;
3962 if (flags & SO_CPU) {
3965 for_each_possible_cpu(cpu) {
3966 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3968 if (!c || c->node < 0)
3972 if (flags & SO_TOTAL)
3973 x = c->page->objects;
3974 else if (flags & SO_OBJECTS)
3980 nodes[c->node] += x;
3986 lock_memory_hotplug();
3987 #ifdef CONFIG_SLUB_DEBUG
3988 if (flags & SO_ALL) {
3989 for_each_node_state(node, N_NORMAL_MEMORY) {
3990 struct kmem_cache_node *n = get_node(s, node);
3992 if (flags & SO_TOTAL)
3993 x = atomic_long_read(&n->total_objects);
3994 else if (flags & SO_OBJECTS)
3995 x = atomic_long_read(&n->total_objects) -
3996 count_partial(n, count_free);
3999 x = atomic_long_read(&n->nr_slabs);
4006 if (flags & SO_PARTIAL) {
4007 for_each_node_state(node, N_NORMAL_MEMORY) {
4008 struct kmem_cache_node *n = get_node(s, node);
4010 if (flags & SO_TOTAL)
4011 x = count_partial(n, count_total);
4012 else if (flags & SO_OBJECTS)
4013 x = count_partial(n, count_inuse);
4020 x = sprintf(buf, "%lu", total);
4022 for_each_node_state(node, N_NORMAL_MEMORY)
4024 x += sprintf(buf + x, " N%d=%lu",
4027 unlock_memory_hotplug();
4029 return x + sprintf(buf + x, "\n");
4032 #ifdef CONFIG_SLUB_DEBUG
4033 static int any_slab_objects(struct kmem_cache *s)
4037 for_each_online_node(node) {
4038 struct kmem_cache_node *n = get_node(s, node);
4043 if (atomic_long_read(&n->total_objects))
4050 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4051 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
4053 struct slab_attribute {
4054 struct attribute attr;
4055 ssize_t (*show)(struct kmem_cache *s, char *buf);
4056 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4059 #define SLAB_ATTR_RO(_name) \
4060 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4062 #define SLAB_ATTR(_name) \
4063 static struct slab_attribute _name##_attr = \
4064 __ATTR(_name, 0644, _name##_show, _name##_store)
4066 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4068 return sprintf(buf, "%d\n", s->size);
4070 SLAB_ATTR_RO(slab_size);
4072 static ssize_t align_show(struct kmem_cache *s, char *buf)
4074 return sprintf(buf, "%d\n", s->align);
4076 SLAB_ATTR_RO(align);
4078 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4080 return sprintf(buf, "%d\n", s->objsize);
4082 SLAB_ATTR_RO(object_size);
4084 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4086 return sprintf(buf, "%d\n", oo_objects(s->oo));
4088 SLAB_ATTR_RO(objs_per_slab);
4090 static ssize_t order_store(struct kmem_cache *s,
4091 const char *buf, size_t length)
4093 unsigned long order;
4096 err = strict_strtoul(buf, 10, &order);
4100 if (order > slub_max_order || order < slub_min_order)
4103 calculate_sizes(s, order);
4107 static ssize_t order_show(struct kmem_cache *s, char *buf)
4109 return sprintf(buf, "%d\n", oo_order(s->oo));
4113 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4115 return sprintf(buf, "%lu\n", s->min_partial);
4118 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4124 err = strict_strtoul(buf, 10, &min);
4128 set_min_partial(s, min);
4131 SLAB_ATTR(min_partial);
4133 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4137 return sprintf(buf, "%pS\n", s->ctor);
4141 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4143 return sprintf(buf, "%d\n", s->refcount - 1);
4145 SLAB_ATTR_RO(aliases);
4147 static ssize_t partial_show(struct kmem_cache *s, char *buf)
4149 return show_slab_objects(s, buf, SO_PARTIAL);
4151 SLAB_ATTR_RO(partial);
4153 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4155 return show_slab_objects(s, buf, SO_CPU);
4157 SLAB_ATTR_RO(cpu_slabs);
4159 static ssize_t objects_show(struct kmem_cache *s, char *buf)
4161 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
4163 SLAB_ATTR_RO(objects);
4165 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4167 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4169 SLAB_ATTR_RO(objects_partial);
4171 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4173 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4176 static ssize_t reclaim_account_store(struct kmem_cache *s,
4177 const char *buf, size_t length)
4179 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4181 s->flags |= SLAB_RECLAIM_ACCOUNT;
4184 SLAB_ATTR(reclaim_account);
4186 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4188 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
4190 SLAB_ATTR_RO(hwcache_align);
4192 #ifdef CONFIG_ZONE_DMA
4193 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4195 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4197 SLAB_ATTR_RO(cache_dma);
4200 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4202 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4204 SLAB_ATTR_RO(destroy_by_rcu);
4206 static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4208 return sprintf(buf, "%d\n", s->reserved);
4210 SLAB_ATTR_RO(reserved);
4212 #ifdef CONFIG_SLUB_DEBUG
4213 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4215 return show_slab_objects(s, buf, SO_ALL);
4217 SLAB_ATTR_RO(slabs);
4219 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4221 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4223 SLAB_ATTR_RO(total_objects);
4225 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4227 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4230 static ssize_t sanity_checks_store(struct kmem_cache *s,
4231 const char *buf, size_t length)
4233 s->flags &= ~SLAB_DEBUG_FREE;
4235 s->flags |= SLAB_DEBUG_FREE;
4238 SLAB_ATTR(sanity_checks);
4240 static ssize_t trace_show(struct kmem_cache *s, char *buf)
4242 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4245 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4248 s->flags &= ~SLAB_TRACE;
4250 s->flags |= SLAB_TRACE;
4255 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4257 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4260 static ssize_t red_zone_store(struct kmem_cache *s,
4261 const char *buf, size_t length)
4263 if (any_slab_objects(s))
4266 s->flags &= ~SLAB_RED_ZONE;
4268 s->flags |= SLAB_RED_ZONE;
4269 calculate_sizes(s, -1);
4272 SLAB_ATTR(red_zone);
4274 static ssize_t poison_show(struct kmem_cache *s, char *buf)
4276 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4279 static ssize_t poison_store(struct kmem_cache *s,
4280 const char *buf, size_t length)
4282 if (any_slab_objects(s))
4285 s->flags &= ~SLAB_POISON;
4287 s->flags |= SLAB_POISON;
4288 calculate_sizes(s, -1);
4293 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4295 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4298 static ssize_t store_user_store(struct kmem_cache *s,
4299 const char *buf, size_t length)
4301 if (any_slab_objects(s))
4304 s->flags &= ~SLAB_STORE_USER;
4306 s->flags |= SLAB_STORE_USER;
4307 calculate_sizes(s, -1);
4310 SLAB_ATTR(store_user);
4312 static ssize_t validate_show(struct kmem_cache *s, char *buf)
4317 static ssize_t validate_store(struct kmem_cache *s,
4318 const char *buf, size_t length)
4322 if (buf[0] == '1') {
4323 ret = validate_slab_cache(s);
4329 SLAB_ATTR(validate);
4331 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4333 if (!(s->flags & SLAB_STORE_USER))
4335 return list_locations(s, buf, TRACK_ALLOC);
4337 SLAB_ATTR_RO(alloc_calls);
4339 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4341 if (!(s->flags & SLAB_STORE_USER))
4343 return list_locations(s, buf, TRACK_FREE);
4345 SLAB_ATTR_RO(free_calls);
4346 #endif /* CONFIG_SLUB_DEBUG */
4348 #ifdef CONFIG_FAILSLAB
4349 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4351 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4354 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4357 s->flags &= ~SLAB_FAILSLAB;
4359 s->flags |= SLAB_FAILSLAB;
4362 SLAB_ATTR(failslab);
4365 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4370 static ssize_t shrink_store(struct kmem_cache *s,
4371 const char *buf, size_t length)
4373 if (buf[0] == '1') {
4374 int rc = kmem_cache_shrink(s);
4385 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
4387 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
4390 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
4391 const char *buf, size_t length)
4393 unsigned long ratio;
4396 err = strict_strtoul(buf, 10, &ratio);
4401 s->remote_node_defrag_ratio = ratio * 10;
4405 SLAB_ATTR(remote_node_defrag_ratio);
4408 #ifdef CONFIG_SLUB_STATS
4409 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4411 unsigned long sum = 0;
4414 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4419 for_each_online_cpu(cpu) {
4420 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
4426 len = sprintf(buf, "%lu", sum);
4429 for_each_online_cpu(cpu) {
4430 if (data[cpu] && len < PAGE_SIZE - 20)
4431 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
4435 return len + sprintf(buf + len, "\n");
4438 static void clear_stat(struct kmem_cache *s, enum stat_item si)
4442 for_each_online_cpu(cpu)
4443 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
4446 #define STAT_ATTR(si, text) \
4447 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4449 return show_stat(s, buf, si); \
4451 static ssize_t text##_store(struct kmem_cache *s, \
4452 const char *buf, size_t length) \
4454 if (buf[0] != '0') \
4456 clear_stat(s, si); \
4461 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4462 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4463 STAT_ATTR(FREE_FASTPATH, free_fastpath);
4464 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4465 STAT_ATTR(FREE_FROZEN, free_frozen);
4466 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4467 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4468 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4469 STAT_ATTR(ALLOC_SLAB, alloc_slab);
4470 STAT_ATTR(ALLOC_REFILL, alloc_refill);
4471 STAT_ATTR(FREE_SLAB, free_slab);
4472 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4473 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4474 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4475 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4476 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4477 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
4478 STAT_ATTR(ORDER_FALLBACK, order_fallback);
4481 static struct attribute *slab_attrs[] = {
4482 &slab_size_attr.attr,
4483 &object_size_attr.attr,
4484 &objs_per_slab_attr.attr,
4486 &min_partial_attr.attr,
4488 &objects_partial_attr.attr,
4490 &cpu_slabs_attr.attr,
4494 &hwcache_align_attr.attr,
4495 &reclaim_account_attr.attr,
4496 &destroy_by_rcu_attr.attr,
4498 &reserved_attr.attr,
4499 #ifdef CONFIG_SLUB_DEBUG
4500 &total_objects_attr.attr,
4502 &sanity_checks_attr.attr,
4504 &red_zone_attr.attr,
4506 &store_user_attr.attr,
4507 &validate_attr.attr,
4508 &alloc_calls_attr.attr,
4509 &free_calls_attr.attr,
4511 #ifdef CONFIG_ZONE_DMA
4512 &cache_dma_attr.attr,
4515 &remote_node_defrag_ratio_attr.attr,
4517 #ifdef CONFIG_SLUB_STATS
4518 &alloc_fastpath_attr.attr,
4519 &alloc_slowpath_attr.attr,
4520 &free_fastpath_attr.attr,
4521 &free_slowpath_attr.attr,
4522 &free_frozen_attr.attr,
4523 &free_add_partial_attr.attr,
4524 &free_remove_partial_attr.attr,
4525 &alloc_from_partial_attr.attr,
4526 &alloc_slab_attr.attr,
4527 &alloc_refill_attr.attr,
4528 &free_slab_attr.attr,
4529 &cpuslab_flush_attr.attr,
4530 &deactivate_full_attr.attr,
4531 &deactivate_empty_attr.attr,
4532 &deactivate_to_head_attr.attr,
4533 &deactivate_to_tail_attr.attr,
4534 &deactivate_remote_frees_attr.attr,
4535 &order_fallback_attr.attr,
4537 #ifdef CONFIG_FAILSLAB
4538 &failslab_attr.attr,
4544 static struct attribute_group slab_attr_group = {
4545 .attrs = slab_attrs,
4548 static ssize_t slab_attr_show(struct kobject *kobj,
4549 struct attribute *attr,
4552 struct slab_attribute *attribute;
4553 struct kmem_cache *s;
4556 attribute = to_slab_attr(attr);
4559 if (!attribute->show)
4562 err = attribute->show(s, buf);
4567 static ssize_t slab_attr_store(struct kobject *kobj,
4568 struct attribute *attr,
4569 const char *buf, size_t len)
4571 struct slab_attribute *attribute;
4572 struct kmem_cache *s;
4575 attribute = to_slab_attr(attr);
4578 if (!attribute->store)
4581 err = attribute->store(s, buf, len);
4586 static void kmem_cache_release(struct kobject *kobj)
4588 struct kmem_cache *s = to_slab(kobj);
4594 static const struct sysfs_ops slab_sysfs_ops = {
4595 .show = slab_attr_show,
4596 .store = slab_attr_store,
4599 static struct kobj_type slab_ktype = {
4600 .sysfs_ops = &slab_sysfs_ops,
4601 .release = kmem_cache_release
4604 static int uevent_filter(struct kset *kset, struct kobject *kobj)
4606 struct kobj_type *ktype = get_ktype(kobj);
4608 if (ktype == &slab_ktype)
4613 static const struct kset_uevent_ops slab_uevent_ops = {
4614 .filter = uevent_filter,
4617 static struct kset *slab_kset;
4619 #define ID_STR_LENGTH 64
4621 /* Create a unique string id for a slab cache:
4623 * Format :[flags-]size
4625 static char *create_unique_id(struct kmem_cache *s)
4627 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4634 * First flags affecting slabcache operations. We will only
4635 * get here for aliasable slabs so we do not need to support
4636 * too many flags. The flags here must cover all flags that
4637 * are matched during merging to guarantee that the id is
4640 if (s->flags & SLAB_CACHE_DMA)
4642 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4644 if (s->flags & SLAB_DEBUG_FREE)
4646 if (!(s->flags & SLAB_NOTRACK))
4650 p += sprintf(p, "%07d", s->size);
4651 BUG_ON(p > name + ID_STR_LENGTH - 1);
4655 static int sysfs_slab_add(struct kmem_cache *s)
4661 if (slab_state < SYSFS)
4662 /* Defer until later */
4665 unmergeable = slab_unmergeable(s);
4668 * Slabcache can never be merged so we can use the name proper.
4669 * This is typically the case for debug situations. In that
4670 * case we can catch duplicate names easily.
4672 sysfs_remove_link(&slab_kset->kobj, s->name);
4676 * Create a unique name for the slab as a target
4679 name = create_unique_id(s);
4682 s->kobj.kset = slab_kset;
4683 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4685 kobject_put(&s->kobj);
4689 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4691 kobject_del(&s->kobj);
4692 kobject_put(&s->kobj);
4695 kobject_uevent(&s->kobj, KOBJ_ADD);
4697 /* Setup first alias */
4698 sysfs_slab_alias(s, s->name);
4704 static void sysfs_slab_remove(struct kmem_cache *s)
4706 if (slab_state < SYSFS)
4708 * Sysfs has not been setup yet so no need to remove the
4713 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4714 kobject_del(&s->kobj);
4715 kobject_put(&s->kobj);
4719 * Need to buffer aliases during bootup until sysfs becomes
4720 * available lest we lose that information.
4722 struct saved_alias {
4723 struct kmem_cache *s;
4725 struct saved_alias *next;
4728 static struct saved_alias *alias_list;
4730 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4732 struct saved_alias *al;
4734 if (slab_state == SYSFS) {
4736 * If we have a leftover link then remove it.
4738 sysfs_remove_link(&slab_kset->kobj, name);
4739 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
4742 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4748 al->next = alias_list;
4753 static int __init slab_sysfs_init(void)
4755 struct kmem_cache *s;
4758 down_write(&slub_lock);
4760 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
4762 up_write(&slub_lock);
4763 printk(KERN_ERR "Cannot register slab subsystem.\n");
4769 list_for_each_entry(s, &slab_caches, list) {
4770 err = sysfs_slab_add(s);
4772 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4773 " to sysfs\n", s->name);
4776 while (alias_list) {
4777 struct saved_alias *al = alias_list;
4779 alias_list = alias_list->next;
4780 err = sysfs_slab_alias(al->s, al->name);
4782 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4783 " %s to sysfs\n", s->name);
4787 up_write(&slub_lock);
4792 __initcall(slab_sysfs_init);
4793 #endif /* CONFIG_SYSFS */
4796 * The /proc/slabinfo ABI
4798 #ifdef CONFIG_SLABINFO
4799 static void print_slabinfo_header(struct seq_file *m)
4801 seq_puts(m, "slabinfo - version: 2.1\n");
4802 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4803 "<objperslab> <pagesperslab>");
4804 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4805 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4809 static void *s_start(struct seq_file *m, loff_t *pos)
4813 down_read(&slub_lock);
4815 print_slabinfo_header(m);
4817 return seq_list_start(&slab_caches, *pos);
4820 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4822 return seq_list_next(p, &slab_caches, pos);
4825 static void s_stop(struct seq_file *m, void *p)
4827 up_read(&slub_lock);
4830 static int s_show(struct seq_file *m, void *p)
4832 unsigned long nr_partials = 0;
4833 unsigned long nr_slabs = 0;
4834 unsigned long nr_inuse = 0;
4835 unsigned long nr_objs = 0;
4836 unsigned long nr_free = 0;
4837 struct kmem_cache *s;
4840 s = list_entry(p, struct kmem_cache, list);
4842 for_each_online_node(node) {
4843 struct kmem_cache_node *n = get_node(s, node);
4848 nr_partials += n->nr_partial;
4849 nr_slabs += atomic_long_read(&n->nr_slabs);
4850 nr_objs += atomic_long_read(&n->total_objects);
4851 nr_free += count_partial(n, count_free);
4854 nr_inuse = nr_objs - nr_free;
4856 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
4857 nr_objs, s->size, oo_objects(s->oo),
4858 (1 << oo_order(s->oo)));
4859 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4860 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4866 static const struct seq_operations slabinfo_op = {
4873 static int slabinfo_open(struct inode *inode, struct file *file)
4875 return seq_open(file, &slabinfo_op);
4878 static const struct file_operations proc_slabinfo_operations = {
4879 .open = slabinfo_open,
4881 .llseek = seq_lseek,
4882 .release = seq_release,
4885 static int __init slab_proc_init(void)
4887 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
4890 module_init(slab_proc_init);
4891 #endif /* CONFIG_SLABINFO */