2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks and only
6 * uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
12 #include <linux/swap.h> /* struct reclaim_state */
13 #include <linux/module.h>
14 #include <linux/bit_spinlock.h>
15 #include <linux/interrupt.h>
16 #include <linux/bitops.h>
17 #include <linux/slab.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/kmemcheck.h>
21 #include <linux/cpu.h>
22 #include <linux/cpuset.h>
23 #include <linux/mempolicy.h>
24 #include <linux/ctype.h>
25 #include <linux/debugobjects.h>
26 #include <linux/kallsyms.h>
27 #include <linux/memory.h>
28 #include <linux/math64.h>
29 #include <linux/fault-inject.h>
31 #include <trace/events/kmem.h>
38 * The slab_lock protects operations on the object of a particular
39 * slab and its metadata in the page struct. If the slab lock
40 * has been taken then no allocations nor frees can be performed
41 * on the objects in the slab nor can the slab be added or removed
42 * from the partial or full lists since this would mean modifying
43 * the page_struct of the slab.
45 * The list_lock protects the partial and full list on each node and
46 * the partial slab counter. If taken then no new slabs may be added or
47 * removed from the lists nor make the number of partial slabs be modified.
48 * (Note that the total number of slabs is an atomic value that may be
49 * modified without taking the list lock).
51 * The list_lock is a centralized lock and thus we avoid taking it as
52 * much as possible. As long as SLUB does not have to handle partial
53 * slabs, operations can continue without any centralized lock. F.e.
54 * allocating a long series of objects that fill up slabs does not require
57 * The lock order is sometimes inverted when we are trying to get a slab
58 * off a list. We take the list_lock and then look for a page on the list
59 * to use. While we do that objects in the slabs may be freed. We can
60 * only operate on the slab if we have also taken the slab_lock. So we use
61 * a slab_trylock() on the slab. If trylock was successful then no frees
62 * can occur anymore and we can use the slab for allocations etc. If the
63 * slab_trylock() does not succeed then frees are in progress in the slab and
64 * we must stay away from it for a while since we may cause a bouncing
65 * cacheline if we try to acquire the lock. So go onto the next slab.
66 * If all pages are busy then we may allocate a new slab instead of reusing
67 * a partial slab. A new slab has noone operating on it and thus there is
68 * no danger of cacheline contention.
70 * Interrupts are disabled during allocation and deallocation in order to
71 * make the slab allocator safe to use in the context of an irq. In addition
72 * interrupts are disabled to ensure that the processor does not change
73 * while handling per_cpu slabs, due to kernel preemption.
75 * SLUB assigns one slab for allocation to each processor.
76 * Allocations only occur from these slabs called cpu slabs.
78 * Slabs with free elements are kept on a partial list and during regular
79 * operations no list for full slabs is used. If an object in a full slab is
80 * freed then the slab will show up again on the partial lists.
81 * We track full slabs for debugging purposes though because otherwise we
82 * cannot scan all objects.
84 * Slabs are freed when they become empty. Teardown and setup is
85 * minimal so we rely on the page allocators per cpu caches for
86 * fast frees and allocs.
88 * Overloading of page flags that are otherwise used for LRU management.
90 * PageActive The slab is frozen and exempt from list processing.
91 * This means that the slab is dedicated to a purpose
92 * such as satisfying allocations for a specific
93 * processor. Objects may be freed in the slab while
94 * it is frozen but slab_free will then skip the usual
95 * list operations. It is up to the processor holding
96 * the slab to integrate the slab into the slab lists
97 * when the slab is no longer needed.
99 * One use of this flag is to mark slabs that are
100 * used for allocations. Then such a slab becomes a cpu
101 * slab. The cpu slab may be equipped with an additional
102 * freelist that allows lockless access to
103 * free objects in addition to the regular freelist
104 * that requires the slab lock.
106 * PageError Slab requires special handling due to debug
107 * options set. This moves slab handling out of
108 * the fast path and disables lockless freelists.
111 #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
112 SLAB_TRACE | SLAB_DEBUG_FREE)
114 static inline int kmem_cache_debug(struct kmem_cache *s)
116 #ifdef CONFIG_SLUB_DEBUG
117 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
124 * Issues still to be resolved:
126 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
128 * - Variable sizing of the per node arrays
131 /* Enable to test recovery from slab corruption on boot */
132 #undef SLUB_RESILIENCY_TEST
135 * Mininum number of partial slabs. These will be left on the partial
136 * lists even if they are empty. kmem_cache_shrink may reclaim them.
138 #define MIN_PARTIAL 5
141 * Maximum number of desirable partial slabs.
142 * The existence of more partial slabs makes kmem_cache_shrink
143 * sort the partial list by the number of objects in the.
145 #define MAX_PARTIAL 10
147 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
148 SLAB_POISON | SLAB_STORE_USER)
151 * Debugging flags that require metadata to be stored in the slab. These get
152 * disabled when slub_debug=O is used and a cache's min order increases with
155 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
158 * Set of flags that will prevent slab merging
160 #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
161 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
164 #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
165 SLAB_CACHE_DMA | SLAB_NOTRACK)
168 #define OO_MASK ((1 << OO_SHIFT) - 1)
169 #define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
171 /* Internal SLUB flags */
172 #define __OBJECT_POISON 0x80000000UL /* Poison object */
174 static int kmem_size = sizeof(struct kmem_cache);
177 static struct notifier_block slab_notifier;
181 DOWN, /* No slab functionality available */
182 PARTIAL, /* Kmem_cache_node works */
183 UP, /* Everything works but does not show up in sysfs */
187 /* A list of all slab caches on the system */
188 static DECLARE_RWSEM(slub_lock);
189 static LIST_HEAD(slab_caches);
192 * Tracking user of a slab.
195 unsigned long addr; /* Called from address */
196 int cpu; /* Was running on cpu */
197 int pid; /* Pid context */
198 unsigned long when; /* When did the operation occur */
201 enum track_item { TRACK_ALLOC, TRACK_FREE };
204 static int sysfs_slab_add(struct kmem_cache *);
205 static int sysfs_slab_alias(struct kmem_cache *, const char *);
206 static void sysfs_slab_remove(struct kmem_cache *);
209 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
210 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
212 static inline void sysfs_slab_remove(struct kmem_cache *s)
220 static inline void stat(struct kmem_cache *s, enum stat_item si)
222 #ifdef CONFIG_SLUB_STATS
223 __this_cpu_inc(s->cpu_slab->stat[si]);
227 /********************************************************************
228 * Core slab cache functions
229 *******************************************************************/
231 int slab_is_available(void)
233 return slab_state >= UP;
236 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
238 return s->node[node];
241 /* Verify that a pointer has an address that is valid within a slab page */
242 static inline int check_valid_pointer(struct kmem_cache *s,
243 struct page *page, const void *object)
250 base = page_address(page);
251 if (object < base || object >= base + page->objects * s->size ||
252 (object - base) % s->size) {
259 static inline void *get_freepointer(struct kmem_cache *s, void *object)
261 return *(void **)(object + s->offset);
264 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
266 *(void **)(object + s->offset) = fp;
269 /* Loop over all objects in a slab */
270 #define for_each_object(__p, __s, __addr, __objects) \
271 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
275 #define for_each_free_object(__p, __s, __free) \
276 for (__p = (__free); __p; __p = get_freepointer((__s), __p))
278 /* Determine object index from a given position */
279 static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
281 return (p - addr) / s->size;
284 static inline struct kmem_cache_order_objects oo_make(int order,
287 struct kmem_cache_order_objects x = {
288 (order << OO_SHIFT) + (PAGE_SIZE << order) / size
294 static inline int oo_order(struct kmem_cache_order_objects x)
296 return x.x >> OO_SHIFT;
299 static inline int oo_objects(struct kmem_cache_order_objects x)
301 return x.x & OO_MASK;
304 #ifdef CONFIG_SLUB_DEBUG
308 #ifdef CONFIG_SLUB_DEBUG_ON
309 static int slub_debug = DEBUG_DEFAULT_FLAGS;
311 static int slub_debug;
314 static char *slub_debug_slabs;
315 static int disable_higher_order_debug;
320 static void print_section(char *text, u8 *addr, unsigned int length)
328 for (i = 0; i < length; i++) {
330 printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
333 printk(KERN_CONT " %02x", addr[i]);
335 ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
337 printk(KERN_CONT " %s\n", ascii);
344 printk(KERN_CONT " ");
348 printk(KERN_CONT " %s\n", ascii);
352 static struct track *get_track(struct kmem_cache *s, void *object,
353 enum track_item alloc)
358 p = object + s->offset + sizeof(void *);
360 p = object + s->inuse;
365 static void set_track(struct kmem_cache *s, void *object,
366 enum track_item alloc, unsigned long addr)
368 struct track *p = get_track(s, object, alloc);
372 p->cpu = smp_processor_id();
373 p->pid = current->pid;
376 memset(p, 0, sizeof(struct track));
379 static void init_tracking(struct kmem_cache *s, void *object)
381 if (!(s->flags & SLAB_STORE_USER))
384 set_track(s, object, TRACK_FREE, 0UL);
385 set_track(s, object, TRACK_ALLOC, 0UL);
388 static void print_track(const char *s, struct track *t)
393 printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
394 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
397 static void print_tracking(struct kmem_cache *s, void *object)
399 if (!(s->flags & SLAB_STORE_USER))
402 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
403 print_track("Freed", get_track(s, object, TRACK_FREE));
406 static void print_page_info(struct page *page)
408 printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
409 page, page->objects, page->inuse, page->freelist, page->flags);
413 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
419 vsnprintf(buf, sizeof(buf), fmt, args);
421 printk(KERN_ERR "========================================"
422 "=====================================\n");
423 printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
424 printk(KERN_ERR "----------------------------------------"
425 "-------------------------------------\n\n");
428 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
434 vsnprintf(buf, sizeof(buf), fmt, args);
436 printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
439 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
441 unsigned int off; /* Offset of last byte */
442 u8 *addr = page_address(page);
444 print_tracking(s, p);
446 print_page_info(page);
448 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
449 p, p - addr, get_freepointer(s, p));
452 print_section("Bytes b4", p - 16, 16);
454 print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
456 if (s->flags & SLAB_RED_ZONE)
457 print_section("Redzone", p + s->objsize,
458 s->inuse - s->objsize);
461 off = s->offset + sizeof(void *);
465 if (s->flags & SLAB_STORE_USER)
466 off += 2 * sizeof(struct track);
469 /* Beginning of the filler is the free pointer */
470 print_section("Padding", p + off, s->size - off);
475 static void object_err(struct kmem_cache *s, struct page *page,
476 u8 *object, char *reason)
478 slab_bug(s, "%s", reason);
479 print_trailer(s, page, object);
482 static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
488 vsnprintf(buf, sizeof(buf), fmt, args);
490 slab_bug(s, "%s", buf);
491 print_page_info(page);
495 static void init_object(struct kmem_cache *s, void *object, u8 val)
499 if (s->flags & __OBJECT_POISON) {
500 memset(p, POISON_FREE, s->objsize - 1);
501 p[s->objsize - 1] = POISON_END;
504 if (s->flags & SLAB_RED_ZONE)
505 memset(p + s->objsize, val, s->inuse - s->objsize);
508 static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
511 if (*start != (u8)value)
519 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
520 void *from, void *to)
522 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
523 memset(from, data, to - from);
526 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
527 u8 *object, char *what,
528 u8 *start, unsigned int value, unsigned int bytes)
533 fault = check_bytes(start, value, bytes);
538 while (end > fault && end[-1] == value)
541 slab_bug(s, "%s overwritten", what);
542 printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
543 fault, end - 1, fault[0], value);
544 print_trailer(s, page, object);
546 restore_bytes(s, what, value, fault, end);
554 * Bytes of the object to be managed.
555 * If the freepointer may overlay the object then the free
556 * pointer is the first word of the object.
558 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
561 * object + s->objsize
562 * Padding to reach word boundary. This is also used for Redzoning.
563 * Padding is extended by another word if Redzoning is enabled and
566 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
567 * 0xcc (RED_ACTIVE) for objects in use.
570 * Meta data starts here.
572 * A. Free pointer (if we cannot overwrite object on free)
573 * B. Tracking data for SLAB_STORE_USER
574 * C. Padding to reach required alignment boundary or at mininum
575 * one word if debugging is on to be able to detect writes
576 * before the word boundary.
578 * Padding is done using 0x5a (POISON_INUSE)
581 * Nothing is used beyond s->size.
583 * If slabcaches are merged then the objsize and inuse boundaries are mostly
584 * ignored. And therefore no slab options that rely on these boundaries
585 * may be used with merged slabcaches.
588 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
590 unsigned long off = s->inuse; /* The end of info */
593 /* Freepointer is placed after the object. */
594 off += sizeof(void *);
596 if (s->flags & SLAB_STORE_USER)
597 /* We also have user information there */
598 off += 2 * sizeof(struct track);
603 return check_bytes_and_report(s, page, p, "Object padding",
604 p + off, POISON_INUSE, s->size - off);
607 /* Check the pad bytes at the end of a slab page */
608 static int slab_pad_check(struct kmem_cache *s, struct page *page)
616 if (!(s->flags & SLAB_POISON))
619 start = page_address(page);
620 length = (PAGE_SIZE << compound_order(page));
621 end = start + length;
622 remainder = length % s->size;
626 fault = check_bytes(end - remainder, POISON_INUSE, remainder);
629 while (end > fault && end[-1] == POISON_INUSE)
632 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
633 print_section("Padding", end - remainder, remainder);
635 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
639 static int check_object(struct kmem_cache *s, struct page *page,
640 void *object, u8 val)
643 u8 *endobject = object + s->objsize;
645 if (s->flags & SLAB_RED_ZONE) {
646 if (!check_bytes_and_report(s, page, object, "Redzone",
647 endobject, val, s->inuse - s->objsize))
650 if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
651 check_bytes_and_report(s, page, p, "Alignment padding",
652 endobject, POISON_INUSE, s->inuse - s->objsize);
656 if (s->flags & SLAB_POISON) {
657 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
658 (!check_bytes_and_report(s, page, p, "Poison", p,
659 POISON_FREE, s->objsize - 1) ||
660 !check_bytes_and_report(s, page, p, "Poison",
661 p + s->objsize - 1, POISON_END, 1)))
664 * check_pad_bytes cleans up on its own.
666 check_pad_bytes(s, page, p);
669 if (!s->offset && val == SLUB_RED_ACTIVE)
671 * Object and freepointer overlap. Cannot check
672 * freepointer while object is allocated.
676 /* Check free pointer validity */
677 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
678 object_err(s, page, p, "Freepointer corrupt");
680 * No choice but to zap it and thus lose the remainder
681 * of the free objects in this slab. May cause
682 * another error because the object count is now wrong.
684 set_freepointer(s, p, NULL);
690 static int check_slab(struct kmem_cache *s, struct page *page)
694 VM_BUG_ON(!irqs_disabled());
696 if (!PageSlab(page)) {
697 slab_err(s, page, "Not a valid slab page");
701 maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
702 if (page->objects > maxobj) {
703 slab_err(s, page, "objects %u > max %u",
704 s->name, page->objects, maxobj);
707 if (page->inuse > page->objects) {
708 slab_err(s, page, "inuse %u > max %u",
709 s->name, page->inuse, page->objects);
712 /* Slab_pad_check fixes things up after itself */
713 slab_pad_check(s, page);
718 * Determine if a certain object on a page is on the freelist. Must hold the
719 * slab lock to guarantee that the chains are in a consistent state.
721 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
724 void *fp = page->freelist;
726 unsigned long max_objects;
728 while (fp && nr <= page->objects) {
731 if (!check_valid_pointer(s, page, fp)) {
733 object_err(s, page, object,
734 "Freechain corrupt");
735 set_freepointer(s, object, NULL);
738 slab_err(s, page, "Freepointer corrupt");
739 page->freelist = NULL;
740 page->inuse = page->objects;
741 slab_fix(s, "Freelist cleared");
747 fp = get_freepointer(s, object);
751 max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
752 if (max_objects > MAX_OBJS_PER_PAGE)
753 max_objects = MAX_OBJS_PER_PAGE;
755 if (page->objects != max_objects) {
756 slab_err(s, page, "Wrong number of objects. Found %d but "
757 "should be %d", page->objects, max_objects);
758 page->objects = max_objects;
759 slab_fix(s, "Number of objects adjusted.");
761 if (page->inuse != page->objects - nr) {
762 slab_err(s, page, "Wrong object count. Counter is %d but "
763 "counted were %d", page->inuse, page->objects - nr);
764 page->inuse = page->objects - nr;
765 slab_fix(s, "Object count adjusted.");
767 return search == NULL;
770 static void trace(struct kmem_cache *s, struct page *page, void *object,
773 if (s->flags & SLAB_TRACE) {
774 printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
776 alloc ? "alloc" : "free",
781 print_section("Object", (void *)object, s->objsize);
788 * Hooks for other subsystems that check memory allocations. In a typical
789 * production configuration these hooks all should produce no code at all.
791 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
793 flags &= gfp_allowed_mask;
794 lockdep_trace_alloc(flags);
795 might_sleep_if(flags & __GFP_WAIT);
797 return should_failslab(s->objsize, flags, s->flags);
800 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object)
802 flags &= gfp_allowed_mask;
803 kmemcheck_slab_alloc(s, flags, object, s->objsize);
804 kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags);
807 static inline void slab_free_hook(struct kmem_cache *s, void *x)
809 kmemleak_free_recursive(x, s->flags);
812 * Trouble is that we may no longer disable interupts in the fast path
813 * So in order to make the debug calls that expect irqs to be
814 * disabled we need to disable interrupts temporarily.
816 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
820 local_irq_save(flags);
821 kmemcheck_slab_free(s, x, s->objsize);
822 debug_check_no_locks_freed(x, s->objsize);
823 if (!(s->flags & SLAB_DEBUG_OBJECTS))
824 debug_check_no_obj_freed(x, s->objsize);
825 local_irq_restore(flags);
831 * Tracking of fully allocated slabs for debugging purposes.
833 static void add_full(struct kmem_cache_node *n, struct page *page)
835 spin_lock(&n->list_lock);
836 list_add(&page->lru, &n->full);
837 spin_unlock(&n->list_lock);
840 static void remove_full(struct kmem_cache *s, struct page *page)
842 struct kmem_cache_node *n;
844 if (!(s->flags & SLAB_STORE_USER))
847 n = get_node(s, page_to_nid(page));
849 spin_lock(&n->list_lock);
850 list_del(&page->lru);
851 spin_unlock(&n->list_lock);
854 /* Tracking of the number of slabs for debugging purposes */
855 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
857 struct kmem_cache_node *n = get_node(s, node);
859 return atomic_long_read(&n->nr_slabs);
862 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
864 return atomic_long_read(&n->nr_slabs);
867 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
869 struct kmem_cache_node *n = get_node(s, node);
872 * May be called early in order to allocate a slab for the
873 * kmem_cache_node structure. Solve the chicken-egg
874 * dilemma by deferring the increment of the count during
875 * bootstrap (see early_kmem_cache_node_alloc).
878 atomic_long_inc(&n->nr_slabs);
879 atomic_long_add(objects, &n->total_objects);
882 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
884 struct kmem_cache_node *n = get_node(s, node);
886 atomic_long_dec(&n->nr_slabs);
887 atomic_long_sub(objects, &n->total_objects);
890 /* Object debug checks for alloc/free paths */
891 static void setup_object_debug(struct kmem_cache *s, struct page *page,
894 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
897 init_object(s, object, SLUB_RED_INACTIVE);
898 init_tracking(s, object);
901 static noinline int alloc_debug_processing(struct kmem_cache *s, struct page *page,
902 void *object, unsigned long addr)
904 if (!check_slab(s, page))
907 if (!on_freelist(s, page, object)) {
908 object_err(s, page, object, "Object already allocated");
912 if (!check_valid_pointer(s, page, object)) {
913 object_err(s, page, object, "Freelist Pointer check fails");
917 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
920 /* Success perform special debug activities for allocs */
921 if (s->flags & SLAB_STORE_USER)
922 set_track(s, object, TRACK_ALLOC, addr);
923 trace(s, page, object, 1);
924 init_object(s, object, SLUB_RED_ACTIVE);
928 if (PageSlab(page)) {
930 * If this is a slab page then lets do the best we can
931 * to avoid issues in the future. Marking all objects
932 * as used avoids touching the remaining objects.
934 slab_fix(s, "Marking all objects used");
935 page->inuse = page->objects;
936 page->freelist = NULL;
941 static noinline int free_debug_processing(struct kmem_cache *s,
942 struct page *page, void *object, unsigned long addr)
944 if (!check_slab(s, page))
947 if (!check_valid_pointer(s, page, object)) {
948 slab_err(s, page, "Invalid object pointer 0x%p", object);
952 if (on_freelist(s, page, object)) {
953 object_err(s, page, object, "Object already free");
957 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
960 if (unlikely(s != page->slab)) {
961 if (!PageSlab(page)) {
962 slab_err(s, page, "Attempt to free object(0x%p) "
963 "outside of slab", object);
964 } else if (!page->slab) {
966 "SLUB <none>: no slab for object 0x%p.\n",
970 object_err(s, page, object,
971 "page slab pointer corrupt.");
975 /* Special debug activities for freeing objects */
976 if (!PageSlubFrozen(page) && !page->freelist)
977 remove_full(s, page);
978 if (s->flags & SLAB_STORE_USER)
979 set_track(s, object, TRACK_FREE, addr);
980 trace(s, page, object, 0);
981 init_object(s, object, SLUB_RED_INACTIVE);
985 slab_fix(s, "Object at 0x%p not freed", object);
989 static int __init setup_slub_debug(char *str)
991 slub_debug = DEBUG_DEFAULT_FLAGS;
992 if (*str++ != '=' || !*str)
994 * No options specified. Switch on full debugging.
1000 * No options but restriction on slabs. This means full
1001 * debugging for slabs matching a pattern.
1005 if (tolower(*str) == 'o') {
1007 * Avoid enabling debugging on caches if its minimum order
1008 * would increase as a result.
1010 disable_higher_order_debug = 1;
1017 * Switch off all debugging measures.
1022 * Determine which debug features should be switched on
1024 for (; *str && *str != ','; str++) {
1025 switch (tolower(*str)) {
1027 slub_debug |= SLAB_DEBUG_FREE;
1030 slub_debug |= SLAB_RED_ZONE;
1033 slub_debug |= SLAB_POISON;
1036 slub_debug |= SLAB_STORE_USER;
1039 slub_debug |= SLAB_TRACE;
1042 slub_debug |= SLAB_FAILSLAB;
1045 printk(KERN_ERR "slub_debug option '%c' "
1046 "unknown. skipped\n", *str);
1052 slub_debug_slabs = str + 1;
1057 __setup("slub_debug", setup_slub_debug);
1059 static unsigned long kmem_cache_flags(unsigned long objsize,
1060 unsigned long flags, const char *name,
1061 void (*ctor)(void *))
1064 * Enable debugging if selected on the kernel commandline.
1066 if (slub_debug && (!slub_debug_slabs ||
1067 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1068 flags |= slub_debug;
1073 static inline void setup_object_debug(struct kmem_cache *s,
1074 struct page *page, void *object) {}
1076 static inline int alloc_debug_processing(struct kmem_cache *s,
1077 struct page *page, void *object, unsigned long addr) { return 0; }
1079 static inline int free_debug_processing(struct kmem_cache *s,
1080 struct page *page, void *object, unsigned long addr) { return 0; }
1082 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1084 static inline int check_object(struct kmem_cache *s, struct page *page,
1085 void *object, u8 val) { return 1; }
1086 static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
1087 static inline unsigned long kmem_cache_flags(unsigned long objsize,
1088 unsigned long flags, const char *name,
1089 void (*ctor)(void *))
1093 #define slub_debug 0
1095 #define disable_higher_order_debug 0
1097 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1099 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1101 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1103 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1106 static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1109 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
1112 static inline void slab_free_hook(struct kmem_cache *s, void *x) {}
1114 #endif /* CONFIG_SLUB_DEBUG */
1117 * Slab allocation and freeing
1119 static inline struct page *alloc_slab_page(gfp_t flags, int node,
1120 struct kmem_cache_order_objects oo)
1122 int order = oo_order(oo);
1124 flags |= __GFP_NOTRACK;
1126 if (node == NUMA_NO_NODE)
1127 return alloc_pages(flags, order);
1129 return alloc_pages_exact_node(node, flags, order);
1132 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1135 struct kmem_cache_order_objects oo = s->oo;
1138 flags |= s->allocflags;
1141 * Let the initial higher-order allocation fail under memory pressure
1142 * so we fall-back to the minimum order allocation.
1144 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1146 page = alloc_slab_page(alloc_gfp, node, oo);
1147 if (unlikely(!page)) {
1150 * Allocation may have failed due to fragmentation.
1151 * Try a lower order alloc if possible
1153 page = alloc_slab_page(flags, node, oo);
1157 stat(s, ORDER_FALLBACK);
1160 if (kmemcheck_enabled
1161 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
1162 int pages = 1 << oo_order(oo);
1164 kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1167 * Objects from caches that have a constructor don't get
1168 * cleared when they're allocated, so we need to do it here.
1171 kmemcheck_mark_uninitialized_pages(page, pages);
1173 kmemcheck_mark_unallocated_pages(page, pages);
1176 page->objects = oo_objects(oo);
1177 mod_zone_page_state(page_zone(page),
1178 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1179 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1185 static void setup_object(struct kmem_cache *s, struct page *page,
1188 setup_object_debug(s, page, object);
1189 if (unlikely(s->ctor))
1193 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1200 BUG_ON(flags & GFP_SLAB_BUG_MASK);
1202 page = allocate_slab(s,
1203 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1207 inc_slabs_node(s, page_to_nid(page), page->objects);
1209 page->flags |= 1 << PG_slab;
1211 start = page_address(page);
1213 if (unlikely(s->flags & SLAB_POISON))
1214 memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
1217 for_each_object(p, s, start, page->objects) {
1218 setup_object(s, page, last);
1219 set_freepointer(s, last, p);
1222 setup_object(s, page, last);
1223 set_freepointer(s, last, NULL);
1225 page->freelist = start;
1231 static void __free_slab(struct kmem_cache *s, struct page *page)
1233 int order = compound_order(page);
1234 int pages = 1 << order;
1236 if (kmem_cache_debug(s)) {
1239 slab_pad_check(s, page);
1240 for_each_object(p, s, page_address(page),
1242 check_object(s, page, p, SLUB_RED_INACTIVE);
1245 kmemcheck_free_shadow(page, compound_order(page));
1247 mod_zone_page_state(page_zone(page),
1248 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1249 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1252 __ClearPageSlab(page);
1253 reset_page_mapcount(page);
1254 if (current->reclaim_state)
1255 current->reclaim_state->reclaimed_slab += pages;
1256 __free_pages(page, order);
1259 static void rcu_free_slab(struct rcu_head *h)
1263 page = container_of((struct list_head *)h, struct page, lru);
1264 __free_slab(page->slab, page);
1267 static void free_slab(struct kmem_cache *s, struct page *page)
1269 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1271 * RCU free overloads the RCU head over the LRU
1273 struct rcu_head *head = (void *)&page->lru;
1275 call_rcu(head, rcu_free_slab);
1277 __free_slab(s, page);
1280 static void discard_slab(struct kmem_cache *s, struct page *page)
1282 dec_slabs_node(s, page_to_nid(page), page->objects);
1287 * Per slab locking using the pagelock
1289 static __always_inline void slab_lock(struct page *page)
1291 bit_spin_lock(PG_locked, &page->flags);
1294 static __always_inline void slab_unlock(struct page *page)
1296 __bit_spin_unlock(PG_locked, &page->flags);
1299 static __always_inline int slab_trylock(struct page *page)
1303 rc = bit_spin_trylock(PG_locked, &page->flags);
1308 * Management of partially allocated slabs
1310 static void add_partial(struct kmem_cache_node *n,
1311 struct page *page, int tail)
1313 spin_lock(&n->list_lock);
1316 list_add_tail(&page->lru, &n->partial);
1318 list_add(&page->lru, &n->partial);
1319 spin_unlock(&n->list_lock);
1322 static inline void __remove_partial(struct kmem_cache_node *n,
1325 list_del(&page->lru);
1329 static void remove_partial(struct kmem_cache *s, struct page *page)
1331 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1333 spin_lock(&n->list_lock);
1334 __remove_partial(n, page);
1335 spin_unlock(&n->list_lock);
1339 * Lock slab and remove from the partial list.
1341 * Must hold list_lock.
1343 static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1346 if (slab_trylock(page)) {
1347 __remove_partial(n, page);
1348 __SetPageSlubFrozen(page);
1355 * Try to allocate a partial slab from a specific node.
1357 static struct page *get_partial_node(struct kmem_cache_node *n)
1362 * Racy check. If we mistakenly see no partial slabs then we
1363 * just allocate an empty slab. If we mistakenly try to get a
1364 * partial slab and there is none available then get_partials()
1367 if (!n || !n->nr_partial)
1370 spin_lock(&n->list_lock);
1371 list_for_each_entry(page, &n->partial, lru)
1372 if (lock_and_freeze_slab(n, page))
1376 spin_unlock(&n->list_lock);
1381 * Get a page from somewhere. Search in increasing NUMA distances.
1383 static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1386 struct zonelist *zonelist;
1389 enum zone_type high_zoneidx = gfp_zone(flags);
1393 * The defrag ratio allows a configuration of the tradeoffs between
1394 * inter node defragmentation and node local allocations. A lower
1395 * defrag_ratio increases the tendency to do local allocations
1396 * instead of attempting to obtain partial slabs from other nodes.
1398 * If the defrag_ratio is set to 0 then kmalloc() always
1399 * returns node local objects. If the ratio is higher then kmalloc()
1400 * may return off node objects because partial slabs are obtained
1401 * from other nodes and filled up.
1403 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1404 * defrag_ratio = 1000) then every (well almost) allocation will
1405 * first attempt to defrag slab caches on other nodes. This means
1406 * scanning over all nodes to look for partial slabs which may be
1407 * expensive if we do it every time we are trying to find a slab
1408 * with available objects.
1410 if (!s->remote_node_defrag_ratio ||
1411 get_cycles() % 1024 > s->remote_node_defrag_ratio)
1415 zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1416 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1417 struct kmem_cache_node *n;
1419 n = get_node(s, zone_to_nid(zone));
1421 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1422 n->nr_partial > s->min_partial) {
1423 page = get_partial_node(n);
1436 * Get a partial page, lock it and return it.
1438 static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1441 int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node;
1443 page = get_partial_node(get_node(s, searchnode));
1444 if (page || node != -1)
1447 return get_any_partial(s, flags);
1451 * Move a page back to the lists.
1453 * Must be called with the slab lock held.
1455 * On exit the slab lock will have been dropped.
1457 static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
1460 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1462 __ClearPageSlubFrozen(page);
1465 if (page->freelist) {
1466 add_partial(n, page, tail);
1467 stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1469 stat(s, DEACTIVATE_FULL);
1470 if (kmem_cache_debug(s) && (s->flags & SLAB_STORE_USER))
1475 stat(s, DEACTIVATE_EMPTY);
1476 if (n->nr_partial < s->min_partial) {
1478 * Adding an empty slab to the partial slabs in order
1479 * to avoid page allocator overhead. This slab needs
1480 * to come after the other slabs with objects in
1481 * so that the others get filled first. That way the
1482 * size of the partial list stays small.
1484 * kmem_cache_shrink can reclaim any empty slabs from
1487 add_partial(n, page, 1);
1492 discard_slab(s, page);
1497 #ifdef CONFIG_CMPXCHG_LOCAL
1498 #ifdef CONFIG_PREEMPT
1500 * Calculate the next globally unique transaction for disambiguiation
1501 * during cmpxchg. The transactions start with the cpu number and are then
1502 * incremented by CONFIG_NR_CPUS.
1504 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1507 * No preemption supported therefore also no need to check for
1513 static inline unsigned long next_tid(unsigned long tid)
1515 return tid + TID_STEP;
1518 static inline unsigned int tid_to_cpu(unsigned long tid)
1520 return tid % TID_STEP;
1523 static inline unsigned long tid_to_event(unsigned long tid)
1525 return tid / TID_STEP;
1528 static inline unsigned int init_tid(int cpu)
1533 static inline void note_cmpxchg_failure(const char *n,
1534 const struct kmem_cache *s, unsigned long tid)
1536 #ifdef SLUB_DEBUG_CMPXCHG
1537 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1539 printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name);
1541 #ifdef CONFIG_PREEMPT
1542 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
1543 printk("due to cpu change %d -> %d\n",
1544 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1547 if (tid_to_event(tid) != tid_to_event(actual_tid))
1548 printk("due to cpu running other code. Event %ld->%ld\n",
1549 tid_to_event(tid), tid_to_event(actual_tid));
1551 printk("for unknown reason: actual=%lx was=%lx target=%lx\n",
1552 actual_tid, tid, next_tid(tid));
1558 void init_kmem_cache_cpus(struct kmem_cache *s)
1560 #if defined(CONFIG_CMPXCHG_LOCAL) && defined(CONFIG_PREEMPT)
1563 for_each_possible_cpu(cpu)
1564 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
1569 * Remove the cpu slab
1571 static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1574 struct page *page = c->page;
1578 stat(s, DEACTIVATE_REMOTE_FREES);
1580 * Merge cpu freelist into slab freelist. Typically we get here
1581 * because both freelists are empty. So this is unlikely
1584 while (unlikely(c->freelist)) {
1587 tail = 0; /* Hot objects. Put the slab first */
1589 /* Retrieve object from cpu_freelist */
1590 object = c->freelist;
1591 c->freelist = get_freepointer(s, c->freelist);
1593 /* And put onto the regular freelist */
1594 set_freepointer(s, object, page->freelist);
1595 page->freelist = object;
1599 #ifdef CONFIG_CMPXCHG_LOCAL
1600 c->tid = next_tid(c->tid);
1602 unfreeze_slab(s, page, tail);
1605 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1607 stat(s, CPUSLAB_FLUSH);
1609 deactivate_slab(s, c);
1615 * Called from IPI handler with interrupts disabled.
1617 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
1619 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
1621 if (likely(c && c->page))
1625 static void flush_cpu_slab(void *d)
1627 struct kmem_cache *s = d;
1629 __flush_cpu_slab(s, smp_processor_id());
1632 static void flush_all(struct kmem_cache *s)
1634 on_each_cpu(flush_cpu_slab, s, 1);
1638 * Check if the objects in a per cpu structure fit numa
1639 * locality expectations.
1641 static inline int node_match(struct kmem_cache_cpu *c, int node)
1644 if (node != NUMA_NO_NODE && c->node != node)
1650 static int count_free(struct page *page)
1652 return page->objects - page->inuse;
1655 static unsigned long count_partial(struct kmem_cache_node *n,
1656 int (*get_count)(struct page *))
1658 unsigned long flags;
1659 unsigned long x = 0;
1662 spin_lock_irqsave(&n->list_lock, flags);
1663 list_for_each_entry(page, &n->partial, lru)
1664 x += get_count(page);
1665 spin_unlock_irqrestore(&n->list_lock, flags);
1669 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1671 #ifdef CONFIG_SLUB_DEBUG
1672 return atomic_long_read(&n->total_objects);
1678 static noinline void
1679 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1684 "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1686 printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1687 "default order: %d, min order: %d\n", s->name, s->objsize,
1688 s->size, oo_order(s->oo), oo_order(s->min));
1690 if (oo_order(s->min) > get_order(s->objsize))
1691 printk(KERN_WARNING " %s debugging increased min order, use "
1692 "slub_debug=O to disable.\n", s->name);
1694 for_each_online_node(node) {
1695 struct kmem_cache_node *n = get_node(s, node);
1696 unsigned long nr_slabs;
1697 unsigned long nr_objs;
1698 unsigned long nr_free;
1703 nr_free = count_partial(n, count_free);
1704 nr_slabs = node_nr_slabs(n);
1705 nr_objs = node_nr_objs(n);
1708 " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1709 node, nr_slabs, nr_objs, nr_free);
1714 * Slow path. The lockless freelist is empty or we need to perform
1717 * Interrupts are disabled.
1719 * Processing is still very fast if new objects have been freed to the
1720 * regular freelist. In that case we simply take over the regular freelist
1721 * as the lockless freelist and zap the regular freelist.
1723 * If that is not working then we fall back to the partial lists. We take the
1724 * first element of the freelist as the object to allocate now and move the
1725 * rest of the freelist to the lockless freelist.
1727 * And if we were unable to get a new slab from the partial slab lists then
1728 * we need to allocate a new slab. This is the slowest path since it involves
1729 * a call to the page allocator and the setup of a new slab.
1731 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1732 unsigned long addr, struct kmem_cache_cpu *c)
1736 #ifdef CONFIG_CMPXCHG_LOCAL
1737 unsigned long flags;
1739 local_irq_save(flags);
1740 #ifdef CONFIG_PREEMPT
1742 * We may have been preempted and rescheduled on a different
1743 * cpu before disabling interrupts. Need to reload cpu area
1746 c = this_cpu_ptr(s->cpu_slab);
1750 /* We handle __GFP_ZERO in the caller */
1751 gfpflags &= ~__GFP_ZERO;
1757 if (unlikely(!node_match(c, node)))
1760 stat(s, ALLOC_REFILL);
1763 object = c->page->freelist;
1764 if (unlikely(!object))
1766 if (kmem_cache_debug(s))
1769 c->freelist = get_freepointer(s, object);
1770 c->page->inuse = c->page->objects;
1771 c->page->freelist = NULL;
1772 c->node = page_to_nid(c->page);
1774 slab_unlock(c->page);
1775 #ifdef CONFIG_CMPXCHG_LOCAL
1776 c->tid = next_tid(c->tid);
1777 local_irq_restore(flags);
1779 stat(s, ALLOC_SLOWPATH);
1783 deactivate_slab(s, c);
1786 new = get_partial(s, gfpflags, node);
1789 stat(s, ALLOC_FROM_PARTIAL);
1793 gfpflags &= gfp_allowed_mask;
1794 if (gfpflags & __GFP_WAIT)
1797 new = new_slab(s, gfpflags, node);
1799 if (gfpflags & __GFP_WAIT)
1800 local_irq_disable();
1803 c = __this_cpu_ptr(s->cpu_slab);
1804 stat(s, ALLOC_SLAB);
1808 __SetPageSlubFrozen(new);
1812 if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1813 slab_out_of_memory(s, gfpflags, node);
1816 if (!alloc_debug_processing(s, c->page, object, addr))
1820 c->page->freelist = get_freepointer(s, object);
1821 c->node = NUMA_NO_NODE;
1826 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1827 * have the fastpath folded into their functions. So no function call
1828 * overhead for requests that can be satisfied on the fastpath.
1830 * The fastpath works by first checking if the lockless freelist can be used.
1831 * If not then __slab_alloc is called for slow processing.
1833 * Otherwise we can simply pick the next object from the lockless free list.
1835 static __always_inline void *slab_alloc(struct kmem_cache *s,
1836 gfp_t gfpflags, int node, unsigned long addr)
1839 struct kmem_cache_cpu *c;
1840 #ifdef CONFIG_CMPXCHG_LOCAL
1843 unsigned long flags;
1846 if (slab_pre_alloc_hook(s, gfpflags))
1849 #ifndef CONFIG_CMPXCHG_LOCAL
1850 local_irq_save(flags);
1856 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
1857 * enabled. We may switch back and forth between cpus while
1858 * reading from one cpu area. That does not matter as long
1859 * as we end up on the original cpu again when doing the cmpxchg.
1861 c = __this_cpu_ptr(s->cpu_slab);
1863 #ifdef CONFIG_CMPXCHG_LOCAL
1865 * The transaction ids are globally unique per cpu and per operation on
1866 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
1867 * occurs on the right processor and that there was no operation on the
1868 * linked list in between.
1874 object = c->freelist;
1875 if (unlikely(!object || !node_match(c, node)))
1877 object = __slab_alloc(s, gfpflags, node, addr, c);
1880 #ifdef CONFIG_CMPXCHG_LOCAL
1882 * The cmpxchg will only match if there was no additonal
1883 * operation and if we are on the right processor.
1885 * The cmpxchg does the following atomically (without lock semantics!)
1886 * 1. Relocate first pointer to the current per cpu area.
1887 * 2. Verify that tid and freelist have not been changed
1888 * 3. If they were not changed replace tid and freelist
1890 * Since this is without lock semantics the protection is only against
1891 * code executing on this cpu *not* from access by other cpus.
1893 if (unlikely(!this_cpu_cmpxchg_double(
1894 s->cpu_slab->freelist, s->cpu_slab->tid,
1896 get_freepointer(s, object), next_tid(tid)))) {
1898 note_cmpxchg_failure("slab_alloc", s, tid);
1902 c->freelist = get_freepointer(s, object);
1904 stat(s, ALLOC_FASTPATH);
1907 #ifndef CONFIG_CMPXCHG_LOCAL
1908 local_irq_restore(flags);
1911 if (unlikely(gfpflags & __GFP_ZERO) && object)
1912 memset(object, 0, s->objsize);
1914 slab_post_alloc_hook(s, gfpflags, object);
1919 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1921 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1923 trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
1927 EXPORT_SYMBOL(kmem_cache_alloc);
1929 #ifdef CONFIG_TRACING
1930 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1932 void *ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, _RET_IP_);
1933 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
1936 EXPORT_SYMBOL(kmem_cache_alloc_trace);
1938 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1940 void *ret = kmalloc_order(size, flags, order);
1941 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1944 EXPORT_SYMBOL(kmalloc_order_trace);
1948 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1950 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1952 trace_kmem_cache_alloc_node(_RET_IP_, ret,
1953 s->objsize, s->size, gfpflags, node);
1957 EXPORT_SYMBOL(kmem_cache_alloc_node);
1959 #ifdef CONFIG_TRACING
1960 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
1962 int node, size_t size)
1964 void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1966 trace_kmalloc_node(_RET_IP_, ret,
1967 size, s->size, gfpflags, node);
1970 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
1975 * Slow patch handling. This may still be called frequently since objects
1976 * have a longer lifetime than the cpu slabs in most processing loads.
1978 * So we still attempt to reduce cache line usage. Just take the slab
1979 * lock and free the item. If there is no additional partial page
1980 * handling required then we can return immediately.
1982 static void __slab_free(struct kmem_cache *s, struct page *page,
1983 void *x, unsigned long addr)
1986 void **object = (void *)x;
1987 #ifdef CONFIG_CMPXCHG_LOCAL
1988 unsigned long flags;
1990 local_irq_save(flags);
1993 stat(s, FREE_SLOWPATH);
1995 if (kmem_cache_debug(s))
1999 prior = page->freelist;
2000 set_freepointer(s, object, prior);
2001 page->freelist = object;
2004 if (unlikely(PageSlubFrozen(page))) {
2005 stat(s, FREE_FROZEN);
2009 if (unlikely(!page->inuse))
2013 * Objects left in the slab. If it was not on the partial list before
2016 if (unlikely(!prior)) {
2017 add_partial(get_node(s, page_to_nid(page)), page, 1);
2018 stat(s, FREE_ADD_PARTIAL);
2023 #ifdef CONFIG_CMPXCHG_LOCAL
2024 local_irq_restore(flags);
2031 * Slab still on the partial list.
2033 remove_partial(s, page);
2034 stat(s, FREE_REMOVE_PARTIAL);
2037 #ifdef CONFIG_CMPXCHG_LOCAL
2038 local_irq_restore(flags);
2041 discard_slab(s, page);
2045 if (!free_debug_processing(s, page, x, addr))
2051 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2052 * can perform fastpath freeing without additional function calls.
2054 * The fastpath is only possible if we are freeing to the current cpu slab
2055 * of this processor. This typically the case if we have just allocated
2058 * If fastpath is not possible then fall back to __slab_free where we deal
2059 * with all sorts of special processing.
2061 static __always_inline void slab_free(struct kmem_cache *s,
2062 struct page *page, void *x, unsigned long addr)
2064 void **object = (void *)x;
2065 struct kmem_cache_cpu *c;
2066 #ifdef CONFIG_CMPXCHG_LOCAL
2069 unsigned long flags;
2072 slab_free_hook(s, x);
2074 #ifndef CONFIG_CMPXCHG_LOCAL
2075 local_irq_save(flags);
2082 * Determine the currently cpus per cpu slab.
2083 * The cpu may change afterward. However that does not matter since
2084 * data is retrieved via this pointer. If we are on the same cpu
2085 * during the cmpxchg then the free will succedd.
2087 c = __this_cpu_ptr(s->cpu_slab);
2089 #ifdef CONFIG_CMPXCHG_LOCAL
2094 if (likely(page == c->page && c->node != NUMA_NO_NODE)) {
2095 set_freepointer(s, object, c->freelist);
2097 #ifdef CONFIG_CMPXCHG_LOCAL
2098 if (unlikely(!this_cpu_cmpxchg_double(
2099 s->cpu_slab->freelist, s->cpu_slab->tid,
2101 object, next_tid(tid)))) {
2103 note_cmpxchg_failure("slab_free", s, tid);
2107 c->freelist = object;
2109 stat(s, FREE_FASTPATH);
2111 __slab_free(s, page, x, addr);
2113 #ifndef CONFIG_CMPXCHG_LOCAL
2114 local_irq_restore(flags);
2118 void kmem_cache_free(struct kmem_cache *s, void *x)
2122 page = virt_to_head_page(x);
2124 slab_free(s, page, x, _RET_IP_);
2126 trace_kmem_cache_free(_RET_IP_, x);
2128 EXPORT_SYMBOL(kmem_cache_free);
2131 * Object placement in a slab is made very easy because we always start at
2132 * offset 0. If we tune the size of the object to the alignment then we can
2133 * get the required alignment by putting one properly sized object after
2136 * Notice that the allocation order determines the sizes of the per cpu
2137 * caches. Each processor has always one slab available for allocations.
2138 * Increasing the allocation order reduces the number of times that slabs
2139 * must be moved on and off the partial lists and is therefore a factor in
2144 * Mininum / Maximum order of slab pages. This influences locking overhead
2145 * and slab fragmentation. A higher order reduces the number of partial slabs
2146 * and increases the number of allocations possible without having to
2147 * take the list_lock.
2149 static int slub_min_order;
2150 static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
2151 static int slub_min_objects;
2154 * Merge control. If this is set then no merging of slab caches will occur.
2155 * (Could be removed. This was introduced to pacify the merge skeptics.)
2157 static int slub_nomerge;
2160 * Calculate the order of allocation given an slab object size.
2162 * The order of allocation has significant impact on performance and other
2163 * system components. Generally order 0 allocations should be preferred since
2164 * order 0 does not cause fragmentation in the page allocator. Larger objects
2165 * be problematic to put into order 0 slabs because there may be too much
2166 * unused space left. We go to a higher order if more than 1/16th of the slab
2169 * In order to reach satisfactory performance we must ensure that a minimum
2170 * number of objects is in one slab. Otherwise we may generate too much
2171 * activity on the partial lists which requires taking the list_lock. This is
2172 * less a concern for large slabs though which are rarely used.
2174 * slub_max_order specifies the order where we begin to stop considering the
2175 * number of objects in a slab as critical. If we reach slub_max_order then
2176 * we try to keep the page order as low as possible. So we accept more waste
2177 * of space in favor of a small page order.
2179 * Higher order allocations also allow the placement of more objects in a
2180 * slab and thereby reduce object handling overhead. If the user has
2181 * requested a higher mininum order then we start with that one instead of
2182 * the smallest order which will fit the object.
2184 static inline int slab_order(int size, int min_objects,
2185 int max_order, int fract_leftover)
2189 int min_order = slub_min_order;
2191 if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
2192 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
2194 for (order = max(min_order,
2195 fls(min_objects * size - 1) - PAGE_SHIFT);
2196 order <= max_order; order++) {
2198 unsigned long slab_size = PAGE_SIZE << order;
2200 if (slab_size < min_objects * size)
2203 rem = slab_size % size;
2205 if (rem <= slab_size / fract_leftover)
2213 static inline int calculate_order(int size)
2221 * Attempt to find best configuration for a slab. This
2222 * works by first attempting to generate a layout with
2223 * the best configuration and backing off gradually.
2225 * First we reduce the acceptable waste in a slab. Then
2226 * we reduce the minimum objects required in a slab.
2228 min_objects = slub_min_objects;
2230 min_objects = 4 * (fls(nr_cpu_ids) + 1);
2231 max_objects = (PAGE_SIZE << slub_max_order)/size;
2232 min_objects = min(min_objects, max_objects);
2234 while (min_objects > 1) {
2236 while (fraction >= 4) {
2237 order = slab_order(size, min_objects,
2238 slub_max_order, fraction);
2239 if (order <= slub_max_order)
2247 * We were unable to place multiple objects in a slab. Now
2248 * lets see if we can place a single object there.
2250 order = slab_order(size, 1, slub_max_order, 1);
2251 if (order <= slub_max_order)
2255 * Doh this slab cannot be placed using slub_max_order.
2257 order = slab_order(size, 1, MAX_ORDER, 1);
2258 if (order < MAX_ORDER)
2264 * Figure out what the alignment of the objects will be.
2266 static unsigned long calculate_alignment(unsigned long flags,
2267 unsigned long align, unsigned long size)
2270 * If the user wants hardware cache aligned objects then follow that
2271 * suggestion if the object is sufficiently large.
2273 * The hardware cache alignment cannot override the specified
2274 * alignment though. If that is greater then use it.
2276 if (flags & SLAB_HWCACHE_ALIGN) {
2277 unsigned long ralign = cache_line_size();
2278 while (size <= ralign / 2)
2280 align = max(align, ralign);
2283 if (align < ARCH_SLAB_MINALIGN)
2284 align = ARCH_SLAB_MINALIGN;
2286 return ALIGN(align, sizeof(void *));
2290 init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
2293 spin_lock_init(&n->list_lock);
2294 INIT_LIST_HEAD(&n->partial);
2295 #ifdef CONFIG_SLUB_DEBUG
2296 atomic_long_set(&n->nr_slabs, 0);
2297 atomic_long_set(&n->total_objects, 0);
2298 INIT_LIST_HEAD(&n->full);
2302 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
2304 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
2305 SLUB_PAGE_SHIFT * sizeof(struct kmem_cache_cpu));
2307 #ifdef CONFIG_CMPXCHG_LOCAL
2309 * Must align to double word boundary for the double cmpxchg instructions
2312 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), 2 * sizeof(void *));
2314 /* Regular alignment is sufficient */
2315 s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2321 init_kmem_cache_cpus(s);
2326 static struct kmem_cache *kmem_cache_node;
2329 * No kmalloc_node yet so do it by hand. We know that this is the first
2330 * slab on the node for this slabcache. There are no concurrent accesses
2333 * Note that this function only works on the kmalloc_node_cache
2334 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2335 * memory on a fresh node that has no slab structures yet.
2337 static void early_kmem_cache_node_alloc(int node)
2340 struct kmem_cache_node *n;
2341 unsigned long flags;
2343 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
2345 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
2348 if (page_to_nid(page) != node) {
2349 printk(KERN_ERR "SLUB: Unable to allocate memory from "
2351 printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2352 "in order to be able to continue\n");
2357 page->freelist = get_freepointer(kmem_cache_node, n);
2359 kmem_cache_node->node[node] = n;
2360 #ifdef CONFIG_SLUB_DEBUG
2361 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
2362 init_tracking(kmem_cache_node, n);
2364 init_kmem_cache_node(n, kmem_cache_node);
2365 inc_slabs_node(kmem_cache_node, node, page->objects);
2368 * lockdep requires consistent irq usage for each lock
2369 * so even though there cannot be a race this early in
2370 * the boot sequence, we still disable irqs.
2372 local_irq_save(flags);
2373 add_partial(n, page, 0);
2374 local_irq_restore(flags);
2377 static void free_kmem_cache_nodes(struct kmem_cache *s)
2381 for_each_node_state(node, N_NORMAL_MEMORY) {
2382 struct kmem_cache_node *n = s->node[node];
2385 kmem_cache_free(kmem_cache_node, n);
2387 s->node[node] = NULL;
2391 static int init_kmem_cache_nodes(struct kmem_cache *s)
2395 for_each_node_state(node, N_NORMAL_MEMORY) {
2396 struct kmem_cache_node *n;
2398 if (slab_state == DOWN) {
2399 early_kmem_cache_node_alloc(node);
2402 n = kmem_cache_alloc_node(kmem_cache_node,
2406 free_kmem_cache_nodes(s);
2411 init_kmem_cache_node(n, s);
2416 static void set_min_partial(struct kmem_cache *s, unsigned long min)
2418 if (min < MIN_PARTIAL)
2420 else if (min > MAX_PARTIAL)
2422 s->min_partial = min;
2426 * calculate_sizes() determines the order and the distribution of data within
2429 static int calculate_sizes(struct kmem_cache *s, int forced_order)
2431 unsigned long flags = s->flags;
2432 unsigned long size = s->objsize;
2433 unsigned long align = s->align;
2437 * Round up object size to the next word boundary. We can only
2438 * place the free pointer at word boundaries and this determines
2439 * the possible location of the free pointer.
2441 size = ALIGN(size, sizeof(void *));
2443 #ifdef CONFIG_SLUB_DEBUG
2445 * Determine if we can poison the object itself. If the user of
2446 * the slab may touch the object after free or before allocation
2447 * then we should never poison the object itself.
2449 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
2451 s->flags |= __OBJECT_POISON;
2453 s->flags &= ~__OBJECT_POISON;
2457 * If we are Redzoning then check if there is some space between the
2458 * end of the object and the free pointer. If not then add an
2459 * additional word to have some bytes to store Redzone information.
2461 if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2462 size += sizeof(void *);
2466 * With that we have determined the number of bytes in actual use
2467 * by the object. This is the potential offset to the free pointer.
2471 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
2474 * Relocate free pointer after the object if it is not
2475 * permitted to overwrite the first word of the object on
2478 * This is the case if we do RCU, have a constructor or
2479 * destructor or are poisoning the objects.
2482 size += sizeof(void *);
2485 #ifdef CONFIG_SLUB_DEBUG
2486 if (flags & SLAB_STORE_USER)
2488 * Need to store information about allocs and frees after
2491 size += 2 * sizeof(struct track);
2493 if (flags & SLAB_RED_ZONE)
2495 * Add some empty padding so that we can catch
2496 * overwrites from earlier objects rather than let
2497 * tracking information or the free pointer be
2498 * corrupted if a user writes before the start
2501 size += sizeof(void *);
2505 * Determine the alignment based on various parameters that the
2506 * user specified and the dynamic determination of cache line size
2509 align = calculate_alignment(flags, align, s->objsize);
2513 * SLUB stores one object immediately after another beginning from
2514 * offset 0. In order to align the objects we have to simply size
2515 * each object to conform to the alignment.
2517 size = ALIGN(size, align);
2519 if (forced_order >= 0)
2520 order = forced_order;
2522 order = calculate_order(size);
2529 s->allocflags |= __GFP_COMP;
2531 if (s->flags & SLAB_CACHE_DMA)
2532 s->allocflags |= SLUB_DMA;
2534 if (s->flags & SLAB_RECLAIM_ACCOUNT)
2535 s->allocflags |= __GFP_RECLAIMABLE;
2538 * Determine the number of objects per slab
2540 s->oo = oo_make(order, size);
2541 s->min = oo_make(get_order(size), size);
2542 if (oo_objects(s->oo) > oo_objects(s->max))
2545 return !!oo_objects(s->oo);
2549 static int kmem_cache_open(struct kmem_cache *s,
2550 const char *name, size_t size,
2551 size_t align, unsigned long flags,
2552 void (*ctor)(void *))
2554 memset(s, 0, kmem_size);
2559 s->flags = kmem_cache_flags(size, flags, name, ctor);
2561 if (!calculate_sizes(s, -1))
2563 if (disable_higher_order_debug) {
2565 * Disable debugging flags that store metadata if the min slab
2568 if (get_order(s->size) > get_order(s->objsize)) {
2569 s->flags &= ~DEBUG_METADATA_FLAGS;
2571 if (!calculate_sizes(s, -1))
2577 * The larger the object size is, the more pages we want on the partial
2578 * list to avoid pounding the page allocator excessively.
2580 set_min_partial(s, ilog2(s->size));
2583 s->remote_node_defrag_ratio = 1000;
2585 if (!init_kmem_cache_nodes(s))
2588 if (alloc_kmem_cache_cpus(s))
2591 free_kmem_cache_nodes(s);
2593 if (flags & SLAB_PANIC)
2594 panic("Cannot create slab %s size=%lu realsize=%u "
2595 "order=%u offset=%u flags=%lx\n",
2596 s->name, (unsigned long)size, s->size, oo_order(s->oo),
2602 * Determine the size of a slab object
2604 unsigned int kmem_cache_size(struct kmem_cache *s)
2608 EXPORT_SYMBOL(kmem_cache_size);
2610 const char *kmem_cache_name(struct kmem_cache *s)
2614 EXPORT_SYMBOL(kmem_cache_name);
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);
2628 for_each_free_object(p, s, page->freelist)
2629 set_bit(slab_index(p, s, addr), map);
2631 for_each_object(p, s, addr, page->objects) {
2633 if (!test_bit(slab_index(p, s, addr), map)) {
2634 printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2636 print_tracking(s, p);
2645 * Attempt to free all partial slabs on a node.
2647 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
2649 unsigned long flags;
2650 struct page *page, *h;
2652 spin_lock_irqsave(&n->list_lock, flags);
2653 list_for_each_entry_safe(page, h, &n->partial, lru) {
2655 __remove_partial(n, page);
2656 discard_slab(s, page);
2658 list_slab_objects(s, page,
2659 "Objects remaining on kmem_cache_close()");
2662 spin_unlock_irqrestore(&n->list_lock, flags);
2666 * Release all resources used by a slab cache.
2668 static inline int kmem_cache_close(struct kmem_cache *s)
2673 free_percpu(s->cpu_slab);
2674 /* Attempt to free all objects */
2675 for_each_node_state(node, N_NORMAL_MEMORY) {
2676 struct kmem_cache_node *n = get_node(s, node);
2679 if (n->nr_partial || slabs_node(s, node))
2682 free_kmem_cache_nodes(s);
2687 * Close a cache and release the kmem_cache structure
2688 * (must be used for caches created using kmem_cache_create)
2690 void kmem_cache_destroy(struct kmem_cache *s)
2692 down_write(&slub_lock);
2696 if (kmem_cache_close(s)) {
2697 printk(KERN_ERR "SLUB %s: %s called for cache that "
2698 "still has objects.\n", s->name, __func__);
2701 if (s->flags & SLAB_DESTROY_BY_RCU)
2703 sysfs_slab_remove(s);
2705 up_write(&slub_lock);
2707 EXPORT_SYMBOL(kmem_cache_destroy);
2709 /********************************************************************
2711 *******************************************************************/
2713 struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
2714 EXPORT_SYMBOL(kmalloc_caches);
2716 static struct kmem_cache *kmem_cache;
2718 #ifdef CONFIG_ZONE_DMA
2719 static struct kmem_cache *kmalloc_dma_caches[SLUB_PAGE_SHIFT];
2722 static int __init setup_slub_min_order(char *str)
2724 get_option(&str, &slub_min_order);
2729 __setup("slub_min_order=", setup_slub_min_order);
2731 static int __init setup_slub_max_order(char *str)
2733 get_option(&str, &slub_max_order);
2734 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
2739 __setup("slub_max_order=", setup_slub_max_order);
2741 static int __init setup_slub_min_objects(char *str)
2743 get_option(&str, &slub_min_objects);
2748 __setup("slub_min_objects=", setup_slub_min_objects);
2750 static int __init setup_slub_nomerge(char *str)
2756 __setup("slub_nomerge", setup_slub_nomerge);
2758 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
2759 int size, unsigned int flags)
2761 struct kmem_cache *s;
2763 s = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
2766 * This function is called with IRQs disabled during early-boot on
2767 * single CPU so there's no need to take slub_lock here.
2769 if (!kmem_cache_open(s, name, size, ARCH_KMALLOC_MINALIGN,
2773 list_add(&s->list, &slab_caches);
2777 panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2782 * Conversion table for small slabs sizes / 8 to the index in the
2783 * kmalloc array. This is necessary for slabs < 192 since we have non power
2784 * of two cache sizes there. The size of larger slabs can be determined using
2787 static s8 size_index[24] = {
2814 static inline int size_index_elem(size_t bytes)
2816 return (bytes - 1) / 8;
2819 static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2825 return ZERO_SIZE_PTR;
2827 index = size_index[size_index_elem(size)];
2829 index = fls(size - 1);
2831 #ifdef CONFIG_ZONE_DMA
2832 if (unlikely((flags & SLUB_DMA)))
2833 return kmalloc_dma_caches[index];
2836 return kmalloc_caches[index];
2839 void *__kmalloc(size_t size, gfp_t flags)
2841 struct kmem_cache *s;
2844 if (unlikely(size > SLUB_MAX_SIZE))
2845 return kmalloc_large(size, flags);
2847 s = get_slab(size, flags);
2849 if (unlikely(ZERO_OR_NULL_PTR(s)))
2852 ret = slab_alloc(s, flags, NUMA_NO_NODE, _RET_IP_);
2854 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
2858 EXPORT_SYMBOL(__kmalloc);
2861 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2866 flags |= __GFP_COMP | __GFP_NOTRACK;
2867 page = alloc_pages_node(node, flags, get_order(size));
2869 ptr = page_address(page);
2871 kmemleak_alloc(ptr, size, 1, flags);
2875 void *__kmalloc_node(size_t size, gfp_t flags, int node)
2877 struct kmem_cache *s;
2880 if (unlikely(size > SLUB_MAX_SIZE)) {
2881 ret = kmalloc_large_node(size, flags, node);
2883 trace_kmalloc_node(_RET_IP_, ret,
2884 size, PAGE_SIZE << get_order(size),
2890 s = get_slab(size, flags);
2892 if (unlikely(ZERO_OR_NULL_PTR(s)))
2895 ret = slab_alloc(s, flags, node, _RET_IP_);
2897 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
2901 EXPORT_SYMBOL(__kmalloc_node);
2904 size_t ksize(const void *object)
2907 struct kmem_cache *s;
2909 if (unlikely(object == ZERO_SIZE_PTR))
2912 page = virt_to_head_page(object);
2914 if (unlikely(!PageSlab(page))) {
2915 WARN_ON(!PageCompound(page));
2916 return PAGE_SIZE << compound_order(page);
2920 #ifdef CONFIG_SLUB_DEBUG
2922 * Debugging requires use of the padding between object
2923 * and whatever may come after it.
2925 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2930 * If we have the need to store the freelist pointer
2931 * back there or track user information then we can
2932 * only use the space before that information.
2934 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2937 * Else we can use all the padding etc for the allocation
2941 EXPORT_SYMBOL(ksize);
2943 void kfree(const void *x)
2946 void *object = (void *)x;
2948 trace_kfree(_RET_IP_, x);
2950 if (unlikely(ZERO_OR_NULL_PTR(x)))
2953 page = virt_to_head_page(x);
2954 if (unlikely(!PageSlab(page))) {
2955 BUG_ON(!PageCompound(page));
2960 slab_free(page->slab, page, object, _RET_IP_);
2962 EXPORT_SYMBOL(kfree);
2965 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2966 * the remaining slabs by the number of items in use. The slabs with the
2967 * most items in use come first. New allocations will then fill those up
2968 * and thus they can be removed from the partial lists.
2970 * The slabs with the least items are placed last. This results in them
2971 * being allocated from last increasing the chance that the last objects
2972 * are freed in them.
2974 int kmem_cache_shrink(struct kmem_cache *s)
2978 struct kmem_cache_node *n;
2981 int objects = oo_objects(s->max);
2982 struct list_head *slabs_by_inuse =
2983 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
2984 unsigned long flags;
2986 if (!slabs_by_inuse)
2990 for_each_node_state(node, N_NORMAL_MEMORY) {
2991 n = get_node(s, node);
2996 for (i = 0; i < objects; i++)
2997 INIT_LIST_HEAD(slabs_by_inuse + i);
2999 spin_lock_irqsave(&n->list_lock, flags);
3002 * Build lists indexed by the items in use in each slab.
3004 * Note that concurrent frees may occur while we hold the
3005 * list_lock. page->inuse here is the upper limit.
3007 list_for_each_entry_safe(page, t, &n->partial, lru) {
3008 if (!page->inuse && slab_trylock(page)) {
3010 * Must hold slab lock here because slab_free
3011 * may have freed the last object and be
3012 * waiting to release the slab.
3014 __remove_partial(n, page);
3016 discard_slab(s, page);
3018 list_move(&page->lru,
3019 slabs_by_inuse + page->inuse);
3024 * Rebuild the partial list with the slabs filled up most
3025 * first and the least used slabs at the end.
3027 for (i = objects - 1; i >= 0; i--)
3028 list_splice(slabs_by_inuse + i, n->partial.prev);
3030 spin_unlock_irqrestore(&n->list_lock, flags);
3033 kfree(slabs_by_inuse);
3036 EXPORT_SYMBOL(kmem_cache_shrink);
3038 #if defined(CONFIG_MEMORY_HOTPLUG)
3039 static int slab_mem_going_offline_callback(void *arg)
3041 struct kmem_cache *s;
3043 down_read(&slub_lock);
3044 list_for_each_entry(s, &slab_caches, list)
3045 kmem_cache_shrink(s);
3046 up_read(&slub_lock);
3051 static void slab_mem_offline_callback(void *arg)
3053 struct kmem_cache_node *n;
3054 struct kmem_cache *s;
3055 struct memory_notify *marg = arg;
3058 offline_node = marg->status_change_nid;
3061 * If the node still has available memory. we need kmem_cache_node
3064 if (offline_node < 0)
3067 down_read(&slub_lock);
3068 list_for_each_entry(s, &slab_caches, list) {
3069 n = get_node(s, offline_node);
3072 * if n->nr_slabs > 0, slabs still exist on the node
3073 * that is going down. We were unable to free them,
3074 * and offline_pages() function shouldn't call this
3075 * callback. So, we must fail.
3077 BUG_ON(slabs_node(s, offline_node));
3079 s->node[offline_node] = NULL;
3080 kmem_cache_free(kmem_cache_node, n);
3083 up_read(&slub_lock);
3086 static int slab_mem_going_online_callback(void *arg)
3088 struct kmem_cache_node *n;
3089 struct kmem_cache *s;
3090 struct memory_notify *marg = arg;
3091 int nid = marg->status_change_nid;
3095 * If the node's memory is already available, then kmem_cache_node is
3096 * already created. Nothing to do.
3102 * We are bringing a node online. No memory is available yet. We must
3103 * allocate a kmem_cache_node structure in order to bring the node
3106 down_read(&slub_lock);
3107 list_for_each_entry(s, &slab_caches, list) {
3109 * XXX: kmem_cache_alloc_node will fallback to other nodes
3110 * since memory is not yet available from the node that
3113 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
3118 init_kmem_cache_node(n, s);
3122 up_read(&slub_lock);
3126 static int slab_memory_callback(struct notifier_block *self,
3127 unsigned long action, void *arg)
3132 case MEM_GOING_ONLINE:
3133 ret = slab_mem_going_online_callback(arg);
3135 case MEM_GOING_OFFLINE:
3136 ret = slab_mem_going_offline_callback(arg);
3139 case MEM_CANCEL_ONLINE:
3140 slab_mem_offline_callback(arg);
3143 case MEM_CANCEL_OFFLINE:
3147 ret = notifier_from_errno(ret);
3153 #endif /* CONFIG_MEMORY_HOTPLUG */
3155 /********************************************************************
3156 * Basic setup of slabs
3157 *******************************************************************/
3160 * Used for early kmem_cache structures that were allocated using
3161 * the page allocator
3164 static void __init kmem_cache_bootstrap_fixup(struct kmem_cache *s)
3168 list_add(&s->list, &slab_caches);
3171 for_each_node_state(node, N_NORMAL_MEMORY) {
3172 struct kmem_cache_node *n = get_node(s, node);
3176 list_for_each_entry(p, &n->partial, lru)
3179 #ifdef CONFIG_SLAB_DEBUG
3180 list_for_each_entry(p, &n->full, lru)
3187 void __init kmem_cache_init(void)
3191 struct kmem_cache *temp_kmem_cache;
3193 struct kmem_cache *temp_kmem_cache_node;
3194 unsigned long kmalloc_size;
3196 kmem_size = offsetof(struct kmem_cache, node) +
3197 nr_node_ids * sizeof(struct kmem_cache_node *);
3199 /* Allocate two kmem_caches from the page allocator */
3200 kmalloc_size = ALIGN(kmem_size, cache_line_size());
3201 order = get_order(2 * kmalloc_size);
3202 kmem_cache = (void *)__get_free_pages(GFP_NOWAIT, order);
3205 * Must first have the slab cache available for the allocations of the
3206 * struct kmem_cache_node's. There is special bootstrap code in
3207 * kmem_cache_open for slab_state == DOWN.
3209 kmem_cache_node = (void *)kmem_cache + kmalloc_size;
3211 kmem_cache_open(kmem_cache_node, "kmem_cache_node",
3212 sizeof(struct kmem_cache_node),
3213 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3215 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
3217 /* Able to allocate the per node structures */
3218 slab_state = PARTIAL;
3220 temp_kmem_cache = kmem_cache;
3221 kmem_cache_open(kmem_cache, "kmem_cache", kmem_size,
3222 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3223 kmem_cache = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3224 memcpy(kmem_cache, temp_kmem_cache, kmem_size);
3227 * Allocate kmem_cache_node properly from the kmem_cache slab.
3228 * kmem_cache_node is separately allocated so no need to
3229 * update any list pointers.
3231 temp_kmem_cache_node = kmem_cache_node;
3233 kmem_cache_node = kmem_cache_alloc(kmem_cache, GFP_NOWAIT);
3234 memcpy(kmem_cache_node, temp_kmem_cache_node, kmem_size);
3236 kmem_cache_bootstrap_fixup(kmem_cache_node);
3239 kmem_cache_bootstrap_fixup(kmem_cache);
3241 /* Free temporary boot structure */
3242 free_pages((unsigned long)temp_kmem_cache, order);
3244 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3247 * Patch up the size_index table if we have strange large alignment
3248 * requirements for the kmalloc array. This is only the case for
3249 * MIPS it seems. The standard arches will not generate any code here.
3251 * Largest permitted alignment is 256 bytes due to the way we
3252 * handle the index determination for the smaller caches.
3254 * Make sure that nothing crazy happens if someone starts tinkering
3255 * around with ARCH_KMALLOC_MINALIGN
3257 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3258 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3260 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3261 int elem = size_index_elem(i);
3262 if (elem >= ARRAY_SIZE(size_index))
3264 size_index[elem] = KMALLOC_SHIFT_LOW;
3267 if (KMALLOC_MIN_SIZE == 64) {
3269 * The 96 byte size cache is not used if the alignment
3272 for (i = 64 + 8; i <= 96; i += 8)
3273 size_index[size_index_elem(i)] = 7;
3274 } else if (KMALLOC_MIN_SIZE == 128) {
3276 * The 192 byte sized cache is not used if the alignment
3277 * is 128 byte. Redirect kmalloc to use the 256 byte cache
3280 for (i = 128 + 8; i <= 192; i += 8)
3281 size_index[size_index_elem(i)] = 8;
3284 /* Caches that are not of the two-to-the-power-of size */
3285 if (KMALLOC_MIN_SIZE <= 32) {
3286 kmalloc_caches[1] = create_kmalloc_cache("kmalloc-96", 96, 0);
3290 if (KMALLOC_MIN_SIZE <= 64) {
3291 kmalloc_caches[2] = create_kmalloc_cache("kmalloc-192", 192, 0);
3295 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3296 kmalloc_caches[i] = create_kmalloc_cache("kmalloc", 1 << i, 0);
3302 /* Provide the correct kmalloc names now that the caches are up */
3303 if (KMALLOC_MIN_SIZE <= 32) {
3304 kmalloc_caches[1]->name = kstrdup(kmalloc_caches[1]->name, GFP_NOWAIT);
3305 BUG_ON(!kmalloc_caches[1]->name);
3308 if (KMALLOC_MIN_SIZE <= 64) {
3309 kmalloc_caches[2]->name = kstrdup(kmalloc_caches[2]->name, GFP_NOWAIT);
3310 BUG_ON(!kmalloc_caches[2]->name);
3313 for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3314 char *s = kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3317 kmalloc_caches[i]->name = s;
3321 register_cpu_notifier(&slab_notifier);
3324 #ifdef CONFIG_ZONE_DMA
3325 for (i = 0; i < SLUB_PAGE_SHIFT; i++) {
3326 struct kmem_cache *s = kmalloc_caches[i];
3329 char *name = kasprintf(GFP_NOWAIT,
3330 "dma-kmalloc-%d", s->objsize);
3333 kmalloc_dma_caches[i] = create_kmalloc_cache(name,
3334 s->objsize, SLAB_CACHE_DMA);
3339 "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3340 " CPUs=%d, Nodes=%d\n",
3341 caches, cache_line_size(),
3342 slub_min_order, slub_max_order, slub_min_objects,
3343 nr_cpu_ids, nr_node_ids);
3346 void __init kmem_cache_init_late(void)
3351 * Find a mergeable slab cache
3353 static int slab_unmergeable(struct kmem_cache *s)
3355 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3362 * We may have set a slab to be unmergeable during bootstrap.
3364 if (s->refcount < 0)
3370 static struct kmem_cache *find_mergeable(size_t size,
3371 size_t align, unsigned long flags, const char *name,
3372 void (*ctor)(void *))
3374 struct kmem_cache *s;
3376 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3382 size = ALIGN(size, sizeof(void *));
3383 align = calculate_alignment(flags, align, size);
3384 size = ALIGN(size, align);
3385 flags = kmem_cache_flags(size, flags, name, NULL);
3387 list_for_each_entry(s, &slab_caches, list) {
3388 if (slab_unmergeable(s))
3394 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
3397 * Check if alignment is compatible.
3398 * Courtesy of Adrian Drzewiecki
3400 if ((s->size & ~(align - 1)) != s->size)
3403 if (s->size - size >= sizeof(void *))
3411 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
3412 size_t align, unsigned long flags, void (*ctor)(void *))
3414 struct kmem_cache *s;
3420 down_write(&slub_lock);
3421 s = find_mergeable(size, align, flags, name, ctor);
3425 * Adjust the object sizes so that we clear
3426 * the complete object on kzalloc.
3428 s->objsize = max(s->objsize, (int)size);
3429 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
3431 if (sysfs_slab_alias(s, name)) {
3435 up_write(&slub_lock);
3439 n = kstrdup(name, GFP_KERNEL);
3443 s = kmalloc(kmem_size, GFP_KERNEL);
3445 if (kmem_cache_open(s, n,
3446 size, align, flags, ctor)) {
3447 list_add(&s->list, &slab_caches);
3448 if (sysfs_slab_add(s)) {
3454 up_write(&slub_lock);
3461 up_write(&slub_lock);
3463 if (flags & SLAB_PANIC)
3464 panic("Cannot create slabcache %s\n", name);
3469 EXPORT_SYMBOL(kmem_cache_create);
3473 * Use the cpu notifier to insure that the cpu slabs are flushed when
3476 static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3477 unsigned long action, void *hcpu)
3479 long cpu = (long)hcpu;
3480 struct kmem_cache *s;
3481 unsigned long flags;
3484 case CPU_UP_CANCELED:
3485 case CPU_UP_CANCELED_FROZEN:
3487 case CPU_DEAD_FROZEN:
3488 down_read(&slub_lock);
3489 list_for_each_entry(s, &slab_caches, list) {
3490 local_irq_save(flags);
3491 __flush_cpu_slab(s, cpu);
3492 local_irq_restore(flags);
3494 up_read(&slub_lock);
3502 static struct notifier_block __cpuinitdata slab_notifier = {
3503 .notifier_call = slab_cpuup_callback
3508 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
3510 struct kmem_cache *s;
3513 if (unlikely(size > SLUB_MAX_SIZE))
3514 return kmalloc_large(size, gfpflags);
3516 s = get_slab(size, gfpflags);
3518 if (unlikely(ZERO_OR_NULL_PTR(s)))
3521 ret = slab_alloc(s, gfpflags, NUMA_NO_NODE, caller);
3523 /* Honor the call site pointer we recieved. */
3524 trace_kmalloc(caller, ret, size, s->size, gfpflags);
3530 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
3531 int node, unsigned long caller)
3533 struct kmem_cache *s;
3536 if (unlikely(size > SLUB_MAX_SIZE)) {
3537 ret = kmalloc_large_node(size, gfpflags, node);
3539 trace_kmalloc_node(caller, ret,
3540 size, PAGE_SIZE << get_order(size),
3546 s = get_slab(size, gfpflags);
3548 if (unlikely(ZERO_OR_NULL_PTR(s)))
3551 ret = slab_alloc(s, gfpflags, node, caller);
3553 /* Honor the call site pointer we recieved. */
3554 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
3561 static int count_inuse(struct page *page)
3566 static int count_total(struct page *page)
3568 return page->objects;
3572 #ifdef CONFIG_SLUB_DEBUG
3573 static int validate_slab(struct kmem_cache *s, struct page *page,
3577 void *addr = page_address(page);
3579 if (!check_slab(s, page) ||
3580 !on_freelist(s, page, NULL))
3583 /* Now we know that a valid freelist exists */
3584 bitmap_zero(map, page->objects);
3586 for_each_free_object(p, s, page->freelist) {
3587 set_bit(slab_index(p, s, addr), map);
3588 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3592 for_each_object(p, s, addr, page->objects)
3593 if (!test_bit(slab_index(p, s, addr), map))
3594 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
3599 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3602 if (slab_trylock(page)) {
3603 validate_slab(s, page, map);
3606 printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3610 static int validate_slab_node(struct kmem_cache *s,
3611 struct kmem_cache_node *n, unsigned long *map)
3613 unsigned long count = 0;
3615 unsigned long flags;
3617 spin_lock_irqsave(&n->list_lock, flags);
3619 list_for_each_entry(page, &n->partial, lru) {
3620 validate_slab_slab(s, page, map);
3623 if (count != n->nr_partial)
3624 printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3625 "counter=%ld\n", s->name, count, n->nr_partial);
3627 if (!(s->flags & SLAB_STORE_USER))
3630 list_for_each_entry(page, &n->full, lru) {
3631 validate_slab_slab(s, page, map);
3634 if (count != atomic_long_read(&n->nr_slabs))
3635 printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3636 "counter=%ld\n", s->name, count,
3637 atomic_long_read(&n->nr_slabs));
3640 spin_unlock_irqrestore(&n->list_lock, flags);
3644 static long validate_slab_cache(struct kmem_cache *s)
3647 unsigned long count = 0;
3648 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3649 sizeof(unsigned long), GFP_KERNEL);
3655 for_each_node_state(node, N_NORMAL_MEMORY) {
3656 struct kmem_cache_node *n = get_node(s, node);
3658 count += validate_slab_node(s, n, map);
3664 * Generate lists of code addresses where slabcache objects are allocated
3669 unsigned long count;
3676 DECLARE_BITMAP(cpus, NR_CPUS);
3682 unsigned long count;
3683 struct location *loc;
3686 static void free_loc_track(struct loc_track *t)
3689 free_pages((unsigned long)t->loc,
3690 get_order(sizeof(struct location) * t->max));
3693 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
3698 order = get_order(sizeof(struct location) * max);
3700 l = (void *)__get_free_pages(flags, order);
3705 memcpy(l, t->loc, sizeof(struct location) * t->count);
3713 static int add_location(struct loc_track *t, struct kmem_cache *s,
3714 const struct track *track)
3716 long start, end, pos;
3718 unsigned long caddr;
3719 unsigned long age = jiffies - track->when;
3725 pos = start + (end - start + 1) / 2;
3728 * There is nothing at "end". If we end up there
3729 * we need to add something to before end.
3734 caddr = t->loc[pos].addr;
3735 if (track->addr == caddr) {
3741 if (age < l->min_time)
3743 if (age > l->max_time)
3746 if (track->pid < l->min_pid)
3747 l->min_pid = track->pid;
3748 if (track->pid > l->max_pid)
3749 l->max_pid = track->pid;
3751 cpumask_set_cpu(track->cpu,
3752 to_cpumask(l->cpus));
3754 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3758 if (track->addr < caddr)
3765 * Not found. Insert new tracking element.
3767 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
3773 (t->count - pos) * sizeof(struct location));
3776 l->addr = track->addr;
3780 l->min_pid = track->pid;
3781 l->max_pid = track->pid;
3782 cpumask_clear(to_cpumask(l->cpus));
3783 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
3784 nodes_clear(l->nodes);
3785 node_set(page_to_nid(virt_to_page(track)), l->nodes);
3789 static void process_slab(struct loc_track *t, struct kmem_cache *s,
3790 struct page *page, enum track_item alloc,
3793 void *addr = page_address(page);
3796 bitmap_zero(map, page->objects);
3797 for_each_free_object(p, s, page->freelist)
3798 set_bit(slab_index(p, s, addr), map);
3800 for_each_object(p, s, addr, page->objects)
3801 if (!test_bit(slab_index(p, s, addr), map))
3802 add_location(t, s, get_track(s, p, alloc));
3805 static int list_locations(struct kmem_cache *s, char *buf,
3806 enum track_item alloc)
3810 struct loc_track t = { 0, 0, NULL };
3812 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3813 sizeof(unsigned long), GFP_KERNEL);
3815 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3818 return sprintf(buf, "Out of memory\n");
3820 /* Push back cpu slabs */
3823 for_each_node_state(node, N_NORMAL_MEMORY) {
3824 struct kmem_cache_node *n = get_node(s, node);
3825 unsigned long flags;
3828 if (!atomic_long_read(&n->nr_slabs))
3831 spin_lock_irqsave(&n->list_lock, flags);
3832 list_for_each_entry(page, &n->partial, lru)
3833 process_slab(&t, s, page, alloc, map);
3834 list_for_each_entry(page, &n->full, lru)
3835 process_slab(&t, s, page, alloc, map);
3836 spin_unlock_irqrestore(&n->list_lock, flags);
3839 for (i = 0; i < t.count; i++) {
3840 struct location *l = &t.loc[i];
3842 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
3844 len += sprintf(buf + len, "%7ld ", l->count);
3847 len += sprintf(buf + len, "%pS", (void *)l->addr);
3849 len += sprintf(buf + len, "<not-available>");
3851 if (l->sum_time != l->min_time) {
3852 len += sprintf(buf + len, " age=%ld/%ld/%ld",
3854 (long)div_u64(l->sum_time, l->count),
3857 len += sprintf(buf + len, " age=%ld",
3860 if (l->min_pid != l->max_pid)
3861 len += sprintf(buf + len, " pid=%ld-%ld",
3862 l->min_pid, l->max_pid);
3864 len += sprintf(buf + len, " pid=%ld",
3867 if (num_online_cpus() > 1 &&
3868 !cpumask_empty(to_cpumask(l->cpus)) &&
3869 len < PAGE_SIZE - 60) {
3870 len += sprintf(buf + len, " cpus=");
3871 len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3872 to_cpumask(l->cpus));
3875 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
3876 len < PAGE_SIZE - 60) {
3877 len += sprintf(buf + len, " nodes=");
3878 len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3882 len += sprintf(buf + len, "\n");
3888 len += sprintf(buf, "No data\n");
3893 #ifdef SLUB_RESILIENCY_TEST
3894 static void resiliency_test(void)
3898 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || SLUB_PAGE_SHIFT < 10);
3900 printk(KERN_ERR "SLUB resiliency testing\n");
3901 printk(KERN_ERR "-----------------------\n");
3902 printk(KERN_ERR "A. Corruption after allocation\n");
3904 p = kzalloc(16, GFP_KERNEL);
3906 printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3907 " 0x12->0x%p\n\n", p + 16);
3909 validate_slab_cache(kmalloc_caches[4]);
3911 /* Hmmm... The next two are dangerous */
3912 p = kzalloc(32, GFP_KERNEL);
3913 p[32 + sizeof(void *)] = 0x34;
3914 printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3915 " 0x34 -> -0x%p\n", p);
3917 "If allocated object is overwritten then not detectable\n\n");
3919 validate_slab_cache(kmalloc_caches[5]);
3920 p = kzalloc(64, GFP_KERNEL);
3921 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3923 printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3926 "If allocated object is overwritten then not detectable\n\n");
3927 validate_slab_cache(kmalloc_caches[6]);
3929 printk(KERN_ERR "\nB. Corruption after free\n");
3930 p = kzalloc(128, GFP_KERNEL);
3933 printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3934 validate_slab_cache(kmalloc_caches[7]);
3936 p = kzalloc(256, GFP_KERNEL);
3939 printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3941 validate_slab_cache(kmalloc_caches[8]);
3943 p = kzalloc(512, GFP_KERNEL);
3946 printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3947 validate_slab_cache(kmalloc_caches[9]);
3951 static void resiliency_test(void) {};
3956 enum slab_stat_type {
3957 SL_ALL, /* All slabs */
3958 SL_PARTIAL, /* Only partially allocated slabs */
3959 SL_CPU, /* Only slabs used for cpu caches */
3960 SL_OBJECTS, /* Determine allocated objects not slabs */
3961 SL_TOTAL /* Determine object capacity not slabs */
3964 #define SO_ALL (1 << SL_ALL)
3965 #define SO_PARTIAL (1 << SL_PARTIAL)
3966 #define SO_CPU (1 << SL_CPU)
3967 #define SO_OBJECTS (1 << SL_OBJECTS)
3968 #define SO_TOTAL (1 << SL_TOTAL)
3970 static ssize_t show_slab_objects(struct kmem_cache *s,
3971 char *buf, unsigned long flags)
3973 unsigned long total = 0;
3976 unsigned long *nodes;
3977 unsigned long *per_cpu;
3979 nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
3982 per_cpu = nodes + nr_node_ids;
3984 if (flags & SO_CPU) {
3987 for_each_possible_cpu(cpu) {
3988 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3990 if (!c || c->node < 0)
3994 if (flags & SO_TOTAL)
3995 x = c->page->objects;
3996 else if (flags & SO_OBJECTS)
4002 nodes[c->node] += x;
4008 lock_memory_hotplug();
4009 #ifdef CONFIG_SLUB_DEBUG
4010 if (flags & SO_ALL) {
4011 for_each_node_state(node, N_NORMAL_MEMORY) {
4012 struct kmem_cache_node *n = get_node(s, node);
4014 if (flags & SO_TOTAL)
4015 x = atomic_long_read(&n->total_objects);
4016 else if (flags & SO_OBJECTS)
4017 x = atomic_long_read(&n->total_objects) -
4018 count_partial(n, count_free);
4021 x = atomic_long_read(&n->nr_slabs);
4028 if (flags & SO_PARTIAL) {
4029 for_each_node_state(node, N_NORMAL_MEMORY) {
4030 struct kmem_cache_node *n = get_node(s, node);
4032 if (flags & SO_TOTAL)
4033 x = count_partial(n, count_total);
4034 else if (flags & SO_OBJECTS)
4035 x = count_partial(n, count_inuse);
4042 x = sprintf(buf, "%lu", total);
4044 for_each_node_state(node, N_NORMAL_MEMORY)
4046 x += sprintf(buf + x, " N%d=%lu",
4049 unlock_memory_hotplug();
4051 return x + sprintf(buf + x, "\n");
4054 #ifdef CONFIG_SLUB_DEBUG
4055 static int any_slab_objects(struct kmem_cache *s)
4059 for_each_online_node(node) {
4060 struct kmem_cache_node *n = get_node(s, node);
4065 if (atomic_long_read(&n->total_objects))
4072 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4073 #define to_slab(n) container_of(n, struct kmem_cache, kobj);
4075 struct slab_attribute {
4076 struct attribute attr;
4077 ssize_t (*show)(struct kmem_cache *s, char *buf);
4078 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4081 #define SLAB_ATTR_RO(_name) \
4082 static struct slab_attribute _name##_attr = __ATTR_RO(_name)
4084 #define SLAB_ATTR(_name) \
4085 static struct slab_attribute _name##_attr = \
4086 __ATTR(_name, 0644, _name##_show, _name##_store)
4088 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4090 return sprintf(buf, "%d\n", s->size);
4092 SLAB_ATTR_RO(slab_size);
4094 static ssize_t align_show(struct kmem_cache *s, char *buf)
4096 return sprintf(buf, "%d\n", s->align);
4098 SLAB_ATTR_RO(align);
4100 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4102 return sprintf(buf, "%d\n", s->objsize);
4104 SLAB_ATTR_RO(object_size);
4106 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4108 return sprintf(buf, "%d\n", oo_objects(s->oo));
4110 SLAB_ATTR_RO(objs_per_slab);
4112 static ssize_t order_store(struct kmem_cache *s,
4113 const char *buf, size_t length)
4115 unsigned long order;
4118 err = strict_strtoul(buf, 10, &order);
4122 if (order > slub_max_order || order < slub_min_order)
4125 calculate_sizes(s, order);
4129 static ssize_t order_show(struct kmem_cache *s, char *buf)
4131 return sprintf(buf, "%d\n", oo_order(s->oo));
4135 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4137 return sprintf(buf, "%lu\n", s->min_partial);
4140 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4146 err = strict_strtoul(buf, 10, &min);
4150 set_min_partial(s, min);
4153 SLAB_ATTR(min_partial);
4155 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4159 return sprintf(buf, "%pS\n", s->ctor);
4163 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4165 return sprintf(buf, "%d\n", s->refcount - 1);
4167 SLAB_ATTR_RO(aliases);
4169 static ssize_t partial_show(struct kmem_cache *s, char *buf)
4171 return show_slab_objects(s, buf, SO_PARTIAL);
4173 SLAB_ATTR_RO(partial);
4175 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4177 return show_slab_objects(s, buf, SO_CPU);
4179 SLAB_ATTR_RO(cpu_slabs);
4181 static ssize_t objects_show(struct kmem_cache *s, char *buf)
4183 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
4185 SLAB_ATTR_RO(objects);
4187 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4189 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4191 SLAB_ATTR_RO(objects_partial);
4193 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4195 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4198 static ssize_t reclaim_account_store(struct kmem_cache *s,
4199 const char *buf, size_t length)
4201 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4203 s->flags |= SLAB_RECLAIM_ACCOUNT;
4206 SLAB_ATTR(reclaim_account);
4208 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4210 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
4212 SLAB_ATTR_RO(hwcache_align);
4214 #ifdef CONFIG_ZONE_DMA
4215 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4217 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4219 SLAB_ATTR_RO(cache_dma);
4222 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4224 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4226 SLAB_ATTR_RO(destroy_by_rcu);
4228 #ifdef CONFIG_SLUB_DEBUG
4229 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4231 return show_slab_objects(s, buf, SO_ALL);
4233 SLAB_ATTR_RO(slabs);
4235 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4237 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4239 SLAB_ATTR_RO(total_objects);
4241 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4243 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4246 static ssize_t sanity_checks_store(struct kmem_cache *s,
4247 const char *buf, size_t length)
4249 s->flags &= ~SLAB_DEBUG_FREE;
4251 s->flags |= SLAB_DEBUG_FREE;
4254 SLAB_ATTR(sanity_checks);
4256 static ssize_t trace_show(struct kmem_cache *s, char *buf)
4258 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4261 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4264 s->flags &= ~SLAB_TRACE;
4266 s->flags |= SLAB_TRACE;
4271 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4273 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4276 static ssize_t red_zone_store(struct kmem_cache *s,
4277 const char *buf, size_t length)
4279 if (any_slab_objects(s))
4282 s->flags &= ~SLAB_RED_ZONE;
4284 s->flags |= SLAB_RED_ZONE;
4285 calculate_sizes(s, -1);
4288 SLAB_ATTR(red_zone);
4290 static ssize_t poison_show(struct kmem_cache *s, char *buf)
4292 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4295 static ssize_t poison_store(struct kmem_cache *s,
4296 const char *buf, size_t length)
4298 if (any_slab_objects(s))
4301 s->flags &= ~SLAB_POISON;
4303 s->flags |= SLAB_POISON;
4304 calculate_sizes(s, -1);
4309 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4311 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4314 static ssize_t store_user_store(struct kmem_cache *s,
4315 const char *buf, size_t length)
4317 if (any_slab_objects(s))
4320 s->flags &= ~SLAB_STORE_USER;
4322 s->flags |= SLAB_STORE_USER;
4323 calculate_sizes(s, -1);
4326 SLAB_ATTR(store_user);
4328 static ssize_t validate_show(struct kmem_cache *s, char *buf)
4333 static ssize_t validate_store(struct kmem_cache *s,
4334 const char *buf, size_t length)
4338 if (buf[0] == '1') {
4339 ret = validate_slab_cache(s);
4345 SLAB_ATTR(validate);
4347 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4349 if (!(s->flags & SLAB_STORE_USER))
4351 return list_locations(s, buf, TRACK_ALLOC);
4353 SLAB_ATTR_RO(alloc_calls);
4355 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4357 if (!(s->flags & SLAB_STORE_USER))
4359 return list_locations(s, buf, TRACK_FREE);
4361 SLAB_ATTR_RO(free_calls);
4362 #endif /* CONFIG_SLUB_DEBUG */
4364 #ifdef CONFIG_FAILSLAB
4365 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4367 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4370 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4373 s->flags &= ~SLAB_FAILSLAB;
4375 s->flags |= SLAB_FAILSLAB;
4378 SLAB_ATTR(failslab);
4381 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4386 static ssize_t shrink_store(struct kmem_cache *s,
4387 const char *buf, size_t length)
4389 if (buf[0] == '1') {
4390 int rc = kmem_cache_shrink(s);
4401 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
4403 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
4406 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
4407 const char *buf, size_t length)
4409 unsigned long ratio;
4412 err = strict_strtoul(buf, 10, &ratio);
4417 s->remote_node_defrag_ratio = ratio * 10;
4421 SLAB_ATTR(remote_node_defrag_ratio);
4424 #ifdef CONFIG_SLUB_STATS
4425 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4427 unsigned long sum = 0;
4430 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4435 for_each_online_cpu(cpu) {
4436 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
4442 len = sprintf(buf, "%lu", sum);
4445 for_each_online_cpu(cpu) {
4446 if (data[cpu] && len < PAGE_SIZE - 20)
4447 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
4451 return len + sprintf(buf + len, "\n");
4454 static void clear_stat(struct kmem_cache *s, enum stat_item si)
4458 for_each_online_cpu(cpu)
4459 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
4462 #define STAT_ATTR(si, text) \
4463 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4465 return show_stat(s, buf, si); \
4467 static ssize_t text##_store(struct kmem_cache *s, \
4468 const char *buf, size_t length) \
4470 if (buf[0] != '0') \
4472 clear_stat(s, si); \
4477 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4478 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4479 STAT_ATTR(FREE_FASTPATH, free_fastpath);
4480 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4481 STAT_ATTR(FREE_FROZEN, free_frozen);
4482 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4483 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4484 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4485 STAT_ATTR(ALLOC_SLAB, alloc_slab);
4486 STAT_ATTR(ALLOC_REFILL, alloc_refill);
4487 STAT_ATTR(FREE_SLAB, free_slab);
4488 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4489 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4490 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4491 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4492 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4493 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
4494 STAT_ATTR(ORDER_FALLBACK, order_fallback);
4497 static struct attribute *slab_attrs[] = {
4498 &slab_size_attr.attr,
4499 &object_size_attr.attr,
4500 &objs_per_slab_attr.attr,
4502 &min_partial_attr.attr,
4504 &objects_partial_attr.attr,
4506 &cpu_slabs_attr.attr,
4510 &hwcache_align_attr.attr,
4511 &reclaim_account_attr.attr,
4512 &destroy_by_rcu_attr.attr,
4514 #ifdef CONFIG_SLUB_DEBUG
4515 &total_objects_attr.attr,
4517 &sanity_checks_attr.attr,
4519 &red_zone_attr.attr,
4521 &store_user_attr.attr,
4522 &validate_attr.attr,
4523 &alloc_calls_attr.attr,
4524 &free_calls_attr.attr,
4526 #ifdef CONFIG_ZONE_DMA
4527 &cache_dma_attr.attr,
4530 &remote_node_defrag_ratio_attr.attr,
4532 #ifdef CONFIG_SLUB_STATS
4533 &alloc_fastpath_attr.attr,
4534 &alloc_slowpath_attr.attr,
4535 &free_fastpath_attr.attr,
4536 &free_slowpath_attr.attr,
4537 &free_frozen_attr.attr,
4538 &free_add_partial_attr.attr,
4539 &free_remove_partial_attr.attr,
4540 &alloc_from_partial_attr.attr,
4541 &alloc_slab_attr.attr,
4542 &alloc_refill_attr.attr,
4543 &free_slab_attr.attr,
4544 &cpuslab_flush_attr.attr,
4545 &deactivate_full_attr.attr,
4546 &deactivate_empty_attr.attr,
4547 &deactivate_to_head_attr.attr,
4548 &deactivate_to_tail_attr.attr,
4549 &deactivate_remote_frees_attr.attr,
4550 &order_fallback_attr.attr,
4552 #ifdef CONFIG_FAILSLAB
4553 &failslab_attr.attr,
4559 static struct attribute_group slab_attr_group = {
4560 .attrs = slab_attrs,
4563 static ssize_t slab_attr_show(struct kobject *kobj,
4564 struct attribute *attr,
4567 struct slab_attribute *attribute;
4568 struct kmem_cache *s;
4571 attribute = to_slab_attr(attr);
4574 if (!attribute->show)
4577 err = attribute->show(s, buf);
4582 static ssize_t slab_attr_store(struct kobject *kobj,
4583 struct attribute *attr,
4584 const char *buf, size_t len)
4586 struct slab_attribute *attribute;
4587 struct kmem_cache *s;
4590 attribute = to_slab_attr(attr);
4593 if (!attribute->store)
4596 err = attribute->store(s, buf, len);
4601 static void kmem_cache_release(struct kobject *kobj)
4603 struct kmem_cache *s = to_slab(kobj);
4609 static const struct sysfs_ops slab_sysfs_ops = {
4610 .show = slab_attr_show,
4611 .store = slab_attr_store,
4614 static struct kobj_type slab_ktype = {
4615 .sysfs_ops = &slab_sysfs_ops,
4616 .release = kmem_cache_release
4619 static int uevent_filter(struct kset *kset, struct kobject *kobj)
4621 struct kobj_type *ktype = get_ktype(kobj);
4623 if (ktype == &slab_ktype)
4628 static const struct kset_uevent_ops slab_uevent_ops = {
4629 .filter = uevent_filter,
4632 static struct kset *slab_kset;
4634 #define ID_STR_LENGTH 64
4636 /* Create a unique string id for a slab cache:
4638 * Format :[flags-]size
4640 static char *create_unique_id(struct kmem_cache *s)
4642 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4649 * First flags affecting slabcache operations. We will only
4650 * get here for aliasable slabs so we do not need to support
4651 * too many flags. The flags here must cover all flags that
4652 * are matched during merging to guarantee that the id is
4655 if (s->flags & SLAB_CACHE_DMA)
4657 if (s->flags & SLAB_RECLAIM_ACCOUNT)
4659 if (s->flags & SLAB_DEBUG_FREE)
4661 if (!(s->flags & SLAB_NOTRACK))
4665 p += sprintf(p, "%07d", s->size);
4666 BUG_ON(p > name + ID_STR_LENGTH - 1);
4670 static int sysfs_slab_add(struct kmem_cache *s)
4676 if (slab_state < SYSFS)
4677 /* Defer until later */
4680 unmergeable = slab_unmergeable(s);
4683 * Slabcache can never be merged so we can use the name proper.
4684 * This is typically the case for debug situations. In that
4685 * case we can catch duplicate names easily.
4687 sysfs_remove_link(&slab_kset->kobj, s->name);
4691 * Create a unique name for the slab as a target
4694 name = create_unique_id(s);
4697 s->kobj.kset = slab_kset;
4698 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4700 kobject_put(&s->kobj);
4704 err = sysfs_create_group(&s->kobj, &slab_attr_group);
4706 kobject_del(&s->kobj);
4707 kobject_put(&s->kobj);
4710 kobject_uevent(&s->kobj, KOBJ_ADD);
4712 /* Setup first alias */
4713 sysfs_slab_alias(s, s->name);
4719 static void sysfs_slab_remove(struct kmem_cache *s)
4721 if (slab_state < SYSFS)
4723 * Sysfs has not been setup yet so no need to remove the
4728 kobject_uevent(&s->kobj, KOBJ_REMOVE);
4729 kobject_del(&s->kobj);
4730 kobject_put(&s->kobj);
4734 * Need to buffer aliases during bootup until sysfs becomes
4735 * available lest we lose that information.
4737 struct saved_alias {
4738 struct kmem_cache *s;
4740 struct saved_alias *next;
4743 static struct saved_alias *alias_list;
4745 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4747 struct saved_alias *al;
4749 if (slab_state == SYSFS) {
4751 * If we have a leftover link then remove it.
4753 sysfs_remove_link(&slab_kset->kobj, name);
4754 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
4757 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4763 al->next = alias_list;
4768 static int __init slab_sysfs_init(void)
4770 struct kmem_cache *s;
4773 down_write(&slub_lock);
4775 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
4777 up_write(&slub_lock);
4778 printk(KERN_ERR "Cannot register slab subsystem.\n");
4784 list_for_each_entry(s, &slab_caches, list) {
4785 err = sysfs_slab_add(s);
4787 printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4788 " to sysfs\n", s->name);
4791 while (alias_list) {
4792 struct saved_alias *al = alias_list;
4794 alias_list = alias_list->next;
4795 err = sysfs_slab_alias(al->s, al->name);
4797 printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4798 " %s to sysfs\n", s->name);
4802 up_write(&slub_lock);
4807 __initcall(slab_sysfs_init);
4808 #endif /* CONFIG_SYSFS */
4811 * The /proc/slabinfo ABI
4813 #ifdef CONFIG_SLABINFO
4814 static void print_slabinfo_header(struct seq_file *m)
4816 seq_puts(m, "slabinfo - version: 2.1\n");
4817 seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4818 "<objperslab> <pagesperslab>");
4819 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4820 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4824 static void *s_start(struct seq_file *m, loff_t *pos)
4828 down_read(&slub_lock);
4830 print_slabinfo_header(m);
4832 return seq_list_start(&slab_caches, *pos);
4835 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4837 return seq_list_next(p, &slab_caches, pos);
4840 static void s_stop(struct seq_file *m, void *p)
4842 up_read(&slub_lock);
4845 static int s_show(struct seq_file *m, void *p)
4847 unsigned long nr_partials = 0;
4848 unsigned long nr_slabs = 0;
4849 unsigned long nr_inuse = 0;
4850 unsigned long nr_objs = 0;
4851 unsigned long nr_free = 0;
4852 struct kmem_cache *s;
4855 s = list_entry(p, struct kmem_cache, list);
4857 for_each_online_node(node) {
4858 struct kmem_cache_node *n = get_node(s, node);
4863 nr_partials += n->nr_partial;
4864 nr_slabs += atomic_long_read(&n->nr_slabs);
4865 nr_objs += atomic_long_read(&n->total_objects);
4866 nr_free += count_partial(n, count_free);
4869 nr_inuse = nr_objs - nr_free;
4871 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
4872 nr_objs, s->size, oo_objects(s->oo),
4873 (1 << oo_order(s->oo)));
4874 seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4875 seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4881 static const struct seq_operations slabinfo_op = {
4888 static int slabinfo_open(struct inode *inode, struct file *file)
4890 return seq_open(file, &slabinfo_op);
4893 static const struct file_operations proc_slabinfo_operations = {
4894 .open = slabinfo_open,
4896 .llseek = seq_lseek,
4897 .release = seq_release,
4900 static int __init slab_proc_init(void)
4902 proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
4905 module_init(slab_proc_init);
4906 #endif /* CONFIG_SLABINFO */