2 * SLOB Allocator: Simple List Of Blocks
4 * Matt Mackall <mpm@selenic.com> 12/30/03
6 * NUMA support by Paul Mundt, 2007.
10 * The core of SLOB is a traditional K&R style heap allocator, with
11 * support for returning aligned objects. The granularity of this
12 * allocator is as little as 2 bytes, however typically most architectures
13 * will require 4 bytes on 32-bit and 8 bytes on 64-bit.
15 * The slob heap is a set of linked list of pages from alloc_pages(),
16 * and within each page, there is a singly-linked list of free blocks
17 * (slob_t). The heap is grown on demand. To reduce fragmentation,
18 * heap pages are segregated into three lists, with objects less than
19 * 256 bytes, objects less than 1024 bytes, and all other objects.
21 * Allocation from heap involves first searching for a page with
22 * sufficient free blocks (using a next-fit-like approach) followed by
23 * a first-fit scan of the page. Deallocation inserts objects back
24 * into the free list in address order, so this is effectively an
25 * address-ordered first fit.
27 * Above this is an implementation of kmalloc/kfree. Blocks returned
28 * from kmalloc are prepended with a 4-byte header with the kmalloc size.
29 * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
30 * alloc_pages() directly, allocating compound pages so the page order
31 * does not have to be separately tracked, and also stores the exact
32 * allocation size in page->private so that it can be used to accurately
33 * provide ksize(). These objects are detected in kfree() because slob_page()
36 * SLAB is emulated on top of SLOB by simply calling constructors and
37 * destructors for every SLAB allocation. Objects are returned with the
38 * 4-byte alignment unless the SLAB_HWCACHE_ALIGN flag is set, in which
39 * case the low-level allocator will fragment blocks to create the proper
40 * alignment. Again, objects of page-size or greater are allocated by
41 * calling alloc_pages(). As SLAB objects know their size, no separate
42 * size bookkeeping is necessary and there is essentially no allocation
43 * space overhead, and compound pages aren't needed for multi-page
46 * NUMA support in SLOB is fairly simplistic, pushing most of the real
47 * logic down to the page allocator, and simply doing the node accounting
48 * on the upper levels. In the event that a node id is explicitly
49 * provided, alloc_pages_node() with the specified node id is used
50 * instead. The common case (or when the node id isn't explicitly provided)
51 * will default to the current node, as per numa_node_id().
53 * Node aware pages are still inserted in to the global freelist, and
54 * these are scanned for by matching against the node id encoded in the
55 * page flags. As a result, block allocations that can be satisfied from
56 * the freelist will only be done so on pages residing on the same node,
57 * in order to prevent random node placement.
60 #include <linux/kernel.h>
61 #include <linux/slab.h>
63 #include <linux/swap.h> /* struct reclaim_state */
64 #include <linux/cache.h>
65 #include <linux/init.h>
66 #include <linux/module.h>
67 #include <linux/rcupdate.h>
68 #include <linux/list.h>
69 #include <linux/kmemtrace.h>
70 #include <linux/kmemleak.h>
71 #include <asm/atomic.h>
74 * slob_block has a field 'units', which indicates size of block if +ve,
75 * or offset of next block if -ve (in SLOB_UNITs).
77 * Free blocks of size 1 unit simply contain the offset of the next block.
78 * Those with larger size contain their size in the first SLOB_UNIT of
79 * memory, and the offset of the next free block in the second SLOB_UNIT.
81 #if PAGE_SIZE <= (32767 * 2)
82 typedef s16 slobidx_t;
84 typedef s32 slobidx_t;
90 typedef struct slob_block slob_t;
93 * We use struct page fields to manage some slob allocation aspects,
94 * however to avoid the horrible mess in include/linux/mm_types.h, we'll
95 * just define our own struct page type variant here.
100 unsigned long flags; /* mandatory */
101 atomic_t _count; /* mandatory */
102 slobidx_t units; /* free units left in page */
103 unsigned long pad[2];
104 slob_t *free; /* first free slob_t in page */
105 struct list_head list; /* linked list of free pages */
110 static inline void struct_slob_page_wrong_size(void)
111 { BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); }
114 * free_slob_page: call before a slob_page is returned to the page allocator.
116 static inline void free_slob_page(struct slob_page *sp)
118 reset_page_mapcount(&sp->page);
119 sp->page.mapping = NULL;
123 * All partially free slob pages go on these lists.
125 #define SLOB_BREAK1 256
126 #define SLOB_BREAK2 1024
127 static LIST_HEAD(free_slob_small);
128 static LIST_HEAD(free_slob_medium);
129 static LIST_HEAD(free_slob_large);
132 * is_slob_page: True for all slob pages (false for bigblock pages)
134 static inline int is_slob_page(struct slob_page *sp)
136 return PageSlobPage((struct page *)sp);
139 static inline void set_slob_page(struct slob_page *sp)
141 __SetPageSlobPage((struct page *)sp);
144 static inline void clear_slob_page(struct slob_page *sp)
146 __ClearPageSlobPage((struct page *)sp);
149 static inline struct slob_page *slob_page(const void *addr)
151 return (struct slob_page *)virt_to_page(addr);
155 * slob_page_free: true for pages on free_slob_pages list.
157 static inline int slob_page_free(struct slob_page *sp)
159 return PageSlobFree((struct page *)sp);
162 static void set_slob_page_free(struct slob_page *sp, struct list_head *list)
164 list_add(&sp->list, list);
165 __SetPageSlobFree((struct page *)sp);
168 static inline void clear_slob_page_free(struct slob_page *sp)
171 __ClearPageSlobFree((struct page *)sp);
174 #define SLOB_UNIT sizeof(slob_t)
175 #define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT)
176 #define SLOB_ALIGN L1_CACHE_BYTES
179 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
180 * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free
181 * the block using call_rcu.
184 struct rcu_head head;
189 * slob_lock protects all slob allocator structures.
191 static DEFINE_SPINLOCK(slob_lock);
194 * Encode the given size and next info into a free slob block s.
196 static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
198 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
199 slobidx_t offset = next - base;
205 s[0].units = -offset;
209 * Return the size of a slob block.
211 static slobidx_t slob_units(slob_t *s)
219 * Return the next free slob block pointer after this one.
221 static slob_t *slob_next(slob_t *s)
223 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
234 * Returns true if s is the last free block in its page.
236 static int slob_last(slob_t *s)
238 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
241 static void *slob_new_pages(gfp_t gfp, int order, int node)
247 page = alloc_pages_node(node, gfp, order);
250 page = alloc_pages(gfp, order);
255 return page_address(page);
258 static void slob_free_pages(void *b, int order)
260 if (current->reclaim_state)
261 current->reclaim_state->reclaimed_slab += 1 << order;
262 free_pages((unsigned long)b, order);
266 * Allocate a slob block within a given slob_page sp.
268 static void *slob_page_alloc(struct slob_page *sp, size_t size, int align)
270 slob_t *prev, *cur, *aligned = NULL;
271 int delta = 0, units = SLOB_UNITS(size);
273 for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
274 slobidx_t avail = slob_units(cur);
277 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
278 delta = aligned - cur;
280 if (avail >= units + delta) { /* room enough? */
283 if (delta) { /* need to fragment head to align? */
284 next = slob_next(cur);
285 set_slob(aligned, avail - delta, next);
286 set_slob(cur, delta, aligned);
289 avail = slob_units(cur);
292 next = slob_next(cur);
293 if (avail == units) { /* exact fit? unlink. */
295 set_slob(prev, slob_units(prev), next);
298 } else { /* fragment */
300 set_slob(prev, slob_units(prev), cur + units);
302 sp->free = cur + units;
303 set_slob(cur + units, avail - units, next);
308 clear_slob_page_free(sp);
317 * slob_alloc: entry point into the slob allocator.
319 static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
321 struct slob_page *sp;
322 struct list_head *prev;
323 struct list_head *slob_list;
327 if (size < SLOB_BREAK1)
328 slob_list = &free_slob_small;
329 else if (size < SLOB_BREAK2)
330 slob_list = &free_slob_medium;
332 slob_list = &free_slob_large;
334 spin_lock_irqsave(&slob_lock, flags);
335 /* Iterate through each partially free page, try to find room */
336 list_for_each_entry(sp, slob_list, list) {
339 * If there's a node specification, search for a partial
340 * page with a matching node id in the freelist.
342 if (node != -1 && page_to_nid(&sp->page) != node)
345 /* Enough room on this page? */
346 if (sp->units < SLOB_UNITS(size))
349 /* Attempt to alloc */
350 prev = sp->list.prev;
351 b = slob_page_alloc(sp, size, align);
355 /* Improve fragment distribution and reduce our average
356 * search time by starting our next search here. (see
357 * Knuth vol 1, sec 2.5, pg 449) */
358 if (prev != slob_list->prev &&
359 slob_list->next != prev->next)
360 list_move_tail(slob_list, prev->next);
363 spin_unlock_irqrestore(&slob_lock, flags);
365 /* Not enough space: must allocate a new page */
367 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
373 spin_lock_irqsave(&slob_lock, flags);
374 sp->units = SLOB_UNITS(PAGE_SIZE);
376 INIT_LIST_HEAD(&sp->list);
377 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
378 set_slob_page_free(sp, slob_list);
379 b = slob_page_alloc(sp, size, align);
381 spin_unlock_irqrestore(&slob_lock, flags);
383 if (unlikely((gfp & __GFP_ZERO) && b))
389 * slob_free: entry point into the slob allocator.
391 static void slob_free(void *block, int size)
393 struct slob_page *sp;
394 slob_t *prev, *next, *b = (slob_t *)block;
398 if (unlikely(ZERO_OR_NULL_PTR(block)))
402 sp = slob_page(block);
403 units = SLOB_UNITS(size);
405 spin_lock_irqsave(&slob_lock, flags);
407 if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
408 /* Go directly to page allocator. Do not pass slob allocator */
409 if (slob_page_free(sp))
410 clear_slob_page_free(sp);
411 spin_unlock_irqrestore(&slob_lock, flags);
414 slob_free_pages(b, 0);
418 if (!slob_page_free(sp)) {
419 /* This slob page is about to become partially free. Easy! */
423 (void *)((unsigned long)(b +
424 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
425 set_slob_page_free(sp, &free_slob_small);
430 * Otherwise the page is already partially free, so find reinsertion
436 if (b + units == sp->free) {
437 units += slob_units(sp->free);
438 sp->free = slob_next(sp->free);
440 set_slob(b, units, sp->free);
444 next = slob_next(prev);
447 next = slob_next(prev);
450 if (!slob_last(prev) && b + units == next) {
451 units += slob_units(next);
452 set_slob(b, units, slob_next(next));
454 set_slob(b, units, next);
456 if (prev + slob_units(prev) == b) {
457 units = slob_units(b) + slob_units(prev);
458 set_slob(prev, units, slob_next(b));
460 set_slob(prev, slob_units(prev), b);
463 spin_unlock_irqrestore(&slob_lock, flags);
467 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
470 #ifndef ARCH_KMALLOC_MINALIGN
471 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long)
474 #ifndef ARCH_SLAB_MINALIGN
475 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long)
478 void *__kmalloc_node(size_t size, gfp_t gfp, int node)
481 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
484 lockdep_trace_alloc(gfp);
486 if (size < PAGE_SIZE - align) {
488 return ZERO_SIZE_PTR;
490 m = slob_alloc(size + align, gfp, align, node);
495 ret = (void *)m + align;
497 trace_kmalloc_node(_RET_IP_, ret,
498 size, size + align, gfp, node);
500 unsigned int order = get_order(size);
502 ret = slob_new_pages(gfp | __GFP_COMP, get_order(size), node);
505 page = virt_to_page(ret);
506 page->private = size;
509 trace_kmalloc_node(_RET_IP_, ret,
510 size, PAGE_SIZE << order, gfp, node);
513 kmemleak_alloc(ret, size, 1, gfp);
516 EXPORT_SYMBOL(__kmalloc_node);
518 void kfree(const void *block)
520 struct slob_page *sp;
522 trace_kfree(_RET_IP_, block);
524 if (unlikely(ZERO_OR_NULL_PTR(block)))
526 kmemleak_free(block);
528 sp = slob_page(block);
529 if (is_slob_page(sp)) {
530 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
531 unsigned int *m = (unsigned int *)(block - align);
532 slob_free(m, *m + align);
536 EXPORT_SYMBOL(kfree);
538 /* can't use ksize for kmem_cache_alloc memory, only kmalloc */
539 size_t ksize(const void *block)
541 struct slob_page *sp;
544 if (unlikely(block == ZERO_SIZE_PTR))
547 sp = slob_page(block);
548 if (is_slob_page(sp)) {
549 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
550 unsigned int *m = (unsigned int *)(block - align);
551 return SLOB_UNITS(*m) * SLOB_UNIT;
553 return sp->page.private;
555 EXPORT_SYMBOL(ksize);
558 unsigned int size, align;
561 void (*ctor)(void *);
564 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
565 size_t align, unsigned long flags, void (*ctor)(void *))
567 struct kmem_cache *c;
569 c = slob_alloc(sizeof(struct kmem_cache),
570 GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
575 if (flags & SLAB_DESTROY_BY_RCU) {
576 /* leave room for rcu footer at the end of object */
577 c->size += sizeof(struct slob_rcu);
581 /* ignore alignment unless it's forced */
582 c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
583 if (c->align < ARCH_SLAB_MINALIGN)
584 c->align = ARCH_SLAB_MINALIGN;
585 if (c->align < align)
587 } else if (flags & SLAB_PANIC)
588 panic("Cannot create slab cache %s\n", name);
590 kmemleak_alloc(c, sizeof(struct kmem_cache), 1, GFP_KERNEL);
593 EXPORT_SYMBOL(kmem_cache_create);
595 void kmem_cache_destroy(struct kmem_cache *c)
598 slob_free(c, sizeof(struct kmem_cache));
600 EXPORT_SYMBOL(kmem_cache_destroy);
602 void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
606 if (c->size < PAGE_SIZE) {
607 b = slob_alloc(c->size, flags, c->align, node);
608 trace_kmem_cache_alloc_node(_RET_IP_, b, c->size,
609 SLOB_UNITS(c->size) * SLOB_UNIT,
612 b = slob_new_pages(flags, get_order(c->size), node);
613 trace_kmem_cache_alloc_node(_RET_IP_, b, c->size,
614 PAGE_SIZE << get_order(c->size),
621 kmemleak_alloc_recursive(b, c->size, 1, c->flags, flags);
624 EXPORT_SYMBOL(kmem_cache_alloc_node);
626 static void __kmem_cache_free(void *b, int size)
628 if (size < PAGE_SIZE)
631 slob_free_pages(b, get_order(size));
634 static void kmem_rcu_free(struct rcu_head *head)
636 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
637 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
639 __kmem_cache_free(b, slob_rcu->size);
642 void kmem_cache_free(struct kmem_cache *c, void *b)
644 kmemleak_free_recursive(b, c->flags);
645 if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
646 struct slob_rcu *slob_rcu;
647 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
648 INIT_RCU_HEAD(&slob_rcu->head);
649 slob_rcu->size = c->size;
650 call_rcu(&slob_rcu->head, kmem_rcu_free);
652 __kmem_cache_free(b, c->size);
655 trace_kmem_cache_free(_RET_IP_, b);
657 EXPORT_SYMBOL(kmem_cache_free);
659 unsigned int kmem_cache_size(struct kmem_cache *c)
663 EXPORT_SYMBOL(kmem_cache_size);
665 const char *kmem_cache_name(struct kmem_cache *c)
669 EXPORT_SYMBOL(kmem_cache_name);
671 int kmem_cache_shrink(struct kmem_cache *d)
675 EXPORT_SYMBOL(kmem_cache_shrink);
677 int kmem_ptr_validate(struct kmem_cache *a, const void *b)
682 static unsigned int slob_ready __read_mostly;
684 int slab_is_available(void)
689 void __init kmem_cache_init(void)