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/cache.h>
64 #include <linux/init.h>
65 #include <linux/module.h>
66 #include <linux/rcupdate.h>
67 #include <linux/list.h>
68 #include <asm/atomic.h>
71 * slob_block has a field 'units', which indicates size of block if +ve,
72 * or offset of next block if -ve (in SLOB_UNITs).
74 * Free blocks of size 1 unit simply contain the offset of the next block.
75 * Those with larger size contain their size in the first SLOB_UNIT of
76 * memory, and the offset of the next free block in the second SLOB_UNIT.
78 #if PAGE_SIZE <= (32767 * 2)
79 typedef s16 slobidx_t;
81 typedef s32 slobidx_t;
87 typedef struct slob_block slob_t;
90 * We use struct page fields to manage some slob allocation aspects,
91 * however to avoid the horrible mess in include/linux/mm_types.h, we'll
92 * just define our own struct page type variant here.
97 unsigned long flags; /* mandatory */
98 atomic_t _count; /* mandatory */
99 slobidx_t units; /* free units left in page */
100 unsigned long pad[2];
101 slob_t *free; /* first free slob_t in page */
102 struct list_head list; /* linked list of free pages */
107 static inline void struct_slob_page_wrong_size(void)
108 { BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); }
111 * free_slob_page: call before a slob_page is returned to the page allocator.
113 static inline void free_slob_page(struct slob_page *sp)
115 reset_page_mapcount(&sp->page);
116 sp->page.mapping = NULL;
120 * All partially free slob pages go on these lists.
122 #define SLOB_BREAK1 256
123 #define SLOB_BREAK2 1024
124 static LIST_HEAD(free_slob_small);
125 static LIST_HEAD(free_slob_medium);
126 static LIST_HEAD(free_slob_large);
129 * is_slob_page: True for all slob pages (false for bigblock pages)
131 static inline int is_slob_page(struct slob_page *sp)
133 return PageSlobPage((struct page *)sp);
136 static inline void set_slob_page(struct slob_page *sp)
138 __SetPageSlobPage((struct page *)sp);
141 static inline void clear_slob_page(struct slob_page *sp)
143 __ClearPageSlobPage((struct page *)sp);
146 static inline struct slob_page *slob_page(const void *addr)
148 return (struct slob_page *)virt_to_page(addr);
152 * slob_page_free: true for pages on free_slob_pages list.
154 static inline int slob_page_free(struct slob_page *sp)
156 return PageSlobFree((struct page *)sp);
159 static void set_slob_page_free(struct slob_page *sp, struct list_head *list)
161 list_add(&sp->list, list);
162 __SetPageSlobFree((struct page *)sp);
165 static inline void clear_slob_page_free(struct slob_page *sp)
168 __ClearPageSlobFree((struct page *)sp);
171 #define SLOB_UNIT sizeof(slob_t)
172 #define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT)
173 #define SLOB_ALIGN L1_CACHE_BYTES
176 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
177 * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free
178 * the block using call_rcu.
181 struct rcu_head head;
186 * slob_lock protects all slob allocator structures.
188 static DEFINE_SPINLOCK(slob_lock);
191 * Encode the given size and next info into a free slob block s.
193 static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
195 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
196 slobidx_t offset = next - base;
202 s[0].units = -offset;
206 * Return the size of a slob block.
208 static slobidx_t slob_units(slob_t *s)
216 * Return the next free slob block pointer after this one.
218 static slob_t *slob_next(slob_t *s)
220 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
231 * Returns true if s is the last free block in its page.
233 static int slob_last(slob_t *s)
235 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
238 static void *slob_new_pages(gfp_t gfp, int order, int node)
244 page = alloc_pages_node(node, gfp, order);
247 page = alloc_pages(gfp, order);
252 return page_address(page);
255 static void slob_free_pages(void *b, int order)
257 free_pages((unsigned long)b, order);
261 * Allocate a slob block within a given slob_page sp.
263 static void *slob_page_alloc(struct slob_page *sp, size_t size, int align)
265 slob_t *prev, *cur, *aligned = NULL;
266 int delta = 0, units = SLOB_UNITS(size);
268 for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
269 slobidx_t avail = slob_units(cur);
272 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
273 delta = aligned - cur;
275 if (avail >= units + delta) { /* room enough? */
278 if (delta) { /* need to fragment head to align? */
279 next = slob_next(cur);
280 set_slob(aligned, avail - delta, next);
281 set_slob(cur, delta, aligned);
284 avail = slob_units(cur);
287 next = slob_next(cur);
288 if (avail == units) { /* exact fit? unlink. */
290 set_slob(prev, slob_units(prev), next);
293 } else { /* fragment */
295 set_slob(prev, slob_units(prev), cur + units);
297 sp->free = cur + units;
298 set_slob(cur + units, avail - units, next);
303 clear_slob_page_free(sp);
312 * slob_alloc: entry point into the slob allocator.
314 static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
316 struct slob_page *sp;
317 struct list_head *prev;
318 struct list_head *slob_list;
322 if (size < SLOB_BREAK1)
323 slob_list = &free_slob_small;
324 else if (size < SLOB_BREAK2)
325 slob_list = &free_slob_medium;
327 slob_list = &free_slob_large;
329 spin_lock_irqsave(&slob_lock, flags);
330 /* Iterate through each partially free page, try to find room */
331 list_for_each_entry(sp, slob_list, list) {
334 * If there's a node specification, search for a partial
335 * page with a matching node id in the freelist.
337 if (node != -1 && page_to_nid(&sp->page) != node)
340 /* Enough room on this page? */
341 if (sp->units < SLOB_UNITS(size))
344 /* Attempt to alloc */
345 prev = sp->list.prev;
346 b = slob_page_alloc(sp, size, align);
350 /* Improve fragment distribution and reduce our average
351 * search time by starting our next search here. (see
352 * Knuth vol 1, sec 2.5, pg 449) */
353 if (prev != slob_list->prev &&
354 slob_list->next != prev->next)
355 list_move_tail(slob_list, prev->next);
358 spin_unlock_irqrestore(&slob_lock, flags);
360 /* Not enough space: must allocate a new page */
362 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
368 spin_lock_irqsave(&slob_lock, flags);
369 sp->units = SLOB_UNITS(PAGE_SIZE);
371 INIT_LIST_HEAD(&sp->list);
372 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
373 set_slob_page_free(sp, slob_list);
374 b = slob_page_alloc(sp, size, align);
376 spin_unlock_irqrestore(&slob_lock, flags);
378 if (unlikely((gfp & __GFP_ZERO) && b))
384 * slob_free: entry point into the slob allocator.
386 static void slob_free(void *block, int size)
388 struct slob_page *sp;
389 slob_t *prev, *next, *b = (slob_t *)block;
393 if (unlikely(ZERO_OR_NULL_PTR(block)))
397 sp = slob_page(block);
398 units = SLOB_UNITS(size);
400 spin_lock_irqsave(&slob_lock, flags);
402 if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
403 /* Go directly to page allocator. Do not pass slob allocator */
404 if (slob_page_free(sp))
405 clear_slob_page_free(sp);
406 spin_unlock_irqrestore(&slob_lock, flags);
409 free_page((unsigned long)b);
413 if (!slob_page_free(sp)) {
414 /* This slob page is about to become partially free. Easy! */
418 (void *)((unsigned long)(b +
419 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
420 set_slob_page_free(sp, &free_slob_small);
425 * Otherwise the page is already partially free, so find reinsertion
431 if (b + units == sp->free) {
432 units += slob_units(sp->free);
433 sp->free = slob_next(sp->free);
435 set_slob(b, units, sp->free);
439 next = slob_next(prev);
442 next = slob_next(prev);
445 if (!slob_last(prev) && b + units == next) {
446 units += slob_units(next);
447 set_slob(b, units, slob_next(next));
449 set_slob(b, units, next);
451 if (prev + slob_units(prev) == b) {
452 units = slob_units(b) + slob_units(prev);
453 set_slob(prev, units, slob_next(b));
455 set_slob(prev, slob_units(prev), b);
458 spin_unlock_irqrestore(&slob_lock, flags);
462 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
465 #ifndef ARCH_KMALLOC_MINALIGN
466 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long)
469 #ifndef ARCH_SLAB_MINALIGN
470 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long)
473 void *__kmalloc_node(size_t size, gfp_t gfp, int node)
476 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
478 lockdep_trace_alloc(gfp);
480 if (size < PAGE_SIZE - align) {
482 return ZERO_SIZE_PTR;
484 m = slob_alloc(size + align, gfp, align, node);
488 return (void *)m + align;
492 ret = slob_new_pages(gfp | __GFP_COMP, get_order(size), node);
495 page = virt_to_page(ret);
496 page->private = size;
501 EXPORT_SYMBOL(__kmalloc_node);
503 void kfree(const void *block)
505 struct slob_page *sp;
507 if (unlikely(ZERO_OR_NULL_PTR(block)))
510 sp = slob_page(block);
511 if (is_slob_page(sp)) {
512 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
513 unsigned int *m = (unsigned int *)(block - align);
514 slob_free(m, *m + align);
518 EXPORT_SYMBOL(kfree);
520 /* can't use ksize for kmem_cache_alloc memory, only kmalloc */
521 size_t ksize(const void *block)
523 struct slob_page *sp;
526 if (unlikely(block == ZERO_SIZE_PTR))
529 sp = slob_page(block);
530 if (is_slob_page(sp)) {
531 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
532 unsigned int *m = (unsigned int *)(block - align);
533 return SLOB_UNITS(*m) * SLOB_UNIT;
535 return sp->page.private;
537 EXPORT_SYMBOL(ksize);
540 unsigned int size, align;
543 void (*ctor)(void *);
546 struct kmem_cache *kmem_cache_create(const char *name, size_t size,
547 size_t align, unsigned long flags, void (*ctor)(void *))
549 struct kmem_cache *c;
551 c = slob_alloc(sizeof(struct kmem_cache),
552 GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
557 if (flags & SLAB_DESTROY_BY_RCU) {
558 /* leave room for rcu footer at the end of object */
559 c->size += sizeof(struct slob_rcu);
563 /* ignore alignment unless it's forced */
564 c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
565 if (c->align < ARCH_SLAB_MINALIGN)
566 c->align = ARCH_SLAB_MINALIGN;
567 if (c->align < align)
569 } else if (flags & SLAB_PANIC)
570 panic("Cannot create slab cache %s\n", name);
574 EXPORT_SYMBOL(kmem_cache_create);
576 void kmem_cache_destroy(struct kmem_cache *c)
578 slob_free(c, sizeof(struct kmem_cache));
580 EXPORT_SYMBOL(kmem_cache_destroy);
582 void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
586 if (c->size < PAGE_SIZE)
587 b = slob_alloc(c->size, flags, c->align, node);
589 b = slob_new_pages(flags, get_order(c->size), node);
596 EXPORT_SYMBOL(kmem_cache_alloc_node);
598 static void __kmem_cache_free(void *b, int size)
600 if (size < PAGE_SIZE)
603 slob_free_pages(b, get_order(size));
606 static void kmem_rcu_free(struct rcu_head *head)
608 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
609 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
611 __kmem_cache_free(b, slob_rcu->size);
614 void kmem_cache_free(struct kmem_cache *c, void *b)
616 if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
617 struct slob_rcu *slob_rcu;
618 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
619 INIT_RCU_HEAD(&slob_rcu->head);
620 slob_rcu->size = c->size;
621 call_rcu(&slob_rcu->head, kmem_rcu_free);
623 __kmem_cache_free(b, c->size);
626 EXPORT_SYMBOL(kmem_cache_free);
628 unsigned int kmem_cache_size(struct kmem_cache *c)
632 EXPORT_SYMBOL(kmem_cache_size);
634 const char *kmem_cache_name(struct kmem_cache *c)
638 EXPORT_SYMBOL(kmem_cache_name);
640 int kmem_cache_shrink(struct kmem_cache *d)
644 EXPORT_SYMBOL(kmem_cache_shrink);
646 int kmem_ptr_validate(struct kmem_cache *a, const void *b)
651 static unsigned int slob_ready __read_mostly;
653 int slab_is_available(void)
658 void __init kmem_cache_init(void)