slob.c revision f8fcc93319faa09272185af100fb24e71b02ab03 
1/*
2 * SLOB Allocator: Simple List Of Blocks
3 *
4 * Matt Mackall <mpm@selenic.com> 12/30/03
5 *
6 * NUMA support by Paul Mundt, 2007.
7 *
8 * How SLOB works:
9 *
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.
14 *
15 * The slob heap is a linked list of pages from alloc_pages(), and
16 * within each page, there is a singly-linked list of free blocks (slob_t).
17 * The heap is grown on demand and allocation from the heap is currently
18 * first-fit.
19 *
20 * Above this is an implementation of kmalloc/kfree. Blocks returned
21 * from kmalloc are prepended with a 4-byte header with the kmalloc size.
22 * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
23 * alloc_pages() directly, allocating compound pages so the page order
24 * does not have to be separately tracked, and also stores the exact
25 * allocation size in page->private so that it can be used to accurately
26 * provide ksize(). These objects are detected in kfree() because slob_page()
27 * is false for them.
28 *
29 * SLAB is emulated on top of SLOB by simply calling constructors and
30 * destructors for every SLAB allocation. Objects are returned with the
31 * 4-byte alignment unless the SLAB_HWCACHE_ALIGN flag is set, in which
32 * case the low-level allocator will fragment blocks to create the proper
33 * alignment. Again, objects of page-size or greater are allocated by
34 * calling alloc_pages(). As SLAB objects know their size, no separate
35 * size bookkeeping is necessary and there is essentially no allocation
36 * space overhead, and compound pages aren't needed for multi-page
37 * allocations.
38 *
39 * NUMA support in SLOB is fairly simplistic, pushing most of the real
40 * logic down to the page allocator, and simply doing the node accounting
41 * on the upper levels. In the event that a node id is explicitly
42 * provided, alloc_pages_node() with the specified node id is used
43 * instead. The common case (or when the node id isn't explicitly provided)
44 * will default to the current node, as per numa_node_id().
45 *
46 * Node aware pages are still inserted in to the global freelist, and
47 * these are scanned for by matching against the node id encoded in the
48 * page flags. As a result, block allocations that can be satisfied from
49 * the freelist will only be done so on pages residing on the same node,
50 * in order to prevent random node placement.
51 */
52
53#include <linux/kernel.h>
54#include <linux/slab.h>
55#include <linux/mm.h>
56#include <linux/cache.h>
57#include <linux/init.h>
58#include <linux/module.h>
59#include <linux/rcupdate.h>
60#include <linux/list.h>
61#include <asm/atomic.h>
62
63/*
64 * slob_block has a field 'units', which indicates size of block if +ve,
65 * or offset of next block if -ve (in SLOB_UNITs).
66 *
67 * Free blocks of size 1 unit simply contain the offset of the next block.
68 * Those with larger size contain their size in the first SLOB_UNIT of
69 * memory, and the offset of the next free block in the second SLOB_UNIT.
70 */
71#if PAGE_SIZE <= (32767 * 2)
72typedef s16 slobidx_t;
73#else
74typedef s32 slobidx_t;
75#endif
76
77struct slob_block {
78	slobidx_t units;
79};
80typedef struct slob_block slob_t;
81
82/*
83 * We use struct page fields to manage some slob allocation aspects,
84 * however to avoid the horrible mess in include/linux/mm_types.h, we'll
85 * just define our own struct page type variant here.
86 */
87struct slob_page {
88	union {
89		struct {
90			unsigned long flags;	/* mandatory */
91			atomic_t _count;	/* mandatory */
92			slobidx_t units;	/* free units left in page */
93			unsigned long pad[2];
94			slob_t *free;		/* first free slob_t in page */
95			struct list_head list;	/* linked list of free pages */
96		};
97		struct page page;
98	};
99};
100static inline void struct_slob_page_wrong_size(void)
101{ BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); }
102
103/*
104 * free_slob_page: call before a slob_page is returned to the page allocator.
105 */
106static inline void free_slob_page(struct slob_page *sp)
107{
108	reset_page_mapcount(&sp->page);
109	sp->page.mapping = NULL;
110}
111
112/*
113 * All (partially) free slob pages go on this list.
114 */
115static LIST_HEAD(free_slob_pages);
116
117/*
118 * slob_page: True for all slob pages (false for bigblock pages)
119 */
120static inline int slob_page(struct slob_page *sp)
121{
122	return test_bit(PG_active, &sp->flags);
123}
124
125static inline void set_slob_page(struct slob_page *sp)
126{
127	__set_bit(PG_active, &sp->flags);
128}
129
130static inline void clear_slob_page(struct slob_page *sp)
131{
132	__clear_bit(PG_active, &sp->flags);
133}
134
135/*
136 * slob_page_free: true for pages on free_slob_pages list.
137 */
138static inline int slob_page_free(struct slob_page *sp)
139{
140	return test_bit(PG_private, &sp->flags);
141}
142
143static inline void set_slob_page_free(struct slob_page *sp)
144{
145	list_add(&sp->list, &free_slob_pages);
146	__set_bit(PG_private, &sp->flags);
147}
148
149static inline void clear_slob_page_free(struct slob_page *sp)
150{
151	list_del(&sp->list);
152	__clear_bit(PG_private, &sp->flags);
153}
154
155#define SLOB_UNIT sizeof(slob_t)
156#define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT)
157#define SLOB_ALIGN L1_CACHE_BYTES
158
159/*
160 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
161 * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free
162 * the block using call_rcu.
163 */
164struct slob_rcu {
165	struct rcu_head head;
166	int size;
167};
168
169/*
170 * slob_lock protects all slob allocator structures.
171 */
172static DEFINE_SPINLOCK(slob_lock);
173
174/*
175 * Encode the given size and next info into a free slob block s.
176 */
177static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
178{
179	slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
180	slobidx_t offset = next - base;
181
182	if (size > 1) {
183		s[0].units = size;
184		s[1].units = offset;
185	} else
186		s[0].units = -offset;
187}
188
189/*
190 * Return the size of a slob block.
191 */
192static slobidx_t slob_units(slob_t *s)
193{
194	if (s->units > 0)
195		return s->units;
196	return 1;
197}
198
199/*
200 * Return the next free slob block pointer after this one.
201 */
202static slob_t *slob_next(slob_t *s)
203{
204	slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
205	slobidx_t next;
206
207	if (s[0].units < 0)
208		next = -s[0].units;
209	else
210		next = s[1].units;
211	return base+next;
212}
213
214/*
215 * Returns true if s is the last free block in its page.
216 */
217static int slob_last(slob_t *s)
218{
219	return !((unsigned long)slob_next(s) & ~PAGE_MASK);
220}
221
222static void *slob_new_page(gfp_t gfp, int order, int node)
223{
224	void *page;
225
226#ifdef CONFIG_NUMA
227	if (node != -1)
228		page = alloc_pages_node(node, gfp, order);
229	else
230#endif
231		page = alloc_pages(gfp, order);
232
233	if (!page)
234		return NULL;
235
236	return page_address(page);
237}
238
239/*
240 * Allocate a slob block within a given slob_page sp.
241 */
242static void *slob_page_alloc(struct slob_page *sp, size_t size, int align)
243{
244	slob_t *prev, *cur, *aligned = 0;
245	int delta = 0, units = SLOB_UNITS(size);
246
247	for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
248		slobidx_t avail = slob_units(cur);
249
250		if (align) {
251			aligned = (slob_t *)ALIGN((unsigned long)cur, align);
252			delta = aligned - cur;
253		}
254		if (avail >= units + delta) { /* room enough? */
255			slob_t *next;
256
257			if (delta) { /* need to fragment head to align? */
258				next = slob_next(cur);
259				set_slob(aligned, avail - delta, next);
260				set_slob(cur, delta, aligned);
261				prev = cur;
262				cur = aligned;
263				avail = slob_units(cur);
264			}
265
266			next = slob_next(cur);
267			if (avail == units) { /* exact fit? unlink. */
268				if (prev)
269					set_slob(prev, slob_units(prev), next);
270				else
271					sp->free = next;
272			} else { /* fragment */
273				if (prev)
274					set_slob(prev, slob_units(prev), cur + units);
275				else
276					sp->free = cur + units;
277				set_slob(cur + units, avail - units, next);
278			}
279
280			sp->units -= units;
281			if (!sp->units)
282				clear_slob_page_free(sp);
283			return cur;
284		}
285		if (slob_last(cur))
286			return NULL;
287	}
288}
289
290/*
291 * slob_alloc: entry point into the slob allocator.
292 */
293static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
294{
295	struct slob_page *sp;
296	struct list_head *prev;
297	slob_t *b = NULL;
298	unsigned long flags;
299
300	spin_lock_irqsave(&slob_lock, flags);
301	/* Iterate through each partially free page, try to find room */
302	list_for_each_entry(sp, &free_slob_pages, list) {
303#ifdef CONFIG_NUMA
304		/*
305		 * If there's a node specification, search for a partial
306		 * page with a matching node id in the freelist.
307		 */
308		if (node != -1 && page_to_nid(&sp->page) != node)
309			continue;
310#endif
311		/* Enough room on this page? */
312		if (sp->units < SLOB_UNITS(size))
313			continue;
314
315		/* Attempt to alloc */
316		prev = sp->list.prev;
317		b = slob_page_alloc(sp, size, align);
318		if (!b)
319			continue;
320
321		/* Improve fragment distribution and reduce our average
322		 * search time by starting our next search here. (see
323		 * Knuth vol 1, sec 2.5, pg 449) */
324		if (prev != free_slob_pages.prev &&
325				free_slob_pages.next != prev->next)
326			list_move_tail(&free_slob_pages, prev->next);
327		break;
328	}
329	spin_unlock_irqrestore(&slob_lock, flags);
330
331	/* Not enough space: must allocate a new page */
332	if (!b) {
333		b = slob_new_page(gfp, 0, node);
334		if (!b)
335			return 0;
336		sp = (struct slob_page *)virt_to_page(b);
337		set_slob_page(sp);
338
339		spin_lock_irqsave(&slob_lock, flags);
340		sp->units = SLOB_UNITS(PAGE_SIZE);
341		sp->free = b;
342		INIT_LIST_HEAD(&sp->list);
343		set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
344		set_slob_page_free(sp);
345		b = slob_page_alloc(sp, size, align);
346		BUG_ON(!b);
347		spin_unlock_irqrestore(&slob_lock, flags);
348	}
349	if (unlikely((gfp & __GFP_ZERO) && b))
350		memset(b, 0, size);
351	return b;
352}
353
354/*
355 * slob_free: entry point into the slob allocator.
356 */
357static void slob_free(void *block, int size)
358{
359	struct slob_page *sp;
360	slob_t *prev, *next, *b = (slob_t *)block;
361	slobidx_t units;
362	unsigned long flags;
363
364	if (unlikely(ZERO_OR_NULL_PTR(block)))
365		return;
366	BUG_ON(!size);
367
368	sp = (struct slob_page *)virt_to_page(block);
369	units = SLOB_UNITS(size);
370
371	spin_lock_irqsave(&slob_lock, flags);
372
373	if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
374		/* Go directly to page allocator. Do not pass slob allocator */
375		if (slob_page_free(sp))
376			clear_slob_page_free(sp);
377		clear_slob_page(sp);
378		free_slob_page(sp);
379		free_page((unsigned long)b);
380		goto out;
381	}
382
383	if (!slob_page_free(sp)) {
384		/* This slob page is about to become partially free. Easy! */
385		sp->units = units;
386		sp->free = b;
387		set_slob(b, units,
388			(void *)((unsigned long)(b +
389					SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
390		set_slob_page_free(sp);
391		goto out;
392	}
393
394	/*
395	 * Otherwise the page is already partially free, so find reinsertion
396	 * point.
397	 */
398	sp->units += units;
399
400	if (b < sp->free) {
401		set_slob(b, units, sp->free);
402		sp->free = b;
403	} else {
404		prev = sp->free;
405		next = slob_next(prev);
406		while (b > next) {
407			prev = next;
408			next = slob_next(prev);
409		}
410
411		if (!slob_last(prev) && b + units == next) {
412			units += slob_units(next);
413			set_slob(b, units, slob_next(next));
414		} else
415			set_slob(b, units, next);
416
417		if (prev + slob_units(prev) == b) {
418			units = slob_units(b) + slob_units(prev);
419			set_slob(prev, units, slob_next(b));
420		} else
421			set_slob(prev, slob_units(prev), b);
422	}
423out:
424	spin_unlock_irqrestore(&slob_lock, flags);
425}
426
427/*
428 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
429 */
430
431#ifndef ARCH_KMALLOC_MINALIGN
432#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long)
433#endif
434
435#ifndef ARCH_SLAB_MINALIGN
436#define ARCH_SLAB_MINALIGN __alignof__(unsigned long)
437#endif
438
439void *__kmalloc_node(size_t size, gfp_t gfp, int node)
440{
441	unsigned int *m;
442	int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
443
444	if (size < PAGE_SIZE - align) {
445		if (!size)
446			return ZERO_SIZE_PTR;
447
448		m = slob_alloc(size + align, gfp, align, node);
449		if (m)
450			*m = size;
451		return (void *)m + align;
452	} else {
453		void *ret;
454
455		ret = slob_new_page(gfp | __GFP_COMP, get_order(size), node);
456		if (ret) {
457			struct page *page;
458			page = virt_to_page(ret);
459			page->private = size;
460		}
461		return ret;
462	}
463}
464EXPORT_SYMBOL(__kmalloc_node);
465
466void kfree(const void *block)
467{
468	struct slob_page *sp;
469
470	if (unlikely(ZERO_OR_NULL_PTR(block)))
471		return;
472
473	sp = (struct slob_page *)virt_to_page(block);
474	if (slob_page(sp)) {
475		int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
476		unsigned int *m = (unsigned int *)(block - align);
477		slob_free(m, *m + align);
478	} else
479		put_page(&sp->page);
480}
481EXPORT_SYMBOL(kfree);
482
483/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
484size_t ksize(const void *block)
485{
486	struct slob_page *sp;
487
488	BUG_ON(!block);
489	if (unlikely(block == ZERO_SIZE_PTR))
490		return 0;
491
492	sp = (struct slob_page *)virt_to_page(block);
493	if (slob_page(sp))
494		return ((slob_t *)block - 1)->units + SLOB_UNIT;
495	else
496		return sp->page.private;
497}
498EXPORT_SYMBOL(ksize);
499
500struct kmem_cache {
501	unsigned int size, align;
502	unsigned long flags;
503	const char *name;
504	void (*ctor)(struct kmem_cache *, void *);
505};
506
507struct kmem_cache *kmem_cache_create(const char *name, size_t size,
508	size_t align, unsigned long flags,
509	void (*ctor)(struct kmem_cache *, void *))
510{
511	struct kmem_cache *c;
512
513	c = slob_alloc(sizeof(struct kmem_cache), flags, 0, -1);
514
515	if (c) {
516		c->name = name;
517		c->size = size;
518		if (flags & SLAB_DESTROY_BY_RCU) {
519			/* leave room for rcu footer at the end of object */
520			c->size += sizeof(struct slob_rcu);
521		}
522		c->flags = flags;
523		c->ctor = ctor;
524		/* ignore alignment unless it's forced */
525		c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
526		if (c->align < ARCH_SLAB_MINALIGN)
527			c->align = ARCH_SLAB_MINALIGN;
528		if (c->align < align)
529			c->align = align;
530	} else if (flags & SLAB_PANIC)
531		panic("Cannot create slab cache %s\n", name);
532
533	return c;
534}
535EXPORT_SYMBOL(kmem_cache_create);
536
537void kmem_cache_destroy(struct kmem_cache *c)
538{
539	slob_free(c, sizeof(struct kmem_cache));
540}
541EXPORT_SYMBOL(kmem_cache_destroy);
542
543void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
544{
545	void *b;
546
547	if (c->size < PAGE_SIZE)
548		b = slob_alloc(c->size, flags, c->align, node);
549	else
550		b = slob_new_page(flags, get_order(c->size), node);
551
552	if (c->ctor)
553		c->ctor(c, b);
554
555	return b;
556}
557EXPORT_SYMBOL(kmem_cache_alloc_node);
558
559static void __kmem_cache_free(void *b, int size)
560{
561	if (size < PAGE_SIZE)
562		slob_free(b, size);
563	else
564		free_pages((unsigned long)b, get_order(size));
565}
566
567static void kmem_rcu_free(struct rcu_head *head)
568{
569	struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
570	void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
571
572	__kmem_cache_free(b, slob_rcu->size);
573}
574
575void kmem_cache_free(struct kmem_cache *c, void *b)
576{
577	if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
578		struct slob_rcu *slob_rcu;
579		slob_rcu = b + (c->size - sizeof(struct slob_rcu));
580		INIT_RCU_HEAD(&slob_rcu->head);
581		slob_rcu->size = c->size;
582		call_rcu(&slob_rcu->head, kmem_rcu_free);
583	} else {
584		__kmem_cache_free(b, c->size);
585	}
586}
587EXPORT_SYMBOL(kmem_cache_free);
588
589unsigned int kmem_cache_size(struct kmem_cache *c)
590{
591	return c->size;
592}
593EXPORT_SYMBOL(kmem_cache_size);
594
595const char *kmem_cache_name(struct kmem_cache *c)
596{
597	return c->name;
598}
599EXPORT_SYMBOL(kmem_cache_name);
600
601int kmem_cache_shrink(struct kmem_cache *d)
602{
603	return 0;
604}
605EXPORT_SYMBOL(kmem_cache_shrink);
606
607int kmem_ptr_validate(struct kmem_cache *a, const void *b)
608{
609	return 0;
610}
611
612static unsigned int slob_ready __read_mostly;
613
614int slab_is_available(void)
615{
616	return slob_ready;
617}
618
619void __init kmem_cache_init(void)
620{
621	slob_ready = 1;
622}
623