blob: 08a9bd91a1aaf517f1b17ff688fa9f612273f873 [file] [log] [blame]
Matt Mackall10cef602006-01-08 01:01:45 -08001/*
2 * SLOB Allocator: Simple List Of Blocks
3 *
4 * Matt Mackall <mpm@selenic.com> 12/30/03
5 *
Paul Mundt6193a2f2007-07-15 23:38:22 -07006 * NUMA support by Paul Mundt, 2007.
7 *
Matt Mackall10cef602006-01-08 01:01:45 -08008 * 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
Nick Piggin55394842007-07-15 23:38:09 -070012 * 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.
Nick Piggin95b35122007-07-15 23:38:07 -070014 *
Paul Mundt6193a2f2007-07-15 23:38:22 -070015 * The slob heap is a linked list of pages from alloc_pages(), and
Nick Piggin95b35122007-07-15 23:38:07 -070016 * 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.
Matt Mackall10cef602006-01-08 01:01:45 -080019 *
20 * Above this is an implementation of kmalloc/kfree. Blocks returned
Nick Piggin55394842007-07-15 23:38:09 -070021 * from kmalloc are prepended with a 4-byte header with the kmalloc size.
Matt Mackall10cef602006-01-08 01:01:45 -080022 * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
Paul Mundt6193a2f2007-07-15 23:38:22 -070023 * alloc_pages() directly, allocating compound pages so the page order
Nick Piggind87a1332007-07-15 23:38:08 -070024 * 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.
Matt Mackall10cef602006-01-08 01:01:45 -080028 *
29 * SLAB is emulated on top of SLOB by simply calling constructors and
Nick Piggin95b35122007-07-15 23:38:07 -070030 * 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
Paul Mundt6193a2f2007-07-15 23:38:22 -070034 * calling alloc_pages(). As SLAB objects know their size, no separate
Nick Piggin95b35122007-07-15 23:38:07 -070035 * size bookkeeping is necessary and there is essentially no allocation
Nick Piggind87a1332007-07-15 23:38:08 -070036 * space overhead, and compound pages aren't needed for multi-page
37 * allocations.
Paul Mundt6193a2f2007-07-15 23:38:22 -070038 *
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.
Matt Mackall10cef602006-01-08 01:01:45 -080051 */
52
Nick Piggin95b35122007-07-15 23:38:07 -070053#include <linux/kernel.h>
Matt Mackall10cef602006-01-08 01:01:45 -080054#include <linux/slab.h>
55#include <linux/mm.h>
56#include <linux/cache.h>
57#include <linux/init.h>
58#include <linux/module.h>
Nick Pigginafc0ced2007-05-16 22:10:49 -070059#include <linux/rcupdate.h>
Nick Piggin95b35122007-07-15 23:38:07 -070060#include <linux/list.h>
61#include <asm/atomic.h>
Matt Mackall10cef602006-01-08 01:01:45 -080062
Nick Piggin95b35122007-07-15 23:38:07 -070063/*
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 */
Nick Piggin55394842007-07-15 23:38:09 -070071#if PAGE_SIZE <= (32767 * 2)
Nick Piggin95b35122007-07-15 23:38:07 -070072typedef s16 slobidx_t;
73#else
74typedef s32 slobidx_t;
75#endif
76
Matt Mackall10cef602006-01-08 01:01:45 -080077struct slob_block {
Nick Piggin95b35122007-07-15 23:38:07 -070078 slobidx_t units;
Nick Piggin55394842007-07-15 23:38:09 -070079};
Matt Mackall10cef602006-01-08 01:01:45 -080080typedef struct slob_block slob_t;
81
Nick Piggin95b35122007-07-15 23:38:07 -070082/*
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
Matt Mackall10cef602006-01-08 01:01:45 -0800155#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
Nick Pigginafc0ced2007-05-16 22:10:49 -0700159/*
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
Nick Piggin95b35122007-07-15 23:38:07 -0700169/*
170 * slob_lock protects all slob allocator structures.
171 */
Matt Mackall10cef602006-01-08 01:01:45 -0800172static DEFINE_SPINLOCK(slob_lock);
Matt Mackall10cef602006-01-08 01:01:45 -0800173
Nick Piggin95b35122007-07-15 23:38:07 -0700174/*
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;
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800181
Nick Piggin95b35122007-07-15 23:38:07 -0700182 if (size > 1) {
183 s[0].units = size;
184 s[1].units = offset;
185 } else
186 s[0].units = -offset;
187}
Matt Mackall10cef602006-01-08 01:01:45 -0800188
Nick Piggin95b35122007-07-15 23:38:07 -0700189/*
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
Paul Mundt6193a2f2007-07-15 23:38:22 -0700222static 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
Nick Piggin95b35122007-07-15 23:38:07 -0700239/*
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)
Matt Mackall10cef602006-01-08 01:01:45 -0800243{
244 slob_t *prev, *cur, *aligned = 0;
245 int delta = 0, units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800246
Nick Piggin95b35122007-07-15 23:38:07 -0700247 for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
248 slobidx_t avail = slob_units(cur);
249
Matt Mackall10cef602006-01-08 01:01:45 -0800250 if (align) {
251 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
252 delta = aligned - cur;
253 }
Nick Piggin95b35122007-07-15 23:38:07 -0700254 if (avail >= units + delta) { /* room enough? */
255 slob_t *next;
256
Matt Mackall10cef602006-01-08 01:01:45 -0800257 if (delta) { /* need to fragment head to align? */
Nick Piggin95b35122007-07-15 23:38:07 -0700258 next = slob_next(cur);
259 set_slob(aligned, avail - delta, next);
260 set_slob(cur, delta, aligned);
Matt Mackall10cef602006-01-08 01:01:45 -0800261 prev = cur;
262 cur = aligned;
Nick Piggin95b35122007-07-15 23:38:07 -0700263 avail = slob_units(cur);
Matt Mackall10cef602006-01-08 01:01:45 -0800264 }
265
Nick Piggin95b35122007-07-15 23:38:07 -0700266 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);
Matt Mackall10cef602006-01-08 01:01:45 -0800278 }
279
Nick Piggin95b35122007-07-15 23:38:07 -0700280 sp->units -= units;
281 if (!sp->units)
282 clear_slob_page_free(sp);
Matt Mackall10cef602006-01-08 01:01:45 -0800283 return cur;
284 }
Nick Piggin95b35122007-07-15 23:38:07 -0700285 if (slob_last(cur))
286 return NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800287 }
288}
289
Nick Piggin95b35122007-07-15 23:38:07 -0700290/*
291 * slob_alloc: entry point into the slob allocator.
292 */
Paul Mundt6193a2f2007-07-15 23:38:22 -0700293static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
Nick Piggin95b35122007-07-15 23:38:07 -0700294{
295 struct slob_page *sp;
Matt Mackalld6269542007-07-21 04:37:40 -0700296 struct list_head *prev;
Nick Piggin95b35122007-07-15 23:38:07 -0700297 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) {
Paul Mundt6193a2f2007-07-15 23:38:22 -0700303#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
Matt Mackalld6269542007-07-21 04:37:40 -0700311 /* Enough room on this page? */
312 if (sp->units < SLOB_UNITS(size))
313 continue;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700314
Matt Mackalld6269542007-07-21 04:37:40 -0700315 /* 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) */
Nick Piggind32ddd82007-11-15 12:32:04 +0100324 if (prev != free_slob_pages.prev &&
325 free_slob_pages.next != prev->next)
Matt Mackalld6269542007-07-21 04:37:40 -0700326 list_move_tail(&free_slob_pages, prev->next);
327 break;
Nick Piggin95b35122007-07-15 23:38:07 -0700328 }
329 spin_unlock_irqrestore(&slob_lock, flags);
330
331 /* Not enough space: must allocate a new page */
332 if (!b) {
Paul Mundt6193a2f2007-07-15 23:38:22 -0700333 b = slob_new_page(gfp, 0, node);
Nick Piggin95b35122007-07-15 23:38:07 -0700334 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 }
Christoph Lameterd07dbea2007-07-17 04:03:23 -0700349 if (unlikely((gfp & __GFP_ZERO) && b))
350 memset(b, 0, size);
Nick Piggin95b35122007-07-15 23:38:07 -0700351 return b;
352}
353
354/*
355 * slob_free: entry point into the slob allocator.
356 */
Matt Mackall10cef602006-01-08 01:01:45 -0800357static void slob_free(void *block, int size)
358{
Nick Piggin95b35122007-07-15 23:38:07 -0700359 struct slob_page *sp;
360 slob_t *prev, *next, *b = (slob_t *)block;
361 slobidx_t units;
Matt Mackall10cef602006-01-08 01:01:45 -0800362 unsigned long flags;
363
Satyam Sharma2408c552007-10-16 01:24:44 -0700364 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800365 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700366 BUG_ON(!size);
Matt Mackall10cef602006-01-08 01:01:45 -0800367
Nick Piggin95b35122007-07-15 23:38:07 -0700368 sp = (struct slob_page *)virt_to_page(block);
369 units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800370
Matt Mackall10cef602006-01-08 01:01:45 -0800371 spin_lock_irqsave(&slob_lock, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800372
Nick Piggin95b35122007-07-15 23:38:07 -0700373 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 }
Matt Mackall10cef602006-01-08 01:01:45 -0800382
Nick Piggin95b35122007-07-15 23:38:07 -0700383 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 }
Matt Mackall10cef602006-01-08 01:01:45 -0800393
Nick Piggin95b35122007-07-15 23:38:07 -0700394 /*
395 * Otherwise the page is already partially free, so find reinsertion
396 * point.
397 */
398 sp->units += units;
Matt Mackall10cef602006-01-08 01:01:45 -0800399
Nick Piggin95b35122007-07-15 23:38:07 -0700400 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:
Matt Mackall10cef602006-01-08 01:01:45 -0800424 spin_unlock_irqrestore(&slob_lock, flags);
425}
426
Nick Piggin95b35122007-07-15 23:38:07 -0700427/*
428 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
429 */
430
Nick Piggin55394842007-07-15 23:38:09 -0700431#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
Paul Mundt6193a2f2007-07-15 23:38:22 -0700439void *__kmalloc_node(size_t size, gfp_t gfp, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800440{
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700441 unsigned int *m;
Nick Piggin55394842007-07-15 23:38:09 -0700442 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
443
444 if (size < PAGE_SIZE - align) {
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700445 if (!size)
446 return ZERO_SIZE_PTR;
447
Paul Mundt6193a2f2007-07-15 23:38:22 -0700448 m = slob_alloc(size + align, gfp, align, node);
Nick Piggin95b35122007-07-15 23:38:07 -0700449 if (m)
Nick Piggin55394842007-07-15 23:38:09 -0700450 *m = size;
451 return (void *)m + align;
Nick Piggind87a1332007-07-15 23:38:08 -0700452 } else {
453 void *ret;
454
Paul Mundt6193a2f2007-07-15 23:38:22 -0700455 ret = slob_new_page(gfp | __GFP_COMP, get_order(size), node);
Nick Piggind87a1332007-07-15 23:38:08 -0700456 if (ret) {
457 struct page *page;
458 page = virt_to_page(ret);
459 page->private = size;
460 }
461 return ret;
Matt Mackall10cef602006-01-08 01:01:45 -0800462 }
Matt Mackall10cef602006-01-08 01:01:45 -0800463}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700464EXPORT_SYMBOL(__kmalloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800465
466void kfree(const void *block)
467{
Nick Piggin95b35122007-07-15 23:38:07 -0700468 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800469
Satyam Sharma2408c552007-10-16 01:24:44 -0700470 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800471 return;
472
Nick Piggin95b35122007-07-15 23:38:07 -0700473 sp = (struct slob_page *)virt_to_page(block);
Nick Piggind87a1332007-07-15 23:38:08 -0700474 if (slob_page(sp)) {
Nick Piggin55394842007-07-15 23:38:09 -0700475 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
476 unsigned int *m = (unsigned int *)(block - align);
477 slob_free(m, *m + align);
Nick Piggind87a1332007-07-15 23:38:08 -0700478 } else
479 put_page(&sp->page);
Matt Mackall10cef602006-01-08 01:01:45 -0800480}
Matt Mackall10cef602006-01-08 01:01:45 -0800481EXPORT_SYMBOL(kfree);
482
Nick Piggind87a1332007-07-15 23:38:08 -0700483/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
Pekka Enbergfd76bab2007-05-06 14:48:40 -0700484size_t ksize(const void *block)
Matt Mackall10cef602006-01-08 01:01:45 -0800485{
Nick Piggin95b35122007-07-15 23:38:07 -0700486 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800487
Christoph Lameteref8b4522007-10-16 01:24:46 -0700488 BUG_ON(!block);
489 if (unlikely(block == ZERO_SIZE_PTR))
Matt Mackall10cef602006-01-08 01:01:45 -0800490 return 0;
491
Nick Piggin95b35122007-07-15 23:38:07 -0700492 sp = (struct slob_page *)virt_to_page(block);
Nick Piggind87a1332007-07-15 23:38:08 -0700493 if (slob_page(sp))
494 return ((slob_t *)block - 1)->units + SLOB_UNIT;
495 else
496 return sp->page.private;
Matt Mackall10cef602006-01-08 01:01:45 -0800497}
498
499struct kmem_cache {
500 unsigned int size, align;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700501 unsigned long flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800502 const char *name;
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700503 void (*ctor)(struct kmem_cache *, void *);
Matt Mackall10cef602006-01-08 01:01:45 -0800504};
505
506struct kmem_cache *kmem_cache_create(const char *name, size_t size,
507 size_t align, unsigned long flags,
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700508 void (*ctor)(struct kmem_cache *, void *))
Matt Mackall10cef602006-01-08 01:01:45 -0800509{
510 struct kmem_cache *c;
511
Paul Mundt6193a2f2007-07-15 23:38:22 -0700512 c = slob_alloc(sizeof(struct kmem_cache), flags, 0, -1);
Matt Mackall10cef602006-01-08 01:01:45 -0800513
514 if (c) {
515 c->name = name;
516 c->size = size;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700517 if (flags & SLAB_DESTROY_BY_RCU) {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700518 /* leave room for rcu footer at the end of object */
519 c->size += sizeof(struct slob_rcu);
520 }
521 c->flags = flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800522 c->ctor = ctor;
Matt Mackall10cef602006-01-08 01:01:45 -0800523 /* ignore alignment unless it's forced */
Christoph Lameter5af60832007-05-06 14:49:56 -0700524 c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
Nick Piggin55394842007-07-15 23:38:09 -0700525 if (c->align < ARCH_SLAB_MINALIGN)
526 c->align = ARCH_SLAB_MINALIGN;
Matt Mackall10cef602006-01-08 01:01:45 -0800527 if (c->align < align)
528 c->align = align;
Akinobu Mitabc0055a2007-05-06 14:49:52 -0700529 } else if (flags & SLAB_PANIC)
530 panic("Cannot create slab cache %s\n", name);
Matt Mackall10cef602006-01-08 01:01:45 -0800531
532 return c;
533}
534EXPORT_SYMBOL(kmem_cache_create);
535
Alexey Dobriyan133d2052006-09-27 01:49:41 -0700536void kmem_cache_destroy(struct kmem_cache *c)
Matt Mackall10cef602006-01-08 01:01:45 -0800537{
538 slob_free(c, sizeof(struct kmem_cache));
Matt Mackall10cef602006-01-08 01:01:45 -0800539}
540EXPORT_SYMBOL(kmem_cache_destroy);
541
Paul Mundt6193a2f2007-07-15 23:38:22 -0700542void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800543{
544 void *b;
545
546 if (c->size < PAGE_SIZE)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700547 b = slob_alloc(c->size, flags, c->align, node);
Matt Mackall10cef602006-01-08 01:01:45 -0800548 else
Paul Mundt6193a2f2007-07-15 23:38:22 -0700549 b = slob_new_page(flags, get_order(c->size), node);
Matt Mackall10cef602006-01-08 01:01:45 -0800550
551 if (c->ctor)
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700552 c->ctor(c, b);
Matt Mackall10cef602006-01-08 01:01:45 -0800553
554 return b;
555}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700556EXPORT_SYMBOL(kmem_cache_alloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800557
Nick Pigginafc0ced2007-05-16 22:10:49 -0700558static void __kmem_cache_free(void *b, int size)
559{
560 if (size < PAGE_SIZE)
561 slob_free(b, size);
562 else
563 free_pages((unsigned long)b, get_order(size));
564}
565
566static void kmem_rcu_free(struct rcu_head *head)
567{
568 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
569 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
570
571 __kmem_cache_free(b, slob_rcu->size);
572}
573
Matt Mackall10cef602006-01-08 01:01:45 -0800574void kmem_cache_free(struct kmem_cache *c, void *b)
575{
Nick Pigginafc0ced2007-05-16 22:10:49 -0700576 if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
577 struct slob_rcu *slob_rcu;
578 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
579 INIT_RCU_HEAD(&slob_rcu->head);
580 slob_rcu->size = c->size;
581 call_rcu(&slob_rcu->head, kmem_rcu_free);
582 } else {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700583 __kmem_cache_free(b, c->size);
584 }
Matt Mackall10cef602006-01-08 01:01:45 -0800585}
586EXPORT_SYMBOL(kmem_cache_free);
587
588unsigned int kmem_cache_size(struct kmem_cache *c)
589{
590 return c->size;
591}
592EXPORT_SYMBOL(kmem_cache_size);
593
594const char *kmem_cache_name(struct kmem_cache *c)
595{
596 return c->name;
597}
598EXPORT_SYMBOL(kmem_cache_name);
599
Christoph Lameter2e892f42006-12-13 00:34:23 -0800600int kmem_cache_shrink(struct kmem_cache *d)
601{
602 return 0;
603}
604EXPORT_SYMBOL(kmem_cache_shrink);
605
Christoph Lameter55935a32006-12-13 00:34:24 -0800606int kmem_ptr_validate(struct kmem_cache *a, const void *b)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800607{
608 return 0;
609}
610
Paul Mundt84a01c22007-07-15 23:38:24 -0700611static unsigned int slob_ready __read_mostly;
612
613int slab_is_available(void)
614{
615 return slob_ready;
616}
617
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800618void __init kmem_cache_init(void)
619{
Paul Mundt84a01c22007-07-15 23:38:24 -0700620 slob_ready = 1;
Matt Mackall10cef602006-01-08 01:01:45 -0800621}