blob: 1264799df5d1c119925a89ac55b7ea6e52e415bb [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 *
Matt Mackall20cecba2008-02-04 22:29:37 -080015 * 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.
20 *
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.
Matt Mackall10cef602006-01-08 01:01:45 -080026 *
27 * Above this is an implementation of kmalloc/kfree. Blocks returned
Nick Piggin55394842007-07-15 23:38:09 -070028 * from kmalloc are prepended with a 4-byte header with the kmalloc size.
Matt Mackall10cef602006-01-08 01:01:45 -080029 * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
Paul Mundt6193a2f2007-07-15 23:38:22 -070030 * alloc_pages() directly, allocating compound pages so the page order
Nick Piggind87a1332007-07-15 23:38:08 -070031 * 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()
34 * is false for them.
Matt Mackall10cef602006-01-08 01:01:45 -080035 *
36 * SLAB is emulated on top of SLOB by simply calling constructors and
Nick Piggin95b35122007-07-15 23:38:07 -070037 * 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
Paul Mundt6193a2f2007-07-15 23:38:22 -070041 * calling alloc_pages(). As SLAB objects know their size, no separate
Nick Piggin95b35122007-07-15 23:38:07 -070042 * size bookkeeping is necessary and there is essentially no allocation
Nick Piggind87a1332007-07-15 23:38:08 -070043 * space overhead, and compound pages aren't needed for multi-page
44 * allocations.
Paul Mundt6193a2f2007-07-15 23:38:22 -070045 *
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().
52 *
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.
Matt Mackall10cef602006-01-08 01:01:45 -080058 */
59
Nick Piggin95b35122007-07-15 23:38:07 -070060#include <linux/kernel.h>
Matt Mackall10cef602006-01-08 01:01:45 -080061#include <linux/slab.h>
62#include <linux/mm.h>
63#include <linux/cache.h>
64#include <linux/init.h>
65#include <linux/module.h>
Nick Pigginafc0ced2007-05-16 22:10:49 -070066#include <linux/rcupdate.h>
Nick Piggin95b35122007-07-15 23:38:07 -070067#include <linux/list.h>
68#include <asm/atomic.h>
Matt Mackall10cef602006-01-08 01:01:45 -080069
Nick Piggin95b35122007-07-15 23:38:07 -070070/*
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).
73 *
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.
77 */
Nick Piggin55394842007-07-15 23:38:09 -070078#if PAGE_SIZE <= (32767 * 2)
Nick Piggin95b35122007-07-15 23:38:07 -070079typedef s16 slobidx_t;
80#else
81typedef s32 slobidx_t;
82#endif
83
Matt Mackall10cef602006-01-08 01:01:45 -080084struct slob_block {
Nick Piggin95b35122007-07-15 23:38:07 -070085 slobidx_t units;
Nick Piggin55394842007-07-15 23:38:09 -070086};
Matt Mackall10cef602006-01-08 01:01:45 -080087typedef struct slob_block slob_t;
88
Nick Piggin95b35122007-07-15 23:38:07 -070089/*
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.
93 */
94struct slob_page {
95 union {
96 struct {
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 */
103 };
104 struct page page;
105 };
106};
107static inline void struct_slob_page_wrong_size(void)
108{ BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); }
109
110/*
111 * free_slob_page: call before a slob_page is returned to the page allocator.
112 */
113static inline void free_slob_page(struct slob_page *sp)
114{
115 reset_page_mapcount(&sp->page);
116 sp->page.mapping = NULL;
117}
118
119/*
Matt Mackall20cecba2008-02-04 22:29:37 -0800120 * All partially free slob pages go on these lists.
Nick Piggin95b35122007-07-15 23:38:07 -0700121 */
Matt Mackall20cecba2008-02-04 22:29:37 -0800122#define SLOB_BREAK1 256
123#define SLOB_BREAK2 1024
124static LIST_HEAD(free_slob_small);
125static LIST_HEAD(free_slob_medium);
126static LIST_HEAD(free_slob_large);
Nick Piggin95b35122007-07-15 23:38:07 -0700127
128/*
129 * slob_page: True for all slob pages (false for bigblock pages)
130 */
131static inline int slob_page(struct slob_page *sp)
132{
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700133 return PageSlobPage((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700134}
135
136static inline void set_slob_page(struct slob_page *sp)
137{
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700138 __SetPageSlobPage((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700139}
140
141static inline void clear_slob_page(struct slob_page *sp)
142{
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700143 __ClearPageSlobPage((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700144}
145
146/*
147 * slob_page_free: true for pages on free_slob_pages list.
148 */
149static inline int slob_page_free(struct slob_page *sp)
150{
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700151 return PageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700152}
153
Matt Mackall20cecba2008-02-04 22:29:37 -0800154static void set_slob_page_free(struct slob_page *sp, struct list_head *list)
Nick Piggin95b35122007-07-15 23:38:07 -0700155{
Matt Mackall20cecba2008-02-04 22:29:37 -0800156 list_add(&sp->list, list);
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700157 __SetPageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700158}
159
160static inline void clear_slob_page_free(struct slob_page *sp)
161{
162 list_del(&sp->list);
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700163 __ClearPageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700164}
165
Matt Mackall10cef602006-01-08 01:01:45 -0800166#define SLOB_UNIT sizeof(slob_t)
167#define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT)
168#define SLOB_ALIGN L1_CACHE_BYTES
169
Nick Pigginafc0ced2007-05-16 22:10:49 -0700170/*
171 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
172 * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free
173 * the block using call_rcu.
174 */
175struct slob_rcu {
176 struct rcu_head head;
177 int size;
178};
179
Nick Piggin95b35122007-07-15 23:38:07 -0700180/*
181 * slob_lock protects all slob allocator structures.
182 */
Matt Mackall10cef602006-01-08 01:01:45 -0800183static DEFINE_SPINLOCK(slob_lock);
Matt Mackall10cef602006-01-08 01:01:45 -0800184
Nick Piggin95b35122007-07-15 23:38:07 -0700185/*
186 * Encode the given size and next info into a free slob block s.
187 */
188static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
189{
190 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
191 slobidx_t offset = next - base;
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800192
Nick Piggin95b35122007-07-15 23:38:07 -0700193 if (size > 1) {
194 s[0].units = size;
195 s[1].units = offset;
196 } else
197 s[0].units = -offset;
198}
Matt Mackall10cef602006-01-08 01:01:45 -0800199
Nick Piggin95b35122007-07-15 23:38:07 -0700200/*
201 * Return the size of a slob block.
202 */
203static slobidx_t slob_units(slob_t *s)
204{
205 if (s->units > 0)
206 return s->units;
207 return 1;
208}
209
210/*
211 * Return the next free slob block pointer after this one.
212 */
213static slob_t *slob_next(slob_t *s)
214{
215 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
216 slobidx_t next;
217
218 if (s[0].units < 0)
219 next = -s[0].units;
220 else
221 next = s[1].units;
222 return base+next;
223}
224
225/*
226 * Returns true if s is the last free block in its page.
227 */
228static int slob_last(slob_t *s)
229{
230 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
231}
232
Paul Mundt6193a2f2007-07-15 23:38:22 -0700233static void *slob_new_page(gfp_t gfp, int order, int node)
234{
235 void *page;
236
237#ifdef CONFIG_NUMA
238 if (node != -1)
239 page = alloc_pages_node(node, gfp, order);
240 else
241#endif
242 page = alloc_pages(gfp, order);
243
244 if (!page)
245 return NULL;
246
247 return page_address(page);
248}
249
Nick Piggin95b35122007-07-15 23:38:07 -0700250/*
251 * Allocate a slob block within a given slob_page sp.
252 */
253static void *slob_page_alloc(struct slob_page *sp, size_t size, int align)
Matt Mackall10cef602006-01-08 01:01:45 -0800254{
255 slob_t *prev, *cur, *aligned = 0;
256 int delta = 0, units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800257
Nick Piggin95b35122007-07-15 23:38:07 -0700258 for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
259 slobidx_t avail = slob_units(cur);
260
Matt Mackall10cef602006-01-08 01:01:45 -0800261 if (align) {
262 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
263 delta = aligned - cur;
264 }
Nick Piggin95b35122007-07-15 23:38:07 -0700265 if (avail >= units + delta) { /* room enough? */
266 slob_t *next;
267
Matt Mackall10cef602006-01-08 01:01:45 -0800268 if (delta) { /* need to fragment head to align? */
Nick Piggin95b35122007-07-15 23:38:07 -0700269 next = slob_next(cur);
270 set_slob(aligned, avail - delta, next);
271 set_slob(cur, delta, aligned);
Matt Mackall10cef602006-01-08 01:01:45 -0800272 prev = cur;
273 cur = aligned;
Nick Piggin95b35122007-07-15 23:38:07 -0700274 avail = slob_units(cur);
Matt Mackall10cef602006-01-08 01:01:45 -0800275 }
276
Nick Piggin95b35122007-07-15 23:38:07 -0700277 next = slob_next(cur);
278 if (avail == units) { /* exact fit? unlink. */
279 if (prev)
280 set_slob(prev, slob_units(prev), next);
281 else
282 sp->free = next;
283 } else { /* fragment */
284 if (prev)
285 set_slob(prev, slob_units(prev), cur + units);
286 else
287 sp->free = cur + units;
288 set_slob(cur + units, avail - units, next);
Matt Mackall10cef602006-01-08 01:01:45 -0800289 }
290
Nick Piggin95b35122007-07-15 23:38:07 -0700291 sp->units -= units;
292 if (!sp->units)
293 clear_slob_page_free(sp);
Matt Mackall10cef602006-01-08 01:01:45 -0800294 return cur;
295 }
Nick Piggin95b35122007-07-15 23:38:07 -0700296 if (slob_last(cur))
297 return NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800298 }
299}
300
Nick Piggin95b35122007-07-15 23:38:07 -0700301/*
302 * slob_alloc: entry point into the slob allocator.
303 */
Paul Mundt6193a2f2007-07-15 23:38:22 -0700304static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
Nick Piggin95b35122007-07-15 23:38:07 -0700305{
306 struct slob_page *sp;
Matt Mackalld6269542007-07-21 04:37:40 -0700307 struct list_head *prev;
Matt Mackall20cecba2008-02-04 22:29:37 -0800308 struct list_head *slob_list;
Nick Piggin95b35122007-07-15 23:38:07 -0700309 slob_t *b = NULL;
310 unsigned long flags;
311
Matt Mackall20cecba2008-02-04 22:29:37 -0800312 if (size < SLOB_BREAK1)
313 slob_list = &free_slob_small;
314 else if (size < SLOB_BREAK2)
315 slob_list = &free_slob_medium;
316 else
317 slob_list = &free_slob_large;
318
Nick Piggin95b35122007-07-15 23:38:07 -0700319 spin_lock_irqsave(&slob_lock, flags);
320 /* Iterate through each partially free page, try to find room */
Matt Mackall20cecba2008-02-04 22:29:37 -0800321 list_for_each_entry(sp, slob_list, list) {
Paul Mundt6193a2f2007-07-15 23:38:22 -0700322#ifdef CONFIG_NUMA
323 /*
324 * If there's a node specification, search for a partial
325 * page with a matching node id in the freelist.
326 */
327 if (node != -1 && page_to_nid(&sp->page) != node)
328 continue;
329#endif
Matt Mackalld6269542007-07-21 04:37:40 -0700330 /* Enough room on this page? */
331 if (sp->units < SLOB_UNITS(size))
332 continue;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700333
Matt Mackalld6269542007-07-21 04:37:40 -0700334 /* Attempt to alloc */
335 prev = sp->list.prev;
336 b = slob_page_alloc(sp, size, align);
337 if (!b)
338 continue;
339
340 /* Improve fragment distribution and reduce our average
341 * search time by starting our next search here. (see
342 * Knuth vol 1, sec 2.5, pg 449) */
Matt Mackall20cecba2008-02-04 22:29:37 -0800343 if (prev != slob_list->prev &&
344 slob_list->next != prev->next)
345 list_move_tail(slob_list, prev->next);
Matt Mackalld6269542007-07-21 04:37:40 -0700346 break;
Nick Piggin95b35122007-07-15 23:38:07 -0700347 }
348 spin_unlock_irqrestore(&slob_lock, flags);
349
350 /* Not enough space: must allocate a new page */
351 if (!b) {
Linus Torvalds7fd27252007-12-09 10:14:36 -0800352 b = slob_new_page(gfp & ~__GFP_ZERO, 0, node);
Nick Piggin95b35122007-07-15 23:38:07 -0700353 if (!b)
354 return 0;
355 sp = (struct slob_page *)virt_to_page(b);
356 set_slob_page(sp);
357
358 spin_lock_irqsave(&slob_lock, flags);
359 sp->units = SLOB_UNITS(PAGE_SIZE);
360 sp->free = b;
361 INIT_LIST_HEAD(&sp->list);
362 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
Matt Mackall20cecba2008-02-04 22:29:37 -0800363 set_slob_page_free(sp, slob_list);
Nick Piggin95b35122007-07-15 23:38:07 -0700364 b = slob_page_alloc(sp, size, align);
365 BUG_ON(!b);
366 spin_unlock_irqrestore(&slob_lock, flags);
367 }
Christoph Lameterd07dbea2007-07-17 04:03:23 -0700368 if (unlikely((gfp & __GFP_ZERO) && b))
369 memset(b, 0, size);
Nick Piggin95b35122007-07-15 23:38:07 -0700370 return b;
371}
372
373/*
374 * slob_free: entry point into the slob allocator.
375 */
Matt Mackall10cef602006-01-08 01:01:45 -0800376static void slob_free(void *block, int size)
377{
Nick Piggin95b35122007-07-15 23:38:07 -0700378 struct slob_page *sp;
379 slob_t *prev, *next, *b = (slob_t *)block;
380 slobidx_t units;
Matt Mackall10cef602006-01-08 01:01:45 -0800381 unsigned long flags;
382
Satyam Sharma2408c552007-10-16 01:24:44 -0700383 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800384 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700385 BUG_ON(!size);
Matt Mackall10cef602006-01-08 01:01:45 -0800386
Nick Piggin95b35122007-07-15 23:38:07 -0700387 sp = (struct slob_page *)virt_to_page(block);
388 units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800389
Matt Mackall10cef602006-01-08 01:01:45 -0800390 spin_lock_irqsave(&slob_lock, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800391
Nick Piggin95b35122007-07-15 23:38:07 -0700392 if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
393 /* Go directly to page allocator. Do not pass slob allocator */
394 if (slob_page_free(sp))
395 clear_slob_page_free(sp);
396 clear_slob_page(sp);
397 free_slob_page(sp);
398 free_page((unsigned long)b);
399 goto out;
400 }
Matt Mackall10cef602006-01-08 01:01:45 -0800401
Nick Piggin95b35122007-07-15 23:38:07 -0700402 if (!slob_page_free(sp)) {
403 /* This slob page is about to become partially free. Easy! */
404 sp->units = units;
405 sp->free = b;
406 set_slob(b, units,
407 (void *)((unsigned long)(b +
408 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
Matt Mackall20cecba2008-02-04 22:29:37 -0800409 set_slob_page_free(sp, &free_slob_small);
Nick Piggin95b35122007-07-15 23:38:07 -0700410 goto out;
411 }
Matt Mackall10cef602006-01-08 01:01:45 -0800412
Nick Piggin95b35122007-07-15 23:38:07 -0700413 /*
414 * Otherwise the page is already partially free, so find reinsertion
415 * point.
416 */
417 sp->units += units;
Matt Mackall10cef602006-01-08 01:01:45 -0800418
Nick Piggin95b35122007-07-15 23:38:07 -0700419 if (b < sp->free) {
Matt Mackall679299b2008-02-04 22:29:37 -0800420 if (b + units == sp->free) {
421 units += slob_units(sp->free);
422 sp->free = slob_next(sp->free);
423 }
Nick Piggin95b35122007-07-15 23:38:07 -0700424 set_slob(b, units, sp->free);
425 sp->free = b;
426 } else {
427 prev = sp->free;
428 next = slob_next(prev);
429 while (b > next) {
430 prev = next;
431 next = slob_next(prev);
432 }
433
434 if (!slob_last(prev) && b + units == next) {
435 units += slob_units(next);
436 set_slob(b, units, slob_next(next));
437 } else
438 set_slob(b, units, next);
439
440 if (prev + slob_units(prev) == b) {
441 units = slob_units(b) + slob_units(prev);
442 set_slob(prev, units, slob_next(b));
443 } else
444 set_slob(prev, slob_units(prev), b);
445 }
446out:
Matt Mackall10cef602006-01-08 01:01:45 -0800447 spin_unlock_irqrestore(&slob_lock, flags);
448}
449
Nick Piggin95b35122007-07-15 23:38:07 -0700450/*
451 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
452 */
453
Nick Piggin55394842007-07-15 23:38:09 -0700454#ifndef ARCH_KMALLOC_MINALIGN
455#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long)
456#endif
457
458#ifndef ARCH_SLAB_MINALIGN
459#define ARCH_SLAB_MINALIGN __alignof__(unsigned long)
460#endif
461
Paul Mundt6193a2f2007-07-15 23:38:22 -0700462void *__kmalloc_node(size_t size, gfp_t gfp, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800463{
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700464 unsigned int *m;
Nick Piggin55394842007-07-15 23:38:09 -0700465 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
466
Nick Piggincf40bd12009-01-21 08:12:39 +0100467 lockdep_trace_alloc(flags);
468
Nick Piggin55394842007-07-15 23:38:09 -0700469 if (size < PAGE_SIZE - align) {
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700470 if (!size)
471 return ZERO_SIZE_PTR;
472
Paul Mundt6193a2f2007-07-15 23:38:22 -0700473 m = slob_alloc(size + align, gfp, align, node);
MinChan Kim239f49c2008-05-19 22:12:08 +0900474 if (!m)
475 return NULL;
476 *m = size;
Nick Piggin55394842007-07-15 23:38:09 -0700477 return (void *)m + align;
Nick Piggind87a1332007-07-15 23:38:08 -0700478 } else {
479 void *ret;
480
Paul Mundt6193a2f2007-07-15 23:38:22 -0700481 ret = slob_new_page(gfp | __GFP_COMP, get_order(size), node);
Nick Piggind87a1332007-07-15 23:38:08 -0700482 if (ret) {
483 struct page *page;
484 page = virt_to_page(ret);
485 page->private = size;
486 }
487 return ret;
Matt Mackall10cef602006-01-08 01:01:45 -0800488 }
Matt Mackall10cef602006-01-08 01:01:45 -0800489}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700490EXPORT_SYMBOL(__kmalloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800491
492void kfree(const void *block)
493{
Nick Piggin95b35122007-07-15 23:38:07 -0700494 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800495
Satyam Sharma2408c552007-10-16 01:24:44 -0700496 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800497 return;
498
Nick Piggin95b35122007-07-15 23:38:07 -0700499 sp = (struct slob_page *)virt_to_page(block);
Nick Piggind87a1332007-07-15 23:38:08 -0700500 if (slob_page(sp)) {
Nick Piggin55394842007-07-15 23:38:09 -0700501 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
502 unsigned int *m = (unsigned int *)(block - align);
503 slob_free(m, *m + align);
Nick Piggind87a1332007-07-15 23:38:08 -0700504 } else
505 put_page(&sp->page);
Matt Mackall10cef602006-01-08 01:01:45 -0800506}
Matt Mackall10cef602006-01-08 01:01:45 -0800507EXPORT_SYMBOL(kfree);
508
Nick Piggind87a1332007-07-15 23:38:08 -0700509/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
Pekka Enbergfd76bab2007-05-06 14:48:40 -0700510size_t ksize(const void *block)
Matt Mackall10cef602006-01-08 01:01:45 -0800511{
Nick Piggin95b35122007-07-15 23:38:07 -0700512 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800513
Christoph Lameteref8b4522007-10-16 01:24:46 -0700514 BUG_ON(!block);
515 if (unlikely(block == ZERO_SIZE_PTR))
Matt Mackall10cef602006-01-08 01:01:45 -0800516 return 0;
517
Nick Piggin95b35122007-07-15 23:38:07 -0700518 sp = (struct slob_page *)virt_to_page(block);
Matt Mackall70096a52008-10-08 14:51:57 -0500519 if (slob_page(sp)) {
520 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
521 unsigned int *m = (unsigned int *)(block - align);
522 return SLOB_UNITS(*m) * SLOB_UNIT;
523 } else
Nick Piggind87a1332007-07-15 23:38:08 -0700524 return sp->page.private;
Matt Mackall10cef602006-01-08 01:01:45 -0800525}
526
527struct kmem_cache {
528 unsigned int size, align;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700529 unsigned long flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800530 const char *name;
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700531 void (*ctor)(void *);
Matt Mackall10cef602006-01-08 01:01:45 -0800532};
533
534struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700535 size_t align, unsigned long flags, void (*ctor)(void *))
Matt Mackall10cef602006-01-08 01:01:45 -0800536{
537 struct kmem_cache *c;
538
Yi Li0701a9e2008-04-25 19:49:21 +0300539 c = slob_alloc(sizeof(struct kmem_cache),
Catalin Marinas5e18e2b2008-12-15 13:54:16 -0800540 GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
Matt Mackall10cef602006-01-08 01:01:45 -0800541
542 if (c) {
543 c->name = name;
544 c->size = size;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700545 if (flags & SLAB_DESTROY_BY_RCU) {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700546 /* leave room for rcu footer at the end of object */
547 c->size += sizeof(struct slob_rcu);
548 }
549 c->flags = flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800550 c->ctor = ctor;
Matt Mackall10cef602006-01-08 01:01:45 -0800551 /* ignore alignment unless it's forced */
Christoph Lameter5af60832007-05-06 14:49:56 -0700552 c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
Nick Piggin55394842007-07-15 23:38:09 -0700553 if (c->align < ARCH_SLAB_MINALIGN)
554 c->align = ARCH_SLAB_MINALIGN;
Matt Mackall10cef602006-01-08 01:01:45 -0800555 if (c->align < align)
556 c->align = align;
Akinobu Mitabc0055a2007-05-06 14:49:52 -0700557 } else if (flags & SLAB_PANIC)
558 panic("Cannot create slab cache %s\n", name);
Matt Mackall10cef602006-01-08 01:01:45 -0800559
560 return c;
561}
562EXPORT_SYMBOL(kmem_cache_create);
563
Alexey Dobriyan133d2052006-09-27 01:49:41 -0700564void kmem_cache_destroy(struct kmem_cache *c)
Matt Mackall10cef602006-01-08 01:01:45 -0800565{
566 slob_free(c, sizeof(struct kmem_cache));
Matt Mackall10cef602006-01-08 01:01:45 -0800567}
568EXPORT_SYMBOL(kmem_cache_destroy);
569
Paul Mundt6193a2f2007-07-15 23:38:22 -0700570void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800571{
572 void *b;
573
574 if (c->size < PAGE_SIZE)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700575 b = slob_alloc(c->size, flags, c->align, node);
Matt Mackall10cef602006-01-08 01:01:45 -0800576 else
Paul Mundt6193a2f2007-07-15 23:38:22 -0700577 b = slob_new_page(flags, get_order(c->size), node);
Matt Mackall10cef602006-01-08 01:01:45 -0800578
579 if (c->ctor)
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700580 c->ctor(b);
Matt Mackall10cef602006-01-08 01:01:45 -0800581
582 return b;
583}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700584EXPORT_SYMBOL(kmem_cache_alloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800585
Nick Pigginafc0ced2007-05-16 22:10:49 -0700586static void __kmem_cache_free(void *b, int size)
587{
588 if (size < PAGE_SIZE)
589 slob_free(b, size);
590 else
591 free_pages((unsigned long)b, get_order(size));
592}
593
594static void kmem_rcu_free(struct rcu_head *head)
595{
596 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
597 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
598
599 __kmem_cache_free(b, slob_rcu->size);
600}
601
Matt Mackall10cef602006-01-08 01:01:45 -0800602void kmem_cache_free(struct kmem_cache *c, void *b)
603{
Nick Pigginafc0ced2007-05-16 22:10:49 -0700604 if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
605 struct slob_rcu *slob_rcu;
606 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
607 INIT_RCU_HEAD(&slob_rcu->head);
608 slob_rcu->size = c->size;
609 call_rcu(&slob_rcu->head, kmem_rcu_free);
610 } else {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700611 __kmem_cache_free(b, c->size);
612 }
Matt Mackall10cef602006-01-08 01:01:45 -0800613}
614EXPORT_SYMBOL(kmem_cache_free);
615
616unsigned int kmem_cache_size(struct kmem_cache *c)
617{
618 return c->size;
619}
620EXPORT_SYMBOL(kmem_cache_size);
621
622const char *kmem_cache_name(struct kmem_cache *c)
623{
624 return c->name;
625}
626EXPORT_SYMBOL(kmem_cache_name);
627
Christoph Lameter2e892f42006-12-13 00:34:23 -0800628int kmem_cache_shrink(struct kmem_cache *d)
629{
630 return 0;
631}
632EXPORT_SYMBOL(kmem_cache_shrink);
633
Christoph Lameter55935a32006-12-13 00:34:24 -0800634int kmem_ptr_validate(struct kmem_cache *a, const void *b)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800635{
636 return 0;
637}
638
Paul Mundt84a01c22007-07-15 23:38:24 -0700639static unsigned int slob_ready __read_mostly;
640
641int slab_is_available(void)
642{
643 return slob_ready;
644}
645
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800646void __init kmem_cache_init(void)
647{
Paul Mundt84a01c22007-07-15 23:38:24 -0700648 slob_ready = 1;
Matt Mackall10cef602006-01-08 01:01:45 -0800649}