blob: 0bfa680a8981127307d1d922231dca3d66f9b2bd [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/*
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800129 * is_slob_page: True for all slob pages (false for bigblock pages)
Nick Piggin95b35122007-07-15 23:38:07 -0700130 */
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800131static inline int is_slob_page(struct slob_page *sp)
Nick Piggin95b35122007-07-15 23:38:07 -0700132{
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
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800146static inline struct slob_page *slob_page(const void *addr)
147{
148 return (struct slob_page *)virt_to_page(addr);
149}
150
Nick Piggin95b35122007-07-15 23:38:07 -0700151/*
152 * slob_page_free: true for pages on free_slob_pages list.
153 */
154static inline int slob_page_free(struct slob_page *sp)
155{
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700156 return PageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700157}
158
Matt Mackall20cecba2008-02-04 22:29:37 -0800159static void set_slob_page_free(struct slob_page *sp, struct list_head *list)
Nick Piggin95b35122007-07-15 23:38:07 -0700160{
Matt Mackall20cecba2008-02-04 22:29:37 -0800161 list_add(&sp->list, list);
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700162 __SetPageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700163}
164
165static inline void clear_slob_page_free(struct slob_page *sp)
166{
167 list_del(&sp->list);
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700168 __ClearPageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700169}
170
Matt Mackall10cef602006-01-08 01:01:45 -0800171#define SLOB_UNIT sizeof(slob_t)
172#define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT)
173#define SLOB_ALIGN L1_CACHE_BYTES
174
Nick Pigginafc0ced2007-05-16 22:10:49 -0700175/*
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.
179 */
180struct slob_rcu {
181 struct rcu_head head;
182 int size;
183};
184
Nick Piggin95b35122007-07-15 23:38:07 -0700185/*
186 * slob_lock protects all slob allocator structures.
187 */
Matt Mackall10cef602006-01-08 01:01:45 -0800188static DEFINE_SPINLOCK(slob_lock);
Matt Mackall10cef602006-01-08 01:01:45 -0800189
Nick Piggin95b35122007-07-15 23:38:07 -0700190/*
191 * Encode the given size and next info into a free slob block s.
192 */
193static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
194{
195 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
196 slobidx_t offset = next - base;
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800197
Nick Piggin95b35122007-07-15 23:38:07 -0700198 if (size > 1) {
199 s[0].units = size;
200 s[1].units = offset;
201 } else
202 s[0].units = -offset;
203}
Matt Mackall10cef602006-01-08 01:01:45 -0800204
Nick Piggin95b35122007-07-15 23:38:07 -0700205/*
206 * Return the size of a slob block.
207 */
208static slobidx_t slob_units(slob_t *s)
209{
210 if (s->units > 0)
211 return s->units;
212 return 1;
213}
214
215/*
216 * Return the next free slob block pointer after this one.
217 */
218static slob_t *slob_next(slob_t *s)
219{
220 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
221 slobidx_t next;
222
223 if (s[0].units < 0)
224 next = -s[0].units;
225 else
226 next = s[1].units;
227 return base+next;
228}
229
230/*
231 * Returns true if s is the last free block in its page.
232 */
233static int slob_last(slob_t *s)
234{
235 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
236}
237
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800238static void *slob_new_pages(gfp_t gfp, int order, int node)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700239{
240 void *page;
241
242#ifdef CONFIG_NUMA
243 if (node != -1)
244 page = alloc_pages_node(node, gfp, order);
245 else
246#endif
247 page = alloc_pages(gfp, order);
248
249 if (!page)
250 return NULL;
251
252 return page_address(page);
253}
254
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800255static void slob_free_pages(void *b, int order)
256{
257 free_pages((unsigned long)b, order);
258}
259
Nick Piggin95b35122007-07-15 23:38:07 -0700260/*
261 * Allocate a slob block within a given slob_page sp.
262 */
263static void *slob_page_alloc(struct slob_page *sp, size_t size, int align)
Matt Mackall10cef602006-01-08 01:01:45 -0800264{
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800265 slob_t *prev, *cur, *aligned = NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800266 int delta = 0, units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800267
Nick Piggin95b35122007-07-15 23:38:07 -0700268 for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
269 slobidx_t avail = slob_units(cur);
270
Matt Mackall10cef602006-01-08 01:01:45 -0800271 if (align) {
272 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
273 delta = aligned - cur;
274 }
Nick Piggin95b35122007-07-15 23:38:07 -0700275 if (avail >= units + delta) { /* room enough? */
276 slob_t *next;
277
Matt Mackall10cef602006-01-08 01:01:45 -0800278 if (delta) { /* need to fragment head to align? */
Nick Piggin95b35122007-07-15 23:38:07 -0700279 next = slob_next(cur);
280 set_slob(aligned, avail - delta, next);
281 set_slob(cur, delta, aligned);
Matt Mackall10cef602006-01-08 01:01:45 -0800282 prev = cur;
283 cur = aligned;
Nick Piggin95b35122007-07-15 23:38:07 -0700284 avail = slob_units(cur);
Matt Mackall10cef602006-01-08 01:01:45 -0800285 }
286
Nick Piggin95b35122007-07-15 23:38:07 -0700287 next = slob_next(cur);
288 if (avail == units) { /* exact fit? unlink. */
289 if (prev)
290 set_slob(prev, slob_units(prev), next);
291 else
292 sp->free = next;
293 } else { /* fragment */
294 if (prev)
295 set_slob(prev, slob_units(prev), cur + units);
296 else
297 sp->free = cur + units;
298 set_slob(cur + units, avail - units, next);
Matt Mackall10cef602006-01-08 01:01:45 -0800299 }
300
Nick Piggin95b35122007-07-15 23:38:07 -0700301 sp->units -= units;
302 if (!sp->units)
303 clear_slob_page_free(sp);
Matt Mackall10cef602006-01-08 01:01:45 -0800304 return cur;
305 }
Nick Piggin95b35122007-07-15 23:38:07 -0700306 if (slob_last(cur))
307 return NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800308 }
309}
310
Nick Piggin95b35122007-07-15 23:38:07 -0700311/*
312 * slob_alloc: entry point into the slob allocator.
313 */
Paul Mundt6193a2f2007-07-15 23:38:22 -0700314static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
Nick Piggin95b35122007-07-15 23:38:07 -0700315{
316 struct slob_page *sp;
Matt Mackalld6269542007-07-21 04:37:40 -0700317 struct list_head *prev;
Matt Mackall20cecba2008-02-04 22:29:37 -0800318 struct list_head *slob_list;
Nick Piggin95b35122007-07-15 23:38:07 -0700319 slob_t *b = NULL;
320 unsigned long flags;
321
Matt Mackall20cecba2008-02-04 22:29:37 -0800322 if (size < SLOB_BREAK1)
323 slob_list = &free_slob_small;
324 else if (size < SLOB_BREAK2)
325 slob_list = &free_slob_medium;
326 else
327 slob_list = &free_slob_large;
328
Nick Piggin95b35122007-07-15 23:38:07 -0700329 spin_lock_irqsave(&slob_lock, flags);
330 /* Iterate through each partially free page, try to find room */
Matt Mackall20cecba2008-02-04 22:29:37 -0800331 list_for_each_entry(sp, slob_list, list) {
Paul Mundt6193a2f2007-07-15 23:38:22 -0700332#ifdef CONFIG_NUMA
333 /*
334 * If there's a node specification, search for a partial
335 * page with a matching node id in the freelist.
336 */
337 if (node != -1 && page_to_nid(&sp->page) != node)
338 continue;
339#endif
Matt Mackalld6269542007-07-21 04:37:40 -0700340 /* Enough room on this page? */
341 if (sp->units < SLOB_UNITS(size))
342 continue;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700343
Matt Mackalld6269542007-07-21 04:37:40 -0700344 /* Attempt to alloc */
345 prev = sp->list.prev;
346 b = slob_page_alloc(sp, size, align);
347 if (!b)
348 continue;
349
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) */
Matt Mackall20cecba2008-02-04 22:29:37 -0800353 if (prev != slob_list->prev &&
354 slob_list->next != prev->next)
355 list_move_tail(slob_list, prev->next);
Matt Mackalld6269542007-07-21 04:37:40 -0700356 break;
Nick Piggin95b35122007-07-15 23:38:07 -0700357 }
358 spin_unlock_irqrestore(&slob_lock, flags);
359
360 /* Not enough space: must allocate a new page */
361 if (!b) {
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800362 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
Nick Piggin95b35122007-07-15 23:38:07 -0700363 if (!b)
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800364 return NULL;
365 sp = slob_page(b);
Nick Piggin95b35122007-07-15 23:38:07 -0700366 set_slob_page(sp);
367
368 spin_lock_irqsave(&slob_lock, flags);
369 sp->units = SLOB_UNITS(PAGE_SIZE);
370 sp->free = b;
371 INIT_LIST_HEAD(&sp->list);
372 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
Matt Mackall20cecba2008-02-04 22:29:37 -0800373 set_slob_page_free(sp, slob_list);
Nick Piggin95b35122007-07-15 23:38:07 -0700374 b = slob_page_alloc(sp, size, align);
375 BUG_ON(!b);
376 spin_unlock_irqrestore(&slob_lock, flags);
377 }
Christoph Lameterd07dbea2007-07-17 04:03:23 -0700378 if (unlikely((gfp & __GFP_ZERO) && b))
379 memset(b, 0, size);
Nick Piggin95b35122007-07-15 23:38:07 -0700380 return b;
381}
382
383/*
384 * slob_free: entry point into the slob allocator.
385 */
Matt Mackall10cef602006-01-08 01:01:45 -0800386static void slob_free(void *block, int size)
387{
Nick Piggin95b35122007-07-15 23:38:07 -0700388 struct slob_page *sp;
389 slob_t *prev, *next, *b = (slob_t *)block;
390 slobidx_t units;
Matt Mackall10cef602006-01-08 01:01:45 -0800391 unsigned long flags;
392
Satyam Sharma2408c552007-10-16 01:24:44 -0700393 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800394 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700395 BUG_ON(!size);
Matt Mackall10cef602006-01-08 01:01:45 -0800396
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800397 sp = slob_page(block);
Nick Piggin95b35122007-07-15 23:38:07 -0700398 units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800399
Matt Mackall10cef602006-01-08 01:01:45 -0800400 spin_lock_irqsave(&slob_lock, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800401
Nick Piggin95b35122007-07-15 23:38:07 -0700402 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);
Nick Piggin6fb8f422009-03-16 21:00:28 +1100406 spin_unlock_irqrestore(&slob_lock, flags);
Nick Piggin95b35122007-07-15 23:38:07 -0700407 clear_slob_page(sp);
408 free_slob_page(sp);
409 free_page((unsigned long)b);
Nick Piggin6fb8f422009-03-16 21:00:28 +1100410 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700411 }
Matt Mackall10cef602006-01-08 01:01:45 -0800412
Nick Piggin95b35122007-07-15 23:38:07 -0700413 if (!slob_page_free(sp)) {
414 /* This slob page is about to become partially free. Easy! */
415 sp->units = units;
416 sp->free = b;
417 set_slob(b, units,
418 (void *)((unsigned long)(b +
419 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
Matt Mackall20cecba2008-02-04 22:29:37 -0800420 set_slob_page_free(sp, &free_slob_small);
Nick Piggin95b35122007-07-15 23:38:07 -0700421 goto out;
422 }
Matt Mackall10cef602006-01-08 01:01:45 -0800423
Nick Piggin95b35122007-07-15 23:38:07 -0700424 /*
425 * Otherwise the page is already partially free, so find reinsertion
426 * point.
427 */
428 sp->units += units;
Matt Mackall10cef602006-01-08 01:01:45 -0800429
Nick Piggin95b35122007-07-15 23:38:07 -0700430 if (b < sp->free) {
Matt Mackall679299b2008-02-04 22:29:37 -0800431 if (b + units == sp->free) {
432 units += slob_units(sp->free);
433 sp->free = slob_next(sp->free);
434 }
Nick Piggin95b35122007-07-15 23:38:07 -0700435 set_slob(b, units, sp->free);
436 sp->free = b;
437 } else {
438 prev = sp->free;
439 next = slob_next(prev);
440 while (b > next) {
441 prev = next;
442 next = slob_next(prev);
443 }
444
445 if (!slob_last(prev) && b + units == next) {
446 units += slob_units(next);
447 set_slob(b, units, slob_next(next));
448 } else
449 set_slob(b, units, next);
450
451 if (prev + slob_units(prev) == b) {
452 units = slob_units(b) + slob_units(prev);
453 set_slob(prev, units, slob_next(b));
454 } else
455 set_slob(prev, slob_units(prev), b);
456 }
457out:
Matt Mackall10cef602006-01-08 01:01:45 -0800458 spin_unlock_irqrestore(&slob_lock, flags);
459}
460
Nick Piggin95b35122007-07-15 23:38:07 -0700461/*
462 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
463 */
464
Nick Piggin55394842007-07-15 23:38:09 -0700465#ifndef ARCH_KMALLOC_MINALIGN
466#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long)
467#endif
468
469#ifndef ARCH_SLAB_MINALIGN
470#define ARCH_SLAB_MINALIGN __alignof__(unsigned long)
471#endif
472
Paul Mundt6193a2f2007-07-15 23:38:22 -0700473void *__kmalloc_node(size_t size, gfp_t gfp, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800474{
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700475 unsigned int *m;
Nick Piggin55394842007-07-15 23:38:09 -0700476 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
477
478 if (size < PAGE_SIZE - align) {
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700479 if (!size)
480 return ZERO_SIZE_PTR;
481
Paul Mundt6193a2f2007-07-15 23:38:22 -0700482 m = slob_alloc(size + align, gfp, align, node);
MinChan Kim239f49c2008-05-19 22:12:08 +0900483 if (!m)
484 return NULL;
485 *m = size;
Nick Piggin55394842007-07-15 23:38:09 -0700486 return (void *)m + align;
Nick Piggind87a1332007-07-15 23:38:08 -0700487 } else {
488 void *ret;
489
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800490 ret = slob_new_pages(gfp | __GFP_COMP, get_order(size), node);
Nick Piggind87a1332007-07-15 23:38:08 -0700491 if (ret) {
492 struct page *page;
493 page = virt_to_page(ret);
494 page->private = size;
495 }
496 return ret;
Matt Mackall10cef602006-01-08 01:01:45 -0800497 }
Matt Mackall10cef602006-01-08 01:01:45 -0800498}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700499EXPORT_SYMBOL(__kmalloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800500
501void kfree(const void *block)
502{
Nick Piggin95b35122007-07-15 23:38:07 -0700503 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800504
Satyam Sharma2408c552007-10-16 01:24:44 -0700505 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800506 return;
507
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800508 sp = slob_page(block);
509 if (is_slob_page(sp)) {
Nick Piggin55394842007-07-15 23:38:09 -0700510 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
511 unsigned int *m = (unsigned int *)(block - align);
512 slob_free(m, *m + align);
Nick Piggind87a1332007-07-15 23:38:08 -0700513 } else
514 put_page(&sp->page);
Matt Mackall10cef602006-01-08 01:01:45 -0800515}
Matt Mackall10cef602006-01-08 01:01:45 -0800516EXPORT_SYMBOL(kfree);
517
Nick Piggind87a1332007-07-15 23:38:08 -0700518/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
Pekka Enbergfd76bab2007-05-06 14:48:40 -0700519size_t ksize(const void *block)
Matt Mackall10cef602006-01-08 01:01:45 -0800520{
Nick Piggin95b35122007-07-15 23:38:07 -0700521 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800522
Christoph Lameteref8b4522007-10-16 01:24:46 -0700523 BUG_ON(!block);
524 if (unlikely(block == ZERO_SIZE_PTR))
Matt Mackall10cef602006-01-08 01:01:45 -0800525 return 0;
526
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800527 sp = slob_page(block);
528 if (is_slob_page(sp)) {
Matt Mackall70096a52008-10-08 14:51:57 -0500529 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
530 unsigned int *m = (unsigned int *)(block - align);
531 return SLOB_UNITS(*m) * SLOB_UNIT;
532 } else
Nick Piggind87a1332007-07-15 23:38:08 -0700533 return sp->page.private;
Matt Mackall10cef602006-01-08 01:01:45 -0800534}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +0200535EXPORT_SYMBOL(ksize);
Matt Mackall10cef602006-01-08 01:01:45 -0800536
537struct kmem_cache {
538 unsigned int size, align;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700539 unsigned long flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800540 const char *name;
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700541 void (*ctor)(void *);
Matt Mackall10cef602006-01-08 01:01:45 -0800542};
543
544struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700545 size_t align, unsigned long flags, void (*ctor)(void *))
Matt Mackall10cef602006-01-08 01:01:45 -0800546{
547 struct kmem_cache *c;
548
Yi Li0701a9e2008-04-25 19:49:21 +0300549 c = slob_alloc(sizeof(struct kmem_cache),
Catalin Marinas5e18e2b2008-12-15 13:54:16 -0800550 GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
Matt Mackall10cef602006-01-08 01:01:45 -0800551
552 if (c) {
553 c->name = name;
554 c->size = size;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700555 if (flags & SLAB_DESTROY_BY_RCU) {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700556 /* leave room for rcu footer at the end of object */
557 c->size += sizeof(struct slob_rcu);
558 }
559 c->flags = flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800560 c->ctor = ctor;
Matt Mackall10cef602006-01-08 01:01:45 -0800561 /* ignore alignment unless it's forced */
Christoph Lameter5af60832007-05-06 14:49:56 -0700562 c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
Nick Piggin55394842007-07-15 23:38:09 -0700563 if (c->align < ARCH_SLAB_MINALIGN)
564 c->align = ARCH_SLAB_MINALIGN;
Matt Mackall10cef602006-01-08 01:01:45 -0800565 if (c->align < align)
566 c->align = align;
Akinobu Mitabc0055a2007-05-06 14:49:52 -0700567 } else if (flags & SLAB_PANIC)
568 panic("Cannot create slab cache %s\n", name);
Matt Mackall10cef602006-01-08 01:01:45 -0800569
570 return c;
571}
572EXPORT_SYMBOL(kmem_cache_create);
573
Alexey Dobriyan133d2052006-09-27 01:49:41 -0700574void kmem_cache_destroy(struct kmem_cache *c)
Matt Mackall10cef602006-01-08 01:01:45 -0800575{
576 slob_free(c, sizeof(struct kmem_cache));
Matt Mackall10cef602006-01-08 01:01:45 -0800577}
578EXPORT_SYMBOL(kmem_cache_destroy);
579
Paul Mundt6193a2f2007-07-15 23:38:22 -0700580void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800581{
582 void *b;
583
584 if (c->size < PAGE_SIZE)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700585 b = slob_alloc(c->size, flags, c->align, node);
Matt Mackall10cef602006-01-08 01:01:45 -0800586 else
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800587 b = slob_new_pages(flags, get_order(c->size), node);
Matt Mackall10cef602006-01-08 01:01:45 -0800588
589 if (c->ctor)
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700590 c->ctor(b);
Matt Mackall10cef602006-01-08 01:01:45 -0800591
592 return b;
593}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700594EXPORT_SYMBOL(kmem_cache_alloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800595
Nick Pigginafc0ced2007-05-16 22:10:49 -0700596static void __kmem_cache_free(void *b, int size)
597{
598 if (size < PAGE_SIZE)
599 slob_free(b, size);
600 else
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800601 slob_free_pages(b, get_order(size));
Nick Pigginafc0ced2007-05-16 22:10:49 -0700602}
603
604static void kmem_rcu_free(struct rcu_head *head)
605{
606 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
607 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
608
609 __kmem_cache_free(b, slob_rcu->size);
610}
611
Matt Mackall10cef602006-01-08 01:01:45 -0800612void kmem_cache_free(struct kmem_cache *c, void *b)
613{
Nick Pigginafc0ced2007-05-16 22:10:49 -0700614 if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
615 struct slob_rcu *slob_rcu;
616 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
617 INIT_RCU_HEAD(&slob_rcu->head);
618 slob_rcu->size = c->size;
619 call_rcu(&slob_rcu->head, kmem_rcu_free);
620 } else {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700621 __kmem_cache_free(b, c->size);
622 }
Matt Mackall10cef602006-01-08 01:01:45 -0800623}
624EXPORT_SYMBOL(kmem_cache_free);
625
626unsigned int kmem_cache_size(struct kmem_cache *c)
627{
628 return c->size;
629}
630EXPORT_SYMBOL(kmem_cache_size);
631
632const char *kmem_cache_name(struct kmem_cache *c)
633{
634 return c->name;
635}
636EXPORT_SYMBOL(kmem_cache_name);
637
Christoph Lameter2e892f42006-12-13 00:34:23 -0800638int kmem_cache_shrink(struct kmem_cache *d)
639{
640 return 0;
641}
642EXPORT_SYMBOL(kmem_cache_shrink);
643
Christoph Lameter55935a32006-12-13 00:34:24 -0800644int kmem_ptr_validate(struct kmem_cache *a, const void *b)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800645{
646 return 0;
647}
648
Paul Mundt84a01c22007-07-15 23:38:24 -0700649static unsigned int slob_ready __read_mostly;
650
651int slab_is_available(void)
652{
653 return slob_ready;
654}
655
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800656void __init kmem_cache_init(void)
657{
Paul Mundt84a01c22007-07-15 23:38:24 -0700658 slob_ready = 1;
Matt Mackall10cef602006-01-08 01:01:45 -0800659}