blob: 9641da3d5e58dfca1d7c79dbef99760364fa98b2 [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
Mel Gorman6484eb32009-06-16 15:31:54 -070049 * provided, alloc_pages_exact_node() with the specified node id is used
Paul Mundt6193a2f2007-07-15 23:38:22 -070050 * 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>
Nick Piggin1f0532e2009-05-05 19:13:45 +100063#include <linux/swap.h> /* struct reclaim_state */
Matt Mackall10cef602006-01-08 01:01:45 -080064#include <linux/cache.h>
65#include <linux/init.h>
66#include <linux/module.h>
Nick Pigginafc0ced2007-05-16 22:10:49 -070067#include <linux/rcupdate.h>
Nick Piggin95b35122007-07-15 23:38:07 -070068#include <linux/list.h>
Zhaolei02af61b2009-04-10 14:26:18 +080069#include <linux/kmemtrace.h>
Catalin Marinas4374e612009-06-11 13:23:17 +010070#include <linux/kmemleak.h>
Nick Piggin95b35122007-07-15 23:38:07 -070071#include <asm/atomic.h>
Matt Mackall10cef602006-01-08 01:01:45 -080072
Nick Piggin95b35122007-07-15 23:38:07 -070073/*
74 * slob_block has a field 'units', which indicates size of block if +ve,
75 * or offset of next block if -ve (in SLOB_UNITs).
76 *
77 * Free blocks of size 1 unit simply contain the offset of the next block.
78 * Those with larger size contain their size in the first SLOB_UNIT of
79 * memory, and the offset of the next free block in the second SLOB_UNIT.
80 */
Nick Piggin55394842007-07-15 23:38:09 -070081#if PAGE_SIZE <= (32767 * 2)
Nick Piggin95b35122007-07-15 23:38:07 -070082typedef s16 slobidx_t;
83#else
84typedef s32 slobidx_t;
85#endif
86
Matt Mackall10cef602006-01-08 01:01:45 -080087struct slob_block {
Nick Piggin95b35122007-07-15 23:38:07 -070088 slobidx_t units;
Nick Piggin55394842007-07-15 23:38:09 -070089};
Matt Mackall10cef602006-01-08 01:01:45 -080090typedef struct slob_block slob_t;
91
Nick Piggin95b35122007-07-15 23:38:07 -070092/*
93 * We use struct page fields to manage some slob allocation aspects,
94 * however to avoid the horrible mess in include/linux/mm_types.h, we'll
95 * just define our own struct page type variant here.
96 */
97struct slob_page {
98 union {
99 struct {
100 unsigned long flags; /* mandatory */
101 atomic_t _count; /* mandatory */
102 slobidx_t units; /* free units left in page */
103 unsigned long pad[2];
104 slob_t *free; /* first free slob_t in page */
105 struct list_head list; /* linked list of free pages */
106 };
107 struct page page;
108 };
109};
110static inline void struct_slob_page_wrong_size(void)
111{ BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); }
112
113/*
114 * free_slob_page: call before a slob_page is returned to the page allocator.
115 */
116static inline void free_slob_page(struct slob_page *sp)
117{
118 reset_page_mapcount(&sp->page);
119 sp->page.mapping = NULL;
120}
121
122/*
Matt Mackall20cecba2008-02-04 22:29:37 -0800123 * All partially free slob pages go on these lists.
Nick Piggin95b35122007-07-15 23:38:07 -0700124 */
Matt Mackall20cecba2008-02-04 22:29:37 -0800125#define SLOB_BREAK1 256
126#define SLOB_BREAK2 1024
127static LIST_HEAD(free_slob_small);
128static LIST_HEAD(free_slob_medium);
129static LIST_HEAD(free_slob_large);
Nick Piggin95b35122007-07-15 23:38:07 -0700130
131/*
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800132 * is_slob_page: True for all slob pages (false for bigblock pages)
Nick Piggin95b35122007-07-15 23:38:07 -0700133 */
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800134static inline int is_slob_page(struct slob_page *sp)
Nick Piggin95b35122007-07-15 23:38:07 -0700135{
Wu Fengguang7303f242009-05-11 09:59:34 +0300136 return PageSlab((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700137}
138
139static inline void set_slob_page(struct slob_page *sp)
140{
Wu Fengguang7303f242009-05-11 09:59:34 +0300141 __SetPageSlab((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700142}
143
144static inline void clear_slob_page(struct slob_page *sp)
145{
Wu Fengguang7303f242009-05-11 09:59:34 +0300146 __ClearPageSlab((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700147}
148
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800149static inline struct slob_page *slob_page(const void *addr)
150{
151 return (struct slob_page *)virt_to_page(addr);
152}
153
Nick Piggin95b35122007-07-15 23:38:07 -0700154/*
155 * slob_page_free: true for pages on free_slob_pages list.
156 */
157static inline int slob_page_free(struct slob_page *sp)
158{
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700159 return PageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700160}
161
Matt Mackall20cecba2008-02-04 22:29:37 -0800162static void set_slob_page_free(struct slob_page *sp, struct list_head *list)
Nick Piggin95b35122007-07-15 23:38:07 -0700163{
Matt Mackall20cecba2008-02-04 22:29:37 -0800164 list_add(&sp->list, list);
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700165 __SetPageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700166}
167
168static inline void clear_slob_page_free(struct slob_page *sp)
169{
170 list_del(&sp->list);
Andy Whitcroft9023cb72008-07-23 21:27:19 -0700171 __ClearPageSlobFree((struct page *)sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700172}
173
Matt Mackall10cef602006-01-08 01:01:45 -0800174#define SLOB_UNIT sizeof(slob_t)
175#define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT)
176#define SLOB_ALIGN L1_CACHE_BYTES
177
Nick Pigginafc0ced2007-05-16 22:10:49 -0700178/*
179 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
180 * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free
181 * the block using call_rcu.
182 */
183struct slob_rcu {
184 struct rcu_head head;
185 int size;
186};
187
Nick Piggin95b35122007-07-15 23:38:07 -0700188/*
189 * slob_lock protects all slob allocator structures.
190 */
Matt Mackall10cef602006-01-08 01:01:45 -0800191static DEFINE_SPINLOCK(slob_lock);
Matt Mackall10cef602006-01-08 01:01:45 -0800192
Nick Piggin95b35122007-07-15 23:38:07 -0700193/*
194 * Encode the given size and next info into a free slob block s.
195 */
196static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
197{
198 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
199 slobidx_t offset = next - base;
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800200
Nick Piggin95b35122007-07-15 23:38:07 -0700201 if (size > 1) {
202 s[0].units = size;
203 s[1].units = offset;
204 } else
205 s[0].units = -offset;
206}
Matt Mackall10cef602006-01-08 01:01:45 -0800207
Nick Piggin95b35122007-07-15 23:38:07 -0700208/*
209 * Return the size of a slob block.
210 */
211static slobidx_t slob_units(slob_t *s)
212{
213 if (s->units > 0)
214 return s->units;
215 return 1;
216}
217
218/*
219 * Return the next free slob block pointer after this one.
220 */
221static slob_t *slob_next(slob_t *s)
222{
223 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
224 slobidx_t next;
225
226 if (s[0].units < 0)
227 next = -s[0].units;
228 else
229 next = s[1].units;
230 return base+next;
231}
232
233/*
234 * Returns true if s is the last free block in its page.
235 */
236static int slob_last(slob_t *s)
237{
238 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
239}
240
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800241static void *slob_new_pages(gfp_t gfp, int order, int node)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700242{
243 void *page;
244
245#ifdef CONFIG_NUMA
246 if (node != -1)
Mel Gorman6484eb32009-06-16 15:31:54 -0700247 page = alloc_pages_exact_node(node, gfp, order);
Paul Mundt6193a2f2007-07-15 23:38:22 -0700248 else
249#endif
250 page = alloc_pages(gfp, order);
251
252 if (!page)
253 return NULL;
254
255 return page_address(page);
256}
257
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800258static void slob_free_pages(void *b, int order)
259{
Nick Piggin1f0532e2009-05-05 19:13:45 +1000260 if (current->reclaim_state)
261 current->reclaim_state->reclaimed_slab += 1 << order;
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800262 free_pages((unsigned long)b, order);
263}
264
Nick Piggin95b35122007-07-15 23:38:07 -0700265/*
266 * Allocate a slob block within a given slob_page sp.
267 */
268static void *slob_page_alloc(struct slob_page *sp, size_t size, int align)
Matt Mackall10cef602006-01-08 01:01:45 -0800269{
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800270 slob_t *prev, *cur, *aligned = NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800271 int delta = 0, units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800272
Nick Piggin95b35122007-07-15 23:38:07 -0700273 for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) {
274 slobidx_t avail = slob_units(cur);
275
Matt Mackall10cef602006-01-08 01:01:45 -0800276 if (align) {
277 aligned = (slob_t *)ALIGN((unsigned long)cur, align);
278 delta = aligned - cur;
279 }
Nick Piggin95b35122007-07-15 23:38:07 -0700280 if (avail >= units + delta) { /* room enough? */
281 slob_t *next;
282
Matt Mackall10cef602006-01-08 01:01:45 -0800283 if (delta) { /* need to fragment head to align? */
Nick Piggin95b35122007-07-15 23:38:07 -0700284 next = slob_next(cur);
285 set_slob(aligned, avail - delta, next);
286 set_slob(cur, delta, aligned);
Matt Mackall10cef602006-01-08 01:01:45 -0800287 prev = cur;
288 cur = aligned;
Nick Piggin95b35122007-07-15 23:38:07 -0700289 avail = slob_units(cur);
Matt Mackall10cef602006-01-08 01:01:45 -0800290 }
291
Nick Piggin95b35122007-07-15 23:38:07 -0700292 next = slob_next(cur);
293 if (avail == units) { /* exact fit? unlink. */
294 if (prev)
295 set_slob(prev, slob_units(prev), next);
296 else
297 sp->free = next;
298 } else { /* fragment */
299 if (prev)
300 set_slob(prev, slob_units(prev), cur + units);
301 else
302 sp->free = cur + units;
303 set_slob(cur + units, avail - units, next);
Matt Mackall10cef602006-01-08 01:01:45 -0800304 }
305
Nick Piggin95b35122007-07-15 23:38:07 -0700306 sp->units -= units;
307 if (!sp->units)
308 clear_slob_page_free(sp);
Matt Mackall10cef602006-01-08 01:01:45 -0800309 return cur;
310 }
Nick Piggin95b35122007-07-15 23:38:07 -0700311 if (slob_last(cur))
312 return NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800313 }
314}
315
Nick Piggin95b35122007-07-15 23:38:07 -0700316/*
317 * slob_alloc: entry point into the slob allocator.
318 */
Paul Mundt6193a2f2007-07-15 23:38:22 -0700319static void *slob_alloc(size_t size, gfp_t gfp, int align, int node)
Nick Piggin95b35122007-07-15 23:38:07 -0700320{
321 struct slob_page *sp;
Matt Mackalld6269542007-07-21 04:37:40 -0700322 struct list_head *prev;
Matt Mackall20cecba2008-02-04 22:29:37 -0800323 struct list_head *slob_list;
Nick Piggin95b35122007-07-15 23:38:07 -0700324 slob_t *b = NULL;
325 unsigned long flags;
326
Matt Mackall20cecba2008-02-04 22:29:37 -0800327 if (size < SLOB_BREAK1)
328 slob_list = &free_slob_small;
329 else if (size < SLOB_BREAK2)
330 slob_list = &free_slob_medium;
331 else
332 slob_list = &free_slob_large;
333
Nick Piggin95b35122007-07-15 23:38:07 -0700334 spin_lock_irqsave(&slob_lock, flags);
335 /* Iterate through each partially free page, try to find room */
Matt Mackall20cecba2008-02-04 22:29:37 -0800336 list_for_each_entry(sp, slob_list, list) {
Paul Mundt6193a2f2007-07-15 23:38:22 -0700337#ifdef CONFIG_NUMA
338 /*
339 * If there's a node specification, search for a partial
340 * page with a matching node id in the freelist.
341 */
342 if (node != -1 && page_to_nid(&sp->page) != node)
343 continue;
344#endif
Matt Mackalld6269542007-07-21 04:37:40 -0700345 /* Enough room on this page? */
346 if (sp->units < SLOB_UNITS(size))
347 continue;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700348
Matt Mackalld6269542007-07-21 04:37:40 -0700349 /* Attempt to alloc */
350 prev = sp->list.prev;
351 b = slob_page_alloc(sp, size, align);
352 if (!b)
353 continue;
354
355 /* Improve fragment distribution and reduce our average
356 * search time by starting our next search here. (see
357 * Knuth vol 1, sec 2.5, pg 449) */
Matt Mackall20cecba2008-02-04 22:29:37 -0800358 if (prev != slob_list->prev &&
359 slob_list->next != prev->next)
360 list_move_tail(slob_list, prev->next);
Matt Mackalld6269542007-07-21 04:37:40 -0700361 break;
Nick Piggin95b35122007-07-15 23:38:07 -0700362 }
363 spin_unlock_irqrestore(&slob_lock, flags);
364
365 /* Not enough space: must allocate a new page */
366 if (!b) {
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800367 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
Nick Piggin95b35122007-07-15 23:38:07 -0700368 if (!b)
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800369 return NULL;
370 sp = slob_page(b);
Nick Piggin95b35122007-07-15 23:38:07 -0700371 set_slob_page(sp);
372
373 spin_lock_irqsave(&slob_lock, flags);
374 sp->units = SLOB_UNITS(PAGE_SIZE);
375 sp->free = b;
376 INIT_LIST_HEAD(&sp->list);
377 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
Matt Mackall20cecba2008-02-04 22:29:37 -0800378 set_slob_page_free(sp, slob_list);
Nick Piggin95b35122007-07-15 23:38:07 -0700379 b = slob_page_alloc(sp, size, align);
380 BUG_ON(!b);
381 spin_unlock_irqrestore(&slob_lock, flags);
382 }
Christoph Lameterd07dbea2007-07-17 04:03:23 -0700383 if (unlikely((gfp & __GFP_ZERO) && b))
384 memset(b, 0, size);
Nick Piggin95b35122007-07-15 23:38:07 -0700385 return b;
386}
387
388/*
389 * slob_free: entry point into the slob allocator.
390 */
Matt Mackall10cef602006-01-08 01:01:45 -0800391static void slob_free(void *block, int size)
392{
Nick Piggin95b35122007-07-15 23:38:07 -0700393 struct slob_page *sp;
394 slob_t *prev, *next, *b = (slob_t *)block;
395 slobidx_t units;
Matt Mackall10cef602006-01-08 01:01:45 -0800396 unsigned long flags;
397
Satyam Sharma2408c552007-10-16 01:24:44 -0700398 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800399 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700400 BUG_ON(!size);
Matt Mackall10cef602006-01-08 01:01:45 -0800401
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800402 sp = slob_page(block);
Nick Piggin95b35122007-07-15 23:38:07 -0700403 units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800404
Matt Mackall10cef602006-01-08 01:01:45 -0800405 spin_lock_irqsave(&slob_lock, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800406
Nick Piggin95b35122007-07-15 23:38:07 -0700407 if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
408 /* Go directly to page allocator. Do not pass slob allocator */
409 if (slob_page_free(sp))
410 clear_slob_page_free(sp);
Nick Piggin6fb8f422009-03-16 21:00:28 +1100411 spin_unlock_irqrestore(&slob_lock, flags);
Nick Piggin95b35122007-07-15 23:38:07 -0700412 clear_slob_page(sp);
413 free_slob_page(sp);
Nick Piggin1f0532e2009-05-05 19:13:45 +1000414 slob_free_pages(b, 0);
Nick Piggin6fb8f422009-03-16 21:00:28 +1100415 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700416 }
Matt Mackall10cef602006-01-08 01:01:45 -0800417
Nick Piggin95b35122007-07-15 23:38:07 -0700418 if (!slob_page_free(sp)) {
419 /* This slob page is about to become partially free. Easy! */
420 sp->units = units;
421 sp->free = b;
422 set_slob(b, units,
423 (void *)((unsigned long)(b +
424 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
Matt Mackall20cecba2008-02-04 22:29:37 -0800425 set_slob_page_free(sp, &free_slob_small);
Nick Piggin95b35122007-07-15 23:38:07 -0700426 goto out;
427 }
Matt Mackall10cef602006-01-08 01:01:45 -0800428
Nick Piggin95b35122007-07-15 23:38:07 -0700429 /*
430 * Otherwise the page is already partially free, so find reinsertion
431 * point.
432 */
433 sp->units += units;
Matt Mackall10cef602006-01-08 01:01:45 -0800434
Nick Piggin95b35122007-07-15 23:38:07 -0700435 if (b < sp->free) {
Matt Mackall679299b2008-02-04 22:29:37 -0800436 if (b + units == sp->free) {
437 units += slob_units(sp->free);
438 sp->free = slob_next(sp->free);
439 }
Nick Piggin95b35122007-07-15 23:38:07 -0700440 set_slob(b, units, sp->free);
441 sp->free = b;
442 } else {
443 prev = sp->free;
444 next = slob_next(prev);
445 while (b > next) {
446 prev = next;
447 next = slob_next(prev);
448 }
449
450 if (!slob_last(prev) && b + units == next) {
451 units += slob_units(next);
452 set_slob(b, units, slob_next(next));
453 } else
454 set_slob(b, units, next);
455
456 if (prev + slob_units(prev) == b) {
457 units = slob_units(b) + slob_units(prev);
458 set_slob(prev, units, slob_next(b));
459 } else
460 set_slob(prev, slob_units(prev), b);
461 }
462out:
Matt Mackall10cef602006-01-08 01:01:45 -0800463 spin_unlock_irqrestore(&slob_lock, flags);
464}
465
Nick Piggin95b35122007-07-15 23:38:07 -0700466/*
467 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
468 */
469
Nick Piggin55394842007-07-15 23:38:09 -0700470#ifndef ARCH_KMALLOC_MINALIGN
471#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long)
472#endif
473
474#ifndef ARCH_SLAB_MINALIGN
475#define ARCH_SLAB_MINALIGN __alignof__(unsigned long)
476#endif
477
Paul Mundt6193a2f2007-07-15 23:38:22 -0700478void *__kmalloc_node(size_t size, gfp_t gfp, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800479{
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700480 unsigned int *m;
Nick Piggin55394842007-07-15 23:38:09 -0700481 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300482 void *ret;
Nick Piggin55394842007-07-15 23:38:09 -0700483
Ingo Molnar19cefdf2009-03-15 06:03:11 +0100484 lockdep_trace_alloc(gfp);
Nick Piggincf40bd12009-01-21 08:12:39 +0100485
Nick Piggin55394842007-07-15 23:38:09 -0700486 if (size < PAGE_SIZE - align) {
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700487 if (!size)
488 return ZERO_SIZE_PTR;
489
Paul Mundt6193a2f2007-07-15 23:38:22 -0700490 m = slob_alloc(size + align, gfp, align, node);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300491
MinChan Kim239f49c2008-05-19 22:12:08 +0900492 if (!m)
493 return NULL;
494 *m = size;
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300495 ret = (void *)m + align;
Nick Piggind87a1332007-07-15 23:38:08 -0700496
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +0200497 trace_kmalloc_node(_RET_IP_, ret,
498 size, size + align, gfp, node);
Nick Piggind87a1332007-07-15 23:38:08 -0700499 } else {
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300500 unsigned int order = get_order(size);
Nick Piggind87a1332007-07-15 23:38:08 -0700501
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800502 ret = slob_new_pages(gfp | __GFP_COMP, get_order(size), node);
Nick Piggind87a1332007-07-15 23:38:08 -0700503 if (ret) {
504 struct page *page;
505 page = virt_to_page(ret);
506 page->private = size;
507 }
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300508
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +0200509 trace_kmalloc_node(_RET_IP_, ret,
510 size, PAGE_SIZE << order, gfp, node);
Matt Mackall10cef602006-01-08 01:01:45 -0800511 }
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300512
Catalin Marinas4374e612009-06-11 13:23:17 +0100513 kmemleak_alloc(ret, size, 1, gfp);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300514 return ret;
Matt Mackall10cef602006-01-08 01:01:45 -0800515}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700516EXPORT_SYMBOL(__kmalloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800517
518void kfree(const void *block)
519{
Nick Piggin95b35122007-07-15 23:38:07 -0700520 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800521
Pekka Enberg2121db72009-03-25 11:05:57 +0200522 trace_kfree(_RET_IP_, block);
523
Satyam Sharma2408c552007-10-16 01:24:44 -0700524 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800525 return;
Catalin Marinas4374e612009-06-11 13:23:17 +0100526 kmemleak_free(block);
Matt Mackall10cef602006-01-08 01:01:45 -0800527
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800528 sp = slob_page(block);
529 if (is_slob_page(sp)) {
Nick Piggin55394842007-07-15 23:38:09 -0700530 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
531 unsigned int *m = (unsigned int *)(block - align);
532 slob_free(m, *m + align);
Nick Piggind87a1332007-07-15 23:38:08 -0700533 } else
534 put_page(&sp->page);
Matt Mackall10cef602006-01-08 01:01:45 -0800535}
Matt Mackall10cef602006-01-08 01:01:45 -0800536EXPORT_SYMBOL(kfree);
537
Nick Piggind87a1332007-07-15 23:38:08 -0700538/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
Pekka Enbergfd76bab2007-05-06 14:48:40 -0700539size_t ksize(const void *block)
Matt Mackall10cef602006-01-08 01:01:45 -0800540{
Nick Piggin95b35122007-07-15 23:38:07 -0700541 struct slob_page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800542
Christoph Lameteref8b4522007-10-16 01:24:46 -0700543 BUG_ON(!block);
544 if (unlikely(block == ZERO_SIZE_PTR))
Matt Mackall10cef602006-01-08 01:01:45 -0800545 return 0;
546
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800547 sp = slob_page(block);
548 if (is_slob_page(sp)) {
Matt Mackall70096a52008-10-08 14:51:57 -0500549 int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
550 unsigned int *m = (unsigned int *)(block - align);
551 return SLOB_UNITS(*m) * SLOB_UNIT;
552 } else
Nick Piggind87a1332007-07-15 23:38:08 -0700553 return sp->page.private;
Matt Mackall10cef602006-01-08 01:01:45 -0800554}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +0200555EXPORT_SYMBOL(ksize);
Matt Mackall10cef602006-01-08 01:01:45 -0800556
557struct kmem_cache {
558 unsigned int size, align;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700559 unsigned long flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800560 const char *name;
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700561 void (*ctor)(void *);
Matt Mackall10cef602006-01-08 01:01:45 -0800562};
563
564struct kmem_cache *kmem_cache_create(const char *name, size_t size,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700565 size_t align, unsigned long flags, void (*ctor)(void *))
Matt Mackall10cef602006-01-08 01:01:45 -0800566{
567 struct kmem_cache *c;
568
Yi Li0701a9e2008-04-25 19:49:21 +0300569 c = slob_alloc(sizeof(struct kmem_cache),
Catalin Marinas5e18e2b2008-12-15 13:54:16 -0800570 GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1);
Matt Mackall10cef602006-01-08 01:01:45 -0800571
572 if (c) {
573 c->name = name;
574 c->size = size;
Nick Pigginafc0ced2007-05-16 22:10:49 -0700575 if (flags & SLAB_DESTROY_BY_RCU) {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700576 /* leave room for rcu footer at the end of object */
577 c->size += sizeof(struct slob_rcu);
578 }
579 c->flags = flags;
Matt Mackall10cef602006-01-08 01:01:45 -0800580 c->ctor = ctor;
Matt Mackall10cef602006-01-08 01:01:45 -0800581 /* ignore alignment unless it's forced */
Christoph Lameter5af60832007-05-06 14:49:56 -0700582 c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0;
Nick Piggin55394842007-07-15 23:38:09 -0700583 if (c->align < ARCH_SLAB_MINALIGN)
584 c->align = ARCH_SLAB_MINALIGN;
Matt Mackall10cef602006-01-08 01:01:45 -0800585 if (c->align < align)
586 c->align = align;
Akinobu Mitabc0055a2007-05-06 14:49:52 -0700587 } else if (flags & SLAB_PANIC)
588 panic("Cannot create slab cache %s\n", name);
Matt Mackall10cef602006-01-08 01:01:45 -0800589
Catalin Marinas4374e612009-06-11 13:23:17 +0100590 kmemleak_alloc(c, sizeof(struct kmem_cache), 1, GFP_KERNEL);
Matt Mackall10cef602006-01-08 01:01:45 -0800591 return c;
592}
593EXPORT_SYMBOL(kmem_cache_create);
594
Alexey Dobriyan133d2052006-09-27 01:49:41 -0700595void kmem_cache_destroy(struct kmem_cache *c)
Matt Mackall10cef602006-01-08 01:01:45 -0800596{
Catalin Marinas4374e612009-06-11 13:23:17 +0100597 kmemleak_free(c);
Paul E. McKenney7ed9f7e2009-06-25 12:31:37 -0700598 if (c->flags & SLAB_DESTROY_BY_RCU)
599 rcu_barrier();
Matt Mackall10cef602006-01-08 01:01:45 -0800600 slob_free(c, sizeof(struct kmem_cache));
Matt Mackall10cef602006-01-08 01:01:45 -0800601}
602EXPORT_SYMBOL(kmem_cache_destroy);
603
Paul Mundt6193a2f2007-07-15 23:38:22 -0700604void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800605{
606 void *b;
607
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300608 if (c->size < PAGE_SIZE) {
Paul Mundt6193a2f2007-07-15 23:38:22 -0700609 b = slob_alloc(c->size, flags, c->align, node);
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +0200610 trace_kmem_cache_alloc_node(_RET_IP_, b, c->size,
611 SLOB_UNITS(c->size) * SLOB_UNIT,
612 flags, node);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300613 } else {
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800614 b = slob_new_pages(flags, get_order(c->size), node);
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +0200615 trace_kmem_cache_alloc_node(_RET_IP_, b, c->size,
616 PAGE_SIZE << get_order(c->size),
617 flags, node);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300618 }
Matt Mackall10cef602006-01-08 01:01:45 -0800619
620 if (c->ctor)
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700621 c->ctor(b);
Matt Mackall10cef602006-01-08 01:01:45 -0800622
Catalin Marinas4374e612009-06-11 13:23:17 +0100623 kmemleak_alloc_recursive(b, c->size, 1, c->flags, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800624 return b;
625}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700626EXPORT_SYMBOL(kmem_cache_alloc_node);
Matt Mackall10cef602006-01-08 01:01:45 -0800627
Nick Pigginafc0ced2007-05-16 22:10:49 -0700628static void __kmem_cache_free(void *b, int size)
629{
630 if (size < PAGE_SIZE)
631 slob_free(b, size);
632 else
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800633 slob_free_pages(b, get_order(size));
Nick Pigginafc0ced2007-05-16 22:10:49 -0700634}
635
636static void kmem_rcu_free(struct rcu_head *head)
637{
638 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
639 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
640
641 __kmem_cache_free(b, slob_rcu->size);
642}
643
Matt Mackall10cef602006-01-08 01:01:45 -0800644void kmem_cache_free(struct kmem_cache *c, void *b)
645{
Catalin Marinas4374e612009-06-11 13:23:17 +0100646 kmemleak_free_recursive(b, c->flags);
Nick Pigginafc0ced2007-05-16 22:10:49 -0700647 if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) {
648 struct slob_rcu *slob_rcu;
649 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
650 INIT_RCU_HEAD(&slob_rcu->head);
651 slob_rcu->size = c->size;
652 call_rcu(&slob_rcu->head, kmem_rcu_free);
653 } else {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700654 __kmem_cache_free(b, c->size);
655 }
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300656
Eduard - Gabriel Munteanuca2b84c2009-03-23 15:12:24 +0200657 trace_kmem_cache_free(_RET_IP_, b);
Matt Mackall10cef602006-01-08 01:01:45 -0800658}
659EXPORT_SYMBOL(kmem_cache_free);
660
661unsigned int kmem_cache_size(struct kmem_cache *c)
662{
663 return c->size;
664}
665EXPORT_SYMBOL(kmem_cache_size);
666
667const char *kmem_cache_name(struct kmem_cache *c)
668{
669 return c->name;
670}
671EXPORT_SYMBOL(kmem_cache_name);
672
Christoph Lameter2e892f42006-12-13 00:34:23 -0800673int kmem_cache_shrink(struct kmem_cache *d)
674{
675 return 0;
676}
677EXPORT_SYMBOL(kmem_cache_shrink);
678
Christoph Lameter55935a32006-12-13 00:34:24 -0800679int kmem_ptr_validate(struct kmem_cache *a, const void *b)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800680{
681 return 0;
682}
683
Paul Mundt84a01c22007-07-15 23:38:24 -0700684static unsigned int slob_ready __read_mostly;
685
686int slab_is_available(void)
687{
688 return slob_ready;
689}
690
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800691void __init kmem_cache_init(void)
692{
Paul Mundt84a01c22007-07-15 23:38:24 -0700693 slob_ready = 1;
Matt Mackall10cef602006-01-08 01:01:45 -0800694}