blob: fa53e9f738935605acf538cdcbbf76f41f410169 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Matt Mackall10cef602006-01-08 01:01:45 -08002/*
3 * SLOB Allocator: Simple List Of Blocks
4 *
5 * Matt Mackall <mpm@selenic.com> 12/30/03
6 *
Paul Mundt6193a2f2007-07-15 23:38:22 -07007 * NUMA support by Paul Mundt, 2007.
8 *
Matt Mackall10cef602006-01-08 01:01:45 -08009 * How SLOB works:
10 *
11 * The core of SLOB is a traditional K&R style heap allocator, with
12 * support for returning aligned objects. The granularity of this
Nick Piggin55394842007-07-15 23:38:09 -070013 * allocator is as little as 2 bytes, however typically most architectures
14 * will require 4 bytes on 32-bit and 8 bytes on 64-bit.
Nick Piggin95b35122007-07-15 23:38:07 -070015 *
Matt Mackall20cecba2008-02-04 22:29:37 -080016 * The slob heap is a set of linked list of pages from alloc_pages(),
17 * and within each page, there is a singly-linked list of free blocks
18 * (slob_t). The heap is grown on demand. To reduce fragmentation,
19 * heap pages are segregated into three lists, with objects less than
20 * 256 bytes, objects less than 1024 bytes, and all other objects.
21 *
22 * Allocation from heap involves first searching for a page with
23 * sufficient free blocks (using a next-fit-like approach) followed by
24 * a first-fit scan of the page. Deallocation inserts objects back
25 * into the free list in address order, so this is effectively an
26 * address-ordered first fit.
Matt Mackall10cef602006-01-08 01:01:45 -080027 *
28 * Above this is an implementation of kmalloc/kfree. Blocks returned
Nick Piggin55394842007-07-15 23:38:09 -070029 * from kmalloc are prepended with a 4-byte header with the kmalloc size.
Matt Mackall10cef602006-01-08 01:01:45 -080030 * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
Paul Mundt6193a2f2007-07-15 23:38:22 -070031 * alloc_pages() directly, allocating compound pages so the page order
Ezequiel Garcia999d8792012-10-19 09:33:10 -030032 * does not have to be separately tracked.
33 * These objects are detected in kfree() because PageSlab()
Nick Piggind87a1332007-07-15 23:38:08 -070034 * 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
Vlastimil Babka96db8002015-09-08 15:03:50 -070049 * provided, __alloc_pages_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>
Christoph Lameter97d06602012-07-06 15:25:11 -050062
Matt Mackall10cef602006-01-08 01:01:45 -080063#include <linux/mm.h>
Nick Piggin1f0532e2009-05-05 19:13:45 +100064#include <linux/swap.h> /* struct reclaim_state */
Matt Mackall10cef602006-01-08 01:01:45 -080065#include <linux/cache.h>
66#include <linux/init.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040067#include <linux/export.h>
Nick Pigginafc0ced2007-05-16 22:10:49 -070068#include <linux/rcupdate.h>
Nick Piggin95b35122007-07-15 23:38:07 -070069#include <linux/list.h>
Catalin Marinas4374e612009-06-11 13:23:17 +010070#include <linux/kmemleak.h>
Li Zefan039ca4e2010-05-26 17:22:17 +080071
72#include <trace/events/kmem.h>
73
Arun Sharma600634972011-07-26 16:09:06 -070074#include <linux/atomic.h>
Matt Mackall10cef602006-01-08 01:01:45 -080075
Glauber Costab9ce5ef2012-12-18 14:22:46 -080076#include "slab.h"
Nick Piggin95b35122007-07-15 23:38:07 -070077/*
78 * slob_block has a field 'units', which indicates size of block if +ve,
79 * or offset of next block if -ve (in SLOB_UNITs).
80 *
81 * Free blocks of size 1 unit simply contain the offset of the next block.
82 * Those with larger size contain their size in the first SLOB_UNIT of
83 * memory, and the offset of the next free block in the second SLOB_UNIT.
84 */
Nick Piggin55394842007-07-15 23:38:09 -070085#if PAGE_SIZE <= (32767 * 2)
Nick Piggin95b35122007-07-15 23:38:07 -070086typedef s16 slobidx_t;
87#else
88typedef s32 slobidx_t;
89#endif
90
Matt Mackall10cef602006-01-08 01:01:45 -080091struct slob_block {
Nick Piggin95b35122007-07-15 23:38:07 -070092 slobidx_t units;
Nick Piggin55394842007-07-15 23:38:09 -070093};
Matt Mackall10cef602006-01-08 01:01:45 -080094typedef struct slob_block slob_t;
95
Nick Piggin95b35122007-07-15 23:38:07 -070096/*
Matt Mackall20cecba2008-02-04 22:29:37 -080097 * All partially free slob pages go on these lists.
Nick Piggin95b35122007-07-15 23:38:07 -070098 */
Matt Mackall20cecba2008-02-04 22:29:37 -080099#define SLOB_BREAK1 256
100#define SLOB_BREAK2 1024
101static LIST_HEAD(free_slob_small);
102static LIST_HEAD(free_slob_medium);
103static LIST_HEAD(free_slob_large);
Nick Piggin95b35122007-07-15 23:38:07 -0700104
105/*
Nick Piggin95b35122007-07-15 23:38:07 -0700106 * slob_page_free: true for pages on free_slob_pages list.
107 */
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500108static inline int slob_page_free(struct page *sp)
Nick Piggin95b35122007-07-15 23:38:07 -0700109{
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500110 return PageSlobFree(sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700111}
112
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500113static void set_slob_page_free(struct page *sp, struct list_head *list)
Nick Piggin95b35122007-07-15 23:38:07 -0700114{
Tobin C. Hardingadab7b62019-05-13 17:16:06 -0700115 list_add(&sp->slab_list, list);
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500116 __SetPageSlobFree(sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700117}
118
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500119static inline void clear_slob_page_free(struct page *sp)
Nick Piggin95b35122007-07-15 23:38:07 -0700120{
Tobin C. Hardingadab7b62019-05-13 17:16:06 -0700121 list_del(&sp->slab_list);
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500122 __ClearPageSlobFree(sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700123}
124
Matt Mackall10cef602006-01-08 01:01:45 -0800125#define SLOB_UNIT sizeof(slob_t)
Sasha Levina6d78152012-12-20 14:11:39 -0500126#define SLOB_UNITS(size) DIV_ROUND_UP(size, SLOB_UNIT)
Matt Mackall10cef602006-01-08 01:01:45 -0800127
Nick Pigginafc0ced2007-05-16 22:10:49 -0700128/*
129 * struct slob_rcu is inserted at the tail of allocated slob blocks, which
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -0800130 * were created with a SLAB_TYPESAFE_BY_RCU slab. slob_rcu is used to free
Nick Pigginafc0ced2007-05-16 22:10:49 -0700131 * the block using call_rcu.
132 */
133struct slob_rcu {
134 struct rcu_head head;
135 int size;
136};
137
Nick Piggin95b35122007-07-15 23:38:07 -0700138/*
139 * slob_lock protects all slob allocator structures.
140 */
Matt Mackall10cef602006-01-08 01:01:45 -0800141static DEFINE_SPINLOCK(slob_lock);
Matt Mackall10cef602006-01-08 01:01:45 -0800142
Nick Piggin95b35122007-07-15 23:38:07 -0700143/*
144 * Encode the given size and next info into a free slob block s.
145 */
146static void set_slob(slob_t *s, slobidx_t size, slob_t *next)
147{
148 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
149 slobidx_t offset = next - base;
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800150
Nick Piggin95b35122007-07-15 23:38:07 -0700151 if (size > 1) {
152 s[0].units = size;
153 s[1].units = offset;
154 } else
155 s[0].units = -offset;
156}
Matt Mackall10cef602006-01-08 01:01:45 -0800157
Nick Piggin95b35122007-07-15 23:38:07 -0700158/*
159 * Return the size of a slob block.
160 */
161static slobidx_t slob_units(slob_t *s)
162{
163 if (s->units > 0)
164 return s->units;
165 return 1;
166}
167
168/*
169 * Return the next free slob block pointer after this one.
170 */
171static slob_t *slob_next(slob_t *s)
172{
173 slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK);
174 slobidx_t next;
175
176 if (s[0].units < 0)
177 next = -s[0].units;
178 else
179 next = s[1].units;
180 return base+next;
181}
182
183/*
184 * Returns true if s is the last free block in its page.
185 */
186static int slob_last(slob_t *s)
187{
188 return !((unsigned long)slob_next(s) & ~PAGE_MASK);
189}
190
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800191static void *slob_new_pages(gfp_t gfp, int order, int node)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700192{
Vlastimil Babka6a486c02019-10-06 17:58:42 -0700193 struct page *page;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700194
195#ifdef CONFIG_NUMA
Ezequiel Garcia90f2cbb2012-09-08 17:47:51 -0300196 if (node != NUMA_NO_NODE)
Vlastimil Babka96db8002015-09-08 15:03:50 -0700197 page = __alloc_pages_node(node, gfp, order);
Paul Mundt6193a2f2007-07-15 23:38:22 -0700198 else
199#endif
200 page = alloc_pages(gfp, order);
201
202 if (!page)
203 return NULL;
204
Vlastimil Babka6a486c02019-10-06 17:58:42 -0700205 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE,
206 1 << order);
Paul Mundt6193a2f2007-07-15 23:38:22 -0700207 return page_address(page);
208}
209
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800210static void slob_free_pages(void *b, int order)
211{
Vlastimil Babka6a486c02019-10-06 17:58:42 -0700212 struct page *sp = virt_to_page(b);
213
Nick Piggin1f0532e2009-05-05 19:13:45 +1000214 if (current->reclaim_state)
215 current->reclaim_state->reclaimed_slab += 1 << order;
Vlastimil Babka6a486c02019-10-06 17:58:42 -0700216
217 mod_node_page_state(page_pgdat(sp), NR_SLAB_UNRECLAIMABLE,
218 -(1 << order));
219 __free_pages(sp, order);
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800220}
221
Nick Piggin95b35122007-07-15 23:38:07 -0700222/*
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700223 * slob_page_alloc() - Allocate a slob block within a given slob_page sp.
224 * @sp: Page to look in.
225 * @size: Size of the allocation.
226 * @align: Allocation alignment.
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700227 * @align_offset: Offset in the allocated block that will be aligned.
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700228 * @page_removed_from_list: Return parameter.
229 *
230 * Tries to find a chunk of memory at least @size bytes big within @page.
231 *
232 * Return: Pointer to memory if allocated, %NULL otherwise. If the
233 * allocation fills up @page then the page is removed from the
234 * freelist, in this case @page_removed_from_list will be set to
235 * true (set to false otherwise).
Nick Piggin95b35122007-07-15 23:38:07 -0700236 */
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700237static void *slob_page_alloc(struct page *sp, size_t size, int align,
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700238 int align_offset, bool *page_removed_from_list)
Matt Mackall10cef602006-01-08 01:01:45 -0800239{
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800240 slob_t *prev, *cur, *aligned = NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800241 int delta = 0, units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800242
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700243 *page_removed_from_list = false;
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500244 for (prev = NULL, cur = sp->freelist; ; prev = cur, cur = slob_next(cur)) {
Nick Piggin95b35122007-07-15 23:38:07 -0700245 slobidx_t avail = slob_units(cur);
246
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700247 /*
248 * 'aligned' will hold the address of the slob block so that the
249 * address 'aligned'+'align_offset' is aligned according to the
250 * 'align' parameter. This is for kmalloc() which prepends the
251 * allocated block with its size, so that the block itself is
252 * aligned when needed.
253 */
Matt Mackall10cef602006-01-08 01:01:45 -0800254 if (align) {
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700255 aligned = (slob_t *)
256 (ALIGN((unsigned long)cur + align_offset, align)
257 - align_offset);
Matt Mackall10cef602006-01-08 01:01:45 -0800258 delta = aligned - cur;
259 }
Nick Piggin95b35122007-07-15 23:38:07 -0700260 if (avail >= units + delta) { /* room enough? */
261 slob_t *next;
262
Matt Mackall10cef602006-01-08 01:01:45 -0800263 if (delta) { /* need to fragment head to align? */
Nick Piggin95b35122007-07-15 23:38:07 -0700264 next = slob_next(cur);
265 set_slob(aligned, avail - delta, next);
266 set_slob(cur, delta, aligned);
Matt Mackall10cef602006-01-08 01:01:45 -0800267 prev = cur;
268 cur = aligned;
Nick Piggin95b35122007-07-15 23:38:07 -0700269 avail = slob_units(cur);
Matt Mackall10cef602006-01-08 01:01:45 -0800270 }
271
Nick Piggin95b35122007-07-15 23:38:07 -0700272 next = slob_next(cur);
273 if (avail == units) { /* exact fit? unlink. */
274 if (prev)
275 set_slob(prev, slob_units(prev), next);
276 else
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500277 sp->freelist = next;
Nick Piggin95b35122007-07-15 23:38:07 -0700278 } else { /* fragment */
279 if (prev)
280 set_slob(prev, slob_units(prev), cur + units);
281 else
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500282 sp->freelist = cur + units;
Nick Piggin95b35122007-07-15 23:38:07 -0700283 set_slob(cur + units, avail - units, next);
Matt Mackall10cef602006-01-08 01:01:45 -0800284 }
285
Nick Piggin95b35122007-07-15 23:38:07 -0700286 sp->units -= units;
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700287 if (!sp->units) {
Nick Piggin95b35122007-07-15 23:38:07 -0700288 clear_slob_page_free(sp);
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700289 *page_removed_from_list = true;
290 }
Matt Mackall10cef602006-01-08 01:01:45 -0800291 return cur;
292 }
Nick Piggin95b35122007-07-15 23:38:07 -0700293 if (slob_last(cur))
294 return NULL;
Matt Mackall10cef602006-01-08 01:01:45 -0800295 }
296}
297
Nick Piggin95b35122007-07-15 23:38:07 -0700298/*
299 * slob_alloc: entry point into the slob allocator.
300 */
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700301static void *slob_alloc(size_t size, gfp_t gfp, int align, int node,
302 int align_offset)
Nick Piggin95b35122007-07-15 23:38:07 -0700303{
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500304 struct page *sp;
Matt Mackall20cecba2008-02-04 22:29:37 -0800305 struct list_head *slob_list;
Nick Piggin95b35122007-07-15 23:38:07 -0700306 slob_t *b = NULL;
307 unsigned long flags;
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700308 bool _unused;
Nick Piggin95b35122007-07-15 23:38:07 -0700309
Matt Mackall20cecba2008-02-04 22:29:37 -0800310 if (size < SLOB_BREAK1)
311 slob_list = &free_slob_small;
312 else if (size < SLOB_BREAK2)
313 slob_list = &free_slob_medium;
314 else
315 slob_list = &free_slob_large;
316
Nick Piggin95b35122007-07-15 23:38:07 -0700317 spin_lock_irqsave(&slob_lock, flags);
318 /* Iterate through each partially free page, try to find room */
Tobin C. Hardingadab7b62019-05-13 17:16:06 -0700319 list_for_each_entry(sp, slob_list, slab_list) {
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700320 bool page_removed_from_list = false;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700321#ifdef CONFIG_NUMA
322 /*
323 * If there's a node specification, search for a partial
324 * page with a matching node id in the freelist.
325 */
Ezequiel Garcia90f2cbb2012-09-08 17:47:51 -0300326 if (node != NUMA_NO_NODE && page_to_nid(sp) != node)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700327 continue;
328#endif
Matt Mackalld6269542007-07-21 04:37:40 -0700329 /* Enough room on this page? */
330 if (sp->units < SLOB_UNITS(size))
331 continue;
Paul Mundt6193a2f2007-07-15 23:38:22 -0700332
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700333 b = slob_page_alloc(sp, size, align, align_offset, &page_removed_from_list);
Matt Mackalld6269542007-07-21 04:37:40 -0700334 if (!b)
335 continue;
336
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700337 /*
338 * If slob_page_alloc() removed sp from the list then we
339 * cannot call list functions on sp. If so allocation
340 * did not fragment the page anyway so optimisation is
341 * unnecessary.
342 */
343 if (!page_removed_from_list) {
344 /*
345 * Improve fragment distribution and reduce our average
346 * search time by starting our next search here. (see
347 * Knuth vol 1, sec 2.5, pg 449)
348 */
Tobin C. Hardingadab7b62019-05-13 17:16:06 -0700349 if (!list_is_first(&sp->slab_list, slob_list))
350 list_rotate_to_front(&sp->slab_list, slob_list);
Tobin C. Harding130e8e02019-05-13 17:16:03 -0700351 }
Matt Mackalld6269542007-07-21 04:37:40 -0700352 break;
Nick Piggin95b35122007-07-15 23:38:07 -0700353 }
354 spin_unlock_irqrestore(&slob_lock, flags);
355
356 /* Not enough space: must allocate a new page */
357 if (!b) {
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800358 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
Nick Piggin95b35122007-07-15 23:38:07 -0700359 if (!b)
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800360 return NULL;
Christoph Lameterb5568282012-06-13 10:24:54 -0500361 sp = virt_to_page(b);
362 __SetPageSlab(sp);
Nick Piggin95b35122007-07-15 23:38:07 -0700363
364 spin_lock_irqsave(&slob_lock, flags);
365 sp->units = SLOB_UNITS(PAGE_SIZE);
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500366 sp->freelist = b;
Tobin C. Hardingadab7b62019-05-13 17:16:06 -0700367 INIT_LIST_HEAD(&sp->slab_list);
Nick Piggin95b35122007-07-15 23:38:07 -0700368 set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE));
Matt Mackall20cecba2008-02-04 22:29:37 -0800369 set_slob_page_free(sp, slob_list);
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700370 b = slob_page_alloc(sp, size, align, align_offset, &_unused);
Nick Piggin95b35122007-07-15 23:38:07 -0700371 BUG_ON(!b);
372 spin_unlock_irqrestore(&slob_lock, flags);
373 }
Miles Chen9f88fae2017-11-15 17:32:10 -0800374 if (unlikely(gfp & __GFP_ZERO))
Christoph Lameterd07dbea2007-07-17 04:03:23 -0700375 memset(b, 0, size);
Nick Piggin95b35122007-07-15 23:38:07 -0700376 return b;
377}
378
379/*
380 * slob_free: entry point into the slob allocator.
381 */
Matt Mackall10cef602006-01-08 01:01:45 -0800382static void slob_free(void *block, int size)
383{
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500384 struct page *sp;
Nick Piggin95b35122007-07-15 23:38:07 -0700385 slob_t *prev, *next, *b = (slob_t *)block;
386 slobidx_t units;
Matt Mackall10cef602006-01-08 01:01:45 -0800387 unsigned long flags;
Bob Liud602dab2010-07-10 18:05:33 +0800388 struct list_head *slob_list;
Matt Mackall10cef602006-01-08 01:01:45 -0800389
Satyam Sharma2408c552007-10-16 01:24:44 -0700390 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800391 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700392 BUG_ON(!size);
Matt Mackall10cef602006-01-08 01:01:45 -0800393
Christoph Lameterb5568282012-06-13 10:24:54 -0500394 sp = virt_to_page(block);
Nick Piggin95b35122007-07-15 23:38:07 -0700395 units = SLOB_UNITS(size);
Matt Mackall10cef602006-01-08 01:01:45 -0800396
Matt Mackall10cef602006-01-08 01:01:45 -0800397 spin_lock_irqsave(&slob_lock, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800398
Nick Piggin95b35122007-07-15 23:38:07 -0700399 if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) {
400 /* Go directly to page allocator. Do not pass slob allocator */
401 if (slob_page_free(sp))
402 clear_slob_page_free(sp);
Nick Piggin6fb8f422009-03-16 21:00:28 +1100403 spin_unlock_irqrestore(&slob_lock, flags);
Christoph Lameterb5568282012-06-13 10:24:54 -0500404 __ClearPageSlab(sp);
Mel Gorman22b751c2013-02-22 16:34:59 -0800405 page_mapcount_reset(sp);
Nick Piggin1f0532e2009-05-05 19:13:45 +1000406 slob_free_pages(b, 0);
Nick Piggin6fb8f422009-03-16 21:00:28 +1100407 return;
Nick Piggin95b35122007-07-15 23:38:07 -0700408 }
Matt Mackall10cef602006-01-08 01:01:45 -0800409
Nick Piggin95b35122007-07-15 23:38:07 -0700410 if (!slob_page_free(sp)) {
411 /* This slob page is about to become partially free. Easy! */
412 sp->units = units;
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500413 sp->freelist = b;
Nick Piggin95b35122007-07-15 23:38:07 -0700414 set_slob(b, units,
415 (void *)((unsigned long)(b +
416 SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK));
Bob Liud602dab2010-07-10 18:05:33 +0800417 if (size < SLOB_BREAK1)
418 slob_list = &free_slob_small;
419 else if (size < SLOB_BREAK2)
420 slob_list = &free_slob_medium;
421 else
422 slob_list = &free_slob_large;
423 set_slob_page_free(sp, slob_list);
Nick Piggin95b35122007-07-15 23:38:07 -0700424 goto out;
425 }
Matt Mackall10cef602006-01-08 01:01:45 -0800426
Nick Piggin95b35122007-07-15 23:38:07 -0700427 /*
428 * Otherwise the page is already partially free, so find reinsertion
429 * point.
430 */
431 sp->units += units;
Matt Mackall10cef602006-01-08 01:01:45 -0800432
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500433 if (b < (slob_t *)sp->freelist) {
434 if (b + units == sp->freelist) {
435 units += slob_units(sp->freelist);
436 sp->freelist = slob_next(sp->freelist);
Matt Mackall679299b2008-02-04 22:29:37 -0800437 }
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500438 set_slob(b, units, sp->freelist);
439 sp->freelist = b;
Nick Piggin95b35122007-07-15 23:38:07 -0700440 } else {
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500441 prev = sp->freelist;
Nick Piggin95b35122007-07-15 23:38:07 -0700442 next = slob_next(prev);
443 while (b > next) {
444 prev = next;
445 next = slob_next(prev);
446 }
447
448 if (!slob_last(prev) && b + units == next) {
449 units += slob_units(next);
450 set_slob(b, units, slob_next(next));
451 } else
452 set_slob(b, units, next);
453
454 if (prev + slob_units(prev) == b) {
455 units = slob_units(b) + slob_units(prev);
456 set_slob(prev, units, slob_next(b));
457 } else
458 set_slob(prev, slob_units(prev), b);
459 }
460out:
Matt Mackall10cef602006-01-08 01:01:45 -0800461 spin_unlock_irqrestore(&slob_lock, flags);
462}
463
Nick Piggin95b35122007-07-15 23:38:07 -0700464/*
465 * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
466 */
467
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300468static __always_inline void *
469__do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller)
Matt Mackall10cef602006-01-08 01:01:45 -0800470{
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700471 unsigned int *m;
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700472 int minalign = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300473 void *ret;
Nick Piggin55394842007-07-15 23:38:09 -0700474
Steven Rostedtbd50cfa2011-06-07 07:18:45 -0400475 gfp &= gfp_allowed_mask;
476
Peter Zijlstrad92a8cf2017-03-03 10:13:38 +0100477 fs_reclaim_acquire(gfp);
478 fs_reclaim_release(gfp);
Nick Piggincf40bd12009-01-21 08:12:39 +0100479
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700480 if (size < PAGE_SIZE - minalign) {
481 int align = minalign;
482
483 /*
484 * For power of two sizes, guarantee natural alignment for
485 * kmalloc()'d objects.
486 */
487 if (is_power_of_2(size))
488 align = max(minalign, (int) size);
489
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700490 if (!size)
491 return ZERO_SIZE_PTR;
492
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700493 m = slob_alloc(size + minalign, gfp, align, node, minalign);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300494
MinChan Kim239f49c2008-05-19 22:12:08 +0900495 if (!m)
496 return NULL;
497 *m = size;
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700498 ret = (void *)m + minalign;
Nick Piggind87a1332007-07-15 23:38:08 -0700499
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300500 trace_kmalloc_node(caller, ret,
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700501 size, size + minalign, gfp, node);
Nick Piggind87a1332007-07-15 23:38:08 -0700502 } else {
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300503 unsigned int order = get_order(size);
Nick Piggind87a1332007-07-15 23:38:08 -0700504
David Rientjes8df275a2010-08-22 16:16:06 -0700505 if (likely(order))
506 gfp |= __GFP_COMP;
507 ret = slob_new_pages(gfp, order, node);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300508
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300509 trace_kmalloc_node(caller, ret,
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200510 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}
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300516
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000517void *__kmalloc(size_t size, gfp_t gfp)
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300518{
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000519 return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, _RET_IP_);
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300520}
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000521EXPORT_SYMBOL(__kmalloc);
Matt Mackall10cef602006-01-08 01:01:45 -0800522
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300523void *__kmalloc_track_caller(size_t size, gfp_t gfp, unsigned long caller)
524{
525 return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, caller);
526}
527
528#ifdef CONFIG_NUMA
David Rientjes82bd5502012-09-25 12:53:51 -0700529void *__kmalloc_node_track_caller(size_t size, gfp_t gfp,
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300530 int node, unsigned long caller)
531{
532 return __do_kmalloc_node(size, gfp, node, caller);
533}
534#endif
Ezequiel Garciaf3f74102012-09-08 17:47:53 -0300535
Matt Mackall10cef602006-01-08 01:01:45 -0800536void kfree(const void *block)
537{
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500538 struct page *sp;
Matt Mackall10cef602006-01-08 01:01:45 -0800539
Pekka Enberg2121db72009-03-25 11:05:57 +0200540 trace_kfree(_RET_IP_, block);
541
Satyam Sharma2408c552007-10-16 01:24:44 -0700542 if (unlikely(ZERO_OR_NULL_PTR(block)))
Matt Mackall10cef602006-01-08 01:01:45 -0800543 return;
Catalin Marinas4374e612009-06-11 13:23:17 +0100544 kmemleak_free(block);
Matt Mackall10cef602006-01-08 01:01:45 -0800545
Christoph Lameterb5568282012-06-13 10:24:54 -0500546 sp = virt_to_page(block);
547 if (PageSlab(sp)) {
Arnd Bergmann789306e2012-10-05 16:55:20 +0200548 int align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
Nick Piggin55394842007-07-15 23:38:09 -0700549 unsigned int *m = (unsigned int *)(block - align);
550 slob_free(m, *m + align);
Vlastimil Babka6a486c02019-10-06 17:58:42 -0700551 } else {
552 unsigned int order = compound_order(sp);
553 mod_node_page_state(page_pgdat(sp), NR_SLAB_UNRECLAIMABLE,
554 -(1 << order));
555 __free_pages(sp, order);
556
557 }
Matt Mackall10cef602006-01-08 01:01:45 -0800558}
Matt Mackall10cef602006-01-08 01:01:45 -0800559EXPORT_SYMBOL(kfree);
560
Nick Piggind87a1332007-07-15 23:38:08 -0700561/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
Marco Elver10d1f8c2019-07-11 20:54:14 -0700562size_t __ksize(const void *block)
Matt Mackall10cef602006-01-08 01:01:45 -0800563{
Christoph Lameterb8c24c42012-06-13 10:24:52 -0500564 struct page *sp;
Ezequiel Garcia999d8792012-10-19 09:33:10 -0300565 int align;
566 unsigned int *m;
Matt Mackall10cef602006-01-08 01:01:45 -0800567
Christoph Lameteref8b4522007-10-16 01:24:46 -0700568 BUG_ON(!block);
569 if (unlikely(block == ZERO_SIZE_PTR))
Matt Mackall10cef602006-01-08 01:01:45 -0800570 return 0;
571
Christoph Lameterb5568282012-06-13 10:24:54 -0500572 sp = virt_to_page(block);
Ezequiel Garcia999d8792012-10-19 09:33:10 -0300573 if (unlikely(!PageSlab(sp)))
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -0700574 return page_size(sp);
Ezequiel Garcia999d8792012-10-19 09:33:10 -0300575
Arnd Bergmann789306e2012-10-05 16:55:20 +0200576 align = max_t(size_t, ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN);
Ezequiel Garcia999d8792012-10-19 09:33:10 -0300577 m = (unsigned int *)(block - align);
578 return SLOB_UNITS(*m) * SLOB_UNIT;
Matt Mackall10cef602006-01-08 01:01:45 -0800579}
Marco Elver10d1f8c2019-07-11 20:54:14 -0700580EXPORT_SYMBOL(__ksize);
Matt Mackall10cef602006-01-08 01:01:45 -0800581
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800582int __kmem_cache_create(struct kmem_cache *c, slab_flags_t flags)
Matt Mackall10cef602006-01-08 01:01:45 -0800583{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -0800584 if (flags & SLAB_TYPESAFE_BY_RCU) {
Christoph Lameter278b1bb2012-09-05 00:20:34 +0000585 /* leave room for rcu footer at the end of object */
586 c->size += sizeof(struct slob_rcu);
Christoph Lameter039363f2012-07-06 15:25:10 -0500587 }
Christoph Lameter278b1bb2012-09-05 00:20:34 +0000588 c->flags = flags;
Christoph Lameter278b1bb2012-09-05 00:20:34 +0000589 return 0;
Matt Mackall10cef602006-01-08 01:01:45 -0800590}
Matt Mackall10cef602006-01-08 01:01:45 -0800591
Fabian Frederickc21a6da2015-04-14 15:44:34 -0700592static void *slob_alloc_node(struct kmem_cache *c, gfp_t flags, int node)
Matt Mackall10cef602006-01-08 01:01:45 -0800593{
594 void *b;
595
Steven Rostedtbd50cfa2011-06-07 07:18:45 -0400596 flags &= gfp_allowed_mask;
597
Peter Zijlstrad92a8cf2017-03-03 10:13:38 +0100598 fs_reclaim_acquire(flags);
599 fs_reclaim_release(flags);
Steven Rostedtbd50cfa2011-06-07 07:18:45 -0400600
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300601 if (c->size < PAGE_SIZE) {
Vlastimil Babka59bb4792019-10-06 17:58:45 -0700602 b = slob_alloc(c->size, flags, c->align, node, 0);
Ezequiel Garciafe74fe22012-10-19 09:33:11 -0300603 trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size,
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200604 SLOB_UNITS(c->size) * SLOB_UNIT,
605 flags, node);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300606 } else {
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800607 b = slob_new_pages(flags, get_order(c->size), node);
Ezequiel Garciafe74fe22012-10-19 09:33:11 -0300608 trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size,
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200609 PAGE_SIZE << get_order(c->size),
610 flags, node);
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300611 }
Matt Mackall10cef602006-01-08 01:01:45 -0800612
Matthew Wilcox128227e2018-06-07 17:05:13 -0700613 if (b && c->ctor) {
614 WARN_ON_ONCE(flags & __GFP_ZERO);
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700615 c->ctor(b);
Matthew Wilcox128227e2018-06-07 17:05:13 -0700616 }
Matt Mackall10cef602006-01-08 01:01:45 -0800617
Catalin Marinas4374e612009-06-11 13:23:17 +0100618 kmemleak_alloc_recursive(b, c->size, 1, c->flags, flags);
Matt Mackall10cef602006-01-08 01:01:45 -0800619 return b;
620}
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000621
622void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
623{
624 return slob_alloc_node(cachep, flags, NUMA_NO_NODE);
625}
626EXPORT_SYMBOL(kmem_cache_alloc);
627
628#ifdef CONFIG_NUMA
629void *__kmalloc_node(size_t size, gfp_t gfp, int node)
630{
631 return __do_kmalloc_node(size, gfp, node, _RET_IP_);
632}
633EXPORT_SYMBOL(__kmalloc_node);
634
635void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t gfp, int node)
636{
637 return slob_alloc_node(cachep, gfp, node);
638}
Paul Mundt6193a2f2007-07-15 23:38:22 -0700639EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000640#endif
Matt Mackall10cef602006-01-08 01:01:45 -0800641
Nick Pigginafc0ced2007-05-16 22:10:49 -0700642static void __kmem_cache_free(void *b, int size)
643{
644 if (size < PAGE_SIZE)
645 slob_free(b, size);
646 else
Américo Wang6e9ed0c2009-01-19 02:00:38 +0800647 slob_free_pages(b, get_order(size));
Nick Pigginafc0ced2007-05-16 22:10:49 -0700648}
649
650static void kmem_rcu_free(struct rcu_head *head)
651{
652 struct slob_rcu *slob_rcu = (struct slob_rcu *)head;
653 void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu));
654
655 __kmem_cache_free(b, slob_rcu->size);
656}
657
Matt Mackall10cef602006-01-08 01:01:45 -0800658void kmem_cache_free(struct kmem_cache *c, void *b)
659{
Catalin Marinas4374e612009-06-11 13:23:17 +0100660 kmemleak_free_recursive(b, c->flags);
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -0800661 if (unlikely(c->flags & SLAB_TYPESAFE_BY_RCU)) {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700662 struct slob_rcu *slob_rcu;
663 slob_rcu = b + (c->size - sizeof(struct slob_rcu));
Nick Pigginafc0ced2007-05-16 22:10:49 -0700664 slob_rcu->size = c->size;
665 call_rcu(&slob_rcu->head, kmem_rcu_free);
666 } else {
Nick Pigginafc0ced2007-05-16 22:10:49 -0700667 __kmem_cache_free(b, c->size);
668 }
Eduard - Gabriel Munteanu3eae2cb22008-08-10 20:14:07 +0300669
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200670 trace_kmem_cache_free(_RET_IP_, b);
Matt Mackall10cef602006-01-08 01:01:45 -0800671}
672EXPORT_SYMBOL(kmem_cache_free);
673
Christoph Lameter484748f2015-09-04 15:45:34 -0700674void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
675{
676 __kmem_cache_free_bulk(s, size, p);
677}
678EXPORT_SYMBOL(kmem_cache_free_bulk);
679
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800680int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
Christoph Lameter484748f2015-09-04 15:45:34 -0700681 void **p)
682{
683 return __kmem_cache_alloc_bulk(s, flags, size, p);
684}
685EXPORT_SYMBOL(kmem_cache_alloc_bulk);
686
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000687int __kmem_cache_shutdown(struct kmem_cache *c)
688{
689 /* No way to check for remaining objects */
690 return 0;
691}
692
Dmitry Safonov52b4b952016-02-17 13:11:37 -0800693void __kmem_cache_release(struct kmem_cache *c)
694{
695}
696
Vladimir Davydov89e364d2016-12-12 16:41:32 -0800697int __kmem_cache_shrink(struct kmem_cache *d)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800698{
699 return 0;
700}
Christoph Lameter2e892f42006-12-13 00:34:23 -0800701
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000702struct kmem_cache kmem_cache_boot = {
703 .name = "kmem_cache",
704 .size = sizeof(struct kmem_cache),
705 .flags = SLAB_PANIC,
706 .align = ARCH_KMALLOC_MINALIGN,
707};
708
Dimitri Gorokhovikbcb4ddb2006-12-29 16:48:28 -0800709void __init kmem_cache_init(void)
710{
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000711 kmem_cache = &kmem_cache_boot;
Christoph Lameter97d06602012-07-06 15:25:11 -0500712 slab_state = UP;
Matt Mackall10cef602006-01-08 01:01:45 -0800713}
Wu Fengguangbbff2e42009-08-06 11:36:25 +0300714
715void __init kmem_cache_init_late(void)
716{
Christoph Lameter97d06602012-07-06 15:25:11 -0500717 slab_state = FULL;
Wu Fengguangbbff2e42009-08-06 11:36:25 +0300718}