blob: 7ab05621052dc5a4b5f3add96a92ca54e76b10c2 [file] [log] [blame]
Thomas Gleixner09c434b2019-05-19 13:08:20 +01001// SPDX-License-Identifier: GPL-2.0-only
Vitaly Wool9a001fc2016-05-20 16:58:30 -07002/*
3 * z3fold.c
4 *
5 * Author: Vitaly Wool <vitaly.wool@konsulko.com>
6 * Copyright (C) 2016, Sony Mobile Communications Inc.
7 *
8 * This implementation is based on zbud written by Seth Jennings.
9 *
10 * z3fold is an special purpose allocator for storing compressed pages. It
11 * can store up to three compressed pages per page which improves the
12 * compression ratio of zbud while retaining its main concepts (e. g. always
13 * storing an integral number of objects per page) and simplicity.
14 * It still has simple and deterministic reclaim properties that make it
15 * preferable to a higher density approach (with no requirement on integral
16 * number of object per page) when reclaim is used.
17 *
18 * As in zbud, pages are divided into "chunks". The size of the chunks is
19 * fixed at compile time and is determined by NCHUNKS_ORDER below.
20 *
21 * z3fold doesn't export any API and is meant to be used via zpool API.
22 */
23
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/atomic.h>
Vitaly Woold30561c2017-09-06 16:24:47 -070027#include <linux/sched.h>
Vitaly Wool1f862982019-05-13 17:22:52 -070028#include <linux/cpumask.h>
Vitaly Wool9a001fc2016-05-20 16:58:30 -070029#include <linux/list.h>
30#include <linux/mm.h>
31#include <linux/module.h>
Vitaly Wool1f862982019-05-13 17:22:52 -070032#include <linux/page-flags.h>
33#include <linux/migrate.h>
34#include <linux/node.h>
35#include <linux/compaction.h>
Vitaly Woold30561c2017-09-06 16:24:47 -070036#include <linux/percpu.h>
Vitaly Wool9a001fc2016-05-20 16:58:30 -070037#include <linux/preempt.h>
Vitaly Woold30561c2017-09-06 16:24:47 -070038#include <linux/workqueue.h>
Vitaly Wool9a001fc2016-05-20 16:58:30 -070039#include <linux/slab.h>
40#include <linux/spinlock.h>
41#include <linux/zpool.h>
Qian Caiaf4798a2020-05-27 22:20:40 -070042#include <linux/kmemleak.h>
Vitaly Wool9a001fc2016-05-20 16:58:30 -070043
Vitaly Wool9a001fc2016-05-20 16:58:30 -070044/*
45 * NCHUNKS_ORDER determines the internal allocation granularity, effectively
46 * adjusting internal fragmentation. It also determines the number of
47 * freelists maintained in each pool. NCHUNKS_ORDER of 6 means that the
Vitaly Woolede93212017-02-24 14:57:17 -080048 * allocation granularity will be in chunks of size PAGE_SIZE/64. Some chunks
49 * in the beginning of an allocated page are occupied by z3fold header, so
50 * NCHUNKS will be calculated to 63 (or 62 in case CONFIG_DEBUG_SPINLOCK=y),
51 * which shows the max number of free chunks in z3fold page, also there will
52 * be 63, or 62, respectively, freelists per pool.
Vitaly Wool9a001fc2016-05-20 16:58:30 -070053 */
54#define NCHUNKS_ORDER 6
55
56#define CHUNK_SHIFT (PAGE_SHIFT - NCHUNKS_ORDER)
57#define CHUNK_SIZE (1 << CHUNK_SHIFT)
Vitaly Woolede93212017-02-24 14:57:17 -080058#define ZHDR_SIZE_ALIGNED round_up(sizeof(struct z3fold_header), CHUNK_SIZE)
59#define ZHDR_CHUNKS (ZHDR_SIZE_ALIGNED >> CHUNK_SHIFT)
60#define TOTAL_CHUNKS (PAGE_SIZE >> CHUNK_SHIFT)
Miaohe Line3c0db42021-06-30 18:50:24 -070061#define NCHUNKS (TOTAL_CHUNKS - ZHDR_CHUNKS)
Vitaly Wool9a001fc2016-05-20 16:58:30 -070062
zhong jiangf201ebd2017-02-22 15:46:51 -080063#define BUDDY_MASK (0x3)
Vitaly Woolca0246b2018-11-16 15:07:56 -080064#define BUDDY_SHIFT 2
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -070065#define SLOTS_ALIGN (0x40)
66
67/*****************
68 * Structures
69*****************/
70struct z3fold_pool;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -070071
72enum buddy {
73 HEADLESS = 0,
74 FIRST,
75 MIDDLE,
76 LAST,
77 BUDDIES_MAX = LAST
78};
79
80struct z3fold_buddy_slots {
81 /*
82 * we are using BUDDY_MASK in handle_to_buddy etc. so there should
83 * be enough slots to hold all possible variants
84 */
85 unsigned long slot[BUDDY_MASK + 1];
Vitaly Woolfc548862020-12-14 19:12:30 -080086 unsigned long pool; /* back link */
Vitaly Wool4a3ac932019-11-30 17:56:11 -080087 rwlock_t lock;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -070088};
89#define HANDLE_FLAG_MASK (0x03)
90
91/*
92 * struct z3fold_header - z3fold page metadata occupying first chunks of each
93 * z3fold page, except for HEADLESS pages
94 * @buddy: links the z3fold page into the relevant list in the
95 * pool
96 * @page_lock: per-page lock
97 * @refcount: reference count for the z3fold page
98 * @work: work_struct for page layout optimization
99 * @slots: pointer to the structure holding buddy slots
Vitaly Woolbb9a3742019-07-16 16:25:48 -0700100 * @pool: pointer to the containing pool
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700101 * @cpu: CPU which this page "belongs" to
102 * @first_chunks: the size of the first buddy in chunks, 0 if free
103 * @middle_chunks: the size of the middle buddy in chunks, 0 if free
104 * @last_chunks: the size of the last buddy in chunks, 0 if free
105 * @first_num: the starting number (for the first handle)
Vitaly Wool1f862982019-05-13 17:22:52 -0700106 * @mapped_count: the number of objects currently mapped
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700107 */
108struct z3fold_header {
109 struct list_head buddy;
110 spinlock_t page_lock;
111 struct kref refcount;
112 struct work_struct work;
113 struct z3fold_buddy_slots *slots;
Vitaly Woolbb9a3742019-07-16 16:25:48 -0700114 struct z3fold_pool *pool;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700115 short cpu;
116 unsigned short first_chunks;
117 unsigned short middle_chunks;
118 unsigned short last_chunks;
119 unsigned short start_middle;
120 unsigned short first_num:2;
Vitaly Wool1f862982019-05-13 17:22:52 -0700121 unsigned short mapped_count:2;
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800122 unsigned short foreign_handles:2;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700123};
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700124
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700125/**
126 * struct z3fold_pool - stores metadata for each z3fold pool
Vitaly Woold30561c2017-09-06 16:24:47 -0700127 * @name: pool name
Domenico Cerasuoloe774a7b2023-06-12 11:38:11 +0200128 * @lock: protects pool unbuddied lists
Vitaly Woold30561c2017-09-06 16:24:47 -0700129 * @stale_lock: protects pool stale page list
130 * @unbuddied: per-cpu array of lists tracking z3fold pages that contain 2-
131 * buddies; the list each z3fold page is added to depends on
132 * the size of its free region.
Vitaly Woold30561c2017-09-06 16:24:47 -0700133 * @stale: list of pages marked for freeing
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700134 * @pages_nr: number of z3fold pages in the pool.
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700135 * @c_handle: cache for z3fold_buddy_slots allocation
Vitaly Woold30561c2017-09-06 16:24:47 -0700136 * @compact_wq: workqueue for page layout background optimization
137 * @release_wq: workqueue for safe page release
138 * @work: work_struct for safe page release
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700139 *
140 * This structure is allocated at pool creation time and maintains metadata
141 * pertaining to a particular z3fold pool.
142 */
143struct z3fold_pool {
Vitaly Woold30561c2017-09-06 16:24:47 -0700144 const char *name;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700145 spinlock_t lock;
Vitaly Woold30561c2017-09-06 16:24:47 -0700146 spinlock_t stale_lock;
147 struct list_head *unbuddied;
Vitaly Woold30561c2017-09-06 16:24:47 -0700148 struct list_head stale;
Vitaly Wool12d59ae2017-02-24 14:57:15 -0800149 atomic64_t pages_nr;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700150 struct kmem_cache *c_handle;
Vitaly Woold30561c2017-09-06 16:24:47 -0700151 struct workqueue_struct *compact_wq;
152 struct workqueue_struct *release_wq;
153 struct work_struct work;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700154};
155
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700156/*
157 * Internal z3fold page flags
158 */
159enum z3fold_page_flags {
Vitaly Wool5a27aa82017-02-24 14:57:26 -0800160 PAGE_HEADLESS = 0,
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700161 MIDDLE_CHUNK_MAPPED,
Vitaly Woold30561c2017-09-06 16:24:47 -0700162 NEEDS_COMPACTING,
Vitaly Wool6098d7e2018-05-11 16:01:46 -0700163 PAGE_STALE,
Vitaly Woolca0246b2018-11-16 15:07:56 -0800164 PAGE_CLAIMED, /* by either reclaim or free */
Miaohe Lin943fb612022-04-29 14:40:43 +0800165 PAGE_MIGRATED, /* page is migrated and soon to be released */
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700166};
167
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800168/*
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800169 * handle flags, go under HANDLE_FLAG_MASK
170 */
171enum z3fold_handle_flags {
172 HANDLES_NOFREE = 0,
173};
174
175/*
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800176 * Forward declarations
177 */
178static struct z3fold_header *__z3fold_alloc(struct z3fold_pool *, size_t, bool);
179static void compact_page_work(struct work_struct *w);
180
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700181/*****************
182 * Helpers
183*****************/
184
185/* Converts an allocation size in bytes to size in z3fold chunks */
186static int size_to_chunks(size_t size)
187{
188 return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
189}
190
191#define for_each_unbuddied_list(_iter, _begin) \
192 for ((_iter) = (_begin); (_iter) < NCHUNKS; (_iter)++)
193
Vitaly Woolbb9f6f62019-05-31 22:30:39 -0700194static inline struct z3fold_buddy_slots *alloc_slots(struct z3fold_pool *pool,
195 gfp_t gfp)
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700196{
Miaohe Linf4bad642022-04-29 14:40:43 +0800197 struct z3fold_buddy_slots *slots = kmem_cache_zalloc(pool->c_handle,
198 gfp);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700199
200 if (slots) {
Qian Caiaf4798a2020-05-27 22:20:40 -0700201 /* It will be freed separately in free_handle(). */
202 kmemleak_not_leak(slots);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700203 slots->pool = (unsigned long)pool;
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800204 rwlock_init(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700205 }
206
207 return slots;
208}
209
210static inline struct z3fold_pool *slots_to_pool(struct z3fold_buddy_slots *s)
211{
212 return (struct z3fold_pool *)(s->pool & ~HANDLE_FLAG_MASK);
213}
214
215static inline struct z3fold_buddy_slots *handle_to_slots(unsigned long handle)
216{
217 return (struct z3fold_buddy_slots *)(handle & ~(SLOTS_ALIGN - 1));
218}
219
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800220/* Lock a z3fold page */
221static inline void z3fold_page_lock(struct z3fold_header *zhdr)
222{
223 spin_lock(&zhdr->page_lock);
224}
225
226/* Try to lock a z3fold page */
227static inline int z3fold_page_trylock(struct z3fold_header *zhdr)
228{
229 return spin_trylock(&zhdr->page_lock);
230}
231
232/* Unlock a z3fold page */
233static inline void z3fold_page_unlock(struct z3fold_header *zhdr)
234{
235 spin_unlock(&zhdr->page_lock);
236}
237
Miaohe Lin767cc6c2021-06-30 18:50:33 -0700238/* return locked z3fold page if it's not headless */
239static inline struct z3fold_header *get_z3fold_header(unsigned long handle)
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800240{
241 struct z3fold_buddy_slots *slots;
242 struct z3fold_header *zhdr;
243 int locked = 0;
244
245 if (!(handle & (1 << PAGE_HEADLESS))) {
246 slots = handle_to_slots(handle);
247 do {
248 unsigned long addr;
249
250 read_lock(&slots->lock);
251 addr = *(unsigned long *)handle;
252 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
Miaohe Lin767cc6c2021-06-30 18:50:33 -0700253 locked = z3fold_page_trylock(zhdr);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800254 read_unlock(&slots->lock);
Miaohe Lin943fb612022-04-29 14:40:43 +0800255 if (locked) {
256 struct page *page = virt_to_page(zhdr);
257
258 if (!test_bit(PAGE_MIGRATED, &page->private))
259 break;
260 z3fold_page_unlock(zhdr);
261 }
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800262 cpu_relax();
Miaohe Lin767cc6c2021-06-30 18:50:33 -0700263 } while (true);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800264 } else {
265 zhdr = (struct z3fold_header *)(handle & PAGE_MASK);
266 }
267
268 return zhdr;
269}
270
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800271static inline void put_z3fold_header(struct z3fold_header *zhdr)
272{
273 struct page *page = virt_to_page(zhdr);
274
275 if (!test_bit(PAGE_HEADLESS, &page->private))
276 z3fold_page_unlock(zhdr);
277}
278
Vitaly Woolfc548862020-12-14 19:12:30 -0800279static inline void free_handle(unsigned long handle, struct z3fold_header *zhdr)
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700280{
281 struct z3fold_buddy_slots *slots;
282 int i;
283 bool is_free;
284
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800285 if (WARN_ON(*(unsigned long *)handle == 0))
286 return;
287
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700288 slots = handle_to_slots(handle);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800289 write_lock(&slots->lock);
290 *(unsigned long *)handle = 0;
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800291
292 if (test_bit(HANDLES_NOFREE, &slots->pool)) {
293 write_unlock(&slots->lock);
294 return; /* simple case, nothing else to do */
295 }
296
Vitaly Woolfc548862020-12-14 19:12:30 -0800297 if (zhdr->slots != slots)
298 zhdr->foreign_handles--;
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800299
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700300 is_free = true;
301 for (i = 0; i <= BUDDY_MASK; i++) {
302 if (slots->slot[i]) {
303 is_free = false;
304 break;
305 }
306 }
Uladzislau Rezkid8f117a2020-05-22 22:23:12 -0700307 write_unlock(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700308
309 if (is_free) {
310 struct z3fold_pool *pool = slots_to_pool(slots);
311
Vitaly Woolfc548862020-12-14 19:12:30 -0800312 if (zhdr->slots == slots)
313 zhdr->slots = NULL;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700314 kmem_cache_free(pool->c_handle, slots);
315 }
316}
317
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700318/* Initializes the z3fold header of a newly allocated z3fold page */
Vitaly Wool63398412019-09-23 15:36:51 -0700319static struct z3fold_header *init_z3fold_page(struct page *page, bool headless,
Vitaly Woolbb9f6f62019-05-31 22:30:39 -0700320 struct z3fold_pool *pool, gfp_t gfp)
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700321{
322 struct z3fold_header *zhdr = page_address(page);
Vitaly Wool63398412019-09-23 15:36:51 -0700323 struct z3fold_buddy_slots *slots;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700324
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700325 clear_bit(PAGE_HEADLESS, &page->private);
326 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
Vitaly Woold30561c2017-09-06 16:24:47 -0700327 clear_bit(NEEDS_COMPACTING, &page->private);
328 clear_bit(PAGE_STALE, &page->private);
Vitaly Woolca0246b2018-11-16 15:07:56 -0800329 clear_bit(PAGE_CLAIMED, &page->private);
Miaohe Lin943fb612022-04-29 14:40:43 +0800330 clear_bit(PAGE_MIGRATED, &page->private);
Vitaly Wool63398412019-09-23 15:36:51 -0700331 if (headless)
332 return zhdr;
333
334 slots = alloc_slots(pool, gfp);
335 if (!slots)
336 return NULL;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700337
Miaohe Linc457cd92021-02-24 12:09:22 -0800338 memset(zhdr, 0, sizeof(*zhdr));
Vitaly Wool2f1e5e42017-02-24 14:57:23 -0800339 spin_lock_init(&zhdr->page_lock);
Vitaly Wool5a27aa82017-02-24 14:57:26 -0800340 kref_init(&zhdr->refcount);
Vitaly Woold30561c2017-09-06 16:24:47 -0700341 zhdr->cpu = -1;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700342 zhdr->slots = slots;
Vitaly Woolbb9a3742019-07-16 16:25:48 -0700343 zhdr->pool = pool;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700344 INIT_LIST_HEAD(&zhdr->buddy);
Vitaly Woold30561c2017-09-06 16:24:47 -0700345 INIT_WORK(&zhdr->work, compact_page_work);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700346 return zhdr;
347}
348
349/* Resets the struct page fields and frees the page */
Vitaly Wool1f862982019-05-13 17:22:52 -0700350static void free_z3fold_page(struct page *page, bool headless)
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700351{
Vitaly Wool1f862982019-05-13 17:22:52 -0700352 if (!headless) {
353 lock_page(page);
354 __ClearPageMovable(page);
355 unlock_page(page);
356 }
Vitaly Wool5a27aa82017-02-24 14:57:26 -0800357 __free_page(page);
358}
359
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700360/* Helper function to build the index */
361static inline int __idx(struct z3fold_header *zhdr, enum buddy bud)
362{
363 return (bud + zhdr->first_num) & BUDDY_MASK;
364}
365
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700366/*
Zhongkun He929e4c32024-02-19 10:44:53 +0800367 * Encodes the handle of a particular buddy within a z3fold page.
368 * Zhdr->page_lock should be held as this function accesses first_num
369 * if bud != HEADLESS.
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700370 */
Vitaly Wool3f9d2b52019-09-23 15:33:02 -0700371static unsigned long __encode_handle(struct z3fold_header *zhdr,
372 struct z3fold_buddy_slots *slots,
373 enum buddy bud)
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700374{
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700375 unsigned long h = (unsigned long)zhdr;
376 int idx = 0;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700377
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700378 /*
379 * For a headless page, its handle is its pointer with the extra
380 * PAGE_HEADLESS bit set
381 */
382 if (bud == HEADLESS)
383 return h | (1 << PAGE_HEADLESS);
384
385 /* otherwise, return pointer to encoded handle */
386 idx = __idx(zhdr, bud);
387 h += idx;
388 if (bud == LAST)
389 h |= (zhdr->last_chunks << BUDDY_SHIFT);
390
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800391 write_lock(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700392 slots->slot[idx] = h;
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800393 write_unlock(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700394 return (unsigned long)&slots->slot[idx];
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700395}
396
Vitaly Wool3f9d2b52019-09-23 15:33:02 -0700397static unsigned long encode_handle(struct z3fold_header *zhdr, enum buddy bud)
398{
399 return __encode_handle(zhdr, zhdr->slots, bud);
400}
401
Vitaly Woolca0246b2018-11-16 15:07:56 -0800402/* only for LAST bud, returns zero otherwise */
403static unsigned short handle_to_chunks(unsigned long handle)
404{
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800405 struct z3fold_buddy_slots *slots = handle_to_slots(handle);
406 unsigned long addr;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700407
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800408 read_lock(&slots->lock);
409 addr = *(unsigned long *)handle;
410 read_unlock(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700411 return (addr & ~PAGE_MASK) >> BUDDY_SHIFT;
Vitaly Woolca0246b2018-11-16 15:07:56 -0800412}
413
zhong jiangf201ebd2017-02-22 15:46:51 -0800414/*
415 * (handle & BUDDY_MASK) < zhdr->first_num is possible in encode_handle
416 * but that doesn't matter. because the masking will result in the
417 * correct buddy number.
418 */
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700419static enum buddy handle_to_buddy(unsigned long handle)
420{
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700421 struct z3fold_header *zhdr;
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800422 struct z3fold_buddy_slots *slots = handle_to_slots(handle);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700423 unsigned long addr;
424
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800425 read_lock(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700426 WARN_ON(handle & (1 << PAGE_HEADLESS));
427 addr = *(unsigned long *)handle;
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800428 read_unlock(&slots->lock);
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700429 zhdr = (struct z3fold_header *)(addr & PAGE_MASK);
430 return (addr - zhdr->first_num) & BUDDY_MASK;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700431}
432
Vitaly Wool9050cce2019-05-13 17:22:43 -0700433static inline struct z3fold_pool *zhdr_to_pool(struct z3fold_header *zhdr)
434{
Vitaly Woolbb9a3742019-07-16 16:25:48 -0700435 return zhdr->pool;
Vitaly Wool9050cce2019-05-13 17:22:43 -0700436}
437
Vitaly Woold30561c2017-09-06 16:24:47 -0700438static void __release_z3fold_page(struct z3fold_header *zhdr, bool locked)
439{
440 struct page *page = virt_to_page(zhdr);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700441 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
Vitaly Woold30561c2017-09-06 16:24:47 -0700442
443 WARN_ON(!list_empty(&zhdr->buddy));
444 set_bit(PAGE_STALE, &page->private);
Vitaly Wool35529352017-10-03 16:15:06 -0700445 clear_bit(NEEDS_COMPACTING, &page->private);
Vitaly Woold30561c2017-09-06 16:24:47 -0700446 spin_lock(&pool->lock);
Vitaly Woold30561c2017-09-06 16:24:47 -0700447 spin_unlock(&pool->lock);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800448
Vitaly Woold30561c2017-09-06 16:24:47 -0700449 if (locked)
450 z3fold_page_unlock(zhdr);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800451
Vitaly Woold30561c2017-09-06 16:24:47 -0700452 spin_lock(&pool->stale_lock);
453 list_add(&zhdr->buddy, &pool->stale);
454 queue_work(pool->release_wq, &pool->work);
455 spin_unlock(&pool->stale_lock);
Miaohe Lin5e36c252022-04-28 23:16:05 -0700456
457 atomic64_dec(&pool->pages_nr);
Vitaly Woold30561c2017-09-06 16:24:47 -0700458}
459
Vitaly Woold30561c2017-09-06 16:24:47 -0700460static void release_z3fold_page_locked(struct kref *ref)
461{
462 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
463 refcount);
464 WARN_ON(z3fold_page_trylock(zhdr));
465 __release_z3fold_page(zhdr, true);
466}
467
468static void release_z3fold_page_locked_list(struct kref *ref)
469{
470 struct z3fold_header *zhdr = container_of(ref, struct z3fold_header,
471 refcount);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700472 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800473
Vitaly Wool9050cce2019-05-13 17:22:43 -0700474 spin_lock(&pool->lock);
Vitaly Woold30561c2017-09-06 16:24:47 -0700475 list_del_init(&zhdr->buddy);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700476 spin_unlock(&pool->lock);
Vitaly Woold30561c2017-09-06 16:24:47 -0700477
478 WARN_ON(z3fold_page_trylock(zhdr));
479 __release_z3fold_page(zhdr, true);
480}
481
Ruan Jinjie73d47192023-08-03 19:38:23 +0800482static inline int put_z3fold_locked(struct z3fold_header *zhdr)
483{
484 return kref_put(&zhdr->refcount, release_z3fold_page_locked);
485}
486
487static inline int put_z3fold_locked_list(struct z3fold_header *zhdr)
488{
489 return kref_put(&zhdr->refcount, release_z3fold_page_locked_list);
490}
491
Vitaly Woold30561c2017-09-06 16:24:47 -0700492static void free_pages_work(struct work_struct *w)
493{
494 struct z3fold_pool *pool = container_of(w, struct z3fold_pool, work);
495
496 spin_lock(&pool->stale_lock);
497 while (!list_empty(&pool->stale)) {
498 struct z3fold_header *zhdr = list_first_entry(&pool->stale,
499 struct z3fold_header, buddy);
500 struct page *page = virt_to_page(zhdr);
501
502 list_del(&zhdr->buddy);
503 if (WARN_ON(!test_bit(PAGE_STALE, &page->private)))
504 continue;
Vitaly Woold30561c2017-09-06 16:24:47 -0700505 spin_unlock(&pool->stale_lock);
506 cancel_work_sync(&zhdr->work);
Vitaly Wool1f862982019-05-13 17:22:52 -0700507 free_z3fold_page(page, false);
Vitaly Woold30561c2017-09-06 16:24:47 -0700508 cond_resched();
509 spin_lock(&pool->stale_lock);
510 }
511 spin_unlock(&pool->stale_lock);
512}
513
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700514/*
515 * Returns the number of free chunks in a z3fold page.
516 * NB: can't be used with HEADLESS pages.
517 */
518static int num_free_chunks(struct z3fold_header *zhdr)
519{
520 int nfree;
521 /*
522 * If there is a middle object, pick up the bigger free space
523 * either before or after it. Otherwise just subtract the number
524 * of chunks occupied by the first and the last objects.
525 */
526 if (zhdr->middle_chunks != 0) {
527 int nfree_before = zhdr->first_chunks ?
Vitaly Woolede93212017-02-24 14:57:17 -0800528 0 : zhdr->start_middle - ZHDR_CHUNKS;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700529 int nfree_after = zhdr->last_chunks ?
Vitaly Woolede93212017-02-24 14:57:17 -0800530 0 : TOTAL_CHUNKS -
531 (zhdr->start_middle + zhdr->middle_chunks);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700532 nfree = max(nfree_before, nfree_after);
533 } else
534 nfree = NCHUNKS - zhdr->first_chunks - zhdr->last_chunks;
535 return nfree;
536}
537
Vitaly Wool9050cce2019-05-13 17:22:43 -0700538/* Add to the appropriate unbuddied list */
539static inline void add_to_unbuddied(struct z3fold_pool *pool,
540 struct z3fold_header *zhdr)
541{
542 if (zhdr->first_chunks == 0 || zhdr->last_chunks == 0 ||
543 zhdr->middle_chunks == 0) {
Vitaly Wool135f97f2020-12-14 19:12:36 -0800544 struct list_head *unbuddied;
Vitaly Wool9050cce2019-05-13 17:22:43 -0700545 int freechunks = num_free_chunks(zhdr);
Vitaly Wool135f97f2020-12-14 19:12:36 -0800546
547 migrate_disable();
548 unbuddied = this_cpu_ptr(pool->unbuddied);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700549 spin_lock(&pool->lock);
550 list_add(&zhdr->buddy, &unbuddied[freechunks]);
551 spin_unlock(&pool->lock);
552 zhdr->cpu = smp_processor_id();
Vitaly Wool135f97f2020-12-14 19:12:36 -0800553 migrate_enable();
Vitaly Wool9050cce2019-05-13 17:22:43 -0700554 }
555}
556
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800557static inline enum buddy get_free_buddy(struct z3fold_header *zhdr, int chunks)
558{
559 enum buddy bud = HEADLESS;
560
561 if (zhdr->middle_chunks) {
562 if (!zhdr->first_chunks &&
563 chunks <= zhdr->start_middle - ZHDR_CHUNKS)
564 bud = FIRST;
565 else if (!zhdr->last_chunks)
566 bud = LAST;
567 } else {
568 if (!zhdr->first_chunks)
569 bud = FIRST;
570 else if (!zhdr->last_chunks)
571 bud = LAST;
572 else
573 bud = MIDDLE;
574 }
575
576 return bud;
577}
578
Vitaly Woolede93212017-02-24 14:57:17 -0800579static inline void *mchunk_memmove(struct z3fold_header *zhdr,
580 unsigned short dst_chunk)
581{
582 void *beg = zhdr;
583 return memmove(beg + (dst_chunk << CHUNK_SHIFT),
584 beg + (zhdr->start_middle << CHUNK_SHIFT),
585 zhdr->middle_chunks << CHUNK_SHIFT);
586}
587
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800588static inline bool buddy_single(struct z3fold_header *zhdr)
589{
590 return !((zhdr->first_chunks && zhdr->middle_chunks) ||
591 (zhdr->first_chunks && zhdr->last_chunks) ||
592 (zhdr->middle_chunks && zhdr->last_chunks));
593}
594
595static struct z3fold_header *compact_single_buddy(struct z3fold_header *zhdr)
596{
597 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
598 void *p = zhdr;
599 unsigned long old_handle = 0;
600 size_t sz = 0;
601 struct z3fold_header *new_zhdr = NULL;
602 int first_idx = __idx(zhdr, FIRST);
603 int middle_idx = __idx(zhdr, MIDDLE);
604 int last_idx = __idx(zhdr, LAST);
605 unsigned short *moved_chunks = NULL;
606
607 /*
608 * No need to protect slots here -- all the slots are "local" and
609 * the page lock is already taken
610 */
611 if (zhdr->first_chunks && zhdr->slots->slot[first_idx]) {
612 p += ZHDR_SIZE_ALIGNED;
613 sz = zhdr->first_chunks << CHUNK_SHIFT;
614 old_handle = (unsigned long)&zhdr->slots->slot[first_idx];
615 moved_chunks = &zhdr->first_chunks;
616 } else if (zhdr->middle_chunks && zhdr->slots->slot[middle_idx]) {
617 p += zhdr->start_middle << CHUNK_SHIFT;
618 sz = zhdr->middle_chunks << CHUNK_SHIFT;
619 old_handle = (unsigned long)&zhdr->slots->slot[middle_idx];
620 moved_chunks = &zhdr->middle_chunks;
621 } else if (zhdr->last_chunks && zhdr->slots->slot[last_idx]) {
622 p += PAGE_SIZE - (zhdr->last_chunks << CHUNK_SHIFT);
623 sz = zhdr->last_chunks << CHUNK_SHIFT;
624 old_handle = (unsigned long)&zhdr->slots->slot[last_idx];
625 moved_chunks = &zhdr->last_chunks;
626 }
627
628 if (sz > 0) {
629 enum buddy new_bud = HEADLESS;
630 short chunks = size_to_chunks(sz);
631 void *q;
632
633 new_zhdr = __z3fold_alloc(pool, sz, false);
634 if (!new_zhdr)
635 return NULL;
636
637 if (WARN_ON(new_zhdr == zhdr))
638 goto out_fail;
639
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800640 new_bud = get_free_buddy(new_zhdr, chunks);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800641 q = new_zhdr;
642 switch (new_bud) {
643 case FIRST:
644 new_zhdr->first_chunks = chunks;
645 q += ZHDR_SIZE_ALIGNED;
646 break;
647 case MIDDLE:
648 new_zhdr->middle_chunks = chunks;
649 new_zhdr->start_middle =
650 new_zhdr->first_chunks + ZHDR_CHUNKS;
651 q += new_zhdr->start_middle << CHUNK_SHIFT;
652 break;
653 case LAST:
654 new_zhdr->last_chunks = chunks;
655 q += PAGE_SIZE - (new_zhdr->last_chunks << CHUNK_SHIFT);
656 break;
657 default:
658 goto out_fail;
659 }
660 new_zhdr->foreign_handles++;
661 memcpy(q, p, sz);
662 write_lock(&zhdr->slots->lock);
663 *(unsigned long *)old_handle = (unsigned long)new_zhdr +
664 __idx(new_zhdr, new_bud);
665 if (new_bud == LAST)
666 *(unsigned long *)old_handle |=
667 (new_zhdr->last_chunks << BUDDY_SHIFT);
668 write_unlock(&zhdr->slots->lock);
669 add_to_unbuddied(pool, new_zhdr);
670 z3fold_page_unlock(new_zhdr);
671
672 *moved_chunks = 0;
673 }
674
675 return new_zhdr;
676
677out_fail:
Ruan Jinjie73d47192023-08-03 19:38:23 +0800678 if (new_zhdr && !put_z3fold_locked(new_zhdr)) {
Miaohe Lin5e36c252022-04-28 23:16:05 -0700679 add_to_unbuddied(pool, new_zhdr);
680 z3fold_page_unlock(new_zhdr);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800681 }
682 return NULL;
683
684}
685
Vitaly Wool1b096e52017-02-24 14:57:20 -0800686#define BIG_CHUNK_GAP 3
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700687/* Has to be called with lock held */
688static int z3fold_compact_page(struct z3fold_header *zhdr)
689{
690 struct page *page = virt_to_page(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700691
Vitaly Woolede93212017-02-24 14:57:17 -0800692 if (test_bit(MIDDLE_CHUNK_MAPPED, &page->private))
693 return 0; /* can't move middle chunk, it's used */
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700694
Vitaly Wool1f862982019-05-13 17:22:52 -0700695 if (unlikely(PageIsolated(page)))
696 return 0;
697
Vitaly Woolede93212017-02-24 14:57:17 -0800698 if (zhdr->middle_chunks == 0)
699 return 0; /* nothing to compact */
700
701 if (zhdr->first_chunks == 0 && zhdr->last_chunks == 0) {
702 /* move to the beginning */
703 mchunk_memmove(zhdr, ZHDR_CHUNKS);
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700704 zhdr->first_chunks = zhdr->middle_chunks;
705 zhdr->middle_chunks = 0;
706 zhdr->start_middle = 0;
707 zhdr->first_num++;
Vitaly Wool1b096e52017-02-24 14:57:20 -0800708 return 1;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700709 }
Vitaly Wool1b096e52017-02-24 14:57:20 -0800710
711 /*
712 * moving data is expensive, so let's only do that if
713 * there's substantial gain (at least BIG_CHUNK_GAP chunks)
714 */
715 if (zhdr->first_chunks != 0 && zhdr->last_chunks == 0 &&
716 zhdr->start_middle - (zhdr->first_chunks + ZHDR_CHUNKS) >=
717 BIG_CHUNK_GAP) {
718 mchunk_memmove(zhdr, zhdr->first_chunks + ZHDR_CHUNKS);
719 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
720 return 1;
721 } else if (zhdr->last_chunks != 0 && zhdr->first_chunks == 0 &&
722 TOTAL_CHUNKS - (zhdr->last_chunks + zhdr->start_middle
723 + zhdr->middle_chunks) >=
724 BIG_CHUNK_GAP) {
725 unsigned short new_start = TOTAL_CHUNKS - zhdr->last_chunks -
726 zhdr->middle_chunks;
727 mchunk_memmove(zhdr, new_start);
728 zhdr->start_middle = new_start;
729 return 1;
730 }
731
732 return 0;
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700733}
734
Vitaly Woold30561c2017-09-06 16:24:47 -0700735static void do_compact_page(struct z3fold_header *zhdr, bool locked)
736{
Vitaly Wool9050cce2019-05-13 17:22:43 -0700737 struct z3fold_pool *pool = zhdr_to_pool(zhdr);
Vitaly Woold30561c2017-09-06 16:24:47 -0700738 struct page *page;
Vitaly Woold30561c2017-09-06 16:24:47 -0700739
740 page = virt_to_page(zhdr);
741 if (locked)
742 WARN_ON(z3fold_page_trylock(zhdr));
743 else
744 z3fold_page_lock(zhdr);
Vitaly Wool5d03a662017-11-17 15:26:16 -0800745 if (WARN_ON(!test_and_clear_bit(NEEDS_COMPACTING, &page->private))) {
Vitaly Woold30561c2017-09-06 16:24:47 -0700746 z3fold_page_unlock(zhdr);
747 return;
748 }
749 spin_lock(&pool->lock);
750 list_del_init(&zhdr->buddy);
751 spin_unlock(&pool->lock);
752
Ruan Jinjie73d47192023-08-03 19:38:23 +0800753 if (put_z3fold_locked(zhdr))
Vitaly Wool5d03a662017-11-17 15:26:16 -0800754 return;
Vitaly Wool5d03a662017-11-17 15:26:16 -0800755
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800756 if (test_bit(PAGE_STALE, &page->private) ||
757 test_and_set_bit(PAGE_CLAIMED, &page->private)) {
Vitaly Wool1f862982019-05-13 17:22:52 -0700758 z3fold_page_unlock(zhdr);
759 return;
760 }
761
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800762 if (!zhdr->foreign_handles && buddy_single(zhdr) &&
763 zhdr->mapped_count == 0 && compact_single_buddy(zhdr)) {
Ruan Jinjie73d47192023-08-03 19:38:23 +0800764 if (!put_z3fold_locked(zhdr)) {
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800765 clear_bit(PAGE_CLAIMED, &page->private);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800766 z3fold_page_unlock(zhdr);
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800767 }
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800768 return;
769 }
770
Vitaly Woold30561c2017-09-06 16:24:47 -0700771 z3fold_compact_page(zhdr);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700772 add_to_unbuddied(pool, zhdr);
Vitaly Wooldcf5aed2020-12-14 19:12:33 -0800773 clear_bit(PAGE_CLAIMED, &page->private);
Vitaly Woold30561c2017-09-06 16:24:47 -0700774 z3fold_page_unlock(zhdr);
775}
776
777static void compact_page_work(struct work_struct *w)
778{
779 struct z3fold_header *zhdr = container_of(w, struct z3fold_header,
780 work);
781
782 do_compact_page(zhdr, false);
783}
784
Vitaly Wool9050cce2019-05-13 17:22:43 -0700785/* returns _locked_ z3fold page header or NULL */
786static inline struct z3fold_header *__z3fold_alloc(struct z3fold_pool *pool,
787 size_t size, bool can_sleep)
788{
789 struct z3fold_header *zhdr = NULL;
790 struct page *page;
791 struct list_head *unbuddied;
792 int chunks = size_to_chunks(size), i;
793
794lookup:
Vitaly Wool135f97f2020-12-14 19:12:36 -0800795 migrate_disable();
Vitaly Wool9050cce2019-05-13 17:22:43 -0700796 /* First, try to find an unbuddied z3fold page. */
Vitaly Wool135f97f2020-12-14 19:12:36 -0800797 unbuddied = this_cpu_ptr(pool->unbuddied);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700798 for_each_unbuddied_list(i, chunks) {
799 struct list_head *l = &unbuddied[i];
800
801 zhdr = list_first_entry_or_null(READ_ONCE(l),
802 struct z3fold_header, buddy);
803
804 if (!zhdr)
805 continue;
806
807 /* Re-check under lock. */
808 spin_lock(&pool->lock);
Vitaly Wool9050cce2019-05-13 17:22:43 -0700809 if (unlikely(zhdr != list_first_entry(READ_ONCE(l),
810 struct z3fold_header, buddy)) ||
811 !z3fold_page_trylock(zhdr)) {
812 spin_unlock(&pool->lock);
813 zhdr = NULL;
Vitaly Wool135f97f2020-12-14 19:12:36 -0800814 migrate_enable();
Vitaly Wool9050cce2019-05-13 17:22:43 -0700815 if (can_sleep)
816 cond_resched();
817 goto lookup;
818 }
819 list_del_init(&zhdr->buddy);
820 zhdr->cpu = -1;
821 spin_unlock(&pool->lock);
822
823 page = virt_to_page(zhdr);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800824 if (test_bit(NEEDS_COMPACTING, &page->private) ||
825 test_bit(PAGE_CLAIMED, &page->private)) {
Vitaly Wool9050cce2019-05-13 17:22:43 -0700826 z3fold_page_unlock(zhdr);
827 zhdr = NULL;
Vitaly Wool135f97f2020-12-14 19:12:36 -0800828 migrate_enable();
Vitaly Wool9050cce2019-05-13 17:22:43 -0700829 if (can_sleep)
830 cond_resched();
831 goto lookup;
832 }
833
834 /*
835 * this page could not be removed from its unbuddied
836 * list while pool lock was held, and then we've taken
837 * page lock so kref_put could not be called before
838 * we got here, so it's safe to just call kref_get()
839 */
840 kref_get(&zhdr->refcount);
841 break;
842 }
Vitaly Wool135f97f2020-12-14 19:12:36 -0800843 migrate_enable();
Vitaly Wool9050cce2019-05-13 17:22:43 -0700844
Vitaly Wool351618b2019-05-13 17:22:46 -0700845 if (!zhdr) {
846 int cpu;
847
848 /* look for _exact_ match on other cpus' lists */
849 for_each_online_cpu(cpu) {
850 struct list_head *l;
851
852 unbuddied = per_cpu_ptr(pool->unbuddied, cpu);
853 spin_lock(&pool->lock);
854 l = &unbuddied[chunks];
855
856 zhdr = list_first_entry_or_null(READ_ONCE(l),
857 struct z3fold_header, buddy);
858
859 if (!zhdr || !z3fold_page_trylock(zhdr)) {
860 spin_unlock(&pool->lock);
861 zhdr = NULL;
862 continue;
863 }
864 list_del_init(&zhdr->buddy);
865 zhdr->cpu = -1;
866 spin_unlock(&pool->lock);
867
868 page = virt_to_page(zhdr);
Vitaly Wool4a3ac932019-11-30 17:56:11 -0800869 if (test_bit(NEEDS_COMPACTING, &page->private) ||
870 test_bit(PAGE_CLAIMED, &page->private)) {
Vitaly Wool351618b2019-05-13 17:22:46 -0700871 z3fold_page_unlock(zhdr);
872 zhdr = NULL;
873 if (can_sleep)
874 cond_resched();
875 continue;
876 }
877 kref_get(&zhdr->refcount);
878 break;
879 }
880 }
881
Miaohe Lin7c61c352022-04-29 14:40:43 +0800882 if (zhdr && !zhdr->slots) {
Miaohe Lin4c6bdb32022-04-29 14:40:43 +0800883 zhdr->slots = alloc_slots(pool, GFP_ATOMIC);
Miaohe Lin7c61c352022-04-29 14:40:43 +0800884 if (!zhdr->slots)
885 goto out_fail;
886 }
Vitaly Wool9050cce2019-05-13 17:22:43 -0700887 return zhdr;
Miaohe Lin7c61c352022-04-29 14:40:43 +0800888
889out_fail:
Ruan Jinjie73d47192023-08-03 19:38:23 +0800890 if (!put_z3fold_locked(zhdr)) {
Miaohe Lin7c61c352022-04-29 14:40:43 +0800891 add_to_unbuddied(pool, zhdr);
892 z3fold_page_unlock(zhdr);
893 }
894 return NULL;
Vitaly Wool9050cce2019-05-13 17:22:43 -0700895}
Vitaly Woold30561c2017-09-06 16:24:47 -0700896
897/*
898 * API Functions
899 */
900
901/**
902 * z3fold_create_pool() - create a new z3fold pool
903 * @name: pool name
904 * @gfp: gfp flags when allocating the z3fold pool structure
Vitaly Woold30561c2017-09-06 16:24:47 -0700905 *
906 * Return: pointer to the new z3fold pool or NULL if the metadata allocation
907 * failed.
908 */
Johannes Weiner6a05aa32022-11-28 11:16:11 -0800909static struct z3fold_pool *z3fold_create_pool(const char *name, gfp_t gfp)
Vitaly Woold30561c2017-09-06 16:24:47 -0700910{
911 struct z3fold_pool *pool = NULL;
912 int i, cpu;
913
914 pool = kzalloc(sizeof(struct z3fold_pool), gfp);
915 if (!pool)
916 goto out;
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700917 pool->c_handle = kmem_cache_create("z3fold_handle",
918 sizeof(struct z3fold_buddy_slots),
919 SLOTS_ALIGN, 0, NULL);
920 if (!pool->c_handle)
921 goto out_c;
Vitaly Woold30561c2017-09-06 16:24:47 -0700922 spin_lock_init(&pool->lock);
923 spin_lock_init(&pool->stale_lock);
Miaohe Line891f602021-06-30 18:50:30 -0700924 pool->unbuddied = __alloc_percpu(sizeof(struct list_head) * NCHUNKS,
925 __alignof__(struct list_head));
Xidong Wang1ec69952018-04-10 16:29:34 -0700926 if (!pool->unbuddied)
927 goto out_pool;
Vitaly Woold30561c2017-09-06 16:24:47 -0700928 for_each_possible_cpu(cpu) {
929 struct list_head *unbuddied =
930 per_cpu_ptr(pool->unbuddied, cpu);
931 for_each_unbuddied_list(i, 0)
932 INIT_LIST_HEAD(&unbuddied[i]);
933 }
Vitaly Woold30561c2017-09-06 16:24:47 -0700934 INIT_LIST_HEAD(&pool->stale);
935 atomic64_set(&pool->pages_nr, 0);
936 pool->name = name;
937 pool->compact_wq = create_singlethread_workqueue(pool->name);
938 if (!pool->compact_wq)
Xidong Wang1ec69952018-04-10 16:29:34 -0700939 goto out_unbuddied;
Vitaly Woold30561c2017-09-06 16:24:47 -0700940 pool->release_wq = create_singlethread_workqueue(pool->name);
941 if (!pool->release_wq)
942 goto out_wq;
943 INIT_WORK(&pool->work, free_pages_work);
Vitaly Woold30561c2017-09-06 16:24:47 -0700944 return pool;
945
946out_wq:
947 destroy_workqueue(pool->compact_wq);
Xidong Wang1ec69952018-04-10 16:29:34 -0700948out_unbuddied:
949 free_percpu(pool->unbuddied);
950out_pool:
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700951 kmem_cache_destroy(pool->c_handle);
952out_c:
Vitaly Woold30561c2017-09-06 16:24:47 -0700953 kfree(pool);
Xidong Wang1ec69952018-04-10 16:29:34 -0700954out:
Vitaly Woold30561c2017-09-06 16:24:47 -0700955 return NULL;
956}
957
958/**
959 * z3fold_destroy_pool() - destroys an existing z3fold pool
960 * @pool: the z3fold pool to be destroyed
961 *
962 * The pool should be emptied before this function is called.
963 */
964static void z3fold_destroy_pool(struct z3fold_pool *pool)
965{
Vitaly Wool7c2b8ba2019-05-13 17:22:49 -0700966 kmem_cache_destroy(pool->c_handle);
Henry Burns6051d3b2019-08-13 15:37:21 -0700967
968 /*
969 * We need to destroy pool->compact_wq before pool->release_wq,
970 * as any pending work on pool->compact_wq will call
971 * queue_work(pool->release_wq, &pool->work).
Henry Burnsb9970522019-08-13 15:37:25 -0700972 *
973 * There are still outstanding pages until both workqueues are drained,
974 * so we cannot unregister migration until then.
Henry Burns6051d3b2019-08-13 15:37:21 -0700975 */
976
Vitaly Woold30561c2017-09-06 16:24:47 -0700977 destroy_workqueue(pool->compact_wq);
Henry Burns6051d3b2019-08-13 15:37:21 -0700978 destroy_workqueue(pool->release_wq);
Miaohe Lindac0d1c2021-06-30 18:50:36 -0700979 free_percpu(pool->unbuddied);
Vitaly Woold30561c2017-09-06 16:24:47 -0700980 kfree(pool);
981}
982
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -0400983static const struct movable_operations z3fold_mops;
984
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700985/**
986 * z3fold_alloc() - allocates a region of a given size
987 * @pool: z3fold pool from which to allocate
988 * @size: size in bytes of the desired allocation
989 * @gfp: gfp flags used if the pool needs to grow
990 * @handle: handle of the new allocation
991 *
992 * This function will attempt to find a free region in the pool large enough to
993 * satisfy the allocation request. A search of the unbuddied lists is
994 * performed first. If no suitable free region is found, then a new page is
995 * allocated and added to the pool to satisfy the request.
996 *
Vitaly Wool9a001fc2016-05-20 16:58:30 -0700997 * Return: 0 if success and handle is set, otherwise -EINVAL if the size or
998 * gfp arguments are invalid or -ENOMEM if the pool was unable to allocate
999 * a new page.
1000 */
1001static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
1002 unsigned long *handle)
1003{
Vitaly Wool9050cce2019-05-13 17:22:43 -07001004 int chunks = size_to_chunks(size);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001005 struct z3fold_header *zhdr = NULL;
Vitaly Woold30561c2017-09-06 16:24:47 -07001006 struct page *page = NULL;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001007 enum buddy bud;
Matthew Wilcox8a97ea542018-04-10 16:29:37 -07001008 bool can_sleep = gfpflags_allow_blocking(gfp);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001009
Miaohe Linf4bad642022-04-29 14:40:43 +08001010 if (!size || (gfp & __GFP_HIGHMEM))
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001011 return -EINVAL;
1012
1013 if (size > PAGE_SIZE)
1014 return -ENOSPC;
1015
1016 if (size > PAGE_SIZE - ZHDR_SIZE_ALIGNED - CHUNK_SIZE)
1017 bud = HEADLESS;
1018 else {
Vitaly Wool9050cce2019-05-13 17:22:43 -07001019retry:
1020 zhdr = __z3fold_alloc(pool, size, can_sleep);
Vitaly Woold30561c2017-09-06 16:24:47 -07001021 if (zhdr) {
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001022 bud = get_free_buddy(zhdr, chunks);
1023 if (bud == HEADLESS) {
Ruan Jinjie73d47192023-08-03 19:38:23 +08001024 if (!put_z3fold_locked(zhdr))
Vitaly Woold30561c2017-09-06 16:24:47 -07001025 z3fold_page_unlock(zhdr);
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001026 pr_err("No free chunks in unbuddied\n");
1027 WARN_ON(1);
Vitaly Wool9050cce2019-05-13 17:22:43 -07001028 goto retry;
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001029 }
Vitaly Wool9050cce2019-05-13 17:22:43 -07001030 page = virt_to_page(zhdr);
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001031 goto found;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001032 }
1033 bud = FIRST;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001034 }
1035
Miaohe Lindf6f0f12022-04-29 14:40:43 +08001036 page = alloc_page(gfp);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001037 if (!page)
1038 return -ENOMEM;
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001039
Vitaly Wool63398412019-09-23 15:36:51 -07001040 zhdr = init_z3fold_page(page, bud == HEADLESS, pool, gfp);
Vitaly Wool9050cce2019-05-13 17:22:43 -07001041 if (!zhdr) {
1042 __free_page(page);
1043 return -ENOMEM;
1044 }
1045 atomic64_inc(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001046
1047 if (bud == HEADLESS) {
1048 set_bit(PAGE_HEADLESS, &page->private);
1049 goto headless;
1050 }
Henry Burns810481a2019-07-11 20:52:14 -07001051 if (can_sleep) {
1052 lock_page(page);
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -04001053 __SetPageMovable(page, &z3fold_mops);
Henry Burns810481a2019-07-11 20:52:14 -07001054 unlock_page(page);
1055 } else {
Miaohe Lin2c0f3512022-04-29 14:40:43 +08001056 WARN_ON(!trylock_page(page));
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -04001057 __SetPageMovable(page, &z3fold_mops);
Miaohe Lin2c0f3512022-04-29 14:40:43 +08001058 unlock_page(page);
Henry Burns810481a2019-07-11 20:52:14 -07001059 }
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001060 z3fold_page_lock(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001061
1062found:
1063 if (bud == FIRST)
1064 zhdr->first_chunks = chunks;
1065 else if (bud == LAST)
1066 zhdr->last_chunks = chunks;
1067 else {
1068 zhdr->middle_chunks = chunks;
Vitaly Woolede93212017-02-24 14:57:17 -08001069 zhdr->start_middle = zhdr->first_chunks + ZHDR_CHUNKS;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001070 }
Vitaly Wool9050cce2019-05-13 17:22:43 -07001071 add_to_unbuddied(pool, zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001072
1073headless:
Vitaly Woold30561c2017-09-06 16:24:47 -07001074 spin_lock(&pool->lock);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001075 *handle = encode_handle(zhdr, bud);
1076 spin_unlock(&pool->lock);
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001077 if (bud != HEADLESS)
1078 z3fold_page_unlock(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001079
1080 return 0;
1081}
1082
1083/**
1084 * z3fold_free() - frees the allocation associated with the given handle
1085 * @pool: pool in which the allocation resided
1086 * @handle: handle associated with the allocation returned by z3fold_alloc()
1087 *
1088 * In the case that the z3fold page in which the allocation resides is under
Miaohe Lined0e5dc2022-04-28 23:16:05 -07001089 * reclaim, as indicated by the PAGE_CLAIMED flag being set, this function
1090 * only sets the first|middle|last_chunks to 0. The page is actually freed
1091 * once all buddies are evicted (see z3fold_reclaim_page() below).
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001092 */
1093static void z3fold_free(struct z3fold_pool *pool, unsigned long handle)
1094{
1095 struct z3fold_header *zhdr;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001096 struct page *page;
1097 enum buddy bud;
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001098 bool page_claimed;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001099
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001100 zhdr = get_z3fold_header(handle);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001101 page = virt_to_page(zhdr);
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001102 page_claimed = test_and_set_bit(PAGE_CLAIMED, &page->private);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001103
1104 if (test_bit(PAGE_HEADLESS, &page->private)) {
Vitaly Woolca0246b2018-11-16 15:07:56 -08001105 /* if a headless page is under reclaim, just leave.
1106 * NB: we use test_and_set_bit for a reason: if the bit
1107 * has not been set before, we release this page
1108 * immediately so we don't care about its value any more.
1109 */
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001110 if (!page_claimed) {
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001111 put_z3fold_header(zhdr);
Vitaly Wool1f862982019-05-13 17:22:52 -07001112 free_z3fold_page(page, true);
Vitaly Woolca0246b2018-11-16 15:07:56 -08001113 atomic64_dec(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001114 }
Vitaly Woolca0246b2018-11-16 15:07:56 -08001115 return;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001116 }
1117
Vitaly Woolca0246b2018-11-16 15:07:56 -08001118 /* Non-headless case */
Vitaly Woolca0246b2018-11-16 15:07:56 -08001119 bud = handle_to_buddy(handle);
1120
1121 switch (bud) {
1122 case FIRST:
1123 zhdr->first_chunks = 0;
1124 break;
1125 case MIDDLE:
1126 zhdr->middle_chunks = 0;
1127 break;
1128 case LAST:
1129 zhdr->last_chunks = 0;
1130 break;
1131 default:
1132 pr_err("%s: unknown bud %d\n", __func__, bud);
1133 WARN_ON(1);
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001134 put_z3fold_header(zhdr);
Vitaly Woold30561c2017-09-06 16:24:47 -07001135 return;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001136 }
1137
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001138 if (!page_claimed)
Vitaly Woolfc548862020-12-14 19:12:30 -08001139 free_handle(handle, zhdr);
Ruan Jinjie73d47192023-08-03 19:38:23 +08001140 if (put_z3fold_locked_list(zhdr))
Vitaly Woold30561c2017-09-06 16:24:47 -07001141 return;
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001142 if (page_claimed) {
1143 /* the page has not been claimed by us */
Miaohe Lined0e5dc2022-04-28 23:16:05 -07001144 put_z3fold_header(zhdr);
Vitaly Wool6098d7e2018-05-11 16:01:46 -07001145 return;
1146 }
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001147 if (test_and_set_bit(NEEDS_COMPACTING, &page->private)) {
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001148 clear_bit(PAGE_CLAIMED, &page->private);
Miaohe Lin4a1c3832022-04-29 14:40:43 +08001149 put_z3fold_header(zhdr);
Vitaly Woold30561c2017-09-06 16:24:47 -07001150 return;
1151 }
1152 if (zhdr->cpu < 0 || !cpu_online(zhdr->cpu)) {
Vitaly Woold30561c2017-09-06 16:24:47 -07001153 zhdr->cpu = -1;
Vitaly Wool5d03a662017-11-17 15:26:16 -08001154 kref_get(&zhdr->refcount);
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001155 clear_bit(PAGE_CLAIMED, &page->private);
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001156 do_compact_page(zhdr, true);
Vitaly Woold30561c2017-09-06 16:24:47 -07001157 return;
1158 }
Vitaly Wool5d03a662017-11-17 15:26:16 -08001159 kref_get(&zhdr->refcount);
Vitaly Wool5b6807d2019-10-06 17:58:22 -07001160 clear_bit(PAGE_CLAIMED, &page->private);
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001161 queue_work_on(zhdr->cpu, pool->compact_wq, &zhdr->work);
1162 put_z3fold_header(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001163}
1164
1165/**
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001166 * z3fold_map() - maps the allocation associated with the given handle
1167 * @pool: pool in which the allocation resides
1168 * @handle: handle associated with the allocation to be mapped
1169 *
1170 * Extracts the buddy number from handle and constructs the pointer to the
1171 * correct starting chunk within the page.
1172 *
1173 * Returns: a pointer to the mapped allocation
1174 */
1175static void *z3fold_map(struct z3fold_pool *pool, unsigned long handle)
1176{
1177 struct z3fold_header *zhdr;
1178 struct page *page;
1179 void *addr;
1180 enum buddy buddy;
1181
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001182 zhdr = get_z3fold_header(handle);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001183 addr = zhdr;
1184 page = virt_to_page(zhdr);
1185
1186 if (test_bit(PAGE_HEADLESS, &page->private))
1187 goto out;
1188
1189 buddy = handle_to_buddy(handle);
1190 switch (buddy) {
1191 case FIRST:
1192 addr += ZHDR_SIZE_ALIGNED;
1193 break;
1194 case MIDDLE:
1195 addr += zhdr->start_middle << CHUNK_SHIFT;
1196 set_bit(MIDDLE_CHUNK_MAPPED, &page->private);
1197 break;
1198 case LAST:
Vitaly Woolca0246b2018-11-16 15:07:56 -08001199 addr += PAGE_SIZE - (handle_to_chunks(handle) << CHUNK_SHIFT);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001200 break;
1201 default:
1202 pr_err("unknown buddy id %d\n", buddy);
1203 WARN_ON(1);
1204 addr = NULL;
1205 break;
1206 }
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001207
Vitaly Wool1f862982019-05-13 17:22:52 -07001208 if (addr)
1209 zhdr->mapped_count++;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001210out:
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001211 put_z3fold_header(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001212 return addr;
1213}
1214
1215/**
1216 * z3fold_unmap() - unmaps the allocation associated with the given handle
1217 * @pool: pool in which the allocation resides
1218 * @handle: handle associated with the allocation to be unmapped
1219 */
1220static void z3fold_unmap(struct z3fold_pool *pool, unsigned long handle)
1221{
1222 struct z3fold_header *zhdr;
1223 struct page *page;
1224 enum buddy buddy;
1225
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001226 zhdr = get_z3fold_header(handle);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001227 page = virt_to_page(zhdr);
1228
Vitaly Wool2f1e5e42017-02-24 14:57:23 -08001229 if (test_bit(PAGE_HEADLESS, &page->private))
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001230 return;
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001231
1232 buddy = handle_to_buddy(handle);
1233 if (buddy == MIDDLE)
1234 clear_bit(MIDDLE_CHUNK_MAPPED, &page->private);
Vitaly Wool1f862982019-05-13 17:22:52 -07001235 zhdr->mapped_count--;
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001236 put_z3fold_header(zhdr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001237}
1238
1239/**
1240 * z3fold_get_pool_size() - gets the z3fold pool size in pages
1241 * @pool: pool whose size is being queried
1242 *
Vitaly Wool12d59ae2017-02-24 14:57:15 -08001243 * Returns: size in pages of the given pool.
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001244 */
1245static u64 z3fold_get_pool_size(struct z3fold_pool *pool)
1246{
Vitaly Wool12d59ae2017-02-24 14:57:15 -08001247 return atomic64_read(&pool->pages_nr);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001248}
1249
Vitaly Wool1f862982019-05-13 17:22:52 -07001250static bool z3fold_page_isolate(struct page *page, isolate_mode_t mode)
1251{
1252 struct z3fold_header *zhdr;
1253 struct z3fold_pool *pool;
1254
Vitaly Wool1f862982019-05-13 17:22:52 -07001255 VM_BUG_ON_PAGE(PageIsolated(page), page);
1256
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001257 if (test_bit(PAGE_HEADLESS, &page->private))
Vitaly Wool1f862982019-05-13 17:22:52 -07001258 return false;
1259
1260 zhdr = page_address(page);
1261 z3fold_page_lock(zhdr);
1262 if (test_bit(NEEDS_COMPACTING, &page->private) ||
1263 test_bit(PAGE_STALE, &page->private))
1264 goto out;
1265
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001266 if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0)
1267 goto out;
Vitaly Wool1f862982019-05-13 17:22:52 -07001268
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001269 if (test_and_set_bit(PAGE_CLAIMED, &page->private))
1270 goto out;
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001271 pool = zhdr_to_pool(zhdr);
1272 spin_lock(&pool->lock);
1273 if (!list_empty(&zhdr->buddy))
1274 list_del_init(&zhdr->buddy);
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001275 spin_unlock(&pool->lock);
1276
1277 kref_get(&zhdr->refcount);
1278 z3fold_page_unlock(zhdr);
1279 return true;
1280
Vitaly Wool1f862982019-05-13 17:22:52 -07001281out:
1282 z3fold_page_unlock(zhdr);
1283 return false;
1284}
1285
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -04001286static int z3fold_page_migrate(struct page *newpage, struct page *page,
1287 enum migrate_mode mode)
Vitaly Wool1f862982019-05-13 17:22:52 -07001288{
1289 struct z3fold_header *zhdr, *new_zhdr;
1290 struct z3fold_pool *pool;
Vitaly Wool1f862982019-05-13 17:22:52 -07001291
Vitaly Wool1f862982019-05-13 17:22:52 -07001292 VM_BUG_ON_PAGE(!PageIsolated(page), page);
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001293 VM_BUG_ON_PAGE(!test_bit(PAGE_CLAIMED, &page->private), page);
Henry Burns810481a2019-07-11 20:52:14 -07001294 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Vitaly Wool1f862982019-05-13 17:22:52 -07001295
1296 zhdr = page_address(page);
1297 pool = zhdr_to_pool(zhdr);
1298
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001299 if (!z3fold_page_trylock(zhdr))
Vitaly Wool1f862982019-05-13 17:22:52 -07001300 return -EAGAIN;
Vitaly Wool4a3ac932019-11-30 17:56:11 -08001301 if (zhdr->mapped_count != 0 || zhdr->foreign_handles != 0) {
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001302 clear_bit(PAGE_CLAIMED, &page->private);
Miaohe Lin4a1c3832022-04-29 14:40:43 +08001303 z3fold_page_unlock(zhdr);
Vitaly Wool1f862982019-05-13 17:22:52 -07001304 return -EBUSY;
1305 }
Henry Burnsc92d2f32019-07-16 16:26:21 -07001306 if (work_pending(&zhdr->work)) {
1307 z3fold_page_unlock(zhdr);
1308 return -EAGAIN;
1309 }
Vitaly Wool1f862982019-05-13 17:22:52 -07001310 new_zhdr = page_address(newpage);
1311 memcpy(new_zhdr, zhdr, PAGE_SIZE);
1312 newpage->private = page->private;
Miaohe Lin943fb612022-04-29 14:40:43 +08001313 set_bit(PAGE_MIGRATED, &page->private);
Vitaly Wool1f862982019-05-13 17:22:52 -07001314 z3fold_page_unlock(zhdr);
1315 spin_lock_init(&new_zhdr->page_lock);
Henry Burnsc92d2f32019-07-16 16:26:21 -07001316 INIT_WORK(&new_zhdr->work, compact_page_work);
1317 /*
1318 * z3fold_page_isolate() ensures that new_zhdr->buddy is empty,
1319 * so we only have to reinitialize it.
1320 */
1321 INIT_LIST_HEAD(&new_zhdr->buddy);
Vitaly Wool1f862982019-05-13 17:22:52 -07001322 __ClearPageMovable(page);
Vitaly Wool1f862982019-05-13 17:22:52 -07001323
1324 get_page(newpage);
1325 z3fold_page_lock(new_zhdr);
1326 if (new_zhdr->first_chunks)
1327 encode_handle(new_zhdr, FIRST);
1328 if (new_zhdr->last_chunks)
1329 encode_handle(new_zhdr, LAST);
1330 if (new_zhdr->middle_chunks)
1331 encode_handle(new_zhdr, MIDDLE);
1332 set_bit(NEEDS_COMPACTING, &newpage->private);
1333 new_zhdr->cpu = smp_processor_id();
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -04001334 __SetPageMovable(newpage, &z3fold_mops);
Vitaly Wool1f862982019-05-13 17:22:52 -07001335 z3fold_page_unlock(new_zhdr);
1336
1337 queue_work_on(new_zhdr->cpu, pool->compact_wq, &new_zhdr->work);
1338
Miaohe Lin943fb612022-04-29 14:40:43 +08001339 /* PAGE_CLAIMED and PAGE_MIGRATED are cleared now. */
1340 page->private = 0;
Vitaly Wool1f862982019-05-13 17:22:52 -07001341 put_page(page);
1342 return 0;
1343}
1344
1345static void z3fold_page_putback(struct page *page)
1346{
1347 struct z3fold_header *zhdr;
1348 struct z3fold_pool *pool;
1349
1350 zhdr = page_address(page);
1351 pool = zhdr_to_pool(zhdr);
1352
1353 z3fold_page_lock(zhdr);
1354 if (!list_empty(&zhdr->buddy))
1355 list_del_init(&zhdr->buddy);
1356 INIT_LIST_HEAD(&page->lru);
Ruan Jinjie73d47192023-08-03 19:38:23 +08001357 if (put_z3fold_locked(zhdr))
Vitaly Wool1f862982019-05-13 17:22:52 -07001358 return;
Miaohe Lin6cf9a342022-04-29 14:40:43 +08001359 if (list_empty(&zhdr->buddy))
1360 add_to_unbuddied(pool, zhdr);
Vitaly Wooldcf5aed2020-12-14 19:12:33 -08001361 clear_bit(PAGE_CLAIMED, &page->private);
Vitaly Wool1f862982019-05-13 17:22:52 -07001362 z3fold_page_unlock(zhdr);
1363}
1364
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -04001365static const struct movable_operations z3fold_mops = {
Vitaly Wool1f862982019-05-13 17:22:52 -07001366 .isolate_page = z3fold_page_isolate,
Matthew Wilcox (Oracle)68f27362022-06-07 15:38:48 -04001367 .migrate_page = z3fold_page_migrate,
Vitaly Wool1f862982019-05-13 17:22:52 -07001368 .putback_page = z3fold_page_putback,
1369};
1370
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001371/*****************
1372 * zpool
1373 ****************/
1374
Domenico Cerasuolo35499e22023-06-12 11:38:13 +02001375static void *z3fold_zpool_create(const char *name, gfp_t gfp)
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001376{
Domenico Cerasuoloe774a7b2023-06-12 11:38:11 +02001377 return z3fold_create_pool(name, gfp);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001378}
1379
1380static void z3fold_zpool_destroy(void *pool)
1381{
1382 z3fold_destroy_pool(pool);
1383}
1384
1385static int z3fold_zpool_malloc(void *pool, size_t size, gfp_t gfp,
1386 unsigned long *handle)
1387{
1388 return z3fold_alloc(pool, size, gfp, handle);
1389}
1390static void z3fold_zpool_free(void *pool, unsigned long handle)
1391{
1392 z3fold_free(pool, handle);
1393}
1394
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001395static void *z3fold_zpool_map(void *pool, unsigned long handle,
1396 enum zpool_mapmode mm)
1397{
1398 return z3fold_map(pool, handle);
1399}
1400static void z3fold_zpool_unmap(void *pool, unsigned long handle)
1401{
1402 z3fold_unmap(pool, handle);
1403}
1404
1405static u64 z3fold_zpool_total_size(void *pool)
1406{
1407 return z3fold_get_pool_size(pool) * PAGE_SIZE;
1408}
1409
1410static struct zpool_driver z3fold_zpool_driver = {
1411 .type = "z3fold",
Tian Taoe818e822021-02-25 17:18:22 -08001412 .sleep_mapped = true,
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001413 .owner = THIS_MODULE,
1414 .create = z3fold_zpool_create,
1415 .destroy = z3fold_zpool_destroy,
1416 .malloc = z3fold_zpool_malloc,
1417 .free = z3fold_zpool_free,
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001418 .map = z3fold_zpool_map,
1419 .unmap = z3fold_zpool_unmap,
1420 .total_size = z3fold_zpool_total_size,
1421};
1422
1423MODULE_ALIAS("zpool-z3fold");
1424
1425static int __init init_z3fold(void)
1426{
Miaohe Lin014284a2021-06-30 18:50:27 -07001427 /*
1428 * Make sure the z3fold header is not larger than the page size and
1429 * there has remaining spaces for its buddy.
1430 */
1431 BUILD_BUG_ON(ZHDR_SIZE_ALIGNED > PAGE_SIZE - CHUNK_SIZE);
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001432 zpool_register_driver(&z3fold_zpool_driver);
1433
1434 return 0;
1435}
1436
1437static void __exit exit_z3fold(void)
1438{
Vitaly Wool9a001fc2016-05-20 16:58:30 -07001439 zpool_unregister_driver(&z3fold_zpool_driver);
1440}
1441
1442module_init(init_z3fold);
1443module_exit(exit_z3fold);
1444
1445MODULE_LICENSE("GPL");
1446MODULE_AUTHOR("Vitaly Wool <vitalywool@gmail.com>");
1447MODULE_DESCRIPTION("3-Fold Allocator for Compressed Pages");